University of Southern California Dissertations and Theses

A theoretical and empirical examination of infant mortality decline in Brazil's northeast region, 1986--1996

(USC Thesis Other)

A theoretical and empirical examination of infant mortality decline in Brazil's northeast region, 1986--1996

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content INFORMATION TO USERS This manuscript has been reproduced from the microfilm master. UM I films the text directly from the original or copy submitted. Thus, some thesis and dissertation copies are in typewriter face, while others may be from any type of computer printer. The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleedthrough, substandard margins, and improper alignment can adversely affect reproductibn. In the unlikely event that the author did not send U M I a complete manuscript and there are missing pages, these w ill we noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion. Oversize materials (e.g., maps, drawings, charts) are reproduced by sectioning the original, beginning at the upper left-hand comer and continuing from left to right in equal sections with small overlaps. Photographs included in the original manuscript have t)een reproduced xerographically in this copy. Higher quality 6” x 9” black and white photographic prints are available for any photographs or illustrations appearing in this copy for an additional charge. Contact U M I directly to order. Bell & Howell Information and teaming 300 North Zeeb Road, Ann Arbor, M l 48106-1346 USA U IVQ 800-521-0600 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A THEORETICAL AND EMPIRICAL EXAMINATION OF INFANT MORTALITY DECLINE IN BRAZIL’S NORTHEAST REGION, 1986-1996 Volume I By Fernando Veiga Prata A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment o f the Requirements for the Degree DOCTOR OF PHILOSOPHY (Political Economy and Public Policy) August 1999 Copyright 1999 Fernando Veiga Prata Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UMI Number: 9955033 UMI UMI Microform9955033 Copyright 2000 by Bell & Howell Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. Bell & Howell Information and Leaming Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, M l 48106-1346 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. TH S 0KAOU A T I IC H O eL LOS A M C L eS . CA U fO SM lA W W This dSsaertûtiOK vrrittgn by îi^WKv4l>ft V feV iA « ■ « e * e « ^ » e » e e * « » a i s « * e e e e e e e » e e e W » e e » * # e e « « " S ® * » w » * « w e « t * * * * « » e 'e i « » e e e e e e e e e e e e under tht iireçtion o f Disseriâtien CammiiHc ond approved by all its mtmben, hat bftn praaehied to and aeeeptad by The C F B d u a te School m partial fulfillment of re- quiremerits for 0ie degree of DOCTOR OF PHUOSOPHY D B S G ^ T IO X C O M M ir r ^ * * » , # # $ # # # » # # # # # # * * * Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. T o the children of Brazil. rich or poor, black or white, bom in the Northeast or not, for the ones who made it and the for the ones who did not. u Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. AKNOWLEDGEMENTS I wish to thawlr Professor Richard Easterlin, chairman, of my dissertation committee, for inspiration, academic guidance, encouragement and expertise. Professor E^asterlin’ s contribution to my graduate trainmg defined the foundation and the underlying conceptual framework of this study. However, his influence extends fer beyond this dissertation. I would not be able to sufficient^' express my profound gratitude to Professor E^asterlin. I am also extremely grateful to the other two members of my dissertation committee: Professor Abraham Lowenthal and Professor David Heer. Professor Lowenthal deeply mfluenced my mtellectual development at U.S.C. I learned immensely from his graduate teachings and scholastic guidance. In addition to influencing my academic development. Professor Abraham Lowenthal taught me to understand Social Science m a gestalt frishion, by focusing not only on the summation and research answers but also on the mdividual structure and questions. I am also deeply indebted to Professor David Heer, Head of the Population Research Laboratory at U.S.C., for his encouragement, technical training and invaluable advice in the production of this work. Professor Heer was also ready to generously aid me in the clarification of demographic and quantitative issues. I am deeply thankful to him I alone am responsible for any errors or omissions present in this study. Very special thanks are also due to Dr. Farideh Motamedi, Associate Director of the Political E k:onom y & Public Policy Program, for her vital guidance and support throughout my graduate studies at the Uhiversily of Southern CaU fom ia. I am also very thankful to my uncde Nüo M . Rolim and to my aunt Luiza V eijg a Rolim for their precious and exemplary moral support. I also wish to thank my godson Lucas Pinheiro Prata, my brother Marœlo and my father Dr. M arcelo Prata dos Santos. I would also like to give very special thanks to my dear friends Professor Selma Holo, Fred Croton, Aires Conceicao and Dr. J.P. Wensel, vdio indirectly but very importantly, contributed to this dissertation. Finally I would like to wholeheartedly thank Mery - M ary A Külh- for all the in:q)iration, love and support, for the past, present and for the future. m Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. CONTENTS VOLUM E! 1. INTRODUCTION 2 2. BACKGROUND 16 2.1. The Physical Environment 16 2.2. The Economic and Social-Political Background 39 in Historical Perspective 3. THE NATURE AND FACTS OF MORTALITY REVOLUTION 88 AND INFANT MORTALITY DECLINE 3.1. Nature o f M ortality Revolution and Demographic Transition 88 3.1.1. Introduction 88 3.1.2. Endogenous Socioeconomic Changes Vs. Exogenous Health- 90 Medical Innovation Changes? Notes on the Nature o f Brazil’s Mortality Decline 3.1.3. Notes on the Timing and Interconnectedness o f Brazil’s Fast, 99 Dramatic and Dynamic Fertility and hifant Mortality Decline 3.2. Facts on M ortality Revolution and Di&nt Mortality Decline 103 in Brazil and in its N ortheast Region IV Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.2.1. Brazil’s Life Expectancy at Birth 103 3.2.2. Northeast’s Life Expectancy at Birth 111 3.2.2. Population Growth 114 3.2.3. Birth Rate and Total Fertility Rate 123 3.2.4. A Comparative Perspective o f Infent M ortality Decline in Brazil 128 and in its Northeast Region 3.2.5. The Impact o f Public Policy and Health on hifent M ortali^ Decline 143 4. CONCEPTUAL FRAMEWORK & LITERATURE REVIEW 156 5. CONCEPTUALIZATION OF VARIABLES 177 6. DATA & METHODS 199 6.1. Data Sources 199 6.2. Sampling Design 201 6.3. Data Manipulation 205 6.4. Analytical Framework 212 6.5. The Logistic Regression Method 215 6.6. The SPSS Logit Statistical Output 226 6.7. The Cox Regression M ethod 228 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. VOLUME n 7. RESULTS OF DESCRIPTIVE AND MULTIVARIATE 239 REGRESSION ANALYSIS 7.1. Introduction 239 7.2. Results o f Descriptive and M ultivariate Regression Analysis Applied 252 to the 1996 Logistic Regression and Cox Data Sets 7.2.1. Descriptive Results 252 7.2.2. Diagnostic Results for 1996 Logistic Regression Models 270 7.23. On the Regression Estimates 272 7.2.4. Logistic Regression Results for 1996 Models 273 7.2.5. Diagnostic Results for 1996 Cox Regression Models 284 7.26. Cox Regression Results fo r 1996 Models 288 7.3. Results o f Descriptive and M ultivariate Regression Analysis Applied 301 to the 1991 Logistic Regression and Cox Data Sets 7.3.1. Descriptive Results 301 7.3.2. Diagnostic Results for 1991 Logistic Regression Models 316 7.3.3. Logistic Regression Results for 1991 Models 318 7.3.4. Diagnostic Results for 1991 Cox Models 326 7.3.5. Cox R ^ e ssio n Results for 1991 Models 328 VI Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 7.4. Results o f Descriptive and Multivariate Regression Analysis ^ p lie d 342 to the 1986 Logistic Regression and Cox D ata Sets 7.4.1. Descriptive Results 342 7.4.2. Diagnostic Results for 1986 Logistic Regression Models 354 7.4.3. Logistic Regression Results for 1986 Models 356 7.4.4. Diagnostic Results for 1986 Cox Models 361 7.4.5. Cox Regression Results for 1986 Models 364 7.5. Results o f Descriptive and Multivariate Regression Analysis Applied 376 to the 1991-1986 Pooled Logistic Regression and Cox Data Sets 7.5.1. Descriptive Results 376 7.5.2. Diagnostic Results for 1991-1986 Logistic Regression Models 388 7.5.3. Logistic Regression Results for 1991-1986 Models 390 7.5.4. Diagnostic Results for 1991-1986 Cox Models 397 7.5.5. Cox Regression Results for 1991-1986 Models 400 7.6. Results ofDescriptive and Multivariate Regression Analysis Applied 412 to the Pooled 1996-1991-1986 Logistic Regression and Cox Data Sets 7.6.1. Descriptive Results 412 7.6.2. Diagnostic Results for 1996-1991-1986 Logistic Regression Models 425 7.6.3. Logistic Regression Results for 1996-1991-1986 Models 427 7.6.4. Diagnostic Results for 1996-1991-1986 Cox Models 434 7.6.5. Cox Regression Results for 1996-1991-1986 Models 437 vu Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 7 . 1. Results ofDescriptive and Nüiltîvariate Regression Analysis ^*plied 449 to the 1996-1991 Pooled Logistic Regression and Cox Data Sets 7.7.1. Descriptive Analysis 449 7.7.2. Diagnostic Results for 1996-1991 Logistic Regression Models 465 7.7.3. Logistic Regression Results for 1996-1991 Models 467 7.7.4. Diagnostic Results for 1996-1991 Cox Models 475 7.7.5. Cox Regression Results fo r 1996-1991 Models 479 8. CONCLUSION 493 Bibliography 518 A ppendix 528 vui Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. LIST OF TABLES Table 1. Brazil, Total Area, Resident Population, Population Density, Mean geometric Rate of Annual bicrease and Percent Distribution, by Major Regions and States (Federative Units) 21 2. Resident Population, Urban and Rural, and Urbanization Rate, by Major Regions and States (Federative Units) - 1991 23 3. Resident Population, by Sex and Sex ratio, by Age Groups - 1991 24 4. Territorial Area, by Major Regions and States (Feder. Units) - 1994 25 5. Administrative EXrolution, by Major Regions and States (Federative Units) - 1940/1995 26 6. Permanent private housing units, by Major Regions - Selected characteristics - 1991 29 7. Ebrtent of the border line, by neighboring countries and the Atlantic Ocean - 1994 30 8. Population by Ethnic Group - 1996 30 9. The Evolution of Brazil's H uman Development Index (H D l), 1970-1996 31 10. The Evolution of Brazil's Life Ebcpectancy at Birth, 1970-1996 32 11. The Evolution of Brazil's Life Expectancy Index for HD, 1970-1996 33 12. The Evolution of Brazil's Adult Literacy Rate, 1970-1996 (% ) 34 13. The Eîvolution of Brazil's Eklucation Index for HD, 1970-1996 35 14. The Evolution of Brazil's Per Capita GDP, 1970-1996 (US $ PPG) 36 15. The Evolution of Brazil's GDP Index for HD, 1970-1996 37 16. PC GDP Order - HDI Order, 1970-1996 38 17. Brazil's and Latin America's GDP PC Annual Average Growth Rates During Maddison's Phases of Development- 1900-1987 (% ) 71 ix Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 18. Brazil's Average Rate of Annual Growth of Population, Aggregate Product and Per Capita Product 72 19. Brazil's Average Rate of Annual Growth of Real Product PC 1940-1954 { % ) 72 20. Brazil's Average Rate of Annual Growth of Aggregate Product -1920-1957 (% ) 73 21. Brazil's Average Rate of Annual Growth of Agpregate Product Per Capita -1920-1957 (% ) 73 22. Brazil's Per Capita GDP Growth Rates, 1950-1990 (% ) 74 23. Brazil's Average Rate of Annual Growth of Aggregate Product Per Capita -1900-02 to 1969-71 (% ) 74 24. Brazil's Per Capita GDP, $ at 1985 US prices, 1900-1989 75 25. Brazil's Average Rate of Annual Growth of Aggregate and Sectoral Product -1900-02 to 1969-71 (% ) 75 26. Brazil's Average Rate of Annual Growth of Aggregate and Industrial Product 1948-52 to 1968 -74 (% ) 76 27. Recent Shares of GDP (%), 1965-1990 76 28. GDP Growth Rates by Sector, 1948-1976 (% ) 77 29. GDP Growth Rates by Sector, 1980-1990 (% ) 78 30. Growth Rates of Population and Labor Force in Brazil and in the Northeast, 1940-1970 (% ) 78 31. Regional Distribution of the Labor Force by Sector, Brazil 1940-1991 (% ) 79 32. Participation Rates and Structure of the Labor Force in Brazil and in its Northeast Region, 1940-1991 (% ) 80 33. General Historical Indicators, Brazil, 1865-1974 81 34. Urban Definition and Calculation 84 35. Urbanization, 1940-1996(%) 85 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 36. Real GDP Growth (%), 1921-1995 86 37. E ^volution. of Brazil's Crude Death Rate ^ e r 1,000 inh.) 1872-1990 106 38. The Evolution, of Brazil's Life Ebqxectangr at Birth according to Main Estimates, 1870-1996 109 39. Absolute and Relative (% ) Life Expectancy Gains by Decade, 1930-1991 110 40. Evolution of Life Ebcpectancy at Birth in Selected Countries - 1953-1993 (in years} 111 41. Life Expectanty at Birth by Great Regions, 1940-1991 (in years) 112 42. Life Expectancy at Birth- Brazil and its Northeast Region - 1940-1990 113 43. Brazil’s and Latin America's Average Rate of Growth of Population -1900-1987 (a.a. % ) 114 44. Brazil's Average Rate of Annual Growth of Population 115 45. Brazil's Mean Geometric Rate of Annual Population Growth 1872/1996 (% ) 117 4 6. General Characteristics - 1996 118 47. Mean Rates of Population Growth of Brazil for Selected Periods, 1799-1996 (a.a. %) 119 48. Brazil's Population- 1872-1999 (millions) 120 49. Urban and Rural Population of Brazil by Major Regions - 1940-1996(% Total Pop.) 121 50. Mean Geometric Rate of Annual Population Growth by Major Regions - Rural & Urban- 1940/1991 (% ) 122 51. Mean Geometric Rate of Annual Population Growth by Major Regions - 1890-1940 (% ) 123 52. Current Contraceptive Use among Women (15 to 49 years old) Living in Union - 1996 125 53. Evolution of Brazil's Crude Birth Rate (per 1,000 inh.) 1872-1990 126 xi Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 54. Total Fertüity Rate by Major Regions, 1940-1996 126 55. Total Fertüity Rate and lâfe Expectancy a t Birth by Sex- 1996 127 56. Infant Mortality by Sex and Ethnic Origin - 1996 127 57. EXrolntion of Brazil's Infant Mortality Rate by Great Regions, 1940-1996 (% ) 129 58. Evolution of Brazil's Infant Mortality Decline by Maj'or Regions, Decade and Selected Periods (% Change) 129 59. Infant Mortality Rates in Brazil and in the Northeast Region, 1965-1996 130 60. Average Annual Infant Mortality Decline Rates (1940-1990) 132 61. Contrasting Gains in Life Expectancy and Infant Mortality m Brazil and in the NE, by Decade (% ) 133 62. Brazil - Contribution of E^ach A ge Group to Life Expectancy Gains, 1940-1990 135 63. Northeast Region - Contribution of Each Age Group to Gains in L ife Expectancy at Birth, 1940-1990 136 63 A . The Evolution of Infant Mortality Rates in the Northeastern States, 1965-1994 (1 of 2) 138 63 B. The Evolution of Infant Mortality Rates in the Northeastern States, 1965-1994(1 of 2) 139 6 4. Observed and Moving Average Adj'usted Infant Mortality Rates in Brazü, 1929-1992 139 65. Observed and Moving Average Adjusted Infant Mortality Rates in the Northeast Region, 1926-1990 141 66. Distribution of Young Women by Eklucational Attainment (% ) 143 67. Brazil’ s Female Labor Force Participation Rate, 1960-1990(%) 144 68. Illiteracy Levels, 1872-1991 144 69. Public Expenditures on Health (%GDP), 1960-1991 149 XÜ Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 70. Data Sample Sizes 242 71. Means and Freq. for all independent variables of 1996 Ne Logistic Regression Data Set 262 72. Means and Freq. for all independent variables of 1996 Ne Cox Regression Data Set 265 73. 1996 Correlation Levels with Dependent Variable 269 74. Diagnostic Results, Predictive EfiBciency and Further Characteristics of the Logistic Regression Models for 1996 Data Set 271 75 A . Logistic Regression Results for 1996 Ne Data Set (1) 279 75 B. Logistic Regression Results for 1996 Ne Data Set (1 with no DPT Immunization) 280 76. Logistic Regression Results for 1996 Ne Data Set (2) 281 77. Logistic Regression Results for 1996 Ne Data Set (3) 282 78. Logistic Regression Results for 1996 Ne Data Set (4) 283 79. Diagnostic Results, Predictive Efficiency and Further Characteristics of the Cox Regression Models for 1996 Data Set (1 of 2) 286 80. Diagnostic Results, Predictive Efficiency and Further Characteristics of the Cox Regression Models for 1996 Data Set (2 of 2) 287 293 294 295 296 297 298 299 300 x iii 81. Cox Regression Results for 1996 Ne Data Set (1 ) 82. Cox Regression Results for 1996 Ne Data Set (2 ) 83. Cox Regression Results for 1996 Ne Data Set (3 ) 84. Cox Regression Results for 1996 Ne Data Set (4 ) 85. Cox Regression Results for 1996 Ne Data Set (5 ) 86. Cox Regression Results for 1996 Ne Data Set (6 ) 87. Cox Regression Results for 1996 Ne Data Set (7) 88. Cox Regression Results for 1996 Ne Data Set (8) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 89. Means and Freq. for all independent variables of 1991 Ne Logistic Regression. Data Set 308 90. Means and Freq. for all independent variables of 1991 Ne Cox Regression Data Set 311 91. 1991 Correlation Levels with Dependent Variable 315 92. Diagnostic Results, Predictive EfSciency and Further Characteristics of the Logistic Regression Models for 1991 Data Set 317 93. Logistic Regression Results for 1991 Ne Data Set (1) 322 94. Logistic Regression Results for 1991 Ne Data Set (2) 323 95. Logistic Regression Results for 1991 Ne Data Set (3) 324 96. Logistic Regression Results for 1991 Ne Data Set (4) 325 97. Diagnostic Results, Predictive Efficiency and Further Characteristics of the Cox Regression Models for 1991 Data Set (1 of 2) 327 98. Diagnostic Results, Predictive Efficiency and Further Characteristics of the Cox Regression Models for 1991 Data Set (2 of 2) 328 99. Cox Regression Results for 1991 N e Data Set (1) 334 100. Cox Regression Results for 1991 Ne Data Set (2) 335 101. Cox Regression Results for 1991 Ne Data Set (3) 336 102. Cox Regression Results for 1991 Ne Data Set (4) 337 103. Cox Regression Results for 1991 Ne Data Set (5) 338 104. Cox Regression Results for 1991 Ne Data Set (6) 339 105. Cox Regression Results for 1991 Ne Data Set (7) 340 106. Cox Regression Results for 1991 Ne Data Set (8) 341 107. Means and Freq. for all independent variables of 1986 N e Logistic Regression Data Set 347 108. Means and Freq. for all independent variables of 1986 Ne Cox Regression Data Set 350 xiv Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 109. 1986 Correlation. Levels with Dependent Variable 353 110. Diagnostic Results, Predictwe Effidenqr and Further Characteristics of the Logistic Regression Models for 1986 Data Set 355 111. Logistic Regression Results for 1986 Ne Data Set (1) 357 112. Logistic Regression Results for 1986 Ne Data Set (2) 358 113. Logistic Regression Results for 1986 Ne Data Set (3) 359 114. Logistic Regression Results for 1986 Ne Data Set (4) 360 115. Diagnostic Results, Predictive Efficiency and Further Characteristics of the Cox Regression Models for 1986 Data Set (1 of 2) 362 116. Diagnostic Results, Predictive Efficiency and Further Characteristics of the Cox Regression Models for 1986 Data Set (2 of 2) 363 117. Cox Regression Results for 1986 Ne Data Set (1 ) 368 118. Cox Regression Results for 1986 Ne Data Set (2 ) 369 119. Cox Regression Results for 1986 Ne Data Set (3 ) 370 120. Cox Regression Results for 1986 Ne Data Set (4 ) 371 121. Cox Regression Results for 1986 Ne Data Set (5 ) 372 122. Cox Regression Results for 1986 Ne Data Set (6 ) 373 123. Cox Regression Results for 1986 Ne Data Set (7 ) 374 124. Cox Regression Results for 1986 Ne Data Set (8 ) 375 125. Means and Freq. for all independent variables of 1991-86 Pooled Ne Logistic Regression Data Set 381 126. Means and Freq. for all independent variables of 1991-86 Pooled Ne Cox Regression Data Set 384 127. 1991-1986 Correlation Levels with Dependent Variable 387 128. Diagnostic Results, Predictive Efficiency and Further Characteristics of the Logistic Regression Models for 1991-1986 Pooled Data Set 389 XV Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 129. Logistic Regression. Results for 1991-1986 Ne Data Set (1) 393 130. Logistic Regression Results for 1991-1986 Ne Data Set (2) 394 131. Logistic Regression Results for 1991-1986 Ne Data Set (3) 395 132. Logistic Regression Results for 1991-1986 Ne Data Set (4) 396 133. Diagnostic Results, Predictive EfGciengr and Further Characteristics of the Cox Regression Models for 1991-86 Data Set (1 of 2) 398 134. Diagnostic Results, Predictive EfSciency and Further Characteristics of the Cox Regression Models for 1991-86 Data Set (2 of 2) 399 135. Cox Regression Results for 1991-1986 Ne Data Set (1) 404 136. Cox Regression Results for 1991-1986 Ne Data Set (2) 405 137. Cox Regression Results for 1991-1986 Ne Data Set (3) 406 138. Cox Regression Results for 1991-1986 Ne Data Set (4) 407 139. Cox Regression Results for 1991-1986 Ne Data Set (5) 408 140. Cox Regression Results for 1991-1986 Ne Data Set (6) 409 141. Cox Regression Results for 1991-1986 Ne Data Set (7) 410 142. Cox Regression Results for 1991-1986 Ne Data Set (7) 411 143. Means and Freq. for all independent variables of 1996-91-86 Pooled Ne Logistic Regression Data Set 418 144. Means and Freq. for all independent variables of 1996-91-86 Ne Cox Regression Data Set 421 145. 1996-1991-1986 Correlation Levels with Dependent Variable 424 146. Diagnostic Results, Predictnre EfSciency and Further Characteristics of the Logistic Regression Models for 1996-91-86 Pooled Data Set 426 147. Logistic Regression Results for 1996-91-86 Ne Data Set (1) 430 148. Logistic Regression Results for 1996-91-86 Ne Data Set (2) 431 149. Logistic Regression Results for 1996-91-86 Ne Data Set (3) 432 xvi Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 150. Logistic Régression Results for 1996-91-86 Ne Data Set (4 ) 433 151. Diagnostic Results, Predictive EfSciency and Further Characteristics of the Cox Regression Models for 1996-91-86 Data Set (1 of 2) 435 152. Diagnostic Results, Predictive EfSciency and Further Characteristics of the Cox Regression Models for 1996-91-86 Data Set 436 153. Cox Regression Results for 1996-91-86 Ne Data Set (1 ) 441 154. Cox Regression Results for 1996-91-86 Ne Data Set (2 ) 442 155. Cox Regression Results for 1996-91-86 Ne Data Set (3 ) 443 156. Cox Regression Results for 1996-91-86 Ne Data Set (4 ) 444 157. Cox Regression Results for 1996-91-86 Ne Data Set (5 ) 445 158. Cox Regression Results for 1996-91-86 Ne Data Set (6 ) 446 159. Cox Regression Results for 1996-91-86 Ne Data Set (7 ) 447 160. Cox Regression Results for 1996-91-86 Ne Data Set (8 ) 448 161. Means and Freq. for all independent variables of 1996-91 Pooled Ne Logistic Regression Data Set 457 162. Means and Freq. for all independent variables of 1996-91 Pooled Ne Cox Regression Data Set 460 163. 1996-1991 Correlation Levels with Dependent Variable 464 164. Diagnostic Results, Predictive Efficiency and Further Characteristics of the Logistic Regression Models for 1996-91 Pooled Data Set 466 165. Logistic Regression Results for 1996-1991 Ne Data Set (1) 471 166. Logistic Regression Results for 1996-1991 Ne Data Set (2) 472 167. Logistic Regression Results for 1996-1991 Ne Data Set (3 ) 473 168. Logistic Regression Results for 1996-1991 Ne Data Set (4) 474 169. Diagnostic Results, Predictive Efficiency and Further Characteristics of the Cox Regression Models for 1996-91 Data Set (1 of 2) 477 xvu Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 170. Diagnostic Results, Predictive Effîciency and Further Characteristics of the Cox Regression Models for 1996-91 Data Set (2 of 2) 478 171. Cox Regression Results for 1996-1991 Ne Data Set (1) 484 172. Cox Regresaon Results for 1996-1991 Ne Data Set (2) 485 173. Cox Regression Results for 1996-1991 Ne Data Set (3) 486 174. Cox Regression Results for 1996-1991 Ne Data Set (4) 487 175. Cox Regression Results for 1996-1991 Ne Data Set (5) 488 176. Cox Regression Results for 1996-1991 Ne Data Set (6) 489 177. Cox Regression Results for 1996-1991 Ne Data Set (7) 490 178. Cox Regression Results for 1996-1991 Ne Data Set (8) 491 179. Place of Residence, 1986 and 1996 496 180. Drinking Water, 1986 and 1996 497 181. Sewage, 1986 and 1996 498 182. Household Crowding, 1986 and 1996 497 183. Mother's Age Risk, 1986 and 1996 498 184. Mother’ s Ekiucational Attainment, 1986 and 1996 498 185. Goods, 1986 and 1996 498 186. Birth Order Risk, 1986 and 1996 499 187. Sex of infant, 1986 and 1996 499 188. Significant Independent Variables in LR Data Sets (1 of 2) 501 189. Significant Independent Variables in Cox Data Sets (2 of 2) 502 190. Significant hidependent Variables in Pooled LR Data Sets (1 of 2) 503 191. Significant Independent Variables in Pooled Cox Data Sets (2 of 2) 504 xvui Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 192. Correlatioii Matrix, Levels of Correlation Among Independent Variables, 1996 Logistic Regression Data Set (1 of 2) 528 193. Correlation Matrix, Levels of Correlation Among Independent Variables, 1996 Logistic Regression Data Set (2 of 2) 529 194. Correlation Matrix, Levels of Correlation Among ^dependent Variables, 1991 Logistic Regression Data Set (1 of 2) 530 195. Correlation Matrix, Levels of Correlation Among bidependent Variables, 1991 Logistic Regression Data Set (2 of 2) 531 196. Correlation Matrix, Levels of Correlation Among Independent Variables, 1986 Logistic Regression Data Set (I of 2) 532 197. Correlation Matrix, Levels of Correlation Among Independent Variables, 1986 Logistic Regression Data Set (2 of 2) 533 XIX Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. “ The One Thing N eedful - Now, w hat I w ant is. Facts. Teach these boys arui g irls n othing but Facts. F acts alone are w anted in life. P lant nothing else, a nd root out everything else. You can only fo rm the m inds o f reasonable anim als upon F acts: nothing else w ill be o f any service to them . S tick to fa c ts. Sir. (C harles D ickens, H ard Times, 1854) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 1 INTRODUCTION Through an empirical and historical method, this study will test two basic underlying hypotheses: 1 ) The nature and timing of infant mortality decline - as well as of the demographic transition itself - in the Northeast is different than the pattern exhibited in the rest of Brazil. 2) The recent acceleration in the decline of infant mortality in the Northeast region would be the result of the association of both socio economic and public health changes. Based on the literature and on an exam ination of the specific characteristics of the region, this research hypothesizes that the most relevant factors in explaining infant survival gains are: a) improvements in mother’ s education; b) household wealth and income (through goods, a proxy for household income); c) prenatal care by a doctor; d) better sewage disposal; e) increasing breastfeeding practices. Other relevant factors that wül be included in the main model of infant mortality determination as explanatory variables are: place of residence (urban X rural ), access to clean drinking water, household crowding, age of the mother, ethnicity, birth order risk, sex of the child and immunization patterns. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. With respect to the first hypothesis, this study will argpe that the same way there is no unique Latin American mortali^ decline path, there is not a specific Brazilian infant mortality path, i For historical and political reasons that wül be discussed in this study, the Northeast region has lagged behind the rest of the country institutionally, socially and economically. The rate of female illiteracy in the region is almost 300% higher than in other four regions of Brazil: 27 against an average of 10 for all the other regions. ^ Per capita income is also about a third of the average of the most developed regions, such as the South and the Southeast. ^ Income, wealth and regional inequality^ are pervasive in Brazil, so it is no surprise that indexes of human development such as infant mortality, life expectancy and fertility are also rather contrasting among its regions. As far as infant mortality is concerned. Northeastern levels would be closer to Sub- Saharan levels than to the rest of Brazil. This research contends that both the timing and nature of demographic transition in the Northeast region are particularly distinct than the rest of ^ Alberto Pallcoi, Health Levels and Care m Lada America; The Case of Infimt Mortality 1900- 1985”, Health Transition, Charter 10, Vol.1. (Camberra: The National Library of Australia, 1989) ^ EBGE values for 1996 values in percentage terms. ^ hi the latter regions per capita mcome is closer to the levels o f odier relatively more developed South American countries, sudi as Argentina and Chile, whereas Nortiieastem income and wealth levels reswible poorer countries sudi as Peru and Paraguay. 3 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Brazil. Improvements in life expectancy, infant mortality and fertility started much, later in the Northeast. The onset of mortality revolution in this region did not occur in the early 1950’ s but rather in the mid 1960’ s. In addition, the rate of decline in infant mortality in the most industrialized and urban regions of Brazil would be higher than in the Northeast until the mid 1980 s when this trend was reversed. The onset of the fertility transition happened in the 1980’ s as well. Propelled in great part by the decline in infant mortality in the previous decades, the phenomenal decline in fertility has been more homogeneous across regions and it is happening at a much faster rate than the decline in infant mortality. From 1940 to 1970 Brazil’ s TFR was around 6.0. In a period of one generation it was reduced to 2.3 (1996). In the Northeast TFR rate fell firom 4.9 in 1985 to 2.9 in 1996. The histitute National D’ Etudes Démographiques calls such a relative homogeneous decline in fertility is a country as diverse as Brazil as surprising. 5 This study does not find the remarkable decline in fertility levels in the Northeast surprising given the widespread usage of modem contraception as ^ For instance, the level o f life expectancy attained m the South region of Brazil in 1930 (50.1) would only be adneved in the Northeast in 1970. (51.6). ^ " Brésil; La Transition Donogr^hique fbq)ide D un Pays Heterogene", Population e Sociétés, April 1999. 4 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. well as the changes in women status. However, of particular interest to this study is the impact of this vigorous fertility decline on the concurrent recent decline in infant mortality in the Northeast. Simoes maintains that despite the recent decline in infant mortality, the relative differences between the Northeast and other regions such as the Southeast are the same as the ones verified before the beginning of the mortality revolution in the 1940's. ^ This correct statement lacks dynamics and obscures the fact that the differential was much higher in previous decades and that it is decreasing by virtue of the relative acceleration in infant mortality decline in the Northeast as of the mid 1980’ s and particularly in the 1990 s. This research is primarily concerned with this period, spanning firom the mid 1980’ s to the mid 1990’ s. Recent studies in Brazil argue that, as of the 1980’ s, social intervention and pubhc policy have been playing a decisive role in explaining infant mortality decline in the Northeast. 7 This study will investigate this claim by evaluating three covariates highly correlated with public programs and state intervention: pre-natal care by a physician, breastfeeding status of the infant and immunization levels (D PT 1, 2 and 3). Other demographic and family independent variables included in the study’ s main model are: sex of the index child, birth order of the infant child. ° Celso Simoes, A Moitalidade Infantil na Tiansicao da Moitalidade no Brasil: Um Estudo Conqiaiativo entre o Noideste e o Sudeste, UFMG/Cedq>lar, Belo Horizonte, 1997. ^ Celso Simoes, Ibid. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. age of the mother at the time of the birth of the child and ethnicity of the mother. The socioeconomic and household independent variables are: mother’ s educational attainment, goods (a p ro ^ for household income), household crowding, sewage disposal, source of drinking water and place of residence (urban x rural). In addition to birth cohort variables, a dichotomous dependent variable based on the surviving status of the mother’ s latest child was devised. The data were collected from the three phases of the Demographic and Health Surveys (D H S) in Brazil; the 1986 PNSMIPF (National Research on Family Planning and Health), the 1991 PSFNe (Research on Family Health in the Northeast) and the 1996 PNDS (National Survey on Demography and Health). The Northeast region represents 30% of the households surveyed in 1986, 100% in 1991 and 36% in 1996 (1,791, 6,222 and 4,771 cases, respectively ). Women are the units of observation, whereas the unit of analysis is the infant child. The retrospective stratified sample surveys interviewed women aged 15 to 49. Individual and household data sets were merged, cases selected and variables recoded originating 6 data files: the 1996, 1991 and 1986 and the pooled 1996-91-86, 1996-91 and 1991-86 datafiles. Since two distinct methods of analysis were used, the six data files were transformed into twelve, six for each particular method. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The methods applied are logistic regression and survival analysis (proportional hazards Cox regression). The conceptual Gramework borrows firom the most part finm that of Mosley and Chen, in which infant mortality is seen as the final outcome of a continual interaction between the infant and its environment. This research hypothesizes that income and educational levels of the mother, as well as sewage conditions, breastfeeding status and prenatal care are together the most powerful determinants of infant survival. Child mortality is often cited as being a more sensitive social indicator than infant mortality, but it is deeply affected by measurement errors. » In addition, infant mortality accounts for a much larger firaction of mortality changes. Thus, even though it is not an ideal indicator of health conditions, infant mortality is the most appropriate one. Furthermore, and unlike other developing areas such as the Sub-Saharan where child and under 5 mortality are comparatively of greater concern to policy-makers, infant mortality seems to be much more troublesome in Brazil (according to the published data computed by the 1996 DHS, Ne’ s IqO was 74 and 4 q l was 16; Brazil’ s IqO and 4ql were respectively 48 and 9). In order to examine the nature and causes of infant mortality one needs to observe not only economic derived indicators such as GDP per capita, wealûi levels and nutrition intake, but also the general social welfare of the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. population., its living conditions, health services and medical care, educational levels, stocks of food, clean water and sanitation and so forth. Brazil’ s infant mortality is in average twice as much as the rates of countries with similar GDP per capita such as Chile and Venezuela. Other Latin American countries with a substantially lower GDP per capita than Brazil, such as Costa Rica and Cuba, have superior chances of child survival. By all standards, infant mortality is high in Brazil: the country ranks 110^ among 191 countries (U N , 1998). Male and female infant mortality rates per 1,000 live births in Brazil are, respectively, 48 and 36.4. Brazil’ s ofScial infant mortality rate per 1,000 live births is 37.5. The gender differential for life expectancy at birth is also substantial: 63.9 and 71.4. The combined life expectan<gr for a Brazilian is 67.6 years . ® Such a high mortality level cannot obscure two facts: a) infant survival chances are contrasting in Brazil: industrial, urban and more developed areas of areas of the Southeast and South typically have much lower mortality rates than areas such as the backlands {Sertao) of the economically lagging Northeast region; b) bifant mortality rates have been falling veiy greatly in the last two decades. * Alberto Pallooi, “ HeaUi Levels and Care in Latin America: The Case o f Tnfant Mortality 1900- 1985”, “Health Transition - The CulturaL Social and Bdiavioural Determmants o f HeaMi”, Chapter 10, Volume I. (Canberra: National Library of Australia, 1989). ’ Most recent values (1996), accordmg to die official data by the Brazilian Institute o f Geognqihy and Statistics QBGE). 8 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Brazil, a middle-mcome country with a per capita GDP of approximately $ 4,800, has seen its quality of life, as measured by UN’ s Human Development Index 1 0 , bettering significantly in recent years, (tables 2.1.14 and 2.1.15) Life expectancy, mortality levels and particularly schooling levels have been improving a great deal in recent decades, (tables 2.1.10 to 2.1.13) GDP per capita growth rates were limited, however, by tiie absence of economic growth in the 1980’ s. As a result, income inequality and poverty increased acutely. Inequality levels in Brazil are inextricably related to institutional, historical and economic factors. Income inequality is also profoundly associated with regional inequality in Brazil. According to the latest data by the Brazilian Institute of Geography and Statistics, the poorest 50% Brazilians hold only 12% of the total income, whereas the richest 20% detain 65% of it. n The HDI is used to measure quality o f life in the 174 country members. It ranges fiom 0 to 1, according to general development in mcome, life expectancy and education. The low HDI group (latest data, 1998) c<m q>rises 44 countries wife a HDI lower than 0.5. 31 offeese countries are in Afiica. The mtermediate level includes 66 countries, wMi a HDI ranging finm 0.5 to 0.8. The h i^ HDI group is fermed by 64 countries. On top o f fee list is Canada, followed by France, Norway and the US. " Income inequality was worsened m fee last three decades. In 1960 these figures were 18% and 54% for fee poorest 50% and fee ridiest 20%, respectively. As fee 1996 U.N. Human Development Rq>ort for Brazil stresses, inequa%r of mcmne is rooted in fee h ^ e r stratum of fee distribution, among fee rife, feat is, if fee mcome of fee richest 10% were not considered, inequality between fee medium and fee low mcome stratums would fee same as Mexico or fee US. Brazil is a very unequal society m fee sense feat fee 10% rifeest are relatively very well ofi^ draining 30 times more mcmne than fee poorest 40%, against 5 times in Germany, 13 times in Costa Rica and 25 times in Peru, for example. 9 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Almost half of Brazil’ s poor (45%) are firom the Northeast region. ^ These 19 m illion people comprise 46% of the total population in the region, Notwithstanding the very unequal character of its society, poverty and the decreasing but still very high levels of infant mortality, in 1998 Brazil entered UN’ s high human development group. According to the UN, Brazil’ s HDI improved firom 0.39 (1960), to 0.5 (1970), to 0.67 (1980), to 0.78 (1991) reaching a level of0.809 in 1995. (tables 2.1.9 and 2.1.16) Infant mortality has been falling rapidly, decreasing 50 % between 1980 and 1996. This study sets out to investigate the underlying causes of infant mortality decline in Brazil’ s most impoverished region, the Northeast, between 1986 and 1996 through an empirical and historical work. The Northeast is Brazil’ s most backward region. Infant mortality in the Northeast is 60.4, almost 300% the South region’ s level of 22. Half of all infant deaths in Brazil occur in the Northeast. Nevertheless, in the past two decades, infant mortality has been falling consistently in the region, imR is falling throughout B ra ^ but as of die late 1980’ s it has been declining more steadfastiy in the Northeast. Evm th o u ^ there is controversy about the total number o f poor Brazilians, the UN estimates that 30 % o f the total population m 1990 or 42 m illion people lived below the poverty line. ^ hr contrast, only 20% o f the total population in the South Region are poor and 23% in the Southeast. According to the ofiScial figures for 1996 released by the IBGE. Infont mortality rates coitqmted by the published report of the 1996 DHS are shortly h i^en 74 and 25. Northeastern Brazil is a lagging region wifom a developing country. Itf one takes the human development mdex as an indicator o f general welfore, the states witfi the worst performance m Brazil are all in foe Northeast It is a r%ion stricken by economic backwardness, droughts and widespread poverty. 10 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Although a higher per capita income and nutrition intake tend to promote significant gains in infant mortalily, advances in public health and medical care as well in the educational attainm en t of the mofiier also seem to play a very strong role in promotmg infant mortality decline. Studies of infant mortality in Brazil have suggested that the causal relationships are reinforcing rather than separable and distinct. Some authors argue that the most significant gains obtained in moving a country firom high to moderate infant mortality rates involve improvements in sanitation, water supply, medical care and especially in education, Similarly, recent studies in Brazil underscore the importance of breastfeeding patterns, mothers’ education, provision of health services such as immunization and prenatal care as well as of improvements in the sources of drinking water and sewage disposal. Kuznets admonishes that the differentiation between economic development and public health creates a false dichotomy, While the average number of schoolmg years for Brazilian men increased 112% between 1960 and 1990, this same measure improved 158% for wmnen. Kathlem Newland, Woridvrâtch Paper 47, hifont Mortality and the Health of Societies, December 1981. "See: Celso Simoes, “A Saude Inftntil no Brasil nos Anos 90”, hifoncia Brasdeira nos Anos 90, UNICEF/IBGE (Brasilia, 1996). L. Ortiz and C. Simoes, A Mbrtalidade hrfimtil no Brasil nos Anos 80”, IBGE/DPE, Vol. 1, No.7 (Rio de Janeiro, 1988). Simon Kuznets, Population Trends and Modem Economic Growth: Notes Towards a Historical Perspective”, Population Debate: Dimensions and Perspectives, ^ e w York: Worid Population Council, 1975). 11 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Infant mortality seems to be closely connected to economic and social conditions as well as to the myriad of complmc interactions between them. Based on causal assumptions generated by the literature, on history and on the facts revealed by survey data, the purpose of this empirical and theoretical study is to investigate and ascertain the relative impact and the statistic significance of various socio-economic and demographic independent in explaining the nature of infant mortality decline in the Brazil’ s Northeast. Chapter Two is the Background chapter. It will depict the physical background as well as the social and economic characteristics of Brazil and its Northeast region in a historical perspective. It will also discuss the nature, tim ing and characteristics of modem economic growth in Brazil. It is divided in two sections: The Physical Environment; The E^conom ic and Social-Political Background in Historical Perspective. Chapter Three will investigate The Nature and Facts of Mortality Revolution and Infant Mortality Decline in Brazil and in its Northeast Region. This chapter is comprised of two sections: a) Nature of Mortality Revolution and Demographic Transition b) Facts on Mortality Revolution and Tnfant Mortality Decline in Brazil and in its Northeast Region. The section Nature of Mortality Revolution and Demographic Transition is comprised of four parts: hitroduction; Endogenous Socioeconomic Changes 1 2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. vs. Exogenous Healtti-Medical Innovation Changes? Notes on the Nature of Brazil’ s Mortality' Decline; Notes on the Timing and Interconnectedness of Brazil’ s Fast, Dramatic and Dynamic Fertility and Infant Mortality Decline. The section Facts on Mortality Revolution and Infant Mortality Decline in Brazil and its Northeast Region is divided in six parts as well: Brazil’ s Life Expectancy at Birth; Northeast’ s Life Expectancy at Birth; Population Growth, Birth Rates and Total Fertility Rate; A Comparative Perspective of Infant Mortality Decline in Brazil and in its Northeast Region; The Impact of Public Policy and Health on Infant Mortality Decline. Chapter Four will introduce the Conceptual Framework, discuss issues and review the Literature. Chapter Five will describe and explain the Conceptualization of the Variables or determ inants adopted in this study. Chapter Six will focus on Data Collection and Methods. This chapter is constituted of six sections: Data Sources; Sampling Design; Data Manipulation; The Analytical Framework; the Logistic Regression and Proportional Risks (Cox) Methods; The SPSS Statistical Output. Chapter Seven will present the Results of the Descriptive and Multivariate Regression Analyses. It will be comprised of six sections, one for each set of data sets: 1996; 1991; 1986; 1996-91-86; 1996-91 and 1991-86. Each section will be divided into five parts: Descriptive Results, Diagnostic 13 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Results for Logistic Régression. Models, Logistic Regression Results, Diagnostic Results for Cox Regression Models and Cox Regression Results. Chapter Eight will conclude and summarize this study. 14 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. “ The fu tu re, then, to which the epoch o f modern econom ic grow th is leading in one o f never ending econom ic growth, a w orld in which ever grow ing abundance is m atched by ever risin g aspirations, a w orld in w hich cultural d ifféren ces are leveled in the constant race to achieve the g ood life o f m aterial plenty. I t is a w orld fo u n d on b e lie f in science cmd the pow er o f rational inquiry a n d in the ultim ate capacity o f hum anity to shape its own destiny. ” (Richcard E asterlin, M odem E conom ic G row th- The Tw enty-first C entury in H isto rica l P erspective, 1996) 15 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 2 BACKGROUND The Physical Environm ent With a total area of 8,547,403.5 sq.km or 3.286.500 sq miles, the Federative Republic of Brazil is world’ s fifth largest country, after Russia, Canada, China and the United States (if Alaska is included). Brazil is also the world’ s fifth most populous country, after China, India, the US and Indonesia. Nevertheless, the country’ s 163.378.860 citizens 2 0 only represent 2.73% of Earth’ s almost 6 billion inhabitants. The country is constituted of 26 states and a Federal District (Brasilia), organized in 5 very distinct regions: South, Southeast, North, Northeast and Central-West. In terms of geographic, economic and population magnitude, Brazil is as great as the rest of South America together. Encompassing nearly half of South America (47%), the country sprawls across the continent’ s central eastern area. It stretches firom the Plata basin in the south to the Guyana highland in the north, firom the Andes in the west to the Atlantic ocean in the east. Brazil is situated between the 05°16'20" parallel of latitude north and the 33"44'32 " parallel of latitude south, and between the 34“ 47'30" and 73°5932 ' meridians west of Greenwich. To the north, west and south Brazil borders all South American countries, except for Chile and Ecuador. 16 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The national language is Portuguese and the national currency is the Real (R$). 2 1 The main Religion is Roman Catholic (85%). Many authors like to characterize Brazil subcontinent as an island surrounded by the Plata and the Amazon (world’ s largest river in volume) basins as well as by 4,600 miles of Atlantic coastline. As Bums notices, the Atlantic serves as Brazil’ s highway to the world, carrying capital, goods, immigrants and ideas. 2 2 Just like their Portuguese ancestors, most Brazilians prefer to remain close to the sea, mushrooming Brazil’ s coastal cities.2 3 Almost 80% of Brazil’ s population live in urban areas, the vast majority of them on the coastal strip. The main cities are: Sao Paulo (11 million in the state of Sao Paulo), Rio de Janeiro (6 million in the state of Rio de Janeiro), Belo Horizonte (2.3 million in the state of Minas Gerais), Salvador (2 million in the state of Bahia), Brasilia (1.8 million, the capital), Fortaleza (1.8 million in the state of Ceara), Recife (1.4 million in the state of Pernambuco), Porto Alegre (1.4 million in the state of Rio Grande do Sul) and Curitiba ( 1.4 million in the state of Parana). For the most part, Brazil’ s territory lies in the tropics. The country is world’ s largest tropical country. None the less, the country’ s geographic configuration, bordered by the Atlantic ocean fi*o m north to south, its continental size and relief features conditioned a great climatic diversity, with IBGE’s Population Clock, March 26,1999,5PM, Brasilia standard time. The exdiange rate has fluctuated a great deal with the recent devaluation o f the real. As of Mardi 26, the exdiange rate is US$ 1,87. “ E. Bradford Bums, “A History of Brazil”. (New York: Columbia University Press, 1970). 17 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. sharp regional differences. The highest annual average temperatures (26° to 28°C, or 78° to 82° F) occur in the Northeast region's interior and the mid and lower Amazon river. The lowest values (under 18°C, or 64° F ) occur in the hilly areas of the Southeast and in most of the South region. The highest absolute values (over 40°C or 104 °F ) are recorded in the Northeast region's low interior lands and in the Southeast region's depressions, vall^s and lowlands. The lowest absolute temperatures (around and a little under 0° C or 32°F) occur on the highest peaks of the Southeast and in most of the South region. In terms of precipitation, the most rainy areas are located in the North region (Western Amazonas and Para's coastal lands) and in parts of the Southeast Region (Serra do Mar in Sao Paulo). Annual pluviométrie values can reach an average of over 3,500 mm. The less rainy areas are situated in the Northeast region, where annual values are under 500 mm. The Northeast, and particularly its Sertao (interior), is drought-stricken, hot and very poor. Due to the vast surface of the Brazilian territory, associated to its tropicality, the country is endowed with an extraordinary florist variety. One of such many eco^stems is the Amazon region and its Amazon Tropical Forest. The Amazon, home of one of the largest wildlife reservations on earth, encompasses an area of 5,217,423 sq.km (over 2 million sq.miles ), including the states of Acre, Rondônia, Amazonas, Para, Roraima, Amapa and Tocantins in the North region, Mato Grosso in the Central-West region and Maranhâo in the Northeast. ^ U n k n o w n 1 7 * century historian: “ Th^ clm g crablike to the beadies”. jg Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The country is well endowed with natural resources. Brazil has a vast and dense hydrographic network with a great number of highlands rivers which possess excellent potential for hydroelectric power production. Production of petroleum is growing and there are abundant depomts of gold, iron ore, manganese, cassiterite, bauxite, copper and other minerals. Brazil is a constitutional Federative Republic comprised of 9,274 districts distributed in 4,974 municipalities, divided into 26 states and a Federal District. Brazil is a Federal country in which states and municipalities hold considerable political and economic power. The Federal District - where the Nation’ s capital, Brasilia, is located- is the seat of government, housing the Executive, Judiciary and Legislative powers. There are 81 senators (3 for each state and Federal District) and 513 Federal deputies. Fifteen political parties are represented in the Congress. The main political parties are: PSDB - Brazilian Social Democratic Party (center- left); PFL - Liberal Front Party (center-right); PMDB - Brazilian Social Democratic Party (center), FT - Workers Party (left) and PDT - Democratic Labor Party (left); PPB- Brazilian Progressive Party (right). The President is Fernando Henrique Cardoso, PSDB but elected through a heterodox party alliance, elected for a second 4 year term in November of 1998. 19 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The Federal government has jurisdiction over national security and defense, foreign affairs and the setting of economic policy (monetary and fiscal) and electoral law. Recently, the Federal government monopoly in the energy and telecommunications industries began to be dismantled with an aggressive privatization process. The two other Federative levels, states and municipalities, have their own three levels of government. For economic, social, cultural and administrative purposes, Brazil is divided into five major regions: North, Central-West, Northeast, South and Southeast 2 4 Enveloping 40% of the Brazilian territory, the North region is Brazil’ s largest. It covers most of the Amazon basin and rain forest and it occupies 45.2 percent of the national territory. 7.2 percent of Brazil’ s population live in this region. The least populated of all five great regions includes 7 states: Roraima, Rondônia, Amapâ, Acre, Amazonas, Para and Tocantins. Covering 18.percent of the territory, the Central West region is home of 6.6 percent of the Brazilian population. The Central West is formed by the 3 states of Mato Grosso do Sul, Mato Grosso and Goiâs, in addition to the Federal District, Brasilia. Occupying 6.7 percent of the territory, the South region is Brazil’ s smallest. It concentrates 15 percent of the population and it is also the most ^ A regional breakdown occurred in 1960. Up until then Sao Paulo was considered to be part of the South and Bahia was part of the Soudieast region. 20 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. culturally and socially homogeneous. The South region is comprised by the states of Paran a, Santa Catarina and Rio Grande do Sul. 43 percent of the Brazilian population live in the economically relatively wealthy Southeast region. Brazil’ s urban-industrial heartland covers 10.8 percent of the national territory. The Southeast is composed of four states: Espirito Santo, Minas Gerais, Rio de Janeiro and Sao Paulo. The three last ones are the three richest states in Brazil. In addition to being Brazil’ s industrial heartland, the state of Sao Paulo- with its $280Bi GDP and 35 million people- is Brazil’ s main agricultural and financial powerhouse. The Northeast region occupies 18.2 percent of the national territory and is formed by 9 states: Maranhâo, Haul, Ceara, Rio Grande do Norte, Paraiba, Pernambuco, Alagoas, Sergipe and Bahia. Around 30 percent of the Brazilian population lives in this region, most of them along the coastal capital cities. 2 1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 1 Brazil, Total Area, Resident Population, Population Density, Mean geometric Rate of Annual Increase and Percent Distribution, by Major Regions and States (Federative Units) ==== Maj or Regions Total area Resident population pop.density mean geo.rate of ann.inc % States (sq.km) 1980 1991 1991 1980/91 Pop.perc.1991 BRAZIL 100.00 (1)(2)8 547 403 119 002 706 146 825 475 17.18 1.93 NORTHEAST 28.94 (1)1 561 177 34 812 356 42 497 540 27.22 1.83 Maranhâo 3.36 333 365 3 996 404 4 930 253 14.79 1.93 Piaui 1.76 252 378 2 139 021 2 582 137 10.23 1.73 Ceara 4.34 146 348 5 288 353 6 366 647 43.50 1.70 Rio Grande 1.65 do Norte 53 306 1 898 172 2 415 567 45.31 2.22 Paraiba 2.18 56 584 2 770 176 3 201 114 56.57 1.32 Pernambuco 4.85 98 937 6 143 272 7 127 855 72.04 1.36 Alagoas 1.71 27 933 1 982 591 2 514 100 90.00 2.18 Sergipe 1.02 22 050 1 140 121 1 491 876 67.66 2.47 Bahia 8.08 567 295 9 454 346 11 867 991 20.92 2.09 NORTH 3 869 637 6 619 152 10 030 556 2.59 3.85 6.83 Rondônia 238 512 491 069 1 132 692 4.75 7.89 0.77 Acre 153 149 301 303 417 718 2.73 3.01 0.28 Amazonas 1 577 820 1 430 089 2 103 243 1.33 3.57 1.43 Roraima 225 116 79 159 217 583 0.97 9.63 0.15 Para 1 253 164 3 403 391 4 950 060 3.95 3.46 3.37 Amapâ 143 453 175 257 289 397 2.02 4.67 0.20 Tocantins 278 420 738 884 919 863 3.30 2.01 0.63 2 2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 1 (cont.) Brazil, Total area. Resident Population, Population Density, Mean Geometric Rate of Annual Increase and Percent Distribution, by Major Regions and States (Federative Units) SOUTHEAST 927 286 51 734 125 62 740 401 67.66 1.77 42.73 Minas Gerais 588 383 13 378 553 15 743 152 26.76 1.49 10.72 Espirito Santo 46 184 2 023 340 2 600 618 56.31 2.31 1.77 Rio de Janeiro 43 909 11 291 520 12 807 706 291.68 1.15 8.72 Sao Paulo 248 808 25 040 712 31 588 925 126.96 2.13 21.51 SOUTH 577 214 19 031 162 22 129 377 38.34 1.38 15.07 Parana 199 709 7 629 392 8 448 713 42.30 0.93 5.75 Santa Catarina 95 442 3 627 933 4 541 994 47.59 2.06 3.09 Rio Grande do Sul 282 062 7 773 837 9 138 670 32.40 1.48 6.22 CENTRAL WEST 1 612 077 6 805 911 9 427 601 5.85 3.01 6.42 Mato Grosso do Sul 358 158 1 369 567 1 780 373 4.97 2.41 1.21 Mato Grosso 906 806 1 138 691 2 027 231 2.24 5.38 1.38 Goiâs 341 289 3 120 718 4 018 903 11.78 2.33 2.74 Federal District 5 822 1 176 935 1 601 094 275.00 2.84 1.09 Source: IBGE, Diretoria de Pesquisas, Population Census Note: Data from the 1980 Population Census relate to the administrative division effective in 1991 (1) Includes 2,977 sq. km corresponding to the area being disputed by the States of Piaui and Ceara. (2) Includes 18 sq. km corresponding to Trindade and Martin Vaz islands. 23 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 2 Resident Population, Urban and Rural, and Urbanization Rate, by Major Regions and States (Federative Units) - 1991 Major Regions and rate States Total Resident population Urban Rural Urbanization (percent) BRAZIL 146 825 475 110 990 990 35 834 485 75.59 NORTHEAST 42 497 540 25 776 279 16 721 261 60.65 Maranhâo 4 930 253 1 972 421 2 957 832 40.01 Piaui 2 582 137 1 367 184 1 214 953 52.95 Ceara 6 366 647 4 162 007 2 204 640 65.37 Rio Grande do Norte 2 415 567 1 669 267 746 300 69.10 Paraiba 3 201 114 2 052 066 1 149 048 64.10 Pernambuco 7 127 855 5 051 654 2 076 201 70.87 Alagoas 2 514 100 1 482 033 1 032 067 58.95 Sergipe 1 491 876 1 002 877 488 999 67.22 Bahia 11 867 991 7 016 770 4 851 221 59.12 NORTH 10 030 556 5 921 837 4 108 719 59.04 Rondônia 1 132 692 659 327 473 365 58.24 Acre 417 718 258 520 159 198 61.89 Amazonas 2 103 243 1 502 754 600 489 71.45 Roraima 217 583 140 818 76 765 64.72 Para 4 950 060 2 595 651 2 354 409 52.44 Amapâ 289 397 234 131 55 266 80.90 Tocantins 919 863 530 636 389 227 57.69 SOUTHEAST 62 740 401 55 225 983 7 514 418 88.02 Minas Gerais 15 743 152 11 786 893 3 956 259 74.87 Espirito Santo 2 600 618 1 924 588 676 030 74.01 Rio de Janeiro 12 807 706 12 199 641 608 065 95.25 Sao Paulo 31 588 925 29 314 861 2 274 064 92.80 SOUTH 22 129 377 16 403 032 5 726 345 74.12 Parana 8 448 713 6 197 953 2 250 760 73.36 Santa Catarina 4 541 994 3 208 537 1 333 457 70.64 Rio Grande do Sul 9 138 670 6 996 542 2 142 128 76.56 CENTRAL WEST 9 427 601 7 663 122 1 764 479 81.28 Mato Grosso do Sul 1 780 373 1 414 447 365 926 79.45 Mato Grosso 2 027 231 1 485 110 542 121 73.26 Goiâs 4 018 903 3 247 676 771 227 80.81 Federal District 1 601 094 1 515 889 85 205 94.68 Source: IBGE, Diretoria de Pesquisas, Population Census. 24 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 3 Resident Population, by Sex and Sex ratio, by Age Groups - 1991 Age groups (years) Resident population Sex ratio Total Males Females (percent) BRAZIL 146 825 475 72 485 475 74 340 353 97.50 0 to 4 16 521 114 8 379 650 8 1 * 464 102.93 5 to 9 17 420 159 8 836 268 8 5b3 891 102.94 10 to 14 17 047 159 8 585 508 8 461 651 101.46 15 to 19 15 017 472 7 460 490 7 556 982 98.72 20 to 24 13 564 878 6 712 435 6 852 443 97.96 25 to 29 12 638 078 6 174 959 6 463 119 95.54 30 to 34 11 063 483 5 406 785 5 656 708 95.58 35 to 39 9 463 763 4 597 824 4 865 939 94.49 40 to 44 7 834 714 3 860 918 3 973 796 97.17 45 to 49 6 124 688 2 994 785 3 129 903 95.68 50 to 54 5 165 128 2 526 581 2 638 547 95.76 55 to 59 4 242 124 2 017 494 2 224 630 90.69 60 to 64 3 636 858 1 715 601 1 921 257 89.30 65 to 69 2 776 060 1 308 343 1 467 717 89.14 70 to 74 1 889 918 872 424 1 017 484 85.74 75 to 79 1 290 218 575 738 714 480 80.58 80 and over 1 129 651 459 319 670 332 68.52 Source: IBGE, Diretoria de Pesquisas, Population Census. Notes: In 1997 the total population, is around 157,000,000. Preliminary projections of resident population for 2000 show a total of 165,715,400 inhabitants, and 200,306,300 for 2020. The aged population (persons sixty years old and over) has increased from 6.1 percent in 1980 to 7.3 percent in 1991. Aged women comprised 6 .4 percent of the female population in 1980 and 7.8 percent in 1991, while the figures for the male population were 5.8 percent in 1980 and 6.8 percent in 1991. The Country’ s most populous m unicipali^, according to the 1991 Census, is Sao Paulo, wiüi 9.6 million inhabitants, followed by Rio de Janeiro (5.4 million persons), Belo Horizonte and Salvador (2 million persons), Fortaleza (1.7 million persons), Brasilia (1.5 million persons), Curitiba (1.3 million persons), Recife, Nova Iguaçu and Porto Alegre (1.2 million persons). 25 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 4 Territorial Area, by Major Regions and States federative Units) - 1994 Major Regions and States Territorial area Absolute Relative (sq. km) Brazil (percent) Regions BRAZIL 8 547 403.5 100.00 NORTHEAST 1 561 177.8 18.27 100.00 Maranhâo 333 365.6 3.90 21.35 Piaui 252 378.5 2.95 16.16 Ceara—Piaui litigation 2 977.4 0.03 0.19 Ceara 146 348.3 1.71 9.37 Rio Grande do Norte 53 306.8 0.62 3.41 Paraiba 56.584.6 0.66 3.62 Pernambuco (1) 98.937.8 1.16 6.33 Alagoas 27 933.1 0.32 1.79 Sergipe 22 050.4 0.26 1.41 Bahia 567 295.3 6.64 36.34 NORTH 3 869 637.9 45.27 100.00 Rondônia 238 512.8 2.80 6.16 Acre 153 149.9 1.79 3.96 Amazonas I 577 820.2 18.45 40.77 Roraima 225 116.1 2.64 5.81 Para I 253 164.5 14.65 32.38 Amapâ 143 453.7 1.67 3.70 Tocantins 278 420.7 3.26 7.20 SOUTHEAST 927 286.2 10.85 100.00 Minas Gerais 588 383.6 6.89 63.45 Espirito Santo 46 184.1 0.54 4.98 Rio de Janeiro 43 909.7 0.51 4.73 Sao Paulo 248 808.8 2.91 26.83 SOUTH 577 214.0 6.76 100.00 Parana 199 709.1 2.34 34.61 Santa Catarina 95 442.9 1.12 16.53 Rio Grande do Sul 282 062.0 3.30 48.86 CENTRAL WEST I 612 077.2 18.86 100.00 Mato Grosso do Sul 358 158.7 4.19 22.22 Mato Grosso 906 806.9 10.60 56.25 Goiâs 341 289.5 3.99 21.17 Federal District 5 822.1 0.07 0.36 Trindade and Martin Vaz islands 10.4 - - Source: IBGE, Diretoria de Geociências, Departmento de Cartografia. Note; Topographic charts readings and geodetic measurements of areas. Territorial division on December 31, 1993. (1) Includes area of the State District of Fernando de Noronha (18.4 sq. km). ______ 26 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 5 Administrative Evolution, by Major Regions and States (Federative Units) - 1940/1995 Major Regions and 1995 Federative Municipalities Units A d m i n i s t r a t i v e Units In Aug. 31, Municipalities created and installed created Not installed 1940(1) 1950 1960 1970 1980 1990 Installed BRAZIL 232 1 574 1 889 2 766 3 952 3 974 4 491 4 974 NORTHEAST 174 584 609 903 1 376 1 375 1 509 1 558 Maranhâo 77 65 72 91 130 130 136 136 Piaui 35 47 49 71 114 114 118 148 Ceara 79 79 142 142 141 178 184 Rio Grande do 10 Norte 42 48 83 150 150 152 152 Paraiba 50 41 41 88 171 171 171 171 Pernambuco 85 91 103 165 165 (2) 168 (2)177 Alagoas 2 33 37 69 94 94 97 100 Sergipe 42 42 62 74 74 74 75 Bahia 150 150 194 336 336 415 415 NORTH 32 88 99 120 143 153 298 398 Rondônia (3) 8 2 2 2 7 23 40 Acre 1 7 7 7 7 12 12 22 Amazonas 28 25 44 44 44 62 62 Roraima 2 - 2 2 2 2 8 8 Para 8 Amapâ 1 53 59 60 83 83 105 128 - 4 5 5 5 9 15 Tocantins 12 79 123 27 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 5 (cont.) Administrative evolution, by Major Regions and States (Federative Units) - 1940/1995 Major Regions and Federative Municipalities Units created A d m i n i s t r a t i v e Units In Aug. 31, 1995 Municipalities created and installed Not installed 1940(1) 1950 1960 1970 1980 1990 Installed SOUTHEAST 641 845 1 085 1 410 1 410 1 432 1 533 18 Minas Gerais 288 386 483 722 722 723 756 Espirito Santo 32 33 37 53 53 67 71 Rio de Janeiro 51 57 62 64 64 70 81 Sâo Paulo 270 369 503 571 571 572 625 11 SOUTH 5 181 224 414 717 719 873 1 058 Parana 49 80 162 288 290 323 371 Santa Catarina C 44 52 102 197 197 217 260 D Rio Grande do Sul 88 92 150 232 232 333 427 CENTRAL WEST O 80 112 244 306 317 379 427 . 3 Mato Grosso do Sul - - - - 55 72 77 Mato Grosso 3 28 35 64 84 38 95 117 Goias 52 77 179 221 223 211 232 Federal District - - 1 1 1 1 1 “ Source: IBGE, Diretoria de Geociências, Departamento de Estudos Territorials. (1) Administrative Units on July 1st. (2) Includes the Federal District and Fernando de Noronha. (3) Municipalities with Administrators appointed by the Governor according to the second paragraph, article 108 of the Federal Constitution. 28 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 6 Permanent private housing units, by Major Regions - Selected characteristics 1991 Brazil North. Northeast Southeast South Central West 34 734 715 1 954 358 9 014 003 15 820 409 5 694 400 2 251 535 Type House and apartment 32 966 728 1 769 466 8 535 437 14 922 220 5 548 870 2 190 735 In agglomerate (1) I 575 766 166 007 445 581 799 158 134 884 30 136 Room 192 221 18 895 32 985 99 031 10 646 30 664 Tenure Owner-occupied 24 261 954 1 531 556 6 814 419 10 409 949 4 069 564 1 436 466 Rented 5 689 170 191 007 1 010 088 3 294 119 807 381 486 575 Lent 4 546 025 219 443 1 129 327 2 019 708 764 048 413 499 Other 237 566 12 362 60 169 96 633 53 407 14 995 Water supply Public supply system 24 562 013 873 773 4 753 637 13 415 164 4 032 234 1 487 205 Well or spring 6 549 363 797 010 1 572 809 1 975 359 1 511 553 692 632 Other 3 623 339 283 585 2 687 557 429 886 150 613 71 698 Sewage disposal Public sewer 12 256 963 26 005 800 840 10 039 479 777 255 613 384 Septic tank 5 941 799 477 832 1 359 542 1 672 051 2 262 104 170 270 Other 11 437 559 1 035 397 3 456 244 3 404 944 2 309 089 1 231 885 None 5 098 394 415 134 3 397 377 703 935 345 952 235 996 Destination of urban refuse Collected 21 739 197 Other 5 418 071 653 493 3 605 965 12 291 859 3 785 354 1 402 359 537 106 2 052 730 1 839 145 551 160 437 930 Source: IBGE, Diretoria de Pesquisas, Population Census. (1) Houses or apartments located in subnormal agglomerate (slum, shantytown, river shack, etc.). 29 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 7 Extent of the border line, by neighboring countries and the Atlantic Ocean - 1994 Neighboring countrries and Atlantic Ocean Extent of Absolute (km) border line Relative (percent) Total 23 086 100.00 Neighboring countries 15 719 67.97 Venezuela 1 495 6.47 Guyana 1 606 6.96 Suriname 593 2.57 French Guiana 655 2.84 Uruguay 1 003 4.34 Argentina 1 263 5.47 Paraguay 1 339 5-80 Bolivia 3 126 13.54 Peru 2 995 12.98 Colombia 1 644 7.12 Atlantic Ocean 7 367 31.91 Source: IBGE, Diretoria de Geociências, Departmento de Cartografia. Table 8 Population by Ethnic Group - 1996 White Black Pardo* Oriental Native Brazilian Indian Brazil 55,2 6.0 38,2 0.4 0.2 Urtian North** 28,5 3,7 67,2 0.4 0.2 Northeast 30,6 6,1 62,9 0.1 0.2 Southeast 65,4 7,4 26,5 0.6 0.1 South 85,9 3.1 10,5 0,4 0.1 Central W est 48,3 4,0 46,6 0,6 0.5 Data sources: IBGE/Pesquisa Nacional p er Amostra de Domici'lios - PNAD -1996. * "Pardo": mixed race or color (muiato, m estizo). Excludes data for the rural area of th e states of Rondônia, Acre. Amazonas, Roraima, Paré and Amapa. 30 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 9 The Evolution of Brazil's Human Development Index (HOI), 1970-1996 Region / State 1970 1980 1991 1995 1996 North 0.425 0.595 0.676 0.720 0.727 Rondônia 0.474 0.611 0.725 0.782 0.820 Acre 0.376 0.506 0.662 0.752 0.754 Amazonas 0.437 0.696 0.761 0.754 0775 Roraima 0.463 0.619 0.687 0.788 0.818 Para 0.431 0.587 0.657 0.709 0.703 Amapa 0.509 0.614 0.767 0.797 0.786 Tocantins — — 0.534 0.578 0.587 Northeast 0.299 0.463 0.557 0.596 0.608 Maranhao 0.292 0.408 0.489 0.546 0.547 Piaui 0.288 0.416 0.494 0.529 0.534 Cearà 0.275 0.477 0.537 0.576 0.590 Rio Grande do Norte 0.266 0.501 0.620 0.666 0.668 P arait» 0.259 0.442 0.504 0.548 0.557 Pernambuco 0.315 0.509 0.590 0.602 0.615 Alagoas 0.263 0.437 0.506 0.538 0.538 Sergipe 0.320 0.493 0.655 0.748 0.731 Bahia 0.338 0.533 0.593 0.632 0.655 Southeast 0.620 0.795 0.832 0.853 0.857 Minas Gerais 0.460 0.695 0.748 0.780 0.823 Espirito Santo 0.485 0.715 0.782 0.819 0.836 Rio de Janeiro 0.657 0.804 0.824 0.842 0.844 Sâo Paulo 0.710 0.811 0.848 0.867 0.868 South 0.553 0.789 0.834 0.855 0.860 Parana 0.487 0.723 0.811 0.844 0.847 Santa Catarina 0.560 0.796 0.827 0.857 0.863 Rio Grande do Sul 0.631 0.808 0.845 0.863 0.869 Central-West 0.469 0.704 0.817 0.839 0.848 Mato Grosso do Sul — 0.725 0.784 0.844 0.848 Mato Grosso 0.458 0.600 0.756 0.768 0.767 Goias 0.431 0.635 0.743 0.765 0.786 Distrito Federal 0.666 0.819 0.847 0.864 0.869 Brazil 0.494 0.734 0.787 0.814 0.830 Source: U N (Human Development Country Reports) 31 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 10 The Evolution of Brazil's Life Expectancy at Birth, 1970-1996 Region / Stale 1970 1980 1991 199S 1996 North 54.06 60.30 65.67 67.03 67.38 Rondônia 5 4 ^ 60.34 65-34 66-71 67.06 Acre 53-15 59-23 65-27 66-68 67.04 Amazonas 54-31 59-66 65-92 67.30 67.65 Roraima 52-32 58-92 64-53 65-93 66.29 Para 54-39 60-72 65-83 67.20 67.56 Amapa 54-79 61-01 66-17 67-50 67.85 Tocantins — — 65-46 66.84 67.19 N ortheast 44.36 57.67 62.71 64.10 64.46 Maranhâo 49-07 56-15 61-94 63.29 63.64 Piaui 49-41 56-71 62.66 64.06 64.42 Ceara 43-14 59-45 63.39 64-78 65.14 Rio Grande do Norte 38.63 59-41 63.42 64.82 65-18 Paraiba 38-91 56-58 61.34 62-79 63-16 Pernambuco 41-13 56.26 60.58 62-03 62-41 Alagoas 40-55 54.92 60.07 61-52 61.89 Sergipe 45-12 58-74 64.22 65.63 65.99 Bahia 48-77 58.82 64-74 66.12 66.47 S o u th east 56.89 64.26 67.71 68.59 68.82 Minas Gerais 54-35 62-74 67.66 68-94 69.27 Espirito Santo 57.92 62-87 67.74 68-91 69.22 Rio de Janeiro 57.29 63.30 66.04 66-78 66.97 Sâo Paulo 58-45 65-67 68.47 69.20 69.39 South 60.26 64.60 68.90 69.94 70.20 Parana 57-50 63-16 67-70 68.91 69.23 Santa Catarina 60-85 65-13 69.29 70.25 70.50 Rio Grande do Sul 64-52 65-80 69-75 70.62 70.84 C entral-W est 55.96 62.22 67.14 68.26 68.54 Mato Grosso do Sul — > 63.26 67.65 68.93 69.26 Mato Grosso 57.86 60-31 66.33 67.66 68.01 Goias 55.28 61.80 67-17 68.30 68-60 Distrito Federal 54-17 64.65 67-54 68.21 68.38 Brazil 52.67 61.76 66.13 67.28 67.58 Source: UN (Human Development Country Reports) 32 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 11 The Evolution of Brazil's Life Expectancy Index for HD, 1970-1996 Region / Sttrte 1970 1980 1991 1995 1996 North 0.404 0.588 0.678 0.701 0.706 Rondônia 0.487 0.589 0.672 0.695 0.701 Acre 0.469 0.571 0-671 0.695 0.701 Am azonas 0.489 0.578 0.682 0.705 0.711 Roraima 0.455 0.565 0.659 0.682 0.688 Para 0.490 0.595 0.681 0.703 0.709 Amapa 0.497 0.600 0.686 0.708 0.714 Tocantins — — 0.674 0.697 0.703 N ortheast 0.323 0.545 0.629 0.652 0.658 Maranhâo 0.401 0.519 0.616 0.638 0.644 Piaui 0.407 0.529 0.628 0.651 0.657 Cearà 0.302 0.574 0.640 0.663 0.669 Rio Grande do Norte 0.227 0.574 0.640 0.664 0.670 Paraiba 0.232 0.526 0.606 0.630 0.636 Pemambuco 0.269 0.521 0.593 0.617 0.624 Alagoas 0.259 0.499 0.585 0.609 0.615 Sergipe 0.335 0.562 0.654 0.677 0.683 Bahia 0.396 0.564 0.662 0.685 0.691 S o u th e a st 0.532 0.654 0.712 0.727 0.730 Minas Gerais 0.489 0.629 0.711 0.732 0.738 Espirito Santo 0.549 0.631 0.712 0.732 0.737 Rio de Janeiro 0.538 0.638 0.684 0.696 0.700 Sâo Paulo 0.558 0.678 0.725 0.737 0.740 S outh 0.588 0.660 0.732 0.749 0.753 Parana 0.542 0.636 0-712 0.732 0.737 Santa Catarina 0.598 0.669 0.738 0.754 0.758 Rio Grande do Sul 0.659 0.680 0.746 0.760 0.764 C entral-W est 0.516 0.620 0.702 0.721 0.726 Mato Grosso do Sul — 0.638 0.711 0.732 0.738 Mato Grosso 0.548 0.589 0.689 0.711 0.717 Goias 0.505 0.613 0.703 0.722 0.727 Distrito Federal 0.486 0.661 0.709 0.720 0.723 Brazil 0.461 0.613 0.686 0.705 0.710 Source: UN (Human Development Countiy Reports) 33 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 12 The Evolution of Brazil* s Adult Literacy Rate, 1970-1996 (%) Region ! Stale 1970 1900 1991 1995 1996 North 63.0 69.4 75.9 0.0 79.2 Rondônia 64.7 68.5 80.4 84.3 85.8 Acre 47.3 55.2 65.7 70.2 70.2 Amazonas 62.6 70.7 76.2 78.8 79.7 Roraima 66.4 74.6 78.3 84.1 85.9 Para 67.7 72.3 76.4 78.3 78.7 Amapa 66.4 75.3 80.7 85.0 85.0 Tocantins — — 69.9 75.4 78.8 N ortheast 46.1 54.1 63.5 69.5 71.3 Maranhâo 40.5 49.0 59.3 68.3 66.9 Piaui 40.4 50.4 59.5 64.9 65.6 Cearà 44.6 54.5 63.9 68.5 69.0 Rio Grande do Norte 45.6 55.6 65.1 70.5 71.6 Paraiba 45.0 50.7 59.4 67.8 68.7 Pem am buco 50.3 57.8 67.1 70.2 73.8 Alagoas 39.1 46.1 56.0 65.2 63.7 Sergipe 46.6 53.5 65.0 73.7 74.9 Bahia 49.4 56.9 65.5 71.7 75.5 S o u th e a st 77.1 83.5 88.2 90.7 91.3 Minas G erais 65.7 75.3 82.5 85.9 87.2 Espirito Santo 67.2 76.0 83.0 85.9 85.9 Rio de Janeiro 83.4 87.2 90.7 93.2 93.7 Sâo Paulo 81.2 86.3 90.2 92.3 92.6 S outh 76.6 84.2 88.7 90.9 91.1 Parana 69.0 79.7 85.7 88.4 88.3 Santa Catarina 81.1 87.2 90.8 92.6 92.7 Rio Grande do Sul 81.6 87.0 90.4 92.2 92.8 C entral-W est 67.5 76.5 83.9 86.6 88.4 Mato Grosso do Sul — 76.4 83.7 86.7 87.6 Mato Grosso 64.2 69.7 81.1 84.7 88.1 Goias 64.4 74.0 82.3 84.8 86.8 Distrito Federal 83.0 88.6 91.3 93.5 93.7 Brazil 67.0 74.7 80.6 84.4 85.3 Source: UN (Human Development Countiy Reports) 34 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 13 The Evolution of Brazil's Education Index for H D , 1970-1996 Region/ State 1970 1060 1091 1995 1996 North 0.567 0.646 0.716 0.771 0.777 Rondônia 0.578 0.626 0.746 0.795 0.807 Acre 0.436 0.530 0.635 0.715 0.709 Amazonas 0.544 0.660 0.713 0.755 0.764 Roraima 0.632 0.713 0.722 0.820 0.838 Para 0.608 0.668 0.714 0.768 0.770 Amapa 0.646 0.725 0.779 0.837 0.845 Tocantins — — 0.714 0.799 0.835 N ortheast 0.433 0.547 0.624 0.696 0.714 Maranhâo 0.385 0.489 0.612 0.697 0.687 Piaui 0.396 0.544 0.614 0.656 0.657 Ceara 0.406 0.564 0.603 0.677 0.714 Rio Grande do Norte 0.456 0.582 0.669 0.723 0.731 Paraiba 0.429 0.559 0.589 0.672 0.682 Pemambuco 0.481 0.578 0.678 0.723 0.750 Alagoas 0.387 0.465 0.565 0.646 0.638 Sergipe 0.456 0.563 0.657 0.736 0.751 Bahia 0.448 0.548 0.617 0.707 0.732 S o u th east 0.702 0.776 0.831 0.871 0.875 Minas Gerais 0.614 0.706 0.776 0.827 0.843 Espirito Santo 0.629 0.736 0.798 0.836 0.839 Rio de Janeiro 0.755 0.820 0.834 0.870 0.867 Sâo Paulo 0.732 0.794 0.859 0.895 0.895 South 0.688 0.764 0.827 0.861 0.870 Parana 0.616 0.728 0.809 0.847 0.851 Santa Catarina 0.726 0.773 0.829 0.864 0.876 Rio Grande do Sul 0.742 0.795 0.841 0.870 0.883 Centrai-V test 0.614 0.720 0.803 0.841 0.860 Mato Grosso do Sul — 0.676 0.807 0.846 0.855 Mato Grosso 0.575 0.647 0.770 0.824 0.841 Goias 0.592 0.715 0.795 0.826 0.854 Distrito Federal 0.778 0.840 0.861 0.894 0.902 Brazil 0.611 0.702 0.763 0.815 0.825 Source: UN (Human Development Country Reports) 35 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 14 The Evolution of Brazil's Per Capita GDP, 1970-1996 (US $ PPC) R egion/State 1S70 1980 1991 1995 1998 North 1302.03C 3068.0 3516.10 4490.0 470& Rondônia 2025.390 3425.799 4185.194 5562.0 6448 Acre 1302.036 2342.604 3767.250 5499.0 5741 Amazonas 1591.378 4680.025 4884.005 5209.0 5718 Roraima 1736.049 3223.672 3767.250 5594.0 6231 Para 1157.366 2783.138 3210.310 4281.0 4268 Amapa 2170.061 2886.793 4605.056 5487.0 5370 Tocantins — — 1255.750 1607.0 1575 N ortheast 868.024 2021.0 2360.043 2905.0 3085. Maranhao 578.683 1264.591 1394.745 2027.0 2158 Piaui 434.012 1036.550 1394.745 1892.0 2004 Ceara 723.354 1674.029 2093.556 2570.0 2667 Rio Grande do Norte 723.354 1964.263 3070.357 3993.0 4083 Paraiba 723.354 1399.343 1812.690 2276.0 2438 Pemambuco 1157.366 2404.797 2790.449 3064.0 3213 Alagoas 868.024 1964.263 2093.556 2387.0 2496 Sergipe 1012.695 2010.908 3628.255 5402.0 5122 Bahia 1012.695 2720.945 2790.449 3305.0 3677 S o u th east 3472.097 6981.0 6867.323 7956.0 8843 Minas Gerais 1591.378 4151.384 4185.194 5083.0 5968 Espirito Santo 1591.378 4296.501 4605.056 5771.0 6251 Rio de Janeiro 3761.439 6841.233 6696.694 7524.0 8653 Sâo Paulo 4629.463 8774.399 8372.306 9716.0 10536 South 2170.061 5235.0 5236.765 6669.0 6865. Parana 1736.049 4446.801 5023.0 6393.0 6485 Santa Catarina 2025.390 5472.986 5023.0 6269.0 6519 Rio Grande do Sul 2748.744 5970.530 5581.857 7131.0 7395 Central-W est 1591.378 4271.0 5575.147 6647.0 7073 Mato Grosso do Sul — 4747.401 4605.056 6279.0 6410 Mato Grosso 1446.707 3135.565 4466.060 5011.0 5003 Goias 1157.366 3218.489 4046.199 4871.0 5238 Distrito Federal 4050.780 7577.183 12000.560 13468.0 14854 Brazil 2314.731 4882.0 5023.0 5986.0 6491 Source; U N (Human Development Country Reports) 36 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 15 The Evolution of Brazil's GDP Index for H D , 1970*1996 R egion/SW * 1970 1900 1901 1995 1990 N orth 0.223 0.550 0.633 0.688 0.697 Rondônia 0.357 0.617 0-757 0.856 0.953 Acre 0.223 0.416 0.680 0.847 0.854 Am azonas 0.277 0.849 0.887 0.801 0.850 Roraim a 0.303 0.579 0.680 0.861 0.928 Para 0.196 0.498 0.577 0.656 0.631 Amapa 0.384 0.517 0.835 0.845 0-798 Tocantins — — 0.214 0.236 0.223 N o rth east 0.142 0.356 0.419 0.440 0.452 M aranhao 0.089 0.216 0.240 0.302 0.311 Piaui 0.062 0.174 0.240 0.281 0.288 C eara 0.116 0.292 0.370 0.387 0.388 Rio Grande do Norte 0.116 0.346 0.551 0.610 0.603 Paraiba 0.116 0.241 0.318 0.341 0.354 Pem am buco 0.196 0.427 0.499 0.465 0.471 A lagoas 0.142 0.346 0.370 0.359 0.363 Sergipe 0.169 0.354 0.654 0.831 0-760 Bahia 0.169 0.486 0.499 0.503 0.541 S o u th e a st 0.625 0.955 0.955 0.963 0.966 Minas Gerais 0.277 0.751 0.757 0-781 0.888 Espirito Santo 0.277 0.778 0.835 0-889 0.931 Rio de Janeiro 0.679 0.954 0.954 0.961 0.965 Sâo Paulo 0.840 0.961 0.960 0.968 0.970 S outh 0.384 0.942 0.942 0.956 0.957 Parana 0.303 0.806 0.913 0.954 0.954 S anta Catarina 0.357 0.946 0.913 0.952 0.954 Rio G rande do Sul 0.491 0.950 0.947 0.959 0.960 C entrai-W èst 0.277 0.773 0.947 0.956 0.959 Mato Grosso do Sul — 0.862 0.835 0.953 0.952 Mato Grosso 0.250 0.563 0.810 0.770 0.742 G oias 0.196 0.578 0.732 0.748 0-778 Distrito Federal 0.733 0.957 0.972 0.979 0.981 Brazil 0.411 0.887 0.913 0.923 0.954 Source: UN (Human Development Country Reports) 37 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 16 PC GDP Order - H D I Order, 1970-1996 R eg io n / State 1S70 tw o 1M1 IM S IMS North _ Rondônia -3.0 -3.0 -2.0 -1.0 3 Acre -2-0 .0 -1.0 -5.0 -4 Amazonas -3.0 -2.0 -3.0 -1.0 -1 Roraima -2-0 .0 .0 -1.0 -1 P ara 1.0 .0 1.0 .0 Amapa -1.0 2.0 -1.0 3.0 1 Tocantins — - 4.0 5.0 4 Northeast — — — — - Maranhao 4.0 -1.0 -2.0 .0 Piaui 4.0 1.0 -1.0 -1.0 -1 C earà -1.0 1.0 .0 -1.0 Rio Grande -2.0 1.0 -0 .0 do Norte Paraiba -4.0 1.0 -1.0 .0 Pem am buco -4.0 .0 -1.0 .0 Alagoas -4.0 -3.0 -2.0 -3.0 •3 Sergipe .0 -1.0 -1.0 -4.0 -1 Bahia 1.0 .0 .0 .0 Southeast — — — — - Minas -1.0 .0 .0 3.0 2 G erais Espirito 2.0 1.0 .0 .0 1 Santo Rio de .0 -1.0 -2.0 -4.0 -4 Janeiro Sâo Paulo .0 -1.0 1.0 1.0 -1 South — — — — - Parana 1.0 1.0 -1.0 .0 Santa 1.0 .0 1.0 3.0 1 Catarina Rio Grande .0 1.0 1.0 1.0 ■ 3 do Sul Central- — — — - West Mato Grosso — .0 1.0 .0 3 do Sul Mato Grosso 1.0 -1.0 .0 3.0 2 Goias .0 2.0 1.0 3.0 3 Distrito .0 1.0 -1.0 -1.0 -1 Federal Brazil - - — — Source: UN (Human Development Country Reports) 38 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The E conntnfc and Social-Political Background in Historical Perspective Brazil is a truly multicultural and ethnically dhrerse society. The main ethnic groups are: Portuguese, Italian, German, Japanese, A&ican and Indigenous people. Accordmg to the IBGE/Pesquisa Nacional por Amostra de Oomicdios - PN A D for 1996, 55% of the Brazilian population is white, 38.2% of mixed race or color 2 5^ g% black, 4% of Oriental (Asian) descent and 2% of Indian descent. The ethnic composition of the population varies a great deal in Brazil’s five regions. 2 6 In the prosperous and endowed with fertile agricultural land South region, German and Italian, among other European influences, are quite visible. The population is predominantly white: 85.9% of the population is white, 10.5% mixed, 4% Asian, 3.1% black and 1% Indian. In the Central-West region, almost half of the population is white and the other half of mixed descent: 48.3% of the population is white, 46.6% mixed, 6% Asian, 5 % Indian and 4% black. The region is traditionally under populated and has been growing rapidly in the last four decades after the construction of Brasilia (in 1960). The North is characterized by a lower concentration of people of white color (28.5%) and by a larger participation of people of mixed race ( 67.2%). 3.7 % of the population is black, 4 % Asian and 2 % hidian. 27 ^ Mixed race or color also known as Pardo. It includes races sudi as Muiato, Mestico, Caboclo e Mameluco. ^ One must notice that such race categories are mudi less rigid m Brazil than m the U S . ” Urban North. Data for the rural areas of foe Notfo t^ o n were not computed m foe 1996 PNAD. 39 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The urban Southeast dominates the country economically in all sectors. Per capita income is 50% higher than the national average. The population is predommantly white: 65.4 % of the population is white, 26.5% mixed, 7.4% black, 1% Indian and 6% Asian. The region has a significant concentration of individuals of Italian and Japanese descent (Sâo Paulo has the largest Japanese population outside Japan). The Northeast was at the heart Brazil’ s colonization. Ever since the X D C century, however, with the economic decline of its main agricultural e ^ o rts (sugar, cotton and others), the region has been struggling with pover^ and underdevelopment. At the turn of the millennium, the Northeast, and particularly its bacldands ( Sertao ), is one the poorest areas in the Americas. Ethnically, most of the Northeastems are of mixed race: 30.6 % of the population is white, 62.9% is of mixed race, 6.5% black, 1% Asian and 2% Indian. As Furtado notices, agrarian structures in Latin America and in Brazil specifically are not only an element of the production ^stem but also the basic feature of the entire social organization. ^ Land ownership is still very concentrated in Brazil, and particularly in its Northeast, which intensifies serious social problems such as migration to urban centers, widespread pover^ and landless rural movements. ^ Celso Fuitado, “Eccnonic Development m Latin America: A Surv^ fixxn Colonial Times to the Cuban Revolution.”. (Cambridge: Cambridge Universfty Press, 1970). 40 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. From the discovery of the country in 1500 th ro n g colonial times (1500- 1822), Empire (1822-1889) and early Republican era (1889-1930) 2», Brazil’ s economic history was characterized by cycles dominated, by a few export commodities. Only after W.W.II, the Brazilian economy was transformed firom a typical agrarian economy into a diversified, mdustrialized and modem one. Most of the colonial period was dominated by the production of sugar. This plantation economy, based on latifundium and African slave labor, flourished in the humid coast of the Northeast region. The first commodity to be ei^loited by the Portuguese was, however. Brazilwood, after which the country was named. According to Sûnonsen sugar exports represented 95% of the total value of exports in 1650, 47% in 1750 and 31% in 1800. In 1970 its share of Brazil’ s exports was 4.9%, and in 1999 (Jan.) the participation of sugar in the value of Brazil’ s exports has decreased to 2.5%. In addition to the sugar cycle, the Northeast also mcperienced export booms of smaller magnitude with cotton and tobacco. As of the 17^ century stock breedmg also became an important economic activity in the Northeast. ^ Period also known as Republica Velha or Old Republic. Roberto Stmonsen, Ifistoria Eccncxnica do Brasil- 1500-1820 (Sao Paulo: Companhia Editera Nacional, 1969). Trade figures obtained frcxn the official “Balança Ccxnercial Brasdeira”, SECEX, MDIC, March 1999 (statistics for January). 41 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The decadence of the Northeastern s u ^ r economy became inevitable in the 18^ century with the emergence of competing plantation economies by the French, Dutch and English. ^ E )ven though in 1999 the sugar economy is still a dominant part of coastal Northeast's economy, its role as a dynamic national force dominating Brazil’ s political, social and economic life had been seriously eroded by mid 1 9 * * > century. Wood and Carvalho ^ observe that by virtue of the capital and labor intensive character of the sugar economy as well as of the indivisibilily of mill and land, sugar plantations were usually not fragmented through friheritance. As a result, land concentration in the region is extremely high and the agrarian structure is controlled by the same famflies for centuries. Many authors claim that the historical roots of stagnation and underdevelopment in the Northeast could be traced to the decline of the sugar cycle during this period. Furtado argues that: The characteristics acquired by the two economic ^stem s in the Northeast - the sugar industry and stock breeding - during the slow process of decline, beginning in the latter half of the seventeenth century, were basic factors in the formation of what was to become the Brazilian economy of the twentieth century. It has already been noticed how the production units, whether in the sugar economy or in stock breeding, tended to preserve their original form throughout the phases of contraction and ^ Bums ( Ibid ) posits that the increased oxnpeCitica frnn European colonies in die Caribbean caused Brazil’s income from sugar to decline by two thirds between 1650 and 1715. ^ Charles Wood & Jose Carvalho, The Danogr^hy of Inequality^ m BrazQ”. (Cambridge: Cambridge Univers^ Press, 1988). 42 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. expansion. Development in both groups of economic activity occurred merely in extension* through the accretion of land and manpower* without entailing structural changes affecting production costs and hence productivity. Another fundamental cause of the collapse of the sugar economy was the discovery of diamonds and gold in the state of Minas Gerais in Brazil’ s Southeast region, The discovery of gold in late 1 8 * * ^ century promoted great commercial* cultural social, urban and population ^ development in the states of NGnas Gerais and Rio de Janeiro and caused the shift of Brazil’ s economic gravity center from the Northeast to the Southeast. The cry of gold and the subsequent mining boom did not change the m o d u s operandi of the Portuguese pattern of colonial development since Portugal did not foster manufacturmg growth in Brazil. ^ ^ Celso Fuitado* Foimacao Economica do Brasil”. (Rio de Janeiro; Editora Fundo c fe Cultura* 1959). Translated fiom Portuguese as “The Economic Growth o f Brazil”: A Survey fiom Colonial to Modem Times.” (Wesport: Greenwood Press, 1963), pg. 68. Gold was first reported m 1695 and diamonds ia 1729. Mmas Gerais is Portuguese for General Mines. ^ The emergence o f die Brazilian Baroque culture and the flowermg o f rococo art m cities sudi as Ouro Preto* Mariana, Sao Joao del Rey and Sahara (all m the state o f Mmas Gerais ) was one o f the main cultural results o f die gold cycle. Furtado observes that between eariy 1 8 * ^ century and early 19* century the Brazilian population increased ten fiild, fiom 300,000 to around 3 million. See: Celso Furtack), Foimacao Economica do Brasil”, ^ o de Janeiro: Editora Fundo de Cultura, 1959) The impact of the gold economy on the social development o f Brazil cannot be underestimated, however. As Joao Antond (“Cultura e Opulencia do Brasil”. First published in Lisbon, 1711. Sao Paulo, 1967), a 18* century dironicler o f die period, describes: “ Every year great number o f Portuguese and fiireigners come m die fleets bound for die mines. From foe cities* villages, mlets, and hinterland of Brazil there go vfoites, browns* blacks, and many hufians who are m foe service of foe Paulistas. The mncture is of every kmd and condMon of person: men and wrxnen, young and old, rich and poor* nobles and commoners* laymen, priests* and monks o f all orders* many o f whom have neifoer convent nor chapter in Brazil.” 43 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. England reaped the benefits of the gold boom, accumulated the gold and financed its industrial growth. The Portuguese Crown was heavily m debt with that country which had the monopoly of selling manufactures in Portuguese territories. ^ Gold production increased yearly until 1760, when output reached a maximum after a rather intensive mineral production. 80% of the 2 million pounds of gold produced m the world in the I S * * » century came firom Brazilian mines. Mining was a strong impetus to urbanization and to the development of a bourgeois class, but after 1760 the gold cycle had faded completely and Brazil began to stagnate economically. The stagnation of Brazil's economy would last for the next 75 years until the mid-nineteenth century Furtado notes that the historical structural roots of economic stagnation in the Northeast could be traced to the 1 7 " ”» and 1 8 ^ * century decadence of the sugar economy and supporting hegemonic political ^stem . There is every indication that durmg the long period between the last quarter of the seventeenth and the beginning of the nineteenth century, the economy of the Northeast underwent a slow process of atrophy, in the sense that the real per capita income of the population declined steadily with the passing of time. Nevertheless, it is of interest to note that such atrophy was in itself the formative process of what, in the nineteenth century, was to become the Brazilian Northeastern economic ^stem , characteristics of which prevail today. The stagnation of the sugar industry did not create a need for emigration, as it did m the West Indies. Since no adequate employment was available in the su^u: region for the entire increased fi^e population, part of the workers were attracted to the mterior by the shifting stock-breeding 39 The M ediuen Treaty^s^ped ia 1 7 0 3 w arranted this m onopoly to E n g lan d . 44 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. firontîer. Hence the more unfavorable conditions in sugar, the greater the tendency to migrate inland. Furtado goes on to say that: In this way the Brazilian Northeast became converted firom a hig^ productivity economic ^stem into an economy in which the major part of the population produced only what was necessary fiar its bare existence. Dispersion of a part of the population through extensive cattle breeding caused an involution in the division of labor and specialization, resulting in a reversion to primitive techniques, even in craftsmanship actmties. The shaping of the population of the Northeast, and its precarious subsistence economy- a basic factor in the Brazilian economic tqrstem of later periods- are therefore linked w ith the slow declme of the great sugar industry, an industry which at its best was perhaps the most profitable colonizing and agricultural business of all time. Until 1763, the central government of the colony (or the viceroyalty as the country was also known) was based in Salvador (in the state of Bahia) in the Northeast. On that year, due to the agricultural and mining development in the Southeast, the capital moved to Rio de Janeiro. Rio had an excellent port and was closer to Minas Gerais. The economic and political importance of the Northeast diminished considerably. In 1808, fleeing Napoleon, the Portuguese royal family moved the court to Brazil. In 1822, Prince Pedro I, heir to the Portuguese throne, declared the independence of Brazil, and was crowned “ C o n stitu tio n a l E m peror a n d P erpetual D e fen d e r o f B ra zil ". The United States was the fikst country to recognize Brazil's independence in 1822. ^ Celso Fuitado, Ibid, 69. Celso Fuitado, Ibid, 71. It also reflected, accoidi Spanish America in the South. Celso Fuitado, Ibid, 71. It also reflected, according to many historians, a great concern with settling the boundaries widi 45 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A s noted by many historians, the most fundamental problem confrontmg the new country was not so much its need to assert its independence as to maintain its unity. Unlike Spanish America, however, Brazil was able to remain united both culturally and politically. The production of cotton, demanding far less resources than sugar, became increasingly popular in the early 1 9 * ^ century. Cotton accounted for 20% of the value of Brazil’ s exports then. Tobacco became an important export commodity as well. To a lesser extent, other agricultural goods that were well exported at the time were cacao, indigo and rice. Export expansion was a fundamental requirement for economic development in Brazil in the first half of the 19^ century. The country lacked a strong technological and consumption market foundation to promote industrial growth. Furtado argues that the damming up of Brazilian exports was the main cause of the great relative backwardness of the Brazilian economy in the first half of the 19^ century. This author postulates the data show that Brazil’ s real per capita income declined considerably in the first half of the 19^ century and that by 1850 per capita income did not exceed that of the colonial period. 43 When the US, in war wMr Great Britam, was unable to supply die European maricet. 46 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The economic development of Latin America^ and of Brazil specifically» in the first half of the century would have been seriously limited by the early stages of the Industrial Revolution. ^ Between 1821/30 and 1841/50 the prices of Brazil's exports declined 40%. During this period the average annual growth of the sterling value of Brazil’ s exports was no more dian 0.8%» whereas population was growing at an annual rate of about 1.3 %. Nevertheless, by early 19^ century another agricultural product had already become Brazil’ s most important commodity: coffee. Based on data computed by Furtado and Simonsen Merrick & Graham ^ observe that by 1830, coffee already contributed for more than 40% of the total value of exports. Coffee was as labor intensive a cultivation as sugar but it did not require as much capital. Land was abundant and fertile. As Furtado puts it: Brazil’ s problem was in finding exportable products in whose production land was a basic factor. In fact, land was the sole production factor abundant in Brazil. There was almost no capital, and manpower was composed of a reservoir of little more than two million slaves, most of whom were mvolved either in the sugar industry or in domestic service. This thesis was also originally put foidi by Celso Fuitado. For a dioiougb evaluation of dûs thesis see Fuitado m “The Economic Development o f Latin America: A S u rv ^ fiom Colonial Times to the Cuban Revolution. (Cambric^: Cam bric^ University Press, 1970), pg. 23. Robert Simonsen, Ibid.“ ^ Thomas Merrick & Douglas Graham, “Population and Econmnic Development in Brazil: 1800 to the Present”. (Baltimore: The John Hopkins IMversity Press, 1979). Celso Furtado, “Econanic Development in Latin America: A Survey from Colonial Tunes to the Cuban Revolution.”. (Cambridge: Cambridge University Press, 1970), pg. 123. 47 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Furtado maintains that the entire expansion in e ^ o rts in the first half of the century was provided by the production of coffee. ^ The bean was introduced late in Brazil and its cycle in the Brazilian economy began modestly. Coffee entered Brazil firom the French Guiana in 1727, but only arrived and fiourished in the states of Rio de Janeiro, Minas Gerais and Sao Paulo (Southeast of Brazil} Vall^ in the late 18^ century when a hike in international prices took place. For clunatic and topographic reasons and for its terra roxu ^ soil, the production of coffee concentrated in the Southeast region, delegating an enormous political and economic ascendancy over the rest of Brazil for its 4 states (and more notably for Sao Paulo). The previous economic cycles and commodities had relied a great deal upon African slaves. The new dynamic coffee cycle was also deeply dependent on slave labor. Brazil imported about a half a million to a million more slaves than did all of Spanish America^ In 1830 Brazil was world's largest slave economy, with more slaves than free men. s o ^ Celso Fuitado, Ibid., 116. This was happening in spite o f the feet that bodi sugar and cotton Rq)oit economies (and die quantum expoitecf) were mqieriencing relatively gcxxi growth rates. Fuitado calculated die growdi o f die sterlmg value o f sugar exports between 1820 and 1850 as 24%, or I.l% aa. Since prices were exporters had to double die cpiantum exported to allow for such a giowdi. This audior also argues that die vahie o f cotton exports was reduced in 50% during the pericxL The diaracteristic "purple soil” o f die Soudieast. A deep, porous, and widi humus soil, very appropriate for coffee cultivation. ^ Thomas Skidmore, " Uma HSstoria do Brasil”. (Sao Paulo: Paz e Terra, 1999). Skidmore observes th at, unlike the US, the reasoning behmd slavery in Brazil was based on pragmatic and not cn strictly racist -or race inforioriy- arguments. 48 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Yet, population growth, and life «cpectancy of slaves in Kiazfl. was low. 5 1 With, the termination of the slave trade in 1850 and the abolition of slavery in 1888, the acute labor shortage became a serious problem. As Antond sz had put it in the 17“» century, slaves were “ th e h a n d s a n d f e e f of Brazd. The abolition of slavery in 1888 and the overthrow of the monarchy m 1889 unleashed a period of deep political, social and economic change in Brazd. This period, spanning from the birth of Republican Brazd to the 1930 revolution which took Getulio Vargas to power, is known as Old Republic. With governmental financial aid, Brazd set out to solve the labor shortage by importing labor firom Europe. Bound either to the emerging new urban centers, to the smaU properties in the South or to the coffee plantations of the Southeast, an large amount of European immigrants arrived as of the mid to late 19“» century, particular^ from Italy, Germany and Spain. From an average of 100,000 immigrants in 1851-1870, to 215,000 in 1871-1880, to 530,000 in 1881-1890, to 1,125,000 immigrants in the last decade of the 19“» century. 5 3 For a carefiil analysis of die nature and causes of the low rate of slave reproduction see F u rtad o in “E c o n < H m c G row th of B razil”, Ib id . ^ Joao A ntonil, Ibid. ^ B ased on estim ates by G iorgio M bitaia and IB G E , 1970. S ee also; G ioigio M ortara, * * M ethods of U sin g C en su s Statistks for the Calculatioa of L ife T ables and O ther Danographic M easure (w ith ^plication to die Population of Brazil ), P opulation S tu d ie s N o.7, N ew Y o ric , 1 9 4 9. 49 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A massive immigration process firom Japan also took place in the first two decades of the 20* ^ century, having as final destiny the urban and rural areas of the state of Sao Paulo. Bums notices that "as had been true throughout its history, Brazil depended - or gambled- once again on a single raw product for sale on a capricious world market." s* The boom of the coffee cycle inscribed a profound transformation in Brazil’ s society and economic tgrstem. hi the second half of the nineteen century, the country experienced high rates of economic and population growth and the modem impact granted in great part by the coffee economy. A decline in cmde death rates was evident in that period, but the contribution of natural increase to the growth of the Brazilian population growth was restricted vis- à-vis the impact of European immigration- Population growth leaped firom an annual average of 1.5% in 1840-70 to 2.4% per year in the last decade of the 19^ century, s e The modernizing impact brought about by the coffee production economy was the antithesis of the old state of affairs set forth by the sugar economy. Brazil was advancing fast and so was the world, which was witnessing the creation of a new world economic system and the establishment of an intemational division of labor. ^ Bums, Ibid, 217. Thomas Merrick and Douglas Graham, Ibid, 15. ^ A more detailed «amination o f the nature and diaracteristics of Brazil’s population growth will be laid out later in diis diapter. 50 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Economie growth and per capita income mcreased in a sustained and regular fashion in countries experiencing “ modem economic growth* s? ÿj which technological progress was pervasive, but were also present in countries using their natural resources in a more rational manner. ^ In addition and favored by to the rise in real per capita mcomes and population growth, conspicuous changes occurred in public services, urbeinization, factor proportions and redistribution of income towards manpower. Easterlin posits that modem economic growth is characterized by rapid and sustained increases in per capita GDP growth rates, accompanied by structural changes in factor proportions, allocation of productive resources, urbanization and technological innovations. ^ Kuznets postulates that one of the most relevant features of this new world economy, in addition to the rise in income growth, was the creation and rapid expansion of a fund of transmissible technological knowledge related to the forms of production. For an elaborate understanding o f the nature o f modem economic growdi see Richard Easterim, “Growth Triunq>hant: die Twenty-First Century in Historical Perspective (Ann Arbor, The University o f Michigan Press, 1996). Celso Furtado, “Econmnic Development in Latm America; A Survey fiom Colonial Times to die Cuban Revolution.”. (Cambridge: Cambridge University Press, 1970), pg. 29. ^ Ridiard Easterlin, Ibid, Chapter 2. " Simon Kuznets, “ Modem Economic Growth”. (New Haven: Yale IMversity Press, 1966), pg. 286. 51 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. As Furtado puts it: * Productive activity gave rise to further productive activity, ju st as one generation gives birth to the next.* The new world economic system was characterized by an increasing intemational specialization and by the expansion of world trade. The value of world trade rose from $3.5 billion in 1840 to over $40 billion before W.W. I. Some authors suggest that the growth rates in Brazil before W.W. II would be comparable to that of MEG. At any rate, however, the advent of modem economic growth did not take place before W.W.I. What was happening in Brazil in the mid to late 1 9 < b century was not a sustained and generalized structural change in the nature of technical inversion and economic growth, but rather a significant per capita income growth in the export sector, a trade boom. Brazil was integrated to the world economy due to the rapid expansion in the demand for tropical commodities such as cacao, sugar and particularly coffee. Brazil’ s coffee production increased from 3.7 million 60kg. bag in 1880 to 16.3 million in 1900, whereas coffee’ s share of the Brazilian total exports rose from 41.4% in 1841-1850 to 64.5% in 1891-1900. S 2 In the half century between 1840 and 1890, the average price of e ^ o rt products increased 46% and the price index for imports decreased 8%. Celso Fuitado, “Economic Development in Latin America: A Survey fixxn Colonial Times to die Cuban Revolution.”. (Cambridge: Cambridge University Press, 1970), pg. 30. ^ Thomas Merrick and Douglas Graham, Ibid. 52 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In addition to this 58% improvement in the terms of trade, the quantum exported rose by 214%, which translates in a total increase of almost 400% in the real income generated by Brazil’ s dynamic «qiort sector. Purtado observes, however, that while real income levels were increasing in the Southeast’ s coffee economy, in the Northeast sugar, cotton and subsistence economies, real per capita incomes were actually falling in the second half of the 1 9 * * » century. One hundred years ago, fueled by the coffee boom, Brazil’ s Southeast and particularly Sao Paulo, was experiencing a notable population, urban and manufacturing growth. At the same time, the Northeast region was still characterized by the same social and economic features of the colonial plantation era. Sugar was still the main economic activity and cotton and cocoa e3q>ort economies were developing at a irregular fluctuating manner. A brief boom in rubber exports also took place in the Amazon region. As noted by Merrick and Graham, in 1900 Brazil had become an export economy in the tidiest sense, The first four decades of the 20* century can be seen as a transition period to industrial development and tidl blown MEG in Brazil. Coffee profits were channeled into industrialization through the banking ^stem and other credit institutions. ^ Figures calculated by Fuitado, Ibid, 1 5 5 . ^ For a com plete analysis see Fuitado, Ibid, 1 5 7 . 65 T hœ nas M errick and D ouglas G raham , Ibid, pg. 1 5 . 53 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Not surprisingly, Sao Paulo, the main coffee producing state, emerged as the most important industrial center by 1920, surpassing Brazil’ s capital, Rio de Janeiro. The coffee wealth, abundant immigrant labor, the existence of a consumption market, a wide variety of raw materials and potential for hydroelectric power, a relatively good transportation ty^stem: Sao Paulo had most of the ingredients necessary for a successful industrialization. During W.W. I more capital became available as trade balance and value of exports grew favorably to Brazil. The value of exports increased firom 26 million to 78 million British pounds between 1914 and 1919. The Brazilian government was supporting the industrializing effort through tariffs, protective measures and exchange controls. Food processing and textile industries grew significantly during this period. After W.W.I Brazilian exports fell as the world became less dependent of Brazil’ s raw materials and primary goods. Meanwhile, the participation of coffee in Brazil’ s total e ^ o rts continued to grow until the Great Depression. In the 1920’ s coffee’ s share of the Brazilian exports was 70%. Ever since the early 20^ century, and at the expense of sectors and regions, Brazil’ s government put together artificial incentives and policies to contain the overproduction of coffee, protect earnings and to somehow manipulate world prices through price-boostmg schemes. The first of such '^coffee valorization” policies was executed in 1906. 54 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The problem, with these policies was that as long as prices were high, profits and investments would also be h i^ , wielding increasing pressure upon total supply. Then, the world crisis of 1929 showed how vulnerable the defense mechanisms were. The value of exports in 1932 had fallen to 37% of the 1927 level. Purtado argues that the 60% decline in international coffee prices was inevitable since supply factors (land and manpower) were growing faster than world’ s demand. As Merrick and Graham put it: Brazilian economic history during the period of coffee expansion - roughly 1850 to 1920- is illustratwe of both the possibilities and problems of a staple model of economic development (Watkins 1963, Lewis 1970). Opinion is mixed about the benefits Üiat accrued to Brazil from this period, though most would agree that a large portion of the economic issues of succeeding decades have roots in it: e.g.. The extent to which the coffee producing sector contributed to later industrialization through the earnings it generated and the immigrant labor force it attracted, as well as the regional inequalities either created or accentuated through the concentration of its benefits in the Southeastern region, especially Sao Paulo. ^ Celso Fuitado, Ibid. bi Act, with the bankruptQr o f die conveidbili^ system, Brazil’s cunenqr was devaluated, cushionmg the inqiact o f the colUqise m prices. Additionally, B razil’s government put together a broad anticyclical K^niesian-type poli<y to boost national mctxne and protect en^Ioyment As a result the immediate mq>act o f die dqiression was not as severe as m the US. ^ Thomas Merrick and Douglas Graham, Ibid, 16. 55 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The world crisis had created a powerful rationale for a strong central government in Brazil, New industrial sectors in the Southeast region (including rival political groups fix > m Minas Gerais), land owning interests from the Northeast, an increasing middle class, new military and administrative groups, new urban interests as well nationalistic political forces from Rio Grande do Sul (in the Soudi region) all conspired against the prevailing balance of power of the “ Old. R ep u b lic“ . In 1930 national elections were manipulated in favor of the candidate from Sao Paulo, Julio Prestes. Getulio Vargas, the nationalistic gaucho governor from Rio Grande do Sul and defeated candidate, representing the new political forces, carried out a successful coup and became the new president. The “ 1 9 3 0 R évo lu tio n ” ended the “ O ld R ep u b lic” politic period. Sao Paulo’ s elites and military attempted to regain control and to eventually separate the state from the rest of the country in a serious armed reaction in 1932, the “ Sao P aulo C o n stitu tio n a l R evo lu tio n ” . This rebellion was crushed by Federal forces. The history of Brazil for the next quarter of a century would be deeply influenced by Getulio Vargas. A . new constitution (Brazil’ s third) was drafted in 1934. In 1937, however, Getulio Vargas, under the allegation of a communist threat, consolidated his power, suspended civil rights and initiated the “ E sta do N ovo” , an authoritarian period which lasted until the end of W .W . n in 1945. 56 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Getulio flirted with nasâ-fascism. and the US feared th at Brazil, with its la rg e German colony, would not support the Allies. This did not materialize, though. Brazil remained neutral until 1942 when it declared war against nazi-fascism, sending soldiers to the battlefields in Europe. President Getulio Vargas was forced out of office once W .W .H was over. The period between 1946 and 1964 is known as the ‘ ^D em ocratization P eriod” in Brazil. Vargas was quite popular among Brazil’ s poor having implemented union, minimum wages and paternalistic labor laws. In 1950 Vargas returned as President democratically elected. Without political support from the emerging industrial elites, the military and other urban groups the former dictator populist President was isolated and threatened with forced resignation. Instead, Getulio Vargas preferred to “ e x it life a n d e n te r th e h isto ry b o o ks” by shooting himself in the heart. During the war period, and benefited by a fixed exchange rate, export prices were increasing more than the domestic price level, favoring the export sectors. In 1947, under the liberal policies of President elected Gaspar Dutra, imports were liberated, however. As a result the import coefEcient rose 15%, which would be impossible to sustain given Brazil’ s import capacity' and ^ Between 1934 and 1938 the trade between Brazil and Germany also rose significantly. The latter primarily importing cotton and exportmg industrial goods. ® Getulio was known as “ IJte Father o f the Poor” . ^ And particularly die UDN party and its most eloquent representative, die Governor of the State of Rio de Janeiro (Guanabara) Carlos Laceida. ^ As he wrote in his suicide note. 57 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. balance of payments limitations. The country had to face two possibilities, either to devaluate the currency or to establish selective import controls. Brazil chose the latter strategy. Getulio Vargas was an ardent proponent of industrialization since the 1930’ s, promoting exchange and commercial policies to protect and expand native industries. In addition to consumer good and textiles, capital goods, chemicals, metals and steel began to be produced. In 1952, during his presidency, the National Bank for Economic Development (B N D E) was created to support industrial investments. Vargas ignored the severe criticism from the liberal orthodoxy at the time against state intervention in the economy, and created state monopolies in electricity and petroleum (Eletrobras and Petrobras respectively). The decision to adopt import controls was of paramount importance for the industrialization process in Brazil, a successful industrialization based on import substitution. Such control policy focused on the imports of finished consumption goods and not on the imports of capital goods and raw materials. Purtado argues that : In this manner the industrial sector was doubly favored: on the one hand, because the possibility from competition abroad was reduced to a Tninimum owing to import controls; and on the other, because raw materials and equipment could be acquired at relatively low prices. A situation extremely favorable to investments in industries connected with the domestic market was thus created. It was this situation that was responsible for 58 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the lise in the investment rate and the mtensification in the process of growth during the post-war period. ^2 President Vargas's suicide actually offered strong support to his industrializing ideas. Juscelino Kubitscheck, a politician from Ahnas Gerais, who shared many of Getulio’ s political views, was elected President in 1955. Ju st as Vargas, Kubitscheck was a visionary, nationabstic and quite popular President. If one were to identify the historical moment when Brazil began to experience Modem E)conomic Growth it could well be in the 1950's, in between the governments of Getulio Vargas and Juscelino Kubitscheck. The importance of foreign trade as a determinant of GDP levels is enfeebled in favor of the domestic market. GDP growth rates before the 1950’ s were already growing at levels compatible with MEG. Maddison estimates that per capita income annual average growth rates increased from 1.4% in the years immediately preceding W .W . I ( 1900-1913) to 2% from 1913-1950. (tables 19 and 20). Haddad sees per capita product growth rates advancing from 2.15% between 1900-02 - 1945-47 to 3.03% between 1945-47 - 1969-71 (table 24). ^ Celso Fuitado, Formacao Economica do Brasil”. (%o de Janeiro: Editora Fundo de Cultura, 1959), pg. 253. “The Economic Growth of Brazil”: A Surv^ from Colonial to Modem Times.” (Wesport: Greenwood Press, 1963), pg. 241. 59 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Purtado argues that per capita aggregate product growth rates were averaging 2.5%a.a. in tiie before 1929 years (1920-1929), falling to 0.3% and 0.7% during the 1929-1937 and 1937-1947 periods, but increasing significantly to 2.% between 1947 and 1957. (table 21). Hence, the evidence shows that long term per capita growth rates for the first half of the twentieth century were indeed somewhat compatible with Kuznets’ long term values for economic growth in Russia (2.45%, 1913-1958) and Japan (2.37%, 1879-1959) but the growth trend was neither harmonic nor sustained. Only as of the mid 1950’ s one can truly identify modem economic growth through unquestionable sustained increases in per capita GDP growth rates accompanied by structural and organizational changes in society. This premises seems to be somewhat accordant with Easterlin’ s assertion that * Among Latin American countries, modem economic growth appears to be underway in Argentina and, perhaps, Mexico before W .W .I, followed with perhaps a half century lag by Brazil.” ^ 3 In the 1950’ s Brazil’ s economy began to grow at a much higher rate fueled by industrial development. The industrial sector became larger, grew in complexity, and integrated consumer, intermediate and capital goods. E )ven though the import substitution model implemented at that period exhausted itself and showed its contradictions in the early 1960’ s when Brazil faced a serious economic, political and institutional crisis - which ultimately led to the 60 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. military coup of 1964- it was a succes^ul effort to industrialize and modernize Brazil. Industrialization also imposed perverse economic implications - such as the inflationaiy way the industrial effort was financed as well as the income and regional inequalities promoted. President Kubitscheck instituted numerous policies to protect and develop infant industry. An expansive fiscal and monetary policy was enforced. Economic policy was aimed specifically at promoting industrial growth. Incentives were given to auto industry foreign companies which ultimately led to strong growth in the sector. This president presented a “ Plan of Goals" 7 4 to develop the country within 5 years (or “ 5 0 y e a rs m 5" as it became known ), which included the transfer of Brazil’ s capital firom Rio de Janeiro to a new city to be built in the center of Brazil- the futurist Brasilia (founded in 1960). The “ P lano d e M etasT devised an ambitious plan of industrial and infirastructure goals, involving the private and the public sector. Most of the import substitution of the period took place in capital and consumer durable goods. In 1949, the share of imports in the total supply of intermediate goods was already 26%, whereas for nondurable goods it was only 4%. For capital goods, however, firom a post W.W.n. level of 64%, the share of imports dropped to 10% at the beginning of the authoritarian regime in 1964. ^ Ridiard Easterim, Ibid., 33. Plano de Metas. 61 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In the durable goods sector, the import substitution, process saw import shares falling for the same period from 65% to 2%. According to many Brazilian and South American economists of that period, industrialization was the only way to reverse the traditional dependence to exports and the decreasing terms of exchange, limited industrial and subsistence agriculture sectors were unable to provide dynamism to domestic aggregate demand. Economic growth was determined in great part by the international demand of export commodities. Industrialization would alleviate dependence and be the historical answer to all of Brazil’ s economic and social problems. Major changes in society did take place, such as intense urbanization and reorganization of the labor markets. Propelled by the industrialization and technological advancement of Brazil’ s society, urbanization and metropolization levels increased a great deal after the 1940s’ . Urbanization levels increased from a level of 31.2% in 1940, to 36.2% (1950), 44.9 % (1960), 55.9% (1970), 67.6% (1980) .7 6 When the last decade census was taken in 1991, the ofGcial urbanization rate was 75.6%. 7 7 (see Tables 34 and 35) See, for example, Maria Conceicao Tavares, “ Da Substituicao das Importacoes ao Capitalismo Fioanceiro”. (Rio de Janeiro: Zahar, 1973). The critenon for Urbanization accordmg to foe ofiEcial IBGE denominaticn is more much flradble than a 20,000 mhabitants classificatioa. Accordmg to my computation foe percentage of the Brazilian population vho lived m urban centers with at least 20,000 people was 15.3% in 1940 and 61.2% in 1991. ^ IBGE’s (kfoution o f an urban center relates to foe localization o f dwellmgs and people in judicially independent chies, towns or villages accordmg to municipal law. 62 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In 1996 78.3% of the Brazilian population reside in urban centers. In the Northeast region, urbanization level is 65.2% and in the Southeast region it is 89.3%. In the last half a century the change in urbanization rate in Brazil increased over 150%. ? » Industrial development and per capita income gains were substantial while social problems and inequality grew. Modem economic growth in Latin America’ s largest economy happened fast, abruptly and brimful of perverse regional and distributive consequences and contradictions. Between 1920 and 1957, the annual per capita income increased at a 1.6% rate, which Fuitado notices is relatively close to the average for the second half of the 19^ century. Most of the growth happened as of the late 1940’ s and early 1950’ s. The average rate of annual growth of real product per capita increased from 1.9% between 1940-1946, to 3.0% between 1946-1949 to 3.5% between 1949-1954. The rising incomes went firom an annual average per capita growth rate of 2.5% in 1920-29, to 0.3% between 1929-37 to a slightly higher level of 0.7% between 1937-47. ^ For a thorou^ analysis see: Richard Morse, “ The Urban Development of Latm America- 1750-1920”, Center for Latm American Studies, Stanford University, 1971. 63 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Between. 1947 and 1957 the rate of GDP growth jumped to 5.3% and the per capita annual growth rate increased fourfold to 2.8% .^® Maddison calculates Brazil’ s annual per capita growth rates during his four Phases of Development as being 1.4% (1900-13), 2.0% (1913-50), 3.8% (1950-73) and 2.2% (1973-87). « J After a short period of institutional and economic crisis in the early 1960’ s, a military coup took over the government on March 31, 1964, with the support of several international and domestic segments of society. The Authoritarian regime lasted for twenty years until 1985, even though the so- called * a b ertu rtf I opening of the regime was gradual and progressive, starting in 1974. From the late 1960’ s up until 1973, the Brazilian economy experienced very strong industrial growth. This period is known as the * BrazU ian M iracle.’ Between 1969 and 1974, the industrial product grew at an annual average of 12.2%.82 Per capita annual GDP growth increased from 8.62% in 1971 to 9.34 % in 1972 to 11,28% in 1973. The auto industry played a very important role during this boom, growing at an aimual rate of 33% between 1967-70. ^ Data consulted by Furtado. The author observes, that this 1.6% rate, although high, would be below the long term rate in die US. Ibid, 261.. Angus Maddison, “Dynamic Forces in Capitalist Development’ * . ( New Yoric Oxford University Press, 1991). See table. The annual inflation rate increased fl^ ran 12.4% m 1958, to 35.9% in 1959 to 89.9% m 1964. See table. ^ Wilson Suzigan, “ hidustrialization and Econranic Policy m Historical Perfective’ .(Rio de Janeiro; EPEA/INPES, 1976). 64 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. As of the late 1960’ s, the model of import substitution was replaced by an aggressive export promoting pobqr. As a result, sp o rts rose firom a level of $1.2 billion a year in the mid 1950’ s to around $8 biUion a year in 1974. * 3 In spite of protective barriers and ejqiort oriented policies, the military government of Brazil put together a comprehensive industrial and development plan in 1974, the * Z f P lano N acional d e D esenvo lvim ento", which promoted further growth in capital goods industries as well as energy and infra structure. The exogenous shocks of the 1970’ s caused serious problems in Brazils’ balance of payments and monetary policy. After the first shock in 1973, when the cost of imports rose substantially, the government decided to finance the deficit by taking out more loans, instead of containing spending. Once a second shock happened in 1979, the Brazilian economy was inevitably heading to a process of staginflation of the 1980’ s. in 1980, oil imports represented 43% of Brazil’ s total imports.*^ The rise in the international interest rates in 1981 aggravated the economic crisis. Per capita GDP grew - 6.5% in 1981, - 1.6% in 1982 and -5.4% in 1983. Under popular pressure, and after 11 years of gradual concessions, the military accepted in 1985 to hand in the Presidency to a civilian, Tancredo Neves, a moderate firom Minas Gerais. Neves became terminally ill the night ^ Thomas Merrick & Douglas Graham, Ibid, 21. * * The so-called “ lost decade". A strategy to substitute methanol alcohol for petroleum was then in^iemented, the " Pro- Alcool". Tancredo won by receiving the necessary votes fimn the “Congressual College”, whidi used to elect m ly candidates indicated by the military. 65 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. before taking ofBce and Mr. Samey, the conservative appointed vice president became Brazil’ s first civilian president after 21 years. During his presidency in the late 1980’ s inflation rates grew tremendously, causing major losses to the lower ranks of society. In 1988 a liberal constitution, which considerably reduced the bargaining power of the Federal government and increased fiscal benefits for states, municipalities, was drafted. This new constitution would also be behind Brazil’ s ballooning fiscal and social security misfortunes of the late 1990’ s as it enshrined labor benefits and job stability to public employees. After a failed stabilization plan - the * C ru za d o P la n extremely high inflation rates, corruption and rising unemployment came to the forefront by the end of President Barney’ s government in the late 1980’ s. In 1990, a young, telegenic and independent President was elected, the first elected president in almost 30 years, Fernando Collor de Mello. When this President took office inflation was almost 100% a month, most wages and contracts were indexed and pegged to the previous monthly inflation rate. The economic situation was rather chaotic, especially for the poor, who did not have any hedges against the inflationary tax. Mcome distribution was becoming even more skewed. Collor was responsible for a new strategy to lower import tariffs and open the country to privatization and international competition. On the other 66 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. hand, his bold stabilization programs ^ failed. In 1990 Brazil’ s economy was in deep recession and per capita GDP fell 6.6%. Ultimately, Collor was forced to resign and to lose all political rights in 1992, charged with corruption. ^ Once again, the vice-president, Itamar Franco, became the substitute President. Mr. Franco’ s appointed Fernando Heniique Cardoso, a Sao Paulo senator and eminent sociologist, to be his Finance Minister. In July of 1994, N h r. Cardoso conceived an extremely successful stabilization plan and created a new currency, the Real. After over a decade of stagnation, the Brazilian economy began to grow vigorously again. Annual inflation rates fell from over 5,000% in 1993 to its lowest post W.W.II level, reaching less than 5% in 1998. The success of the “ R ea l P la rf was the credential necessary to elect Fernando Henrique Cardoso as President of Brazil in 1995. Particularly the poor, who carried the burden of the inflationary tax in their pockets, and who saw their real incomes rising significantly with the new currency, supported die new President. Between 1993 and 1997, the Brazilian economy expanded, stimulated by the increasing aggregate demand brought about with the “ R eed P lan” . The success of the stabilization plan, and particularly the falling inflation rate, reduced income inequality levels during this period and m ust bi his first attraqX upon taking oath his government firoze all assets. Including savings and diecking accounts. ^ Good sign for the Brazilian mfont demociary, he was one o f foe fow democratically elected Presidents to be impeadied democratically m foe Western Hemi^here. 67 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. certainly have had a positive impact on infant mortality. Between 1953, when the industrialization effort began to intensify, and 1979, inflation remained at double digits during all years. From 1979 to 1987, inflation reached triple digits levels and from 1987 to 1993 the so called “ necesscary evU” reached four and even five digits levels. Inflation rates were very high, but not out of control, in the sense that prices, financial assets and most wages were indexed to the previous month inflation rate. The losses for the working class were significant though. The advancement of modem economic growth in Brazil was executed through a formidable process of regional and income concentration. Purtado maintains that: Although by the middle of the century the Brazilian economy had achieved a certain degree of articulation between the various regions, the disparity in regional income levels had also increased sharply. As industrial development took the place of coffee prosperity, the tendency toward regional concentration of income increased....The industrialization process started in Brazil concomitantly in almost every region. However, after the first trial stages were over , the industrial process naturally tended to concentrate in a single region.... The decisive stage of concentration apparently occurred during World War I, when the first phase of speeding up industrial development occurred... In 1955, the state of Sao Paulo, with its population of 10,330,000 inhabitants, had a gross product 2.3 times larger than the entire Northeast, whose population for the same year was 20,100,000. The per capita income in the Sao Paulo region was consequently 4.7 times higher than that of the Northeast. This growing disparity between the living standards of the main population groups in Brazil may give rise to serious regional tension. Celso Fuitado, "the Economie Growth of Brazil”, Ibid., 265. 68 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. This pattern of regional inequality was only worsened in the following three decades. In fact, the intense industrial growth was financed, to a certain extent, by the increasing income and regional inequality. Import Substitution-led industrial protectionism and chronic inflation are among the main causes of income and regional inequality. The historical unequal landholding pattern also plays important role, particularly in Brazil’ s Northeast. Nevertheless, between 1994 and 1995, with falling inflation, economic growth and increasing mir>imum wages real incomes rose whereas poverty and income inequality actually fell considerably. In 1994, 33% of the people living in Brazil’ s six major cities were poor. In 1996 this ratio had fallen to 25%. During this period, interest rates were kept very high, the currency overvalued, and a significant growth in foreign direct investments occurred. In spite of the fact that the privatization process was considerable, the necessary adjustments in public spending were not implemented. Such fiscal inconsistencies, which were also aggravated by the 1988 Constitution, made the Brazilian economy quite vulnerable to international speculative swings. ” The government more than doubled minimum wages during the early phase o f the Real Plan (94-97). According to IPEA/INPES. 69 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In 1998 with, the Asian Crisis, the government was able to defend its currency by increasing interest rates to very high levels and by using its international reserves. Under these critical auspices, in November of 1998, President Cardoso was re-elected, the first democratically re-elected president in Brazil’ s history. Immediately after taking office, the President changed its exchange policy, devalued the currency, raised interest rates and made an agreement with the I.M.F. As a result, the economy of Brazil is facing a situation of rising moderate inflation, unemployment and recession in 1999. However, and given the country has faced crises in the past, there are no evident indicators it will not recuperate, that is, that the currency will not strengthen, inflation controlled or that economic growth will not return in the year 2000. The distributive impact of the crisis seems to be even more troublesome, though. With unemployment on the rise, cuts in public spending and public programs (including health and education) and inflation losses, the new economic crisis w iU mostly affect the poor. With the success of the “ M ono R ea l’ ’ , this country, which has maybe the worst income distribution in the world as measured by the Gini coefficient, witnessed considerable gains in income distribution. The poorest 20% of the Brazilian population hold only 2.5% of the total income whereas the richest 20% hold 64% of the country’ s product, accx>rding to the IBGE (1995 values). 70 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Nevertheless, between 1994 and 1996 the poorest half of Brazil’ s 170 million people had a 55.6% increase in their incomes, as opposed to a 22.7% increase for the wealthiest 10% of the population. Between 1996 and 1998, the income gains of the poorest half of the population rose 36%. Yet, in 1996 53 million Brazilians were still the poverty line, half of them in the Northeast. ^ At the turn of the millennium, all such gains will be either diluted or neutralized. Inequality in this rather unequal country is again on the rise which will certainly have a negative impact on child survival chances in the next few years. Table 17 Brazil's and Latin America's GDP PC Annual Average Growth Rates During Maddison's Phases of Development- 1900-1987 (% ) Brazil Latin America * 1 90 0-1 91 3 1.4 2 -1 19 1 3 -1 9 5 0 2 1.4 1 95 0-19 73 3.8 2.5 1973-19 87 2.2 * Latin America's average including Brazil. 0.8 Source; Angus Maddison (1989) 92 This threshold being defined as earning a montfaly income of less than 65 reais at die tune. 71 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 18 BrazH's Average Rate of Annual Growth of Population, Abrogate Product and Per Capita Product -1900-1987 (%) GDP Populat. G D P/ PC 1900-1950 4 2.1 1.8 1950-1987 6 2.7 3.2 1900-1987 4.8 2.4 2.4 Sorce; Angus Maddison ( 1989) Table 19 Brazil's Average Rate of Annual Growth of Real Product PC 1940-1954 (% ) 1940-1946 1.9 1946-1949 3 1949-1954 3.5 Source: Celso Furtado ( 1959) 72 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 20 Brazil's Average Rate of Annual Growth, of Aggregate Product -1920-1957 (% ) 1920-1929 4.5 1929-1937 2.3 1937-1947 2.9 1947-1957 5.3 1920-1957 3.9 Source: Celso Furtado ( 1959) Table 21 Brazil's Average Rate of Annual Product Per Capita -1920-1957 Growth of Aggregate (%) 1920-1929 2.5 1929-1937 0.3 1937-1947 0.7 1947-1957 2.8 1920-1957 1.6 Source: Celso Furtado ( 1959) 73 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 22 Brazil s Per Capita GDP Growth Rates, 1950-1990 (% ) 1950 0 1964 0 1978 2.58 1951 3.09 1965 -2.54 1979 4.34 1952 4.94 1966 0.97 1980 6.75 1953 -0.32 1967 2-75 1981 -6.54 1954 5.78 1968 7.56 1982 -1.6 1955 3.7 1969 6.87 1983 -5.44 1956 0.92 1970 6.63 1984 3.02 1957 4.52 1971 8.62 1985 5.69 1958 5.48 1972 9.34 1986 5.08 1959 2.88 1973 11.28 1987 1.3 1960 5.21 1974 5-78 1988 -2.31 1961 6.54 1975 2.59 1989 1.17 1962 2.38 1976 7.66 1990 -6.57 1963 -1.77 1977 2.42 Source; IBGE Research D epartm ent (1993) Table 23 Brazil's Average Rate of Annual Growth of Aggregate Product Per Capita -1900-02 to 1969-71 (% ) Y ears Product P er Capita 1900-02 to 1910-12 1.74 1910-12 to 1929-22 1.75 1920-2210 1930-32 1.95 1930-3210 1940-42 2.56 1940-4210 1945-47 3.02 1900-02 to 1945-47 2.15 1945-4710 1969-71 3.03 1900-0210 1969-71 2.46 Source: Claudio Haddad (1974) 74 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 2 4 Brazil’ s Per Capita GDP, $ at 1985 US prices, 1900-1989 1900 1913 1950 1973 1989 586 700 1441 3363 4241 Source: A.Maddison (1989) Table 25 Brazil's Average Rate of Annual Growth of Aggregate and Sectoral Product -1900-02 to 1969-71 (% ) Agricult. Industry Total Product 1900-02 to 1910-12 2.31 6.25 3.96 1910-12 to 1929-22 3.29 5.65 3.97 1920-22 to 1930-32 3.3 3.51 4.04 1930-32 to 1940-42 2.86 7.2 4.71 1940-42 to 1945-47 2.42 9.18 5.49 1900-02 to 1945-47 2.88 6.03 4.3 1945-47 to 1969-71 4.39 8.31 6.45 1900-02 to 1969-71 3.56 6.82 5.05 Source: Claudio Haddad (1974) 75 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 26 Brazil's Average Rate of Annual Growth of Aggregate and Industrial Product 1 9 4 8 -5 2 to 1 9 6 8 - 7 4 (%) Industry Total Product 1948-52 8.8 7 1953-57 8-1 6-1 1958-62 11.2 7.6 1963-67 2.9 3.4 1968-74 12.2 10.1 Source: Wilson Suzigan (1976) Table 27 Recent Shares o f G D P (% ), 1 9 6 5 -1 9 9 0 1965 1973 1980 1986 1987 1990 GDP m.p 100.0 100.0 100.0 100.0 100.0 100.0 Net indirect Taxes 10.9 14.3 9.6 10.8 10.7 10.1 Agricult. 16.5 11.4 9.8 10. 11.1 11.2 Industry 29.9 33.4 37.1 34.2 33.2 32.6 (of wtiich Manuf.) 23.3 25.5 28.4 24.8 - - Services 42.6 40.8 43.6 45.1 45. 46.1 76 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 28 GDP Growth. Rates b y Sector, 1948-1976 (% ) Growth R ates * Year industry Agriculture Services Total 1948 11.3 6.9 5.8 7.5 1949 10.3 4.5 6. 6.6 1950 11.3 1.5 7.1 6.5 1951 6.4 .7 9.9 5.9 1952 5. 9.2 10.8 8.7 1953 8.7 .2 -.01 2.5 1954 8.7 7.9 13. 10.1 1955 10. 7.7 3.5 6.9 1956 66.9 -2.4 4.7 3.2 1957 5.7 9.3 9. 8.1 1958 16.2 2. 5.4 7.7 1959 11.9 5.3 1.2 5.6 1960 9.6 4.9 13. 9.7 1961 10.6 7.6 11.9 10.3 1962 7.8 5.5 3.3 5.2 1963 2. 1. 2.9 1.6 1964 5.2 1.3 2. 2.9 1965 -4.7 13.8 1.3 2.7 1966 9.8 -15. - 3.8 1967 3. 9.2 - 4.8 1968 13.3 4.5 - 11.2 1969 12-1 3.8 - 10. 1970 10.4 1. - 8.8 1971 14.3 11.4 - 13. 1972 13.4 4.1 - 11.7 1973 15.8 3.5 - 14. 1974 9.9 8.5 - 9.8 1976 9:4 = 6:6 1976 10.8 4.2 - 9. 1977 3.9 9.6 - 4.7 * Ot)s. These estim ates for GDP growth seem s to be th e sam e as the ones used by ECLAC, and they differ a little from the official IBGE values. Source; World Bank, Human Resources Special Report (1977). 77 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 2 9 GDP Growth Rates by Sector, 1980-1990 (%) Growth Rates Y ear Industry Agriculture* C onstruct Trade** Total 1977 - - - - - 1978 - - - - - 1979 - - - - - 1980 9.1 9.5 9. 8.5 9.2 1981 -10.4 8.1 -6. -6.4 -4.4 1982 -.2 -.5 -1.3 .2 .6 1983 -5.8 -.5 -14.2 -4.2 -3.4 1984 6.2 3.0 -.6 4.1 5.3 1985 8.3 9.8 10.9 7.8 7.9 1986 11.3 -8.1 17.5 8.1 7.5 1987 1. 15. 1.1 2.6 3.6 1988 -3.4 -.4 -3. -2.8 -.1 1989 3.1 2.2 7.6 2.9 3.3 1990 -9.5 ^Includes Forestry and Fishing ** Includes W holesale and Retail Source: Brian Mitchell (1983) -4.4 -12.3 -6.5 -4.6 Table 30 Growth Rates of Population and Labor Force in Brazil and in the Northeast, 1940-1970 (% ) Average Growth Rate of Labor Force (a.a) Brazil Pop. Growth Total Agricult. Industry Serv 1940-50 2.34 1.7 .51 4.25 3.57 1950-60 3.05 2.92 1.97 2.38 6.35 1960-70 2.88 2.61 .54 7.66 3.07 Northeast 1940-50 1.77 1.22 .63 1.77 3.86 1950-60 2.78 2.39 1.83 1.45 4.8 1960-70 2.2 1.67 .54 5.58 3.38 Source: World Bank, Human Resources Special Report (1977). 78 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 31 Regional Distribution of the Labor Force by Sector, Brazil 1940-1991 (%) Percentage of the Labor Force in Region Northeast S outheast Frontier Total 1940 3 4 2 59.3 6.4 1950 32.8 60.8 6.4 1960 31.1 61.3 7.6 1970 28.4 62.9 8.7 1991 21.4 68.3 10.3 Agricult. 1940 40. 52.9 7.1 1950 40.5 51-7 7.8 1960 39.9 50.9 9.2 1970 39.9 49.3 10.8 1991 35. 47.3 17.7 Industry 1940 20.7 74.8 4.4 1950 16.3 80.8 3. 1960 18.4 76.8 4.8 1970 15.1 79.9 5. 1991 13.2 81.6 5.2 Services 1940 24. 70.5 5.4 1950 24.7 70. 5.3 1960 21.3 72.7 5.9 1970 22. 69.6 8.4 1991 18.6 72.3 9.1 Source: IBG E. R esearch DepL(1994) 79 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 32 Participation. Rates and Structure of the Labor Force in Brazil and in its Northeast Region, 1940-1991 (%) Labor Force Percentage of the Labor Force in Participation Agriculture Industry Services Rate (%) Brazil 1940 34.9 67.9 12.6 19.9 1950 32.8 59.9 16.1 23.9 1960 32.4 54.5 12.4 33.1 1970 31.6 44.5 20. 34.6 1991 35.1 33.8 28.2 38. Northeast 1940 34.3 78.5 7.6 13.9 1950 32.5 74. 8. 18. 1960 31.3 70. 7.3 22.7 1970 29.7 62.5 10.6 26.8 1991 33.5 55.1 14.8 30.1 Source: BGE and World Bank, Human R esources Special Report (1977). 80 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 33 General Historical hidicators, Brazil, 1865-1973 R ailroad. Telegraph Postal Proportion of World Trade (.00001) (jperSq.MI.) (t=mill.sent) (m=mill.sent) 1865 0. _ - 2154 1866 1. - - 2220 18671. - - 2094 1868 1. - - 2235 1869 1. - 10 2355 1870 1. - 10 2302 1871 2. .1 12 2037 1872 2. .1 13 1862 1873 3. .1 14 1785 1874 3. .1 13 1701 1874 4. .1 15 1959 1876 5. 2 15 2140 1877 5. 2 16 2358 1878 6. .3 19 2506 1879 7. .3 20 2510 1880 7. .4 24 2084 1881 8. .3 24 1998 1882 9. .3 25 1684 1883 10. .4 32 1665 188411. .4 35 1623 1885 11. - - 1685 1886 12. .7 56 1691 1887 12. .5 40 1883 1888 13. .6 44 1971 1889 13. .8 50 2075 1890 14. 1. 64 1799 1891 14. 1.2 87 1869 1892 17. 1.1 72 1915 1893 20. 1.3 69 1830 1894 23. 1.5 75 1331 1895 26. 1.7 105 1567 1896 27. 1.7 135 1411 1897 28. 1.4 198 1463 1898 28. 1.4 197 1495 1899 29. 1.4 278 1501 1900 29. 1.2 320 1560 1901 30. 1.2 326 1528 1902 30. 1.4 347 1396 1903 31. 1.5 380 1457 1904 30. 1.5 390 1530 Source; Arthur Banks (1971) for Railroad and Trade Data & B. for Postal and Telegraph Data. Mitchell (1983) Si Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 33 (cont) Railroad. Telegraph Postal Proportion of World T rade (.00001) (perSq.MI.) (l=mill.sent) (m=mill.sent) 31. 1.5 394 1631 1905 33. 1.7 472 1630 1906 35. 1.9 520 1659 1907 37. 2.3 567 1463 1908 39. 2.4 481 1767 1909 41. 2.8 544 1768 1910 43. 2.8 608 1745 1911 45. 3.7 612 1911 1912 47. 3.8 634 1767 1913 - 4. 653 - 1914 - 3.7 443 - 1915 - 3.9 479 - 1916 - 4.4 466 - 1917 - 5.4 514 - 1918 54. 5.6 546 1837 1919 54. 6.6 642 1466 1920 55. 6.1 624 1094 1921 56. 6.6 773 1248 1922 56. 6.9 873 1040 1923 57. 7.2 1225 1169 1924 58. 7.6 1746 1220 1925 58. 7.4 1861 1019 1926 59. 7.5 1912 988 1927 60. 6.5 2152 1071 1928 60. 6. 2105 1001 1929 60. 5.5 1909 1467 1930 61. 7.1 1821 1341 1931 61. 8.1 1403 1485 1932 62. 8.6 1708 1742 1933 62. 8.9 1834 1654 1934 63. 9.9 2554 1577 1935 63. 10. 2555 1688 1936 64. 11. 2308 1688 1937 64. 11. 3004 1691 1938 64. 11. 3141 1652 1939 - 11. 655 - 1940 - 13. 592 - 1941 - 15. 546 - 1942 - 18. 689 82 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 33 (cont.) R ailroad. Telegraph Postal Proportion o f World Trade (.00001) (per Sq.MI.) (t=mill.sent) (m=miil.sent) 1943 - 20. 932 _ 1944 - 24. 1183 - 1945 66. 27. 1280 2884 1946 67. 29. 1394 2704 1947 67. 30. 1850 2366 1948 68. 26. 1865 2065 1949 68. 27. 2017 2352 1950 68. 29. 2160 2647 1951 69. 29. 2300 2363 1952 69. 31. 2183 1963 1953 69. 29. 2349 2095 1954 70. 32. 2508 1625 1955 70. 24. 1962 1449 1956 71. 21. 3884 1394 1957 71. 22. 1687 1313 1958 72. 20. 2734 1255 1959 72. 26. 4948 1137 1960 70. 27. 3997 1131 1961 68. 24. 4398 990 1962 67. 25. 4390 976 1963 65. 28. 4534 810 1964 63. 18. 2325 744 1965 62. 30. 2964 818 1966 - 37. 2327 - 1967 - 22. 2755 - 1968 - 18. 3385 - 1969 - 17. - - 1970 - 19. - - 1971 - 20. - - 1972 - 21. - - 1973 - 20. - - Source; Arthur Banks (1971) for Railroad and T rade Data & B. Mitchell (1983) for Postal and Telegraph Data. 83 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 3 4 U itan Definition and Calculation Cens.Y ear Own, based on C ensus Tabulation (vary) IBGE Definition (official) Merrick & Graham (20,000+) Arthur Banks (20,000+) 1940 20,000+ 15.3 31.2 16.1 - 1940 100,000+ 10.3 - - - 1940 500,000+ 6.7 - - - 1940 1000,000+ 6.7 - - - 1940 20,000+ 2 0 2 3 6 2 21.1 2 0 2 1950 100,000+ 13.2 - - - 1950 500,000+ 9.4 - - - 1950 1000,000+ 5.4 - - - 1950 2000,000+ 5.4 - - - 1950 20,000+ 28. 44.9 28.8 29.4 1960 100,000+ 16.4 - - - 1960 500,000+ 12.8 - - - 1960 1000,000+ 9.2 - - - 1960 2000,000+ 9 2 - - - 1960 20,000+ 39.4 55.9 38.8 29.6 (1966) 1970 500,000+ 15.6 - - - 1970 1000,000+ 13.8 - - - 1970 4000,000+ 10.3 - - - 1970 20,000+ 51.4 67.6 - - 1980 500,000+ 21.5 - - - 1980 1000,000+ 19.9 - - - 1980 5000,000+ 11.4 - - - 1980 20,000+ 6 1 2 75.6 - - 1991 1000,000+ 20.2 - - - 1991 2000,000+ 13.1 - - - 1991 5000,000+ 10.3 - - - 1991 78.4 84 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 35 Urbanization, 1940-1996(%) 1940 1950 Region %in Brazil Urban % Rural % Region %in Brazil Urt>an% Rural % Pop-On Pop. million) Brazil 41.2 - 31.2 68.8 52. 36.2 63.8 South 5.7 13.8 28. 72. 7.8 15.1 29.5 70.5 Southeast 18.3 44.4 39.3 60.7 22.7 43.6 47.1 52.9 Mid-West 1.26 3.1 21.4 78.6 1.72 3.3 24.4 75.6 Northeast 14.4 35. 22.9 77.1 17.9 34.4 26.2 73.8 North 1.54 3.7 26. 74. 1.88 3.6 30.8 69.2 1950 1960 Total Pop. %in Brazil Urtan% Rural % Total Pop. %in Brazil Urt>an % Rural % Brazil 52. - 36.2 63.8 70.2 44.9 55.1 South 7.8 15.1 29.5 70.5 11.8 16.8 37.3 62.7 Southeast 22.7 43.6 47.1 52.9 30.7 43.8 57. 23. Mid-West 1.72 3.3 24.4 75.6 2.95 4.2 33.8 66.2 Northeast 17.9 34.4 26.2 73.8 22.2 31.6 33.7 66.3 North 1.88 3.6 30.8 69.2 2.55 3.6 37.6 62.4 1970 1980 Total Pop. %in Brazil Urban % Rural % Total Pop. %in Brazil Urt>an % Rural % Brazil 93.1 - 55.9 44.1 119. 67.6 32.4 South 16.5 17.7 44.2 55.8 19.1 16. 62.3 37.7 Southeast 39.9 42.8 72.7 27.3 51.8 43.6 82.6 17.4 Mid-West 5. 5.4 48. 52. 7.5 6.3 68. 44. Northeast 28.1 30.2 42. 58. 34.8 29.2 50.5 49.5 North 3.6 3.9 45.2 54.8 5.8 4.9 51.7 48.3 1991 1996 Total Pop. %in Brazil Urban % Rural % Total Pop. %in Brazil Urt>an % Rural % Brazil 146.9 - 75.6 24.5 157.1 78.4 21.6 South 22.1 15. 74.2 25.8 23.5 15. 77.2 22.8 Southeast 62.7 42.7 87.8 12.2 67.0 42.6 89.3 10.7 Mid-West 9.4 6.4 80.1 19.9 10.5 6.7 84.4 15.6 Northeast 42.4 28.9 60.6 39.4 44.7 28.5 65.2 34.8 North 10.3 7. 57.2 42.8 11.3 7.2 62.4 37.6 IBGE's official definition of an urban center relates to the localization of dwellings and people in judicially independent cities, towns or villages according to municipal law. 85 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 36 Real GDP Growth ECLA Years ECLA IBGE IMF OECD 1921 ^ 1951 5.9 6.57 - 5.1 1922 5. 1952 8.7 8.35 - 5.7 1923 5.8 1953 2.5 2.84 - 3.2 1924 .2 1954 10.1 9.07 - 7.6 1925 4.2 1955 6.9 6.91 - 6.8 1926 .2 1956 3.2 4.03 - 1.9 1927 5.3 1957 8.1 7.67 - 6.9 1928 8.2 1 9587.7 8.64 - 6.6 1929 .7 1959 3. 5.97 - 7.3 1930 -3.4 1960 12.5 8.39 - 9.7 1931 -.6 1961 10.3 9.82 - 10.3 1932 1.1 1962 5.2 5.56 - 5.2 1933 5.6 1963 1.6 1.27 - 1.5 1934 6.8 1964 2.9 3.01 2.6 2.9 1935 2.8 1965 2 .7 .26 23.1 2-7 1936 9.1 1966 3.8 3.75 3.5 3.8 1937 2.6 1967 4.9 5.45 5.4 4.9 1938 4.1 1968 11.2 10.34 10.8 11.2 1939 2.8 1969 9.9 9.61 9.8 9.9 1940 1. 1970 8.8 9.29 2.6 8.9 1941 4.9 1971 13.3 11.33 12.2 12. 1942 -2.8 197211.7 12. 10.9 11.1 1943 5.8 1973 13.9 13.97 13.5 13.6 1944 4.6 1974 9.8 8.31 9.7 9.7 1945 .9 1975 5.7 5.03 5.6 5.4 1946 7.8 1976 9. 10.22 9.7 9.7 1947 2.4 1977 4.7 4.85 2.9 5.7 1948 7.4 1978 6. 4.98 5. 5. 1949 6.6 1979 6.4 6.77 6.8 6.4 1950 6.5 1980 7.2 9.21 9.2 7.2 1981 -3.4 -4.4 -4.4 -1.6 1982.9 .64 .7 .9 1983 -2.3 -3.3 -3.4 -3.2 1984 5.7 5.32 5. 5.7 1985 8.3 8. 8.3 8.4 1986 8. 7.36 7.5 8. 1987 2.9 3.46 3.6 2.9 1988 - -.26 - - 1989 - 3.26 3.3 - 1990 - -4.69 -4.1 - 1991 - .24 - - 1992 - -.78 - - 1993 - 4.12 - - 1994 - 5.7 - - 1995 - 4.2 - - 86 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. “ I am trying to show how a society can begin to move fo rw a rd a s it is, in spite o f w hat it is a n d because o f w hat it is. " (A lbert H irschm an, on B ra zil's Northeast, Journeys Tow ard P rogress) 87 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 3 THE NATURE AND FACTS OF THE MORTALITY REVOLUTION & INFANT MORTALITY DECLINE IN BRAZIL AND IN ITS NORTHEAST REGION Nature o f Mortality Revolution and Dem ographic Transition Introduction. Easterlin asserts that the timing of the fe r tility tra n sitio n is closely determined by the timing of m ortality revo lu tio n , and, more specifically by the great reduction in infant and mortality decline that accompanied the m o rta lity revo lu tio n . ^ The fast population growth following mortality decline would be a transient state in the sense that fertility decline would follow mortality decline with a lag, that in the case of developing countries such as Brazil, would be of a few decades. The evidence seems to strongly corroborate this claim. The transition between high and low levels of mortality and fertility is often descriptively defined as the d em ographic tra n sitio n . Demeny defines dem ographic tra n sitio n through its connection with a modernization process: * In traditional societies fertility and mortality are high. In modem societies, fertility and mortality are low. In between, there is demographic transition.” 9 4 ” Ridiard Easterlin, Ibid, 1996, pg. 95. Paul Demeny (1968), 502. 88 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Notestein is recurrently cited as one of the first authors to present the dem ographic tra n sitio n in theoretical terms. According to this author, the onset of industrialization would characterize a progressive decline in mortality by virtue of im p ro vin g nutritional and livings standards, whereas fertility would remain at first at high levels. 9 s Notestein points out that, after a period of rapid population growth, the associated impact of mdustrializadon, urbanization, modernization and mortally decline would bring about a fertility reduction; "...under the impact of urban life, the social aim of perpetuating the family gave away progressively to lhat of producing health, education and material welfare of the individual child; family limitation became widespread; and the end of the period of (population) growth came in sight." ^ Caldwell suggests that a broader and more encompassing term than m o rta lity tra n sitio n would be the term epidem iologiccd tra n sitio n since it would embrace not only the levels of mortality but also of sickness. This author proceeds to suggest a even more suitable term which would involve not only "purely outcome measures* but also social and behavioral changes, the term h e a lth tra n sitio n . ” Notestein (1945). (Quoted by Szreter (1993) and Camaiano (1996). ” Camaiano observes that Notestein would later alter his mitial ^ r o a d i, seeing the dnrd phase of demographic transition less as a strictly dependent phase and more as a period Wridi could be affected by govemment-^onsored policies of fiunily pUummg. Caldwell, 1990. 89 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Focusing more on mortality than health, PaUoni argues that the term h e a lth tra n sitio n has a veiy specific meaning in the experience of Latin American countries. Unlike the convergent and stable trajectories of the h e a lth tra n sitio n in developed countries, the passage firom high to low mortality regimes in Latin America would be characterized, according to Palloni, by four basic properties: multiplicity of paths, vulnerability (and discontmuity), increasing mortality differentials by social class and the close association between h e a lth tra n sitio n and reproductive tra n sitio n , ^ Endogenous Socioeconomic Changes vs. Ebcogenous Health-Medical Innovation Changes? Notes on the Nature of Brazil*» Mortality neH fne Many authors associate the tim ing of d em og ra p hic tra n sitio n with a modernization process, but in so doing they usually ascribe a dependent character for the former and an economic-driven character for the latter. Nevertheless, modernization is at best an imprecise concept. A more encompassing definition than the ones set forth by some of the so-called classic formulators of d em og ra p hic tra n sitio n is offered by Inkeles This author associates modernization with a '‘ syn d ro m e o f a ttitu d e s , vcdues a n d b eh a vio r ”, having education (in a broader sense) as one of its main forces. ^ PaUoni, 1990. Inkeles, 1969. Exammg modernization m developmg countries, Fuitado defines Modernization as this process of adoption o f more sophisticated patterns of consumption (private and public) wfaidi occurs without the correspondmg process o f co ital accumulation and tedinical progress m productive methods.” 90 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In spite of the fact that the patterns of spread are similar, dem ographic tra n sitio n changes are much faster and wider than economic changes. In addition, demographic changes are much more evenly distributed, often lead to convergence, are more cost effective, involve a higher participation of women and labor and demand less institutional requirements than economic growth. Easterlin underscores th at what demographers often call dem ographic tra n sitio n is the shift between initially high to eventually low levels of mortality and fertility, and that, with the notable exception of some Sub-Saharan countries in which the process has started in recent times, such a transition is well developed in most of the world. 102 Preston well points out that there is much more consensus on the fact of mortality decline in developing countries than on its causes. G rosso m odo there are two basic interpretations about the determination of mortality levels: one view attributes mortality decline to endogenous changes in living standards and to socioeconomic development; the alternative view contends that the driving force behind m ortality re vo lu tio n would be a revo lu tio n in public health and changes in medical technology, which would be per se exogenous changes. See Celso Fuitado, “ Underdevelopment and Dependence; The Fundamental Connection”, Wbridng Paper No. 17, Centre of Latin American Studies, Universiqrof Cambridge, 1973. Easterlin (1996), Ibid., 95. Samuel Preston, “ Causes and Consequences of Mortality Declmes m LDC Durmg the Twentieth Century” in “Population and Econcxnic Change m Developing Countries”, edited by R _ Easterlin. (Chicago: Univers^ o f Chicago Press, 1980). 91 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Investigating mortality decline in England in the nineteenth and early twentieth centuries McKeown advocates that the changes produced by the d em ographic tra n sitio n , more specifically» increasing mortality gains, are inextricably and endogenously related to economic growth. Per capita income growth, higher living standards, and the resulting upgrading quantity and quality of clothing and food intake would be the main forces behind falling mortality rates. McKeown argues that only as of the early decades of the twentieth century advances in medical technology would have an impact on life expectancy levels in Britain. Palloni maintains that such results would be, in all likelihood, generalizable to other area in Western Europe and North America, los In contrast, in an empirical and historical work, Preston & Haines assert that changes in the first phase of the d em og ra p hic tra n sitio n in late nineteenth- century US would have been caused not by income growth itself but by new medical technologies and knowledge, as well as by the dissemination of health practices, water control and public intervention, These authors often maintain that mortality levels are not determined by lack of resources but rather by lack of know how. McKeown (1976). Alberto Palloni, “ Moitality^ in Latm America; Emerging Patterns”, Population and Development Review 7, No. 4, December 1981. Samuel Preston & Midiael Hames, ‘ Tatal Years”.(Prmceton: Prmceton Lbiversity Press, 1991.) 92 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Preston contends that the contribution of per capta income growth to changes in life expectancy levels would not account for more than 25% of the improvements. For a given level of pci, life expectancy would only change through health innovations. Using a sample of developed and developing nations, Preston estimates that during the 1930-1970 period, 80% of the improvements in life e^ectancy can be attributed to changing technologies. Using a smaller sample of developing counties for the period 1940-1970, this author calculates the relative importance of changing technologies as 50%. Can such results be generalized and valid for understanding mortality decline in Latin America and in Brazil in particular? Palloni argues that the impact of changing technologies on life expectancy gains in Latin America between 1950 and 1970 would also be around 50%. Easterlin posits, however, that the specific nature of the underlying health technologies, the level of institutional intervention and public health would be among the cardinal causes behind the d em o g ra p h ic tra n sitio n . Immunization measures, sanitation, water and food control and personal hygiene would be largely independent of economic improvements. Samuel Preston, “The Changing Relation Between Mortality and Level o f Economic Development ”, Population Studies, No. 2, 231-248,1975. Alberto Palloni, “ Mortality Decline in Latin America”, Ptqier presented a t the Annual Meeting of the Population Association of America, 1979. 93 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Similarly, Arriaga & Davis 10» and Stolnitz 11° maintain that the magnitude of mortality gains in the first phase of the mortality decline in less developing countries would have been determined by the dissemination of medical technology, by the development of specific vaccination and immunization campaigns, by government intervention and by public health improvements which would be independent of changes m the levels of well being. As a result of the formidable gains in life expectancy promoted by such improvements in public health, the transition firom very high to much lower mortality levels in these countries was accomplished within a few decades. In accordance with Easterlin, Arriaga & Davis and other proponents of this view, Preston contends that the discrepancy between the expected and the observed levels of life expectancy in developing countries would be associated with the availabüiy of low-cost technologies to improve health conditions. Assessing the historical course of mortality decline in Latin Am erican, Palloni maintains that there appears to be some degree of independence between the effects of medical innovation and the effects of socioeconomic development. This author observes, however, that whereas the latter can contribute to mortality decline in the absence of the former, as in the English case, medical innovation would be conditioned by socioeconomic development. Eduardo Arriaga & Kmgsley Davis, The Pattern o f M oitaliy Change m Latin America”, Demography, Vol. 6, No. 3., 1969. G. Stolnitz, “ Recent Mortality Trends m Latin America, Asia and Africa”, Population Studies, Vol. 19, 1965. Samuel Preston, Ibid., 1980. 94 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Palloni m aintains that; Not only does the stock of knowledge to produce efScient medical interventions depend on the achievement of some degree of industrialization- a relationship that is becoming less and less relevant in that backward economic ^stem s can to various degrees utilize new technology produced elsewhere - but, more importantly, a low level of socioeconomic development sets boundaries on the possibility of absorbing efGdently a certain type of technology. “ 2 This research sustains that a correlation between public health, income growth and mortality decline does not necessarily imply the existence of an underlying causal explanation. One needs to prevent the pitfalls of generalizing different historical experiences and oversimplifying specific mortality transition contexts. It seems that in the case of the leading mortality revolution countries in Western Europe and North America, socioeconomic development and rising living standards did have a substantial impact on mortality levels, and particularly on life expectancy. Historically mortality decline preceded medical breakthroughs and innovations in health policies, which would also initially arise in these same leading countries. Such seemingly exogenous medical improvements were made possible by virtue of the accumulation of knowledge facilitated by the leadership presented by these very same countries in industrial development and modem economic growth. Alberto PaUoni, Mortality m Latin America: Emetgmg Patterns”, 626. 95 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. To generalize or overestimate the impact of endogenous economic changes on mortality levels in countries which had distinct historical experiences, as far as the timing of both MEG and the dissemination of public health initiatives and medical breakthroughs, is an impropriety. Even in Southern and Eastern Europe where the timing of mortality revolution was different, and mortality decline concentrated between 1920 and 1950, the impact of living standards or endogenous changes on mortality decline was not as robust as in Western Europe, Canada and die US. Analogously, there is no uniform mortality transition path, there is no generalizable primacy of economic development on the determination of mortality decline. In the leading countries of the mortality revolution, which not incidentally were also the same leading countries in the development and dissemination of modem economic growth, endogenous causes driven by economic changes seem to have played an important role. It could not be otherwise since advances in public health and medical technologies only happened decades later. E)ven in LD C’ s, such as Brazil, in which mortality transition only developed full blown after W.W.II, income status did play a w eighs role since mortality and morbidity levels are certainly affected by social class and income inequality. 96 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. It seems evident that the diffusion of new technologies in the medical science, as well as immunization and public health policies, do exert an immediate positive impact on mortality levels even in the absence of economic development. Conversely, a low level of socioeconomic development does set limits on the efBciency of technical absorption. Isolated technological innovations can also worsen morbidity and mortality levels related to some causes, It also seems indisputable that income growth and rising living standards per se do not cause mortality levels or infant mortality in particular to decline. Mortality patterns in Western, Eastern and Southern European countries, in the US, or in any LDC’ s for that matter, are all peculiar, as far as the nature of the mortality revolution is concerned, for the timing and social- historical forces at play were also contrasting. Likewise, in Latin America there was no unique pattern. Palloni identifies four different paths or groups according to the starting levels and the nature of the decline before 1940. Brazil would be part of the last group to admit major gains in mortality levels, even though the country’ s path was also dissimilar in its category since mortality gains accrued even before 1940. n* Alberto Palloni, Ibid., 627. Alberto Palloni, Health Levels and Care m Latin America: The Case of Infant Mortality 1900-1985”, Healdi Transition, Cluqjter 10, Vbl.1. (Camberra: The National Library of Australia, 1989) 97 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The veiy way infant mortality was perceived as a poh(gr issue changed over time. Infant mortality in the late nineteenth centuiy and early twentieth century was seen less as a medical problem and more as a reason for social concern. This situation would change in the 1930’ s and 1940's when public policy to ameliorate child survival began to be more directed to implementing technical innovation than to changing environmental lislqr factors. Somewhat the responsibility for decreasing infant mortality shifted from the public health field to the medical field, "s Some of the inquiries that need to be addressed as far as this study is concerned are: 1 ) Is this discussion on the role of endogenous and exogenous changes on mortality decline transferable in to to or germane for understanding the specific historical evolution of Brazil’ s decline in infant mortality? 2) What is the approximate relative impact of wealth and income growth at the household level vis a vis other measures of ‘so cia l p o lic y-te ch n ica l changé" such as access to pre-natal care, breastfeeding patterns, sanitation levels or educational attainment of the mother. Usually infant mortality* decline patterns do resemble and determine the overall mortality trends in Brazil b u t this need not to be always true, n ® See Xue Cheo, “ Ihfimt Mortality and its Oetennmants m Los Angeles County, 1989-1991”, PJiJ3. dissertatian, U S C , 1997. See Alberto PaUoni, 1991, Ibid 98 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Furthermore, a pressing need exists for an integrated epistemologjr of the causes of infant survwal, b^ond the ambivalence of exogenous vs. endogenous causes. Kuznets admonishes that the distinction between economic development and public health creates a false dichotomy. Moreover, the very differentiation of relevant variables as affected or being the by-product of the social-economic development as opposed to what Preston calls * so cia l p o licy-tech n ica l changesT seems somewhat arbitrary and peripheral. Notes on the Timiny and Interconnectedness of Brazil's Fast, Dramatic and Dynamic Fertility and Infant Mortality Decline Departing firom EZasterlin's conceptual framework on the specific nature of the m o rta lity revo lu tio n and fe r tility tra n sitio n , it is convenient to briefly scrutinize the meaning and association of these terms. Simon Kumets, “ Population Trends and Modem Econmnic Growth: Notes Towards a Ifistorical Perspective”, Population Debate: Dimensions and Perspectives. (New Yoric World Population Council, 1975). Also mentioned in Samuel Preston, Ibid, 1980, 291. 99 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Camarano notes that mortality and the interactive effect of fertility and mortality would receive relatively less attention than the fertility decline itself: “ demographic transition became a matter of how to moderate fertility.* The World Bank exhorts that “ no country has completed the transition to low fertility without parallel declines in infant and child m ortali^ rates. * 120 The three phases concept embedded in the dem ographic tra n sitio n ’ s descriptive model imply differential mortality and fertility rates in the intermediate phase and stabilization in the third phase. They also entail the notion of an onset and a completion. Easterlin emphasizes that rapid population growth in the second phase of the demographic transition is a transient moment and that fertility levels are falling throughout the world due to a shift not only on the level but also on the very nature of fertility patterns. This scholar stresses the impact of m o rta lity revo lu tio n - or of falling child and infant mortality- on fertility through the increasing interest to control birth using contraceptive means. "* Camaiano, Ibid. Van de Walle observes diat the POPLINE records betwem 1974 and 1992 have shown a progressive increase m the use of the term fertility transition as opposed to the term demographic transition. Fertility transition is more specific a tennmology since it ùnplies a shift fiom natural to coitrolled fertility. Henry defines ‘ natural fe rtility ' as die absence of parity specific control and ‘ controlled fe rtility ’ as the purposive stoppmg o f diildbearmg conditional upon a couple’s previous fertility history. ™ World Bank, “Population and Development : hrgilications fer the World Bank”. (Washmgton DC: 1994). 100 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The impact of infant and child m o r ta lity o n fertility depends on biological effects through breastfeeding but also on behavioral effects through the usage of contraception, 121 Easterlin & Crimmins, th ro u ^ their supply-demand theory of fertility determination, show how essential child survival is in determining the use of birth control. 1 2 2 In an empirical examination of the historical decline of infant mortality in Latin America between 1900 and 1985, Palloni posits that * (in Latin America) as in Western Europe, there can be no question that changes in infant mortality precede rather than follow changes in fertility.* 1 2 3 This author, cautions, however, that the relationships are dramatically different in the sense that while in Western Europe the transition to levels of mortality below 150 per thousand live births occurred under fertility levels of around .27. most Latin American countries begin in earnest the decline of infant mortality with fertility values of a t least .35. Notwithstanding that there are contrasting views about the timing of the onset of mortality decline in Brazil- with some authors such as Mortara 124 and For a thorough analysis see: Susan Cochrane & K C Zachariah, Worid Bank Staff Working Paper # 556, ‘ Tnfent and Child Mortality as Determinant of Fertility: The Policy hiq)lications”. (Washington DC: World Bank, 1992) Ridiard Easterlin & Eileen Crimmins, “ The Fertility Revolutian: A Siqiply -Demand Analysis". (Chicago: University of Chicago Press, 1985). Alberto PaUoni, “ Health Levels and Care m Latm America: The Case o f hifent Mortality 1900-1985”, Health Transition, Chapter 10, V0I.I. (Camberra: The National Library o f Australia, 1989) Gioigio Mortara, Contribuicoes para o Estudo de Dempgrafia no Brasil”. IBGE, 1970. 101 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Carvalho izs seeing notable gains in life ^cpectancy even before the 1930’ s and others such as Arriaga seeing dramatic changes only after 1930’ s - there is a consensus that infant m ortally decline was only greatly advanced in Brazil after W.W.II once new medical technologies and public health practices were intensely introduced and developed. The onset of the mortality revolution in Brazil could be identified with this post-W.W.n period. As of the mid 1940’ s, mortally rates began to fall dramatically in the most urban, industrialized and developed areas of Brazil’ s) five regions ( primarily the South and Southeast regions ) and across different social classes and races propelled by major improvements in infant mortality rates. Greatly determined by infant mortali^ improvements in the 1970’ s, the dramatic decline in birth and fertility rates as of the 1980’ s, also helped to nourish a further a acceleration in infant mortality decline during the same period. According to recent studies and to the latest 1996 DHS survey, the decline in fertility affected positively Brazil’ s infant and child mortally rates mostly through three factors: parturition decline, greater relative reduction of births in riskier (35+) age groups and increase in the interval between births. Jose Alberto de Carvalho, “ Analysis of Regional Trends in Fertility, Mortality and Migration in Brazil - 1950-1970”, Mi.D. Dissertation, London School ofEconmrdcs, 1973. ™ Eduardo Arriaga, New Lifo Tables for Latin American Population m the Nineteenth and Twentieth Centuries, Ibid. 102 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Facts on Mortality Revolution and Infant Mortality Decline in Brazil and its Northeast Region. Brazil's Life E^ectancv at Birth The decline of infant mortality in Brazil is directly related to the historical evolution of mortality, measured by the improvements in life expectancy. Mortality rates have been falling in Brazil ever since the beginning of the century. This process began slowly with moderate and gradual improvements in life expectancy, (tables 38 and 39) The changes in mortality patterns between 1940 and 1970 were sensational. Infant mortality decline, as well as other factors such as thriving urbanization, changes in women’ s education, employment and contraceptive patterns, caused shifts in the preference maps for children, imposmg a dramatic drop in fertility levels in the 1980 s and 1990’ s. Driven by the robust improvement in child survival, the fertilily transition, by its turn, had a reinforcâig impact on the level as well on the relative contribution of infant mortality to gains in life e^ectancy in the last two decades. Demographic data in Brazil are fiagmented, contentious and subject to adjustments and corrections. Often even the ofQcial IBGB figures are revised. As far as life expectancy at birth is concerned, there are conflicting values for the pre-1940’ s period as well as for the 1970’ s. 103 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Estimation of mortality levels in Brazil before 1940 is sketchy at best. The estimates adopted here for the period panning from the late nineteenth century (1870) until the 1940’ s are the ones developed by Mortara 1 2 7, de Carvalho 12» and Arriaga 1 2 9, (table 38) Although Mortara’ s estimates are higher for the early twentieth century and do not change as drastically as Arriaga’ s in the 1930’ s and 1940 s, they will match Arriaga’ s in 1950. Arriaga’ s mortality estimates for Brazil before the 1940’ s are lower than Mortara’ s and de Carvalho’ s, implying that mortality gains and, more specifically, life expectancy, improved more dramatically only once the medical innovations and health measures were adopted in the late 1930’ s, early 1940’ s. Carvalho’ s and Mortara’ s also detect such substantial mortality decline in the 1940’ s but not as strongly as Arriaga’ s, (table 39) Arriaga estimates that the life span of the average Brazilian at the turn of this century would be under 30 years (29.4). Mortara estimates that Brazil’ s life eiq>ectancy at birth between 1900 and 1920 would be 39.3 and that between 1940 and 1950, Brazil’ s life expectancy at birth would be 43.7. Giorgio Mortara. “ Estudos sobre a Otilizacao do Censo Demografico para a Reconstrucao de Estatisticas do Movùnento da Populacao do Brasil”, VI. Sinopse da Dmamica da Populacao do Brasil nos Ultimos Cem Anos, Revista Brasileita de Estatistica. (Rio de Janeiro, 1941). Jose Alberto de Carvalho, “ Fecundidade e Mortalidade no Brasil ”, Ford Foundation Research Rq>ort. (Belo Horizonte, 1977). Eduardo Arriaga, “ New Life Tables for Latin American Population in the Nmeteenth and Twentieth Centuries. (Berkeley; hastitute for Inteinational Studies, 1968). 104 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. De Carvalho does not have estimates for before 1940, but he estimates life expectancy to be 41.2 in 1940 and 43.6 in 1950- The first Brazilian Census was carried out in 1872 but life expectancy reliable estimates were calculated by the Brazilian Institute of Geography and Statistics (IB G E) only as of the 1930’ s census. These official estimates collected directly fi-om the decade demographic censuses have been revised constantly particularly for the 1930'-1950’ s period but also for the 1970’ s. According to the IBGE, Brazil’ s life expectancy at birth increased firom a level of 41.2 in 1930 to 42.8 in 1940, reaching 45.9 in 1950. The rate of average gain in life expectancy for the decade doubled from 7% in the 1940’ s to almost 14% in the 1950’ s (table 3.3.). As a result, life expectancy reached 52.3 in 1960, according to the official data. In the 1960 s the gains in life expectancy for Brazil as a whole were insignificant. In contrast, in the 1970’ s such gains were extensive, with the average life of a Brazilian growing in length from 52.7 to 62 years, or 18%. The improvement in life expectancy continues in the 1980’ s and 1990’ s but at a lower tempo. From 1930 to the latest IBGE estimate (1996) of Brazil’ s life expectancy at birth increased 64%, reaching the current level of 67.6. Thomas Merrick and Douglas Graham, “ Population and Econcxnic Development in Brazil- 1800 to the Present Baltimore: The John Hopkms U nivers^ Press, 1979). Pg 42. 105 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The h isto rica l evolution, o f B razil’ s cru d e d eath rates in d icate a sim flar pattern of mortality decline. According to de Carvalho & Wood Brazil’ s crude death rate decreased 20% between the periods of 1920-1940 and 1940- 1950. Between the periods of 1940-1950 and 1950-1960, the crude death rate would have fallen an additional 40%. Similarly, the IBGE estimates that Brazil’ s crude death rate fell almost 30% in the 1950’ s. (table 37) Table 37 1872-1890 30. 1890-1900 28. 1900-1920 26. 1900-1905 24.6 1915-1920 22. 1920-1940 25. 1940-1950 20. 1950-1955 16.7 1950-1960 12. 1965-1970 9.5 Evolution of Brazil's Crude Death Rate (per 1,000 inh.) 1872-1990 196311.16 1964 10.48 1965 10.23 1966 9.87 19679.52 1968 9.44 1969 9.18 1970 10.12 1971 10.33 19729.96 1973 9.5 Change in Crude Death R ate( 1990-1941): * Values estim ated by Armin Ludwig. Remaining values collected from various IBGE's Anuarios Estatisticos do Brasil. * * Values estim ated by Jo sé Magno de Carvalho & C harles Wood____________ 1941 19.25 195213.92 194218.96 1953 14.14 1943 18.54 195414.6 1944 19.29 1955 13.26 1945 18.46 1956 14.42 1946 18.15 1957 14.79 1947 18.14 1958 14.2 1948 17.36 1959 12.36 1949 17.02 196011.36 1950 14.45 1961 10.96 1951 15.21 196210.78 1974 9-41 1985 7.72 1975 9.4 1986 7.61 1976 9.26 1987 7.51 1977 9.25 1988 7.42 1978 9.15 1989 7.34 1979 9.1 1990 7.27 1980 8.87 1981 8.65 19828.41 1983 8.17 1984 7.94 -62.2% 131 106 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. We will also consider the estimates computed by Simoes for the 1940- 1990 based on IBGE data as well as on his own research. Simoes’ estimates for the 1940’ s and 1950’ s are slightly higher than the ones projected by Arriaga or the ofScial IBGE values and are somewhat closer to the ones by Mortara and de Carvalho. This author also has a higher estimate for H fe expectancy in 1970 (55.2 and not 52.7), which implies non residual gains for the 1960’ s and lower overall gains for the 1970’ s - 8% and 13%, respectively, (table 3.3.) For the 1980’ s and 1990’ s there is agreement between Simoes’ and the official estimates by the Brazilian Institute of Geography and Statistics. This study regards Arriaga’ s estimates for the 1870-1920 period to be quite plausible. However, firom 1920 to 1940, we find his estimates to be too low. Arriaga claims that mortality gains were more clustered in the 1930’ s as opposed to more homogeneously distributed in the 1920’ s and 1930’ s. Taking into consideration the data, the alternative estimates and the historical context, we reckon that firom a level a little higher than 30 in 1920, life expectancy would have gradually improved to a level slightly over 40 in the early 1930’ s. This level would be compatible with the levels proposed by both de Carvalho and the IBGE for the mid 1930’ s: around 42. Celso Simoes is one of the foremost researdiers on mortality in Brazil and is a donographer working for IBGE’s Research Department. 107 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Hence, life expectancgr would have increased around 30% between the 1920’ s and the 1930’ s, most likely as a result of both economic and organizational changes in society as well as of the introduction of health and sanitary innovations during this period. However dramatic the pre-W.W. H decline in mortality levels was, it took place in the absence of harmonic socioeconomic development, In the 1940’ s life expectancy would have increased slowly, reaching a level of around 45 in 1950. This estimate is compatible with Mortara’ s, Arriaga’ s and de Carvalho’ s levels of 43.7, 43.0 and 43.6 respectively for 1950. This mid-century level of 45 would be ju st a little below the IBGE’ s 45.9 estimate and Simoes’ level of 47.5 for the same year. In the 1950’ s, however, mortality decline accelerates in Brazil. IBGE’ s ofGcial estimates seem to be quite reasonable indicating that life expectancy at birth would be around 52 in 1960. In contrast, mortality decline certainly diminished in the 1960 s. If one accepts the IBGE estimate of 52.7 for 1970 it w iU be evident that there were no gains at all in the 1960’ s. Simoes’ revised estimate of 55 seems to be more plausible indicating th at life expectancy gains were small but non residual in the decade. At any rate, the 1970’ s were characterized by extraordinary gains in mortality in Brazil. Life expectancy would reach 62 in 1980, a gain of 13% in comparison with 1970 (or of 18% if the official data are taken into consideration). As of the early 1980’ s the gains See Alberto Palloni, “ Mortality in Latm America: Emergmg Patterns”, Ibid. 108 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. continue but at a lower pace. Brazil’ s life e x p e c ta n c y at birth in 1 9 9 6 was 67.6. In short, life expectancy began to improve in the 1920’ s and 1930’ s, growing slowly but steadily in the 1940’ s, 1960’ s and 1980’ s and much more vigorously in the 1950’ s and 1970’ s. In the 1990’ s Brazil’ s overall life expectancy at birth seems to be advancing at a more limited tempo. Table 38 The Evolution of Brazil's life Expectancy at Birth according to Main Estimates, 1870-1996 Arriaga Mortara Carvalho Sim oes IBGE (official Years 1870 27.4 33.9 estim ates) 1880 27.6 33.9 - - - 1890 27.8 39.3 - - - 1900 29.4 39.3 - - - 1910 30.6 39.3 - - - 1920 32 - - - - 1930 34 - - - 41.2 1940 36.7 43.7 41.2 44.9 42.8 1950 43 43.7 43.6 47.3 45.9 1960 55.5 - 49.6 51.2 52.3 1970 - - 53.4 55.2 52.7 1980 - - - 62.3 62 1991 * * - - - 65.6 65.5 1996*** - - - - 67.6 * Mortara: 1870-1890= 33.9; 1890-1920= 39.3; 1940-1950= 43.7 * * The decade census for 1990 was exceptionally taken in 1991. According to th e 1996 PNAD. Life Expectancy: The average numt)er of years newtx>m children would live if sutyect to mortality risks prevalent for the population a t each ag e at the tim e of their txrth. Sources: Sim oes (1 9 9 ^, d e Carvalho (1977), Arriaga (1968), Mortara (1941) and IBGE (Demographic Censuses). 109 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 39 Absolute and Relative (% ) Life Ebq)ectancy Gains by Decade, 1930-1991 Decade 1930's Arriaga Carvalho Sim oes IBGE* Years 2.7 - - 1.6 % 1940*3 7.90% - - 3.90% Years 6.3 2-4 2-4 3.1 % 1950*3 17.20% 5.80% 5.30% 7.20% Years 12.5 6 3.9 6.4 % 1960*3 27.90% 13.80% 8.20% 13.90% Years - 3.8 4 0.4 % 1970*3 - 7.70% 7.80% 0.10% Years - - 7.1 9.3 % 1980*3 - - 12.90% 17.60% Years - - 3.6 3.5 % - - 5.80% 5.60% Note: Mortara's change com pares the average for the 1890-1920 period (39.3) with the average for the 1940-1950 period (43.7) the 1940-1950 period (43.7). an estim ated gain of 4.4 years or 11.2%._______________ Mortality levels in Latin America and in Brazil in particular were still very high in the first decades of this century. In 1920, while life expectancy at birth in the leading mortality revolution countries was around 58, in Latin America as a whole the average person's life span was 31.1 years or 53.6% lower. 1 3 4 ^ Luis Rosero & H. Camaano, ‘ Tablas de ^ d a de Costa Rica, 1900-1980", Associacicn Demografica Costarricense. ( San Jose, 1984) 110 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. As of the 1920’ s and 1930’ s and particularly as of the 1950’ s, mortality decline accelerates robustly in Brazil. The improvement in life expectanqr in Brazil from the early 1950’ s to the early 1990’ s have been over 50%. (table 40) Table 40 Evolution of Life Ejqxectancy at Birth in Selected Covmtxies- 1953-1993 (in years) 1953 1960 1975 1993 (1953/93) Years* Sweden 69.3 73. 75.2 77.7 12-1% 8.4 U.S. 65.9 70.3 72. 75.9 15.2% 10. Argentina 59.1 65. 68.4 71.1 20.3% 12. Portugal 54.4 63.3 70.1 74. 36.1% 20.4 Ttiailand 50.3 60.4 66. 68.7 36.6% 18.4 Brazil 43.7 55. 61.2 65.8 50.1% 22.1 Source: Adapted from U N. Annuaire Démographique yeartxx)ks. ( Brazil's estim ates tiave not been revised.)___________________ Northeast’ s Life Rypectancv at Birth The trajectoiy of mortality rates and more specifically of life expectancy in the NorÜieast region is rather different than the trends in other regions of Brazil. The level of life expectancy reached in the South region in 1930 (50.1) would only be achieved in the Northeast in 1970 (51.6). The provision of health services, reforms in infirastructure and sanitation as well as socio-economic development would only be expanded in the Northeast as of the 1960’ s. I l l Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The difTerential between Brazil’ s life expectancy and the Northeast’ s which was around 4 years in the 1930’ s and 1940’ s actually doubled to 8 years in the 1950’ s, indicating that the onset of mortality revolution happened later in this lagging region, (table 41) As of the 1960’ s, however, life e3q>ectancy in the Northeast has been iacreasing significantly. From a level of 44.3 in 1960 it improved 16% to 51.6 in 1970. In the following two decades life expectancy at birth rose an additional 14 % and 9% in the Northeast. As a result of these enormous gains, the differential between the national average and Northeastern life expectancy fell fro m 18% in 1960 to 2% in 1991 (or from a difference of 8 to 1.3 years), Table 41 Life Expectancy at Birth by Great Regions, 1940-1991 (in years) 1930 1940 1950 1960 1970 1980 1991 1996 Brazil 41.2 42.8 45.9 52.3 52.7 62. 65.5 67.6 South 50.1 50.1 53.3 60.3 60.2 65.3 69. 70.2 Southeast 43.5 44. 48.9 57. 57. 64.5 67.5 68.8 Central- W est 46.9 48.2 51. 56.4 56. 63.4 67.8 68.5 Northeast 36.7 38.7 43.5 44.3 51.6 58.7 64.2 64.5 North 39.8 40.4 44.2 52.3 52.7 61.3 67.3 67.4 Source: IBGE Demographic C ensuses 135 Accordmg to the IBGE’s ofBcial infant mortality estimates. 1 1 2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Using estimates slightly different than IBGE's, Simoes investigated the evolution of life e^ectancy in the Northeast and in Brazil hom 1940 to 1990. In addition to the conspicuous recent gains in life expectancy and the shortening of regional differences, it is interesting to note how the gender differential in life expectancy has remained at a level of 6 /7 years for the past three decades both in Brazil as a whole as in the Northeast, (table 42) Table 42 Life Expectancy at Birth- Brazil and its Northeast Region- 1940-1990 Years eo Gains eo Gains eo Gains DUTerence B ySei Men Women Total Brazil 1940 42,75 2,10 47,15 2,67 44,89 238 4,41 1950 44,85 4,31 49,82 4,98 4737 4,64 4,97 1960 49,16 3,23 54,80 3,31 5131 337 5,64 1970 52,39 6,82 58,10 7,47 55,18 7,13 5,72 1980 59,20 2,93 65,57 4J9 6231 339 637 1990 62,13 69,86 6530 7,73 Nordeste 1940 1,80 42,49 1,71 38,87 2,72 6,58 1950 371 2,30 44,20 233 41,58 2,01 6,49 1960 4 5,35 46,53 5,00 43,59 539 632 1970 45,35 8,60 51,53 833 48,88 835 6,17 1980 53,95 6,01 60,46 6,61 57,14 6,04 6,51 1990 59,96 67,07 63,17 7,11 Source: Adapted fn m Cebu Sbnoes & 113 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Population Growth Population growth increased substantially in Brazil after the WWn, which is compatible with the advent of the m o rta lity r e v o lu tio n . In most LDC’ s, the rate of population growth in between the emergence of a progressive decline in mortality rates and the fe r tility d eclin e has been estimated to be around 2- 2.5%a.a. According to Maddison, Brazil’ s population growth from 1950 to 1973 was 2.9 %. This author estimates the long term annual rate of population growth until the mid twentieth century (1870-1950) to be 2.1% a.a., (table 43) In a like manner Furtado calculates the average annual population growth between 1920 and 1937 to be 2.0% a.a., increasing to 2.2% a.a. between 1937-1947 and reaching 2.4% a.a. between 1947-1957. (table 44) Table 43 Brazil's and Latin America's Average Rate of Growth of Population -1900-1987 (a.a. % ) Rate of Population Growth Latin America Average 1900-1913 2.1 1.9 1913-1950 2.1 1.8 1950-1973 2.9 2.6 1973-1987 2.5 2.2 Source: M addison ( 1991) * Including Brazil. Angus Maddison, “Dynamic Forces m Capftalist Development”. (New Yoric Oxford Itaiversity Press, 1991). 137 Celso Furtado, “The Economic Growth of Brazil”, Ibid, 260. 114 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. T able 4 4 Brazil's Average Rate of Annual Growth of Population -1920-1957 (% ) 1920-1929 2 1929-1937 2 1937-1947 2.2 1947-1957 2.4 Source: Celso Furtado ( 1959) An examination of the ofGcial data by the Brazilian Institute of Geography and Statistics (IBGE), shows that, if population growth was already high in the 1940’ s -2.4%a.a. - it certainly increased to a much h i^ e r level in the 1950’ s and 1960’ s - 3.0% a.a. Using data by Mortara Merrick and Graham observe that, while in the US early declines in mortality b ro u ^ t rapid natural increase in the Grst half of the nineteenth century, in Brazil this process would only take place in the mid twentieth century. The average annual rate of population growth would have increased firom 1.85% in the 1850-1900 period to 2.12% between 1900 and 1950. Between 1950 and 1970, Brazil’ s population growth leaped to 2.92% a.a., according to Mortara’ s estimates. Gioigio Mortara, IbicL Thomas Merrick & Douglas Graham, Ibid., 32. 115 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The population of Brazil, and particular^ the urban population, increased at a very fast pace after W.W.II. (tables 3.10 - 3.15.) It is known, however, that this process of population growth, propelled in great part by decreasing infant and child mortality rates, is a transient state and that, in less developing countries, a generalized fertility decline is likely to accrue within a period of no longer than about a generation firom the onset of the mortality revolution. In the 1970’ s the rate of population growth was again back down to the pre-WWn levels of 2.4%a.a., falling to an annual rise average of almost 2% until the late 1980’ s. Brazil’ s population growth has decreased been rather dramatically in the past two decades under the impact of an extraordinary decline in both fertility and infant mortality rates. Recent data released by the IBGE indicate that Brazil’ s population is presently (91-96) growing at a rate of only 1.3% a.a, or almost half the rate of two decades ago, or at one of the country’ s lowest rates of population growth ever, (tables 3.9. to 3.11.). Boosted by rapid decline in both fertility and mortality levels, the current rate of population growth in the Northeast, is ven lower, or 1.1% a. a. As a result of these demographic changes, the Brazilian population is growing at a much lower tempo than estimated a few years ago: in 1980, demographers estimated that Brazil’ s population would be 192 million in 1997; the actual total population in 1997 was 157 million, or 35 million people less For a thorough analysis see Easterlin, Ibid., 1996. 116 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. than what was expected under an already rapid decline in the growth of population. While the latest population census (1991) showed that 41% of Brazil's population was young (0 to 17), this proportion has been decreasing at a very dramatic cadence. The median age increased firom 21.7 in 1991 to 23.2 in 1996. Brazilians over 65 made up 4.8% of the population in 1991, only five years later they make up 5.5%. This trend will have in the future profound implications for health, social security, welfare and the setting of public policy in the coming years. If this rate is to be kept, some demographers estimate that within 25 years Brazil wül have the World’ s fifth largest elderly population. A look at Brazil’ s 5 year age pyramid for 1995 reveals an abrupt narrowing of the base as of 1985 or a proportionally smaller 0-4 and 5-9 age groups. Table 45 Brazil's Mean Geometric Rate of Annual Population Growth 1 87 2 /1 9 9 6 (% ) Census Period Brazil 18 7 2 /1 8 9 0 2.06 18 9 0 /1 9 0 0 2-42 19 0 0 /1 9 2 0 2.09 1 9 20 /1 94 0 2.04 1 9 40 /1 95 0 2.39 1 9 50 /1 96 0 2.99 1 9 60 /1 97 0 2.89 1 9 7 0 /1 9 8 0 2.48 1 9 8 0 /1 9 9 1 1 .9 3 1 99 1/19 96 1 .3 8 Source: Derived from C ensus Tabulation (1994,1982,1975,1960,1950) 117 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. T able 4 6 Regions Total Population General Characteristics - 1996 Population Urbanization Population Age Growth Rate (%) S ex Ratio Dependency ( a.a. %) Ratio Brazil 157 079 573 1,4 78,4 97,3 58,7 North* 11290 573 2,4 62,4 102,9 73,7 Northeast 44 768 201 1,1 65,2 95,8 69,6 Southeast 67 003 069 1,4 89,3 96,5 52,0 South 23 516 730 1,2 77,2 98,2 54,2 Central West 10 501 480 2,2 84,4 100,2 55,7 Data sources; IBGE/ContagemNacîonal de Populaçâo - 1996 * Excludes data for the rural area o f the states o f Rondônia, Acre, Amazonas, Roraima, Para and Amapa. Notes m Definitions ; Average annual growdi rate o f the population: average annual increment o f the population, measured by the expression i= n sQ root { P(t-Hi) / P(t) }, P(t+n) and P(t) bemg populations corresponding to two successive dates, and n die tune interval between these dates, measured in years and fiactions of year. Urbanization rate: proportion of die urban population to the total population. Populatim sex ratio: ratio of women to men in a population. Age dqiendmcy ratio - calculated as the ratio o f dqiendents - die population under age 15 and above age 65 -to the working-age population - those aged 15-64. 118 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. T able 4 7 Mean Rates of Population Growth of Brazü for Selected Periods, 1799-1996 (a.a. % ) Period % Growth Source 1799 1.1 Richard Morse 1808 1.8 Richard Morse 1830 1.8 Richard Morse 1854 1.5 Richard Morse 1872 1.5 Richard Morse 1890 2 Richard Morse 1900 1.9 Richard Morse 1920 2.9 Richard Morse 1808-1890 1.9 IBGE Resumo Historico 1890-1900 2.4 IBGE Brasil em Numéros 1900-1920 2.1 IBGE Brasil em Numéros 1920-1940 2 IBGE Brasil em Numéros 1890-1950 2.2 IBGE Resumo Historico 1940-1950 2.4 Derived from IBGE Census 1950-1960 3 Derived from IBGE Census 1960-1970 2.9 Derived from IBGE Census 1950-1970 3 IBGE Resum o Historico 1900-1913 2.1 Angus Maddison 1913-1950 2.1 Angus Maddison 1950-1973 2.9 Angus Maddison 1973-1987 2.5 Angus Maddison 1970-1980 2.4 Derived from IBGE Census 1980-1991 1.9 Derived from IBGE Census 1800-1900 1.7 Merrick & Graham 1900-1970 2.4 Merrick & Graham 1960-1970 2.8 Merrick & Graham 1991-1996 1.4 Derived from IBGE PNAD 119 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. T able 4 8 Brazil's Population- 1872-1999 (miUions) 08-1872 9.93 1916 25.2 Sep-60 70.2 1873 10.2 1917 25.8 1961 71.9 1874 10.5 1918 26.2 1962 74.1 1875 10.7 1919 26.8 1963 76.5 1876 10.9 Sep-20 27.4 1964 78.7 187711.1 1921 27.8 1965 81. 187811.4 1922 28.5 1966 82.9 1879 11.5 1923 29.1 1967 85.2 1880 11.8 192429.7 1968 87.6 1881 12. 1925 30.3 1969 90. 1882 12.2 1926 30.9 Sep-70 93.1 1883 12.4 1927 31.6 1971 95.1 1884 12.7 1928 32.2 1972 97.8 1885 13. 1929 32.9 1973 99.9 1886 13.1 1930 33.6 1974 102.4 1887 13.4 1931 34.2 1975 104.9 1888 13.7 1932 35. 1976 107.5 1889 14. 1933 35.7 1977 110.2 12-1890 14.4 1934 37. 1978 112.9 1891 14.5 1935 38. 1979 115.7 1892 14.9 1936 37.9 Sep-80 119. 1893 15.2 1937 38.6 1981 124. 1894 15.6 1938 39.4 1982 126.8 1895 16. 1939 40.2 1983 129.6 1896 16.3 Sep-40 41.2 1984132.6 1897 16.7 1941 42.1 1985 135.5 1898 17.1 1942 43.1 1986 138.5 1899 17.3 194344. 1987141.4 DeoOO 17.5 1944 45.1 1988 143.5 1901 18.3 1945 46.2 1989 144.6 1902 18.8 1946 47.3 1990 145.9 1903 19.2 1947 48.4 Sep-91 146.9 1904 19.5 1948 49.6 1992151.2 1905 20. 1949 50.1 1993 152.5 1906 20.4 Jan-50 52. 1994 153.7 1907 20.9 1951 53.7 1995 155.1 1908 21.3 1952 55.1 1996 157.1 1909 21.8 1953 56.8 1997 1910 22.2 1954 58.4 1998 1911 22.7 1955 60.1 1999 1912 23.2 1956 62. 1913 23.7 195764. 1914 24.1 1958 65.8 1915 24.7 1959 68. Source: Collected from Demographic C ensuses and Anuânos Estatisticos do Brasil (1872-1998). 1 20 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. T able 49 U rb a n and Rural Population of Brazil b y Major Regions - 1 9 4 0 -1 9 9 6 (% Total Pop.) 1940 1950 Region %in Brazil Urban % Rural % Region %in Brazil Uiban% Rural % Pop.(in Pop. million) Brazil 41.2 31.2 68.8 52. 36.2 63.8 South 5.7 13.8 28. 72. 7.8 15.1 29.5 70.5 Southeast 18.3 44.4 39.3 60.7 22.7 43.6 47.1 52.9 Central- 1.26 3.1 21.4 78.6 1.72 3.3 24.4 75.6 W est Northeast 14.4 35. 22.9 77.1 17.9 34.4 26.2 73.8 North 1.54 3.7 26. 74. 1.88 3.6 30.8 69.2 1950 1960 Total Pop %in Brazil Urt)an% Rural % Total Pop. %in Brazil Urban % Rural % Brazil 52. 36.2 63.8 70.2 44.9 55.1 South 7.8 15.1 29.5 70.5 11.8 16.8 37.3 62.7 Southeast 22.7 43.6 47.1 52.9 30.7 43.8 57. 23. Central- 1.72 3.3 24.4 75.6 2.95 4.2 33.8 66.2 W est Northeast 17.9 34.4 26.2 73.8 22.2 31.6 33.7 66.3 North 1.88 3.6 30.8 69.2 2.55 3.6 37.6 62.4 1970 1980 Total Pop. %in Brazil Urban % Rural % Total Pop. %in Brazil Urban % Rural % Brazil 93.1 55.9 44.1 119. 67.6 32.4 South 16.5 17.7 44.2 55.8 19.1 16. 62.3 37.7 Southeast 39.9 42.8 72.7 27.3 51.8 43.6 82.6 17.4 Central- 5. 5.4 48. 52. 7.5 6.3 68. 44. W est Northeast 28.1 30.2 42. 58. 34.8 29.2 50.5 49.5 North 3.6 3.9 45.2 54.8 5.8 4.9 51.7 48.3 1991 1996 South 22.1 15. 74.2 25.8 23.5 15. 77.2 22.8 Southeast 62.7 42.7 87.8 12.2 67.0 42.6 89.3 10.7 Central- 9.4 6.4 80.1 19.9 10.5 6.7 84.4 15.6 W est Northeast 42.4 28.9 60.6 39.4 44.7 28.5 65.2 34.8 North 10.3 7. 57.2 42.8 11.3 7.2 62.4 37.6 Source: IBGE's Demographic Censuses. * IBGE's official definition of an urban center relates to the localization of dwellings and people in judicially independent cities, towns o r villages according to municipal law. * An urban center is often defined a s a city (municipal capital), village (district capital) or an isolated urban agglomeration. 1 2 1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 5 0 M ea n Geometric Rate of Annual Population Growth by Major Regions -Rural & Urban- 1 9 4 0 /1 9 9 1 (%) 1940/1950 1950/1960 Total Urt)an Rural Total Urban Rural Brazil 2.39 3.91 1.6 2.99 5.15 1.55 North 2.29 3.71 1.84 3.34 5.04 2.37 Northeast 2.27 3.51 1.84 2.08 4.63 1.02 Southeast 2.14 4.08 .64 3.06 4.91 1.06 South 3.25 3.88 2.97 4.07 6.44 2.9 Center- 3.41 4.65 2.94 5.36 8.9 3.89 W est 1960/1970 1970/1980 Total Urt)an Rural Total Urban Rural Brazil 2.89 5.22 .57 2.48 4.44 .62 North 3.47 5.44 2.11 5.02 6.44 3.7 Northeast 2.4 4.57 1.1 2.16 4.1 .53 Southeast 2.67 5.19 1.88 2.64 3.99 2. South 3.45 5.29 2.2 1.44 4.98 2.48 Center- 5.6 9.94 6.14 4.05 7.69 .81 W est 1980/1991 Total Urban Rural Brazil 1.93 2.96 .67 North 3.85 5.37 2.04 Northeast 1.83 3.34 -.28 Southeast 1.77 2.32 -1.52 South 1.38 2.97 -2.01 Center- 3.01 4.3 -1 .0 6 W est SourcezlBGE (1992.1982) 1 2 2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. T able 51 Mean Geometric Rate of Annual Population Growth by Major Regions - 1890-1940 (% ) B raz il 2.15 N o rth 2.29 N ortheast 1 .97 Southeast 1 .64 South 3.16 C entral- 2.78 West Source: D eriv ed fro m Census T abulation (1940,1921) Birth Rates and Total Fertility Rate Recent studies in Brazil have emphasized that between 1940 and 1970 fertflity and birth rates were already declining in Brazil, but at a very slow pace and concentrated regionally as well as socially and that, only as of the 1980’ s, such rates would have fallen more dramatically as well as encompassing all regions and social groups. The data seem to confirm this claim. Fertility transition began in the 1980’ s and it is advancing in Brazil. In the Northeast the decline in fertility, also initiated in the 1980’ s, accelerated in the 1990’ s. Until the mid 1970’ s, Brazil’ s overall birth and fertility rates were still quite high. As of the late 1970’ s and particularly as of the 1980’ s, however, the decline in both birth and fertility rates have been extremely significant. Brazil’ s crude average birth rate descended from around 35 in 1970 to an estimated 21 in 1996, a fall twice as rapid as in most developed countries, (table 53) E)ven more steadfast has been the decline in die TFR: firom levels around 6.0 in the early 1960’ s to 4.3 in 1980, to 2.7 in 1990. (table 54). 123 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Fertilily decline has been facilitated by the generalization of modem usage of contraception, even though the government of Brazil has never had a defined policy on birth control. According to the 1996 DHS Survey, almost 77% of the Brazilian women at reproductive age living in union use a modem method of birth control. Interestmgly the most popular method by far is the female sterilization: over 40% of all women age 15-49 would be sterilized. In the Northeast this rate would be close to 44%. (table 52) There is still a great variability in the levels of fertility in Brazil. Total fertility rates are much higher in rural areas, in la^png regions such as the Northeast and among specific ethnic and socio-economic groups. Nevertheless, on average, the decline in fertility has been much more converging and homogeneously distributed across Brazil’ s 5 regions than the decline in infant mortality. From 1940 to 1996 fertility levels dropped 63% in Brazil as a whole and 59% in the Northeast. In 1991 TFR for Brazil and the NE were, respectively, 2.7 and 4.0. Five years later the fertility differential had fallen to 0.6 year, with. Northeastem fertility levels collapsing almost 30% in half a decade arriving at a level of 2.9 in 1996. (table 54 and 55) Fertility transition took place in a rather dramatic fashion in Brazil: the most recent data available show that the TFR for women age 15-49 (1996) declined 60% both in Brazil as well as in the NE in the past 25 years, (table 54 and 55) 1 2 4 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In just one generation the number of children an average Brazilian woman would have during her lifetime fell from almost 6 to almost 2 (5.76 and 2.3 respectively). This sensational fertility decline started in the early 1980’ s and it is still developing in the late 1990’ s. Brazil’ s TFR was around 6.0 from 1940 to 1970. It fell to moderately in the 1970’ s reachmg 4.35 in 1980. This trend changed dramatically with the onset of fertility transition in then 1980’ s. TFR dropped 35% in Brazil and 35% in the Northeast region the 1980’ s. Brazil’ s TFR fell 63% from 1940-96, 60% from 1970-96 and 47% from 1980-96. In the Northeast the average decline from 1980 to 1996 was slightly higher, or 53%. Fertility decline in the NE in the 1990’ s has been stronger, though, leading to fertility levels less divergent than the national averages. Table 52 Current Contraceptive Use among Women (15 to 49 years old) Living in Union - 1996 Some method None Female Pill* sterilization Male O thers sterilization Brazil 76,7 23,3 40,1 20,7 2,4 13,5 Urban North* 72,3 27,7 51,3 11,1 0,0 9,9 Northeast 78,2 31,8 43,9 12,7 0,4 11,2 South 80,3 19,7 29,0 34,1 3,5 13,7 Central West 84,5 15,5 59,5 16,1 1,8 7,1 Rio de Janeiro 83,0 17,0 46,3 22,5 1,0 13,2 Sao Paulo 78,8 21,2 33,6 21,4 5,3 18,5 Data source: Adapted from PesquisaNadonal sobre Demografia e Saude, PNDS, BENFAM,DHS,BraziI,1996 __________ ____________________ 125 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. T able 5 3 E^rolution of Brazil's Crude Birtb Rate ^ e r 1,000 inh.) 1872-1990 1872-1890 47. 1971 40.5 1981 31. 1890-1900 46. 197240.5 1982 30.2 1900-1920 45. 1973 40. 1983 29.4 1920-1940 44. 1974 39.9 1984 28.6 1940-1950 44. 1975 39.6 1985 27.8 1950-1955 46.4 •* 1976 39.5 1986 26.9 1950-1960 42. 1977 39.6 1987 25.9 1960-1970 40.5 * ** 1978 36.2 1988 25.1 1965-1970 37.7 1979 33.5 1989 24.4 1970 40.8 1980 31.8 1990 23.7 Change in Crude Birth R ate (1990-1940): -0.46% * Brasil em Nümeros-iBGE-1960. Remaining values collected from various IBGE's Anuàrios Estatisticos do Brasil. ** Value estim ated by Jo sé Magno de Carvalho & C harles Wood ^ Estimated by Giorgio Mortara in Merrick & G raham . Crude Birth Rate: The num ber of live births, per y ear per 1.000 of population. Table 54 Total Fertility Rate by Major Regions, 1940-1996 1940 1950 1960 1970 1970 1970 1980 1980 Total Total Total Urban Rural Total Urban Rural Brazil 6.16 6.21 6.28 4.55 7.74 5.76 3.63 6.4 South 5.07 5.7 5.89 4.06 6.86 5.42 3.2 4.55 Southeast 5.69 5.45 6.34 3.83 7.14 4.56 3.17 5.46 Cent-W est 6.36 6.86 6.74 5.31 7.71 6.42 3.97 5.98 Northeast 7.15 7.5 7.39 6.44 8.45 7.53 4.94 7.66 North 7.17 7.97 8.56 6.62 9.59 8.18 5.24 8.04 1980 1991 1996 1996/40 1996/1970 1996/1980 Total Total Total Brazil 4.35 2-7 2.3 -63% -60% -47% South 3.63 2.3 2.1 -59% -61% -42% Southeast 3.45 2.4 1.9 -66% -58% -45% Cent-W est 4.51 2.9 2.3 -64% -64% -49% Northeast 6.13 4 2.9 -59% -61% -53% North 6.45 4 2.8 -61% -66% -57% Total Fertility Rate: T he num ber of children an average woman would have if during her lifetime her chilbearing behavior were the sam e a s th at of the cross section of women at the tim e o f th e observation. Source: IBGE 126 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 55 Total Fertility Rate and Life E^ectancy at Birth by Sex- 1996 Total fertility rate Life expectancy at birth by sex male/fem. male female Brazil 2,28 67,6 63,9 71,4 Urban North* 2,77 67,4 64,5 70,4 Northeast 2,89 64,5 61,5 67,5 Southeast 1,99 68,8 64,4 73,4 South 2,1 70,2 66,5 74,1 Central W est 2,29 68,5 65,3 71,9 Data sources: Based on Projected IBGE/PNAD 1996 * Excludes data fertfae rural area of the states of Rondônia, Acre, Amazonas, Roraima, Para and Amapa. Table 56 Infant Mortality by Sex and Ethnic Origin - 1996 Infant mortality rate (1,000) Ethnic Origin Male/fem Male Female W hite Other Black/"Pardo Brazil 37,5 48,0 36,4 37,3 62,3 Urban North* 36,1** 45,2 34,6 - - Northeast 60,4 71,7 60,8 68,0 96,3 Southeast 25,8 27,7 17,2 25,1 43,1 South 22,8 25,2 14,8 28,3 38,9 Central West 25,8 29,5 19,3 27,8 42,0 D ata sources: Based on projected IBGE/PNAD 1996 * Excludes data fbrdie rural area o f the states o f Rondônia, Acre, Amazonas, Roraima, Para and Amapa. ** Result estimated ferdie North r%ion as a Wiole. -"Pardo": mixed race or color (naulato. mestizo).______________ 127 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A Comparative Perspective of Infant Mortality DecJine in Rragni and in its Northeast Region. W ith, the onset of the mortality revolution as of the 1940’ s, infant mortality has been falling rapidly in all of Brazil's five regions. The declme in infant mortality has not been homogeneous across time or regions, thougpi. This prolonged decline in infant mortality accelerated quite dramatically as of the 1970’ s. (tables 57 to 59) Brazil’ s infant m ortali^ rate has decreased fi'om a level of 163.4 in 1940 to 146 in 1950. In the 1950 s infant mortality rate fell 17%. In 1960 this rate had been lowered to 121 according to IBGE estimates, decreasing at a slower pace to 113 in 1970. (table 57) In the 1970’ s infant mortality rate began to fall very rapidly, which ignited the beginning of a robust fertility decline in the 1980’ s. (table 58) In this decade Brazil’ s infant mortality decreased 39%. By 1980, the infant mortality rate was 6 9 .1 . ^^2 During the 1980 s, while real incomes stagnated, infant mortality rates fell an additional 28% . In the last 20 years Brazil’ s infant mortality was cut in more than 50%, The data oo infimt mortality in Brazil and particularly in its Nordieast are disputable at best. For the period before 1940, the estimates are scarce. We will rely iq>on the official IBGE figures but sudi numbers have also subject to constant dianges. At any rate, dûs researdi considers die official IBGE estimates as well as die estimates developed by Simoes in the 1990’s to be die best ones available. According to IBGE data collected direcdy fiom Demographic Censuses and Anuarios Estatisticos. The IBGE often accepts a shortly h i^er level o f 73 for 1980’s IMR. 128 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. declining from a level around 85 in 1976 to the current level of 37.5 (1996). (tables 57 and 59) Table 57 Evolution of Brazil's Infant Mortality Rate by Great Regions, 1940-1996 (% ) 1940 1950 1960 1970 1980 1990 1996 Brazil 163.4 146.4 121.1 113.8 69.1 49.7 37.5 South 131.1 116.3 87 88.1 43.7 26.7 22.8 Southeast 154.1 131.5 100.6 98.3 47.4 30 25.8 Central-W est 136.3 123.2 101.2 92.3 47.9 33 25.8 Northeast 176.4 175.2 166 146.3 106.8 88.2 60.4 North 167.3 150.3 114.2 109.1 62.8 53.2 36.1 Source: IBGE, Anuarios Estatisticos do Brasil (1960,1981,1996). Note: In more recent IBGE estim ates the IMR is 73 for 1980 and it is 47.2 for 1990, indicating a greater infant mortality decline in th e 1980's._______________________ Table 58 Evolution of Brazil's Infant Mortality Decline by Major Regions, Decade and Selected Periods (% Change) 1950/1940 1960/1950 1970/1960 1980/1970 1990/1980 1996/1990 Brazil -10.40% -17.30% -6.00% -39.30% -28.10% -24.50% South -11.30% -25.20% 1.30% -50.40% -38.90% -14.60% Southeast -14.70% -23.50% -2.30% -51.80% -36.70% -14.00% Central-W est -9.60% -17.86% -8.80% -48.10% -31.10% -21.80% Northeast -0.60% -5.20% -11.90% -27.00% -17.40% -31.50% North -10.20% -24.00% -4.50% -42.40% -15.30% -32.10% 1996/1990 1990/1980 1980/1970 1970/1960 1960/1950 1950/1940 Brazil -24.50% -28.10% -39.30% -6.00% -17.30% -10.40% South -14.60% -38.90% -50.40% -1.30% -25.20% -11.30% Southeast -14.00% -36.70% -51.80% -2.30% -23.50% -14.60% Central-W est -21.80% -31.50% -48.10% -8.80% -17.60% -9.60% Northeast -31.50% -17.40% -27.00% -11.90% -5.20% -0.70% North -32.00% -15.30% -42.40% -4.50% -24.00% -10.20% 1996/1940 1990/1940 1996/1980 1996/1970 1960/1940 Brazil -77% -69.60% -46.30% -67.00% -10.40% South -82.60% -79.60% -47.80% -74.20% -33.60% Southeast -83.30% -85.20% -45.60% -73.70% -34.70% Central-W est -81.10% -83.30% -48.10% -72.00% -25-70% Northeast -65.80% -50.00% -43.40% -58-70% -5.90% North -78.40% -68.20% 42.50% -66.90% -31.80% 129 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 59 Infant Mortality Rates in Brazfl. and in the Nordieast Region, 1965-1996 Y ears B razil N o rth ea st 1965 117,9 153,9 1966 115,3 151,4 1967 112,6 148,8 1968 109,9 145,9 1969 107,0 143,0 1970 104,1 139,8 1971 101,1 138,6 1972 98,0 133,2 1973 94,9 129,6 1974 91,7 126,0 1975 88,6 122,3 1976 85,4 118,5 1977 82,3 114,7 1978 79,1 110,0 1979 76,1 107,0 1980 73,0 103,2 1981 70,0 99,5 1982 67,1 95,9 1983 64,2 92,3 1984 61,5 88,9 1985 58,9 85,6 1986 56,3 82,4 1987 53,9 79,4 1988 51,6 76,6 1989 49,3 73,9 1990 47,2 71,4 1991 45,3 69,1 1992 43,4 66,9 1993 41,6 64,9 1994 40,0 63,1 1995 38.3 61.8 1996 37,5 60,4 Source: Celso Simoes (1997) and IBGE. Obs. Brazil’s infiuit mortality rates for foe late 1970’s have been estimated to be slig)itly higher in other EBGE’s studies. 130 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The structure and timing of infant mortali^ decline in quite distinct than that of the fertility decline. The decline in fertility initiated in the 1980's was caused among other reasons by the remarkable improvement in infant survival chances in the 1970*s- From the 1980 s onwards both infant mortality and fertility rates have been declining very significantly with mutual reinforcing impact. Further, the decrease in infant mortality rates occur in a less homogeneous marmer both temporally and regionally than the fertili^ decline . The process of decline in infant mortality in Brazil happened in two “ waves", with its onset taking place in the 1950’ s with the outbreak of mortality revolution in Brazil, and a second “ wave* developing from the 1970’ s until present times. The pattern of decline in infant mortality in the Northeast is rather unique: infant mortality decline and the very onset of mortality revolution began only in the mid 1960’ s in the Northeast; in addition, the rate of infant mortality decline in the most urbanized and industrialized regions of the country has been stronger than in the Northeast until the mid 1980’ s, when this situation reversed, (table 58) The data on infant mortality before the 1940’ s are limited. In Brazil as a whole, the decline in infant mortality started slowly in the 1940’ s and developed at a very fast pace only in the 1950’ s. 131 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Using revised IBGE data, Simoes asserts that the rate of average annual decline in Brazil’ s infant m ortally changed hrom -1.01% aa between 1940/65 to -0.96% aa between 1965/1975. Between 1975/1985 the rate of average annual decline would have increased to -4.98% aa, reaching a level of -4.14% aa between 1985 and 1990. hi the Northeast region the rate of average decline in infant m ortally until the 1980’ s increased more slowly than in other regions, ranging firom a decline of -0.61% aa (1940/1955), to -1.17% aa (1965/1975) to -3.37% aa (1975/1985). Infant mortality decline was further accelerated as of the mid 1980’ s increasing to -4.53% a.a. between 1985 and 1990. (table 60) Table 60 Average Annual Infant Mortali^ Decline Rates (1940-1990) Years Brazil Northeast 1 94 0 /5 5 -1 .01 -0 .6 1 1 95 5 /6 5 -1 .01 -1.05 1 96 5 /7 5 -0.96 -1.17 1 97 5 /8 5 -4.98 -3.37 1 98 5 /9 0 -4.14 -4.53 Source: Celso Simoes (1997), IBGE. 143 Computed by IBGE/Celso Simoes based oa die demographic census firan 1940 and 1991. 132 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. As far as the national rates are concerned, infant mortality pattern mirrors the decline in life expectancy. Brazil’ s life expectancy at birth improved 7.2% in the 1940’ s and 14.% in the 1950 s. In a like manner, infant mortality decreased 10.4% in the 1940’ s and 17.3% in the 1950 s. (table 61) Table 61 Contrasting Gains in Life Ebcpectancy and Infant Mortality in Brazil and in the NE, by Decade (% ) 1940's 1950's 1960's 1970's 1980's* 1991-96 e Brazil 7.2 14 0.7 17-7 5.6 3.2 NE 12.4 1.9 16.5 13.8 9.4 0.4 IMR Brazil 10.4 17.3 6 39.3 28.1 24.5 NE 0.6 5.2 11.9 27 17.4 31.5 •1980-1991. In the Northeast the decline in infant mortality started later than the initial improvements in life expectancy: life expectancy at birth rose 12.4% in the 1940’ s and 1.9 % in the 1950 s while infant mortality declined 0.6% in the 1940’ s and 5.2% in the 1950’ s. The onset of mortality revolution in the Northeast also happened later. It can be traced to the mid 1960’ s and not the 1950’ s. 133 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. While life e^qxectancy in Brazil improved 0.7% in the 1960’ s, it grew 16.5% in the N E. Infant mortality conditions improved 6% in Brazil and almost 12% in the NE in the 1960’ s. Although the trajectory of the evolution of life expectancy is not necessarily the same as the trajectory of infant mortality’ s, there are many similarities. Moreover, the contribution of infant mortality changes to mortality decline have been significant, particularly in the initial phases of the mortality revolution, hi Brazil this contribution has also grown significandy as of the 1980’ s. Improving survival conditions of infants, whose mortality rates are higher than that of individuals in other age brackets, will have a proportionally higher impact on increasing life expectancy at birth than a similar change in a different age group. Using the Pollard method to evaluate the proportional impact by sex of each age group on life expectancy gains from 1940 to 1990, Simoes concludes that the contribution of the reduction in infant mortality to gains in life expectancy in Brazil changed from levels higher than 20% both for men and women in the 1960’ s and 1970’ s to 43% and 77.5% respectively for women and men. (table 62) Simoes contends that this substantial gender difference is due not only to m ânt mortality declme itself but to the increase in mortality o f young male adults in urban centers due to external causes such as violence, AIDS, etc. 134 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In the Northeast region the contribution of infant mortali^ decline to gains in life expectancy would have increased fourfold in the 1980’ s, from a level around 10% to a level over 40% (41.71 and 44.2%, respectively for women and men), (table 63) Table 62 Brazfl. - Contribution of E)ach Age Group to life Ebqxectancy Gains, 1940-1990 Contribution of Each Age Group to Gains in Life ExpectanQr at Birth Sei& 1940-1950 1950-1960 1960-1970 1970-1980 1980-1990 Age Years % Years % Years % Years Years % <1 0 3 9 3 5 , 7 7 033 13 , 02 0 3 3 2 6 , 0 4 13 7 2 2 3 5 1. 7 7 4 3 4 3 1- 4 0 . 4 6 18 45 14 2 2 3 4 0 038 18 34 1 4 3 1 7 3 7 0 4 0 9 , 7 8 5- 14 0 4 3 5 4 1 040 6 4 3 046 540 049 54 9 0 4 2 239 15 - 19 0 , 0 9 330 046 5 4 2 0 4 2 339 040 338 0 3 5 1 4 0 2 0- 29 0 4 5 932 034 1 7 4 3 0 4 7 1 1 3 9 039 1 1 3 7 0 4 5 3 , 7 5 3 0 - 1 9 0 . 41 1 6 4 3 1 4 2 2 3 4 3 035 2045 136 2 2 4 1 0 3 1 1 4 3 9 50e+ 048 1 1 3 2 037 1 1 , 7 8 0 , 4 8 1 5 . 0 1 14 0 1 6 4 3 0 3 8 2 3 3 6 T o t a l 240 1 0 0 , 0 0 435 10 0, 00 3 4 1 1 0 0 3 0 7 4 3 1 0 0 . 0 0 4 . 09 1 0 0 , 0 0 M e n < 1 0 4 5 1 7 , 4 9 0 , 7 9 1 9, 0 8 0 3 0 2 5 . 7 » 1 , 4 9 2 2 38 2 4 7 7 7 4 9 1- 4 0 , 4 5 2 2 3 1 036 2 0, 71 0 3 2 1 6 3 8 031 1 3 3 3 0 4 9 1 4 3 8 5 - 14 0 4 3 633 046 6 4 9 0 4 7 5 . 4 4 042 4 , 7 8 - 0 4 4 - 4 3 1 15- 19 0 , 0 8 332 045 3 , 7 8 0 4 1 341 040 339 - 0. 09 - 3 , 0 7 2 0- 29 0 4 8 1 4 4 9 035 1 3 4 6 045 1 1 4 6 036 937 - 0 . 0 6 248 3 0 - 4 9 048 2 3 . 7 2 039 2 3 3 2 0 , 7 0 2235 137 2 3 . 7 3 0 3 7 2 0 4 3 50e + 0 4 3 1 1 3 4 034 1 2 3 5 046 1 4 3 7 1 , 4 6 2 2 3 2 • 0 3 4 1 3 5 T o i a l 2 3 1 1 0 0 3 0 444 1 0 0 3 0 3 4 1 1 0 0 3 0 6 3 1 1 0 0 . 0 0 2 3 0 1 0 0 3 0 Source: C L Simoes. Dem. Censuses lMO-1991 135 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 63 Northeast Region - Contribution of Each Age Group to Gains in life Expectancy a t Birth, 1940-1990 Sex& Gende ContribHtioB o f Each Age Group to Gains in Life E xpectant at Birth 1940-1950 1950-1960 1960-1970 1970-1980 1980-1990 Years % Years % Years % Years % Years % W o <1 045 940 048 1 7 4 2 146 2 6 4 8 0 4 3 941 249 4 1 , 7 1 1 - 4 0 , 4 4 2 7 , 0 3 04 5 2 4 , 7 5 048 2 0 4 3 1 4 5 22 48 044 1 4 4 5 5 - 1 4 04 1 640 04 4 643 046 547 04 4 643 047 446 1 5 - 1 9 040 54 1 0 4 2 542 040 448 0 4 7 543 047 2 , 7 2 2M9 042 1 9 4 8 0 4 7 1 6 4 9 0 , 6 2 1 2 4 4 1 4 9 1 6 4 8 041 840 3049 046 2 1 , 7 1 0 , 4 4 2 0 , 0 0 040 1 9 , 0 1 2 4 7 2 4 , 7 6 1 4 3 1 7 4 0 5 0 e+ 047 1 0 4 6 0 4 1 940 043 1 1 4 8 1 4 3 1 5 4 1 0 , 7 2 1 1 4 6 Total 1 > 4 1 0 0 , 0 0 2 4 1 1 0 0 4 0 4 , 7 5 Men 1 0 0 4 0 8 , 7 7 1 0 0 4 0 6 , 4 4 1 0 0 4 0 d 046 945 0 4 9 1 8 4 4 1.0# 2140 048 1148 246 4 4 4 0 1-4 0 , 4 4 2 6 , 0 4 0 4 4 2440 1.11 2140 140 1040 042 1 5 4 7 5 - 1 4 041 6 A1 045 647 0 4 2 841 040 8 4 0 049 5 , 0 6 1 5 - 1 9 0 . 0 7 3 , 8 9 0 4 9 4 , 0 3 0 4 0 3 40 041 3.70 048 3 4 3 2 0 - 2 9 046 1 5 4 5 04 2 14 46 0.00 13,48 1.10 13.18 048 946 3049 047 2 2 , 1 9 0 , 4 0 2246 1.13 2248 242 28,87 14 0 1 7 4 1 5 0 e+ 048 1 6 4 7 0 4 0 944 0.83 1043 1.07 20,02 046 4 , 4 6 Total 1 , 6 9 1 0 0 3 1 0 1 0 0 . 0 0 8.00 10040 - . 04» . _ 10040 5 . 7 » 1 0 0 4 0 S o u rc e : C rb o S fa B o n . I B G E . Dm. C o h o m s 1 9 4 0 - 1 9 9 1 136 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Besides having started later than in other more urban and industrialized regions of Brazil, infant m ortally decline in the Nordieast region until the 1980’ s always lagged behind the average decline of other regions. In the 1950s’ infant mortality declined in average 25% in the industrialized, urban and relatively developed South and Southeast regions and only 5% in the Northeast. In the 1970’ s, infant mortality fell over 50% in the South and Southeast region and 27% in the Northeast. As of the mid 1980’ s, however, the decline in infant mortality in the Northeast became more robust than that of other areas of Brazil. Infant mortality fell 39.3% and 28% in Brazil in the 1970’ s and 1980’ s. In the Northeast it decreased 27% and 17.4%, respectively, (table 61) In the period from 1991-1996, however, infant mortality decline in the Northeast advanced 31.5%, against a 14% decline in the Southeast region and a 24.5% decline in the national rates. The empirical and theoretical investigation of the underlying causes of this change, or of the recent acceleration in infant mortality decline in the Northeast, is the main objective of this study. 137 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 6 3 A T he E^volution o f In fa n t M ortality R ates in th e N o rth ea ste rn S tates, 1965-1994 îars Maianhao Piaui Ceaia Paraiba Rio Grande do Norte 1965 127,5 126,5 160,4 175,2 189,9 1966 125,9 124,8 159,1 173,3 186,5 1967 124,2 122,9 157,5 171,2 182,8 1968 122,4 120,9 155,8 168,9 178,8 1969 120,6 118,8 153,8 166,4 174,3 1970 118,7 116,6 151,6 163,7 169,4 1971 116,7 114,2 149,1 160,7 164,2 1972 114,7 111,8 146,2 157,5 158,5 1973 112,6 109,2 143,1 154,1 152,5 1974 110,4 106,5 139,6 150,5 146,1 1975 108,2 103,8 135,8 146,7 139,5 1976 106,0 100,9 131,6 142,6 132,6 1977 103,6 98,0 127,1 138,4 125,6 1978 101,3 95,0 122,3 134,0 118,4 1979 98,9 91,9 117,1 129,4 111,2 1980 98,5 88,8 111,8 124,8 104,1 1981 94,1 85,7 106,2 120,0 97,0 1982 91,6 82,6 100,4 115,2 90,1 1983 89,2 79,5 94,5 110,5 83,5 1984 86,7 76,4 88,7 105,7 77,2 1985 84,2 73,3 82,8 100,9 71,2 1986 81,8 70,3 77,1 96,3 65,5 1987 79,4 67,3 71,6 91,8 60,3 1988 76,9 64,4 66,2 87,5 55,4 1989 74,5 61,6 61,2 83,3 51,0 1990 72,2 58,9 56,5 79,3 46,9 1991 69,9 56,3 52,1 75,5 43,2 1992 67,6 53,8 48,0 71,9 39,9 1993 65,4 51,4 44,3 68,6 37,0 1994 63,2 49,1 40,9 65,5 34,3 Source: Celso Simoes (1995) from IBGE- Demognçhic Censuses 1940-1991, PNAD’ s 1992-1995. 138 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 63 B The E^olutioii of Infant Mortality Rates in the Northeastern States, 1965-1994 Y ears Alagoas Seigÿe Bahia Pernambuco 1965 165,9 151,2 133,4 173,6 1966 163,9 148,2 131,2 171,3 1967 161,8 145,1 128,9 188,8 1968 159,6 141,9 126,4 166,2 1969 157,4 138,6 123,9 163,5 1970 155,0 135,2 121,2 160,5 1971 152,5 131,7 118,4 157,4 1972 150,0 128,1 115,5 154,1 1973 147,3 124,4 112,5 150,6 1974 144,6 120,7 109,4 146,9 1975 141,8 116,9 106,2 143,1 1976 138,9 113,1 103,0 139,2 1977 135,9 109,3 99,7 135,2 1978 132,9 105,5 96,3 131,0 1979 129,8 101,6 93,0 126,7 1980 126,7 97,8 89,6 122,4 1981 123,5 94,1 88,3 118,1 1982 120,3 90,3 83,0 113,7 1983 117,1 86,7 79,0 109,3 1984 113,9 83,1 76,6 104,9 1985 110,6 79,8 73,4 100,6 1986 107,4 76,2 70,4 96,4 1987 104,2 72,9 67,5 92,2 1988 101,0 69,7 64,6 88,2 1989 97,9 66,6 61,9 84,3 1990 94,8 63,6 59,3 80,5 1991 91,8 60,0 56,8 76,9 1992 88,8 58,0 54,5 73,4 1993 85,9 55,4 52,3 70,1 1994 83,0 53,0 50,2 67,0 Source: Celso Simoes (1995) fîrom IBGE- Demographic Censuses 1940-1991, PNAD’ s 1992-1995. 139 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 6 4 Observed and Moving Average Adjusted Infant Mortality Rates in Brazil, 1929-1992 Years Total Moving Average 1929,4 0,1624 0,1624 1932,1 0,1544 0,1584 1934,7 0,1527 0,1548 1936,8 0,1574 0,1549 1938,6 0,1546 0,1547 1939,3 0,1520 0,1511 1942,0 0,1468 0,1476 1944,6 0,1439 0,1454 1946,7 0,1454 0,1444 1948,5 0,1440 0,1458 1949,2 0,1480 0,1416 1952,0 0,1328 0,1364 1954,7 0,1285 0,1297 1956,9 0,1279 0,1271 1956,9 0,1250 0,1262 1958,8 0,1256 0,1248 1959,9 0,1239 0,1237 1962,5 0,1216 0,1205 1964,9 0,1161 0,1189 1966,4 0,1190 0,1161 1966,9 0,1132 0,1140 1968,6 0,1098 0,1135 1969,4 0,1174 0,1134 1972,1 0,1130 0,1136 1974,7 0,1104 0,1067 1976,8 0,0966 0,0995 1977,6 0,0914 0,0921 1979,2 0,0882 0,0875 1980,5 0,0828 0,0850 1981,3 0,0840 0,0798 1982,9 0,0727 0,0756 1983,1 0,0702 0,0732 1983,5 0,0768 0,0724 1983,9 0,0701 0,0692 1985,5 0,0607 0,0629 1985,6 0,0579 0,0604 140 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 6 4 (Cent.) Observed and Moving Average Adjusted Infant Mortality Rates in Brazil, 1929-1992 Years Total Moving Average 1986,0 0,0627 0,0598 1986,5 0,0589 0,0580 1987,6 0,0525 0,0542 1987,6 0,0512 0,0520 1988,5 0,0522 0,0526 1988,6 0,0544 0,0519 1989,6 0,0492 0,0512 1990,5 0,0500 0,0483 1990,6 0,0456 0,0474 1992,5 0,0466 0,0474 Source: Celso Simoes (1997) from IBGE-Demographic Censuses 1940-1991 and PNAD’s 1992-1995. Table 65 Observed and Moving Average Adjusted Infant Mortality Rates in the Northeast Region, 1926-1990 Years Total Moving Average 1926,8 0,1904 0,1920 1929,8 0,1944 0,1973 1932,4 0,1899 0,1904 1935,0 0,1938 0,1912 1937,0 0,1970 0,1889 1938,8 0,1900 0,1842 1936,5 0,1809 0,1824 1939,5 0,1849 0,1877 1942,2 0,1834 0,1839 1944,7 0,1885 0,1860 1946,7 0,1922 0,1842 1946,8 0,1703 0,1717 1949,8 0,1730 0,1757 1952,4 0,1703 0,1708 1955,0 0,1719 0,1696 1957,1 0,1719 0,1656 1958,8 0,1686 0,1633 141 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 6 5 (Cont.) Observed and Moving Average Adjusted Infant Mortality Rates in the Northeast Region, 1926-1990 Years Total Moving Average 1959,2 0,1606 0,1639 1960,2 0,1607 0,1641 1962,8 0,1545 0,1553 1965,2 0,1533 0,1509 1967,1 0,1512 0,1491 1968,8 0,1443 0,1430 1969,7 0,1476 0,1557 1972,3 0,1371 0,1422 1974,8 0,1319 0,1335 1976,9 0,1295 0,1307 1978,7 0,1339 0,1348 1977,1 0,1082 0,1352 1980,1 0,0989 0,1185 1982,7 0,0867 0,1001 1985,3 0,0811 0,0954 1987,4 0,0740 0,0844 1989,3 0,0682 0,0793 1977,5 0,0998 0,1250 1980,4 0,0981 0,1176 1983,1 0,0864 0,0997 1985,6 0,0791 0,0931 1987,7 0,0736 0,0840 1978,2 0,1012 0,1268 1981,1 0,0914 0,1096 1983,8 0,0856 0,0988 1986,5 0,0774 0,0912 1988,6 0,0706 0,0806 1980,3 0,1048 0,1311 1983,2 0,0957 0,1146 1985,8 0,0798 0,0922 1988,4 0,0708 0,0835 1990,6 0,0629 0.0719 Source: Celso Simoes (1997) from IBGE- Demi PNAD’s 1992-1995. 142 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Thft Tmoact of Public Policy and Health on Infant Mortality Decline The Brazilian Institute for Geography and Statistics has been recording an increase in education levels across all age groups. For instance, 67% of children between 15 and 17 are now in school against 55% only five years ago. Without an ofGcial family planning program, but boosted by this rapid urbanization process, by the dissemination of modem contraception methods, by improvements in women educational and labor status, by a steady decline in infant and child mortality rates as well as by a vigorous public and institutional health intervention, the demographic transition seems to be well advanced in Brazil. The astonishing fall in the levels of fertility and infant mortality rates seem to be particularly influenced by two structural factors. The first one is the dynamic redefinition of the roles and status of women in Brazil’ s society. In 1960 the proportion of women age 20-29 in the labor market was 20%, this rate increased to 40% in 1970 and to almost 60% in 1996. The share of women within this age bracket with no education has fallen from 40% in 1960 to 7% in 1996. (table 66- 68) Table 66 Distribution of Young Women by Educational Attainment (% ) (Women 20-29 by years in school) 1960 1980 None 40.7 18. Primary (1-5) 50.2 44.5 Secondary (6+) 9.1 37.5 1990 9.13 Source: For 1980 & 1980 -Merrick (1985). For 1990 Census Tabulation. * In 1990, the # years in school cells were changed to 1-2,2-4,4-8. 143 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 6 7 B razil’ s Fem ale L ab o r F orce P articipation. R ate, 1960-1990(% ) Age 1960 1970 1980 1990 15-19 23.4 24.2 31.1 3 2 2 20-29 20.6 26.1 37-7 49.8 30-44 17.1 21.2 34.2 49.5 Source: Adapted from IBGE-Census Data T able 68 Illiteracy Levels, 18 7 2 -1 9 9 1 1872 84.2 1890 85.2 1900 65.1 1910 63.8 1920 62.7 1930 60.2 1940 55.4 1950 50.7 1960 39.5 1970 34.1 1980 26.1 1991 18.4 SourcerlBGE (Census Years) The second structural factor is the widespread expansion of public health which includes the supply water and sanitation services, vaccination and control of infectious diseases, social legislation, urban infrastructure, and, particularly since the 1980’ s, the new public policies and programs to promote the health of the mother and child -such as breast-feeding, oral rehydration, diarrhea control, pre-natal care, etc. 144 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Although the dimension of the impact of medical innovations and public health measures on infant mortali^ before the onset of mortality revolution in Brazil is not clear, there is evidence that after 1940 such changes did exert a profound impact on infant mortality decline. With the end of the ‘ O ld R ep itb lic ” in 1930, the structure of the Brazilian state is redefined, and public intervention to promote economic development becomes pervasive. Major investments are implemented in public health, infrastructure, social policies and sanitation. Prior to this period, the overall health and hygiene conditions, particularly in urban areas, were very poor. To help to coordinate such a public effort to improve health conditions, the Ministry of Health is created. Government intervention in public health coincided with an intense economic and urban growth process in Brazil. However, the continuous decline in mortality levels after 1940 was not driven by socioeconomic factors or changes, but rather by the conjugation of two sets of forces: public policy and government intervention as well as by the dissemination of medical innovations. Technical changes in the medical field were institutionalized. Immunization, vaccination campaigns, and new pharmaceutical products were ag g a y essrvely introduced reducing and controlling the spreading of many infectious diseases. 145 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. As a result, the structure of ca u ses o f d e a th for adults and infants alike is dramatically transformed in the absence of major changes in the economic environment. Technological knowledge applied to medical sciences itself as well as the categorical government role disseminating these innovations, improving sanitary infiastructure and promoting public health, are the dynamic forces explaining infant mortality decline in Brazil between the 1940 s and the 1960’ s. In spite of the importance of medical innovations and public health campaigns (such as the successful campaigns to eradicate malaria ) to explain mortality decline in LDC’ s and in Brazil in particular after W .W .n, it was certainly an exaggeration to claim that: These countries do not have to develop and maintain a major medical establishment of their own; rather, they can “ import* new techniques, discoveries or drugs from more advanced nations, as well as receive international or financial aid. Hence the public health variable is almost independent of the country’ s economy; it depends a great deal on medical progress and development in other countries, hi other words, assuming that most new medical discoveries occur in the most advanced countries, the public health of an underdevelopment country is related more to the economy of the advanced countries that it is to its own economy. 1 4 5 In stark contrast to this view and, while comparing the historic decline of infant mortality in Latin America to that of England and Wales, Palloni contends that * infant mortality declined during the first three or four decades Eduardo Arriaga, “ M oitali^ Declme and its Demographic Effects in Latm America”. (Berkeley: Institute for International Studies, 1970). Other authors sudi as Stolnrtz (1965) claimed that mortality decline was absolutely independent of economic development. 146 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. after W.W.Il by as much, as it would have in the absence of applications of new medical technologies; its relative magnitude is ju st about the same as the decline that took place in England and Wales before the introduction of important advances in the control of diseases." Aside from the ideological implications of these two views they reflect, once again, the two distinct interpretations of the driving forces behind infant mortality decline. Palloni’ s adamant claim that infant m ortah^ gains in Latin America could only be derived from medical innovations after the 1950’ s seems to contrast with the facts, at least for the specific case of Brazil, Nevertheless, this author’ s assertion on the existence of a limit to further infant mortality decline in Latin America without socioeconomic change does fit the facts as far as Brazil’ s historical experience is concerned. Dramatic reduction in m o rta lity need not be a thing of the past. The influence of socioeconomic factors on mortality at various ages is paramount in any prognosis for further reductions. Social and economic factors related to the rate of adult literacy appear to be by far the most relevant; they are surely among those to be manipulated if any further improvements are to be realized. After W .W .n, infant mortality gains were continuous but not uniformly distributed. In the case of Brazil, and principally of its Northeast region, the Alberto Palloni, Ibid., 1991,191. "^AlbertoPalloni, Ibid, 1991,191. 14% And against die view diat a elimination of the gap between mortality levels m LDC and developed countries would be immment Alberto Palloni, “ Mortality in Latin America: Emergmg Patterns”, Ibid., 1981,645. 147 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. impact of public health, and medical innovations on infant mortality did have a differentiated impact on different age groups but also on different social and income strata. Mortality levels and particularly infant mortality rates in the Northeast region did not fall as fast as in other more developed regions of Brazil. For instance, the difference in life expectancy between the Northeast and the Southeast increased from an average of 7 years in 1940 (38.9 vs. 45.9) to 11.5 years in 1960 (43.6 vs. 55) . Infant mortality rate in the Northeast in 1940 was 24% over the national average ( 185.0 vs. 149.), rising to 32% in 1955 ( 168.7 vs. 128.).i5o By virtue of the economic crisis of the early 1960’ s, the decay of urban infrastructure and services and the perverse characteristics of Brazil’ s model of economic development in terms of regional, income and social inequali^, the process of mortality decline saturates and slows down in the 1960’ s. isi The desaceleration in the rate of mortality decline between 1965 and 1975 reflects the limits of mortality gains in the absence of major socioeconomic reforms or in the presence of intense income and regional inequaüly. Source: These are the revised estimates calculated by Simoes (IBGE), 1997. The official figures from Brazil’s demographic census are sligfitly higher. Celso Simoes, ‘ ^Aqiectos Metodologicos das Estimativas de Mortalidade hifimtil no Brasil”. (Rio de Janeiro: IBGE, 1990). 148 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 69 Public E^enditures on Health (%GDP), 1960-1991 19601- 19741.6 19801.3 1986 2.7 19651.2 19752.2 1981 1.5 19872.9 19701.3 19761.2 19821.7 1988 2.5 1971 1.4 19771.1 1983 1.3 1989 2.9 19721.3 19781.3 19841.6 1990 2.9 19731.2 19791.4 1985 2.1 1990 2.9 Source; Economic Council for Latin American and the Caribbean -CEPAL-ECLAC Yeart)ooks (1980-1993). The Brazilian government new strategy of public intervention and socioeconomic development adopted as of mid 1970’ s would have a very positive impact on infant mortality levels in the 1980’ s and 1990 s. The success of such a strategy does not indicate the primacy of exogenous over endogenous causes on bringing down infant mortality in Brazil. For the most part of the period, and particularly during the 1980’ s, living standards did not grow. It does not s u re s t that endogenous forces and public health per se determine child survival either. What it truly indicates is that the coordination of policies and programs to promote both socioeconomic development and health intervention targeting the poorest segments and areas of Brazil - such as breastfeeding and immunization campaigns, sanitation reform and educational improvements - constitute an efScient way to improve infant survival rates. 149 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Simoes isz and Jaguaiibe isa argue that by 1974 the military government of Brazil had set a new strategy for social policy and government intervention through the creation of several programs and investments. These programs would have a very positive impact on Brazil’ s and particularly on the Northeast’ s infant mortality rates in the following decades. Among such programs and organisms Simoes cites: - The Council for Social Development (C D S) linked to the Presidency, 1974; - The Program to Support Social Development Projects (FA S), 1974; - The Program for Social Development in the Northeast, 1974; - The System of Housing Financing (B N H ), 1975; - The Program of Sanitary Action for the Northeast, 1975; - The National Sanitary Plan (Planasa), 1975; - The Program of Sanitary Action for the Amazon region; - The Program of Epidemiological Vigilance Actions, 1975; - The Program of Labor Actions, 1975; - The National Program of Food and Nutrition, 1976; - The Program of Environmental Sanitation, 1976; - The Program of hiteriorization of Health and Sanitation Actions in the Northeast (PIASS), 1976. - The F ’ rogram of Infant and Maternal Health, 1977. Celso Simoes, “ A Mortalidade DrAntil na Transicao da Mortalidade no Brasil; Um Estudo Compaiativo entre o Nordeste e o Sudeste”, Ph.D. Dissertation m Doncgraphy. ^ l o Horizonte: Cedeplar, 1997). Helio Jaguaribe, Iritroducao ao Desenvolvimento Social”. (Rio de Janeiro: Paz e Terra, 1978). 150 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Despite of the fact that some of these programs would be later abandoned, others such as the Program of Infant and Maternal Health, would have a quite positive impact on infant survival, through the improvement of health, sanitation, nutrition, immunization and education levels, especially in poor rural communities of the Northeast. More importantly, these programs laid the foundation for new programs, strategies and government actions to improve infant m ortali^ levels and achieve other social objectives developed in the 1980’ s and 1990’ s. As of the 1980’ s infant mortality decline accelerates significantly in Brazil as a whole and in the Northeast region in particular. This study is particularly concerned with this period. In analyzing the historical evolution of infant mortality between 1950 and 1990, Simoes claims that, while in some other regions of Brazil the relative contribution of infant mortality to gains in life expectancy were considerable throu^out the period, in the Northeast this contribution only became important after 1980. Temporal and regional differences in infant mortality rates are further reduced in the 1990’ s. iss The decline in infant mortality in the Northeast expands in both absolute and relative terms. Celso Simoes, Ibid., 1997. Simoes argues that infant mortality differences between die Nordieast and die Southeast are the same between 1940 and 1950, m spite o f die decline m Northeastern infant mortality levels in the 1980’s. This is true but if one studies infant mortality regional differentials between 1990 and 1950’s or 1960’s instead, it is possible to verify that die trend indicates a greater decline in die Northeast or a falling differential. Observing infant mortalify trends m die Northeast after 1990 can further support diis hypothesis. In short, not only inftmt mortalify decline accelerated in absolute terms but also in relative terms as o f die 1980’s and particularly in the 1990’s. 1 5 1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. This research hypothesizes, through an empirical analysis of household survey data from 1986 to 1996, that the robust improvement in infant survival conditions in the Northeast is due to a myriad of possible factors or determinants. Some of these causes or variables would be endogenous, such as income levels (estimated through a proagr, "household goods* ) and household crowding (which would reflect the interactive impact of the recent fertility decline). Most of the possible causes, however, would reflect social- institutional changes and health advancement promoted through public policy and government action, such as gains in educational attainment of the mother, sanitation conditions, pre-natal medical care, immunization and breastfeeding practices. Several successful health and educational programs were instituted between the early 1980’ s and 1996. The vast majority of them maintained by government agencies but many by NGO’ s such as UNICEF and the Catholic Church. One of the main non government organizations engaged in bringing about lower infant mortality rates is the Catholic Church. Its "C hild’ s P astoral” was created in 1983. One of the main objectives of the "C hild’ s P astoral” is to improve infant and child survival through nutrition, literacy, vaccination and educational programs. 152 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In 1996 the ‘ ^C hild’ s P astoral’ ’ was present in all of Brazil’ s 27 states and in half of its municipalities, accompanymg directly 2 million families through 77,000 '^com rnunxtarian ctg en ts.’ ’ In 1984 The Ministry of Health instituted the Program of Integral Assistance to the Health of the Child. In 1994 the Basic Plan of Action for the Protection of the Child and Adolescent was created. The government and particularly the Nhnistry of Health began to develop and promote a "Pact for Infancy” involving different organizations of socie^ and distinct social programs, having the reduction of infant m ortali^ as its main objective. In an aggressive stance, the Ministry of Eklucation has been promoting substantial gains in literacy levels, specially in the Northeast. One of the most successful examples of recent engaged and coordinated public intervention to reduce infant mortality is the Ministry of Health’ s PRMI or Project for the Reduction of Infant Mortality. Instituted in 1995 the PRM I envisages actions and measures for the development of human resources and for the promotion and reorganization of public health, all integrated in a wide and multi-sectorial strategy involving other government institutions and Ministries and having as its main objective improving infant mortality conditions. The PRMI divides its policies and investments into five main categories: sanitation; food and nutrition; immunization ; health assistance to the mother and child; communitarian health agents. 153 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The PRM I is developed in specific poor mimicipalities where infant mortalily is high. From the 914 municipalities covered in 1996, 582 or 57% were located in the Northeast region. The PRMI assigns communitarian agents and nurses to work on a preventive basis visiting low income areas periodically. Having as main objective of public policy the reduction of infant mortaliy^, the PRMI allocates investments towards sanitation, nutrition and immunization programs. The PRMI is an extremely high cost/benefit ratio program, often cited as exerting a very positive impact on infant mortality decline in the Northeast in the 1990's. In short, the reduction in infant mortality in the Northeast seem to have been positively affected by an association of efforts and programs involving non government organizations such as the Catholic Church, international organizations such as UNICEF, and public policy intervention at the Municipal, State and Federal level. See among others the fordicoming bode by USAID’s economist Judith Tendler. “Gocxl Government in the Tropics” 154 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. “ The N eed fo r a P roject to Achieve a Socioeconom ic Self-Tran^orm ation D evelopm ent- besides being a productivity -increasing phenom enon o f the labor fa c to r th at in terests the econom ist- is an adaptation p ro cess o f the socioeconom ic stru ctures to a n expanding horizon o f p o ssibilities open to m an. The tw o dim ensions o f developm ent- the econom ic a n d cu ltu ra l- cannot be p erceived except a s a whole. A n d fo r reasons o f m ethodological sim plicity, the econom ist concentrates h is attention on the m easurable aspects o f developm ent; tha t is, he fa v o rs the variables th a t are quantifiable. I t is im p licit that the other elem ents o f the process rem ain unchanged or that they do n o t a ffe c t sig n ifica ntly the to ta l o f the process during the p erio d in which observations are m ade. Such a reduction o f rea lity to a sim plified schem e is com m on in scientific work. H owever, in the study o f developm ent, that m ethod involves a special risk arising fro m the unquestionably sig nifica nt im portance o f the time fa cto r. Thus the relations am ong econom ic variables are establishedfrom non-econom ic data, such a s the structure o f the population, consum er habits, institutional fram ew ork, etc. -a ll isolated fro m the tim e fa cto r. Im m ediately thereafter econom ists consider the behavior in time o f a certain econom ic variable by keeping in m ind only the observable behavior o f a lim ited num ber o f other variables. A t the end o f the tim e p erio d , the non-econom ic d a t a are again observed a s i f they h a d evolved a ll by them selves independently o f the behavior o f the econom ic variables, a n d a s i f such a behavior co u ld be explained w ithout having to take into account the perm anent transform ations o f the non-econom ic data. In sum mary, the interaction o f the econom ic w ith the noneconom ic, which undoubtedly is the m ost im portant fa c to r in the process o f developm ent, is c lo x d out fro m the range o f observations o f the econom ist. W hat a t the beginning w as a necessary m ethodological sim plification is soon tranfform ed into an obstacle to the perception o f the rutture o f the problem itself. ” (C elso Furtado, A Stucfy o f tin Case o f B razil, in O bstacles to D evelopm ent in L a tin Am erica, 1970) 155 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 4 CONCEPTUAL FRAMEWORK & LITERATURE REVIEW This study sets out to investigate the impact of several socioeconomic and demographic determinants on recent infant mortality decline in Brazil’ s Northeast based on the three Demographic and Health. Surveys of 1986, 1991 and 1996. In the early 1970’ s, Brazil's infant mortality rate was still one of Latin America’ s highest. In a period of one generation, infant mortality declined in Brazil from 105 (1970) to 39 (1996). During the same period urbanization levels increased firom 55 to 78%. Several studies have emphasized the paramount importance of public policy in sanitation and health in promoting such a improvement in infant and child mortality rates. In Brazil the allocation of these services has been concentrated predominantly in the most afduent South and Southeast regions, as well as in urban areas. In 1970, the differentials in child survival by residence were quite small. In most of the regions, urban infant and child According to die data released by die 1996 DBS survey for the five-year period prior to the survey. Such as DHS/BEMFAM, 1996 National Research on Health & Danography ( Rio de Janeiro, 1997); C Simoes and D. Vetter, “Acesso aos Serviços de Saneamento Basico e Mortalidade”, Revista Brasileira de Estatistica, V61.42, No.169 (1982): 17-35. 156 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. mortalily were even slightly higher than in rural areas, hi contrast, infant survival chances in urban settings in 1996 are almost 100% higher than in 159 urban areas (61 and 32/000, respectively). A comprehensive framework to understand and explain the underlying forces behind the trajectory of infant mortality rates in Brazil and more specifically in the Northeast region m ust be both multidisciplinaxy and multidimensional. Mosley and Chen argue that a thorough conceptual framework must incorporate and unify research methods used by both medical and social scientists. In addition, an effective conceptual framework must be able to address both the micro and macro level determinants of child survival. These authors claim that malnutrition and infectious diseases are often assumed by medical researchers to be the causes of infant and child deaths, and that, instead, these factors are the consequences of a b io so d a l in tera ctio n . While a great deal of research and policies have been formulated to measure and prevent risk factors, the social interactions are often taken as given. 159 1996 DHS, forthe five-year period prior to Ae surv^. 160 H. M o sI qt and L. Chen, “An Analytical Framework for Ae SAcfy of Child Survival in Developmg Countries”, Population and Development Review, Child Research- Strategies for Research, Supplement to V61.10 (1984): 25-47. 157 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Mosley and Foster show th at most medical control strategies for diarrheal diseases have focused on efTective, but merely curative oral rehydration procedures, neglecting socio-economic and cultural interactions - which are the ultimate determ inants of both an efScient use of health facilities as well as of the interruption of contamination and malady through structural changes at the household level. In the realm of the social sciences, Palloni observes that macro-level research has focused on aggregate data on existing data sets, without much attention to data analysis and research design. With the advent major demographic surveys such as the WFS’ s and DHS’ s, new conceptual frameworks were devised, shifting the emphasis from describing and documenting mortality differentials to understanding its underpinnings at the household level. This relatively new micro-level social research focuses on modeling the causal paths between the socio-economic environment and mortality patterns. Individual and household data are scrutinized. Data on water supply, disposal of sewage and use of health services come to the forefront. 161 Henry Mosl^r, “Child Survival: Research and Poliqr, Population and Development Review, Child Research- Strategies for Research, SupplCTMntto Vol. 10 (1984): 3-23. 162 S. Foster, “Immunization and Respiratory Diseases and Child Mortality”, Population and Develc^ment Review, Child Research- Strat%ies for Research, Supplemmt to Vol. 10 (1984): 119-139. 163 A. Palloni, “Mortalhy in Latin America: Emerging Patterns”, Population and Development Review, Vol.7, No. 4 (1981): 623-649. 158 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. With regards to the role of living standards, per capta income and nutrition intake vis-à-vis technological progress in the medical field, McKeown is often cited as one of the main proponents of the view that demographic changes in mortality would be endogenous to economic growth. In stark contrast, Arriaga & Davis Stolnitz Davis among others, advocate that the unprecedented mortality changes set forth by the demographic transition in developing countries would be largely independent of socio-economic development and determined primarily by medical innovations and by the generalization of modem medical technology. Van de Walle & Mbacké consider such views which assume that medical innovations and public health could improve mortality patterns in a sustained fashion even in the absence of a major increase in individual incomes to be overly optimistic. 164 T. McKeown, The Modem Rise of Pc^ulatica (New Yoric: Academic Press, 1976). E. Arriaga and K. Davis, ^Patterns o f M ortally Change in Latm America”, Demography, No.6 (1969): 223-242. 166 G. Stolnitz, ‘ Thtemational Mortality Trends: Some Mam Trends and Implications”, UN, The Peculation Debate: Dimensions and Perspectives, P ^ers o f the World Population Conference, Vol. 2 ( Budiarest, 1974). G. Stolnitz, “Recent Mortality Trends in Latin America, Asia and Africa” , Population Studies, No. 19 (1965): 117-138. 167 K. Davis, “The Amazing Declme o f Mortality in Underdeveloped Areas”, American Economic Review, No. 46 (1956): 305-318. E. Van de Walle and C. Mbacké, “Socio-Economic Factors and Health Service Use in Mortality and Society in Sub-Saharan Africa” (Oxford: Clarenckm Press, 1992): 123-144. 159 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Preston and Palloni advocate that both technological medical progress and income growth play a role in explaining historical improvements in life e3q)ectancy. hivestigating infant mortality in early 2 0 * * ^ century US, Preston and Haines posit that mortality differentials and child survival can be well explained not by nutritional status and per capta income, but rather by socio-class analysis, public health and interventions m healüi and sanitation. Not lack of resources, but lack of knowledge about causation, transmission and prevention of diseases would be behind dismal child survival. Palloni and Gwatkin both argue that the economic stagnation would be an impediment to further mortality decline in developing countries. Palloni propounds that there appears to be some degree of independence between the effects of socio-economic development and the effects of medical innovation in Latin America. According to this author, whereas the former can contribute to mortality decline in the absence of the 169 S. Prestw, “The Changmg Relations Between Mortality' and the Level of Econcxnic Development”, Population Studies, Vol. 29, No. 2 (1980): 231-248. S. Preston, “Causes and Conseqpiences of Mortality Declines in Less Developed Countries During the Twentieth Caitury”, Population and Economic Change m Developmg Countries, ed. R. Easterlin, (Chicago: University of Chicago Press, 1980): 289-359. 170 A. Palloni, “Mortality Decline in Latin America”, Paper presented at the Annual Meeting of the Population Association o f America (Washington D C , 1979). S. Prestcxi, and M. Hames, Fatal Years: Child Mortality m Late Twentietii Century America, (New Jersey: Prmceton IMversity Press, 1991). 172 A. Palloni, Ibid., 1981. D. Gwatkin, “hidicators of Change in Developing Countries Mortality Trends: The End of an Era?”, Population and Development Review, Vol.6. No. 4 (1980): 615-644. 1 7 4 A. Palloni, Ibid., 1981. 160 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. latter (as in the Knglish case), the effects of medical innovation would be conditioned by socio-economic development. A low level of socio-economic development would set boundaries on the possibility of efSciently absorbing certain types of technologies. Similarly, recent studies in Brazil recognize the importance of medical improvements, prenatal care and. public health for mortality decline, but stress that the impact on child survival is concentrated regionally and socio economically, and that the access to and efScient use of health services would be negatively affected by an uneven prevailing socio-economic structure. As a result, there would be a structural limit to further gains in infant and child mortality. Such studies have emphasized the role of parturition decline, of the decrease of births at riskier (under 20 and 35+) age groups, as well as of higher birth intervals, in bringing down Brazil’ s infant mortality. In what Mosley calls o n e o f th e m ore co m p reh en sive n a tio n a l s tu d ie s o f m o rta lity to u n d e rta k e n to d a te DaVanzo attempts to e3q>lain the mechanisms through which socio-economic determinants influence mortality. 175 Such as: C. Simoes, “A Saude Tnfentil no Brasil nos Anos 90”, hiância Brasileira nos Anos 90, UNICEF/IBGE (Brasilia, 1996). L. Ortiz and C Simoes, “A Mortalidade Ihfantil no Brasil nos Anos 80”, IBGE/DPE, Vol. 1, No.7 (Rio de Janeiro, 1988). 176 H. Mosley, Ibid, 1984. J. DaVanzo, * * A Housdiold Survey of Child Determinants m Malaysia”, Population and Development Review, Child Research- Strat%ies forReseardi, Siq»planaitto Vol.10 (1984): 307- 323. 16 1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. DaVanzo sets out to review Malaysian surveys on child survival and finds a strong correlation between socio-economic status and infant mortality. DaVanzo notes, however, that such correlation is particularly significant primarily in low income families. Other determinant variables frequently cited in die Brazilian literature are; breastfeeding patterns; mothers’ education; the provision of health services in immunization, prenatal care as well as oral rehydration, maternal and infant health campaigns; sources of drinking water and sewage disposal. The sanitary situation in Brazil, and particularly in the Northeast is very unsatisfactory. Nevertheless, the proportion of Brazil’ s households with permanent sewage disposal increased firom 26% in 1970 to 40% (1980), to over 55% in 1996. In the Northeast these proportions increased firom 8% in 1970, to 17% (1980) to less than 20% in 1996. In the rural areas of Brazil’ s most impoverished region, in which child survival is dreary, the respective sewage rates are 0.3%, 1.8% to 3%. While 95% of the total Brazilian homes have piped water, in the Northeast this proportion is 50%. E ^ v en though substantial improvements are nationally taking place in electricity, sewage disposal and water supply, the urban-rural differentials as well as regional differentials between Brazil and its Northeast region are still rather high. 162 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Behm & Primante using census and survey data for many Latin American countries, argue that in two-thirds of the countries the risk of mortality in the first two years of life in rural areas was 30-60 higher than in urban areas. The 1996 DHS/BEMFAM maintains that Brazil’ s urban-rural differential in infant mortality has been increasing substantially in the 1990’ s by virtue of the relative improvement in the provision of public health, sanitation and pre-natal care in urban areas. According to Behn and Vargas the urban-rural differential of infant mortality in Latin America tends to vanish when one controls for social class and mother’ s education. Another variable very firequently cited as having high explanatory power for the understanding of Brazil’ s infant mortality rate is education. Van de Walle & Mbacké consider maternal education to be the most important variable accounting for the variance of child and infant mortality. Inkeles associates modernization with a ‘syn d ro m e o f a ttitu d e s, v a lu e s a n d b eh a vio r", having education (in a broader sense) as one of its main forces. ITS H. Bdun and D Primante, “Mortalidad en Los Primeros Anos de Vida ea La America Latina”, Notas de Poblacion Vol.6, No. 16 (1978). 179 H. Bdim and E. Vargas, “Guatemala: Difetencias Socioeconomicas de La Mortalidad de Los Mmores de Dos Anos 1968-1976”, Ser A, No.1044 (San José, 1984). ISO E. Van de Walle and C Mbacké, Ibid, 1992. A. Inkeles, “Making Men Modem: On the Causes and Consequences o f Individual Change m Six Developing Countries”, American Journal of Sociology, 75, No. 2 (1969): 208-225. 1 6 3 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The Brazilian Institute for Geography and Statistics has been recording an increase in education levels across all age groups. 67% of children between 15 and 17 are in school (1996) against 55% only five years ago. Notably the improvement in female schooling levels reflects the current dynamic redefinition of the roles and status of women in Brazil’ s socie^. hi 1960 the proportion of women age 20-29 in the labor market was 20%, this rate increased to 40% in 1970 and to almost 60% in. 1996. The share of women within this age bracket with no education has fallen fi* om 40% in 1960 to 7% in 1996. The differentials in infant mortality rates by schooling levels seem to have an impact even greater than by urban-rural residence differential. The infant mortality rate of children whose mothers have no education is 93 in Brazil. With four years of education, this rate falls to 42. For women with 12 or more years of education, the infant mortality rate plunges to a level of only 5 children per 1,000 births. Behm shows an inverse relationship between child mortality and maternal education. Palloni observes that such a inverse relation is contingent on the social setting and that in poor environments an uneducated mother is in a far greater disadvantage than more educated mothers. 182 H. Bdun, “SocioecooOTiic Detenninants of Mortality in Latin America”, Proceedings of the M eting on Socioeconmnic Determinants and Consequences o f Mortality, UN and WHO (Mexico City, 1979): 19-25. 183 A- Palloni, Ibid, 1981. 164 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Mosley & Chen argue that maternal educational level can affect child survival by influencing her choices and increasing her skills in health care practices related to hygiene, contraception, nutrition, preventive care and disease treatment. The positive impact of parent’ s education -and particularly of maternal education- on diildren’ s survival has been emphasized in many studies. Using data from Kenya, Mosley asserts that an increased amount of female education accounts for an important share of the gains in infant and child mortalily. This author claims that the impact of maternal education on the proximate determinants of infant and child mortality is so intense that there would exist a “social synergy" process. Preston contends that, since a substantial part of the differential in urban and rural infant and child mortality in the surveys, are determined by residence differentials in education levels, the effectiveness of public policy focusing on medical investments in urban areas must be reevaluated. IS4 H. Mosley and L Chen, Ibid, 1984. H. Mosl^r, H., “Les Soins de Santé Primaires Peuvent-ils Réduire La Mortalité Inûmtile? Bilan Critique de Quelques Programmes Ahicams et Asiatiques” m La Lutte Contre La Mort, J. Vallin and A. Lopez (eds). Travaux et Documents No. 108,101-136, ENED-PUF QParis, 1985). VS6 H. Mosley, Ibid, 1984. S. Preston, “M ortal^ in Childhood: Lessons firxn WHS”, Reproductive Change m Developing Countries, J. Cleland and J. Hobcraft (eds), (Oxford University Press, 1985): 253-272. 165 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Caldwell observes, however, that ‘demographers may have begun to concentrate too much on maternal education rather than on its larger context*, and that schooling would be nothing but a catalyst which promotes an optimal usage of health facilities and practices (according to this author, the dilemma I s it b e tter to build, a sc h o o l o r a h o sp ita l? would be, thus, a false dilemma). Caldwell claims that education fosters child survival through a reduction in fatalism, through a more enlightened understanding of the Modern world’ and through a shift in traditional balance of family power relationships that gives mothers more influence in issues such as child care, food allocation and illness treatment. Ruzicka stresses that one of the main methodological shortcomings of analysis of empirical data in the studies of child mortality differentials by social and economic strata has been that the main stratification indicators are highly correlated with each other: parental occupation is closely correlated to education, individual income to occupation, and so on. Mosley and Chen argue that, even though income and education often correlate to infant and child survival in developing countries, the exact J. Caldwell, “ Routes to Low Moitality”, Populatioa and Development Review, Vol. 12, No. 4 (1986): 395-413. IS9 L. Ruzicka, ‘Tioblems and Issues in the Study of Mortality Differentials”, Differential Mortality: MetbodoI<%ical Issues and Biosocial Factors, L. Ruzicka & G. Wunsch & P. Kane, (eds.), (Oxford: Clarendon Press, 1989). 190 H. Mosley and L. Chen, Ibid., 1984. 166 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. mechanisms through which such socio-economic determinants as well as others operate remain a “ b la d e h o x T . Medical science attributes mortality to specific biological processes of disease such as infections and malnutrition. The methodology and research focus hinges on dietary intake as well as on the mechanisms of disease transmission in the environment. The most commonly adopted dependent variable is morbidity or the manifestations of sickness among survivors. An example of a conceptual model to explain infant and child mortality which includes not only socioeconomic and biological factors as well as institutional development is the one developed by Pool to evaluate mortality trends in New Zealand. This author distinguishes between explanatory variables and intermediate variables. The latter including preventive and curative medicine, health measures and disease mechanisms. The former is divided into macro and micro-level variables. The macro variables are health care infiastructure and technology, health administration and regulations, environment, sanitation and socio-economic organization. The micro-level explanatory variables in this firamework are hygiene, medical practices, individual socioeconomic status, genetic factors and biosocial characteristics (gender, age and parity). Pool’ s conceptual firamework and many other ones are criticized, however, for 191 I. P ool, “ Is N ew Zealand a HealAy C ountry?’ ’ , N ew Zealand Population R e v ie w , V 6 1 .8 . N o.2, (1982): 2 -2 7 . 167 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Ignoring or nor explaining well enough the biological risks of child mortality and morbidi^ as well as the way through which such risks are influenced by human behavior. Chen & Mosley propose a new conceptual framework incorporating both social and medical sciences methodologies. Such a framework is often cited as the most comprehensive and ^stem atic conceptual framework on child survival. The framework has as its main a priori assumption the belief that all socioeconomic determinants (independent variables) of morbidity and children mortality (dependent variables) operate through a set of behaviorally mediated common biological mechanisms: the proximate determinants (intermediate variables). The proximate determinants are highly interactive basic mechanisms (a biological sy n e rg y p ro cess) which influence the likelihood of disease and the outcome of disease processes. Under optimal economic, social, environmental and biological conditions, 97 % of all children would survive the first five years of life. Child survival (or the dependent variables growth faltering and mortality) would be the cumulative consequence of multiple disease processes, including their biosocial consequences. A single disease state is not the ‘cause ‘ of death, but rather a reflection of the operation of intermediate variables or proximate determinants. 168 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Mosley & Chen’ s proximate determinants firamework identify 14 proximate determ in an ts of child survival aggregated in 5 categories: 1 ) maternal factors: age, parity and birth interval; 2) injury: accidental and intentional; 3) personal illness control: personal preventive measures and medical treatment; 4) nutrient deficiency micronutrients (vitamins and minerals), calories and protein; 5) environmental contamination: air, food / fingers / water, insect vectors, skin/sod/ inanimate objects. All maternal factors stem firom the biological interdependence between infant and mother’ s health and can be measured directly by interviews. The fi-equency of injuries in a given population would reflect, according to the M & C firamework, environmental risks, which vary according to socioeconomic contexts. Injuries are either accidental or intentional, involve poisoning, physical injury and bum s and can be measured by the record of recent incidences. In terms of the intermediate variable personal illness control, the preventive measures factor includes immunization, prophylactic actions, natal, pre/neo/postnatal care. Medical treatment relates to curative measures. Both factors can be measured by the reported use of the preventive or curative practices, by environmental analysis and primarily by surveying methods. The fourth set of proximate determinants relates to the nutritional availabilify to the infant as well as to the mother during pregnancy and lactation. 169 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The assessment is usually obtained by interviews, observation, weighing of foods or by biochemical exams. Environmental contamination involves the contamination of children and mothers by infectious agents through insect vectors, skin contamination, air (respiratory diseases primarily) and through food/water/fingers (intestinal diseases mainly). Measurement is often carried out through microbiological examination of samples of air, vectors, skin washing, water and food. The incidence of a specific disease in the cohort of children under study can also be scrutinized. Mosley & Chen observe that usually the potential exposure to disease can be approximated and measured by physical indexes correlated to environmental contamination such as household crowding (air contamination and respiratory diseases), source of water supply, use of soap, presence of toilets (intestinal diseases), household habits, etc. These data can be gathered by household survey and census information. The dependent variable in M&C's proximate determinants conceptual fi-amework combines the emphasis of medical research on diseases and nutritional status of survivors with the social science focus on the death outcome. Ideally, child survival would be estimated by the creation of a health index which would include growth faltering (and not malnutrition, since nutritional intake is not an efficient p ro ^ for morbidity) and mortali^. 170 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. This composite dependent variable would have a high explanatory power for the understanding of the health status of th.e entire population -deceased or not- and would reflect both the concerns of medical and social science, strengthening the effectiveness of public policy. "D ie socioeconomic determinants of child survival are the independent variables which intervene through the proximate determinants. In the M & C conceptual framework the socioeconomic determinants are grouped into three levels: individual, household and community-level variables. Community-level variables can be divided into ecological ^stem , political economy and health system variables. Health system variables refer to certain mechanisms which can operate at many different levels. One of the most important mechanisms of health system variables, according to this framework, is the category public education/ motivation/ information. It affects child survival by enhancing skills and attitudes, by changing resource allocation and strategies and by improving the skills of health workers. Another mechanism is constituted by cost subsidies to change the relative prices of health services and goods. The third mechanism is the existence of institutionalized actions such as epidemic control measures and immunizations. Another integrating part of community-level variables is the political economy variable that refers to the state of physical infrastructure, organization of production and political institutions. 1 7 1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The last set of community-level variables are the ones related to the ecological setting such as climate, soil, rainfall, seasonality and other physical characteristics. Household-level variables are the independent variables which operate at the household level. The conjunction income/wealth aggregates all of these variables and is often assumed to be one of the most important socioeconomic determinants of child survival, acting through various proximate determinants, hicluded under income /wealth are food, water, housing, hygienic/ preventive/ curative care, energy/fuel and information access. Individual-level variables are sub-divided in this firamework into two categories: individual productivity and traditions/ norms and attitudes. The latter considers factors which dictate health and economic individual choices according to norms and attitudes. One of the most important mechanisms is the intra-family power status within the household. Beliefs about disease causation and spread, food preferences and value of children are the other three mechanisms included in this category. Individual productivity variables are the individual independent variables determined by the time, skills (schooling as its common proxy) and health endowments of the childbearing and childbearing adult (usually the mother) and other adults (usually the father). M & C argue that for fathers, schooling levels correlate strongly with occupation and with household income as a result. 172 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Father’ s education, particularly in urban settings, correlates with health outcomes through income and consumption preference effects. The authors claim that the greater the educational advantage of fathers over mothers, the more significant wiU be the impact on child survival. With respect to the mothers, skills, tinie and health endowments would operate not indirectly but rather directly on the proximate determinants. Maternal education affects choices and increases the efSciency of health practices in contraception, nutrition, preventive care and the like. Mosley maintains that the education level of the mother influences so many proximate determinants of child survival that a so cia l sy n e rg y would exist between them. The concept of social synergy in risk factors reflects the fact that a single socioeconomic determinant can independently have an impact on several proximate determinants, affecting the risk of mortality in a manner much greater than the consolidated influence of the operation of each proximate determinant (intermediate variable). The impact of maternal education on child survival would be a strong indicative of such a phenomenon. Some authors claim that this social energy would be one of the explanations behind the limited effect of direct medical interventions, as well as the often extraordinary impact of improvements in women’ s education (and of social status, for that matter) on child survival in the absence of economic changes. 192 H. Mosley, Ibid, 1984. 193 Sudi as L. Ruzicka and J. Caldwell. 173 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Maternal health status and time would affect child survwal directly through biological links during the gestation, through breastfeeding as well as through time allocated toward prenatal care, cleaning and so on. The unpact of the maternal time endowment on child care and survival hinges greatly on women’ s social status as well as on labor and income and cultural constramts. Ruzicka observes that the translation of the M &C firamework variables into measurable indicators is often difficult and may not be possible to be done in the same field survey. Also, the treatment of explanatory theory into empirical testing requires a careful ^stematization of indicators and operational definitions. Ruzicka suggests that the investigation of the mechanism through which social synergism operates may be better carried out with anthropological methodologies and research designs (such as the ones developed by Caldwell). In addition, some of the social mechanisms -schooling as an example- act differently across countries and time. This study employs a conceptual fiamework which is influenced by Mosley & Chen’ s. It attempts to measure and quantify infant mortality decline in Brazil’ s Northeast region through the analysis of survey data (1986, 1991 and 1996 Demographic & Health Surveys) In addition to a dichotomous dependent variable and to birth cohort variables, a set of independent variables was conceived. Such variables include family-level demographic as well as household socioeconomic variables. L. Ruzicka, Ibid.1989. 174 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The demographic and. fami^ independent variables are : sex of the index child; birth order of the infant child; breastfeeding status of the infant; age of the mother at the time of the birth of the child; prenatal care of the infant by a physidaa; DPT immunization; ethnicity of the mother. The socioeconomic and household independent variables are; mother’ s educational attainment; goods ( a p ro xy for household income); household crowding; sewage disposal; source of drinking water; place of residence (urban X rural). 175 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. To measure is the fir s t step to improve ' (Sir W illiam P etty, 1654) 176 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 5 CONCEPTUALIZATION OF VARIABLES Birth. Cohort Variables For each survey year and for each data set, the birth cohort variables were devised in two different ways: a) a cohort variable takes four values firom 0 to 3. For instance, for the 1991 DHS survey applying logistic regression, the children bom between 1985 and 1990 are assigned a value 3. For the 1980-1984, 1975-1979 and before 1979 periods, the values are, respectively : 2, 1 and 0. b) 3 sets of dichotomous birth cohort variables are constructed, taking the value 1 if the child was bom within that specific time period and 0, otherwise. The older time bracket works as a dummy variable. Sex of the index child The first demographic covariate included in the model was the sex of the index child. The index child is the mother’ s latest child. Even though there is no theoretical evidence linking gender and infant mortality, the inclusion of the former has been quite common in studies of child survival in Brazil. 177 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In 1980, Brazil’ s total infant mortality rate was 65.8. The male and female infant mortality rates were 72.8 and 58.5, respectively (male excess infant mortality of 24%). Ih the Northeast the same rates were 106.8, 114.6 and 98.8 (male excess infant mortality of 16%). In 1991, Brazil’ s total infant mortality rate had fallen 22% to 51.6. The male and female infant mortality rates had decreased 19.5% and 24% to 58.7 and 44.3, respectively (male excess infant mortality increased to 32%). In Brazil’ s Northeast region the same rates had fallen 17%, 16.5% and 18.5% to 88.2, 95.6 and 80.6, respectively (male excess infant mortality increased to 18 % ).W 5 According to the figures released by BEM FAM based on the 1996 DHS survey, Brazil’ s total, female and male infant mortality rates are 48, 44 and 51 (16% excess male infant mortality). Preston (1976) argues that the male excess mortality is often greater in infancy, Data on the sex of the index child were obtained without difficulty firom the Reproduction section (birth history) of the surveys. The original codes for male and female were recoded from 1 and 2 to 0 and 1, respectively. The variable was named sex.1 for the most recent child a woman gave birth to. All data relate to the number of yearly deaths of in&nts aged 0-12 months per 1,000 births. Confuted firmn IBGE’s censuses of 1980 and 1991. ^ Samuel Preston, 1976. 178 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Mother's age group at birth Maternal age or mother's age at the birth of the child is often considered one of the most important determinants of infant mortality. For very young (15- 19 age bracket) as well as for older women (35-49 years of age), the odds for the survival of the child are much less favorable. Behavioral and strictly physiological problems would contribute to increasing the risks of infant death. Data on this demographic covariate were obtained firom the Respondent's Basic Data section (Individual Recode). The current age in completed years is calculated firom the century month code (CM C) of the date of birth of the respondent and the CMC of the date of the interview. In some cases the age reported by the respondent was different than the age in the data file, when the respondent's birthday was in the month of the interview but after the latter. In such cases the age was rounded up by the interviewers. From the variable moüier's age (mothroge), and the variable mother's age groups {m otaggro, which includes 7 categories in 5 years increments firom 15 to 49 years old), another variable was created, the age risk for the mother [agem tr2). The moftiers were divided, according to their respective ages, into 2 categories, high and low risk. The low risk factor, labeled 1, includes the intermediate age groups (20-24, 25-29 and 30-34) and the high risk factor, labeled 0, includes the teenager age bracket (15-19) as well as the older women age groups (35-39, 40-44 and 45-49). 179 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Birth order Birth order or the rank of the child is often, indicated as one of the most important physiological determinants of infant mortality. Initially the measure employed for the number of children a woman has ever had was another demographic variable, the variable p a rity , derived firom the Reproduction section’ s c/uZdbom variable. If the respondent had more than 3 children, the parity was considered high and labeled 0. If the mother had 3 or less than 3 children the parity was assumed low and labeled 1. According to the summary report put forth by BEMFAM based on the 1996 DHS survey, mortality risks for children of women who had more than 3 children are significantly higher than the children of women who had 3 children or less. Demographers have pointed out the inverse, U or J relationship between infant mortality risks and parturition levels, Sastry (1997) maintains that high parity births are clearly associated with higher infant mortality levels in Brazil’ s Northeast region. J. Trussel and C Hammerslough, A Hazards-Modei Analysis o f die Covariates of IhAnt and Child Modality in Sri Lanka, Demography 20, no. 1 (Feb. 1993). Celso Sûnôes and luri Leite, “Padrao Rqirodutivo, Serviços de Saüde e Mortalidade infimtil- Nordeste”, Pesquisa sobre Saude Familiar no Nordeste, 1991. Narayan Sastry, “Family-Level Clustermg of Childhood Mortality Risk in Nordieast Brazil”, Population Studies 51 ^ o v . 1997): 245-261. 180 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Since mortality risks for the first newborn are higher than for the succeeding children and since child spacing also plays a rather important role in determining child survival chances, instead, of p a rity we adopted b irth o rd er as our demographic variable to explain both level and timing of parturition. For biological and socioeconomic reasons, first and higher order births ( four and more) are usually characterized by higher infant mortality rates. 2 0 0 Among first-time mothers the greater risks of infimt mortali^ would be associated with physical, socioeconomic as well as p^chological factors. The strength of the higher birth order effect is less dear cut (than the first birth order effect) and would vary substantially firom country to country, indicating that socioeconomic factors would be more important than strictly biological ones in explaining the greater infant mortality of high rank births. 201 Since the relation between birth order and infant mortality is U-shaped it is common to define four reference categories: first child, 2-3, 4-6 and 7+. According to the 1996 DHS summary report published by BEM FAM , Brazil's infant mortality for the 2-3 category would be 44 increasing to 86 for the 7+ birth order level .2 0 2 in the Northeast the mortality differential according to birth order would tend to be even higher. ^ J. Hobciaft, J. McDonald, J. and S. Rutstein ( Ibid ) assert, however, that, Wiile the mortality of first-born diiidren is particularly h i^ e r durmg the first year of life, the greater mortality o f higher birth orders may be due to otiier fectors sudi as spacmg patterns. S. Rutstem, J. Sullivan and C. Bic%o, “bfiuit and Child Mortality”, DHS Conq>arative Stucty 15 ( Calverton, June 1994). riortothesurvty. 181 For the ten year period prior to the survty. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. From the birth order variable (b irto rd l), another variable was devised: the b o d rrk l variable or the birth order risk for the most recent child which assumes the variable 0 for high risk and. over 3 « * births) and 1 for low risk levels for the child (2“* and 3 "^ births). 2 0 3 Urban and Rural Residence This socioeconomic variable has been included in many studies of infant mortality risks. 2 0 4 The socioeconomic conditions in urban and rural areas are often quite different, particularly in developing nations. With industrialization and modernization the differences tend to be substantially reduced, but in an highly underdeveloped area such as Brazil’ s Northeast there are still significant socioeconomic contrasts between rural and urban areas. One may question the impact the differences of geographic residence would have on infant mortality. 2 0 5 The odds of dying during infancy are usually higher in rural areas than in urban ones. This discrepancy would be due not only to the better provision Similarly, for the second most recent child, the variables are biitord 2 and bodnkZ. B. Bradley and R. Jdinson, “Socioeconomic Classifications for die Sturh^ of Mortality Differentials”, Proceedings of die Meetings on Socioeconomic Determmants and Consequences of Mortality (Mexico Cfty, June 1979). ^ Ihiited Nations, Department of Ihtemational Econcxnic and Social Affeirs, Child Mortality in Devel(^ing Countries: Socioeconmnic Differentials, Trends and Implications (New York, 1991). 1 8 2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. of health services 2 0 6 in urban areas but to the interaction between the latter and other socioeconomic variables such as sanitation, piped water and mothers’education. 2 0 7 Since living conditions in rural areas of Brazil’ s Northeast are often much harsher than in urban ones, it would be expected that the risks of infant death in the latter areas to be lower than in the former. Based on the 1996 survey, BEMFAM/DHS infant mortality for Brazil’ s was computed as 48. In urban areas this rate would be 42. In Brazil’ s rural areas infant mortality would be 55% higher, or 65. 20s IBGE’ s ofEcial definition of an urban center relates to the localization of dwellings and people in judicially independent cities, town or villages, according to municipal law. 209 The DHS surveys follow IBGE’ s terminology and categorization of place of residence. The data on urban and rural residence were collected firom both the Household According to the figures released by BEMFAM based on the 1996 DHS survey, 78% of all Brazilian duldren bom withm five years before the su rv ^ were assisted by a doctor during childbirth. In foe rural areas o f foe Northeast foe rate of doctor assistance during birth fidls to 55%. A stufor evaluatmg foe socioeconomic determinants of duld mortality m Kenya finind that childrm vfoose parents have little or no education and ifoo live in rural areas are mudi more likely to die at an early age than otherwise. See K. Venkatadiarya, K., Child Mortality m Developmg Countries, Socioeconomic Differentials in Child Mortality' - Kenya , United Nations, Department o f International Economic and Social Affeirs (New York, 1991). Rates per 1,000 for foe ten years prior to foe survey. The overall infimt mortality for foe Northeast was calculated as 74, but there foe breakdown by place of residence was not released. ^ That is, foe official urban definition is also ge% r^hic and administrative. Often centers gathering a population much smaller than 20,000 are assumed to be urban. Somethnes municipalities with only 2,000 irfoabitants are assumed to be urban, according to such crheria. 1 83 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Data and the respondent’ s Basic Data. The nrhxrur variable was simply recoded firom the de facto place of residence variables and the labels changed to 0 (rural) and 1 (urban). Drinking Water The Northeast of Brazil is a region in which socioeconomic development is lacking and breastfeeding levels are particularly low. The availabili^ of piped water for drinking and nondrinking purposes becomes vital, since access to good water dramatically improves the health and hygiene of the infant. Untreated drinking water and contaminated food disseminate bacteria transmission, increasing diarrhea incidence and ultimately endangering the life of the infant. As some authors have pointed out, 2 1 0 survey data on piped water often underestimate the impact of the latter on child survival since, firequently, the access to water is erratic or deficient. Measures of drinking water are indeed crude measures. The selection and creation of the d rin k w a 2 variable was done in stages. The original data firom the surveys characterized the access to water through several different categories: well water, rainwater, bottled water, surface water, dam, pond, piped water into household, piped water into yard, tanker truck, etc. See John Marcotte, “The kq)act of Pÿed Water on Child Mortality and Breastfeedmg in Ecuador and Brazil”, (PhJ). dissertation. University of Wisconsin-Madison, 1988). 1 8 4 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The categoiizatioii also changed across the surveys. An intermediate variable - d rin kw a t- was generated. The water supply was dmded into 4 categories: 1- piped water (including piped water into household, into yard and into public tap); 2- public water (including well, spring, river, dam, rain and pond); 3- other (including tanker truck, bottled water and other nonpublic water); 4- missing values. The d rin k w a t was then recoded into a new one: drm kw o2. The new labels are 0 for water not good for drinking and 1 for water good for drinking (labeled 1 and 3 under the previous variables). Sewage Low quality sanitary conditions, that is, missing or deficient sanitary facilities, are oftentimes cited as risk factors undermining child survival, Modem sanitary facilities and a clean water supply are epidemiologically associated with lower mortality rates. 2 1 2 Working with household data firom Malaysia, DaVanzo (1988) 213 concluded that improvements in sanitation as well as in water supply and in United Nations, Socioeconomic Differentials in Child M ortally in Developing Countries, E.85. Xm.7 (New York, 1985). R. Serrano and C Puffer, Caracteristicas de la Mortalidad en la Ninez, Pan American Healdr Association (Washington, D C , 1973). Julie DaVanzo. Ibid. 185 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the mother’ s education, levels are the most important factors accounting for both regional and over time inverse changes in infant mortality. 2 1 4 Investigating infant mortality decline in the Northeast through the analysis of causes of death, Simoes maintains that recent improvements could be traced to the relative decline of avoidable causes related to first infancy, respiratory infections and sewage. This author argues that further gains in infant survival in the Northeast would be fostered by investments in public health and sanitation. 2 1s hi stark contrast, many studies on infant mortality in Brazil claim that the impact of sanitation facilities on infant m ortaliy is neither substantial nor significant. 2 1 6 Modem sanitation facilities are present in less than 50% of the Brazilian households according to the 1996 DHS. In the Northeast region, however, the share of households with adequate sanitation is about 25%. In Brazil’ s rural areas the absence of household toilet facilities would be the norm in almost two thirds of the households. In some rural areas of Brazil’ s Northeast, the la actually, increases in maternal education would have a stronger in ta c t on infimt mortality rates when tenq>oraI decline is scrutinized, vdiereas regional dififermces m water and sanitation would be relatively more in^ortantwhen t%ional IMR differences are to be e?q>lained. Celso Simoes, Ibid., 1995, pg. 130. See among others; - Thomas Merrick, Ibid. - C.G. Victora, J.P. Vaughan and P. Smith, Ibid.. - Narayan Sastry, N. Goldman and L. Moreno, “ The Relationshÿ between Place of Residence and Child Survival m Brazil”, hitemational Population Conference, no. 3, lUSSP (Liège, 1993): 293- 322. 186 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. presence of modem, sanitation facilities is very restricted. In such, areas, infant mortality rates tend to be much higher. The selection and creation of the se u x ig e 2 variable was also done in stages. The original survey questionnaires labeled the type of toilet facilities through several different categories: own flush toilet, shared flush toilet, toilet to sewer, toilet to open space, to lake, to river, latrine to sewer, latrine no- connected, traditional latrine, traditional pit toilet, bush, no facilities at all. 2 1 7 Through an intermediate variable, se w a g e , the type of sanitation facility was classified under 4 categories: 1- flush toilet (own, shared, to sewer, to open space or river); 2- pit toilet latrine (traditional and non-traditional); 3- no facility and rudimentary facility; 4-missing values. Taking the values 1 (modem sewage, previous flush toilet) and 0 (not modem sewage, other categories), the new sanitation variable se w a g e 2 was then devised. The mean values for modem sewage conditions in the Northeast for 1991 seem to be excessively high. In 1986 the distribution of modem X non modem sewage conditions were respectively 12.2 % and 87.8%, according to the 1996 Logistic Regression data set, and 11.6% and 88.4%, according to the Cox Regression Data set. In 1996, according to the Logistic Regression Data set, 23.8% of the Northeastem households had modem sewage and 76.2% had not. The Cox Regression data set indicate similar values of 23.1% and 76.9%. 217 The labeling also dbanged across the three DHS su rv is. 187 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In 1991, however, 63.2% of the selected Northeastern, households for the Logistic Regression data set had modem sewage and 36.8% had not, while for the Cox Regression data set, these rates were , respectively, 61% and 39%. Since there were no computation or recoding errors, and to prevent spurious results, aU the final regressions were run with and without the sewage variable. Household Crowding Measures of household crowding are used to indicate exposure to health hazards such as infectious and respiratory diseases. Cramped dwellings facilitate the spreading of diseases. Crowded sleeping and living conditions tend to be associated with reduced chances of child survival. This household-level variable would ideally correlate the number of rooms to the number of people living in the household. Questions regarding the number of rooms were present in the DHS 1 1 1 survey but not in the previous two surveys. Thus, instead of dwellers per room, the overall number of people living in the household was used as a measure of household crowding. Another adjustment was made due to the interaction between the number of living people in the household and the survival status of the child (the dependent variable). If the child survived, 1 is subtracted firom the household crowding variable. If the surviving children were to be included in the household crowding variable, the final effects would alter the statistical 188 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. significance of other variables. The numerical variable created, h h crS , is the household crowding variable that will be adopted in the statistical analysis. For comparative purposes another household variable was devised, hhcr3 2. This dichotomous measure of household crowding is labeled 0 and 1, respectively. Households with five or more individuals are considered to be crowded, posing a negative impact on infant mortality, while households with less than five individuals are considered to be non-crowded. Ethnicity Both the DHS codes and questions concerning ethnic background are specific to Brazil. Race is self reported in the surveys. Brazil has an ethnically diverse society: 55% of the population is reportedly white, 8% of Afidcan descent, 3.5 % Asian, 1.5% Indian and 32% mixed. This distribution, however, varies a great deal across states and regions. According to the data computed from the 1991 Northeastern survey, 71% of the 3842 women who have ever had a child were of mixed race, 8% black and 21% white. For the 1996 survey, from the 3875 mothers surveyed in the Northeast, 67% were of mixed race, 5% black and 29% white. In contrast, among the 1094 women surveyed in the South region for the 1996 DHS, only 2% were black, 30% mixed and 68% white. By mixed it is understood the ciossmg between the races white, black, yellow and Indian. The Brazilian categories are pardo, mulato, caboclo, cafiiza, mameluca, mestiço or moreno. 189 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Many authors believe that for certain countries there is a strong association between ethnicity and infant mortality. 219 Other authors observe that the influence of ethnid^ on mortality would be greater in countries in which the differentiation between races is stronger. 2 2 0 E)ven th o u ^ ethnic differentiation is not clear-cut in Brazil, an intense social and economic polarization is evident across races, having white individuals on one side and black and mixed race individuals on the other. From the original 5 categories white, mixed, black, Asian and Indian (e th n ic it variable), another binomial variable was created: e th n ic i2 taking the value 1 for white and 0 for non-white (all other 4 races). 2 2 1 if the five categories were to be maintained, the standard error of the estimates would increase and we would lose degrees of freedom. Dichotomizing this variable is usually standard in Brazil and it does not imply in any considerable loss of information. The ethnicity variable was not included in the 1986 models since ethnic origin questions were not present in the first phase of the DHS’ s in Brazil. See among others J. Trussel and C Hammerslough, Ibid. ™ United Nations, Socio-economic Differentials in Child Mortality m Developing Countries, (New York, 1985): 86. Since edinic classifications vary widely fiom country to country, this researdi decided to use the broad white and non-^^ute di&rentiation. P. Hauser and E. Kitagawa adopted a similar classification in Differential Mortality^ in the U.S.: A Stucfy m Socio-economic ^idemiology (Cambridge, 1973). 190 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Education Attainment Education, of the mother is considered to be one of the most important socioeconomic determinants of child survival. The level of formal maternal education is inversely correlated to child mortality. 2 2 2 Several studies on Latin American mortality substantiate this negative relationship. 2 2 3 Maternal education contributes directly to a greater health care knowledge. Maternal education is also an important factor in the use of medical facilities. It would also be associated with better nutrition and personal hygiene. 224 The impact of the mother’ s education, however, goes beyond health knowledge and relates to a process of change in values, beliefs, preferences and resource allocation. Education leads to social change and modernization. 2 2 5 In addition, maternal education is also correlated with higher income and other socioeconomic indicators as well as to quality of life. ^ United Nations, Department o f International Economic and Social Af&irs, Child Mortality in Developing Countries: Socioeconomic Differentials, Trends and Duplications, Ibid. “ See Hugo Befam, Ibid. F. Shorter and T. Belgm, Determinants o f Child Mortality: A Study o f Squatter Settlements in Jordan” in Child Survival - Strategies for Research, edited by Henry Mosley and L. Chen, Pc^ulatim and Develc^ment Review 10 (1985). ^ See among others: - J. Caldwell, P. Caldwell and P. Reddy, “The Social Cmnponent o f Mortality^ Declme: An hivestigation in South India enq>loying Alternative Methodologies”, Population Studies 37, no. 2, (July 1983): 185-205. - J. Caldwell, “Education as a Factor m Mortality Declme: an Exammation of Nigerian Data”, Peculation Studies 33, no. 3 (1979): 395-413. - K. Stratfield, M. Singarimbun and I. Diamond, “Maternal Education and Child Dnunization, Demography 27, no.3 (1990): 447-455. 191 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The DHS surveys gather information, on schooling through several questions which quantify the highest level of school attended or the number of years attended in school. The independent variable used in this study - ecLgroup- measures education in single years, assigning the values 0, 1, 2 and 3 for no education, primary, secondary and higher education, respectively. As noted by the DHS program, the number of years that constitute primary education vary from country to country. 2 2 6 in Brazil primary education includes 8 years of schooling. A new dichotomous variable for the education level of the mother was created, ed.gr2, which assumes the value 0 for low education levels- no education and primary education- and 1 for h i ^ education levels- secondary and higher. However, to prevent loss of valuable information, the covariate adopted to measure the mother’ s educational level was etL graup. The education level of the father has also an inverse impact on the mortality of the child but is not assumed to be as determinant a factor as the maternal education. In addition the amount of missing data was considerable. Father’ s education was not included as one of the covariates in the model. As an alternative to maternal education, literacy levels are also inversely correlated with infant and child mortality. 226 S. Rutstein, J. Sullivan and G. Bic%o, Ibid. 192 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Illiteracy levels in Brazil and particularly in the Northeast region are extremely high, in spite of the recent gains. The United Nations Educational, Scientific and Cultural Organization (Unesco) considers literate a person who * can, with understanding, both read and write a short, simple statement on his or her everyday life. * Preston (1978) concluded, after studying 120 countries around 1970, that an increase of 10% m literacy levels would raise life expectancy at birth in over two years. 2 2 7 The impact of literacy on mortality would fluctuate according to the age bracket. Palloni (1981) points out that in Latin America the impact of literacy on child mortality would be greater than on infant mortality. 2 2» Literacy questions with yes’ and ^ o ’ answers would ideally be followed by additional questions or by cards with test sentences. The DHS standard literacy question is usually asked to women who attended primary school. Otherwise the mother is considered to be illiterate. The variable literacy evolves from the following question: * Can you read a letter or newspaper easily, with difficulty or not at all?” The values 1, 2 and 3 are assigned for each of die three answers. ^ Samuel Preston, “Mortali^, Morbidity^ and D evelc^m aifPaper presented to the Semmar on Population and Development in the ECWA Region (SqX. 1978). Mentioned m Proceedings o f the Meeting m Socioeconomic Determinants and Consequences of Mortality (Mexico City, 1979): 154. ^ A. Palloni, "Mortality m Latm America: Emeigmg Patterns”, Population and Development Review 7, no. 4 (1981): 623-649. 193 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Another covariate was then, created, lxtera2, assuming the values 0 - or illiterate- if the respondent cannot read or reads with difficulty, and 1, otherwise. Nevertheless, since literacy level is quite correlated with educational attainment of the mother and since, from the data, it can be shown that the latter is a stronger determinant of infant mortality levels, preference was given to the schooling variable and the literaQr variable was not included in the model. Goods In spite of the fact that the relative importance and the mechanisms through which income and wealth affect infant mortality are controversial, higher levels of household income have been often associated with lower infant mortality. 2 2 9 Higher per capita income would lead to better living conditions, general hygiene and nutrition intake. Ideally the total household income would be used as the income covariate, but the DHS surveys did not include questions on real income. The father's employment status has been used as a p ro x y for income but that would depend on a further investigation of marital status of the mothers. ^S ee: - Thomas Merrick, Ibid. ' D.Thomas, J. Strauss and M. Henriques, Ibid. 1 9 4 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The relationship between income and infant mortality is not straightforward and is characterized by many ambiguities and measurement problems. Since specific data on income were not collected in the surveys and since some studies maintain that it is disparities in wealth rather than income that most determine disparities in health outcomes, this research will use different a proxy for household consumption and wealth levels based on ownership of three important household goods which have also been associated to improved child survival through a better access to health care and information: car, radio and TV . 2 3 0 Each of the three goods is worth one point for a maximum of three points. Prenatal Care bv a Phvsician It is well accepted that there is an inverse relationship between an effective access and use of health services and infant mortality. Prenatal consultation with a doctor - d r2 .1 variable- can be shown to be a very significant determinant of infant mortality levels in the Northeast. Pre-natal consultation with a nurse was also considered but not included in the model for prenatal care by a physician is a much stronger determinant of infant survival rates. ™ For a similar approadi see B Fidzani, "Socioeconmnic Determmants of Ihfimt Mortality in Botswana, 1978-1988”, USC (M ay 1996). 195 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The great amount of missing data in the original variable d rp m cu l ( a yes/no variable) was the reason why another independent variable was created, the d r 2 .1 . The missing data were assumed to be a negative answer, or no visits to a doctor. However, the dichotomous variable d r 2 ,l was not present in the 1986 survey, because no questions on the subject were present in the interview. Immunization Levels (DPT 1. 2. 3 1 Another important public health independent variable used in this research is immunization. This variable aims at capturing the impact of public policy - and particularly of vaccination programs promoted by the government - on child survival. Data were collected on whether the infant received or not the three shots of DPT ( 1, 2 and 3). I f positive, the d p tl2 3 variable assumes the value 1. When the child does not receive all three dpt immunization rounds and when the data are missing, a 0 value is assigned. There are no also data on immunization for the 1986 survey, so this variable was not included in the 1986 data sets. Breastfeeding The survival of the infant depends a great deal on the child’ s capacity to absorb nutrients and resist to diseases. 196 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Breastmilk significantly enhances survival chances by providing major nutrients, and by boosting the child’ s immune ^stem with maternal antibodies. In addition to this extremely important effect, mcreasing breastfeeding levels also have a positive impact on infant mortality through two other effects. Breastfeeding reduces the child’ s exposure to water and food borne diseases and infectious agents that would be consumed otherwise. Analogously, improvements in the access to piped water have been associated with a reduction in breastfeeding. 2 3 1 Prolonged breastfeeding would also extend postpartum anovulation and lengthen the spacing between births. Longer birth intervals, in turn, would also be beneficial for optimal breastfeeding. 2 3 2 As pointed out by Curtis et al (1993) , 2 3 3 birth interval effects are very significant in Brazil and particularly in the Northeast where child spacing is very restricted. The covariate months of breastfeeding, b f.lm th s , is one of the most important independent variables included in the model and it is the only time varying one. It takes the value of the number of months the child was reported as being breastfed by the mother. Jdm Marcotte, Ibid. A. Pailoni and S. Millman, “E fiE bcts o f loter-Birdi bitervals and Breastfeeding on Inânt and Early Childhood Mortality, Population Studies 40 (1986): 215-236. S. Curtis, I. Diamond and J. McDonald, “Birth Interval and Family Efibcts on Posmeonatal Mortality in Brazil”, Danognq>hy 30, no.l (February 1993): 33-43. 197 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ' For so it is, oh L ord m y G od, I m easure it; but w hat is it t h a t I m easure? (St. Augustine, C onfessions) 198 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C huter 6 DATA AND M ETHODS D ata Source The data for the analysis of infant mortality in Brazil’ s Northeast region originated in the three surveys promoted by the Demographic and Health Surveys (DH S) and conducted by the Civil Society for Family Welfare and Planning in Brazil (B E M FA M ): the 1986 National Research on Family Planning and Health (PNSM IPF), the 1991 Research on Family Health in the Northeast (PSFNe) and the 1996 National Survey on Demography and Health (PNDS). Census data by the Brazilian Institute of Geography and Statistics (IBGE) will also be used for the analysis of long term trends, but the main thrust of this empirical research wül rely on the DHS survey data for the 1986-1996 period. The sample surveys are administered as part of a periodic program maintained by DHS through which a probabilistic sample of households is drawn from the census universe for separate questioning. The DHS surveys collect specified retrospective information firom a sample of the population through a much longer, thorough and deeper interviewing process than the IBGE national censuses and household surveys (P N A D ’ s). The DHS sample surveys also include questions, issues and specific questionnaires not examined by IBGE’ s household and census data. 199 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The three phases (DHSI-H H I) allow for the exploration of more questions pertaining to infant mortality than the censuses since they compile an enormous amount of socioeconomic-institutional, demographic-behavioral and health data. The DHS/BEMFAM program collects data at the household, community and individual levels (female respondents). Unlike most of the WPS surveys, the DHS surveys are not restricted to ever-married (B M W ) samples, involving all woman at reproductive age in the surveyed household. The surveys follow the general DHS policies on sampling design: they are supposed to be as simple as possible, they use pre-existing sampling universes, samples are self-weighting and scientific probabilistic sampling is always used. 2 3 4 While the universe of the 1986 survey (DHS I) covers all women age 15- 44, the following two surveys expanded the reproductive age upper limit to 49. DHS I and H I are national surveys, gathering data on all of Brazil’ s five great regions, whereas the 1991 survey is limited to the 9 states of the Northeast region. The sample coverage of the DHS I was 95% of the national territory since it excluded some rural areas of the North and Central-West regions. In 1996 the DHS m covered 100% of the Brazilian territory. ^ See Samuel Baum, ed., "DHS Sampling Policy”, Readings m Population Research Methodology (New York, Vol. 1, Chapter 3, 9): 41. 200 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The 1986 survey was conducted between the months of May and August and was comprised of a total of 5,892 complete interviews, 30% of each (1784 women) were carried out in the Northeast. The 1991 survey was conducted in Northeastem Brazil between the months of August and December. It surveyed 6,064 households and its sample size was 6,222 women. The 1996 DHS survey interviewed 12,612 women age 15-49 in a total of 13,283 households between the months of February and May. 4,772 or 37.8 % of these women were firom Northeastem states. Thus, the original sample sizes of the survey data used in this research were 1784 (30% of the sample), 6,222 (100% of the sample) and 4,772 (37.8% of the sample), respectively for the 1986, 1991 and 1996 DHS su rv is. Sam pling D esign DHS’ sampling universe is selected out of a sub-sample of IBGE’ s General Household Surveys (PN A D ’ s). The clustered, multistage sampling design of the surveys is based on selecting Prùnary Sampling Units (U PA ’ s) firom the census files. The UPA’ s are stratified and grouped by place of residence (rural/urban) and state of origin. The geographical stratification allows for creating strata with lower intemal variability. The sampling error will depend not on the population variance between stratas but rather on the variance within a given strata. 201 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. DHS’ stratification process attenuates sampling error and yield more precise estimates. DHS surveys are probabilistically constructed in two stages; in a first stage the census groups, primary sampling units and clusters are conscripted, taking into consideration the probability and stratified proportion of each, firaction of the sampling total; in a second stage the households within each census group are randomly chosen. The selection of the census sectors for the DHS surveys is proportional to the selection probability of IBGE’ s censuses and household surveys (PN A D ’ s), resulting in a sub-sample with probability proportional to the size and characteristics of the original sample. For each Brazilian state a rate of sub-sampling ( s ) was established by the division between DHS’ number of census sectors ( a ) and the total number of census sectors firom IBGE’ s PNAD’ s ( b ). s = a / b The selection probability for the ith census sector in IBGE’ s P N A D (Pli ) is given by the expression: Pli = (b *mi)/(Smi) 202 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. where: mi = Number of households in the ith census sector from the (decade*s) national census. Z mi = Number of households in the total selected census sectors according to the (decade’ s) national census. The selection probability for the ith census sector in the DHS survey ( P*li ) is given by the expression: P*li = s ( b * mi ) / (Z mi ) = (a * mi) / (Z mi ) where: a = Number of census sectors from the DHS survey. mi = Number of households in the ith census sector according to the (decade’ s) national census. Z mi = Number of households in the total selected census sectors according to the (decade’ s) national census. The number of households selected for the DHS survey within a given census sector is denoted as (ni), whereas the number of households selected within a given census sector for the previous PN A D is denoted as ( Li ). 203 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The household selection internal probability for the ith census sector (P*2ij ) in the DHS survey is given by the expression: P*2ij = ni / Li The total sampling fraction for each census factor in a DHS survey ( F ) is then calculated as the product of the selection probability for the ith census sector ( P*li ) and the household selection internal probability for the ith census sector ( P*2ij ). F = P*1I *P*2ij and, from this expression: ni = ( b * F * Li ) / ( a * P li ) where: ni = The number of households selected for the DHS survey within a given census sector. a = Number of census sectors frnm the DHS survey. Li = Number of households selected within a given census sector for the previous PNAD. b = Total number of census sectors for the previous PN A D . 204 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. P li = The selection probability for the ith census sector in the P N A D = (# Municipalities selected in the P N A D ) * ( Probability of selection of a municipality in the P N A D ) * ( # of census sectors selected in the municipality in the P N A D ) * ( probability of selection of a census sector in the PN A D ). F = The total sampling fraction for each census factor in a DHS survey D ata M anipulation Using a standard DHS core questionnaire appended with questions specifrc to Brazil, and consisting of a wide range of retrospective questions concerning infant and maternal mortality, contraception, fertility, health indicators and behavior as well as household and community variables, the DHS survey data are expected to offer independent and highly representative estimates. In an preliminary phase of the data analysis process, the Civil Society for Family Welfare and Planning in Brazil (BEM FA M ) was contacted so that information on the design and administration of the surveys could be gathered. The original reports for all three surveys and the original questionnaires in Portuguese were collected and scrutinized. Macro International, the private institutional responsible for the development of the Demographic and Health Survey, was then contacted and a authorization was given to download the raw data sets. 2 3 5 The downloading and manipulatian of the original data files were time consuming due to mormous size of the data sets. 205 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The data are formatted in rectangular files (fixed number of records for each case in the data file) designed to be used with the Statistical Package for the Social Sciences (SPSS). Due to its power and convenience, this software package was adopted for use in this research. The two most important data file types are the Household Raw Data and the Individual Raw Data. The data sets are presented in a zipped format. Once the data sets were unzipped and transformed from its raw form into actual data, six laigge data sets were selected: BRIROIRT and BRHROIRT (DHS I - 1986); BRIR21RT and BRHR21RT (DHS H -1991); BRIR31RT and BRHR31RT (DHS m - 1996). After investigating the DHS model questionnaires, the coding standards, the different sections and variables, the variables labels and descriptions and the rationale for recoding, the household and the individual (standard recode) level variables or the two basic data types were merged onto the CASEID and the HH numbers. For each individual female respondent, the respective household information was appended, creating three mega data sets. Often the same variables were either labeled or coded differently, which created problems during the merging process. In addition to these three data sets, a complete single data set was created from the merging of the three. Data regarding to respondents fi*om the Center-West, Southeast, South and North regions were discarded. 206 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The sampling sizes for the four Northeastern data sets ( both 86, both 91, both 96 and the 86-91-96) were then substantially reduced to 1,791 (30% of total survey), 6,222 (100%), 4,772 (37.8% of total survey) and 12,785 respondents, respectively. The next stage involved the filtering of all cases or all women who had never had a child: 37%, 38% and 19%, respectively for DHS I-II-in, of all female respondents at reproductive age in the Northeast have never given birth. Such individuals were also discarded for the purpose of this research. The final sample size for 1986, 1991, 1996 and the aggregate 1986-91- 96 data sets for Brazil’ s Northeast after these adjustments are: 1,125, 3,843, 3876 and 8,844. In addition to these four main data sets two other ones were created: 1986-91 and 1991-96. Another stage of the data operationaUzatiorL focused on integrating the information contained in several relevant variables, creating uniform labels and recoding them into new variables. In most cases the new variables were defined in a simple and binary manner to allow for an immediate understanding of the impact of the covariates on infant mortality. The label 0 usually indicates an impact that would be theoretically considered deleterious to child survival, and the label 1 would signal an opposite influence for the covariate. Some independent variables, however, are not dichotomized since that would implicate in loss of information and degrees of fi%edom. 207 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. These variables are education attainment of the mother, household crowding and goods (and breastfeeding in the case of survival analysis). Urbanization, drinking water, sewage, birth order, mother’ s age at birth, ethnicity and prenatal care by a physician, immunization and sex of the child are presented in a binary manner. Recoding and restructuring the independent variables made the data suitable for the methods of analysis which will be employed in this investigation. The data sets were then edited to eliminate variables of no direct interest to this empirical research such as contraceptive and fertility preferences. One of the main difficu lties of the data analysis was dealing with missing data. Missing data was a problem not only because often respondents refused or did not know how to answer certain questions (nonresponse and omission), but most notably because some variables present in the some surveys are not present in others at all. As far as the variables relevant for this research are concerned, this is particularly the case for the data on inununization. Small missing rates in some covariates can sum up to a substantial loss of data since the respective entire ca seid . could be deleted from the statistical analysis. The strategy adopted was to try as much as possible to avoid discarding missing data, an active strategy. In many situations the missing data were associated with the data expected not to present a positive impact on survival chances (or to a value 0 to the variable). 208 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. However, when, a question or a variable supposedly relevant for the model is lacking on a given survey year, the strategjr adopted for dealing with the missing data was passive, either excluding the missing data altogether or removing the associated variable firom the model for that year. According to the estimates calculated by BEMFAM from the 1996 DHS survey, Brazil’ s neonatal (N N ), infant (IqO) and child mortality (4q0) are, respectively, 22, 48 and 9 (per 1,000). In the Northeast region, however, neonatal mortality rate is 28, infant mortality* 74 and child mortality 16. infant mortality rate is much higher in the NE vis-à-vis the rest of the country. These rates are still a stark contrast with other regions such as the South region where mortality rates are, respectively, 15, 25 and 5. Most of the deaths in the Northeast region occur during the first year of a child’ s life. Based our own computation of infant mortality through the creation of a new dependent variable and the usage of two methods of analysis for the three DHS surveys, the data show, indeed, that less than 10% of children’ s deaths in the N E occurred after the first year of life (in 1986 this proportion was 6%, and in 1991 and 1996 it increased to 9%). 2 3 6 In addition to creating the binary d ea d X a li dependent variable, cohort variables were introduced to qualify the different risks associated with infant mortality through the years. This variable was introduced in two different ways. In the first, a cohort variable assuming 4 values was devised. For instance, for the 1996 data set, if the child was bom between 1990 and 1996, the value is 3. 236 Computed only for the most recent child of each respondent. 209 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Between 1985-1989 and 1980-1984, the values are respectively 2 and 1. For children bom before 1980, the value is 0. Another way the cohort variable was introduced was through the creation of 3 time cohorts dichotomous variables : 90-96, 85-89 and 80-84. Each of them takes the value 0 if the child was bom outside of the reference time frame and 1 otherwise. The before 1980 period works as a dummy variable m the 1996 data set. The original 6 data sets for Brazil’ s Northeast region (1996, 1991, 1986, 1996-1991-1986, 1996-1991 and 1991-1986) were tumed into 12, since the two methods used in this study, survival and logistic analyses, call for different data structures. Cases inconsistent with the definition of the dependent variable in each of the two methods were deleted. In the data sets designed for logistic analysis, if the child was not subject to risks for a full year, the respective case was removed from the data set. The 1996 DHS survey was taken in May, so all the children bom between April of 1995 and May of 96 were not considered. 2 3 7 Similarly, the children bom between November of 1990 and December of 1991 and between July of 1985 and August of 1986, respectively for the 1991 and 1986 DHS surveys, were also taken out of the data files. As a result, the size of the samples were reduced from 3876 to 3269 (84.3%) in 1996, from 3843 to 3212 (83.6%) and from 1125 to 826 (73.4%) in 1986. T he exact age and D O B of each child w ere calculated fr«n the C M C (century m onth co d e) eq u atio n C M C = (YY * 1 2 ) + M M 210 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In the data sets designed for survival method, by virtue of the fact that the analysis pertains only to monthly proportional risks, only the children bom in the month previous to the survey were not considered (that is, in April and May of 1996, in November and December of 1991 and in July and August of 1986). The sample sizes were reduced in the Cox analysis to 3822 in 1996 (98.6%), 3811 (99.1%) in 1991 and to 1120 (95.5%). The next step was to merge these 6 data sets (1996, 1991 and 1986 for logistic and survival analysis) and to obtain the pooled data sets for 1996-1991-1986, 1996-1991 and 1991-1986. Once the 12 data files were ready, 12 sets of descriptive stats including means, standard deviations, correlation, crosstabulation analysis and frequencies for all independent variables were calculated. Thus, this study sets out to empirically investigate the demographic and socio-economic underpinnings of infant mortality in the Northeast of Brazil in the 1986-1996 period. 211 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A n alytical F ram fw nrk Infant Mortality is generally described as the probability of dying between birth and exact age one (IqO), per 1,000 births. Assuming ( By ) as the total number of U ve births within a given year ( y ) and ( D O y ) as the total number of non-fetal deaths between birth and age one among residents in a community during the same year ( y ), then the infant mortality rate wül be: DOy / By * k where k is usually taken as 1,000. 2 3 » The estimation of infant mortality in this research, however, will rely only on the empirical data obtained from the proportion of children ever bom. Explained or dependent variables computed directly from the survey data provide robust estimates of death rates in the early infancy. 2 3 9 Preston and Trussell (1984) observe diat data on children ever bom and surviving children allow for a powerful analysis of the impact of covariates on This expression yields a reliable measure of infant mortality when the number of births do not change considerably from one year to the nmct. Since many Wants bom in a given calendar year will (miy die m the following year and since many deaths in the same calendar year are of infonts bom in the precedmg year (while in both cases foe infants are still m their first year of lifo), if foe number of birfos fluctuate substantially, an adjusted mfimt mortality rate must be calculated. Sudh rate would be given by foe expression { DOy (1-f 0 ) / By + DOy f 0 / By-1 } k, WierefO is foe prc^ortion o f infimt deaths wfthm a calendar year among bhfos from foe preceding year (usually this proportion varies around 0.1 and 0.3). For a conq>lete analysis, see Mortimer Spiegelmen, “ Further Considerations of Mortality’ ', Readings in Population Researdi Mefoodol(%y (New York, Vol. 1, Cluq>ter 7, 14): 75. W. Bras, Uses o f Census and Survty Data for foe Estimation o f ^ ta l Rates”, African Seminar on V tal Statistics (Addis Abeba, 1964): 14-19 212 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. infant and child mortality. 2 4 0 Infant mortality is assessed through the retrospective data collected in both the reproduction and the birth sections of the DHS questionnaires. In the reproductive section, a series of Brass questions were asked to determme the number of children ever bom to the respondent, such as the number of daughters and the number of sons living at home, living elsewhere as well as the num ber of deceased children. In the birth section data specific to each live birth were gathered such as date of birth, sex, age of death and survival status. The internal consistency of the reproductive data can be checked in the birth section. Women are the units of observation and the unit of analysis is the infant child. This study investigates the survival of the most recent child of each of the respondents. Two new variables were created: the age of death in months and the d e a d x a liv e variable. The djea d xa live variable is the binary dependent variable which can be described in terms of success and failure, life and death. This dependent variable is categorical, nominal, nonlinear and dichotomous, taking the value 1 for each incidence of death and 0 otherwise. Thirteen independent variables were included in the model. The individual, maternal and family-level demographic independent ^ Samuel Preston and J. Trussell, “Estimatmg the Covariates of Childhood Mortality from Retrospective Rqiorts of Mothers”, Methodologies for the Collection and Analysis o f Mortality Data, edited by J. Vallin, J. Pollard, and L. Heligman, (1984); 331-336. 2 1 3 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. variables are sex of the index child, mother’ s age group a t birth, birth order, pre-natal medical care by a physician, DPT immunization and breastfeeding status. Birth spacing was not included as a covariate due to the substantial amount of missing data but the impact of excess children can be captured by the birth order variable. 2 4 1 The individual and household socioeconomic independent variables are goods (a projgr for household income) 2 4 2 ^ mother’ s educational attainment 2 4 3^ mother’ s ethnicity, place of residence (urban x rural ) 2 4 4 as well as sanitary and household covariates such as household crowding, source of drinking water and sewage. See JJM . HobcraA, J. McDonald, J. and S. Rutstem, “ Dmnographic Detenninants of Inânt and Early Child Mortality A Comparative Analysis”, Population Studies 39 (1985): 363-385. Several studies have concluded that m Brazil hi^er levels o f household mcome have a sigpificant impact on infant mortality rates. Among such studies: - C.G. Victora, P. Smith and J.P. Vaughan, “Social and Enviromnental Influences on Child Mortality in Brazil: Logistic Regression Analysis fi’ om Census Files’’ , Journal o f Biosocial Science 18, no. 1 (1986): 87-101. - Thomas Merrick, “The Effect o f Pçed Water on Early Childhood Mortalfty in Urban Brazil, 1970 to 1976”, World Bank Staff Working Papers 594 (Washmgton, D C , 1983). - D. Thomas, J. Strauss and M Henriques, “Child Survival, Height for Age and Housdiold Characteristics in Brazil”, Journal o f Development Economics 33, no. 2 (1990): 197-234. See: - J. Caldwell, “Education as a Factor in Mortality' Decline: An Exammation of Nigerian Data”, Peculation Studies 33, no. 3 (1979): 395-413. - K . Stratfield, K , I. Diamond and M. Smgarimbum, “Maternal Education and Child Immunization”, Demography 27, no. 3 (1990): 447-455. - J. Hobcraft, ‘ Women’s Education, Child Welfare and Child Survival: A Review of die Evidence”, Health Transhion Review 3. no. 2 (1993): 159-175. See: - Hugo Bdm, “Socioeconomic Determinants of Mortality in Latin America”, Proceedings of the Meetings on Socioeconomic Detennmants and Consequences o f Mortality (Mexico City, June 1979). - Julie DaVanzo, “Infant Mortality and Socioeconomic Development: Evidence fiom Malaysian Housdiold Data”, Demography 25, no. 4 (1988): 581-595. 214 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In addition, to these thirteen independent variables, a birth cohort variable was included in all the twelve data sets. The Logistic R egression Method The main method of statistical analysis adopted in this empirical study is logistic regression. We are mostly concerned with the impact of nominal, binary, dichotomous independent variables on direct estimates of infant mortality calculated from data collected on surviving children. In this section we seek to show that logistic (or logit) regression offers an efhcient modeling strategy to understand the impact of a set of categorical independent variables on infant mortality. Despite the similarities with ordinary least squares regression methods, logit regression is built upon a different underlying mathematical foundation. Moreover, the employment of the traditional OLS method of linear estimation with a binary dependent variable such as d ea d xa li would entail several problems, which we will discuss here. Under certain ideal conditions, OLS is ejected to perform better than other regression methods, but if such conditions do not hold, linear regression is clearly inadequate and do not offer a best unbiased estimator. OLS methods do not work well, however, with dichotomous variables since many of its I x y assumptions are violated. 215 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. OLS estimates parameters of linear models by calculating the expected value (population mean) of a dependent variable Y given a ith set of XI, X 2 . JGc values. E [Yi] =a + plXil +p2Xi2 + +PkXü c Since, Yi = E[Yi) + 8i Yi =a + piXil + P2 Xi2 + +Pk Xik + ei Under certain conditions OLS is BLUE, that is, the best (most efficient ) linear unbiased estimator. The necessary conditions or expected assumptions are the following; 2 4 5 1. No measurement errors: All independent and dependent variables are measured without errors. 2. No specification errors: A U independent and dependent variables are measured without errors, meaning that no irrelevant predictors of the independent variables have been included, no relevant ones have been - WJ}. Berry, Uiderstanding R^ression Assumptions (Newbury Park, CA: Sage IMversity^ Papers no. 92, 1993): 7. - WJ>. Berry and S. Feldman, Mukqile Regression in Practice Lbderstanding R%ressicn Assunqpticns Beverly Hills, CA: Sage University^ Papers no. 50, 1985): 7. - Michael Lewis-Beck, ^ p lie d R%ression (Thousand Oaks, Ca: Sage University^ Papers no. 22, 1980): 26. 2 1 6 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. excluded and there is a linear (allowing for transformations ) relationship between independent and dependent variables. 3. Fixed X : Random samples would have the same X values but different Ys since the ei values are different as well. 2 4 6 4. Errors have zero mean: The expected value of the error ( e ) is zero. 5. Homoscedasticity: The variance of the error term, 8, is constant for all values of Xi: Var [ ei ] = o2 6. Normality of errors: Errors are normally distributed for all sets of values of the independent variables. 7. N o autocorrelation: Errors terms from values of the independent variables are uncorrelated with each other: Cov [ ei, g ) 0 where i 8. N o correlation between the independent variables and the error terms: E [ ei, Xi ) = 0. 9. Absence of perfect multicoUinearity: No independent variable is a perfect linear combination of the other independent variables. Assumptions 3 and 4 ensure the independence of errors and X variables, which is a sufBcient condition for providing unbiased O LS estimates of all the parameters. Unbiasedness implies that sample estimates bk will be very close to the true parameters values P k in the long run: E ( bk ) = P k. With randwn X values, mstead o f E [ Yl ), there would be a ccnditionai expectation of Yi given a X vector Xi, or E [ Yi/Xi ]. 217 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Assumptions 5 and 7 ensure that the standard errors are unbiased and that O L S is more efficient 2 4 7 than any other linear unbiased estimator, or that OLS is BLUE. Assumptions 3, 4, 5 and 7 combined allow OLS to be BLUE, which is also called the Gauss- Markov theorem. 24g Many of the above OLS assumptions are violated when modeling a dichotomous dependent variable such as infant mortality. Menard (1995) 2 4 9 that the OLS model becomes a linear probabilistic model when the dependent variable is dichotomous. In this case, the mean of the dependent variable becomes a function of the probability that a given case win fall into the higher of the two categories for the variable. When the values of the variable are coded 1 and 0, the mean of the variable becomes the proportion of cases in the higher of the two categories. The conditional mean or the predicted value of the dichotomous dependent variable becomes the predicted probability that * a case falls into the higher of the two categories on the dependent variable, given its value on the independent variable 2 5 0 As a result, the predicted values for the dependent variable could be much h i^ e r or lower than the possible values of the dependent variable. bk is more efficient than ok if both are unbiased estimators o f bk and Var [b k ] < Var (ck ), or that E [ bk - bk ] < E [ ck - bk ] 2. Efficiency increase the likelihood that a sample estimate will be close to the true parameter. ^ Even if OLS is BLUE, there may be more efficient and robust estimators than OLS. Also, certain biased estimators can have a smaller mean square error than OLS. ^ Scott Menard, Applied Logistic Regression Analysis (Thousand Oaks, CA: Sage University Papers no. 106,1995): 13. Scott Menard, Ibid: 6. 2 1 8 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Using OLS method to predict the probabdi^ that a binary event will happen would imply impossible probabilities exceeding the {0, 1 } boundaries. Moreover, several of the OLS assumptions would be violated, indicating that the estimators could be unbiased but would certainly not be efBdent. The shortcomings of using OLS regression with dichotomous dependent variables can be indicated through a simple example. Assume the data are collected through survey sampling. Suppose the variable Y is a measure of infant mortality, or the odds of survival (IM ) , taking the value 0 for death and 1 for survival. Assume the X variable is a not dichotomous variable expressing the length of time the child's mother has attended school, or the number of schooling years ( Sch ). Sch fluctuates from 0 to 15. The regression line Y ’ = .069 — .007 Xil or IM = .069 — .007 Sch shows a negative relationship between the infant mortality and schooling. For any values of Sch ( X ), there are only two possible values for the residuals: If IM ( Y ) = 1, then ei = 1 - (.069 - .007 Sch ) If IM ( Y ) = 0, then ei = (.069 - .007 Sch ) With any dichotomous Y variable, the errors will have only two possible values for each Xi. The variance of the residuals will depend on E ( IM ) and on Sch, which makes the homoscedasticity or constant variance for the error term assumption (the Gaussian condition) implausible. Additionally, the variability 219 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. of the error terms will depend upon the size of the independent variable. This phenomenon- called heteroscedasticity - implies that the estimates for the regression coefScients would be unbiased but not efficient or the best estimates. Since the standard errors will be biased, not normally distributed and the sampling variances will not be calculated correctly, hypothesis testing and confidence intervals will also be invalid. 2si In addition to all the problems created by heteroscedasticity and non Gaussian errors, the employment of the OLS method with dichotomous variables will impose an even more fundamental shortcoming: the predicted values of the dependent variables or the predicted probabilities will exceed the {0,1 } boundaries. In our example, for a mother who did not attend school at all : IM = .069 — .007 (0) = .069, or a probability of 6.9% or 69/000. As the independent variable Sch increases, however, the predicted probabilities will take on impossible values since , per definition, a probability cannot be less than 0, or greater than 1. If Sch = 1, IM = .069 - .007 (1) = .062. If Sch = 3, IM = .069 - .007 (3) = .048. If Sch = 8, IM = .069 - .007 (8) = .013 LJ3. Sdiroeder, DX. Sjoquist and PX. Stq>han, Lbderstanding Regression Analysis: An Mroductoiy Guide Beverly Hills, CA: Sage Univers^ Papers no. 57, 1986): 7. 220 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. If Sch = 10, IM = .069 - .007 (10) = -.001 If Sch = 15, IM = .069 - .007 (15) = -.036 The employment of O LS or any linear probability method to estimate binary dependent variables is clearly unrealistic. The fact th at linear predictions exceed the { 0,1} boundaries imply that the parameters being estimated do not exist or that the relationship between the variables is non linear. Scott Long ( 1997) notes 4 main shortcomings of the application of a linear OLS type regression to a binary dependent variable (application also known as a linear probability model -LPM ) are: 252 - Heteroscedasticity: The variance of the errors is not constant and is conditional on X’ s; the residuals are biased; the test statistics are erroneous and the OLS estimators of the parameters are inefficient and not BLUE. - Normality: Errors cannot be normally distributed. Functional form: Since the outcome is a probability, the effects of the independent variables will have diminishing returns as the predicted probabilities approaches 0 or 1. The predicted probabilities of a binary dependent variable will be a S-shaped function of the independent variables. The model should be non-linear. Scott Long, R^ressiaa Models for Cat^orical and Limited Dqiendent Variables (Thousand Oaks, CA; Sage Publications, 1997): 39. 221 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. - Nonsensical predictions: Negative and greater than 1 values for the dependent variables are predicted. The method used in this empirical study will be, instead, logistic (logit) regression. Logistic regression circumvents the problems posed by linear regression models through the modeling of predicted probabilities as a S- shaped, non linear cumulative logistic distribution of the independent variables X. D. Collett (1991) observes that linear logistic model is a member of a class of models known as generalized linear models, introduced by J.A. Nelder and R.W. Wedderbum (1972). 253 It is important to note, however, that, even th o u ^ the estimation method for logistic regression differs from OLS, some of the main structural assumptions of the latter are maintained, such as: independence between cases; true specification and measurement of the independent variables; no X variables can be linear functions of the other independent variables (no perfect multicoUinearity). W e will now set out to show the main conceptual features of the logistic model. Logit coefScients can be seen in terms of probabilities, odds or odd ratios. Scott Menard (1995) notes that in order to understand the structure of a given logistic model, one must understand that * the probability, tiie odds - D. Collett, Modeling Binary Data ( London, UK: Clubman & Hall, 1991): 56. - JA . Nelder and R.W. Wedderbum, “Gmeralized Lmear Models", Journal of the Royal Statistics Society, A 135 ( 1972). 222 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. and the logit are three different ways of expressing exactly the same thing. * 2 5 4 Given the conceptual structure of the model, we seek to investigate whether the death of the child can be predicted by the set of categorical independent variables. Since the dependent variable is dichotomous, takmg the value 0 for death and 1 for survival, we are really interested is estimating the binary response in terms of the odds of success or failure. If P ( Y=1 ) indicates the probability that the infant’ s survival status variable {0,1 } Y takes the value 1 (child is alive), the odds favoring this event are: a ( Y = 1 ) = P ( Y = 1 ) / 1 - P ( Y = 1 ) 2 5 5 By tflldng the natural logarithm of the odds, we obtain the logit of the dependent variable Y , or the log odds of the ratio of the probability of death to that of survival. L = loged = ln{(P/(l-P)} The logit of Y range firom - 00 ( as the odds decrease firom 1 t o 0 ) t o qo (a s the odds increase firom 0 to 1). ^ Scott Mmard, Ibid: 13. If, instead of odds, probabilities were used, in a model where die probability' that Y= 1 equals to P (Y=l) = a + b X i, die predicted values could be greater dian 1 or less than 0. Thus, lmear regression would not work successfully because linear predictions can exceed {0,1} boundaries. In addition, with didiotcMnous covariates, the errors cannot be mqiected to be Gaussian or to have constant variance. 223 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Using as the dependent the natural logarithm, of the odds that Y ~ 1, the logit regression or the equation for the general logistic model takes the form: Logit ( Y ) = log e Ô * In {( H / ( 1 - H ) } = a + p i Xil + 32 Xi2 + ... + 3k Xi, k (1 ) where: a is the intercept value; Pi is the probability of death for the ith child; 3 1 , 3 2 , ..., 3k are the parameter estimates; Xil, Xi2,..., 5 Ü , k are the corresponding independent variables. The general logistic model can be converted back to odds by direct exponentiation and back to probability as well, ^ se yielding the following equations, respectively: a ( Y = 1 ) = e * * In {» ( Y = 1 ) } = e * * a + p i Xil + p2 Xi2 + ... + pkXi, k ( 2 ) and P ( Y = 1 ) = e * * a + 3 1 Xil + 3 2 Xi2 + ... + 3kXi, k / 1 + e * * a + piX il + 3 2 Xi2 + ... + 3k Xi, k ( 3 ) 2 5 7 “ ^Byusmg J ( Y = l) = e * logit (Y ) to getthe odds and P ( Y = 1 ) = J ( Y = 1 ) /{ 1 + J ( Y = 1) } to get the probabilities. The predicted probabilities estimated m (3) will not reach or exceed die { 0, 1 } boundaries. 224 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Since the logit values for the dependent variable in. (1) can be infinitely large or small, OLS cannot be used to estimate the parameters. 2 5 8 Logistic regression analysis, through the form in (1 ) Logit ( Y ) = log e 8 = In {( Pi / (l-Pi)} = a + 3 l Xil + 32 Xi2 + ...+ 3k k , is one of the best ways to understand and analyze dichotomous dependent variables, such as the infant's death or survival. Logistic models do not use the method of least squares. Instead of solving directly for the parameters using OLS, an iterative and converging estimation process using maximum likelihood is adopted. A likelihood function manifests the probability of obtaining the observed sample or the likelihood of the unknown parameters in the model for the sample data. This function is the joint probability of the observed data seen not as a function of the data but rather as a function of the unknown parameters in the model. Through ma-yiTnum likehhood, the values of the logarithm of the likelihood fiinction (log-likelihood) are maximized to indicate what parameter values make the sample more likely or how likely the values for the dependent variable are, given the values of the parameters a + 3 1 + + 3k and the independent variables. The main question posed by maximum likelihood estimation is: What parameter values make the observed sample most likely? The logit regression curve and die OLS line would b%in very close together but would separate near the 0 probability. The OLS Ime would contmue below probability 0, which is inqiossible. In contrast, the logit curve would never readi 0. 225 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Iteratively, the maximum likelihood estimates - or the best approximations for the parameter values - are foimd through the maximization of the log- likehhood: Log e L = S {Yi log e Pi + (1 - Yi ) log e ( 1- Pi ) } To get the maximum Ukelihood estimates, we take the derivatives of the log likelihood with respect to each of the estimated parameters, fmding two sets of simultaneous equations; = Z(Yi-K) = Oand 2 ( Yi - Pi ) Xik = 0 Since the equations are non linear for the parameters, the best approximations for the parameter values must be found not directly (as in OLS) but through an iterative computer process. The SPSS Lofidt S tatistical O utput Next we will briefly examine the main statistical components of the SPSS logistic regression output. The statistical results are divided into two parts: the regression diagnostics or the summary statistics for evaluating the model and the maximum likelihood estimates or logistic regression coefficients. 226 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The maùi elements of the model summary statistics are the initial log likelihood statistics ( initial -2LL ), the - 2 log likelihood statistics (-2LL or Dm ), the model chi-square (model X2 or Gm) and the classification table. Scott Menard ( 1995 ) notes that in logistic regression analysis often the focus is not only how well the set of independent variables explain the dependent variable but also whether the predicted values (predicted means or predicted probabilities ) are correct and close to the observed 0 and 1 values of the dependent variables. 2 5 9 The diagnostics statistics provide an understanding of the overall goodness of fit of the model. In logistic regression, instead of the sum of the square errors (SSE), the main criterion for selecting parameters is the log-likelihood. The SPPS output uses not the log-likelihood itself but rather -2LL since the log-hkelihood when multiplied by -2 is distributed as a chi-square (X 2). The first statistic to appear in the logistic regression output is the “ initial log likelihood function - 2 log likelihood", also called intercept only, D O or initial -2LL. This statistic, similar to the total sum of squares (SST) in linear regression, is the -2 log likelihood statistic only with a constant term and with no independent variables at all in the equation. Do = -2 {(n y=l) In [ P (y=l) + (n y=0) In ( P (y=0) ]} where y=l is the total number of cases Y=l, y=0 is the total number of cases Y=0 and P ( Y = 1 ) = n y=l/N is the probability that Y=l. Scott Menard, Ibid; 17. 227 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. After the listing of the independent variables, the - 2 log likelihood statistics (-2LL or Dm ) and the goodness of fit statistics are presented. The -2LL includes all the independent variables as well as the intercept. This statistic is similar to the SSE in linear regression and indicates "how p o o rly the model fits with all the independent variables in the equation.” 2 0 0, Larger values for the log-likelihood statistics indicate relatively worse predictions for the dependent variable. The goodness of fit statistic investigates the difference (residual) between the predicted probabilities and the observed ones. This summary statistics is defined as: Z2 = I { Residual2 / Pi (1-Pi) } Another important diagnostics statistic presented in the SPSS logit output is the model chi-square (Gm ) and its significance. This statistic relates to both the P test and the regression sum of squares (SSR) in linear regression and can be defined as the difference between the initial log-likefihood -2LL (D O ) and - 2LL (D m ). If the model chi-square (G m ) is statistically significant ( p ( .005 ), we must reject the null hypothesis (H o) and assume that the independent variables improve the predictability power of the model. Otherwise, we faü to reject the null hypothesis since there is not enough ^ Scott M en ard , Ib id : 20. This author observes that the statistical significance of D m is sim ilar to die statistical sig/aiBcaace of the unexplained variance of O L S r% ressions. 228 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. evidence that the variance explained in the model is not due to a random sample variation. In order to estimate whether adding a specific covariate improves the predictability of the model, another diagnostic regression statistic can be used: the model improvement statistic. The latter is the change in Dm ( - 2LL or - 2 times the log likelihood ) obtained through successive model building steps. Basically it tests the null hypothesis that the coef&cients for the added variable are 0. When the null hypothesis is rejected it can be stated that the added variable is indeed statistically significant. The SPSS statistical output also presents a four cell classification table, comparing the values predicted and observed for both values of the dichotomous deaXalil dependent variable. The classification table evaluates the predictive efficiency of the model. The most important part of the SPSS statistical logit output is labeled Variables in the equation” and involves the maximum likelihood estimates for all the predictors. Its main components are: the parameter estimates, the B coefficients; the standard errors of such estimates; the Wald statistic; the odds ratio associated with each coefficient, Exp (B ) ; the significance or p value of each estimate. 229 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The B coeffîcieat, or the unstandardized logistic regression coefficient is the non-linear change in the dependent variable logit ( Y ) 2 0 1 associated with a one-unit change in the independent variable . 262 Scott Menard (1995) observes that "unstandardized regression coefficients are especially useful for evaluating the practical impact of one variable on another and for comparing the effects of the same variable in different samples.” 263 Exp (B ) is the odds ratio ( the probability / 1- probability or & ( Y = 1 ) = P ( Y = I) / 1 - P ( Y = 1 ) ) , or the maximum likelihood estimate of die odds ratio. Thus, the parameter estimate B is the log of the odds ratio Exp (B). The odds ratio Exp (B ) is the number by which we must multiply the odds of dying for each one-unit increase in the independent variable. If the odds ratio (Exp (B )) is greater than 1, the odds of dying increase as the value for the independent variable increases. Analogously, if the odds ratio is less than 1, the odds of being dead decrease as the value for the independent variable increases. For instance, if the Exp (B ) is 1.0301 (impact of drinking water’ s odds ratio for 86-91-96 NE data) the odds of being dead increase in 3% (odds are multiplied by 1.0301) for each one unit increase in the variable drinkwa2. The odds ratio offers the same information as the logistic regression coefficient or probability. 261 Logk(Y)= logeJ = ln { (P i/(l-P i)} ^ In contrast, the standardized r%ression coefficient mdicates how many standard deviations a given d^endent variable dianges as a result of a 1 unit standard deviation diange m die indqiaident variable. Scott Menard, Ibid: 37. 230 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The coefficient standard errors of the parameter estimates are also calculated through the TnaviTnum likelihood estimation process and presented in this section. 2 6 4 A nested-models strategy 2 6 5 similar to F-tests in OLS can be used for setting hypothesis testing in logistic regression models. We assume that log e Lk is the log-likelihood of a logistic model with k parameters. In order to test whether this model is better, or improved than a simpler model with less parameters (0 ( v ( k ) , we compare the log-likelihood of both models (Lk and Lk-v ). The null hypothesis that the v omitted X variables have no effect, H O : pi = 32 = .... = 3v = 0 is tested by : X2v = - 2 (log e L k-v - log e Lk ), which is distributed as a chi-square with v degrees of freedom. The test whether the X’ s coefficients are zero is similar to the F test in OLS regression. If the null hypothesis is rejected, the more complex model fits better. The most accurate test to evaluate the statistical significance of the contribution of an independent variable to the explanation of a dependent variable is considered to be the likelihood ratio test, 2 6 6 which is equal to the Gm (model chi-square) for the model with an additional variable minus the GM The standard errors are actually the square roots of the diagonal elanents of the inverse information matrix, vdiich is the matrix of negatives of die second partial derivatives of the likelihood function. ^ Lawrence C Hamilton, Regression widi Graphics ( Pacific Grove, CA: Cole Publishmg Conqiany, 1992): 229. Scott Menard, Ibid: 38. 231 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. for a model without it. Testing hjrpothesis in this fashion can be troublesome since we have to estimate two models and calculate by hand twice (minus) the differences in log-likelihood. The SPSS output for logistic regression presents an alternative to the likelihood ratio test: the Wald statistic. This statistic, very similar to the OLS standard t-test, is distributed asymptotically as a chi-square distribution and can be computed as the square of the parameter estimate divided by its standard deviation. 2 6 7 Wk = ( Bk / s.e Bk ) 2 The degrees of freedom (df) associated with each variable will be equal to the number of categories for each independent variable minus 1. The Wald test is the square of a t-test of the form t = Bk / s.e. Bk. For large samples, the Wald results are similar to the likelihood ratio test 2 6 8 and to the Lagrange multiplier for that matter. Scott Long ( 1997) notes that logit Wald test estimates the model with no constraints and then assesses the constraints by considering two things. First, it measures the distance between the unconstrained and the constrained estimates (the larger the distance, the less likely is that the constraint is true). ^ Scott Menard observes that one o f the mam. shoitconmgs of the Wald statistic is that for a large B, the estimated standard error will be mflated and we will foil to reject the null hypothesis imhen the null hypodiesis is actually folse. Ibid: 39. Andto X2v=- 2 dog e L k-v - loge Lk). 232 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Then, this distance is weighted by the curvature of the log-likelihood function, indicated by the second derivative o2 InL / aP2. ^ The p-value or the significance level is an important statistic also presented in the "variables in the equation* part of the SPSS output. It is also derived firom the Wald statistic and mdicates, at the 95% level of confidence, whether the independent variable has a significant impact on the dependent variable, or whether we must reject or not the null hypothesis. At a p-value of .005 or less, one may conclude that the independent variable has a statistically significant impact on the dependent variable. The Cox R egression Method Another method used in this research is event history or survival analysis. This method is particularly designed to study the occurrence of qualitative changes or events using time series. The qualitative change being observed is the death of a child. The results of survival or event analysis should enforce and be compatible with the logistic regression results. One of the main caveats of the logistic regression method is that it can not be used in models that include time-varying explanatory variables. Logit models using a dichotomous dependent variable do not take into account the censoring Scott Long. Ibid: 87. 233 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. of the data and are not appropriate for evaluating the impact of time-dependent covariates. Breastfeeding is the only time-varying independent variable investigated in our model but the employment of event history analysis is justifiable for that the impact of breastfeeding on the chances of child survival is far reaching. It is important to know not only whether the child was breastfed but for how long. The liîning of this time-dependent explanatory variable is of paramount relevance. Richard Breen ( 1996 ) observes that the estimation of models for censored data was pioneered by Tobin in 1958. 2 7 0 Most of all variables remain constant in our model but not the breastfeeding practices. Using the logistic regression method with time- dependent covariates would entail biased estimates and loss of information because Logit models ignore the timing of the event (the death of the child). Paul Allison ( 1995 ) argues that the ad-hoc methods that deal with censored data, such as discarding the censored cases, are clearly inefScient and unsatisfactory. 2 7 1 Methods of event history analysis, however, do allow for censoring of the data. ^ Richard Breen, Regression Models, Censored, Sample Selected or Truncated Data (Thousand Oaks, CA: Sage University Papers no. 101,1996): 6. ^ Paul Allison, Survival Analysis Using die SAS System- A Practical Guide ( Cary, NC: SAS Institute, 1995): 4. 234 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Influenced by the study of life tables, D. R. Cox ( 1972 ) 2 7 2 developed the proportional hazards model which will be adopted here. The Cox method of partial likelihood estimation, is considered to be one of the most influential regression methods. 273 In order to apply the Cox regression method it is important to define the risk set, or the set of children who are at risk of event occurrence (death) at each point in time. The risk set includes all the children (most recent bom) at the beginning of the first month and, at the end of the following twelve months, decreases progressively with the exclusion of the children who eventually do not survive (and who are no longer at risk of death). The data must be censored to exclude such cases. A sample is said to be censored if we have observed the values of the X variables for a ll sample observations. 274 Censoring the data is achieved through the creation of two variables: d ea th d u r and d ea th sta L The death duration is the quantitative time variable or the age in months of the child at the time of the death. It ranges firom 1 to 12. If the infant is still alive at the end of the first year, the deathdur value is 13, meaning that the subject did not make the transition (did not die) at the end of the first year. The child’ s age at death in months is right censored at the end of twelve months. ^ D R. Cox, “ R^ression Models and Life Tables”, Journal of fee Royal Statistical Society, Series B, 34 (2). ^ Paul Allison, Ibid: 12. Richard Breen, Ibid: 2. 235 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The death status variable is a categorical variable which takes on the values 0 and 1. If the infant makes the transition, if the event (of death) occurs, the value is 1. Otherwise, the variable is censored and the value for the d e a th s ta t variable is 0. The hazard rate is the probability that the event of death will occur to the infant at a particular point in time, given that the infant is at risk at that time. Cox’ s proportional hazards model employs partial likelihood estimation to obtain atymptotically unbiased and normally distributed estimators. The main assumptions of this method are that the observations m ust be independent, and the hazard rate must be constant across time, meaning that the proportionality of hazards firom one case to another should not vary over time (also known as the proportional hazards assumption). The utilization of Cox regression allows us to observe the impact a time-varying variable, breastfeeding, on infant mortality. All the other categorical covariates considered in the logistic regression method are maintained in the Cox regression. The proportional hazard model's basic form is : 8 (t, z) = 5 0 (t) e * * ( pi z 1 + 32 z 2 + ............+ P k z k ) = 5 0 (t) e * * (p z ) where: 8 (t, z) is the hazard function at time t for an infant with k discrete or continuous covariates z. 236 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. p is the set of regression, coefficients which expresses the impact of shifting the time-vaiying baseline hazard function S 6 (t) upwards or downwards. The SPSS output for the Cox regression is very similar to the logistic regression one, including the log likelihood functions, the chisquare and the main regression estimates. 237 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A THEORETICAL AND EMPIRICAL EXAMINATION OF INFANT MORTALITY DECLINE IN BRAZIL’S NORTHEAST REGION, 1986-1996 Volume H By Fernando Veiga Prata A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment o f the Requirements for the Degree DOCTOR OF PHILOSOPHY (Political Economy and Public Policy) August 1999 Copyright 1999 Fernando Veiga Prata Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. “ The basis fo r an interpretation o f causality cannot be fo u n d in the data themselves, even in an experim ent, but ra th er depends upon our g en eral understanding o f the situation together w ith our research purposes . That is, th e la b el “ causality” is a judgm ent. (Julian Sim on, The C oncept o f C ausality in the Social Science) 238 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter? RESULTS OF DESCRIPTIVE AND MULTIVARIATE REGRESSION ANALYSIS Introductioii The results are based on logistic regression, Cox regression and standard statistical analysis applied originally to six data sets: the 1986, 1991, 1996 and the pooled 1996-1991, 1991-1986 and 1996-1991-1986. Twelve data sets were then created, six formatted for logistic and six for Cox regression analysis. The rationale behind the establishment of twelve data sets is that logistic and proportional Cox analysis are quite different in structure requiring the selection of a different number of cases for the dependent variable through the creation of different cohort variables. The 1986 DHS was carried out between May and August; the 1991 DHS was implemented between August and December and the 1996 DHS between April and May. The data sets are restricted to include only women in union who were or had been mothers. In the data sets designed for logistic analysis, if the child was not subject to risks for a full year, the respective case was removed from the data set. The 1996 DHS survey was taken in May, so all the children bom between April of 1995 and May of 1996 were not considered. 239 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Similarly, the children bom between November of 1990 and December of 1991 and between July of 1985 and August of 1986, respectively for the 1991 and 1986 DHS surveys, were also taken out of the data £Q es. As a result, the sample sizes designed for logistic regression were reduced from 3876 to 3269 (84.3%) in 1996, from 3843 to 3212 (83.6%) and from 1125 to 826 (73.4%) in 1986. In the data sets designed for survival method, by virtue of the fact that the analysis pertains only to monthly proportional risks, only the children bom in the month previous to the survey were not considered (that is, in April and May of 1996, in November and December of 1991 and in July and August of 1986). The sample sizes were reduced in the Cox analysis to 3822 in 1996 (98.6%), 3811 (99.1%) in 1991 and to 1120 (95.5%). The pooled data sets for 1996-1991-1986, 1996-1991 and 1991-1986 were created by the merging of these 6 data sets (1996, 1991 and 1986 for logistic and survival analysis). The dependent variable is the survival status of the mother’ s latest child as reported by the respondent. After the year and month of birth of the child were calculated (using the Century Month Code-CMC variables from the DHS surveys), the cohort variables for both the Cox and the LR data sets were introduced through two alternative procedures: 240 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. a) a sin^e variable taking four possible categorical values according to the time period of the child’ s birth (fix > m 0 to 3, 0 being the oldest birth cohort period ). For instance, for the 1996 data formatted for survival analysis, the possible values are 0 ( 79 and before ), 1 (80- 84), 2 (85-89) and 3 ( 90- March 96). b) three dichotomous cohort variables each taking the value 1 if the child was bom within the given time period and 0 otherwise. For the 96 survival data set the cohort variables are 90-96, 85-89 and 80-84 (the oldest cohort, or children bom before 1979, acts as the dummy variable). A single cohort variable seems to be more indicated since it requires fewer degrees of freedom. None the less, all the regressions for the Logistic and Cox analyses were run with both the single and the multiple birth cohorts. The birth cohorts periods for the 1996 logistic regression data set are as follows: 1979 and before (0), 1980-1984 (1), 1985-1989 (2) and 1990- March 1995 (3). For the Cox data set the cohorts are the identical with the exception of the latest one which encompasses a longer period, 1990- March 1996. In the 1991 logistic regression data set, the birth cohort periods are: 1974 and before (0), 1975-1979 (1), 1980-1984 (2) and 1985- October 1990 ( 3). Analogously in the 1986 LR data set, the birdi cohort periods are: 1969 and before (0), 1970-1974 (1), 1975-1979 (2) and 1980- June 1985 (3). 241 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In the 1986-1991 Ne Pooled LR data set the birth cohort periods are: 1974 and before (0), 1975-1979 (1), 1980-1984 (2) and 1985- October 1990. For the 1996-1991-1986 and 1996- 1991 Ne Pooled data sets the birth cohort periods are as follows: 1990- March 1995 (3), 1985-1989 (2), 1980- 1984 (2) and 1979 and before (0). Table 70 DATA SETS DATA SAMPLE SIZES 1996 1991 1986 1996 1991 1986 1996 1991 1991 1986 Final Data S ets Before Separation for Cox and LR (Bef. Cohort Adjst.) 3876 3843 1125 8844 7719 4908 Logistic Regression Data Sets 3269 3212 826 7307 6481 4038 Cox Regression Data Sets 3822 3811 1120 8753 7633 4931 242 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. For each, of the twelve different data files, a complete set of statistical measures involving all mdependent and dependent variables was computed. The main estimates calculated through descriptive statistics are; means, frequencies, contingency tables, chi-square tests, Pearson correlation coefficients, significan ce levels and standard deviations. W e are particularly concerned here with the means for all independent variables and respective categories or with the percentage of deceased children associated with each explanatory variable. The firequencies and proportional shares for all independent variables and categories will be examined. In addition, the descriptive results will analyze the correlation levels between the dependent and the independent variables. The correlation levels will explore how each of the independent variables correlates with infant mortality - as measured by the dependent dichotomous variable d e a d or a live. An analysis of the bivariate relationships between the dependent variable and the explanatory variables indicate that if we pair the values for these variables in a plot we will definitely see a trend, more particularly a linear relationship. The dependent and the independent variables in this study's regression models are correlated, and usually inversely related or correlated negatively. Since the category that it is expected to exert a negative impact on infant mortality is firequently labeled 0, a change in the dichotomous variable firom 0 to 1 would lead to a decline in the odds of dying for the infant. 243 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Hence, most of the explanatory variables will be negatively correlated with the dependent variable. The exception is the variable household crowding, a non dichotomized variable, that is positively correlated with infant mortality. As the household becomes more crammed, chances are the odds of dying will be greater for the child, or as the value for the variable increases so does the dependent variable and the odds of reaching a transition or obtaining a 1 value, death. Pearson’ s product moment coefiScient, also known as ( r ) or correlation coefScient is a measure of linear association, measuring how variables or rank orders are related. Pearson’ s correlation coefhcient ( r ) is defined as: r X, y = S xy - 1/n (S x) (Sy) / (n-l)s x S y, where: - Z xy is the sum of the x y cross-products; - n is the sample size or number of data pairs; - (Z x) and (S y) are the sum of the xi and yi; -S x and S y are the respective sample standard deviations of x and y. It is meaningful to investigate the level of confidence associated with the correlation coefficient. It would be expected that for most variables the respective correlation coefficients are significant (2 tail) at the 0.01 level (99% confidence level), but some coefficients will be less significant or will only be 244 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. significant at the 0.05 level ( 95 % confidence level). However, i f the correlation is not significant at all, then the likelihood that the coefficient resulted firom chance is too high implymg that both the direction of the correlation coefficient (sign) and its magnitude are , reflecting only sampling error. The correlation coefficients indicate both the direction and the strength of the statistical relationship between the explained and the explanatory variables. Correlation levels offer an additional powerful descriptive statistic measure, but it does not imply necessarily a causal relationship among the variables. Regression analysis will examine causality and the relative impact of the interaction of the explanatory variables on the dependent variable. As far as the regression results from the logistic regression and survival Cox methods are concerned, the main statistics are: the parameter estimates or B coefficients, the odds ratio associated with each coefficient or Exp ( B ), as well as the standard errors and the significance levels. The odds ratio is the number by which one multiplies the odds of dying for each unit increase in the independent variable whereas the unstandardized B coefficients are the log of the odds ratios or the non linear change in the dependent variable Logit ( Y ) associated with a one-unit change in the independent variable. 245 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The significance level or p value of each covariate is derived from the Wald Statistic ( which is the square of the parameter estimate divided by its standard error) and evaluates the null hypothesis that the independent variable has no statistically significant effect on the dependent variable. The smaller the significance level, the more stringent the test is, or the greater the confidence that the finding is significant. If the p value is .05, the probability of obtaining the result by chance is I in 20, or the confidence level is 95% or the probabili^ of rejecting the null hypothesis H O (that the independent variable has no impact on infant mortality) when the null is actually true is 5% (also called error type I). Diagnostic statistics for all models are also computed. Among them the -2 log Likelihood, the Goodness of Fit and the Model chisquare coefficients.. Through the classification tables, the predictive efSciency of each model is also calculated. The -2LL statistic is similar to the SSE in OLS Regression, indicating how poorly the model fits with all the independent variables included. Larger values indicate worse predictions for the dependent variable. The goodness of fit statistic compares the observed probabilities with those predicted in the model. The model chisquare is similar to the F test in ordinary least square regressions and equals to the difference between the initial log-likelihood (with a constant) and the -2LL for the whole model. 246 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. When the model chisquare is significant we rqect the null hypothesis and conclude that the independent variable has a statistically significant effect on the dependent variable. The diagnostic statistics provide measures to evaluate the overall goodness of fit of the models since we are not only interested in estimating how well the set of independent variables ^q>lain the dependent variable but also whether the predicted probabilities (or values) are correct and close to the observed binary values of the dependent variable. The standard regression models are comprised of thirteen independent variables, in addition to a birth cohort variable. The non dichotomous independent variables are household crowding, goods (the proagr for household income and wealth) and breastfeeding in months. The six individual, maternal and family-level demographic independent variables are: sex of the most recent child, mother’ s age group at the time of the birth, birth order risk, pre-natal medical care by a physician, DPT immunization and breastfeeding status (the last one is the only time-varying explanatory variable and it is included only in Cox’ s proportional hazard- survival model ). 2 7 5 The seven individual and household socioeconomic independent variables are: goods, mother’ s educational attainment, mother’ s ethnicity. The immunizatioD. variable (<^123) is not mcluded in the standard model since the in tact of prmatal care by a physician is stronger than immunization and deq)ly correlated widi k. 247 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. place of residence (urban x rural ) as well as sanitary and household covariates such as household crowding, source of drinking water and sewage. Note that for the 1986 data sets, as well as for the pooled 1996-91-86 and 1991-86 data sets, the independent variables ethnicity, prenatal care by a doctor and immunization were not included since the questions pertaining to these variables were absent in the first phase (1986) of the DHS in Brazil. Note also th at all the regressions are ru n with and without the sewage variable due to the surprising high level of modem sewage levels in the 1991 survey. 2 7 6 Thus, for each of the logistic regression six data set, four regressions are run: a) all independent variables with single birth cohort; b) the same model as (a) but without sewage; c) all independent variables with multiple birth cohort variables; d) same model as (c ) but without sewage. In addition, for each Cox regression four data sets, eig^t regression are run: a) all independent variables with single birth cohort; b) the same model as (a) but without sewage; c) the same model as (a) but without breastfeeding; d) the same model as (a) but without both breastfeeding and sewage; e) all independent variables with multiple birth cohort variables; f) the same model as (e ) but without sewage; g) the same model as (e) but without bre^tfeeding; h) the same model as (e) but without both breastfeeding and sewage. 276 See Chapter 5, pg. 183. 248 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The main OLS assumptions are also to be met in both Cox and logistic regressions. More specifically: 1 ) independence between cases; 2) true specification and measurement of the independent variables; 3) absence of perfect collinearity, that is, independent variables cannot linear functions of other independent variables. To control for the possibility of multicollinearity, all set of regressions are run with and wifiiout the variables sewage and breastfeeding. The possibility of any coUinearity between immunization, breastfeeding and pre natal care by a doctor will be carefully investigated. One of the main hypothesis or statement thesis tested and examined in previous chapters was that the evolution of infant mortality in the Northeast of Brazil is rather contrasting with the patterns shown in other regions of the country, both in timing and nature. W e have found strong evidence supporting this thesis. The other principal hypothesis, to be empirically tested in this study is that infant mortality as measured as is determined by a myriad of complex interactions between demographic, socio-economic and health factors and that, among these factors, the most relevant ones, or the conditions that have the most impact on the odds of infant survival are: 1 ) the education status of the mother; 2) the level of household wealth and income; 3) access to prenatal care by a doctor; 4) breastfeeding status and 5) sewage disposal conditions. 249 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The data show that the dependent variable for infant mortali^, or the odds of survival in the first year of life, indicate that in 1986 only 6% of the deaths occurred after the first year of life. In 1991 and in 1996 this rate would increase to 9%. The relevance of the research subject is directly related to the its importance for policy prescription. In the next sections we will examine the results of the descriptive and multivariate regression analysis applied to all six data-years. We seek to estimate the relative impact and the proportional effect of the explanatory variables on infant mortality, and we are particularly concerned with testing the hypothesis that some independent variables - education, ownership of durable consumer goods, prenatal care by a doctor, sewage and breastfeeding status. In each of the six data-years, the results will be divided in three parts: a) The descriptive results for the variables, which will scrutinize the mean infant mortality values or the odds of dying for all explanatory variables and categories, as well as the respective firequencies and proportions for them. In addition, the correlation levels and significance between the dependent and all independent variables in both the logistic and the Cox regression formatted data sets wül be examined. 250 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. b) The diagnostic results for all the 72 regression models (4 logistic and 8 Cox regressions for each data year) will also be presented. The ggoodness of fit associated with each and every model will be analyzed through fire investigation of the following statistics: - 2 log likelihood, goodness of fit, model chisquare and predictive efficiency. c) The logistic and Cox regression results for all the 72 models. Four statistical measures will be examined: the unstandardized B coefficient, the Exp (B ) or the odds ratio, the standard error and the significance of each in every variable in aU regression models. 251 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. R esults o f D escriptive and M ultivariate R egression Analysis Applied to 1996 Logistic Regression and Cox D ata Sets Descriptive Results The statistic results of the descriptive analysis indicate that all the selected independent variables when examined separately have shown to exert an effect on infant mortality (tables 71 e 72 ) According to the logistic regression data set (3269 cases), the birth cohort variable clearly indicates a decline in the odds of death as we move from children bom between 1980-84, 1985-90 and 1990-95. 8.5% of the children bom between 1980 and 1984 (16.2% of sub-sample) died; 6.7% between 1985 and 1989 ( 27.8% of sub-sample) and 4.2% in the latest birth cohort period ( 48.3% of sub-sample). The values for the oldest birth cohort (1979 and before) in all regressions will invariably be higher than it would be expected due primarily to sample category size (in this case, 253 cases or 7.7%) and self- reporting errors. All regressions were run excluding the oldest birth cohort, but since there were no major changes in the statistic results, the oldest birth cohort were maintained. This variable take on values firom 0 to 3, the former being the oldest cohort, and the latter the newest. According to the Cox data set (3822 cases), the means for the birth cohort categories are exactly the same as Üie logistic regression ones, even th o u ^ the time period for the latest cohort is one year greater (1990-1996). 252 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The most recent cohort represents 55.8% of the sub-sample and included 2131 cases or 553 more than the logistic regression one. The chances of survival for an infant bom in the most recent birth cohort is still 4.2%. The place of residence variable for the logistic regression data set shows that 5.1% of the children bom in urban areas (73.8% of sub-sample) and 7.9% of the ones bom in rural areas died. This variable takes the value 0 for rural and I for urban residence. According to the Cox data set, which puts more weight on recent cases, the percentage of deceased children from urban areas is slightly lower, or 4.7%, whereas the rural rate remains the same. As far as the drinking water, the impact seems to be relevant but not extremely strong. 5.6% of the children bom in households with clean drinking water (73.4%) died, whereas 6.5% of those with no access to good drinking water lost their lives. The values are 0 for not good for drinking and 1 for water good for drinking. In the Cox data set there are no changes in the percentage of children with no access to good drinking water who died, but the odds of death for those who do have access to good drinking water are reduced from 5.6 to 5.2%. One of the possible shortcomings with some of the independent variables presented in this study is that they lack dynamics and they do not reveal the changes in terms of, say, access to clean water or sewage, between the latest 253 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. child’ s first year of life and the time of the survey. In addition to misreporting, this would underestimate the impact of such variables on the dependent variable. An examination of the evolution of the means for each independent variable between 1986 and 1996 will indicate eventual changes. The possible underestimating of the effect of a given independent variable would be relatively substantial if most of the latest children bom pertained to old cohorts which is clearly not the case. In this data set, over 75% of the children were bom in the most recent two cohorts. According to the descriptive results, the availability of modem sewage disposal seems to have a greater impact than water. This variable takes the value 0 for not modem and 1 for modem sewage. 3.7% of the children in households with modem toilet facility and 6.5% in the ones with no modem toilet facility died in the logistic regression formatted data set. In the Cox data set, these rates are lower, 3.5 and 6.2%, respectively. For social-behavioral and physiological reasons, infant mortally rates of children of young mothers (19 years old and under) as well as of older mothers (over 35 years of age ), have been fi-equently pointed out as being higher than that of women age 20 to 34. According to the logistic regression data set descriptive results, 7.6% of the children of high risk mothers die as opposed to 3.8% of the children of mothers at their reproductive peak (low risk). 254 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The Cox regression data set indicates similar rates, or 7.4 and 3.7%, respectively. This demographic variable takes the value 0 for high risk age (outside of the 10-34 good age group) and 1 for low risk. The household crowding variable was created to indicate exposure to health hazards such as infectious and respiratory diseases. Crowded sleeping and living conditions facilitate the spreading of diseases. This variable is a measure indirectly affected by the total fertility rate and related both to birth order and birth spacing. This independent variable was not dichotomized in the regression models but a binary alternative variable taking the value 1 for non crowded homes (less than 5 individuals) and 0 for crowded ones (5 or more) was computed for comparative reasons (6b in the tables). Initially there was a concern that by using more than two sub-categories we would lose degrees of freedom and the results would vary erratically but there did not turn out to be the result and the household crowding covariate was maintained non dichotomized. The means distribution for each total household number clearly shows a progression for the mean value of deceased children per household number. In 2 and 4 individuals households, on average 5% of the latest children of each respondent die before their first year of life. As we increase the number of residents firom 5 to 7, the mean values for the odds of death grow firom 6.9 to 7.9 and to 9 respectively. 255 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In households with 8 or more individuals, the percentage of children who died changes more erratically and in a downward manner but the number of cases is relatively small (households with 8 or more individuals only comprise 7% of the total for the logistic regression data set). The proportional hazards regression results derived firom the Cox data set indicates a very similar statistic progression. The altematwe dichotomous crowding variable shows means of 4.8% (not crowded) and 7.5% (crowded) in the logistic data set, and 4.6% and 7.2% in the Cox formatted data set, respectively. The next variable is considered by many authors in the literature to be perhaps the variable which has the strongest effect on infant mortality, the education level of the mother as indicated by the number of years in school. This variable is divided into the same 4 categories it is often used in standard surveys and ofScial statistical data: higher education (college level and over); secondary (high school level completed); primary (including primary complete and secondary incomplete, I to 11 years); no education. According to the data sets for LR and Cox analysis, only 4.2% and 3.9%, respectively of the Northeast respondents had h i^ e r education. Among those mothers with higher education, .7% and 1.3% of their youngest infants died according to the logistic and Cox regression results, respectively. 256 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. This difierence is meaningless since the sample size for the sub category is very small: 5% for the logistic and 3.9% for the Cox horn a total number of cases of 136 and 150 respectively. 2 7 7 However, the other sub-categories reveal a very robust and progressive influence: the odds of survival decrease from 12, to 6.4 to 3.7% as we move &om no education to primary , to secondary education in the logistic data . In the proportional hazard 1996 model the means are very similar: 12, 6.1 and 3.5%. The relative proportions of educational attainment are also fairly identical. This variable takes values from 0 to 3, the latter being higher education and the former, no education at all. In 83% of the cases in both data sets, the mothers have either primary or secondary education only. The variable goods is a proxy for household income and wealth. It inquiries about ownership of 3 durable consumer goods: car, TV and radio. For each positive answer, one point was assigned. The values range from 0 to 3. Since it is expected that a value 2 indicates ownership of radio and TV, it could hypothesized that a change from 2 to 3 imply great gains in wealth and, analogously, great losses in mortalily rates. The data seem to corroborate this claim. 16.3% of the households in the logistic data set and 15% in the Cox data set had all 3 goods. Similarly, 11.3% and 12.6% of the mothers respectively in the logistic and Cox data sets had none of the three goods. Such mothers will very likely be very poor. ^ Actually the occurrence of an event, or death, only changed between the logistic and the Cox data sets Amn 1 to 2 cases. 257 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The mean values of deceased chfldien who had 0 and I good is 7 and 8.8% in the logistic data set, and 7.3 and 7.8% in the Cox one. With 2 goods, the percentage of children who lose their lives as infants is cut in half to 4.2 and 4%, respectively for the logistic and the Cox formatted data files. If the mother has access to all 3 goods the rate of deceased infants is reduced to 1.7% in both data sets. This economic variable is likely to have a rather significant impact on the odds of survival. The next variable is the infant s birth order risk. 3.9 and 7.4% of the children who, respectively, have a low (2 « * or 3 ^ ^ child) and hig^h order risk ( 1 » * ^ and over 3 " * ) die in the logistic regression data set. The values for the Cox data set are very close: 3.9 and 6.9%. Female infant mortality is usually much greater than male infant mortality in Brazil. As far as the descriptive results of the data sets, 5.3% of the girls and 6.3% of the boys died in the logistic data set. In the Cox data set, these rates are 4.9% and 6.2%, respectively. On average, 52% of the children in the data sets are males. This study hypothesizes that prenatal care by a physician is one of the independent variables that have the greatest impact on infant mortality. Information on this variable (as well as immunization and ethnicity ) was not collected in the 1986 DHS survey. 258 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A passive strategy was implemented to deal with incoherent answers or missing data in preparation for this variable, meaning lhat "don' t know* answers were associated with 0 or no access to prenatal care by a doctor (a similar procedure was carried out with the immunization variable). Prenatal care by a nurse was considered to be included in the model but statistic tests showed that access to a doctor would be a much better measure of health care. 70% of the respondents m the logistic data set and 63% in the Cox data set did not see a physician for the prenatal care of the infant. This difference could be ascribed to the possibility that access to a doctor has been facilitated in recent years (the Cox data set include children bom between March of 1995 and March of 1996). Remarkably the percentage of deceased children in both data sets is the same: 2.4 and 7.1% (for the assigned 0 and 1 values). The variable ethnicity is presented in a binary form as well as in 4 sub- categoiies. Since the percentage of people of black, Asian and Indian descent is relatively low (around 4.6%) and since ethnic background is usually presented in demographic studies in Brazil in a dichotomized manner, the same was done here. There is an association between race and socio-economic status in Brazil. Around 71% of the respondents are not white. The Northeast has the highest concentration of people of non-white descent in Brazil. 259 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. This distributioii is very close to IBGE’ s official data, on ethnic background for 1996 (table 2.1.1), which indicate a value of 69.4% of non white people in Northeast’ s population. The odds of death for infants of white heritage is 4.7% in the logistic and 4.3% in the Cox data sets. In non-white households such respective rates are 6.3 and 6.1%. The last independent variable is immunization or whether the infant was vaccinated for all three DPT shots. If positive the value assigned was 1, and 0 otherwise. According to the descriptive results, there is a strong correlation between immunization status and infant mortality. Around 69% of the children did not receive all three shots of the DPT vaccine in both data sets. Not even one child out of the 31% of those who were immunized died. In contrast, the mean of deceased children among the ones not immunized was 8.4 and 8.1%, respectively in the logistic and in the Cox data sets. The Cox regression data set attests the importance of breastfeeding for the life of the child. 57.6% of the infants or 2,200 cases were not breast-fed at all. 4.2% of the children were breast-fed for just 1 month, 5.2% for 2 and 5.9% for three months. Only 1.8% of the children were breast-fed for 12 months or over. The mean value of deceased children per months of breastfeeding decrease in a progressive manner: 7.7 % (0), 5.6% (1), 4.1% (2), 3.1% (3), 1.5% (4) and so on. The variation in this value after 5 or 6 months could have been influenced by self-reporting errors. None of the children who were breast-fed for 11 months or more died. 260 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Next we will examine the correlation levels with the dependent variable according to the 1996 Cox and logistic data sets, (table 73) According to the data, the variables sex and e th n ic ity and d rin k in g w a te r are not even correlated with the dependent variable. As expected, the correlation coefficient for the non-binary household crowding variable is significant (0.048 in both data sets) and positively correlated with the dependent variable. All the other variables are significant at the 0.01 level indicating a very low probability that the results could be attributed to chance. The variables which show a strongest level of correlation are: hnmunization levels ( -0.166 and -0.163, respectively in the logistic and Cox data sets); duration of breastfeeding ( -0.101); educational attainment (-0.12 in the logistic and - 0.115 in the Cox data set); goods ( -0.099 and -0.092) and pre-natal care ( -0.094 and -0.099, respectively in the logistic and Cox data sets). 261 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 71 262 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. sss 263 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. H its 264 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 72 l 7 S 265 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 266 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. M o . } ' ■ N .}■ \ f j, . " / , ' n ' ' ' \ 7 ' " ; . \ ■ ' ' '’^ \ * Reproduced with perm ission of the copyright owner. Further reproduction prohibited without permission. 268 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 73 1996 Correlation Levels witti D ependent Variable Logistic Regression Data Set Cox Regression Data S et Variables Cohort* -0.059 -0.062 Urban X Rural -0.054 -0.064 Drinking W ater -0.017 X -0.026 X Sewage -0.05 -0.049 Household Crowding* 0.048 0.048 Household Crowding (B) -0.055 -0.056 Mother's Age Risk -0.081 -0.08 Educational Level* -0 .1 2 -0.115 Goods* -0.099 -0.092 Birth Order Risk -0.075 -0.067 Sex of Child -0.021 X -0.028 X Breastfeeding** N/A -0.101 Ethnicity* 0.023 X 0.028 X Ethnicity (B) -0.03 X -0.036# Dr's Prenatal Care -0.094 -0.099 Immunization (DPT123) -0.166 -0.163 * non dichotomous * * in months A H vansWes Significant St the 0.01 levet g tailed) unless indicated otheneise # Significant at the O .O S level ( 2-taHad) XNotConeMed 269 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Diapiostic Results for 1996 LR Models Tlie diagnostics statistics reveal that all models fit the facts well, (table 74) The -2 Log Likelihood statistic is the analogue of the Sum of the Square Errors (SSE) in OLS regression and expresses how poorly the models fit with all the included variables. The larger tiie -2LL, the worse is relatively the prediction power for the dependent variable. Menard observes that the log- Hkelihood statistic is the most accurate way to investigate the significance of the contribution of the independent variables to the explanation of the dependent variable. 2 7 8 In the logistic data set, the data indicate that the models including sewage ( 1 & 3 ) are clearly better. The best prediction power for the dependent variable is found in model 3 with the multiple birth cohort variables, the one with the h ip e st number of degrees of fi*eedom and the lowest -2LL. The models chisquare indicate that the null hypothesis that the value of the coefiBcients of the variables is equal to zero should be rejected. All models have excellent goodness of fit. 3240 cases were selected and 29 missing. The prediction efficiency of the models is 94.29%. 2 7» Scott Menard, Ibid., 39. 270 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 74 Diagnostic Results, Predictive Efficiency and FurttterCtiaracterfstics of th e Logistic Regression Models for 1996 D ata S et Item Stat- Model 1 Model 2 Model 3 Model 4 -2Log Likelihood 1196.056 1200.704 1194.798 1199.513 Goodness of Fit 2180.764 2157.707 2185.15 2163.273 Model Chisquare 222.474 217.826 223.732 219.017 (Signif) 0 0 0 0 (df) 13 12 15 14 Predictive Efficiency 94.29% 94.29% 94.29% 94.29% N (numljer of cases included ) 3240 3240 3240 3240 271 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. O n. the Regression Estimates The regression, results for both the logistic and the Cox method are expressed through the maximum likelihood estimates for all the predictors. The key results are: the parameter estimates or B statistics; the respective standard errors of these estimates; the odds ratios (E b c p (B )) associated with each coefficient and the p value or statistic significance of each estimate. The p value or significance level is perhaps the most powerful statistic of all and it is derived fi* om the Wald statistic. The latter is the analogue of the t- test in OLS regression and it is measured as the square of the parameter estimate divided by its standard error. The Wald statistic is asymptotically distributed as a chisquare with degrees of fi*eedom equal one less than the total amount of categories. The underlying null hypothesis associated with the p value professes that, at a given level of confidence, the explanatory variable has no effect on the dependent variable associated with infant mortality. A significance level of 0.05 or less indicates that we must reject the Ho and accept that the independent variable does have a statistically significant impact on infant mortality. The significance level is also associated with the likelihood of obtaining the regression estimates by chance. 272 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The B coefficient is the unstandardized regression coefficient or the non linear change in the dependent variable Logit ( Y ) associated with a one unit increase in the independent variable. The B coefficient is usually employed to investigate the effects of a given variable in different samples or models. The parameter estimate can be computed as the log of the odds ratio E ix p (B). Its standard error is also examined and, in a like manner, it is calculated through maximum likelihood estimation. The odds ratio (Exp(B)) or the M LE of the odds ratio is the number by which we must multiply the odds of dying for each one-unit increment in die value of the independent variable. When the odds ratio is less than 1, the odds of dying for the infant decrease as the value of the independent variable increases. In contrast, if the E b q > (B ) is greater than I, the odds of dying increase as the value of the explanatory variable increases. Logistic Regression Results for 1996 Models The 2 main logistic regression models for 1996 (model 1 and 3) are comprised of 12 independent variables in addition to the birth cohort variable or variables. The other 2 regression models (model 2 and 4) include the same variables except the important sewage variable, (tables 75 to 78) Among the 3 non-pooled logistic data sets (the 3 others are pooled), the 1996 data set, derived from the most recent phase of the DHS in Brazil (1996) and constituted of a sub-sample size of 3269 children, is perhaps the one with the highest explanatory power. 273 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Notice that the 9 out of the 12 explanatory variables are coded in a dichotomous fashion. The value 0 is associated with the category which is supposed to have a negative ùnpact on infant mortality and the value 1 related to the opposite expectation. The 3 non binary variables are: mother s education (4 categories ranging from 0 - no education- to 3 - higher education); goods (4 categories ranging from 0 - none of the 3 goods - to 3 - ownership of radio, T V and car ); household crowding (n). The regression results for the 4 models indicate that the independent variables that have a statistically significant effect on infant mortality at the 0.05 level are: mother^s education, goods, prenatal care by a doctor and sewage. These results do confirm the main hypothesis of this dissertation. In model 1 (complete model with single birth cohort), the variables with the lowest p values are, respectively: 1 ) goods (0.0003); 2) Dr’ s prenatal care (0.0015); 3) mother’ s education (0.0019); 4) sewage (0.0382). (table 75 A ) In model 3 (complete model with multiple birth cohorts), the variables ■ w ith the lowest p values are, respectively: 1 ) goods (0.0004); 2) mother’ s education (0.0018); 3) Dr’ s prenatal care ( 0.0058); 4) sewage (0.0369). (table 77) In model 2 (single birth cohort with no sewage variable), the variables with the lowest p values are, respectively: 1 ) goods (0.0002); 2) mother’ s education (0.0011); 3) Dr’ s prenatal care (0.0012). (table 76) 274 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In model 4 (multiple birtfa. cohorts with, no sewage variable), the variables with the lowest p values are, respectively: 1 ) goods (0.0002); 2) mother's education (0.001); 3) Dr’ s prenatal care (0.0045). (table 78) An analysis of the B coefficients’ signs in all 4 models would indicate that drinking water, household crowding and prenatal care are positively correlated with in fan t mortality, implying that better water, prenatal care by a physician and crowding Imng conditions would increase the likelihood of death for the infant. The odds ratios for the three variables (greater than 1 ) would indicate the same positive relationship between the explanatory variables and the dependent variable. The results for the household crowding variable do fit the facts well. The results for water (even if one accepts that negative impact of better drinking water on weaning) 279 and prenatal care would seem surprising and unfitting, though. However, since only the prenatal care variable is statistically significant, the results for the other two independent variables could be ascribed to sampling error and chance. If the p levels of significance are not sufficiently small (confidence level of at least 95%), the hypothesis testing is just not stringent enough and the likelihood that the results were obtained by chance is too high. As suggested by John Marcotte m The Impact of Piped Water on Child M oitali^ and Breastfeeding in Ecuador and Brazil, Ph D dissertation. University of ^fisconsm-Madisoo, 1988. 275 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A scrutiny of the regression coefScients for Üie independent variables which are significant at the 0.05 level m model 1 indicates that, when modem toilet facilities are available, and given the impact of other explanatory variables, the odds of dying decrease m 63% in comparison to what they would be if there were no modem sewage, (table 75 A ) Similarly, the odds of dying for the infant child decrease in 69% as the mother improves her educational attainment firom one category to another, (table 75 A ) As the household acquires one additional consumer good, a radio, a TV or a car, the odds of dying for the family's child are also reduced in 68% in comparison to what they would be if that consumer good was not available. (table 75 A ) In contrast, the significant prenatal care covariate would indicate that the unstandardized B coefficient is positive (0.8774) and the odds would be much greater than 1. If this result is factual, they would contradict conventional wisdom, the literature and other statistical procedures examined in this study, (table 75 A ) On the other hand this finding would corroborate the results attained recently by Chen while investigating infant mortality in the Los Angeles Co. 2» o Since the results may be due to some level of colUinearity between prenatal care and immunization, and in order to further delve into the prenatal ^ Xue Hong Chen, Ibid., 1997. 276 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. care variable, all the regressions were run again, first with the DPT immunization variable but wiüiout the prenatal care by a doctor, and then without the DPT immunization variable. The findings can be extended to all the other data sets involving such variables (96, 91, 96-91 LR and Cox). When the logistic regressions are run without prenatal care variable, there are no substantial changes as far as significance levels, standard errors, B coefficients or odds ratios for the other independent variables. The very same variables are statistically significant. As far as the immunization variable itself, and taking model 1 as the comparative model, the odds ratio for DPT immunization remains the same (0.0001), the significance p level does not change much (firom 0.266 to 0.2903 without the prenatal care variable), neither does the standard error (firo m 8.3391 to 8.34370) or the B coefficient (firom - 9.2751 in model 1 to -8.8224). Conversely, table 75 B examines the regression results for model 1 without the DPT immunization variable. Goods, mother’ s education, sewage and prenatal care are still the only 4 variables which have a statistically significant impact on infant mortaliy. Again there are no changes at all as far as the significance levels and the regression coefficients results for all the other independent variables. Hence, in the 96 Cox Regressions, as well as in the 91 and 96-91 logistic and Cox data sets, we will only show the Exp (B ), B, Se and p coefficients for the prenatal care variable when the immunization variable is omitted as opposed to showing all the coefficient results in a table such as 75 B. 277 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. We need to be sure whether the 4 regression statistics for the prenatal care in all 4 models do reflect the facts or whether they were altered by the DPT variable. Table 75 B for Model I indicates that the direct correlation between prenatal care and the dependent variable shown in table 75 A was indeed caused by the presence of the DPT immunization variable and do not reflect the facts. As far as the significan ce levels for the prenatal care variable there are no major changes: the p value increases slightly firom 0.0015 to 0.0026. The standard error decreases firom 0.276 to 0.2561. However, the unstandardized B coefficient changes fi* o m 0.8774 in model 1 (75 A ) to -.7710 in the alternative model (75 B). The odds ratio is also not greater than 1 (2.4047 ) but much lower, or .4625. Thus, access to prenatal care by a doctor does improve the odds of survival. In a model with all variables included but immunization, the odds of survival for the infant would increase in 46% as the infant receives proper medical prenatal care. In short, the 1996 logistic regression models indicate that the following variables have an effect on the likelihood of infant mortality at the 5% level of significance: goods, mother’ s education, sewage and prenatal care by a doctor. These regression results fit nicely with this study’ s main hypothesis. 278 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 75 A Logistic Regression Results for 1996 Ne D ata S et (1) Variable Name B Standard Error Exp(B) Significance (jp) Birth Cohort -0.0447 0.1039 0.9563 0.6673 Urban X Rural -0.0707 0.1982 0.9318 0.7214 Drinking W ater 0.2863 0.1989 1.3316 0.15 Sewage -0.4597 0.2218 0.6315 0.0382 Household Crowding 0.0566 0.0367 1.0582 0.1234 Age Group of Mother -0.2688 0.1973 0.7643 0.1731 Ethnicity -0.0357 0.188 0.9649 0.8492 Mother's Education -0.3672 0.1185 0.6927 0.0019 Goods -0.3898 0.1088 0.6772 * 0.0003 Birth Order Risk -0.2597 0.1807 0.7713 0.1507 Dr's Prenatal care 0.8774 0.276 2.4047 * 0.0015 Sex of Child -0.2441 0.1581 0.7834 0.1225 DPT Immz. -9.2751 8.3391 0.0001 0.266 Constant -1.4275 0.3228 - 0 Source: 1996 DHS for Brazil's Northeast region from a sub-sam ple total of 3269 cases (3240 selected, 29 missing). 279 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 7 5 B Logistic Regression Results for 1996 Ne Data S et (1 with no DPT Immunization) Variable Name B Standard Error E xpO ) Significance (p) Birth Cohort -.1629 .1014 .8497 .1083 Urban X Rural -.0255 .1951 .9748 .8959 Drinking W ater .2653 1955 1.3038 .1747 Sewage -.4960 .2200 .6090 .0242 Household Crowding .0535 .0358 1.0550 .1352 Age Group of Mother -.3457 .1960 .7077 .0777 Ethnicity -.0400 .1852 .9608 .8291 Mother's Education -.3505 .1852 .7044 .0029 Goods -.3831 .1067 .6817 .0003 Birth Order Risk -.2797 .1772 .7560 .1145 Dr's Prenatal care -.7710 .2561 .4625 • .0026 Sex of Child -.2391 .1558 .7874 .1250 Constant -1.3166 .3180 - 0 Source: 1996 DHS for Brazil's Northeast region from a sub-sam ple total of 3269 cases (3240 selected, 29 missing). 280 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 76 Logistic Regression Results for 1996 Ne D ata S e t (2) Variable Name B Standard Error Exp(B) Significance (p) Birth Cohort -0.0306 0.1034 0.9698 0-767 Urban X Rural -0.0902 0.1977 0.9138 0.6482 Drinking W ater 0.2585 0.1986 1.295 0.193 Household Crowding 0.0542 0.0365 1.0557 0.1379 Age Group of Mother -0.2673 0.1972 0.7654 0.1753 Ethnicity -0.0468 0.1875 0.9543 0.8031 Mother's Education -0.3856 0.1184 0.68 0.0011 Goods -0.4038 0.1083 0.6678 0.0002 Birth Order Risk -0.275 0.1803 0.7596 0.1272 Dr's Prenatal care 0.8916 0.2753 2.4391 0.0012 Sex of Child -0.2449 0.1578 0.7828 0.1207 DPT Immz. -9.2919 8.35 0.0001 0.2658 Constant -1.4507 0.3215 - 0 Source: 1996 DHS for Brazil’ s N ortheast region from a sub-sam ple total of 3269 cases (3240 selected. 29 missing). 281 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 77 Logistic Regression R esults for 1996 Ne Data S et (3) Variable Name B Standard Error Exp(B) Significance (p) Cohort 90-95 0.0066 0.3438 1.0066 0.9848 Cohort 85-89 -0.1085 0.3009 0.8972 0.7185 Cohort 80-84 0.1485 0.303 1.1601 0.6241 Urt>an X Rural -0.0596 0.1987 0.9421 0.7643 Drinking W ater 0.2874 0.1992 1.333 0.149 Sewage -0.4632 0.2219 0.6293 0.0369 Household Crowding 0.0552 0.0368 1.0567 0.1333 Age Group of Mother -0.2579 0.1995 0.7727 0.196 Ethnicity -0.0287 0.1882 0.9717 0.8787 Mother's Education -0.3705 0.1187 0.6904 0.0018 Goods -0.3848 0.109 0.6806 0.0004 Birth O rder Risk -0.2611 0.1807 0.7702 0.1484 Dr's Prenatal care 0.8108 0.294 2.2498 0.0058 S ex of Child -0.2408 0.1582 0.786 0.1279 DPT Immz. -9.2858 8.3463 0.0001 0.2659 Constant -1.513 0.3631 - 0 Source: 1996 DHS for Brazil's Northeast region from a sub-sample total of 3269 cases (3240 selected, 29 missing). 282 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 78 Logistic Regression Results for 1996 Ne Data S et (4) Variable Name B Standard Error Exp(B) Significance (p) Cohort 90-95 0.0478 0.3428 1.049 0.889 Cohort 85-89 -0.0684 0.3 0.9339 0.8197 Cohort 80-84 0.1691 0.3024 1.1843 0.576 Urban X Rural -0.0799 0.1981 0.9232 0.6869 Drinking W ater 0.2589 0.1988 1.2955 0.1927 Household Crowding 0.0529 0.0366 1.0543 0.1482 Age Group of Mother -0.2561 0.1992 0.7741 0.1987 Ethnicity -0.0398 0.1877 0.961 0.8321 Mother's Education -0.3893 0.1186 0.6775 0.001 Goods -0.3993 0.1085 0.6708 0.0002 Birth Order Risk -0.2752 0.1803 0.7595 0.127 Dr's Prenatal care 0.8322 0.2932 2.2983 0.0045 Sex of Child -0.2424 0.1579 0.7847 0.1247 DPT Immz. -9.3016 8.3569 0.0001 0.2657 Constant -1.5416 0.3627 - 0 Source: 1996 DHS for Brazil's Northeast region from a sutxsam ple total of 3269 cases (3240 selected, 29 missing). 283 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Diagnostic Results for 1996 Cox Models The diagnostics statistics reveal th at all 8 Cox regression models fît the facts well, (table 79) One of the main caveats of the logistic model is that it does not take into account time-varying variables such as duration of breastfeeding in months. In contrast, the Cox method, and event history or survival analysis in general, does not ignore the timing of the event (death of the infant). Cox proportional risks method of partial likelihood defines a risk set and censors the data for the occurrence of events (the qualitative transition change). The quantitative death duration variable (or the age of death in months) and the categorical dichotomous death status variable are censored to exclude the cases in the data set (sub-sample size of 3822 cases) in which the infant died or to exclude the occurrence of events. In the 1996 Cox models with no breastfeeding variable ( models 3,4,7 and 8), 23 cases were dropped and 3799 selected. 95.1% of these children are censored and survive the fîrst year of life whereas 186 do make the transition and do not survive. In the models including the duration of breastfeeding in months (models 1, 2, 5 and 6), 35 cases were dropped and 3787 selected. 95.1% of the values are censored and 185 children do make the transition. In the Cox data set, the data indicate that the complete models including sewage and breastfeeding ( 1 & 5 ) are clearly better. The respective -2LL of these two models are 2796.589 and 2794.125. 284 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Overall, the best prediction power for the dependent variable is found in model 5 with the multiple birth cohort variables, the one with the highest number of degrees of freedom and the lowest > 2L L . The model chisquare or Gm is similar to the F test in linear regression and it is the difference between the initial log-likelihood (D O ) and the -2LL for the model (Dm). The models chisquare are very significant, denoting that the null hypothesis that the value of the coefScients of the variables is equal to zero should be rejected. The independent variables do improve the predictability of the models. The Cox regression models with the highest chisquare are models 5 and 1: 169.761 and 168.046, respectively. 285 Reproduced with permission of the copyright o w n er. Further reproduction prohibited without permission. Table 79 Diagnostic Results. Predictive Efficiency and Furttier Ctiaracteristlcs of ttie Cox Regression Models for 1996 D ata S et (1 of 2) Item Stat. Model 1 Model 2 Model 3 Model 4 Events 185 185 186 186 Censored 3602 3602 3613 3613 (%) 95.1 95.1 95.1 95.1 -2Log Likelittood 2796.589 2802.157 2843.194 2848.227 Chisquare (Overall ) 168.046 167.753 160.268 156.263 (Signif) 0 0 0 0 (df) 14 13 13 12 N (numtier of cases included ) 3787 3787 3799 3799 286 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 80 Diagnostic Results, Predictive Efficiency and Further Characteristics of the Cox Regression Models for 1996 Data S et (2 o f 2) Item Stat. M odels Model 6 Model 7 M odels Events 185 185 186 186 Censored 3602 3602 3613 3613 (%) 95.1 95.1 95.1 95.1 -2Log Likelihood 2794.125 2799.927 2842.167 2847.226 Chisquare (Overall ) 169.761 165.446 161.851 157.852 (Signif) 0 0 0 0 (df) 16 15 15 14 N (number of cases included ) 3787 3787 3799 3799 287 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Cox’ s proportional hazards method allows for the inclusion of Hie extremely important time-dependent breastfeeding variable. The results for the regression method should be similar to the logistic regression results. The 2 standard Cox regression models for 1996 (model 1 and 5) are comprised of 13 independent variables plus the birth cohort variable or variables. 2 regression models (model 2 and 6) include the very same variables with the exception of sewage and 2 others ( models 3 and 7) consist of the same models without the duration of breastfeeding variable. The last 2 models (models 4 and 8) exclude bodi sewage and breastfeeding, (tables 81 to 88). Among the 6 Cox data sets (3 of which are pooled), the 1996 data set, also derived firom the most recent phase of the DHS in Brazil (1996) and constituted of a sub-sample size of 3822 children, is maybe the one with the highest explanatory power. The regression results for the 8 models indicate that the independent variables that have a statistically significant effect on infant mortality are: breastfeeding, mother’ s education, goods, prenatal care by a doctor and sewage. In model 1 (complete model with single birth cohort), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0001); 2) goods (0.0002); 3) mother’ s education (0.0058); 4) Dr’ s prenatal care (0.0072); 4) sewage (0.0252). (table 81) 288 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In model 5 (complete model with multiple birth c o h o rts), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0001); 2) goods (0.0003); 3) mother’ s education (0.0054); 4) sewage (0.0252); 5) Dr’ s prenatal care (0.0263). (table 85) In model 2 (single birth cohort with no sewage variable), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0001); 2) goods (0.0001); 3) mother’ s education (0.0036); 3) Dr’ s prenatal care (0.0068). (table 82) In model 6 (multiple birth cohorts with no sewage variable), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0001); 2) goods (0.0001); 3) mother’ s education (0.0033); 3) Dr’ s prenatal care (0.0068). (table 86) In model 3 (single birth cohort with no breastfeeding), the variables with the lowest p values are, respectively: 1 ) goods (0.0006); 2) mother’ s education (0.006); 3) sewage (0.0329). (table 83) In model 7 (multiple birth cohorts with no breastfeeding), the variables with the lowest p values are, respectively: 1 ) goods (0.0007); 2) mother’ s education (0.0053); 3) sewage (0.0324). (table 87) In model 4 (single birdi cohort without sewage and breastfeeding), the variables with the lowest p values are, respectively: 1 ) goods (0.0003); 2) mother’ s education (0.0037). (table 84) 289 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In model 8 (multiple birth, cohorts without sewage and breastfeeding), the variables with the lowest p values are, respectwely: 1 ) goods (0.0003); 2) mother’ s education (0.0032). (table 88) An analysis of the B coefScients’ signs in all 8 Cox regression models would indicate that drinking water, household crowding and prenatal care are positively correlated with in fan t mortality meaning that better water, prenatal care by a physician and crowding living conditions would increase the likelihood of death for the infant. The odds ratios for the three variables (greater than 1) would indicate the same positive relationship between the explanatory variables and the dependent variable. The results for the household crowding variable do fit the facts well. The results for water (even if one accepts that negative impact of better drinking water on weaning) and prenatal care would seem surprising and unfitting, though. Among these 3 covariates, the prenatal care variable is the only statistically significant one (in models 1, 2, 5 and 6), so the results for the other two independent variables could be ascribed to sampling error and chance. The question which arises here is not so much why prenatal care seems to be positively correlated with infant mortality but rather why prenatal care by a doctor is not even statistically significant in the remaining 4 models. W e then hypothesized again that these results were caused by a collinearity effect between immunization and prenatal care. 290 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The regressions (using the multiple cohort models 7 and 8 as references) were run without the immunization variable. As a result, not only did prenatal care become significant ( at a level of 0.011 for model 7 and 0.0122 for model 8) but the B coefEdents became negative ( -.6099 and -.6015, respectively) and the odds ratios became lower than 1 ( .5434 and .548 for the adjusted models 7 and 8). Such results contradict the positive correlation results between infant mortality and prenatal care reached by Chen in the Los Angeles county, Hence, the hypothesis of a certain degree of collinearity between immunization and prenatal care is accepted. One may conclude that access to prenatal care with a doctor improves the odds of survival and does have a statistically significant impact on infant mortality in all 8 models. The variable sex is almost statistically significant at the 95% level of confidence in models 3 and 7 (no breastfeeding) and 4 and 8 (no breastfeeding and no sewage). A scrutiny of the Cox regression coefficients for the independent variables which are significant at the 0.05 level in model 1 indicates that, when modem toilet facilities are available, and given the impact of other e^lanatoiy variables, the odds of dying decrease in 61% in comparison to what they would be if there were no modem sewage, (table 81) Similarly, the odds of dying for the infant child decrease in 73% as a mother improves her educational attainment fiom one category to another, (table 81) Xue Hong Chen, Ibid., 1997 291 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. As the household acquires one additional consumer good, a radio, a TV or a car, the odds of dying for the family's child is also reduced in 68% in comparison to what they would be if that consumer good was not available. All these unstandardized Cox regression coefScients match the logistic regression ones. In sum, the 1996 Cox regression models indicate that the following variables have an effect on the likelihood of infant mortality a t the 5% level of significance: breastfeeding duration, goods, mother's education, sewage and prenatal care by a doctor. The regression results for the Cox proportional hazards method conform both with the logistic results as well as with this study's main hypothesis. 292 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 81 Cox Regression Results for 1996 Ne D ata S et (1) Variable Name 8 Standard Error Exp(B) Significance (jp ) Birth Cohort 0.0493 0.1019 1.0505 0.6286 Urban X Rural -0.2596 0.1805 0.7714 0.1503 Drinking Water 0.253 0.1836 1.2879 0.1682 Sewage -0.4923 0.22 0.6112 0.0252 Household Crowding 0-0547 0.0335 1.0562 0.1024 Age Group of Mother -0.193 0.1786 0.8245 0.2799 Ethnicity -0.2149 0.1815 0.8067 0.2366 Mother's Education -0.3087 0.1119 0.7344 0.0058 Goods -0.386 0.1034 0.6797 0.0002 Birth Order Risk -0.2896 0.171 0.7486 0.0904 Dr's Prenatal care 0.6582 0.2451 1.9313 0.0072 Sex of Child -0.2762 0.149 0.7587 0.0638 DPT Immz. -12.0998 56.5262 5.56E-06 0.8305 Breastfeeding in Months -0.2118 0.0553 0.8091 0.0001 Source: 1996 DHS for Brazil's Northeast region from a sub-sam ple total of 3822 cases (3787selected, 35 dropped). 293 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 82 Cox Regression Results for 1996 Ne Data S et (2) Variable Name B Standard Error Exp(B) Significance (p) Birth Cohort 0.0608 0.1015 1.0627 0.549 Urban X Rural -0.2771 0.1804 0.758 0.1244 Drinking W ater 0.218 0.1833 1.2436 0.2344 Household Crowding 0.052 0.0333 1.0534 0.1185 Age Group of Mother -0.1865 0.1785 0.8298 0.2962 Ethnicity -0.2263 0.1814 0.7975 0.2122 Mother's Education -0.326 0.112 0.7218 0.0036 Goods -0.4042 0.1029 0.6675 0.0001 Birth Order Risk -0.3008 0.1707 0.7402 0.078 Dr's Prenatal care 0.6608 0.2441 1.9363 0.0068 Sex of Child -0.2735 0.149 0.7607 0.0664 DPT Immz. -12.1141 56.5264 5.48E-06 0.8303 Breastfeeding in Months -0.2096 0.055 0.8109 0.0001 Source: 1996 DHS for Brazil's N ortheast region from a sub-sam ple total of 3822 cases (3787selected, 35 dropped). 294 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 83 Variable Name Cox Regression Results for 1996 Ne D ata Set (3) B Standard Error Exp (B) Significance (p) Birth Cohort -0.0371 0.0991 0.9636 0.7085 Urban X Rural -0.1736 0.1804 0.8407 0.3359 Drinking W ater 0.2381 0.1813 1.2688 0.1891 Sewage -0-4703 0.2205 0.6248 0.0329 Household Crowding 0.0381 0.0329 1.0388 0.2468 Age Group of Mother -0.2338 0.1785 0.7915 0.1903 Ethnicity -0.1961 0.1815 0.8219 0.2798 Mother's Education -0.3071 0.1027 0.7355 0.006 Goods -0.351 0.1454 0.704 0.0006 Birth O rder Risk -0.2801 0.1708 0.7557 0.1009 Dr's Prenatal care 0.1731 0.2214 1.189 0.4344 Sex of Child -0.2844 0.1484 0.7525 0.0553 DPT Immz. -12.6315 60.0021 3.27E-06 0.8333 Source: 1996 DHS for Brazil's Northeast region from a sut)-sam ple total o f3822 cases (3799 selected, 23 dropped). 295 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 84 Cox Regression Results for 1996 Ne D ata S e t (4) Variable Name B Standard Error Exp(B) Significance (p) Birth Cohort -0.025 0.0988 0.9753 0.8005 Urban X Rural -0.1925 0.1801 0.8249 0.2851 Drinking W ater 0.2076 0.181 1.2307 0.2515 Household Crowding 0.0355 0.0327 1.0362 0.2774 Age Group of Mother -0.2261 0.1785 0.7976 0.2051 Ethnicity -0.2063 0.1813 0.8136 0.2551 Mother's Education -0.324 0.1117 0.7232 0.0037 Goods -0.3686 0.1022 0.6917 0.0003 Birth O rder Risk -0.2927 0.1705 0.7462 0.086 Dr's Prenatal care 0.1862 0.2212 1.2047 0.3999 Sex of Child -0.281 0.1483 0.755 0.0582 DPT Immz. -12.6456 59.9877 3.22E-06 0.833 Source: 1996 DHS for Brazil's N ortheast region from a sub-sam ple total of 3822 cases (3799 selected, 23 dropped). 296 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 8 5 Cox Regression Results for 1996 Ne Data S et (5) Variable Name 8 Standard Error Exp(B) Significance (p) Cohort 90-96 0.2878 0.3404 1.3335 0.3978 Cohort 85-89 -0.0287 0.3104 0.9717 0.9263 Cohort 80-84 0.2003 0.314 1.2217 0.5236 Urban X Rural -0.2486 0.181 0.7799 0.1696 Drinking W ater 0.2571 0.1838 1.2932 0.1618 Sewage -0.5023 0.2201 0.6051 * 0.0225 Household Crowding 0.0535 0.0334 1.055 0.1095 Age Group of Mother -0.1845 0.1805 0.8315 0.3067 Ethnicity -0.2091 0.1817 0.8113 0.2497 Mother's Education -0.3118 0.1121 0.7321 * 0.0054 Goods -0.3756 0.1035 0.6869 0.0003 Birth Order Risk -0.2939 0.1707 0.7454 0.0852 Dr's Prenatal care 0.562 0.2529 1.7541 0.0263 Sex of Child -0.2695 0.1491 0.7637 0.0706 DPT Immz. -12.1167 56.59 5.47E-06 0.8305 Breastfeeding in Months -0.2216 0.0562 0.8013 0.0001 Source: 1996 DHS for Brazil's Northeast region from a sub-sample total of 3822 cases (3787selected, 35 dropped). 297 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 86 Cox Regression R esults for 1996 Ne Data S et (6) Variable Name B Standard Error Exp(B) Significance (p) Cohort 90-96 0.3211 0.34 1.3787 0.345 Cohort 85-89 0.0098 0.31 1-0098 0.9748 Cohort 80-84 0.2163 0.3139 1.2415 0-4907 Urban X Rural -0.2674 0.1808 0-7654 0-1392 Drinking W ater 0.2203 0.1835 1.2465 0.23 Household Crowding 0.0508 0.0333 1.0522 0.1264 Age Group of Mother -0-1779 0.1803 0.837 0.3238 Ethnicity -0.2204 0-1815 0.8022 0.2246 Mother's Education -0.3293 0-1121 0.7194 0.0033 Goods -0.3945 0-103 0.674 0.0001 Birth Order Risk -0.3035 0-1704 0.7382 0.0749 Dr's Prenatal care 0.5716 0.2522 1.7711 • 0.0235 S ex of Child -0.2683 0-149 0.7646 0.0717 DPT Immz. -12.1306 56.6026 5.39E-06 0.8303 Breastfeeding in Months -0.2187 0.0559 0-8036 0.0001 Source: 1996 DHS for Brazil's N ortheast region from a sub-sam ple total of 3822 cases (3787selected, 35 dropped). 298 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 87 Cox R egression Results for 1996 Ne D ata S e t (7) Variable Name B Standard Error Exp(B) Significance (p) Cohort 90-96 0.0387 0.3423 1.0395 0.9099 Cohort 85-89 0.0238 0.3098 1.0241 0.9387 Cohort 80-84 0.2335 0.3137 1.263 0.4567 Urban X Rural -0.1663 0.1807 0.8468 0.3575 Drinking W ater 0.238 0.1813 1.2686 0.1893 Sewage -0.4715 0.2205 0.624 0.0324 Household Crowding 0.0365 0.033 1.0372 0.2686 Age Group of Mother -0.2194 0.1799 0.803 0.2226 Ethnicity -0.1931 0.1816 0.8244 0.2876 Mother's Education -0.3121 0.1119 0.7319 0.0053 Goods -0.3495 0.1032 0.7051 0.0007 Birth Order Risk -0.2831 0.1707 0.7535 0.0974 Dr's Prenatal care 0.1655 0.2375 1.18 0.4859 Sex of Child -0.2827 0.1484 0.7537 0.0567 DPT Immz. -12.642 60.2561 3.23E-06 0.8338 Source: 1996 DHS for Brazil's N ortheast region from a sub-sam ple total of 3822 cases (3799 selected, 23 dropped). 299 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 88 Cox Regression R esults for 1996 Ne Data S et (8) Variable Name B Standard Error Exp(B) Significance (p) Cohort 90-96 0-074 0.3419 1.0768 0.8287 Cohort 85-89 0.0596 0.3094 1.0614 0.8472 Cohort 80-84 0.2489 0.3135 1.2826 0.4274 Urban X Rural -0.1858 0.1804 0.8305 0.303 Drinking W ater 0.2067 0.1809 1J2296 0.2533 Household Crowding 0.0342 0.0328 1.0348 0.2975 Age Group of Mother -0.2118 0.1797 0.8091 0.2385 Ethnicity -0.2035 0.1814 0.8159 0.2619 Mother's Education -0.3293 0.1119 0.7194 0.0032 Goods -0.3676 0.1027 0.6924 0.0003 Birth O rder Risk -0.2933 0.1705 0.7458 0.0854 Dr's Prenatal care 0.1845 0.2374 1.2026 0.4369 S ex of Child -0.2795 0.1483 0.7562 0.0595 DPT Immz. -12.655 60.2405 3.19E-06 0.8336 Source; 1996 DHS for Brazil's Northeast region from a sub-sample total o f 3822 cases (3799 selected. 23 dropped). 300 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. R esults of D escriptive and Multivariate R egression Analysis Applied to 1991 Logistic Regression and Cox D ata S ets Descriptive Results The statistic results of the descriptive analysis indicate that all the selected independent variables when examined separately have shown to have an impact on infant mortality, (tables 89 and 90) According to the logistic regression data set (3212 cases), the birth cohort variable clearly indicates a decline in the odds of death as we move hom children bom between 1975-79, 1980-84 and 1985-90. 8.5% of the children bom between 1975 and 1979 (10.6% of sub-sample) died; 7.3% between 1980 and 1984 ( 23.7% of sub-sample) and 5.7% in the latest birth cohort period ( 60% of sub-sample). The value for the oldest birth cohort (1974 and before) as expected, is very high, or 19% (the sample size is small, 180 cases or 5.6% of the sub-sample). The birth cohort variable takes on values from 0 to 3, the former being the oldest cohort, and the latter the newest. According to the Cox data set (3811 cases), the means for the birth cohort categories are exactly the same as the logistic regression. The most recent cohort (1985-91) represents 66.3 % of the sub-sample, representing 2527 cases or 599 more than the logistic regression one. The place of residence variable for the logistic regression data, set indicates that 6.6% of the children bom in urban areas (68.6% of sub-sample) 301 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. and 8.2% of the ones bom in rural areas died. This variable takes the value 0 for rural and I for urban residence. According to the Cox data set, the percentage of deceased children from urban areas is very similar, or 6.5%, whereas the rural rate is lower or 7.8%. As far as the drinking water, the impact seems to be substantial. 6.1% of the children bom in households with clean drinking water (71.1%) died, whereas 9.4% of those with no access to good drmking water lost their lives. The values are 0 for not good for drinking and 1 for water good for drinking. In the Cox data set the percentage of children with access to good drinking water who died is almost the same, or 6.2% but the odds of death for those who did not have access to good drinking water is reduced from 9.4 to 8.6 %. According to the descriptive results, the availability of modem sewage disposal seems to have a comparable impact as drinking water. This variable takes the value 0 for not modem and 1 for modem sewage. 6.1% of the children in households with no modem sewage and 8.9% in the ones with modem sewage died in the logistic regression formatted data set. In the Cox data set, these rates are ju st as high, 6.0 and 8.3%, respectively. Note the stark difference in the odds of death between the 1996 and the 1991 data sets, as far as sewage is concerned. This fact could be related to the oddly high level of mothers who have access to modem sewage in the 1991 302 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. DHS: 63.3 and 61% respectively in the logistic and Cox data sets ( as opposed to 23.8 and 23.1% in the 1996 data sets). According to the logistic regression data set descriptive results, 9.1% of the children of high risk mothers die as opposed to 4.8% of the children of mothers at their reproductive peak (low risk). The Cox regression data set indicates ^cactly the same rates, even though the percentage of children whose mothers are considered to have a high age risk is lower than in die logistic data set (50% against 54% in the logistic set). This demographic variable takes the value 0 for high risk age (outside of the 10-34 good age group) and 1 for low risk. The means distribution for each total household number clearly shows a slow but steady progression as far as the mean value of deceased children per household number. In 2 to 4 individuals households, on average, 6 to 7% of the latest children of each répondent die before their first year of life. As we increase the number of residents from 5 to 9, the mean values for the odds of death grow firom 7.6 to 9.2%, repectively. The proportional hazards regression results derived from the Cox data set indicates a very similar statistic progression. The alternative dichotomous crowding variable shows means of 6.2% (non crowded) and 8 % (crowded) in the logistic data set, and 5.9% and 8 % in the Cox formatted data set, repectively. 303 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. It is interesting to notice that the percentage levels of deceased children for aU household crowding numbers is much higher than in the 1996 data sets. Overall the impact of this variable seem to be higher in 1991 than in 1996. The next variable is educational attainment, which is divided into 4 categories: higher education (college level and over); secondary (high school level completed); primary (including primary complete and secondary incomplete, 1 to 11 years); no education. According to the data sets for LR and Cox analysis, only 3.9% and 3.6%, respectively of the Northeast respondents had higher education. Among those mothers with h i^ e r education, .8% and .7% of their youngest infants died according to the logistic and Cox regression results, respectively. The other sub-categories indicate a strong and progressive effect: the odds of survival decrease hrom 9.8, to 7.2 to 3.6% as we move from no education to primary, to secondary education in the logistic data. In the proportional hazard 1996 model the means are very similar: 9.7, 6.9 and 3.9%. What is interesting to observe is the remarkable improvement in the percentage of women with secondary education between 1991 and 1996. In 1991 only 13.1 and 12.8% of the mothers in the logistic and Cox data sets, respectively, had completed their secondary education, hi 1996, this share would increase to 41.1 and 41.2%. 304 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. According to the descriptive statistic analysis, the variable goods seems to have a very significant impact on infant mortality. 11.9 % of the households in the logistic data set and 11% in the Cox data set had all 3 goods. Similarly, 9.1 % and 8.5 % of the mothers respectively in the logistic and Cox data sets had none of the three goods. The mean values of deceased diildren who had 0 and 1 good is 9.1 and 8.7% in the logistic data set, and 8.5 and 8.1% in the Cox one. With 2 goods, the percentage of children who lose their lives as infants is reduced to 5.8 and 5.9%, respectively for the logistic and the Cox formatted data files. If the mother has access to all 3 goods the rate of deceased infants does not decrease very much according to the 1991 data sets and it is reduced to 4.7 and 4.5% in the logistic and Cox data sets, respectively. The next variable is the infant’ s birth order risk. 5.4. and 8.1% of the children who, respectively, have a low (2“* or 3 "* child) and high order risk ( 1 » * ^ and over 3 " * ) die in the logistic regression data set. The values for the Cox data set are very close: 5.2. and 8 %. The descriptive results for the gender variable indicate that 6% of the girls and 8.1% of the boys died in the logistic data set. In the Cox data set, these rates are 6.2% and 7.6%, respectively. On average, 52% of the children in the data sets are males. The male excess mortality seems to be very high according to these data sets. 305 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The prenatal care variable shows that 3.6 and 3.4% of the children of mothers who received prenatal care by a doctor, respectively in the Cox and in the logistic data sets, died. For the ones who did not receive medical care, such rates increase to 8.7 and 8.6% respectively. 69.2 % of the respondents in the logistic data set and 63.7% in the Cox data set did not see a physician for the prenatal care of the infant. The odds of death for infants of white heritage is 5.5% in the logistic as weU as in the Cox data sets. In non-white households the respective rates are 7.6 and 7.3%. The proportion of mothers of white descent is only 21.5% in the 1991 data sets (in 1996, this proportion is almost 30%). The last independent variable is immunization or whether the infant was vaccinated for all three DPT shots. According to the descriptive results, there is a strong correlation between immunization status and infant mortality. Around 67% of the children did not receive aU three shots of the DPT vaccine in both data sets. The mean of deceased children among the ones not immunized was 8.4 and 10%, respectively in the logistic and in the Cox data sets. Among the immunized children (33.8 and 32.5% of the total sample, respectively), the mean values of children deceased is 0 and 0.6 for the logistic and Cox data sets. 306 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. According to the descriptive results, the duration of breastfeeding in months has a clear impact on the child’ s surviving status. 54.8% of the children were not breast-fed a t all. 10% of these children died m the first year of life. The mean values for deceased children fall progressively, but the main difference seems to be between the ones who were not breast-fed at all and the ones who were breast-fed for I month. Self-reporting errors in the number of breastfeeding months could be the cause of variations in the mean values of deceased children. Next we will examine the correlation levels with the dependent variable according to the 1991 Cox and logistic data sets, (table 7.21.) In both data sets, the variables p la ce o f re sid e n ce and e th n ic ity are not correlated with the dependent variable. The variable s e x is not correlated with the dependent variable in the Cox formatted data set and it is only significant at the 0.05 level in the logistic data set. Both correlation coefficients for the household crowding variables in the logistic set are only significant at the 0.05 level (2-tailed). The variable immunization has the strongest correlation coefficient among all variables: - 0.182 and -0.174, respectively for the logistic and the Cox data sets. Correlation levels for breastfeeding duration are very strong: -0.114. Prenatal care indicates correlation coefficients very significant as well: -0.0094 and -0.097. Eklucation levels and mother’ s age risk at the time of the birth coefficients are also relatively strong in magnitude. 307 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 89 308 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 309 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 310 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 90 311 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ^ ‘ ' ' , ; p » •/' i , j ., y P / - ■ s i ; / P y -V ^ \ ■ . ; ; j P; • > ' ;/ , ^ ■ , r j .p , jpp E m m # # 312 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. .i^u! : u ) / / / 11 ; " n ' ' ' \ ' 3 i i - . j - " " 313 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. \ h \ ) r s ., / 314 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 91 1991 Correlation Levels with Dependent Variable Logistic Regression D ata S et Cox Regression Data S et Variables Cohort* -0-103 -0.096 Uiban X Rural -0-03 X -0.024 X Drinking W ater -0.058 -0.044 Sewage -0.053 -0.045 Household Crowding* 0.038# 0.043 Household Crowding (B) -0.035# -0.04# Mother's Age Risk -0.082 -0.084 Educational Level* -0.085 -0.081 Goods* -0.061 -0.052 Birth O rder Risk -0.05 -0.053 Sex of Child -0.04# -0.028 X Breastfeeding** N/A -0.114 Ethnicity* 0.025 X 0.021 X Ethnicity (B) -0.034 X -0.029 X Dr's Prenatal Care -0.094 -0.097 Immunization (DPT123) -0.182 -0.174 * non dichotomous * * in months AHvarôbiesSk|nificantatlhe0.01 level (2 tiled) unless ihdKatoclathenww » SignMlcant at the 0X6 level ( 2-toled) XNatCcrraMed 315 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Diagnostic Results for 1991 LR Models The diagnostics statistics reveal that all models fît the facts well, (table 92) The -2 Log Likelihood statistic is the analogue of the Sum of the Square Errors (SSE) in OLS regression and eatresses how poorly the models fît with all the included variables. In the 1991 logistic data set, the data indicate that the best prediction power for the dependent variable is found in model 3 wiür the multiple birth cohort variables, the one with the highest number of degrees of freedom and the lowest -2LL: 1415.717. The models chisquare indicate that the null hypothesis that the value of the coefficients of the variables is equal to zero should be rejected. A ll models are very significant. The highest model chisquare is found in model 3: 211.28. A U . models have excellent goodness of fît. 3190 cases were selected and 22 missing. The prediction efficiency of the models is 92.95%. 316 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 92 Diagnostic Results, Predictive Efficiency and Further Characteristics of the LR Models for 1991 Data S et Item StaL Model 1 Model 2 Model 3 Model 4 -2Log Likelihood 1428-724 1428.999 1415.717 1415.797 G oodness of Fit 3316.326 3321.108 3198.819 3201.751 Model Chisquare 198.274 197.998 211.28 211.2 (Signif) 0 0 0 0 (df) 13 12 15 14 Predictive Efficiency 92.95% 92.95% 92.95% 92.95% N ((num tier of cases included ) 3190 3190 3190 3190 317 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Logistic Regression Results for 1991 Models The 2 main logistic regression models for 1991 (model 1 and 3) consist of 12 independent variables in addition to the birth cohort variable or variables. The other 2 regression models (model 2 and 4) mclude the very same regressions without the important sewage variable, (tables 93 to 96) The regression results for the 4 models indicate that the independent variables that have a statistically significant effect on infant mortality are: household crowding, prenatal care by a doctor, birth cohort and DPT immunization. Goods is statistically significant in all models but model 4. Sex of child does have a significant impact on infant mortality in the first 2 models. Sewage is not significant at all, but this is no surprise given that the levels of modem sewage in 1991 are excessively high. In this particular data set, the models with and without the type of toilet facility variable are very similar as far as the levels of significance of the explanatory variables. DPT Immunization is a variable which has a significant impact on infant mortality in the 1991 but not in the 1996 data set. This result is in agreement with Simoes’ examination of the number of years lost according to death causes in the Northeast of Brazil. According to this author, between the 1980’ s and the 1990’ s the contribution of immunization to the decline in infant mortality of males in the Northeast of Brazil fell firom 0.05 to 0.03 year. ^ 282 Celso Simoes, Ibid., 120. 318 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. What is most surprising is the fact that mother’ s educational attainm en t is not a statistically significant variable according to the 91 logistic data set. In model I (complete model with sm ^e birth cohort), the variables with the lowest p values are, respectwely: I) DPT (0.); 2) Dr’ s prenatal care (0.0014); 3) birth cohort (0.0046); 4) household crowding (0.0266); 5) goods (0.0392); 6) sex of the child (0.0442). (table 93) In model 3 (complete model with multiple birth cohorts), the variables with the lowest p values are, respectively: 1 ) DPT (0.); 2) birth cohorts (0.0006; 0 and 0.0007, respectively for 85-90, 80-84 and 75-79); 2) Dr’ s prenatal care (0.0215); 4) household crowding (0.0272). (table 95) In model 2 (smgle birth cohort with no sewage variable), the variables with the lowest p values are, respectively: 1) DPT (0.); 2) Dr’ s prenatal care (0.0015); 3) birth cohort (0.0051); 4) goods (0.0168); 5) household crowding (0.0283); sex of the child (0.0432). (table 94) In model 4 (complete model with multiple birth cohorts), the variables with the lowest p values are, respectively: 1 ) DPT (0.); 2) birth cohorts (0.0006; 0 and 0.0007, respectively for 85-90, 80-84 and 75-79); 2) Dr’ s prenatal care (0.0223); 4) household crowding (0.0279); 5) goods (0.0372). (table 96) An analysis of the B coefScients’ signs in all 4 models would indicate that household crowding and prenatal care are positively correlated with in fan t mortality, meaning that access to prenatal care by a physician and crowding living conditions would increase the likelihood of death for the infant. 319 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The odds ratios for the two variables (greater than 1) suggest the same positive relationship between the explanatory variables and the dependent variable. The results for the household crowding variable do fit the facts well. However, the results for prenatal care may reflect a coUinearity effect with the DPT immunization variable. To investigite such a possibility, the model I logistic regression was run without the immunization variable. As a result, the significance levels for the prenatal care variable changes firom 0.0014 to 0.0163. The standard error decreases fi* o m 0.4419 to 0.2175. The unstandardized B coefficient switches from 0.7589 in model 1 (table 93) to -.5227. The odds ratio is also not greater than I (2.136 ) but less than 1 ( .5929). Thus, access to prenatal care by a doctor does improve the odds of survival. L n a model with all variables included but immunization, the odds of survival for the infant would increase in 60% as the infant receives proper medical prenatal care. The direct correlation between prenatal care and the dependent variable shown in table 93 was indeed caused by the mclusion of the DPT immunization variable and do not reflect the facts. An analysis of the independent variables r^ression coefficients significant at the 0.05 level in model 1 indicates that when the infant’ s birth cohort changes from a relatively older to a more recent cohort, and given the impact of other explanatory variables, the odds of dying decrease in 77% vis-a- 320 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. vis to what th^r would be had the child been bom in the older birth cohort. (table 93) In a like mannar, the odds of dying for the infant child decrease in 69% when the infant is a female as opposed to what it would be otherwise, (table 93). As far as the household crowding is concerned, when an additional member is added to the household, the odds of dying for the infant increase 7% in comparison to what they would be without the additional dweller, (table 93) In short, the 1991 logistic regression models indicate that the following variables have an effect on the likelihood of infant mortality at the 5% level of signi&cance: DPT immunization, prenatal care by a doctor, household crowding and birth cohort. The variable goods is significant in all but one of the logistic regression models. The regression results for the 1991 logistic method do not conform completely with this study’ s main hypothesis. Particularly surprising is the lack of significance of the education variable. It may have to do with the fact that gains in female schooling increased significantly only in recent years but that is not clear. The possibility of coUinearity with other variables and particularly with the goods variable was investigated but there seems to be none. The level of significance of the sewage variable may also have been affected by the abnormaUy high proportion of households with modem toilet facilities in 1991. 321 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 93 Variable Name Logistic Regression Results for 1991 Ne D ata S et (1) B Standard Error Exp (B) (P) Significance Cohort -0.2559 0.0903 0.7742 0.0046 Urban X Rural 0.2318 0.2199 1.2609 0.2919 Drinking W ater -0.3761 0.2121 0.6865 0.0762 Sewage -0.1114 0.2118 0.8945 0.5988 Household Crowding 0.0678 0.0306 1.0701 0.0266 Age Group of Mother 0.0124 0.1824 1.0125 0.9457 Ethnicity -0.1842 0.1909 0.8318 0.3346 Mother's Education -0.2017 0.1224 0.8173 0.0992 Goods -0.2062 0.1 0.8137 0.0392 Birth Order Risk -0.1621 0.1698 0.8504 0.3399 Prenatal care by Doctor 0-7589 0.2371 2.136 * 0.0014 Sex of Child -0.2912 0.1447 0.7474 * 0.0442 DPT Immz- -3.2904 0.4419 0.0372 * 0 Constant -1.208 0.282 0 Source; 1991 DHS for Brazil's Norttieast region from a sut>-sample total of 3212 cases (3190 selected, 22 m issing). 322 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 9 4 Variable Name L o g ^ îc Regression Results for 1991 N e Data S e t (2) B Standard Error Exp (B) (P) Significance Cohort -0.2521 0.09 0.7771 • 0.0051 Urtian X Rural 0.1983 0.2103 1.2193 0-3457 Drinking W ater -0.4004 0.2068 0.6701 0.0529 Sewage •0.1062 0.1874 0.8992 0.5709 Household Crowding 0.0669 0.0305 1.0692 * 0.0283 Age Group of Mother 0.0119 0.1823 1.0119 0.9481 Ethnicity •0.1884 0.1907 0.8283 0.3232 Mother's Education -0.2117 0.1211 0.8092 0.0804 Goods -0.2242 0.0938 0.7991 * 0.0168 Birth Order Risk -0.1668 0.1695 0.8464 0.3253 Prenatal care by Doctor 0.7526 0.2369 2.1225 * 0.0015 Sex of Child -0.2926 0.1447 0.7463 * 0.0432 DPT Immz. -3.2874 0.442 0.0374 0 Constant -1.2018 0.282 0 Source: 1991 OHS for Brazil's N orttieast region from a sub-sam ple total of 381 le a se s (3795 selected, 16 dropped). 323 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 95 Variable Nam e Logistic Regression R esults for 1991 Ne Data S et (3) B Standard Error Exp (B) (P) Significance Cohort 85-90 -0.9098 0.2637 0.4026 0.0006 Cohort 80-84 -1.1664 0.2474 0.3115 0 Cohort 75-79 -0.9361 0.2769 0.3922 0.0007 Urtian X Rural 0.243 0.2207 1.2751 0.2709 Drinking W ater -0.3578 0.2141 0.6992 0.0947 Sew age -0.0607 0.2144 0.9411 0.7771 Household Crowding 0.0675 0.0306 1.0699 * 0.0272 Age Group o f Mother -0.0594 0.1844 0.9423 0.7472 Ethnicity -0.1798 0.1914 0.8354 0.3476 M other's Education -0.2019 0.1234 0.8172 0.1017 Goods -0.1869 0.1003 0.8296 0.0625 Birth O rder Risk -0.1812 0.1701 0.8342 0.2867 Prenatal care by Doctor 0.5534 0.2407 1.7391 * 0.0215 S ex of Child -0.2827 0.1453 0.7538 0.0518 DPT Immz. -3.3382 0.4399 0.0355 * 0 Constant -0.8722 0.2915 — 0.0028 Source: 1991 DHS for Brazil's N ortheast region from a sutxsam ple total o f 3212 cases (3190 selected, 22 m ksing). 324 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 96 Logistic Regression Results for 1991 Ne D ata S e t (4) Variable Name B Standard Error Exp(B) Significance (p) Cohort 85-90 -0.9051 0.2631 0.4045 * 0.0006 Cohort 80-84 -1-1659 0.2473 0.3116 * 0 Cohort 75-79 -0.9385 0.2768 0.3912 * 0.0007 Uriaan X Rural 0.2252 0.2115 1.2526 0.2869 Drinking W ater -0.3713 0.2086 0.6899 0.0752 Household Crowding 0.0671 0.0305 1.0694 • 0.0279 Age Group of Mother -0.0601 0.1843 0.9417 0.7445 Ethnicity -0.1816 0.1913 0.834 0.3426 Mother's Education -0.2076 0-1219 0.8125 0.0885 Goods -0.1923 0.0943 0.8216 0.0372 Birth Order Risk -0.1839 0.1698 0.832 0.279 Prenatal care by Doctor 0.549 0.2403 1.7314 * 0.0223 Sex of Child -0.2832 0.1453 0.7534 0.0513 DPT Immz. -3.3366 0.4399 0.0356 * 0 Constant -0.8669 0.2909 - 0.0029 Source: 1991 DHS for Brazil's N ortheast region from a sub-sam ple total of 3212 cases (3190 selected, 22 m issing). 325 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Diagnostic Results for 1991 Cox Models The diagnostics statistics reveal that all 8 Cox regression, models fît the facts well, (table 97-98) hi the Cox proportional risks method of partial likelihood, the data are censored for the occurrence of events (the qualitative transition change). The quantitative death duration variable (or the age of death in months) and the categorical dichotomous death status variable are censored to exclude the cases in the data set (sub-sample size of 3811 cases) in which the infant died or to exclude the occurrence of events. In the 1991 8 Cox models, 16 cases were dropped and 3795 selected. 93.8% of these children are censored and survive the fîrst year of life whereas 237 do make the transition. The data indicate that the complete models including sewage and duration of breastfeeding are clearly the best ones in their respective birth cohort structures (models 1 & 5 ). The -2LL of these two models are: 3655.295 and 3636.405. Overall, the best prediction power for the dependent variable is found in model 5 with the multiple birth cohort variables, the one with the highest number of degrees of fîreedom and the lowest -2LL. The models chisquare are very significant, indicating that the null hypothesis that the value of the coefScients of the variables is equal to zero should be rejected. The independent variables do improve the predictability of the models. The Cox regression model with the h ip e st chisquare is model 5: 185.108. 326 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 97 Diagnostic Results, Predictive Efficiency and Further C haracteristics of the Cox Regression Models for 1991 Data S et (1 o f 2) Item Stat. Model 1 Model 2 Model 3 Model 4 Events 237 237 237 237 Censored 3558 3558 3558 3558 (%) 93.8 93.8 93.8 93.8 -2Log Likelihood 3655.295 3655.756 3694.464 3694.536 Chisquare (Overall ) 162.058 161.535 147.383 147.205 (Signif) 0 0 0 0 (df) 14 13 13 12 N (numt>er of cases included ) 3795 3795 3795 3795 327 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 9 8 Diagnostic Results, Predictive Effiaency and Further Characteristics o f the C ox Regression Models for 1991 D ataS et (2 of 2) M odels M odels Model 7 M odels Item S ta l Events 237 237 237 237 Censored 3558 3558 3558 3558 (%) 93.8 93.8 93.8 93.8 -2Log Likelihood 3636.405 3636.613 3682.283 3682.283 Chisquare (Overall ) 185.108 184.916 167.705 167.689 (Signif) 0 0 0 0 (df) 16 15 15 14 N (numt)er of cases included ) 3795 3795 3795 3795 328 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Cox Regression. Results for 1991 Models Cox’ s proportional hazards method allows for the inclusion of the vital time-dependent breastfeeding variable. The results for the Cox regression method are considered to be even more powerful than the logistic regression results. The 2 main Cox regression models for 1991 (model 1 and 5) are comprised of 13 independent variables plus the birth cohort variable or variables. 2 regression models (model 2 and 6) include the very same variables with the exception of sewage and 2 others ( models 3 and 7) consist of the same models without the duration of breastfeeding variable. The last 2 models (models 4 and 8) exclude both sewage and breastfeeding, (tables 99 to 106). The regression results for the 8 models indicate that the independent variables that have a statistically significant effect on infant mortality are: breastfeeding, DPT immunization, household crowding, mother’ s education and goods (except m model 7). The birth cohort variable have an effect on the probability of dying at the 0.05 level in all 4 models with multiple birth . The prenatal care by a doctor is also statistically significant models 1 and 2. Type of toilet facility is not significant at all, but this is no surprise given that the levels of modem sewage in 1991 are excessively high. 329 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. DPT Immunization is a variable which has a veiy significant impact on infant mortality in the 1991 but not in the 1996 data set. In model 1 (complete model with smgle birth cohort), the variables with the lowest p values are, respectively: 1) breastfeeding (0.0); 2) DPT immunization (0.0); 3) Dr’ s prenatal care (0.0088); 4) household crowding (0.0103); 5) goods (0.0164); 6) mother’ s education (0.0429). (table 99) In model 5 (complete model with multiple birth cohorts), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0); 2) DPT immunization (0.0); 3) birth cohorts (0.0001 and 0.006, respectively for the 80/84 and the 75/79 cohorts); 4) household crowding (0.0106); 5) goods (0.0322); 6) mother’ s education (0.039). (table 103) In model 2 (single birth cohort with no sewage variable), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0); 2) DPT immunization (0.0); 3) Dr’ s prenatal care (0.0088); 4) household crowding (0.0103); 5) goods (0.0164); 6) mother’ s education (0.0429). (table 100) In model 6 (multiple birth cohorts with no sewage variable), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0); 2) DPT immunization (0.0); 3) birth cohorts (0.0001 and 0.0057, respectwely for the 80/84 and the 75/79 cohorts); 4) household crowding (0.0111); 5) goods (0.0135); 6) mother’ s education (0.0306). (table 104) 330 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In model 3 (single burth. cohort with no breastfeedmg), the variables with the lowest p values are, respectnrely; I) DPT immunization (0.0); 2) household crowding (0.0202); 3) goods (0.0372); mother's education (0.0481). (table 101) In model 7 (multiple birth cohorts with no breastfeeding), the variables with the lowest p values are, respectively: 1 ) DPT immunization (0.0); 2) birth cohorts (0.021, 0.0001 and 0.0056, respectively for the 85/91, 80/84 and the 75/79 cohorts); 3) household crowding (0.0225); 4) mother’ s education (0.0481). (table 105) In model 4 (single birth cohort without sewage and breastfeeding), the variables with the lowest p values are, respectively: 1) DPT immunization (0.0); 2) goods (0.0197); 3) household crowding (0.0204); mother’ s education (0.0418). (table 102) hi model 8 (multiple birth cohorts without sewage and breastfeeding), the variables with the lowest p values are, respectively: 1 ) DPT immunization (0.0); 2) birth cohorts (0.0208, 0.0001 and 0.0056, respectively for the 85/91, 80/84 and the 75/79 cohorts); 3) household crowding (0.0224); 4) mother’ s education (0.0453); 5) goods (0.0494). (table 106) An analysis of the B coefScients’ signs in all 8 Cox regression models would indicate that household crowding and prenatal care are positively correlated with infant mortality meaning that access to prenatal care by a physician and crowding living conditions would increase the likelihood of death 331 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. for the infant. The odds ratios for the three variables (greater than I) would indicate the same positive relationship between the explanatory variables and the dependent variable. The results for the household crowdmg variable do fît the facts well. Again ju st as in the 1996 data set, the regression coefficients results for prenatal care would have been affected by some degree of coUinearity with the immunization variable. Nevertheless, the data show that prenatal care only has an effect on the probability of dying at the 0.05 signifîcance level in the fîrst 2 models. The lack of statistical signifîcance for the prenatal care by a doctor variable in the other 6 models was caused by this coUinearity effect between immunization and prenatal care. A scrutiny of the Cox regression coefficients for the independent variables which are signifîcant at the 0.05 level in model 1 indicates t±iat, when the infant is breast-fed by the mother for an additional month, and given the impact of other explanatory variables, the odds of dying decrease in 83% in comparison to what they would be if there were no modem sewage, (table 99) Similarly, the odds of dymg for the infant child decrease in 79% as a mother improves her educational attainment from one category to another, (table 99) 332 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. As the household acquires one additional consumer good, a radio, a TV or a car, the odds of dymg for the family's child is also reduced in 81% in comparison to what they would be if that consumer good was not available, (table 99) When an additional member is added to the household, the odds of dying for the infant increase 7% in comparison to what they would be without the additional dweller, (table 99) When the infant recenses aU three shots of the DPT vaccine, the odds of dying are reduced in 4% vis-a-vis what they would be otherwise. Education is not signifîcant in the logistic regression fîrst model but aU these unstandardized Cox regression coefScients (including education) for model 1 do match the logistic regression ones. In sum, according to the 1996 Cox regression models the following variables have an effect on the likelihood of infant mortality at the 5% level of signifîcance; breastfeeding duration, DPT immunization, household crowding mother’ s education, goods as well as prenatal care by a doctor. The regression results for the Cox proportional hazards method conform well with the logistic results except for the mother’ s education variable. 333 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 99 Cox Regression Results for 1991 Ne Data S et (1) Variable Name 8 Standard Error Exp(B) Significance (p) Cohort -0.0548 0.0878 0.9467 0.5327 Urban X Rural 0-187 0.2034 1.2056 0.3581 Drinking W ater -0.2986 0.1978 0.7419 0.1311 Sewage -0.1293 0.1902 0.8787 0.4964 Household Crowding 0.0656 0.0256 1.0678 * 0.0103 Age Group of Mother 0.0091 0.1611 1.0091 0.9551 Ethnicity -0.1122 0.1727 0.8939 0.5161 Mother's Education -0.2324 0.1148 0.7927 * - 0.0429 Goods -0.2165 0.0902 0.8054 * 0.0164 Birth Order Risk -0.0907 0.1543 0.9133 0.5568 Prenatal care by Doctor 0.5112 0.1952 1.6673 * 0.0088 Sex of Child -0.2353 0.1324 0.7904 0.0755 DPT Immz. -3.2091 0.5921 0.0404 * 0 Breastfeeding in Months -0.183 0.0442 0.8328 * 0 Source: 1991 DNS for Brazil's N ortheast region from a sub-sam ple total of 3811 cases (3795 selected, 16 dropped). 334 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 100 Cox Regression Results for 1991 Ne Data S et (2) Variable Name B Standard Error Exp(B ) Significance (jp) Birth Cohort -0.0515 0.0877 0.9498 0.5571 Urban X Rural 0.1477 0-1947 1.1592 0.4482 Drinking W ater -0.325 0.1935 0.7225 0.0931 Household Crowding 0.065 0.0256 1.0671 * 0.0111 Age Group of Mother 0.0094 0.161 1.0095 0.9533 Ethnicity -0.1154 0.1726 0.891 0.5039 Mother's Education -0.2441 0.1137 0.7834 0.0318 Goods -0.2383 0.0841 0.788 0.0046 Birth O rder Risk -0.0943 0.1542 0.9101 0.5411 Prenatal care by Doctor 0.5029 0.1949 1.6534 0.0099 Sex of Child -0.2387 0.1323 0.7877 0.0711 DPT Immz. -3.2083 0.5922 0.0404 0 Breastfeeding in Months -0.182 0.0441 0.8336 0 Source: 1991 DHS for Brazil's N ortheast region from a sub-sam ple total of 3811 cases (3795 selected, 16 dropped). 335 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 101 Cox R egression Results for 1991 Ne D ata S et (3) Variable Name B Standard Error ExpCB) Significance (p) Birth Cohort -0.1284 0.0864 0.8795 0.1375 Urban X Rural 0.2326 0.2029 1.2618 0.2517 Drinking W ater -0.2555 0.1965 0.7745 0.1935 Sewage -0.0515 0.1922 0.9498 0-7887 Household Crowding 0.0604 0.026 1.0622 * 0-0202 Age Group of Mother -0.0646 0.1609 0-9375 0.6883 Ethnicity -0.0906 0.1726 0.9134 0.5997 Mother's Education -0.2255 0.1142 0.7981 * - 0.0483 Goods -0.1877 0.0901 0.8288 * 0.0372 Birth Order Risk -0.088 0.1549 0.9158 0.5699 Prenatal care by Doctor 0.2307 0.1844 1.2595 0.2109 Sex of Child -0.2317 0.1322 0.7932 0.0797 DPT Immz. -3.6341 0.5886 0.0264 » 0 Source: 1991 DHS for Brazil's N ortheast region from a sub-sam ple total of 3811 cases (3795 selected, 16 dropped). 336 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 102 Cox Regression R esults for 1991 Ne Data S et (4) Variable Name B Standard Error E x p O ) Significance (p) Birth Cohort -0.1266 0.0862 0.881 0.1419 Urban X Rural 0.2171 0.1944 1.2424 0.2642 Drinking W ater -02664 0.1922 0.7661 0.1657 Household Crowding 0.0603 0.026 1.0621 * 0.0204 Age Group of Mother -0.0644 0.1609 0.9376 0.6888 Ethnicity -0.0918 0.1725 0.9123 0.5948 Mother's Education -0.23 0.113 0.7945 • 0.0418 Goods -0.1963 0.0842 0.8218 * 0.0197 Birth Order Risk -0.0897 0.1547 0.9142 0.5621 Prenatal care by Doctor 0.2276 0.1841 1.2556 0.2163 Sex of Child -0.2327 0.1322 0.7924 0.0783 DPT Immz. -3.6324 0.5886 0.0265 * 0 Source: 1991 DHS for Brazil’ s Northeast region from a sub-sample total o f 3811 cases (3795 selected, 16 dropped). 337 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 103 C ox Regression Results for 1991 Ne Data S et (5) Variable Name B Standard Error Exp(B) Significance (p) Cohort 85-91 -0.3806 0.2441 0.6835 0.119 Cohort 80-84 -0.9541 0.2426 0.3852 0.0001 Cohort 75-79 -0-7613 0.2772 0.467 * 0.006 Uiban X Rural 0.1874 0.2034 1.2061 0.357 Drinking W ater -0.2705 0.1997 0.763 0.1756 Sewage -0.0875 0.1918 0.9162 0.6481 Household Crowding 0.0652 0.0255 1.0674 » 0.0106 Age Group of Mother -0.0587 0.1622 0.943 0.7173 Ethnicity -0.1145 0.1725 0.8918 0.5067 M other's Education -0.2381 0.1153 0.7881 * 0.039 Goods -0.1932 0.0902 0.8243 * - 0.0322 Birth Order Risk -0.1123 0.1534 0.8938 0.464 Prenatal care by Doctor 0.3204 0.1952 1.3777 0.1007 Sex of Child -0.2215 0.1323 0.8013 0.0941 DPT Immz. -3.2416 0.5912 0.0391 * 0 Breastfeeding in Months -0.201 0.0455 0.8179 * 0 Source; 1991 DHS for Brazil's N ortheast region from a sub-sam ple total of 3811 cases (3795 selected. 16 dropped). 338 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 104 Cox Regression R esults for 1991 Ne Data S et (6) Variable Name B Standard Error Exp(B) Significance (p) Cohort 85-91 -0.3772 0.2439 0.6858 0.122 Cohort 80-84 -0.9548 0.2426 0.3849 *■ 0.0001 Cohort 75-79 -0-7664 0.2769 0.4647 # 0.0057 Urban X Rural 0.1619 0.1954 1.1757 0.4073 Drinking W ater -0.2891 0.1952 0.7489 0.1385 Household Crowding 0.0648 0.0255 1.0669 * 0.0111 Age Group of Mother -0.0587 0.1621 0.943 0.7172 Ethnicity -0.1156 0.1725 0.8908 0.5027 M other's Education -0.2465 0.114 0.7815 * 0.0306 Goods -0.2079 0.0842 0.8123 « 0.0135 Birth Order Risk -0.1147 0.1533 0.8916 0.4543 Prenatal care by Doctor 0.3138 0.1947 1.3686 0.1072 S ex of Child -0.2235 0.1322 0.7997 0.091 DPT Immz. -3.2409 0.5913 0.0391 * 0 Breastfeeding in Months -0.2002 0.0455 0.8185 • 0 Source: 1991 DHS for Brazil's Northeast region from a sub-sample total of 3811 cases (3795 selected, 16 dropped). 339 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 105 Cox Regression Results for 1991 Ne D ata S et (7) Variable Name 8 Standard Error Exp(B) Significance (p) Cohort 85-91 -0.5686 0.2463 0.5663 0.021 Cohort 80-84 -0.9211 0.2422 0.3981 0.0001 Cohort 75-79 -0.7673 0.2771 0.4643 0.0056 Urban X Rural 0.2383 0.2026 1269 0 2 3 9 7 Drinking Water -0.2355 0.1978 0.7902 02338 Sewage -9.1 IE-04 0.1938 0.9991 0.9962 Household Crowding 0.0592 0.026 1.061 0.0225 Age Group of Mother -0.1292 0.162 0.8788 0.4254 Ethnicity -0.0872 0.1724 0.9165 0.6131 Mother's Education -0.2266 0.1146 0.7973 0.0481 Goods -0.1657 0.0901 0.8473 0.0659 Birth Order Risk -0.1042 0.1543 0.901 0.4992 Prenatal care by Doctor 0.0672 0.1873 1.0695 0.7197 Sex of Child -0.221 0.1322 0.8018 0.0947 DPT Immz. -3.6806 0.5881 0.0252 0 Source; 1991 DHS for Brazil's Northeast region from a sub-sam ple total of 3811 cases (3795 selected. 16 dropped). 340 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 106 Variable Name Cox Regression Results for 1991 Ne D ata S e t (8) B Standard Error Exp (B) Significance (p) Cohort 85-91 -0.5685 0.246 0.5664 * 0.0208 Cohort 80-84 -0.9211 0.2422 0.3981 * 0.0001 Cohort 75-79 -0.7674 0.2769 0.4642 * 0.0056 Urban X Rural 0.238 0.1948 1.2687 0.2218 Drinking W ater -0.2357 0.1933 0.79 0.2227 Household Crowding 0.0592 0.026 1.061 * 0.0224 Age Group of Mother -0.1292 0.162 0.8788 0.4254 Ethnicity -0.0872 0.1724 0.9165 0.613 Mother's Education -0.2266 0.1132 0.7972 * ■ 0.0453 Goods -0.1659 0.0844 0.8472 * 0.0494 Birth Order Risk -0.1043 0.1541 0.901 0.4986 Prenatal care by Doctor 0.0672 0.1867 1.0695 0.7191 Sex of Child -0.221 0.1322 0.8017 0.0946 DPT Immz. -3.6805 0.5881 0.0252 * 0 Source: 1991 DHS for Brazil's Norttieast region from a sutxsam ple total o f 3811 cases (3795 selected, 16 dropped). 341 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. R esults o f D escriptive and M ultivariate Régression A nalysis Applied to 1986 Logistic R egression and Cox Data Sets Descriptive Analysis The statistic results of the descriptive analysis indicate that all the selected independent variables when examined separately have shown to have an impact on infant mortality. However, the sample sizes for the 1986 data sets are much smaller than the 1996 and the 1991’ s data sets; 826 and 1120, respectively for the logistic and the Cox formatted data sets, (tables 107, 108) In both 1986 data sets the birth cohort variable does not indicate great changes in the odds of death between the two most recent cohorts. Around 10% of the children bom in the 1975-79 and 1980-1985 (1980-86 for the Cox data set) did not survive. These two cohort categories represent 93.4% of the cases in the logistic data set and 95.2% in the Cox data set. For children bom before 1975 (the other two categories), the number of cases is fairly small. It is interesting to see that the percentage of deceased children remained relatively stable between 1975 and 1986. The place of residence variable for the logistic regression data set indicates that 8.9% of the children bom in urban areas (60.9% of sub-sample) and 11% of the ones bom in rural areas died. This variable takes the value 0 for rural and 1 for urban residence. According to the Cox data set, the percentage of deceased children foom urban areas is very similar, or 8.8%, whereas the rural rate is the same. 342 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. As far as the drinking water is concerned, 8.8% of the children bom in households with dean drinking water (65.9%) died, whereas 12% of those with no access to good drinking water lost their lives. The values are 0 for not good for drinking and 1 for water good for drinking. In the Cox data set the percentage of children with no access to good drinking water who died is the same, but the odds of death for those who did have access to good drinking water is slightly lower, or 8.6%. The proportion of households with modem sewage according to the 1986 data sets seem to be more accurate than proportion indicated by the 1991 results: 12.2 and 11.6%, respectively for the logistic and Cox data sets. The mean values of deceased children who did not have modem sewage in their households is 11% in both data sets. When modem sewage is available, the odds are 4% in the logistic and 4.6% in the Cox data sets. According to the logistic regresaon data set descriptive results, 12% of the children of high risk mothers die as opposed to 8% of the children of mothers at their reproductive peak (low risk). The rates in the Cox data set are the same. The means distribution for each total household number also shows a slow but steady progression as far as the mean value of deceased children per household number is concemed. in 2 to 3 individuals households, on average, 6 % of the latest children of each respondent die before their first year of life. 343 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. As we increase the number of residents firom 4 to 5, the mean values (or the odds of death grow from 9.3 to 13%. The proportional hazards regression results derived frt> m the Cox data set indicates a very sfmilar statistic progression. The alternative dichotomous crowding variable indicates means of 7.4% (non crowded) and 12 % (crowded) in the logistic data set, and 8% and 12 % in the Cox formatted data set, respectively. The next variable is educational attainment. According to the 1986 data sets for LR and Cox regressions, only 1.6% and 2% (or 13 and 20 women, respectively) of the Northeastern respondents had higher education. All children of this high educational group survived. The other sub-categories indicate a strong and progressive effect: the odds of survival decrease from 14, to 10 to 1.1% as we move from no education to primary, to secondary education in the logistic data. Again, in 1986 the most conspicuous changes in the odds of dying occur as women finish their secondary studies. In the proportional hazard 1996 model the means are almost the same : 14, 10 and 1.7%. According to the descriptive statistic analysis, the variable goods seems to have a very significant impact on infant mortality. The descriptive results for the means suggest that the contribution to mortality decline is boosted as the household acquires a 2nd good, typically a IV. Ownership of a car, typically the 3 "^ good entails another major cut in the odds of dymg. 344 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 9.4 % of the households in the logistic data set and 8.4% in the Cox data set had all 3 goods. Similarly, 19 and 12 % of the mothers respectwely in the logistic and Cox data sets had none of the three goods. The mean values of deceased children who had 0 and 1 good is 13 and 12% in the logistic data set, and 12 and 11% m the Cox one. With 2 goods, the percentage of children who lose their Iwes as infants is reduced to 7.2 and 8.3%, respectively for the logistic and the Cox formatted data files. If the mother has access to all 3 goods the rate of deceased infants is reduced again to 3.8 and 3.2% in the logistic and Cox data sets, respectively. The next variable is the infant's birth order risk. 6.9 % of the low (2 » * or 3 "* child) and 11% of the high ( 1 » * and over 3 "* ) birth order risk children in the sample died. The values for the Cox data set are almost identical: 6.8 and 11%. Most of the children are considered to be exposed to a high birth order risk factor. 68.4 and 67.4% in the logistic and proportional risks data sets. The descriptive results for the gender variable indicate that 9.1% of the females and 11% of the males in both data sets did not survive. The duration of breastfeeding in months variable indicates that over 40% of all mothers did not breast-feed their infants. 17% of these 451 children did not survive. Most of the respondents who did breastfeed, did it for 2 months and the percentage of deceased children is 6%. Again due to small sample size the distribution of mean values seems to be irregular. 345 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. An analysis of correlation between the explanatory variables and the dependent indicates that according to the data the b irth cohort variable ux addition to u rb a n X ru ra l, d ririktn g m ater, b irth o rd e r r is k and s e x o f th e c h ild are not correlated with the dependent variable, (table 109) The correlation coefficient associated with household crowding is not correlated in the logistic data set and it is correlated in the Cox data set at the 0.05 level of significance. Sewage and mother*s age risk at the time of the birth are also only correlated at the same level of significance. The sample size may have affected the outcome in the 1986 data sets. The larger the number of data pans, the larger the sample size, greater are the chances a given statistic will be significant. Immunization, prenatal care as well as ethnicity variables were not included in the 1986 survey. Educational attainment and goods are the only coefficients statistically significant at the 0.01 significance level. The correlation coefficients for the education variable in the logistic and Cox data sets are, respectively: -0.108 and -. 106. The same statistics for the g o o d s variable are -.095 and .08. 346 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 107 .K-\/ r f r ' , < ! : i r ) ( U ■ l i » " ' ' : J r : : . .)/ r . f / i ,r s < - A V < . .r \ i ■' 347 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 348 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 349 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 108 350 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. < j/;>; " M'C fD' J '' f ; .j , < nf T ^ *'(/' % ‘' f ' t ' ' # # # # . ' • > - î ÿ ï f ô ÿ ? 351 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. i \ h \ l , y ' '. / r ’. ; ''i. ' . y " "U !‘ -r A' , - i ^ ; 5 . ■ • ■ • • • B # # # # # # 352 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 109 1986 Correlation Levels with D ependent Variat)le Logistic Regression Data Set_____________ C ox Regression Data Set Variables Cohoit* 0-014 X 0-013 X Urban X Rural -0.036 X -0-042 X Drinking W ater -0-047 X -0-054 X Sewage -0-073# -0-064# Household Crowding* 0-063 X 0-068# Household Crowding (B) -0-08# -0-065# Mother's Age Risk -0-069# -0.068# Educational Level* -0-108 -0.106 Goods* -0-95 -0.08 Birth Order Risk -0.066 X -0-071 # Sex of Child -0-025 X -0-031 X Breastfeeding** N/A -0-156 Ethnicity* N/A N/A Ethnicity (B) N/A N/A Dr's Prenatal Care N/A N/A Immunization (DPT123) N/A N/A ' non dichotomous ^ in months A H vansbbs SignWkant at the 0.01 levei (2 taBed) unless nicaiedolhefwise * Significant at the 0.06 ievsi ( 2-laiiad) X Not Correlated 353 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. D iagnostic R esu lts for 1986 LR M odels Tlie diagnostics statistics reveal Üiat all models fit the facts well, (table 110 ) The -2 Log Likelihood statistic is the analogue of the Sum of the Square Errors (SSE) in OLS regression and expresses how poorly the models fit with, all the included variables. The larger the -2LL, the worse is relatively the prediction power for the dependent variable. In the logistic data set, the data indicate that the model with the best prediction power for the dependent variable is model 2, with the birtii cohort variable and no sewage. Its -2LL is the lowest or 498.175. The models chisquare indicate that the null hypothesis that the value of the coefGcients of the variables is equal to zero should be rejected. All models are statistically significant but model 2 is the one that has the lowest probability that the results were caused by chance- its significance level is 0.0118. All models have excellent goodness of fit. This data set has only 826 cases, 3240 cases were selected and 29 missing. The prediction efQciency of the models is lower than in the 1996 or in the 1991 data sets; 90.22%. 354 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 110 Diagnostic Results, Predictive Efficiency and Further Characteristics of the Logistic Regression Models for 1986 Data S e t Item Stat. Model 1 Model 2 M odels Model 4 -2Log Likelihood 502.841 498.175 504.402 500.022 Goodness of Fit 789.798 767.353 788.028 766.787 Model Chisquare 21.04 25.706 19.479 23.86 (SigniO 0.0208 0.0118 0.0214 0.0133 (df) 10 9 12 11 Predictive Efficiency 90.22% 90.22% 90.22% 9022% N (number of cases included ) 818 818 818 818 355 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. L ogistic Regression R esu lts for 1986 M odels The 2 standard logistic regression models for 1986 (model 1 and 3) consist of 10 independent variables in addition to the birth cohort variable or variables. The other 2 regression models (model 2 and 4) include the very same variables except the sewage variable, (tables 111 to 114) The regression results for the 4 models indicate that the only independent variable which has a statistically significant effect on infant mortality at the 0.05 level of significance is the income pro:;y variable goods. In model 1 (standard logistic regression model with a single birth cohort variable) the variable goods is statistically significant at the 0.048 level. When an additional consumer good is present in the household, the odds of dying for the infant are reduced to 76% to what they would be otherwise, (table 111) In model 3 (standard logistic regression model with multiple birth cohort variables) the same covariate is statistically significant at the 0.0343 level of significance level. When an additional consumer good, either a radio, a TV or a car, is present in the household, the odds of dying for the infant are reduced to 74% to what they would be if the good was not available in the household, (table 113) The lack of statistical significance of most independent variables is due to the limitations of applying logistic regression to a sub-sample size of only 826 cases. As the number of cases increases so does the likelihood that the explanatory variables will be significant. 356 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 111 Logistic Regression R esults for 1986 Ne Data S et (1) Variable Name B Standard Error Exp(B) Significance (jp ) Birth Cohort 0.0581 0.2105 1.0598 0.7825 Urban X Rural 0.134 0.3087 1.1434 0.6643 Drinking W ater -0.0289 0.3082 0.9715 0.9254 Sewage -0.6373 0.5475 0.5287 0.2444 Household Crowding 0.0407 0.0498 1.0415 0.414 Age Group of Mother -0.3677 0.2718 0.6923 0.176 Mother's Education -0.407 0.2392 0.6657 0.0889 Goods -0.2757 0.1601 0.7591 0.0852 Birth Order Risk -0.1711 0.3054 0.8427 0.5753 Sex of Child -0.1637 0.2394 0.849 0.4942 Constant -1.6008 0.6671 - 0.0164 Source: 1986 OHS for Brazil's Northeast region from a sub-sample total o f 826 cases (818 selected, 8 missing). 357 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 112 Logetic Regression Results for 1986 Ne Data S et @) Variable Name B Standard Error ExpCB) Significance (p) Birth Cohort 0.0521 0.2108 1.0535 0.8048 Urban X Rural 0.1137 0.309 1.1204 0.713 Drinking W ater -0.0591 0.3088 0.9426 0.8482 Household Crowding 0.0433 0.0497 1.0443 0.383 Age Group of Mother -0.3599 02715 0.6978 0.185 Mother's Education -0.4265 0.236 0.6528 0.0707 Goods -0.3114 0.1575 0.7324 0.048 Birth Order Risk -0.1883 0.3046 0.8283 0.5365 Sex of Child -0.1692 0.2392 0.8444 0.4793 Constant -1.5568 0.6665 - 0.0195 Source: 1986 OHS for Brazil's N ortheast region from a sub-sample total of 826 cases (818 selected, 8 missing). 358 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 113 Logistic Regression R esults fo r 1986 Ne Data Set (3) Variable Name B Standard Error Exp(B) Significance (p) Cohort 80-85 5.01 8.8109 149.9007 0.5696 Cohort 75-79 5.0106 8.8133 149.99 0.5697 Cohort 70-74 5.5735 8.8213 263.3642 0.5275 Urban X Rural 0-1387 0.307 1.1488 0.6514 Drinking W ater -0.0104 0.3071 0.9896 0.9729 Sewage -0.6922 0.5496 0.5005 0.2078 Household Crowding 0.0446 0.0499 1.0456 0.371 Age Group of Mother -0.3433 0.2729 0.7094 0.2084 Mother's Education -0.3976 0.2392 0.672 0.0965 Goods -0.2959 0.1611 0.7439 0.0663 Birth Order Risk -0.1751 0.3069 0.8394 0.5683 Sex of Child -0.1608 0.2399 0.8515 0.5026 Constant -6.4859 8.8161 - 0.4619 Source: 1986 OHS for Brazil’ s Northeast region from a sub-sam ple total of 826 cases (818 selected, 8 missing). 359 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 114 Logistic Regression R esults fo r 1986 Ne Data S et (4) Variable Name B Standard Error Exp (8) Significance (jp) Cohort 80-85 4.9521 8.8134 141-4696 0.5742 Cohort 75-79 4.9633 8.8158 143.0614 0.5734 Cohort 70-74 5.4826 8.8235 240-4591 0.5344 Urban X Rural 0-1181 0.3074 1-1253 0.7009 Drinking W ater -0.0451 0.3078 0.9559 0.8836 Household Crowding 0.0465 0.0498 1.0476 0.3504 Age Group of Mother -0.3369 0.2728 0.714 0.2169 M other's Education -0-4192 0.2356 0.6576 0.0753 Goods -0-3354 0-1584 0.7151 0.0343 Birth Order Risk -0-1944 0.306 0.8234 0.5254 S ex of Child -0.1682 0.2397 0.8452 0.4829 Constant -6.3904 8.8181 - 0.4686 Source; 1986 DNS for Brazil's Northeast region from a sub-sample total o f 826 cases (818 selected, 8 missing). 3 6 0 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. D iagnostic R esults for 1986 Cox M odels The diagnostics statistics reveal that all 8 Cox regression models fît the facts well, (table 115) In the 1996 Cox models with no breastfeeding variable ( models 3, 4 , 7 and 8), 1114 cases were selected and only 6 dropped and selected. 90.8% of these children are censored and survive the first year of life whereas 103 do make the transition. In the models which include the duration of breastfeeding in months (models 1, 2 , 5 and 6), 9 cases were dropped and 1111 selected. 90.9% of the values are censored while 101 children do make the transition. In the Cox data set, the data indicate that the complete models including sewage and breastfeeding ( 1 & 5 ) are clearly better. The respective -2LL of these two models are 1313.314 and 1311.136. Overall, the best prediction power for the dependent variable is found in model 5 with the multiple birth cohort variables, the one with the lowest -2LL. The model chisquare or Gm is similar to the F test in linear regression and it is the difference between the initial log-Iikelihood (D O ) and the -2LL for the model (Dm ). The models chisquare are veiy significant, denoting that the nuU hypothesis that the value of the coefficients of the variables is equal to zero should be rejected. The independent variables do improve the predictability of the models. The Cox regression model with the highest chisquare is model 5: 62.27. 361 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 115 Diagnostic Results, Predictive Efficiency and Fudher Characteristics of the Cox R egression Models for 1986 Data S et (1 o f 2 ) Item Stat. Model 1 Model 2 Model 3 Model 4 Events 101 101 103 103 Censored 1010 1010 1011 1011 (%) 90.9 90.9 90.8 90.8 -2Log Likelihood 1313.314 1313.791 1410.542 1411.088 Chisquare (Overall ) 61.385 61.022 25.841 25.693 (Signif) 0 0 0 0 (df) 11 10 10 9 N (num ter of cases included ) 1111 1111 1114 1114 362 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 116 Diagnostic Results. Predictive Efficiency and Further Ctiaracteristics of th e C ox Regression Models for 1986 D ata S et (2 of 2) Item S ta t M odels Model 6 Model 7 Model 8 Events 101 101 103 103 Censored 1010 1010 1011 1011 (%) 90.9 90.9 90.8 90.8 -2Log Likelihood 1311.136 1311.697 1408.403 1409.016 Chisquare (Overall ) 62.27 61.846 26.521 26.332 (Signif) 0 0 0 0 (df) 13 12 12 11 N (numt)er of cases included ) 1111 1111 1114 1114 363 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Cox R egression R esults for 1986 M odels The Cox method should present results similar to the logistic method. However, the Cox procedure draws results from a larger data set (1120 cases) and it is more powerful in its explanatory power. It also allows for the inclusion of the highly important time-dependent breastfeeding variable. In this sense it is expected that a greater number of independent variables will show a statistically significant effect on the dependent variable in this data set than what was indicated in the logistic data set. The 2 main Cox regression models for 1986 (model 1 and 5) are comprised of 10 independent variables in addition to the birth cohort variable or variables. 2 regression models (model 2 and 6) include the very same variables with the exception of sewage. 2 other models ( models 3 and 7) consist of the same variables with the exception of the duration of breastfeeding variable. The last 2 models (models 4 and 8) exclude both sewage and duration of breastfeeding variables, (tables 117 to 124) Among the 6 Cox data sets (3 of which are pooled), the 1986 data set, derived from the first phase of the DHS in Brazil (1986) and constituted of a sub-sample size of 1120 children, is the one with the lowest explanatory power. The regression results for the 8 models indicate that the independent variables that have a statistically significant effect on infant mortality at the 95% level of confidence are: breastfeeding and mother’ s education. 364 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Goods and household crowding are also significant the 0.05 level but only in the main models (1 and 5) as well as in the ones which include breastfeeding (2 and 6). In model 1 (complete model with single birth cohort), the variables with the lowest p values are, respectively; I) breastfeeding (0.0); 2) household crowding (0.0155); 3) mother^ education (0.0235); 4) goods (0.0498). (table 117) In model 5 (complete model with multiple birth cohorts), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0); 2) household crowding (0.0092); 3) mother’ s education (0.0261); 4) goods (0.0479). (table 121) In model 2 (single birth cohort with no sewage variable), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0); 2) household crowding (0.0096); 3) birth cohort (0.0189); 4) mother’ s education (0.0197); 4) goods (0.0316). (table 118) In model 6 (multiple birth cohorts with no sewage variable), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0); 2) household crowding (0.0085); 3) mother’ s education (0.0215); 4) goods (0.0293). (table 122) 365 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. la model 3 (single birth cohort with no breastfeeding), the only variable that has a statistically significant impact on infant mortality at the 0.05 level is mother’ s education ( 0.0472). (table 119) Similarly, in model 7 (multiple birth cohorts with no breastfeeding), the only variable that has a statistically significant impact on infant m ortali^ at the 0.05 level is also mother’ s education ( 0.0497). (table 123) In model 4 (sin^e birth cohort without sewage and breastfeeding), once again, the only variable that has a statistically significant unpact on infant mortality at the 0.05 level is mother’ s education ( 0.0411). (table 120) In model 8, the education attainment of the mother is also the only independent variable that has a statistically significant effect on infant mortality at the 0.05 level of significance (0.0429). (table 124) The fact that the variable goods is not statistically significant at the 95% level of confidence in the 1986 Cox regression models when the duration of breastfeeding variable is not present (models 3, 4, 7 and 8) may have something to do with a degree of coUinearity between this variable and education in smaller samples. To investigate such a hypothesis the 1986 Cox regression were run without breastfeeding and also wifii without the education variable. As a result, the variable goods did become significant at the 0.05 level of significance 366 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. E X re n in small data, set such as the 1986 one, it is evident that some independent variables have a substantial impact on infant mortality. Using model 5, the best of all 8, as a reference, one can notice that when some changes occur at the household level and in the mother's behavior the odds of survival for the infant are greatly improved, (table 121) As the household acquires one additional consumer good, a radio, a TV or a car, the odds of dying for the family’ s child is also reduced in 76% in comparison to what they would be if that consumer good was not available. Similarly, and given the impact of other explanatory variables, the odds of dying for the infant child decrease in 64% as the mother improves her educational attainment from one category to another, (table 121) As the household receives an additional dweller, the odds of dying for the child are reduced in 10%. And as the infant is breast-fed for an additional month, the probability of death decreases 72% vis-a-vis what it would be otherwise. 367 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 117 Cox Regression R esults for 1986 Ne Data S e t (1) Variable Name 8 Standard Error Exp(B) Significance (p) Cohort 0.5774 0.2444 1.7815 * 0.0181 Urban X Rural 0.0947 0.258 1.0994 0.7135 Drinking W ater ■0.2366 0.2541 0.7893 0.3518 Sew age -0.2972 0.4452 0.7429 0.5044 Household Crowding 0.1002 0.0392 1.1054 * 0.0155 Age Group of Mother -0.089 0.224 0.9149 0.6912 Mother's Education -0.4533 0.2001 0.6356 * 0.0235 Goods -0.2692 0.1372 0.764 * 0.0498 Birth O rder Risk -0.154 0.2572 0.8573 0.5494 Sex of Child -0.1463 0.2016 0.8639 0.4681 Breastfeeding in Months -0.3198 0.0599 0.7263 * 0 Source: 1986 DHS for Brazil's N ortheast region from a sub-sample total of 1120cases (1111 selected. 9 dropped). 368 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 118 Cox Regression R esults for 1986 Ne Data Set (2) Variable Name B Standard Error Exp (8) Significance (p) Cohort 0.5741 0.2445 1.7756 * 0.0189 Urban X Rural 0.0836 0.2579 1.0872 0-7457 Drinking W ater -0.2502 0.2539 0-7786 0.3244 Household Crowding 0.1012 0.0391 1.1065 0.0096 Age Group of Mother -0.0871 0.2241 0.9166 0.6974 Mother's Education -0.4626 0.1983 0.6296 * 0.0197 Goods -0.2888 0.1344 0.7492 0.0316 Birth Order Risk -0.1673 0.2565 0.8573 0.5142 Sex of Child -0.1473 0.2016 0.863 0.4651 Breastfeeding in months -0.3204 0.0599 0.7258 0 Source: 1986 DHS for Brazil's Northeast region from a sub-sample total of 1120cases (1111 selected, 9 dropped). 369 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 119 Cox R egression R esults for 1986 Ne Data S e t (3) Variable Name B Standard Error Exp(B) Significance (p) Cohort 0.2102 0.2251 1.234 0.3504 Urban X Rural 0.1697 0.2569 1.1849 0.509 Drinking W ater -0.2715 0.2531 0.7623 0.2835 Sew age -0.3163 0.4444 0.7288 0.4766 Household Crowding 0.0455 0.0389 1.0465 0.2421 Age Group of Mother -0.2706 0.2177 0.7629 0.2139 M other's Education -0.3901 0.1966 0.677 0.0472 Goods -0.1903 0.1366 0.8267 0.1634 Birth O rder Risk -0.21 0.2502 0.8106 0.4012 S ex of Child -0.2655 0.1991 0.7668 0.1824 Source: 1986 DHS for Brazil's N ortheast region from a sub-sam ple total of 1120 cases (1114 selected, 6 dropped). 370 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 120 C ox Regression Results for 1986 Ne D ata S e t (4) Variable Name B Standard Error Exp(B) Significance (p) Cohort 0.2083 0.2255 1.2316 0.3555 Urban X Rural 0.1571 0.2569 1.1701 0.5407 Drinking W ater -0.2886 0.2529 0.7493 0.2538 Household Crowding 0.0452 0.0388 1.0463 0.2435 Age Group of Mother -0.2705 0.2178 0.763 0.2142 Mother's Education -0.3981 0.1949 0.6716 0.0411 Goods -0.2193 0.1339 0.8096 0.1147 Birth Order Risk -0.21 0.2497 0.8031 0.3799 S ex of Child -0.2683 0.1991 0.7647 0.1777 Source; 1986 DHS for Brazil's Northeast region from a sub-sam ple total of 1120 cases (1114 selected, 6 dropped). 371 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 121 Cox Regression Results for 1986 Ne Data S et (5) Variable Name B Standard Error Exp(B) Significance (p) Cohort 80-86 9-7323 82.1719 16852.96 0.9057 Cohort 75-79 8.9983 82-1723 8089.412 0.9128 Cohort 70-74 9.2338 82.1737 10237.47 0.9105 Uiban X Rural 0.0942 0.2569 1.0988 0.7139 Drinking W ater -0.2275 0.2531 0.7965 0.3687 Sewage -0.3217 0.4457 0.7249 0.4705 Household Crowding 0.102 0.0391 1.1074 * 0.0092 Age Group of Mother -0.0908 0.2243 0.9132 0.6856 Mother's Education -0.4455 0.2002 0.6405 0.0261 Goods -0.2723 0.1376 0.7617 0.0479 Birth Order Risk -0.1553 0.2575 0.8561 0.5464 Sex of Child -0.1481 0.2016 0.8624 0.4625 Breastfeeding in Months -0.3221 0.0602 0.7246 0 Source: 1986 DHS for Brazil's N ortheast region from a sub-sam ple total of 1120 cases (1111 selected. 9 dropped). 372 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 122 Cox Regression Results for 1986 Ne Data S et (6) Variable Name 8 Standard Error Exp(B) Significance (p) Cohort 80-86 9.6993 82.1084 16305.46 0.906 Cohort 75-79 8-9721 82.1088 7879.794 0.913 Cohort 70-74 9.1875 82.1102 9774.245 0.9109 Urt>an X Rural 0.0826 0.2568 1.0862 0-7476 Drinking W ater -0.2428 0.2529 0.7844 0.337 Household Crowding 0.1029 0.0391 1.1083 * 0.0085 Age Group of Mother -0.0894 0.2244 0.9145 0.6902 Mother's Education -0.4557 0.1982 0.634 * 0.0215 Goods -0.2936 0.1347 0.7455 * 0.0293 Birth Order Risk -0.1689 0.2567 0.8446 0.5107 Sex of Child -0.1494 0.2016 0.8612 0.4587 Breastfeeding in Months -0.3227 0.0602 0.7242 0 Source: 1986 DHS for Brazil's Northeast region from a sub-sam ple total of 1120 cases (1111 selected, 9 dropped). 373 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 123 Cox Regression R esults for 1986 Ne Data S e t (7) Variable Name B Standard Error Exp(B) Significance (p) Cohort 80-86 9.0369 73.472 8407-594 0-9021 Cohort 75-79 8.9015 73-473 7343-011 0-9036 Cohort 70-74 9.0289 73-474 8340-971 0-9022 Urban X Rural 0.1686 0-2556 1-1837 0-5093 Drinking W ater -0.2625 0.2522 0-7692 0.2979 Sewage -0.3353 0-4451 0-7151 0-4512 Household Crowding 0.0467 0-0389 1-0478 0.23 Age Group of Mother -0.26 0.2184 0-771 0.2339 Mother’ s Education -0.3854 0-1964 0-6802 0.0497 Goods -0.1989 0-1368 0-8196 0.1459 Birth O rder Risk -0.2076 0.2504 0-8126 0.4071 Sex of Child -0.2641 0-1992 0-7679 0.1848 Source; 1986 DHS for Brazil's Northeast region from a sub-sam ple total o f 1120 cases (1114 selected, 6 dropped). 374 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 124 Cox Regression Results for 1986 Ne D ata S et (8) Variable Name 8 Standard Error Exp (8) Significance (p) Cohort 80-86 9.0072 73.392 8161.932 0.9023 Cohort 75-79 8-8767 73.3925 7163-404 0.9037 Cohort 70-74 8.9839 73.394 7973.514 0.9026 Urban X Rural 0-1557 0.2556 1.1685 0.5423 Drinking W ater -0.2812 0.252 0.7549 0.2645 Household Crowding 0.0463 0.0388 1.0474 0.233 Age Group of Mother -0.26 0.2185 0.771 0.2341 Mother's Education -0.394 0.1946 0.6744 0.0429 Goods -0.2212 0.1341 0.8016 0.099 Birth O rder Risk -0.2175 0.2499 0.8045 0.384 Sex of Child -0.2672 0.1991 0.7655 0.1796 Source: 1986 DHS for Brazil's Northeast region from a sub-sam ple total of 1120 cases (1114 selected, 6 dropped). 375 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. R esults o f Descriptive and M ultivariate Regression A nalysis Applied to the 1991-1986 Pooled Logistic Regression and Cox D ata S ets Descriptive Analysis The sample sizes of the logistic and the Cox data sets is 4038 and 4931, respectively. The birdi cohort variable indicates a decline in the odds of death, particularly as we move from children bom between the most recent cohort (1985-90) and the other ones. In both data sets, the odds of dying for children bom in the second (1980-84) and third (1975-79) cohorts is relatively similar: 8.3 and 8.9% and 1990-95. For children bom before 1974, the odds increase to almost 20%, but the number of cases is fairly small (5.8% of the sub sample). 50.6 % of the children in the logistic and 59.6% in the Cox data set were bom in the most recent birth cohorts. The odds of dying in this cohort is 5.9 and 6.4 % in each data set, respectively. The place of residence variable for the logistic regression data set shows that 7% of the children bom in urban areas (67% of sub-sample) and 8.9% of the ones bom in rural areas did not survive. This variable takes the value 0 for rural and 1 for urban residence. According to the Cox data set, which puts more weight on recent cases, the percentage of deceased children from rural areas is slightly lower, or 8.8%, whereas the urban rate remains the same. As far as the drinking water, the effect seems to be pertinent. 6.7% of the children bom in households with clean drinking water (70%) died, whereas 376 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 10% of those with no access to good drmking water lost their lives. The label values are 0 for not good for drinking and 1 for water good for drinking. In the Cox data set there are no changes in die percentage of children with access to good drinking water who died, but the odds of death for those who did not have access to good drinking water are reduced &om 10 to 9.5%. According to the descriptwe results, the avaflabflity of modem sewage disposal seems to have an impact as great as drinking water. This variable takes the value 0 for not modem and 1 for modem sewage. 6% of the children in households with modem toilet facility and 9.5% in the ones with no modem toilet facili^ died in the logistic regression formatted data set. In the Cox data set, these levels are very close, 5.9 and 9.2%, respectively. 52.8 and 49.8% of the households in the logistic and Cox data sets, respectively, are endowed with modem sewage disposal. The next variable is the age of the mother at the time of the birth. According to the logistic regression data set descriptive results, 9.6% of the children of high risk age mothers die as opposed to 5.6% of the children of mothers at their reproductive peak (low risk). The Cox regression data set indicates similar rates, or 9.7 and 5.7%, respectively. This demographic variable takes the value 0 for high risk age (outside of the 10-34 good age group) and 1 for low risk. 51.9 and 47.9% of the infants in each of the data sets are considered to be children of h i^ risk age mothers. 377 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The household crowding variable indicates that 8.9% of the children who live in crowded homes are likely not to survive in the logistic data set. In the Cox data set this level is 8.8%. hi non crowded (less than 5 indhnduals) households the odds of dying are 6.5 and 7%, respectively. The means distribution for each total household number clearly shows a progression for the mean value of deceased children per household number. The next e3q>lanatoiy variable is considered to have a weighty impact on infant mortality, the mother’ s educational attainment. According to the data sets for LR and Cox analysis, only 3.4% and 3.2%, respectively of the Northeast respondents had higher education. Among those mothers with higher education, .7% and .6 % of their youngest infants died according to the logistic and Cox regression results, respectively. The other schooling categories are notably distinct as far as the odds of dying for the infant. The odds decrease from 10, to 8 to 3.1% as we move from no education to primaxy, to secondary education in the logistic data. In the proportional hazard 1996 model, the descriptive results are similar: 10, 7.8 and 3.5%. In 84% of the cases in both data sets, the mothers have either only primary or no education at all. As mothers acquire a secondary education, the odds of survival for their children is considerably raised. The variable goods indicates that when more than 1 of the 3 consumer goods is present in the household, the odds of dying for the infant reduce substantially 378 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. For no goods or ju st 1 good, the level of deceased children is 9.9 and 9.5% in the logistic set and 9 and 9.5% in the Cox data set. Around 50% of the children live in such homes. When a 2“* good is purchased, usually a TV, chances are the household average income and wealth are higher. 39.7 and 36.9% of the total samples for the logistic and the Cox data sets are associated with * 2 goods’ households. The odds of dying decrease to 6.1 and 6.3%, respectively in the logistic and Cox data sets. When all the three goods (in around 10% of the cases) are present, only 4.6 and 4.3% of the children do not survive their first year in the respective data sets. The next variable is the infant’ s birth order risk. 5.7 and 8.8% of the children who, respectively, have a low (2 « ^ or 3”* ^ child) and high order risk ( 1 » ^ and over 3^) die in the logistic regression data set. The values in the Cox data set are the same for the high birth order risk age group and sli^tly lower for the low risk group, 5.5%. As far as the descriptive results for the gender variable, 6.7% of the girls and 8.6% of the boys died in the logistic data set. In the Cox data set, these levels are 6.8 and 8.3 %, respectively. The Cox regression data set scrutinizes the importance of breastfeeding for the survival of the child. 51.5% of the infants or 2,537 children were not breast-fed at all. 7.7% of the children were breast-fed for just 1 month, 7.6% for 2 and 6.5% for three months. Only 2.4% of the children were breast-fed for 12 months or over. 379 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The proportional risks model censors the data, meaning that if the infant is stiU alive at the end of the year, she ^e) did not make the transition (did not die) at the end of the first year. The mean value of deceased children per months of breastfeeding decrease in a progressive manner: 11 % (0), 6.1% (1), 5.9% (2), 4.7% (3), 3.2% (4) and so on. According to the descriptive results, children who were not breast-fed seem to be significantly more at risk of death. An analysis of the correlation levels between the dependent and the model’ s explanatory variables indicate that the variables s e x and urban X rural are only correlated with the dependent variable at the 0.05 significance level (table 127) As expected, the correlation coefficient for the non-binary household crowding variable is significant (0.045 in the logistic data set) and positively correlated with the dependent variable. All the other variables are significant at the 0.01 level indicating a very low probability lhat the results could be attributed to chance. The variables which show a strongest level of correlation are: duration of breastfeeding ( -0.122); educational attainment (-0.089 in the logistic and -0.086 in the Cox data set); mother’ s age risk ( -0.076 in both sets) and goods ( -0.072 and - 0.063). The birth cohort coefficient is also very significant a t the 0.01 level: -0.088 and -0.077. 380 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 125 l):;: r- j a ” ■ 381 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. /V/ojns j r \ i 382 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 383 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 126 384 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. , ? p/ f ■ , ; < - p'V ’ ,'iV ^ 'o, i, • •. / 385 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. MoJi>s jr:d * ’•'■ L i j': , ' u! ' u-'': : . j ' S' ' .',1 ■ :>n:. ■ I', .i ,»} i » 386 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 127 1991-1986 Correlation Levels with D ependent VariatWe Logistic Regression Data Set C ox Regression Data S et Variables Cohort* -0.088 -0.077 UrtKin X Rural -0.034» -0.33» Drinking W ater -0.057 -0.05 Sewage -0.067 -0.062 Household Crowding* 0.045 0.05 Household Crowding (B) -0.046 -0.047 Mother's Age Risk -0.076 -0.076 Educational Level* -0.089 -0.086 Goods* -0.072 -0.063 Birth Order Risk -0.055 -0.059 Sex of Child -0.036# -0.028 » Breastfeeding** N/A -0.122 Ethnicity* N/A N/A Ethnicity (B) N/A N/A Dr's Prenatal Care N/A N/A Immunization (DPT123) N/A N/A * non dichotomous “ in months A H variables Significant at the 0.01 level (2 tailed) unless indfcaled othenwse # Significant at the a06 ievei (2-taled) XNotConeMed 387 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. D ia^ostic Results for 1991-1986 L R Models The diagnostics statistics reveal that all models fît the facts well, (table 128) According to the 1991-1986 logistic data set, the data denote that model 4, which excludes sewage and includes multiple birth cohorts, has the lowest -2LL - 2065.707 and, consequently, the best prediction power for the dependent variable. The models chisquare indicate that the null hypothesis that the value of the coeffîcients of the variables is equal to zero should be rejected. The variance explained in the model is not caused by chance. All 4 models are very signifîcant. In addition, all models have excellent goodness of fît. 4011 cases were selected and 2 7 m issin g . The prediction efScienty of the models is 92.35% . 388 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 128 Diagnostic Results, Predictive Efficiency and FurtlierCliaracterlstlcs of th e Logistic Regression Models for1991-1986 Pooled Data Set Item Stat. Model 1 Model 2 Model 3 Model 4 -2Log Likelihood 2067.549 2068.504 2068.81 2065-707 Goodness of Fit 3938.411 3936.392 3936.145 3936.572 Model Chisquare 100.278 99.323 103.017 102.12 (Signif) 0 0 0 0 (df) 10 9 12 11 Predictive Efficiency 92.35% 92.35% 92.35% 92.35% N (numt)er of cases included ) 4011 4011 4011 4011 389 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Logistic Regression. R esults for 1 9 9 1 -1 9 8 6 M odels The 2 standard logistic regression models for the pooled 1991-1986 data set (model 1 and 3) are comprised of 9 independent variables plus the birth cohort variable or variables. The other 2 regression models (model 2 and 4) include the same variables with the exception of sewage disposal, (tables 129 to 132) The regression results for the 4 1991-1986 logistic regression models indicate that the mdependent variables that have a statistically signffîcant effect on infant mortality at the 5 percent level of confidence are: birth cohort, goods, mother’ s education, household crowding and sex of the child. The variables ethnicity, DPT immunization and prenatal care by a doctor are not present in the 1991-1986 pooled data set because no data on them were collected in 1986. Sewage is not significant a t all, but this may have to due with the fact that the levels of modem sewage in 1991 were exceptionally high. D rinking water almost has a significant impact on the dependent variable in the models in which the sewage variable is absent (p value of 0.0549 in model 2 and 0.0535 in model 4) In model 1 (complete model with single birth cohort), the variables with the lowest p values are, respectively: 1 ) birth cohort (0.0); 2) goods (0.0068); 3) mother’ s education (0.0082); 4) household crowding (0.0167); 5) sex of the child (0.0316). (table 129) 390 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In model 3 (complete model with multiple birth cohorts), the variables with the lowest p values are, respectively: 1) birth cohort (0.0, 0.0 and 0.0019, respectively for the 85-90, 80-84 and 75-79 cohorts); 2) goods (0.008); 3) mother’ s education (0.0098); 4) household crowding (0.0149); 5) sex of the child (0.0295). (table 131) In model 2 (single birth cohort with no sewage variable), the variables with the lowest p values are, respectively: 1) birth cohort (0.0); 2) goods (0.0015); 3) mother’ s education (0.0057); 4) household crowding (0.0152); 5) sex of the child (0.0304). (table 130) In model 4 (complete model with multiple birth cohorts), the variables with the lowest p values are, respectively: 1 ) birth cohort (0.0, 0.0 and 0.0018, respectively for the 85-90, 80-84 and 75-79 cohorts); 2) goods (0.0018); 3) mother’ s education (0.007); 4) household crowding (0.0139); 5) sex of the chüd (0.0282). (table 132) With the exception of the household crowding variable, all other explanatory variables are inversely correlated with the dependent variable as it would be ejected . A scrutiny of regression coefficients for the independent variables (taking model 3 as a reference) that have an effect on infant m ortali^ at the 0.05 level indicates that when an additional household member is present the odds of dying for the infant increase in over 6% in comparison to what thQr would be otherwise, (table 131) 391 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Similarly, and given the impact of other explanatory variables, the odds of dying for the infant child decrease in 76% when the infant is a female as opposed to what they would be otherwise, (table 131) The odds of dying for the infant- child decrease in 75% as the mother improves her educational attainment from one category to another, (table 131) In a like manner, as the household acquires one additional consumer good, a radio, a T V or a car, the odds of dying for the family's child are also reduced in 68% in comparison to what they would be if that consumer good was not available. In short, the 4 1991-1986 pcxiled logistic regression models very consistently indicate that the following variables have a effect on the likelihood of infant mortality at the 5% level of significance: birth cohort, goods, mother’ s education, household arowding and s«c of the child. 392 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 129 Logistic Regression Results for 1991-1986 Ne D ata S et (1) Variable Name B Standard Error Exp(B) Significance (p) Birth Cohort -0-3764 0.0717 0.6863 0 Urban X Rural 0.1996 0.1742 1.2209 0.2517 Drinking W ater -0.297 0-1704 0.743 0.0813 Sewage -0.152 0.1554 0.859 0.328 Household Crowding 0.0606 0.0253 1.0625 ♦ 0.0167 Age Group of Mother -0.2 0.1456 0.8187 0.1696 Mother's Education -0.2819 0.1066 0.7544 * 0.0082 Goods -0.2256 0.0833 0.798 • 0.0068 Birth O rder Risk -0.1671 0.1456 0.8461 0.251 Sex of Child -0.2618 0.1218 0.7697 0.0316 Constant -1.042 0.0717 - 0 Source: 1991 & 1986 DNS for Brazil’ s Northeast region from a sub-sam ple total of 4038 cases (4011 selected, 27 missing). 393 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 130 Logistic Regression Results for 1991-1986 Ne Data S et (2) Variable Name B Standard Error Exp(B) Significance (p) Cohort -0.3822 0.0714 0.6823 * 0 Urban X Rural 0.162 0.1699 1.1759 0.3402 Drinking W ater -0.3232 0.1683 0.7238 0.0549 Household Crowding 0.0613 0.0253 1.0632 * 0.0152 Age Group of Mother -0.1837 0.1448 0.8321 0.2046 Mother's Education -0.2928 0.106 0.7461 * 0.0057 Goods -0.2515 0.079 0.7776 0.0015 Birth Order Risk -0.1791 0.145 0.836 0.2167 Sex of Child -0.2636 0.1218 0.7683 0.0304 Constant -1.0214 0.2347 - 0 Source; 1991 & 1986 DHS for Brazil's Northeast region from a sub-sam ple total of 4038 cases (4011 selected, 27 missing). 394 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 131 Logistic Regression R esults for 1991-1986 Me D ata S et (3) Variable Name B Standard Error Ex p C B ) Significance (p) Cohort 85-90 -1.2757 0.2282 0.2792 0 Cohort 80-84 -0.8932 0.2174 0.4093 0 Cohort 75-79 -0.7595 0.2451 0.4679 * 0.0019 Urban X Rural 0.2018 0.1742 1.2236 0.2468 Drinking W ater -0.2991 0.1742 0.7415 0.0797 Sewage -0.1482 0.1707 0.8622 0.3431 Household Crowding 0.0616 0.0253 1.0636 * 0.0149 Age Group of Mother -0.214 0.1458 0.8073 0.1422 Mother's Education -0.2757 0.1068 0.759 * 0.0098 Goods -0.2215 0.0836 0.8013 0.008 Birth O rder Risk -0.1745 0.1459 0.8399 0.2317 Sex of Child -0.2654 0.1219 0.7669 0.0295 Constant -0.8878 0.2555 - 0.0005 Source; 1991 & 1986 DHS for Brazil's Northeast region from a sutxsam ple total of 4038 cases (4011 selected, 27 m issing). 395 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 132 Logistic Regression Results for 1991-1986 Ne Data S et (4) Variable Name B Standard Error Exp(B) Significance (p) Cohort 85-90 -1.2914 0.2276 0.2749 0 Cohort 80-84 -0.8909 0.2172 0.4103 0 Cohort 75-79 -0-7642 0.245 0.4657 0.0018 Urban X Rural 0.1657 0.17 1.1802 0.3298 Drinking W ater -0.3253 0.1684 0.7223 0.0535 Household Crowding 0.0622 0.0253 1.0642 0.0139 Age Group of Mother -0.1988 0.145 0.8197 0.1706 Mother's Education -0.2864 0.1062 0.751 * 0.007 Goods -0.2468 0.0792 0.7813 0.0018 Birth O rder Risk -0.1859 0.1453 0.8304 0.2008 S ex of Child -0.2675 0.1219 0.7653 0.0282 Constant -0.872 0.2549 - 0.0006 Source: 1991 & 1986 DHS for Brazil's Northeast region from a sub-sam ple total of 4038 cases (4011 selected, 27 missing). 396 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. D iagnostic R esu lts for 1991-1986 Cox M odels The diagnostics statistics mdicate that all 8 Cox regression models fit the facts weU. (table 133 - 134) In the 1991-1986 pooled Cox models with no breastfeeding variable ( models 3, 4 , 7 and 8), 18 cases were dropped and 4913 selected. 4570 children- 93% of the sample - are censored and survive the first year of life whereas 343 do make the transition, hi the models including the duration of breastfeeding in months (models 1, 2, 5 and 6), 23 cases were dropped and 4908 selected. In a like manner 95.1% of the values ( 4569 cases) are censored and 185 children do not survive the first year of life. In the Cox data set, the data reveal that the complete models, which include the sewage and breastfeeding variables ( 1 & 5 ) have the best predictive power. The respective -2LL for these two models are 5499.798 and 5492.463. The Gms or models chisquare are very significant, indicating that the null hypothesis that the value of the coefficients of the variables is equal to zero should be rejected. The independent variables do improve the predictability o f the models. The Cox regression models with the highest chisquare are models 5 and 1: 169.553 and 163.191, respectively. 397 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 133 DiagnosUc Results. Predictive Efficiency and Fuittier Ctiaracteristics of the Cox Regression Models for 1991-86 Data S et (1 of 2) Item Stat. Model 1 Model 2 Model 3 Model 4 Events 339 339 343 343 Censored 4569 4569 4570 4570 (%) 93.1 93.1 93 93 -2Log Likelihood 5499.798 5503.624 5721.812 5722.948 Chisquare (Overall ) 163.191 160.71 89.019 88.066 (Signif) 0 0 0 0 (df) 11 10 10 9 N (number of cases included ) 4908 4908 4913 4913 398 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 134 Diagnostic Results. Predictive Efficiency and Further Characteristics of the C ox Regression Models for 1991-86 Data S e t (2 of 2) M odels M odels M odel? M odels Item Stat. Events 339 339 343 343 Censored 4569 4569 4570 4570 C % ) 93.1 93.1 93 93 -2Log Likelihood 5492.463 5496.498 5719.819 5720.875 Chisquare (Overall ) 169.533 166.835 92.884 91.935 (Signif) 0 0 0 0 (df) 13 12 12 11 N (number of cases included ) 4908 4908 4913 4913 399 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Cox Régression R esults for 1991-1986 M odels The Cox method presents results whichr are very similar to results derived through the logistic method. The proportional hazards’ procedure includes the h%hly important tune-dependent breastfeeding variable. The sample size for the Cox data set is larger than the logistic one. Cox’ s explanatory power is also expected to be higher. The 2 main Cox regression models for the 1991-1986 data set (model 1 and 5) are comprised of 10 independent variables in addition to the birth cohort variable or variables. 2 regression models (model 2 and 6) include the same variables with the exception of sewage. 2 other models ( models 3 and 7) also consist of the same variables with the exception of the duration of breastfeeding variable. The last 2 models (models 4 and 8) exclude both sewage and duration of breastfeeding variables, (tables 135 to 142) The variables ethnicity, DPT immunization and prenatal care by a doctor are not present in the 1991-1986 pooled data set because no data on them were collected in 1986. The regression results for the 8 1991-1986 Cox models are similar to the logistic results and they indicate that the following independent variables have a statistically signifîcant effect on infant mortality a t the 5 percent level of signifîcance : breastfeeding, goods, mother’ s education and household crowding. 400 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The birth cohort variable has a statistically significant impact on the dependent variable in all models but 1 and 2. E)ven though the statistical signffîcance of the variable sewage disposal may have been affected by the exceptionally high level of modem toilet facilities in the 1991 data set, this variable has associated p values of 0.0503 and 0.0444- respectwely in the standard models 1 and 5. Sex of the child is also a significant variable at the 0.05 level in the models without the breastfeeding variable (3,4,7 and 8). In model 1 (complete model with smgle birth cohort), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0); 2) household crowding (0.0001); 3) goods (0.001); 4) mother’ s education (0.0053). (table 135) In model 5 (complete model with multiple birth cohorts), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0); 2) household crowding (0.0); 3) goods (0.0018); 4) mother’ s education (0.0059); 5) birth cohort (0.0128 and 0.0123, respectively for the 80-84 and 75-79 cohorts); 6) sewage (0.0444); (table 139) In model 2 (single birth cohort with no sewage variable), the variables with the lowest p values are, respective^ 1 ) breastfeeding (0.0); 2) goods (0.00); 3) household crowding (0.0001); 4) mother’ s education (0.026). (table 136) 401 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In model 6 (multiple birth cohorts with no sewage variable), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0); 2) household crowding (0.0); 3) goods (0.0); 4) mother’ s education (0.0028); 5) birth cohort (0.0112 and 0.011, respectnrely for the 80-84 and 75-79 cohorts); (table 140) In model 3 (sin^e birth cohort with no breastfeeding^, the variables with the lowest p values are, respectively: 1 ) birth cohort (0.0002); 2) mother’ s education (0.0034); 3) household crowding (0.0055); 4) goods (0.0067); 5) sex of the child ( 0.031). (table 137) In model 7 (multiple birth cohorts with no breastfeedmg), the variables with the lowest p values are, respectively: 1 ) mother’ s education (0.0039); 2) household crowding (0.051); 3) goods (0.079); 4) birth cohort (0.0033 and 0.0184, respectively for the 80-84 and 75-79 cohorts); 5) sex of child (0.0295). (table 141) In model 4 (single birth cohort without sewage and breastfeeding), the variables with the lowest p values are, respectively: 1 ) birth cohort (0.0001); 2) goods (0.0012); 3) mother’ s education (0.0023); 3) household crowding (0.005); 5) sex of the child ( 0.0292). (table 138) In model 8 (multiple birth cohorts with no breastfeeding and no sewage), the variables with the lowest p values are, respectively: 1) goods (0.0015); 2) mother’ s education (0.0027); 3) household crowding (0.0047); 4) birth cohort 402 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (0, 0.0034 and 0.017, respectively for the 85-91, 80-84 and 75-79 cohorts); 5) sexofchfld (0.0277). (table 142) With the exception of the household crowding variable, all other explanatory variables are inversely correlated with the dependent variable as it would be expected. An analysis of the regression coefScients for the independent variables (taking model 5 as a reference) that have an impact on the dependent variable at the 0.05 level of significance indicates that when an additional household member is present the odds of dying for the infant increase in over 9% in comparison to what they would be otherwise. Analogously, and given the impact of other explanatory variables, the odds of dying for the infant child decrease in 76% as the mother improves her educational attainment fiom one category to another. As the household wealth mcludes one additional consumer good- either a radio, a TV or a car- the odds of dying for the family’ s child are also reduced in 79% in comparison to what they would be if that consumer good was not available, (table 139) These results are very similar to the logistic regression results. In sum, the 8 1991-1986 pooled Cox regression models indicate that the following variables have an effect on the likelihood of infant mortality at the 5% level of significance: breastfeeding, goods, mother’ s education and household crowding. Birth cohort and sex of the child are also statistically significant in some of these regression models. 403 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 135 Cox Regression Results for 1991-1986 Ne Data S et (1) Variable Name B Standard Error ExpCB) Significance (jp) Cohort -0.0363 0.07 0.9644 0.6043 Uiban X Rural 0.1446 0.1583 1.1556 0.3611 Drinking W ater -0.2526 0-1563 0-7768 0.106 Sewage -0.2808 0.1434 0.7552 0.0503 Household Crowding 0.0846 0.0212 1.0883 0.0001 Age Group of Mother -0.0319 0.1292 0.9686 0.8051 Mother's Education -0.2726 0.0978 0.7614 0.0053 Goods -0.2462 0.0749 0.7818 * 0.001 Birth Order Risk -0.1219 0.1314 0.8853 0.3538 Sex of Child -0.2013 0.1101 0.8177 0.0676 Breastfeeding -0J2605 0.0345 0.7707 * 0 Source: 1991 & 1986 DNS for Brazil's Northeast region from a sub-sam ple total of 4931 cases (4908 selected. 23 dropped). 404 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 136 Cox Regression Results for 1991-1986 Ne D ata S et (2) Variable Name B Standard Error Exp(B) Significance (p) Cohort -0.0469 0.0698 0.9542 0.5015 Urtrao X Rural 0.0775 0.1548 1.0805 0.6169 Drinking W ater -0.3016 0.1545 0.7396 0.0509 Household Crowding 0.0845 0.0212 1.0881 0.0001 Age Group of Mother -0.0114 0.129 0.9887 0.9297 Mother's Education -0.2926 0.0972 0.7463 0.0026 Goods -0.2969 0.0703 0.7431 0 Birth Order Risk -0.1398 0.1311 0.8695 0.2859 Sex of Child -0.208 0.1101 0.8122 0.0588 Breastfeeding in months -0.2573 0.0344 0.7731 0 Source; 1991 & 1986 OHS for Brazil's Northeast region from a sub-sam ple total of 4931 cases (4908 selected. 23 dropped). 405 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 137 Cox Regression Results for 1991-1986 Ne Data S et (3) Variable Name B Standard Error ExpCB) Significance (p) Cohort -0.2512 0.0666 0.7779 * 0.0002 Urban X Rural 0.199 0.1562 1.2202 0.2026 Drinking W ater -0.2618 0.1521 0.7697 0.0853 Sewage -0.1512 0.1418 0.8597 0.2864 Household Crowding 0.0598 0.0216 1.0616 * 0.0055 Age Group of Mother -0.1998 0.1276 0.8189 0.1174 Mother's Education -0.285 0.0972 0.752 * 0.0034 Goods -0.203 0.0748 0.8162 * 0.0067 Birth Order Risk -0.1264 0.1306 0.8813 0.3334 Sex of Child -0.2362 0.1095 0.7896 0.031 Source: 1991 & 1986 DNS for Brazil's N ortheast region from a sub-sample total of 4931 cases (4913 selected, 18 dropped). 406 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 138 Variable Name Cox Regression Results for 1991-1986 Ne D ata S e t (4) B Standard Error Exp (8) Significance (p) Cohort -0.2557 0.0664 0.7744 0.0001 Urt)an X Rural 0.1624 0.1526 1.1763 0.2873 Drinking W ater =0=2866 91808 0=7807 0=0867 Household Crowding 0.0605 0.0215 1.0623 * 0.005 Age Group of Mother -0.1874 0.1272 0.8291 0.1408 Mother's Education -0.2951 0.0967 0.7444 * 0.0023 Goods -0.2293 0.0706 0.7951 * 0.0012 Birth O rder Risk -0.1368 0.1302 0.8721 0.2934 S ex of Child -0.2387 0.1094 0.7877 * 0.0292 Source; 1991 & 1986 DHS for Brazil's Northeast region from a sut>-sample total of 4931 cases (4913 selected, 18 dropped). 407 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 139 Variable Name Cox Regression Results for 1991-1986 Ne Data S et (5) B Standard Error Exp(B) Significance (p) Cohort 85-91 -0.3897 0.2178 0.6772 0.0736 Cohort 80-84 -0.5327 0.2139 0.587 * 0.0128 Cohort 75-79 -0.6227 0.2487 0.5365 * 0.0123 Urban X Rural 0.1421 0.1584 1.1527 0.3696 Drinking W ater -0.2432 0.1568 0.7841 0.1209 Sewage -0.2893 0.1439 0.7488 * 0.0444 Household Crowding 0.0868 0.0211 1.0907 * 0 Age Group of Mother -0.0471 0.1299 0.954 0.717 Mother's Education -0.2701 0.098 0.7633 * 0.0059 Goods -0.2342 0.075 0.7912 * 0.0018 Birth Order Risk -0.1397 0.1312 0.8697 0.2872 Sex of Child -0.2002 0.1102 0.8186 0.0692 Breastfeeding -0.2705 0.0353 0.763 * 0 in months Source: 1991 & 1986 DHS for Brazil's Northeast region from a sut>-sample total of 4931 cases (4908 selected, 23 dropped). 408 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 140 Cox Regression R esults for 1991-1986 Ne D ata S et (6) Variable Name B Standard Error Exp(B) Significance (p) Cohort 85-91 -0.4183 0.2174 0.6582 0.0543 Cohort 80-84 -0.5421 0.2137 0.5815 0.0112 Cohort 75-79 -0.6317 0.2485 0.5317 * 0.011 Urban X Rural 0.0748 0.155 1.0777 0.6295 Drinking W ater -0.2956 0.1548 0.7441 0.0561 Household Crowding 0.0863 0.0211 1.0901 * 0 Age Group of Mother -0.0279 0.1298 0.9725 0.8298 M other's Education -0.2913 0.0975 0.7473 « 0.0028 Goods -0.2868 0.0703 0.7506 * 0 Birth O rder Risk -0.1577 0.1309 0.8541 0.2283 S ex of Child -0.2069 0.1101 0.8131 0.0602 Breastfeeding in months -0.2671 0.0352 0.7656 * 0 Source; 1991 & 1986 DHS for Brazil's Northeast region from a sub-sam ple total of 4931 cases (4908 selected, 23 dropped). 409 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 141 Variable Name Cox Regression Results for 1991-1986 Ne Data S et (7) B Standard Error Exp (B) Source; 1991 & 1986 DHS for Brazil's Norttieast region from a sut>-sampie total of 4931 cases (4913 selected, 18 dropped). Significance (p) Coftort 85-91 -0.8875 0.217 0.4117 0 Cohort 80-84 -0.6303 0.2148 0.5324 0.0033 Cohort 75-79 -0.5861 0.2485 0.5565 * 0.0184 Urban X Rural 0.2 0.1561 1.2214 0.1999 Drinking W ater -0.2635 0.1522 0.7684 0.0835 Sew age -0.1462 0.1423 0.8639 0.3041 Household Crowding 0.0603 0.0215 1.0621 * 0.0051 Age Group of Mother -0.2088 0.1277 0.8115 0.1019 Mother's Education -0.2809 0.0973 0.7551 * 0.0039 Goods -0.1992 0.075 0.8194 0.0079 Birth Order Risk -0.1331 0.1307 0.8754 0.3086 Sex of Child -0.2383 0.1095 0.7879 0.0295 410 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 142 Cox Regression Results for 1991-1986 Ne Data S e t (8) Variable Name B Standard Error Exp(B) Significance (jp) Cohort 85-91 -0.8997 0.2166 0.4067 * 0 Cohort 80-84 -0.6289 0.2147 0.5332 • 0.0034 Cohort 75-79 -0.593 0.2484 0.5527 • 0.017 Urban X Rural 0.1651 0.1525 1.1795 0.279 Drinking W ater -0.2883 0-1504 0.7495 0.0553 Household Crowding 0.0608 0.0215 1.0627 0.0047 Age Group of Mother -0.1972 0.1273 0.821 0.1215 Mother's Education -0.2907 0.0968 0.7478 0.0027 Goods -0.2248 0.0708 0.7987 * 0.0015 Birth O rder Risk -0.1431 0.1303 0.8667 0.2722 Sex of Child -0.241 0.1095 0.7859 » 0.0277 Source: 1991 & 1986 DHS for Brazil's Northeast region from a sub-sam ple total of 4931 cases (4913 selected, 18 dropped). 411 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. R esults of Descriptive and M ultivariate Regression Analysis Applied to the Pooled 1996-91-86 Logistic R egression and Cox D ata S ets Descriptive Analysis The statistic results of the descriptive analysis applied to the pooled data sets for all three surveys indicate that all the selected independent variables when examined separately have shown to have an impact on infant mortality. The 1996-91-86 pooled data sets, with 7307 and 8753 cases, are the largest to be used in this study. According to the logistic regression data set (7307 cases), the birth cohort variable clearly indicates a linear decline in the odds of death as we move from children bom between in the oldest birth cohort (1979 and. before) and 1980-84, 1985-90 and 1990-95. 10% of the 950 subjects who were bom before 1979 (constituting 13% of the sample) died; 8.4% of the children bom between 1980 and 1984 (25% of sub-sample) did not survive; 6.2% between 1985 and 1989 ( 34.5% of sub-sample) and 4.7% in the latest birth cohort period ( 27.6% of sub-sample). This variable takes on values from 0 to 3, the former being the oldest cohort, and the latter the newest. According to the Cox data set (8753 cases), the means for the birth cohort categories are very close to the logistic regression ones: 10, 8.4, 6.6 and 4.7% for the four categories, the latter relating to the latest 1990-96 birth cohort. The most recent cohort covers 27.6% of the sub-sample in the logistic data set and 36.2% in the Cox one (one year longer). 412 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The place of residence variable for the logistic r^ression data set shows that 6.1% of the children bom m urban areas (70% of sub-sample) and 8.5% of the ones bom in rural areas died. This variable takes die value 0 for rural and I for urban residence. According to the Cox data set, which puts more w e i^ t on recent cases, the levels of deceased children fiom urban and rural areas is sli^tly lower: 5.9 and 8.4%, respectively. The next independent variable is drinking water. 6.2% of the children bom in households with clean drinking water (71.5%) died, while 8.5% of those with no access to good drinking water lost their lives. The values are 0 for not good for drinking and I for water good for drinking. In the Cox data set these odds are quite similar: 6 and 8.3%. Type of toilet facflity or sewage is the next independent variable which takes the values 0 for not modem and I for modem sewage. 5.4% of the children in households with modem toilet facility and 7.8% of the ones with no modem toilet facility died in the logistic regression formatted data set. In the Cox data set, these rates are very close: 5.3 and 7.6 %, respectively. According to the logistic regression data set descriptive results, 8.7% of the children of high risk mothers die as opposed to 4.8% of the children of mothers at their reproductive peak (low risk). The Cox regression data set indicates identical rates. 413 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. This demographic variable takes the value O for high risk age (outside of the 10-34 good age group) and I for low risk. In the logistic data set 52.2% of the mothers had them children at a rislqr age. In the Cox data set the proportion of mothers in the high risk age group decreases to 48.6%. The household crowding variable was created to investigate the children’ s exposure to health hazards such as infectious and respiratory diseases. The means distribution for each total household number in these large data sets clearly shows a progression for the mean value of deceased children per household number. In 2 and 4 individuals households, on average, 5/6% of the latest children of each respondent die before their first year of life. As we increase the number of residents firom 5 to 6, the mean values for the odds of death increase to around 8% and, as we move to 7 and 8 residents, it grows to 9%. The proportional hazards regression results derived firom the Cox data set indicates a very similar statistic progression. The dichotomous crowding variable suggests means of 5.6 (non crowded) and 8.4% (crowded) in the logistic data set, and 5.5 and 8.2% in the Cox formatted data sets, respectively. The next independent variable considered is the education level of the mother as indicated by the number of years in school. 414 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. This variable is divided into the same 4 categories: higher education (college level and over); secondary (high school level completed); primary (including primary complete and secondary incomplete, 1 to 11 years); no education. According to the data sets for LR and Cox analysis, only 3.8 and 3.5%, respectively of the Northeast respondents had higher education. Among those mothers with higher education, only .7% and 1 % of their youngest infants did not survive accordmg to the logistic and Cox regression results. However, the other sub-categories reveal a significant progression: the odds of survival decrease from 11, to 7.4 to 3.6% as we move firom no education to prfinary, to secondary education in the logistic data. In the proportional hazard 1996 model, the means are almost the same: 11, 7.2 and 3.5%. Around 70% of the mothers in both data sets had no schooling or just completed the primary level. 29.2 The following variable is the proxy for household income and wealth. It reflects the ownership of 3 durable consumer goods: car, TV and radio. For each positive answer, one point was assigned. The values range from 0 to 3. Once again the change from 1 to 2 goods, which usually implies the purchase of a TV set, has the strongest impact on the odds of dying for the infant. 13.6% of the households in the logistic data set and 12.4% in the Cox data set had all 3 goods. Analogously, 14.6% and 16.7% of the mothers respectively in the logistic and Cox data sets had none of the three goods. 415 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The mean, values of deceased children who had 0 and I good is 8.9 and 9.2% in the logistic data set, and 8.7 and 8.4% in the Cox one. With 2 goods, the percentage of children who did not survive is reduced to 5.3 and 5.4% respectively for the logistic and the Cox formatted data files. If the mother has access to all 3 goods the rate of deceased infants is reduced to 3 and 2.9% respectively in the logistic and Cox data sets. The next variable is the infant’ s birth order risk. Approximately 60% of the infants in both data sets are considered to be exposed to a high birth order risk and over 3 "*). The odds of dying are 4.8 and 8.2% in the logistic and 4.7 and 8% in the Cox data sets. The lower figures relate to children who were of a low birth order risk ( 2 ™ * or 3 ^ ^ child). The pooled data sets for both the logistic and the Cox formatted regressions corroborate the male excess m ortally 7.6 and 7.4 % of the males as opposed to 6.1 and 6% of the females did not survive according to the logistic and the Cox data sets respectively. On average, 52% of the children in the data sets are males. The Cox regression data set attests the importance of breastfeeding for the survival of the child. 54.1% of the in fan ts or 4 ,7 3 7 cases were not breast fed at all. Almost 10% of these children died before the end of their first year. 6.2% of the children were breast-fed for ju st 1 month, 6.5% for 2 and 6.2% for 3 months. Only 1.8% of die children were breast-fed for at least 12 months. 416 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The mean value of deceased children per months of breastfeeding decrease in a linear fashion: 9-6 % (0), 5-9% (1), 5-2% (2), 4% (3), 2-4% (4) and so on- Next we will examine the correlation levels with the dependent variable according to the pooled 1996-1991-1986 Cox and logistic data sets, (table 145)- According to the data, the variable se x is only correlated with the dependent variable at the 0.05 level of significance- As expected, the correlation coefScient for the non dichotomous household crowding variable is significant (0.05 m both data sets) and positively correlated with the dependent variable- The Pearson’ s product moment coefBcient for all the other variables are significant at the 0-01 level indicating a very low probability that the results could be attributed to chance. The mother’ s age risk at the time of the birth and the birth cohort variable have relatively high levels of correlation with the odds of dying according to this pooled survey. Nevertheless, the covariates which show the strongest level of correlation are: duration of breastfeeding ( -0-113); educational attainment (-0-107 in the logistic and -0.104 in the Cox data set) and goods ( -0.085 and -0-077). 417 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 143 Moans andt a d . r ■ ( K ' i ' o r : , i ‘--r : . a r m m : 418 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. M c^ins a n d f roq rm .ill '..Ill ihii < of .'jn \ i ' I otihHl ; r, y , s ,' ' ' ’-li'qi c : K i l . i b , i.lll'i'- 'Vli '. l l ■ ■ i i q o - i'-i y 419 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. M o .in s ,]lld ‘ .'■• U '() ■ .1, I'it^'Prruh • ’s ' -J -y' f f - ' n , , / . ■./ I,; / s f ' I ' J{-!' 420 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 144 J / r - 'nr .>n ■ r : ili'f • ! , . n u ) t ^ V V f i 1 - S h Av ' ' : o x ^ > v .-, / r i ' .Jl^n A '. \j;; - L '- y [ , 421 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. /V/eans j n , i ‘ / .■h; .1'' f . j ' < j ; t / ^ ^ s '.i!* 7 v f . i r , ,V-■ ^ ’ 0 » ^ .^ s ^ ■ i >p fj / . y ^ p r . ' . y / 422 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 423 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 145 1996-1991-1986 Correlation Levels with Dependent Variable Logistic Regression Data Set Cox Regression Data S et Variables Cohort* -0.074 -0.072 Urban X Rural -0.044 -0.048 Drinking W ater -0.042 •0.043 Sewage -0.047 -0.044 Household Crowding* 0.05 0.054 Household Crowding (B) -0.053 -0.055 Mother's Age Risk -0.077 -0.076 Educational Level* -0.107 -0.104 Goods* -0.085 -0.077 Birth O rder Risk -0.067 -0.065 Sex of Child -0.03# -0.028 # Breastfeeding** N/A -0.113 Ethnicity* N/A N/A Ethnicity (B) N/A N/A Dr’ s Prenatal Care N/A N/A Immunization (DPT123) N/A N/A * non dichotomous * * in months A n variaUes Significant at the 0.01 level (2 tailed) untesa indicated othanwiae # Significant at the 0.05 level ( 2-taied) XNotCcfreMed 424 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. D ia ^ o stic R esu lts for 1996-1991-1986 LR M odels The diagnostics statistics reveal that all models fit the facts well, (table 146) The -2 Log Likelihood statistic is the analogue of the Sum of the Square Errors (SSE) in OLS regression and «qxresses how poorly the models fit with aU the included variables. The laiger the -2LL, the worse is the prediction power for the dependent variable. In the pooled 1996-1991-1986 logistic data set, the data show that the models which include the sewage variable ( 1 & 3 ) are clearly better. The prediction power for the dependent variable in these models is very similar. The value for the -2LL statistic in model 3 is 3439.855, and in model 1 it is 3439.855. The models chisquare indicate that the null hypothesis that the value of the coefficients of the variables is equal to zero should be rejected. All models are very significant. The probability of obtaining these results by chance is nil. All models have excellent goodness of fit. 7258 cases were selected and 49 missing. The prediction efficiency of the models is 93.15%. 425 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 146 Diagnostic Results, Predictive Efficiency and Furttier Characteristics o f the Logistic Regression Models for 1996-91-86 Pooled Data Set Item Stat. Model 1 Model 2 Model 3 Model 4 -2Log Likelihood 3439.855 3443.914 3439.832 3443.832 Goodness of Fit 7022.738 7010.819 7023.342 7012.984 Model Chisquare 184.489 180.431 184.512 180.513 (Signif) 0 0 0 0 (df) 10 9 12 11 Predictive Efficiency 93.15% 93.15% 93.15% 93.15% N (number of cases included ) 7258 7258 7258 7256 426 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Logistic Regression. R esu lts for 1 9 9 6-1991-1986 M odels The 1996 1991 1986 pooled data set is by far the largest of all logistic- designed data sets. Its sample size is 7307 cases. The 2 standard logistic regression models for this data set (model 1 and 3) are comprised of 9 independent variables in addition to the birth cohort variable or variables. The other 2 regression models (model 2 and 4) consist of the same variables with the exception of sewage disposal, (tables 147 to 150). The regression results for the 4 1996-1991-1986 logistic regression models indicate that most of the independent variables have a statistically significant impact on infant mortality at the 5 percent level of confidence. Place of residence and drinking water are not significant in any of the models, though. Birth order risk is also not statistically significant in the 2 basic models (1 and 3). Birth cohort, goods, mother’ s education, household crowding, sewage, sex of the child as well as age group of the mother variables all have an effect on the probability of infant mortality a t the 5% level of significance. The variables ethnicity, DPT immunization and prenatal care by a doctor are not present in the 1996-1991-1986 pooled data set for no data on them were collected in 1986. Birth order is also significant in the models wifii no sewage (2 and 4). 427 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. D a model 1 (complete model with, single birth cohort), the variables with the lowest p values are, respectwely: 1 ) birth cohort (0.0); 2) goods (0.0); 3) mother’ s education (0.0); 4) age group of mother (0.0045); 5) sex of the child (0.0089); 6) household crowdmg (0.0139); 7) sewage (0.045). (table 147) In model 3 (complete model with multiple bùth cohorts), the variables with the lowest p values are, respectively: 1) birth cohort (0.0, 0.0001 and 0.045, respectively for 90-95, 85-89 and 80-84 cohorts); 2) goods (0.0); 3) mother’ s education (0.0); 4) age group of mother (0.0047); 5) sex of the child (0.0091); 6) household crowding (0.0139); 7) sewage (0.045). (table 149) In model 2 (single birth cohort with no sewage variable), the variables with the lowest p values are, respectwely: 1 ) birth cohort (0.0); 2) goods (0.0); 3) mother’ s education (0.0); 4) age group of mother (0.0045); 5) sex of the child (0.0084); 6) household crowding (0.0174); 7) birth order risk (0.0423) (table 148) In model 4 (complete model with multiple birth cohorts), the variables with the lowest p values are, respectively: 1 ) birth cohort (0.0 and 0.0002, respectively for the 90-95 and 85-89 cohorts); 2) goods (0.0); 3) mother’ s education (0.0); 4) age group of mother (0.0048); 5) se x of the child (0.0086); 6) household crowding (0.0173); 7) birth order risk (0.0425). (table 150) In addition to the household crowding variable, place of residence is also inversely correlated with the dependent variable, meaning that, according to the data, living m urban areas would increase infant mortality. 428 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. An examination of the (B ) for the urban X rural variable in model 3 suggests that moving from a rural to an urban area would increase the odds of dying in 0.4%. (table 149) Similarly, and given the impact of other explanatory variables, the odds of dying for the infant child decrease in 78% when the infant is a female as opposed to what the odds would be otherwise, (table 149) The odds of dying for the infant child decrease in 71% as the mother improves her educational attainment from one category to another, (table 149) The odds of dying for the infant child would also decrease in 73% if the mother is a low age risk mother as opposed to what they would be if she was either too young or too old. (table 149) In a like manner, as the household acquires one additional consumer good, a radio, a TV or a car, the odds of dying for the family^s child are also reduced in 76% in comparison to what they would be if that consumer good was not available. In sum, according to the pooled 1996-1991-1986 logistic regression models the variables birth cohort, goods, mother’ s education, household crowding, sewage, sex of the child as well as age group of the mother all have an impact on the probability of infant mortality at the 5% level of significance. 429 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 147 Logistic Regression R esults for 1996-91-86 Ne Data S et (1) Variable Name B Standard Error Exp(B) Significance (p) Birth Cohort -0.2794 0.0531 0.7562 0 Urban X Rural 0.0443 0.1274 1.0453 0.7283 Drinking W ater -0.0131 0.1259 0.987 0.9173 Sewage -0.227 0.1132 0.7969 0.045 Household Crowding 0.0493 0.0201 1.0506 0.0139 Age Group of Mother -.3150 0.1109 0.7298 0.0045 Mother's Education -0.3377 0.0768 0.7134 0 Goods -0.2738 0.0639 0.7605 0 Birth Order Risk -0.2167 0.1117 0.8052 0.0524 Sex of Child -0.2489 0.0952 0.7797 0.0089 Constant -1.2965 0.1737 - 0 Source; 1996,1991 and 1986 DHS for Brazil's Northeast region from a sub sam ple total of 7307 cases (7258 selected, 49 missing). 430 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 148 Logistic Regression Results for 1996-91-86 Ne Data S et (2) Variable Name B Standard Error Exp(B) Significance (p) Birth Cohort -0.2693 .0530 . 0.7639 0 Urban X Rural 0.0057 0.126 1.0058 0.9636 Drinking W ater -0.0455 0.125 9555 0.7158 Household Crowding 0.0476 0.02 1.0488 0.0174 Age Group of Mother -0.3151 0.1110 . 0.7297 0.0045 Mother's Education -0.339 0.0768 0.7125 0 Goods -0.3018 0.0624 0.7395 0 Birth Order Risk -0.2263 0.1115 0.7974 0.0423 Sex of Child -0.2508 0.0951 0.7782 0.0084 Constant -1.2948 0.1737 - 0 Source: 1996,1991 and 1986 DHS for Brazil's N ortheast region from a sub sam ple total of 7307 cases (7258 selected. 49 missing). 431 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 149 Variable Name Logistic Regression R esults for 1996*91-86 Ne D ata S et (3) B Standard Error Exp (B) Significance (p) Cohort 90-95 -0.8346 0.1705 0.4341 * 0 Cohort 85-89 -0.5741 0.1485 0.5632 * 0.0001 Cohort 80-84 -0.2863 0.1428 0.751 * - 0.045 Urtian X Rural 0.0436 0.1275 1.0446 0.7322 Drinking W ater -0.0131 0.1259 0.987 0.9171 Sew age -0.2261 0.1136 0.7976 0.0466 Household Crowding 0.0495 0.0201 1.0508 * 0.0139 Age Group of Mother -0.3142 0.1111 0.7304 * 0.0047 M other's Education -0.3383 0.0769 0.713 ♦ 0 Goods -0.2738 0.0639 0.7604 * 0 Birth O rder Risk -0.2167 0.1117 0.8052 0.0524 Sex of Child -0.2484 0.0952 0.7801 * 0.0091 Constant -1.2911 0.1845 - 0 Source: 1996,1991 and 1986 DHS for Brazil's Northeast region from a sub sam ple total of 7307 cases (7258 selected, 49 missing). 432 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 150 Variable Name Logistic Regression Results for 1998-91-86 Ne Data S et (4) B Standard Error Exp(B) Significance (p) Cohort 90-95 -0.795 0.1695 0.4516 0 Cohort 85-89 -0.561 0.1485 0.5706 0.0002 Cohort 80-84 -0.2657 0.1612 0.7667 * 0.0619 Urban X Rural 0.0054 0.126 1.0054 0.9661 Drinking W ater -0.0456 0.125 0.9555 0.7154 Household Crowding 0.0478 0.0201 1.049 * 0.0173 Age Group of Mother -0.3138 0.1112 0.7307 * 0.0048 Mother's Education -0.3399 0.0769 0.7118 * 0 Goods -0.3016 0.0624 0.7397 * 0 Birth O rder Risk -0.2261 0.1115 0.7976 0.0425 Sex of Child -0.2501 0.0952 0.7788 0.0086 Constant -1.2923 0.1845 _ 0 Source: 1996,1991 and 1986 OHS for Brazil's Northeast region from a sub sam ple total of 7307 cases (7258 selected. 49 missing). 433 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. D iapnostic R esults for 1 9 9 6 -1 9 9 1 -1 9 8 6 Cox M odels The diagnostics statistics reveal that all 8 Cox regression models fît the facts well, (table 151 - 152) According to the pooled Cox data set for all years, in the models with no breastfeeding variable included ( models 3, 4 , 7 and 8), 31 cases were dropped and 8722 selected. 93.9% of these children (8187) are censored and survive the first year of life, whereas 535 infants do make the transition, hi the models in which the duration of breastfeeding in months variable is included (models 1, 2, 5 and 6), 57 cases were dropped and 8696 selected. 94% of the values (8172) are censored and 524 children do not survive and make the transition. In the Cox data set, the data attest that the complete models including sewage and breastfeeding ( 1 & 5 ) are clearly better. The respective -2LL of these two models are very similar: 9071.912 and 9071.68. The prediction power of these models is very high» The models chisquare are all very significant, denoting that the null hypothesis that the value of the coefficients of the variables is equal to zero must be rejected. The independent variables do improve the predictability of the models and the odds that these results were obtained by chance are nil. The Cox regression models with the highest chisquare are models 5 and 1: 288.2 and 288.369, respectively. 434 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 151 Diagnostic Results, Predictive Efficiency and Further Characteristics of the Cox Regression Models for 1996-91-86 D ata S et (1 of 2) Item Stat. Model 1 Model 2 M odels Model 4 Events 524 524 535 535 Censored 8172 8172 8187 8187 (%) 94 94 93.9 93.9 -2Log Likelihood 9071.912 9076.266 9498.755 9501.753 Chisquare (Overall ) 288.2 283.465 181.286 177.846 (Signif) 0 0 0 0 (df) 11 10 10 9 N (numtjer of cases included ) 8696 8696 8722 8722 435 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 152 Diagnostic Results. Predictive Efficiency and Fudher C haracteristics of the Cox Regression Models for 1996-91-86 Data S et (2 of 2) M odels Model 6 M odel? M odels Item Stat. Events 524 524 535 535 Censored 8172 8172 8187 8187 (%) 94 94 93.9 93.9 -2Log Likelihood 9071.68 9076.064 9498.311 9501.288 Chisquare (Overall ) 288.369 283.56 181.428 177.983 (Signif) 0 0 0 0 (df) 13 12 12 11 N (number of cases included ) 8696 8696 8722 8722 436 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Cox Regression R esults for 1996-19^ 1-1986 M odels The Cox method presents results which are very similar to results derived through the logistic method. The sample size for the pooled 1996-1991- 1986 data set is the largest of all 12 data sets examined in this study. The Cox procedure takes mto consideration Üie highly important time- dependent breastfeeding variable. Cox’ s explanatory power is also expected to be higher. The 2 main Cox regression models for the 1996-1991-1986 data set (model 1 and 5) are comprised of 10 independent variables in addition to the birth cohort variable or variables. 2 regression models (model 2 and 6) include the same variables with the exception of sewage. 2 other models ( models 3 and 7) consist of the same variables with the exception of the duration of breastfeeding variable. The last 2 models (models 4 and 8) exclude both sewage and duration of breastfeeding variables, (tables 153 to 160) The variables ethnicity, DPT immunization and prenatal care by a doctor are not present in the 1991-1986 pooled data set since they were absent in the 1986 DHS. The regression results for the 8 1996-1991-1986 Cox models are similar to the logistic results and they indicate that the following independent variables have a statistically significant effect on infant mortality at the 95 percent level of confidence: breastfeeding, goods, mother’ s education, household crowding, and sex of the child. The birth cohort is also significant in most of the models. 437 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. According to this data set, the age group of Üie mother variable has a statistically significant hnpact on the dependent variable in all models which exclude the time varying duration of the duration of breastfeeding variable (models 3, 4, 7 and 8). In contrast, birth order risk is only s%nificant in the models in models 1, 2, 5 and 6. The sewage variable is statistically significant in the main models: 1 and 5. In model 1 (complete model with single bùrth cohort), the variables with the lowest p values are, respectively: 1 ) breastfeedmg (0.0); 2) household crowding (0.0); 3) goods (0.0); 4) mother’ s education (0.00); 5) sex of the child (0.0122); 6) sewage (0.0382); 7) birth order risk (0.0448). (table 153) In model 5 (complete model with multiple birth cohorts), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0); 2) household crowding (0.0); 3) goods (0.0); 4) mother’ s education (0.00); 5) sex of the child (0.0128); 6) sewage (0.0376); 7) birth order risk (0.044). (table 157) In model 2 (single birth cohort with no sewage variable), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0); 2) household crowding (0.0); 3) goods (0.0); 4) mother’ s education (0.00); 5) sex of the child (0.0113); 6) birth order risk (0.0376). (table 154) In model 6 (multiple birth cohorts with no sewage variable), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0); 2) household crowding (0.0); 3) goods (0.0); 4) mother’ s education (0.00); 5) s«c of the child (0.0118); 6) birth order risk (0.0369). (table 158) 438 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In model 3 (smgle birtii cohort with no breastfeeding)» the variables with the lowest p values are, respectwely: 1 ) birth cohort (0.0); 2) mother’ s education (0.0034); 3) household crowding (0.0055); 4) goods (0.0067); 5) sex of the chüd ( 0.037). (table 155) In model 7 (multiple birth cohorts with no breastfeeding), the variables with the lowest p values are, respectively: 1 ) goods (0.0); 2) mother’ s education (0.00); 3) household crowding (0.0033); 4) sex of the child (0.0035); 5) age group of mother (0.0088) 6) birth cohort (0 and 0.0111 respectively for 90-96 and 85-89 cohorts), (table 159) In model 4 (single birth cohort without sewage and breastfeeding), the variables with the lowest p values are, respectively: 1) birth cohort (0.0); 2) mother’ s education (0.0034); 3) goods (0.0); 4) household crowding (0.0032); 5) sex of the child ( 0.035). (table 156) In model 8 (multiple birth cohorts with no breastfeeding and no sewage), the variables with the lowest p values are, respectively: 1 ) goods (0.0); 2) mother’ s education (0.00); 3) sex of the child (0.0033); 4) household crowding (0.0039); 5) age group of mother (0.0091) 6) birth cohort (O.Ollland 0.0144 respectively for 90-96 and 85-89 cohorts), (table 160) With the exception of the household crowding variable, all other explanatory variables are inversely correlated with the dependent variable as it would be ejected. 439 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. An evaluation of the regression coefScients for the independent variables (taking model 5 as a reference) that have an effect on the dependent variable at the 0.05 level of significance indicates that when an additional household member is present the odds of dying for the infant increase in over 8% in comparison to what th ^ would be otherwise, (table 157) Similarly, and given the ünpact of other explanatory variables, the odds of dying for the infant child decrease in 71% as the mother improves her educational attainment firom one category to another, (table 157) As the household purchases one additional consumer good- either a radio, a TV or a car- the odds of dying for the family's duld are also reduced in 73% in comparison to what they would be if that consumer good was not available. These results are very M T nflar to the logistic regression results. In sum, the 8 1996-1991-1986 pooled Cox regression models indicate that the following variables have an impact on the likelihood of infant mortality at the 5% level of significance; breastfeeding, goods, mother’ s education, sex of the child and household crowding. Birth cohort, sewage and mother’ s age at the time of the birth are also statistically significant at the 0.05 level in some of the Cox regression models. 440 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 153 Cox Regression Results for 1996-91-86 Ne Data S et (1) Variable Name 8 Standard Error Exp(B) Significance (p) Birth Cohort -0.0289 0.0505 0.9715 0.5688 Urban X Rural -0.078 0-1174 0.925 0.5068 Drinking W ater -0.0063 0.1168 0.9937 0.9569 Sewage -0.2223 0.1073 0.8007 * 0.0382 Household Crowding 0.0798 0.0175 1.083 * 0 Age Group of Mother -0.0793 0.1027 0.9238 0.4399 Mother's Education -0.345 0.072 0.7082 * 0 Goods -0.3114 0.0595 0.7324 * 0 Birth Order Risk -0.2088 0.1041 0.8116 * 0.0448 Sex of Child -0.2217 0.0885 0.8012 * 0.0122 Breastfeeding in Months -0.2563 0.0282 0.7739 * 0 Source; 1996,1991 and 1986 DHS for Brazil's Northeast region from a sub sam ple total of 8753 cases (8696 selected. 57 dropped). 441 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 154 Cox Regression R esults for 1996-91-86 Ne Data Set (2) Variable Name B Standard Error Exp(B) Significance (p) Birth Cohort -0.0207 0.0504 0.9796 0.6818 Urban X Rural -0.1137 0.1163 0.8925 0.328 Drinking W ater -0.0399 0.116 0.9609 0.731 Household Crowding 0.0775 0.0174 1.0806 0 Age Group of Mother -0.0827 0.1028 0.9207 0.4212 M other's Education -0.346 0.0721 0.7075 0 Goods -0.3403 0.0578 0.7115 0 Birth O rder Risk -0.216 0.1039 0.8057 0.0376 Sex of Child -0.224 0.0885 0.7993 * 0.0113 Breastfeeding in Months -0.2561 0.0283 0.774 * 0 Source: 1996,1991 and 1986 DHS for Brazil's N ortheast region from a sul> sam ple total of 8753 cases (8696 selected. 57 dropped). 442 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 155 Variable Name Cox Regression R esults for 1996-91-86 Ne Data S e t (3) B Standard Error Exp (B) Significance (p) Cohort -0.222 0.0477 0.8009 * 0 U rtan X Rural -0.0204 0.1162 0.9798 0.8607 Drinking W ater -0.0031 0-1143 0.9969 0.9786 Sewage -0.1842 0.107 0.8318 0.0852 Household Crowding 0.0528 0.0175 1.0542 * 0.0026 Age Group of Mother -0.2576 0.1001 0.7729 0.01 Mother's Education -0.3417 0.0717 0.7105 0 Goods -0.2554 0.0589 0.7746 * 0 Birth Order Risk -0.1771 0.1029 0.8377 0.0854 Sex of Child -0.2542 0.0875 0.7756 * 0.0037 Source; 1996,1991 and 1986 DHS for Brazil's Northeast region from a sub sam ple total of 8753 cases (8722 selected, 31 dropped). 443 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 156 Cox Regression Results for 1996-91-86 Ne Data S et (4) Variable Name B Standard Error ExpCB) Significance (p) Cohort -0.2156 0.0476 0.806 0 Urban X Rural -0.052 0.1149 0.9493 0.6506 Drinking W ater -0.0286 0-1136 0.9719 0.8015 Household Crowding 0.0517 0.0175 1.0531 0.0032 Age Group of Mother -0.257 0.1002 0.7734 * 0.0103 Mother's Education -0.343 0.0717 0.7096 • 0 Goods -0.2785 0.0574 0.7569 * 0 Birth Order Risk -0.1847 0.1028 0.8314 0.0724 Sex of Child -0.2555 0.0875 0.7745 0.0035 Source: 1996,1991 and 1986 DHS for Brazil's Northeast region from a sub sam ple total of 8753 cases (8722 selected, 31 dropped). 4 4 4 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 157 Variable Name Cox Regression R esults fo r 1996-91-86 Ne Data S e t (5) B Standard E nor Exp (8) Significance (p) Cohort 90-96 -0.1089 0.1613 0.8969 0.4997 Cohort 85-89 -0.0972 0.1447 0.9074 0.5018 Cohort 80-84 -0.0924 0.1433 0.9117 0.5191 Urban X Rural -0.0804 0.1176 0.9228 0.4942 Drinking W ater -0.0054 0.1169 0.9946 0.9633 Sew age -0.2231 0.1073 0.8 * 0.0376 Household Crowding 0.0803 0.0175 1.0836 » 0 Age Group of Mother -0.0786 0.1028 0.9244 0.4447 Mother's Education -0.3454 0.072 0.7079 * 0 Goods -0.3109 0.0595 0.7328 0 Birth O rder Risk -0.2095 0.104 0.811 0.044 Sex of Child -0.2205 0.0885 0.8022 0.0128 Breastfeeding -0.2575 0.0285 0.773 0 in Months Source: 1996,1991 and 1986 DHS for Brazil's N ortheast region from a sub sam ple total of 8753 cases (8696 selected, 57 dropped). 445 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 158 Cox R egression Results for 1996-91-86 Ne Data S et (6) Variable Name B Standard Error Exp(B) Significance (p) Cohort 90-96 -0.0833 0.161 0.9201 0.6049 Cohort 85-89 -0.0765 0.1446 0.9264 0.5968 Cohort 80-84 -0.0804 0.1432 0.9227 0.5744 Urban X Rural -0.116 0.1164 0.8905 0.319 Drinking W ater -0.0392 0.116 0.9616 0.7357 Household Crowding 0.078 0.0175 1.0811 0 Age Group of Mother -0.082 0.1029 0.9213 0.4256 M other's Education -0.3463 0.0721 0.7073 0 Goods -0.34 0.0578 0.7118 0 Birth O rder Risk -0.2168 0.1039 0.8051 0.0369 Sex of Child -0.2229 0.0885 0.8002 0.0118 Breastfeeding in Months -0.2572 0.0285 0.7732 0 Source: 1996,1991 and 1986 DHS for Brazil's Northeast region from a sub sam ple total of 8753 cases (8696 selected, 57 dropped). 446 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 159 Cox Regression R esults for 1996-91-86 Ne Data S e t (7) Variable Name B Standard Error Exp(B) Significance (p) Cohort 90-96 -0.6517 0.1563 0.5212 0 Cohort 85-39 -0.3709 0.146 0.6901 0.0111 Cohort 80-34 -0.1647 0.144 0.8482 0.2527 Urban X Rural -0.0178 0.1163 0.9823 0.8781 Drinking W ater -0.0032 0.1143 0.9968 0.9777 Sewage -0.1837 0.1071 0.8322 0.0862 Household Crowding 0.0519 0.0176 1.0532 0.0033 Age Group of Mother -0.2619 0.1 0.7696 0.0088 Mother's Education -0.34 0.0718 0.7118 0 Goods -0.2557 0.0589 0.7744 0 Birth O rder Risk -0.1767 0.103 0.838 0.0861 Sex of Child -0.2562 0.0876 0.774 0.0035 Source: 1996,1991 and 1986 DHS for Brazil's Northeast region from a sub sam ple total of 8753 cases (8722 selected. 31 dropped). 447 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 160 Cox Regression Results for 1996-91-86 Ne Data S et (8) Variable Name 8 Standard Error Exp (8) Significance (p) Cohort 90-96 -0.6284 0.1559 0.5335 0.0001 Cohort 85-89 -0.3572 0.1459 0.6996 0.0144 Cohort 80-84 -0.1494 0.1437 0.8613 0.2986 Urban X Rural -0.0493 0.115 0.9519 0.668 Drinking W ater -0.0287 0.1136 0.9717 0.8008 Household Crowding 0.0507 0.0176 1.0521 0.0039 Age Group of Mother -0.2614 0.1002 0.77 0.0091 Mother's Education -0.3414 0.0718 0.7108 0 Goods -0.2788 0.0574 0.7567 0 Birth Order Risk -0.1841 .1028 . 8318 0.0734 Sex of Child -0.2576 0.0876 0.7729 0.0033 Source: 1996,1991 and 1986 DHS for Brazil's Northeast region from a sub sam ple total of 8753 cases (8722 selected, 31 dropped). 448 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. R esults o f D escriptive and M ultivariate Regression A nalysis Applied to 1996-1991 Pooled Logistic Reizression and Cox D ata S e ts Descriptive Analysis The 1996-1991 pooled data sets well represent the recent evolution of infant mortality and its determinants in Brazil’ s Northeast region. The samples sizes are 6481 and 7633. The statistic results of the descriptive analysis indicate that all the selected independent variables when examined separately have shown to have an impact on infant mortality. According to the logistic regression data set (6481 cases), the birth cohort variable shows a steady decline in the odds of death as we move from children bom before 1979 to the other birth cohorts (1980-84, 1985-90 and 1990-95). 10% of the children bom before 1979 (12% of the sample) did not survive. 7.8% of the children bom between 1980 and 1984 (19.9% of sub sample) died; 6% between 1985 and 1989 ( 37% of sub-sample) and 4.7% in the latest birth cohort period ( 31.1% of sub-sample). The birth cohort variable takes on values from 0 to 3, die former being the oldest cohort, and the latter the newest. According to the Cox data set (7633 cases), the means for the birth cohort categories are exactly the same as the logistic regression ones, except for the most recent cohort (1990-1996) which has a slightly higher mean or 4.8%. The 1990-1996 birth cohort represents 41.5% of the Cox sub-sample and included 3167 cases or 1552 more cases than the logistic regression one. 449 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The place of residence variable for the logistic repression data set shows that 5.8% of the children bom in urban areas (71.2% of sub-sample) and 8.1% of the ones bom in rural areas died. This variable takes the value 0 for rural and 1 for urban residence. According to the Cox data set, which puts more weight on recent cases, the percentage of deceased children is sli^ tly lower, or 5.5 and 7.8%, respectively. 72.3% of the households had access to good drinking water in the logistic set and 71% in the Cox data set. 5.9% of the children bom in households with clean drinking water did not survive, whereas 8% of those with no access to good drinking water lost their lives. The values are 0 for not good for drinking and 1 for water good for drinking. The odds of dying in the Cox data set are similar; 5.7 and 7.6 %, respectively. The next explanatory variable is sewage or the type of toilet facility. This variable takes the value 0 for not modem and 1 for modem sewage. Note that in 1991 the percentage of households with modem sewage according to the DHS was surprisingly high. The 1986 and 1996 levels of modem sewage in the DHS's and in our sub-samples as well as the ofScial data by the Brazilian Institute of Geography and Statistics indicate a significantly lower level of adequate toilet facility. 450 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. According to the Cox data sets for 1986, 1991 and 1996, 11.6, 61 and 23.1% of the respective sub-sample households were equipped with modem sewage. The pooled 96-91 data sets for Cox and logistic regressions indicate that respectively 42 and 43.3% of the households had modem sewage disposal. 5.4% of the children in households with modem toilet facility and 7.2% in the ones with no modem toilet facility did not survive in the logistic regression formatted data set. In the Cox data set, these rates are slightly lower, 5.3 and 6.9 %, For social-behavioral and physiological reasons, infant mortality rates of children of young mothers (19 years old and under) as well as of older mothers (over 35 years of age ), have been frequently pointed out as being higher than those of women age 20 to 34. According to the logistic regression data set descriptive results, the mortality risks associated with the high risk age group are indeed fairly high: 8.3% of the children of mothers at a high risk age group die as opposed to 4.3% of the children of mothers at their reproductive peak (low risk). The Cox regression data set indicates similar rates, or 8.3 and 4.3%. The household crowding variable seems to be more significant in the 1996-91 pooled data set than in previous data years. The dichotomized crowding variable for the logistic and Cox data sets shows that respectively 42.8 and 43.1% of the children lived in crowded dwellings (5 or more). 451 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The means distribution for each total household number shows a clear and definite progression for the mean value of deceased children per household number. In 2 to 4 individuals households, on average 5/6% of the latest children of each respondent die before their first year of life. As we increase the number of residents firom 5 to 7%, the mean values for the odds of death grow firom 7.3 to 7.5 and to 8.3%, respectively. In households with 8 or more individuals, the percentage of children who made the transition (who did not survive) increases to almost 10%. The proportional hazards regression results derived from the Cox data set indicates a very similar statistic progression. The next variable is the mother’ s educational attainment. This variable is divided into 4 categories: higher education (college level and over); secondary (high school level completed); primary (including primary complete and secondary incomplete, 1 to 11 years); no education. According to the 1996-1991 data sets for LR and Cox analysis, only 4% and 3.8%, respectively of the Northeast respondents had higher education. Among those mothers with higher education, .8% and 1% of their youngest infants died according to the logistic and Cox regression results. The odds of survival improve substantially as mothers become more educated. 11% of the children of women with no education do not survive m both the logistic and the Cox data sets (comprising 18.1 and 17.8% of the total samples respectively). 452 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. As women acquire at least 5 years of schooling (primary complete) the odds of dying decrease to 6.6 and 6.9% for each data set. For the 27% of the mothers who completed secondary school (12 years of schooling) the percentage of deceased children fall to 3.7 and 3.6% in the logistic and Cox formatted data sets. In around 70% of the cases, die educational attainment is limited to primary or no education at all. The variable goods is a p ro xy for household income and wealth. It inquiries about ownership of 3 durable consumer goods: car, TV and radio. For each positive answer, one point was assigned. The values range from 0 to 3. Changes from I to 2, or firom 2 to 3 goods seem to have a strong effect on the odds of survival. In contrast, the change fi* om 0 to I, typically the existence of a radio in the household, does not seem to have a profound impact as far as the descriptive results are concerned. 14.1% of the households in the logistic data set and 13% in the Cox data set had all 3 goods. The level of deceased children for such mothers is only 2.9%. With no goods or with ju st 1, these odds are 8.2 and 8.8% in the logistic and 8% ( 0 and 1 ) in the Cox data set. 38.3% of the mothers in the logistic data set and 36.7% in the Cox data set had 2 of these goods at home. The respective percentages of deceased children are 5.1 and 5%. 453 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The next variable is the infant’ s birth order risk. 4.6 and 7.8% of the children who, respectively, have a low (2“* or 3 * ^ child) and high order risk (1« and over 3” < ) die in the logistic regression data set. The values for the Cox data set are very close: 4.5 and 7.5%. Female infant mortality is usually much higher than male infant mortality in Brazil. The descriptive results of the data sets indicate that 5.7% of the girls and 7.2% of the boys died in the logistic data set. In the Cox data set, these rates are 5.5% and 6.9%, respectively. This study hypothesizes that prenatal care by a physician is one of the independent variables that have the greatest impact on infant mortality. A passive strategy was implemented to deal with mcoherent answers or missing data in preparation for this variable, meaning that "don’ t know” answers were associated with 0 or no access to prenatal care by a doctor (a similar procedure was carried out with the immunization variable). 69.9% of the respondents in the logistic data set and 63.4% in the Cox data set did not see a physician for the prenatal care of the infant. This differential between the sets could be ascribed to the possibility Üiat access to a doctor has been greatly facilitated in recent years (the Cox data set includes the children bom most recently, or between March of 1995 and March of 1996). 454 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The percentage of deceased children who did not have access to prenatal care by a doctor in both data sets is the same : 7.9%. Only 2.9 and 3% of the children who were under the prenatal care of a doctor did not survive, according to the logistic and Cox data sets. The odds of death for infants of white heritage is 4.7% in the logistic and 4.8% in the Cox data sets. In non-white households such respective rates are 6.9 and 6.7%. Three quarters of the respondents are not white. The last independent variable is immunization or whether the infant was vaccinated for aU three DPT shots. ^ positive the value assigned was 1, and 0 otherwise. According to the descriptive results, there is a strong correlation between immunization status and infant mortality. Around 68% of the children did not receive all three shots of the DPT vaccine in both data sets. The percentage of deceased among the ones who did receive all three shots is 0.3% in the logistic and 2.9% in the Cox data set. For some reason the positive impact of immunization was reduced in the last Cox survey year (March 95-96). The mean value of deceased children among the ones not immunized was 9.4 and 9%, respectively in the logistic and in the Cox data sets. The Cox regression data set endorses the importance of breastfeeding for the life of the child. 56.2% of the infants or 4,286 cases were not breast-fed at all. 8.8% of these children did not survive. 5.6% of the children were breast-fed for just 1 month, 6% for 2 and 6.1% for three months. 455 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Only 1.6% of the children were breast-fed for the entire 12 months. The mean value of deceased children per months of b r e a s t f e e d i n g s decrease in a progressive manner: 8.8 % (0), 5.8% (1), 5.1% (2), 3 % (3), 2% (4) and so on. According to the 1996-1991 pooled Cox and logistic data sets, all variables included in the models, with the exception of sex, have Pearson’ s correlation coefGcients th at are statistically significant at the 99% level of confidence suggesting a very low probability that the statistic results could be attributed to chance (table 163). The correlation coefficient associated with the sewage variable in the Cox data set is also only significant at the 95% level of confidence (0.032). The dichotomous household crowding variable is the only covariate that has a positive linear relation with the dependent variable: 0.046 and 0.049, respectively in the logistic and Cox data sets. The independent variables with the strongest level of correlation are: Immunization levels ( -0.173 and -0.168, respectively in the logistic and Cox data sets); duration of breastfeeding ( -0.107); educational attainment (-0.105 in the logistic and -0.101 in the Cox data set); pre-natal care ( -0.093 and - 0.098); mother’ s age risk (-0.082 in both data sets) and goods ( -0.08 and - 0.072)). 456 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 161 . i n U ‘ r . \ ) f u r . // r n 11 ' f u ' r u ! >■'r i . i ' :. ) r I ’ ' ^ ^ 1 ^ ^ 6 - V ; c n / ' s ^ ' ; x/rV'M Yyyô 457 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 458 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. y - - r . ' \ ' ' ,z ' 459 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 162 . 'V / c J / 'î S ‘ J . ' i 460 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. /V’e . i / ’ s .}'uf ‘ ' r - ; '‘O ' ii: ' ’ <- n : . r / j o / » ' " ^ o' * " - Q T ^ %'•" ( o x • ^ o c / ' - ‘ s s / o o j 461 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ■ V Z p . î d s . ? ■ ■ i. r < • p * - ■ ■. r • v ; . i t - ‘ rv-i' -i r.' iS K 462 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. /Vit’. i n s . I f s / : l Ê m m 463 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 163 1996-1991 Correlation Levels with D ependent Variable Logistic Regression Data S et Cox Regression Data S et Variables Cohort* -0.72 -0.71 Urban X Rural -0.042 -0.044 Drinking W ater -0.039 -0.037 Sewage -0.037 0.032* Household Crowding* 0.046 0.049 Household Crowding (B) -0.047 -0.05 Mother's Age Risk -0.082 -0.082 Educational Level* -0.105 -0.101 Goods* -0.08 -0.072 Birth Order Risk -0.64 -0.062 Sex of Child -0.031 * -0.028 * Breastfeeding** N/A -0.107 Ethnicity* 0.026* 0.027* Ethnicity (B) -0.034 -0.035 Dr's Prenatal Care -0.093 -0.098 Immunization (DPT123) -0.173 -0.168 * non djchotomous * * in months A N variables Significant at the 0.01 level (2 tailed) unless mdkatedothefwiBe * Significant at the 0.06 level ( 2-talsd) XNotConaWsd 464 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. D iagnostic R esults for 1 9 96 -1 9 9 1 LR M odels The diagnostics statistics reveal that all models fît the facts veiy weU. (table 164.) The -2 Log Likelihood statistic, the analogue of the Sum of the Square Errors (SSE) in OLS regression, expresses how poorly the models fit with all the included variables. The larger the -2LL, the worse is relatively the prediction power for the dependent variable. In the logistic data set, the data indicate Üiat the model with the best prediction power for the dependent variable is model 3, the one with the multiple birth cohort variables and the sewage variable. This model has the highest number of degrees of fi^edom (15) and the lowest -2LL (2648.01). The models chisquare indicate that the null hypothesis that the value of the coefficients of the variables is equal to zero should be rejected. All models are very significant. Model 3 also has the highest model chisquare (401.382). All models have excellent goodness of fit. 6430 cases were selected and 51 missing. The prediction efficiency of the models is high: 93.62%. 465 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 164 Diagnostic Results, Predictive Efficiency and Further C haracteristics of the Logistic R egression Models for 1996-91 Pooled Data S et Item Stat. Model 1 Model 2 Model 3 Model 4 -2Log Likelihood 2652-701 2654.046 2648.01 2649.253 Goodness of Fit 6465.717 6428.957 6467.264 6426.237 Model Chisquare 397.691 396.346 402.382 401.139 (Signif) 0 0 0 0 (df) 13 12 15 14 Predictive Efficiency 93.62% 93.62% 93.62% 93.62% N (number of cases included ) 6430 6430 6430 6430 466 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Logistic Regression. R esu lts for 1996-1991 M odels The 2 main logistic regression models for the pooled 1996-1991 data, set (model 1 and 3) are comprised of 12 independent variables plus the birth cohort variable or variables. The other 2 regression models (model 2 and 4) include the very same regressions widiout the important sewage variable, (tables 165 to 168) Among the 3 pooled logistic data sets, the 1996-1991 data set, derived firom the 1996 and the 1991 DHS and constituted of a sub-sample size o f6481 children, is perhaps the one with the highest explanatory power. Notice that the 9 out of the 12 explanatory variables are coded in a dichotomous fashion. The value 0 is associated with the category which is supposed to have a negative impact on infant mortality and the value 1 related to the opposite expectation. The 3 non binary variables are; mother’ s education (4 categories ranging firom 0 - no education- to 3 - higher education); goods (4 categories ranging firom 0 - none of the 3 goods - to 3 - ownership of radio, T V and car ); household crowding (n). The regression results for the 4 models indicate that the independent variables that have a statistically significant effect on infant mortality a t the 0.05 level of significance are: mother’ s education, goods, household crowding, prenatal care by a doctor, immunization and sex of the infant. 467 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In model 1 (complete model with smgie birth cohort), the variables with the lowest p values are, respectively: 1 ) Dr’ s prenatal care (0.0); 2) immunization (0.0); 3) mother’ s education (0.0001); 4) goods (0.0002); 5) sex of child (0.0091); 6) household crowdmg (0.0125). (table 165) In model 3 (complete model with multiple birth cohorts), the variables with the lowest p values are, respectwely: 1 ) immunization (0.0); 2) Dr’ s prenatal care (0.0001); 3) mother’ s education (0.0001); 4) goods (0.0002); 5) household crowdmg (0.0089); 6) sex of child (0.0115); 7) birth cohort (0.0304 and 0.0191 respectively for 85-89 and 80-84). (table 167) In model 2 (multiple büth cohorts with no sewage variable), the variables with the lowest p values are, respectively: 1 ) immunization (0.0); 2) Dr’ s prenatal care (0.0); 3) goods (0.0); 4) mother’ s education (0.0001); 5) sex of child (0.0091); 6) household crowding (0.0167). (table 166) In model 4 (sin^e birth cohort with no sewage variable), the variables with the lowest p values are, respectively: 1 ) immunization (0.0); 2) mother’ s education (0.0001); 3) goods (0.0001); 4) Dr’ s prenatal care (0.0002); 5) sex of child (0.0115); 6) household crowding (0.0118); 7) birth cohort (0.0458 and 0.0238 respectively for 85-89 and 80-84). (table 168) It is interesting to see that, according to this data set, the sex of the child is an important determinant of the odds of survival. 468 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Sewage is not a significant explanatory variable, but this result may have been afiected by the oddly high level of modem sewage in the 1991 data sets. An analysis of the B coefEcients’ signs in all 4 models would indicate that household crowding and prenatal care are posithrely correlated with mfant mortality, implying that better prenatal care by a physician and crowding Imng conditions would increase the likelihood of death for the infant. The odds ratios for the three variables (greater than 1 ) would indicate the same positive relationship between the explanatory variables and the dependent variable. Both variables are statistically significant. The results for the household crowding variable do fit the facts well. The results for prenatal care would contradict the literature as well as other statistical results examined in this study. On the other hand, if access to prenatal care by a doctor has a negative statistical impact on infant mortality, the finding would corroborate the results achieved by Chen while investigating infant mortality in the Los Angeles Co. The conclusion reached here after all the regressions were run without the DPT immunization variable is the same as the one reached in all other data sets in which both variables were included: the odds ratios and B coefficients for the prenatal variable are affected by some degree of collinearity with immunization. When the latter is excluded and taking model 1 as a reference. ^ Xue Hong Chen, Ibid., 1997. 469 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the standard error for the prenatal care variable does not change much. Nor does the significance level (changing horn 0 to 0.0001). The unstandardized regression coefficient and the odds ratio, however, change to -.6613 (horn .7535) and to .5136 (from 2.1243). The direct correlation between prenatal care and the dependent variable shown in table 7.91.was indeed caused by the presence of the DPT immunization variable and do not reflect the facts. Thus, access to prenatal care by a doctor does unprove the odds of survival. An analysis of the regression coefficients for the independent variables which are significant at the 0.05 level in model 1 indicates that the odds of dying for the infant child decrease in 73% as a mother improves her educational attainment firom one category to another, given the impact of other explanatory variables, (table 165) As the household acquires one additional consumer good, a radio, a TV or a car, the odds of dying for the family’ s child is also reduced in 75% in comparison to what they would be i f that consumer good was not available, (table 165) According to the 1996-1991 pooled logistic regression models, the following variables have an effect on the likelihood of infant mortality a t the 5% level of significance: immunization, goods, mother’ s education, household crowding, s ^ of the child as well as prenatal care by a doctor (once DPT immunization is controlled). 470 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 165 Logistic Regression R esults for 1996-1961 Ne Data S et (1) Variable Name B Standard Error Exp(B) Significance (p) Birth Cohort -0.0867 0.0678 0.917 0.2013 Urisan X Rural 0.0439 0.1451 1.0449 0.762 Drinking W ater -7.30E-05 0.1428 0.9999 1 Sew age -0.1466 0.1265 0.8637 0.2467 Household Crowding 0.0578 0.0231 1.0595 0.0125 Age Group of Mother -0.148 0.1317 0.8624 0.2609 Ethnicity -0.1277 0.133 0.8801 0.3372 M other's Education -0.3167 0.083 0.7286 * 0.0001 Goods -0.271 0.0717 0.7626 * 0.0002 Birth Order Risk -0.214 0.1235 0.8074 0.0832 Dr's Prenatal care 0.7725 0.18 2.1651 * 0 S ex of Child -0.2773 0.1064 0.7578 * 0.0091 DPT immz. -3.9697 0.4287 0.0189 * 0 Constant -1.5579 0.2005 - * 0 Source: 1996 & 1991 DHS for Brazil's N ortheast region from a sub-sam ple total of 6481 cases (6430 selected. 51 m issing). 471 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 166 Logistic Regression Results for 1996-1991 Ne Data Set (2) Variable Name B Standard Error Exp(B) Significance (jp) Birth Cohort -0.0682 0.0659 0.9341 0.3008 Urban X Rural 0.0193 0-1434 1.0195 0.8931 Drinking W ater -0.0236 0.1414 0.9767 0.8673 Household Crowding 0.055 0.023 1.0566 0.0167 Age Group of Mother -0.1597 0.1311 0.8524 0.2234 Ethnicity -0.1291 0.133 0.8789 0.3318 Mother's Education -0.3159 0.0831 0.7291 0.0001 Goods -0.2884 0.0701 0.7494 * - 0 Birth Order Risk -0.2178 0.1234 0.8043 0.0777 Dr's Prenatal care 0.7535 0.1791 2.1243 * 0 Sex of Child -0.2775 0.1063 0.7577 * 0.0091 DPT Immz. -3.9701 0.4287 0.0189 * 0 Constant -1.4884 0.2006 - 0 Source: 1996 & 1991 DHS for Brazil's Northeast region from a sub-sample total of 6481 cases (6430 selected, 51 missing). 472 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 167 Logistic Regression Results for 1996-1991 Ne D ata S e t (3) Variable Name B Standard Error Exp(B) Significance (p) Cohort 90-95 -0.2753 0.2065 0.7594 0.1825 Cohort 85-89 -0.361 0.1667 0-697 * ■ 0.0304 Cohort 80-84 -0.3798 0.1621 0.684 * ■ 0.0191 Urban X Rural 0.0435 0.1452 1.0445 0.7644 Drinking W ater 0.0022 0.1429 1.0022 0.9879 Sewage -0.1415 0.127 0.8681 • 0.2654 Household Crowding 0.0605 0.0231 1.0624 0.0089 Age Group of Mother -0.1524 0-1327 0.8587 0.2509 Ethnicity -0.1318 0.1331 0.8765 0.3221 Mother's Education -0.3183 0.083 0.7274 0.0001 Goods -0.2654 0.0719 0.7669 0.0002 Birth Order Risk -0.2143 0.1235 0.8071 0.0826 Dr's Prenatal care 0.6987 0.183 2.0111 0.0001 Sex of Child -0.269 0.1065 0.7641 0.0115 DPT Immz. -3.9794 0.4283 0.0187 * 0 Constant -1.345 0.2073 - * 0 Source: 1996 & 1991 DHS for Brazil's N ortheast region from a sub-sam ple total of 6481 cases (6430 selected, 51 missing). 473 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 168 Logistic Regression Results fo r 1996-1991 Ne Data S et (4) Variable Name B Standard Error Exp(B) Significance (p) Cohort 90-95 -0.2225 0.2008 0.8005 0.2678 Cohort 85-89 -0.3275 0-164 0.7207 * 0.0458 Cohort 80-84 -0.3651 0.1615 0.6941 * 0.0238 Urban X Rural 0.0201 0.1435 1.0203 0.8887 Drinking W ater -0.0206 0.1415 0.9796 0.884 Household Crowding 0.0578 0.023 1.0595 * 0.0118 Age Group of Mother -0.1632 0.1322 0.8494 0.2171 Ethnicity -0.1332 0.133 0.8753 0.3167 Mother's Education -0.3177 0.0831 0.7278 • 0.0001 Goods -0.282 0.0703 0.7543 * 0.0001 Birth O rder Risk -0.2183 0.1234 0.8039 0.0768 Dr's Prenatal care 0.6805 0.182 1.9748 * 0.0002 Sex of Child -0.2691 0.1065 0.7641 * 0.0115 DPT Immz. -3.9794 0.4283 0.0187 * 0 Constant -1.354 0.2074 - 0 Source; 1996 & 1991 DHS for Brazil's Northeast region from a sub-sam ple total of 6481 cases (6430 selected, 51 missing). 474 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. D iagnostic Results for 1 9 9 6-1 9 9 1 Cox M odels The diagnostics statistics reveal that all 8 Cox regression models fît the facts very well, (table 79) Among them, the models that seem to predict infant m o rta lity better are the ones including all independent variables (1 and 5). One of the main shortcomings of the logistic model is that it does not take into account time-varying variables such as duration of breastfeeding in months. The Cox models seem to have a more powerful explanatory power since they include all the logistic regression variables in addition to breastfeeding status. Eivent history or survival analysis does not ignore the riming of the event (death of the infant). Cox proportional risks method of partial likelihood defines a risk set and censors the data for the occurrence of events (the qualitative transition change firom life to death). In the 1996-1991 pooled Cox models with no breastfeeding variable ( models 3, 4 , 7 and 8), 39 cases were dropped and 7594 selected. 94.4% of these children or 7171 cases are censored and survive the first year of life and 423 infants do make the transition. In the models including the duration of breastfeeding in months (models 1, 2, 5 and 6), 51 cases were dropped and 7582 selected. The same percentage of cases, 94.4% or 7160 children, are censored whereas 422 children do make the transition. In the Cox data set, the data indicate that the complete models including sewage and breastfeeding ( 1 & 5 ) are clearly the best models. 475 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The respective -2LL for these two models are 7045.164 and 7050.119. Overall, the best prediction power for the dependent variable is found m . model 5 with the multiple birth cohort variables, the one with the highest number of degrees of fireedom and the lowest -2LL. The model chisquare or Gm is similar to the F test in Imear regression and it is the difference between the initial log-lücelihood (D O ) and the -2LL for the model (Dm). The models chisquare are very significant, indicating that the nuU hypothesis that the value of die coefficients of the variables is equal to zero should be rejected. The independent variables do improve the predictabflily of the models. The Cox regression models with the highest chisquare are a ^ in models 5 and 1: 324.381 and 320.801, respectively. 476 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 169 Diagnostic Results, Predictive Efficiency and Further Characteristics o f th e Cox Regression Models for 1996-91 D ataS et (1 of2) Item Stat. Model 1 Model 2 Model 3 Model 4 Events 422 422 423 423 Censored 7160 7160 7171 7171 (%) 94.4 94.4 94.4 94.4 -2Log Likelihood 7050.119 7051.031 7138.316 7138-795 Chisquare (Overall ) 320.801 319.466 297.373 296.688 (Signif) 0 0 0 0 (df) 14 13 13 12 N (number of cases included ) 7582 7582 7594 7594 477 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 170 Diagnostic Results. Predictive Efficiency and Further Characteristics of the Cox Regression Models for 1996-91 D ata S e t (2 o f 2) M odels Model 6 M odel? M odels Item Stat. Events 422 422 423 423 Censored 7160 7160 7171 7171 (%) 94.4 94.4 94.4 94.4 -2Log Likelihood 7045.164 7046.057 7135.998 7136.435 Chisquare (Overall ) 324.381 323-111 300.288 299.667 (Signif) 0 0 0 0 (df) 16 15 15 14 N (number of cases included ) 7582 7582 7594 7594 478 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Cox Regression. R esults for 1 9 9 6 -19 9 1 Models Cox’ s proportional hazards method allows for the mdusion of the extremely important time-dependent breastfeeding variable. The results for the regression method should be sim ila r but with a relatively higher explanatory power than the logistic regression results. The 2 main Cox regression models for 1996-1991 (model 1 and 5) are comprised of 13 independent variables plus the birth cohort variable or variables. 2 regression models (model 2 and 6) include the veiy same variables with the exception of sewage and 2 others ( models 3 and 7) consist of the same models without the duration of breastfeeding variable. The last 2 models (models 4 and 8) exclude both sewage and breastfeeding, (tables 171 to 178) Among the 3 pooled Cox data sets, the 1996-1991 data set, derwed G ro m the 1996 and 1991 DHS in Brazil (1996), and constituted, of a sub-sample size of 7633 children, is perhaps the most representative sample of all data sets. The regression results for the 8 models indicate that the independent variables that have a statistically significant effect on infant mortality are: breastfeeding, DPT immunization, mother’ s education, goods, household crowding, sex of the child as well as prenatal care by a doctor. Sex of the child has a rather important effect on the infant’ s odds of survival. In contrast, sewage is not a significant explanatory variable, but this 479 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. result may be due to the fact that the level of modem sewage in. the 1991 data sets was exceptionally high. In model 1 (complete model with single birth cohort), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0); 2) immunization (0.0); 3) mother's education (0.0); 4) goods (0.0); 5) sex of child (0.0); 6) Dr's prenatal care (0.004); 7) household crowding (0.0027). (table 171) In model 5 (complete model with multiple birth cohorts), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0); 2) immunization (0.0); 3) mother's education (0.0); 4) goods (0.0); 5) household crowding (0.0019); 6) Dr's prenatal care (0.0023); 7) sex of child (0.0126). (table 175) In model 2 (single birth cohort with no sewage variable), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0); 2) immunization (0.0); 3) goods (0.0); 4) mother's education (0.0001); 5) Dr's prenatal care (0.0005); 6) household crowding (0.0034) 7) sex of child (0.0101). (table 172) In model 6 (multiple birth cohorts with no sewage variable), the variables with the lowest p values are, respectively: 1 ) breastfeeding (0.0); 2) immunization (0.0); 3) goods (0.0); 4) mother's education (0.0001); 5) household crowding (0.0024); 6) Dr's prenatal care (0.003); 7) sex of child (0.0122). (table 176) 480 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In model 3 (single birth cohort with no breastfeeding), the variables with the lowest p values are, respectively: 1 ) immunization (0.0); 2) mother*s education (0.0001); 3) goods (0.0002); 4) s«c of child (0.0097); 5) household crowding (0.0129). (table 173) In model 7 (multiple bûrth cohorts with no breastfeeding), the variables with the lowest p values are, respectively: 1 ) immunization (0.0); 2) mother’ s education (0.0001); 3) goods (0.0002); 4) sex of child (0.0103); 6) household crowding (0.0113). (table 177.) In model 4 (single birth cohort without sewage and breastfeeding), the variables with the lowest p values are, respectively: 1 ) immunization (0.0); 2) mother’ s education (0.0001); 3) goods (0.0001); 4) sex of child (0.0096); 5) household crowding (0.0148). (table 174) In model 8 (multiple birth cohorts without sewage and breastfeeding), the variables with the lowest p values are, respectwely: 1 ) immunization (0.0); 2) mother’ s education (0.0001); 3) goods (0.0001); 4) sex of child (0.0102); 5) household crowding (0.0128). (table 178) An analysis of the B coefScients’ signs in all 8 Cox regression models would indicate that drinking water, household crowding and prenatal care are positively correlated with in fan t m ortali^ meaning that better water, prenatal care by a physician and crowding living conditions would increase the likelihood of death for the in fan t. 481 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The odds ratios for the three variables (greater than 1 ) would indicate the same positive relationship between the explanatory variables and the dependent variable. The results for the household crowding variable do fit the facts well. The results for water (even if one accepts that negative hnpact of better drinking water on weaning) and prenatal care would seem surprising and unfittmg, though. Since drinking water is not statistically significant, die results could be ascribed to sampling error and chance. The same conclusion reached m the 1996 Cox data set is valid here: prenatal care by a doctor seems to be positively correlated with infant mortality and not significant when the breastfeeding variable is not included in the regression models (models 3, 4, 7 and 8) due to a certain degree of collinearity between prenatal care and DPT immunization. A ll the regressions using the proportional risks method were run again without the immunization variable, and not only was prenatal care by a doctor always significant but also inversely correlated with the dependent variable. The high level of negative bwariate correlation between these 2 variables shown by table 7.89. would endorse this result. One may infer that access to prenatal care with a doctor improves the odds of survival and does have a statistically significant impact on infant mortali^ in all 8 models. 482 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. An analysis of the regression coefficients for the independent variables which are significant at the 0.05 level m model 1 indicates that the odds of dying for the infant child decrease in 73% as a mother improves her educational attainment firom one category to another, given the impact of other explanatory variables, (table 171) As the household acquires one additional consumer good, a radio, a T V or a car, the odds of dying for the family's child is also reduced in 76% in comparison to what they would be if that consumer good was not available, (table 171) These unstandardized Cox regression coefficients match the logistic regression ones. According to the 1996-1991 pooled Cox regression models, the following variables have an effect on the likelihood of infant mortality at the 5% level of significance: breastfeeding, immunization, goods, mother’ s education, household crowding , sex of the child and prenatal care by a doctor (once immunization is controlled). The regression results for the Cox proportional hazards method conform well with the logistic results. 483 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 171 Cox Regression Results for 1996-1991 Ne D ata S et (1) Variable Name B Standard Error Exp(B) Significance (p) Birth Cohort 0.0361 0.0641 1.0368 0.5728 Urban X Rural -0.0879 0.1332 0.9159 0.5092 Drinking W ater 0.0229 0.1328 1.0232 0.8629 Sewage -0.1123 0.1178 0.8938 0.3404 Household Crowding 0.0601 0.02 1.062 * 0.0027 Age Group of Mother -0.0943 0.1184 0.91 0.4258 Ethnicity -0.1785 0.1251 0.8365 0.1535 Mother's Education -0.3172 0.0782 0.7282 0 Goods -0.2784 0.0664 0.757 * 0 Birth Order Risk -0.1899 0.1146 0.8271 0.0976 Dr’ s Prenatal care 0.5547 0.1563 1.7414 * 0.0004 Sex of Child -0.2532 0.0988 0.7763 * 0.0104 DPT Immz. -3.7779 0.5859 0.0229 * 0 Breastfeeding in Months -0.1951 0.0348 0.8227 0 Source: 1996 & 1991 DHS for Brazil's Northeast region from a sub-sam ple total of 7633 cases (7582 selected, 51 dropped). 484 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 172 C ox Regression Results for 1996-1991 Ne Data S e t (2) Variable Name 8 Standard Error Exp(B) Significance (p) Birth Cohort 0.0477 0.0629 1.0489 0.4479 Urisan X Rural -0-1062 0.1318 0.8993 0.4205 Drinking W ater 0.0041 0.1314 1.0041 0.9751 Household Crowding 0.0584 0.0199 1.0601 * 0.0034 Age Group of Mother -0.1013 0.1181 0.9037 0.3912 Ethnicity -0.1781 0.125 0.8368 0.1543 Mother's Education -0.3166 0.0783 0.7287 * 0.0001 Goods -0.2925 0.0647 0.7464 * 0 Birth Order Risk -0.1918 0.1146 0.8254 0.094 Dr's Prenatal care 0.5363 0.155 1.7097 * 0.0005 Sex of Child -0.2542 0.0988 0.7756 * 0.0101 DPT Immz. -3.7781 0.5859 0.0229 • 0 Breastfeeding in Months -0.1946 0.0347 0.8231 * 0 Source: 1996 & 1991 DHS for Brazil's Northeast region from a sub-sam ple total of 7633 cases (7582 selected. 51 dropped). 485 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 173 Cox Regression Results for 1996-1991 Ne Data S et (3) Variable Name 8 Standard Error Exp (8) Significance (p) Birth Cohort -0.0504 0.0623 0.9508 0.4184 Urban X Rural -0.0124 0.1333 0.9877 0.9258 Drinking W ater 0.0403 0.1317 1.0412 0.7593 Sewage -0.0819 0.1184 0.9214 0.4894 Household Crowding 0.0501 0.0201 1.0514 * 0.0129 Age Group of Mother -0.1425 0.1185 0.8672 0.2292 Ethnicity -0.1546 0.125 0.8568 0.2162 Mother's Education -0.3018 0.0779 0.7395 * 0.0001 Goods -0.2464 0.0664 0.7816 * 0.0002 Birth Order Risk -0.1836 0.1148 0.8323 0.1097 Dr's Prenatal care 0.2135 0.1459 1.238 0.1433 Sex of Child -0.2548 0.0986 0.7751 * 0.0097 DPT Immz. -4.2237 0.5841 0.0146 * 0 Source: 1996 & 1991 OHS for Brazil's Northeast region from a sub-sam ple total of 7633 cases (7594 selected, 39 dropped). 486 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 174 Cox Regression Results for 1996-1991 Ne Data S et (4) Variable Name B Standard Error Exp (8) Significance (p) Birth Cohort -0.0419 0.0611 0.9589 0.4924 Urban X Rural -0.0263 0-1318 0-974 0.8417 Drinking W ater 0.0272 0.1304 1.0276 0.8347 Household Crowding 0.049 0.0201 1.0502 0.0148 Age Group of Mother -0.1475 0-1182 0.8628 0.2119 Ethnicity -0.1541 0-125 0.8572 0.2174 Mother's Education -0.3014 0.0779 0.7398 0.0001 Goods -0.2564 0.0647 0.7738 0.0001 Birth O lder Risk -0.1855 0.1147 0.8307 0.1058 Dr's Prenatal care 0.2036 0.1452 1.2258 0-1608 Sex of Child -0.2551 0.0986 0.7748 * 0.0096 DPT Immz. -4.2229 0.5841 0.0147 0 Source; 1996 & 1991 DHS for Brazil's Northeast region from a sub-sample total of 7633 cases (7594 selected, 39 dropped). 487 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 175 Cox R egression Results for 1996-1991 N e D ata S et (5) Variable Name B Standard Error Exp(B) Significance (p) Cohort 90-96 0.0245 0.1948 1.0248 0.9 Cohort 85-89 -0.0991 0.1626 0.9057 0.5425 Cohort 80-84 -0.3013 0-164 0-7398 0.0861 Urban X Rural -0.0932 0.1333 0-911 0.4846 Drinking W ater 0.0286 0-1331 1.029 0.83 Sewage -0.1114 0.1181 0.8946 0.3456 Household Crowding 0.0623 0.02 1.0643 0.0019 Age Group of Mother -0.1022 0.1189 0.9028 0.3901 Ethnicity -0.183 0-125 0.8328 0.1433 Mother's Education -0.3172 0.0782 0.7282 0 Goods -0.2738 0.0666 0.7605 0 Birth Order Risk -0.1908 0.1144 0.8253 0.0932 Dr's Prenatal care 0.4942 0.1619 1.6392 * 0.0023 Sex of Child -0.2467 0.0989 0.7814 * - 0.0126 DPT Immz. -3.7683 0.586 0.0231 * 0 Breastfeeding in Months -0.2027 0.0357 0.8165 * 0 Source: 1996 & 1991 DHS fo r Brazil's Northeast region from a sub-sam ple total of 7633 cases (7582 selected. 51 dropped). 488 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 176 Cox Regression Results fo r 1996-1991 Ne Data S et (6) Variable Name B Standard Error Exp(B) Significance (p) Cohort 90-96 0.0572 0.1917 1.0588 0.7655 Cohort 85-89 -0.0744 0.1605 0.9283 0.643 Cohort 80-84 -0.2911 0.1636 0.7475 0.0752 U itan X Rural -0.111 0.132 0.895 0.4004 Drinking W ater 0.0098 0.1317 1.0099 0.9406 Household Crowding 0.0605 0.0199 1.0624 * 0.0024 Age Group of Mother -0.1088 0.1186 0.8969 0.359 Ethnicity -0.1826 0.125 0.8331 0.1441 Mother's Education -0.3167 0.0783 0.7285 * 0.0001 Goods -0.2878 0.0649 0.7499 * 0 Birth Order Risk -0.1941 0.1143 0.8236 0.0896 Dr's Prenatal care 0.477 0.1607 1.6113 * 0.003 Sex of Child -0.2477 0.0988 0.7806 * 0.0122 DPT Immz. -3.7687 0.586 0.0231 * 0 Breastfeeding in Months -0.2021 0.0356 0.817 » 0 Source: 1996 & 1991 DHS for Brazil's Northeast region from a sub-sam ple total of 7633 cases (7582 selected, 51 dropped). 489 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 177 Cox Regression Results for 1996-1991 Ne D ata S et (7) Variable Name B Standard Error Exp(B) Significance (p) Cohort 90-96 -0.2326 0.1918 0.7925 0.2252 Cohort 85-89 -0.1459 0.1637 0.8642 0.3728 Cohort 80-84 -0.2691 0.1638 0.764 0.1004 Urban X Rural -0.0151 0.1334 0.985 0.9099 Drinking W ater 0.0447 0.1318 1.0457 0.7345 Sewage -0.0783 0.1186 0.9247 0.5092 Household Crowding 0.0511 0.0202 1.0524 • 0.0113 Age Group of Mother -0.1504 0.1187 0.8603 0.2048 Ethnicity -0.1566 0.125 0.855 0.21 Mother's Education -0.3019 0.0779 0.7394 * 0.0001 Goods -0.2453 0.0666 0.7825 * 0.0002 Birth Order Risk -0.1843 0.1147 0.8317 0.1081 Dr's Prenatal care 0.1997 0.1522 1.221 0.1894 Sex of Child -0.2531 0.0986 0.7764 * 0.0103 DPT Immz. -4.2349 0.5843 0.0145 * 0 Source: 1996 & 1991 DHS for Brazil's Northeast region from a sub-sam ple total of 7633 cases (7594 selected, 39 dropped). 490 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 17& Cox Regression Results for 1996-1991 Ne Data S et (8) Variable Name 8 Standard Error Exp (8) Significance (p) Cohort 90-96 -0.2092 0.1884 0.8113 0.2669 Cohort 85-69 -0.1282 0.1615 0.8797 0.4273 Cohort 80-84 -0.2621 0.1634 0.7695 0.1089 Urban X Rural -0.0284 0.1319 0.972 0.8297 Drinking W ater 0.0322 0.1305 1.0327 0.8052 Household Crowding 0.05 0.0201 1.0512 * 0.0128 Age Group of Mother -0.1552 0.1184 0.8562 0.1898 Ethnicity -0.1562 0.1249 0.8554 0.2112 Mother's Education -0.3016 0.0779 0.7396 * 0.0001 Goods -0.2548 0.0649 0.775 * 0.0001 Birth Order Risk -0.1862 0.1142 0.8301 0.1042 Dr's Prenatal care 0.1908 0.1515 1.2102 0.2079 Sex of Child -0.2534 0.0986 0.7762 0.0102 DPT Immz. -4.2346 0.5843 0.0145 0 Source; 1996 & 1991 DHS for Brazil's Northeast region from a sub-sam ple total of 7633 cases (7594 selected. 39 dropped). 491 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. “ H istory , i f view ed a s a repo sito ry fo r more than anecdote o r chronology, a m id produce a decisive tranrform ation m the image o f science by which we are now possessed... I f science is the co nstella tio n o f fa cts, ttw ories, a n d m ethods collected in current texts, then scientists a re th e m en who, successfulfy o r not, have striven to contribute one or another elem ent to th a t particular constellatian. S cien tific develofm ent becom es the piecem eal process b y w hich these item s have been added, singly a nd in com bination, to the ever grow ing sto ckp ile that con stitu tes x ie n tific technique a n d know ledge. " (Thom as K uhn, The Structure o f S cien tific R evolutions, 1962) 492 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 8 CONCLUSION This study is an attempt to delve into the facts and nature of the recent decline in infant mortality in Brazil’ s Nordieast, by far the country’ s most impoverished region. Through a theoretical and empirical work it investigates to what extent the acceleration m infant mortality decline m die region was brought about by an association of demographic, health and socio-economic determinants. In Chapter 2 the historical backdrop against which the Northeast became one of the most l a ^ ng areas in the Americas as far as both economic and human development is scrutinized. One of the underlying hypotheses examined in this research, is that the very nature and timing of infant mortality decline and of the mortali^ revolution itself in the Northeast are unique and distinct to other regions of Brazil. In chapter 3 it was demonstrated that mortality' levels in Brazil began to improve in the 1920 s and 1930’ s as a result of both economic and organizational changes in society as well as of the introduction of health and sanitary innovations. Between these decades life ea^ectancy increased around 30% in Brazil. In the 1940’ s life mcpectancy in Brazil increased slowly (7%), reaching a level close to 45 in 1950. 493 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The onset of the m ortali^ revolution in Brazil can be traced to the late 1940 s, early I950’ s. In the 1950's the average gain m life expectancy for the decade doubled to 14%. IBGE’ s (Brazilian Institute of Geography and Statistics) ofScial data indicate a life expectancy level of 52.3 in I960, in contrast, m . the 1960’ s there were practically no improvements according to the IBGE, and life expectancy would have remained around 52 until 1970. Other estimates indicate small but non residual gains in the 1960 s. At any rate, life expectancy was greatly increased in the 1970’ s (18% according to the ofScial data), reaching a level of 62 in 1980. As of the 1980’ s life expectancy at birth continue to improve but at a lower tempo reaching 67.6 in 1996. Brazil’ s infant mortality also decreased slowly in the 1940 s, changing from a level of 163.4 in 1940 to a level of 146 in 1950. In the 1950’ s Brazil infant mortality fell 17%, reachmg 121 in 1960. in the 1970’ s infant m ortaliy fell 39%, which also fostered a robust fertility decline in the 1980’ s. By 1980, Brazil’ s infant mortality was 69.1. Between 1976 and 1996, infant mortaliy declined 50%, declining from a level of 85 in 1976 to the current level of 37.5 (1996). hi the Northeast region the onset of mortality revolution took place a decade later than the pattern displayed in the rest of the country. The differential between life expectancy a t birth in Brazil as a whole and in the Northeast actually doubled between the 1940’ s and the 1950’ s (from ^ C elso S im o e s, Ibid, 1 3 7 . 494 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. around 4 to 8 years), bi the 1960*s life «cpectancy in the r ^ o n improved 16%, rising feom 44.3 in I960 to 51.6 in 1970. In the following two decades Northeast’ s life expectanQr rose an additional 14 and 9%, respectively. Even though the differential in life expectancy levels between Brazil and the Northeast declined significantly between 1960 and 1991 (fiom 18 to 2%), the levels reached in the South region in 1930 (51.1), for instance, would only be achieved in the region in 1970 (51.6). As far as infant mortality, there was also a timing ^ p of at least a decade between the changes in the rest of the country and in the Northeast. Whereas the first "wave" of infant mortality decline in Brazil happened in the 1950’ s and the second started in the 1970’ s until present times, in the NE infant mortality only started to decline consistently in the 1960’ s. In the 1950’ s infant mortality declined in average 25% in the South and Southeast and only 5% in the Northeast region. Infant mortality decline in the Northeast as of the mid 1980’ s became for the first time more robust than that of other areas of Brazil. Between 1991 and 1996 infant mortality fell 24.5% in Brazil, 14% in the Southeast region and 31.5% in the Northeast region. Thus, the nature and the timing of infant mortality decline in the Northeast, as well as of the demographic transition itself, are mdeed quite distinct than the pattern revealed in other regions of the country. 495 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In this dissertation it is also argued that the very differentiation between economic development and public health, or between endogenous and exogenous infant mortality determinants is misleading in Üie sense that it obscures historical peculiarities and the compile myriad of interactive demographic, health and socio-economic causes of infant mortality decline. This study attempts to quantity the impact of thirteen independent variables on infant mortality, hypothesizing that some socio-economic, health or institutional factors would be more pervasive than others. It is clear that substantial unprovements occurred with all independent variables, and particularly with maternal educational attainment levels. (Table 186). Following are the tables indicating the proportional change in all independent variables between 1986 and 1996, according to the respective logistic regression data sets, (tables 179 - 187) Table 179 Place of Residence, 1986 and 1996 P Iace of Residence (•/.) 1986 1996 Urban 60-9 73.8 Rural 39.1 26.2 Source; 1986 & L996 LR Data Sets. 2t5 Data on edinicity, prenatal care and DPT immunizaticn were not present in 1986. 496 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 180 Drinking Water, 1986 and 1996 IrinldnE W ater (%) 1986 1996 Good 65.9 73.5 Not Good 34.1 26.5 Source; 1986 & 1996 LRData Sets. Table 181 Sewage, 1986 and 1996 Sewage (% ) 1986 1996 Modem 12.2 23.8 Not Modem 87.8 162 Source: 1986 & 1996 LRData Sets. Table 182 Household Crowding, 1986 and 1996 Household CrowdinB(%) # o f persons per household 1986 1996 2 11.1 14.1 3 18.4 24.6 4 16.9 21.1 5 16.3 15.2 6 11.1 8.9 7 9.3 5.1 8 4.8 3.4 Source; 1986 & 1996 LRData Sets. 497 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 183 Mother’ s Age Risk, 1986 an d 1996 Mother’s Aee R isk(% ) 1986 1996 Low Risk (20-34) 56.1 47.4 High Risk (15-19 & 35-49 43.9 52.6 Source: 1986 & 1996 LR Data Sets. Table 184 Mother’ s Educational Attainment, 1986 and 1996 Mother’s Educatioiud Level {*/•) 1986 1996 No Education 112 12.9 Primary 70.7 41.8 Secondary 10.6 41.1 Higher 1.6 4J2 Scarce: 1986 & 1996 LRData Sets. Table 185 Goods, 1986 and 1996 Goods-addhive Sun of Radio* ' rv and car (%) 1986 1996 0 19.0 11.3 1 41.2 37.7 2 30.4 34.6 3 9.4 16.4 Source: 1986 & 1996 LRData Sets. 498 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 186 Birth Order Risk, 1986 and 1996 Birth (JrderR isk 1986 1996 Low Risk (2 or 3 child) 31.6 46.1 Hig^ Risk (1 and over 3 " * d iild ) 68.4 53.9 Source: 1986 S c , 1996 LR Data Sets. Table 187 Sex of infant, 1986 and 1996 Sex of Infant 1986 1996 Fmnale 49.4 48.0 Male 50.6 52.0 Source: 1986 & 1996LRData Sets. The dissertation’ s main hypothesis is that the most significant factors promoting advances in infant survival are improvements in education of the mother, in household economic levels, in the quality of sewage disposal as well a s o f prenatal care by a physician and breastfeeding status. What is being measured are the odds of survival for the latest child ever b o m , as reported by the mother in the DHS’ s of 1986, 1991 (NE only) and 1996. 499 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Preston, and Trussel ^ a i^ e that techniques and statistical procedures based on surviving children or children ever bom are a robust and powerful for investigating the effect of covariants on infant mortality. Tables 188 and 189 indicate which explanatory variables have a statistically signffîcant impact on the dependent variable in the logistic and Cox data sets for 1996, 1991 and 1986. Following these tables we will scrutinize the descriptive and regression time series results for all independent variables. ^ J. Trussel and Samuel Preston, “Estimatmg the Covariates o f Childhood Mortality^ fimn Retrospective Rqx>its o f Mothers”, Methodologies for die CoUection of Mortality Data by J Vallm, & J Pollard & L Heligan (eds).(Belghim: Ordina Editions, lUSSP, 1984). 500 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 188 Significant Didependent Variables in LR Data Sets (1 o f 2) 1986 (826) 1991 (3212) 1996 (3269) Logistic Regression 5% level (model 1) Logistic Regression 5% level (model 3) Goods Goods DPT, Prenatal, BC, HH, Goods S«c DPT, BC, Prenatal, HH Goods, Prenatal Education Sewage Goods, Education Prenatal, Sewage Logistic Regression 1% level (model I) DPT, Prenatal, BC Goods, Prenatal, Education Logistic Regression 1% level (model 3) DPT, BC Goods, Education, Prenatal 501 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 189 Significant Independent Variables in Cox Data Sets (2 of 2) 1986 (1120) 1991 (3811) 1996 (3822) Cox Regression 5% level (model 1) BF,HC, Education Goods BF,DPT, Prenatal, HC, Goods, Education BF, Goods, Education Prenatal Sewage Cox Regression 5% level (model 3) BF, HC, Education Goods BF, DPT, BC, Prenatal, HC, Goods, Education BF, Goods Education, Prenatal Cox Regression 1% level (model 1) BF,HC BF, DPT, Prenatal BF, Goods Education, Prenatal Cox Regression 1% level (model 3) BF,HC BF, DPT, Prenatal BF, Goods Education Prenatal 502 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 190 Significant Independent Variables in Pooled LR Data Sets (1 o f 2) 91-86 (4038) 96-91-86 (7307) 96-91 (6481) Logistic Regression 5% level (model 1) BC, Goods, Education, HC, Sex Goods, Education, Age, Sex, HC, Sewage Prenatal, DPT, Education, Goods, Sex,HC Logistic Regression 5% level (model 3) BC, Goods, Education, HC, Sex BC, Goods, Education, Age, Sex, HC, Sewage Prenatal, DPT, Education, Goods, HC, Sex, BC Logistic Regression 1% level (model 1) BC, Goods, Education BC, Goods, Education, Age, Sex Prenatal, DPT, Education, Goods, Sex, Logistic Regression 1% level (model 3) BC, Goods, Education BC, Goods, Education, Age, Sex DPT, Prenatal, Education, Goods, HC 503 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 191 Significant Independent Variables in Pooled Cox Data Sets (2 of 2) 91-86 (4931) 96-91-86 (8753) 96-91 (7633) Cox Regression 5% level (model 1 ) BF,HC, Goods, Education BF, HC, Goods, Education, Sex, Sewage, BO BF,DPT, Education, Goods, Sex, Prenatal, HC Cox Regression 5% level (model 5) BF, HC, Goods, Education, BC, Sewage BF, HC, Goods, Education, S«c, Sewage, BO BF, DPT, Education, Goods, HC, Prenatal, Sex Cox Regression 1 % level (model 1 ) BF, HC, Goods, Education BF, HC, Goods, Education BF, DPT, Education, Goods, Sex, Prenatal, HC Cox Regression 1 % level (model 5) BF, HC, Goods, Education BF,HC, Goods, Education, BF, DPT, Education, Goods, HC, Prenatal 504 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Urbanization. According to the logistic regression data sets, between 1986 and 1996 the rate of urbanization in the Northeast region increased from 60.9 to 73.8% (table 179). The u rb a n X n a rtl vaimble mdicate that 8.9% of the children who lived in urban areas deceased, in 1986 and 5.1% in 1996. For children living in rural areas these respective means are 11.0 and 7.9% (1996) (tables 71 and 107). However, this variable has not shown to have a statistically sipiifrcant effect on the dependent variable in any of the models or data sets, (tables 188, 189) Drinking Water In the 1986 LR data set 65.9% of the households had access to good drinking water. In 1996 this proportion had increased to 73.5%. (table 180). The proportion of children who died in households with good drinking water was 8.6% in 1986 and 5.6% in 1996. The respective proportions in household with inadequate drinking water are 12 and 6.5%. (tables 71 and 107). According to the data, the d rin k in g w a ter variable does not have a statistically significant impact on infant mortality in any of the data sets, logistic or Cox formatted, pooled or not. (tables 188, 189) Drinking water is not even correlated with the dependent variable in 1996 and 1986. (tables 73 and 109) 505 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Sewage 12.2% of the Northeastem households in 1986 and 23.8% in 1996 had access to a modem type of toilet facility, (table 181). 11% of the children living in homes with inadequate sewage died in 1986 and 6.5% in. 1996. hi households with m o d em se w tig e these levels are, respectwely, 4.6 and 3.7%. (tables 71 and 107). Improvements in sewage disposal were found to have a statistically significant effect on infant mortality between 1996 and 1986 but not as much as expected. The peculiarly high level of modem sewage according to the 1991 data sets might have skewed the 1991 results, th o u ^ . In 1996, type of toilet facility only has an impact on the odds of survival at the 5% level of significance, (table 188 and 189) Recent investments in sanitation in the Northeast are having a positive impact on infant mortality, but, as far as the data are concerned, this effect is proportionally not as strong as the impact of other explanatory variables. Age of Mother at Birth The proportion of high risk births has decreased from 68.4 % in 1986 to 53.9% in 1996. (table 183) In 1986, 12% of the children of high age risk mothers and 8% of the children of low age risk mothers died, hi 1996 the children of high age risk mothers were twice as much likely to die than the children of low age risk mothers - 7.6 and 3.8%, respectively, (tables 71 and 107). 506 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The a g e o f th e m o th e r a t b irth variable was found to have a statistical^ significant impact a t the 1% level of significance only in pooled logistic regression 1996-91-86 data set (tables 188 and 189). Household C!rnwding The proportion of households with only 2, 3 or 4 members increased from 46.4% in 1986 to 59.8% in 1996. (table 182) The data indicate that, in recent years, the household crowding variable became less important in explaining infant mortality as measured by the dependent variable. The non dichotomous HC variable had a statistically significant effect on the dependent variable at the 5% level of significance in the 1986 Cox regression models. The variable is significant at the 5% level in the LR pooled data sets and at the 1% level of significance in the Cox regression formatted data sets. The impact of the h o u se h o ld cro w d in g variable decreases in 1991 and HC does not have a statistically significant effect on the odds of survival in the 1996 logistic and Cox data sets. During the past few years crowding household conditions in the region might have been improved by the decline in fertility levels, reflected by the decline in importance of infectious and respiratory diseases. 507 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Etfanicitv In 1991, 78.3% of the interviewed mothers were non-white and the proportion of deceased children coming hom such ethnic background was 7.6% (5.5% in "w hite* households). In 1996, 71.3% of the subjects were non white and the proportion of non white children who died had fallen to 6.3% (4.7% in "white’ households). ( tables 71 and 89). No data on e th n icity were collected in 1986. This variable does not have a significant impact on child survival at all according to the logistic and Cox regression data sets, (table 188). Birth Order Risk Factor This demographic covariate only has a significant impact at the 5% level on infant mortality in the pooled 1996-91-86 Cox data set (table 188 and 189). The proportion of high risk children decreased from 67.4% in 1986 to 53.9% in 1996. (tables 186, 107 and 71). 11% of the h ig h risk and 6.9% of the low r is k children died according to the 1986 LR data set, and 7.4 and 3.9%, respectively, in the 1996 LR data set. Sex of Infant The g e n d er o f th e irfix n t has a statistically significant impact on the odds of survival at the 5% level in the 1991 LR data set as well as in most pooled data sets, (tables 188 and 189). The proportion of female children decreased from 49.4% in 1986 to 48% in 1996. (table 187). 508 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The male over infant mortalily is strong in the region: 11% (against 9.1% for females) in 1986 and 6.3% ( against 5.3% for females) in 1996. (tables 71 and 107) DPT Tinimim>.atinn Immunization levels, measured by variable DPT 1-2-3, or whether the child was given the three shots of DPT or not, had a very significant impact on the variable in 1991, but no statistically significant effect at all in 1996. (tables 188 and 189) This result is in agreement with Simoes’ investigation of the number of years lost according to death causes in the Northeast of Brazil which shows that the proportional contribution of immunization to the decline in mortality in the NE declined fiom 0.92% of the total number of deaths in 1980 to 0.73% in 1990. The decline in importance for the DPT variable does not indicate that vaccination no longer has an impact on infant survival but rather that the continuous immunization campaigns and programs developed as of the mid 1980’ s by the Health Ministry had a very positive effect on infan t mortality, which was reflected by the significance levels of the variable in the 1991 data sets and that this effect diminished substantially in recent years as more and more infants are immunized. The im m u n iza tio n variable was not present in the 1986 surveys. 509 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Prenatal Care bv a Doctor The p re n a ta l ca re b y a d o cto r variable has a statistically significant impact on the dependent variable in the 1991 data sets and this effect seem to have been maintained or grown m recent years. The percentage of infants who were seen by a doctor during prenatal care remained around 36/37% between 1991 and 1996, and no data were collected in 1986. Purtha* research should be carried out to investigate the changes in the unpact of this explanatoiy variable across time. Goods Goods or the p ro ^ for household income has been a very important factor in explaming time series changes in infant mortality between 1986 and 1996. This study found strong evidence that the family's per capita income, as expressed by ownership of a radio, a TV set and a car, has a significant statistical effect on infant mortality. This variable is the only variable significan t at the 5% level in the 1986 LR data sets. It is also significant in the 1986 Cox data sets. In 1991 the variable goods is significant a t the 5% level. Ih 1996 as well as in all pooled data sets, die socio-economic variable goods has a statistically significant ünpact on the odds of survival at the 1% level of confidence, (tables 188 and 189). 510 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The very recent ^ in s in meome distribution, and the growth in real per capita incomes for Brazil’ s poor after the creation of the Real Plan in 1994 have certainly had a positive effect on infant mortality'. According to Ûie Cox regression data sets, the percentage of the households surveyed that had none of the three consumer goods (radio, T V and car) decreased ftom 22% in 1986 to 19% in 1991 to 12% in 1996. Between 1986 and 1996, according to the logistic data sets, the proportion of households with none of the three g o o d s fell ftom 19 to 11.3%, while the proportion of households with all three goods increased ftom 9.4 to 16.4%. (table 185) Breastfeeding The proportional hazard method confirms a robust positive impact of duration of breastfeeding on infant mortality in all of the Cox data sets. The continuous efforts to and programs to promote b re a stfee d in g developed by the Ministry of Health as of the mid 1980’ s - such as the Program of Integral Assistance to the Health of the Child and the Project for the Reduction of Infant Mortality- as well as by States, Municipalities and NGO’ s have been playing an important role in increasing breastfeeding levels. The strongest infant mortality differentials are found between no breastfeeding and 1 month of breastfeeding. 511 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. At any rate, duration, of breastfeeding in months, the only time varying variable in the model, is consistently a variable that exerts a very significant statistical effect on infant mortality. Maternal Eîducation As far as explaining infant mortality gains during the 1986-1991 period, the independent variable that had the most profisund effect on the dependent variable is unquestionably the improvement in the educational attainment of the mother, (tables 188 and 189) According to the multivariate logistic regression analysis, the variable m a tern al ed uca tio n did not have a statistically significant impact on the dependent variable in 1986 and m 1991. However, in the Cox formatted data sets, education of the mother is significant at the 95% level of confidence both in 1986 and in 1991. In the I990’ s extraordinary gains in women’ s educational levels, obtained through aggressive and very successful policies and campaigns executed by the Education Mmistry to eradicate illitemcy and to improve educational attainment levels in Brazil’ s Northeast region, did have a considerable positive impact on infant mortality. According to the Cox formatted data sets, the percentage of mothers who had secondary or higher education, increased firom 12% in 1986, to 17% in 1991 to 45% in 1996. (tables 71, 89 and 107). 512 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Between 1986 and 1996, in the logistic regression data sets, the proportion of mothers with high education increased from 1.6% to 4.2%, while the proportion of mothers with secondary education increased fbur-fold from 10.6 to 41.1%. (table 184). To investigate the claim that education of the mother, and particularly enrollment in secondary levels, is the most important factor in explaining infant mortality decline in Brazil's Northeast between 1986 and 1996 we will estimate the relative impact of the improvements in the educational attainment of the mother in explaining the decline in infant mortality during the period. The 1996 regression coefScients will be used in the 1986 standard regression model (with 9 independent variables). The minimum and the maxim um impact of the improvements in education on the dependent variable will be computed. The mean value for education increased &om .97 in 1986 to 1.37 in 1996. The actual probability of death declined from 98/1000 ( 81/826) in 1986 to 58/1000 ( 190/3269) in 1996. Running the standard logistic regression model for 1986 (model 1, table 111) only with mother’ s educational attainm ent as the explanatory variable would yield an infant m ortali^ odds value o f. 10148 in 1986. Log odds = - 1.641 - .6669 X (.97) = -2.2879 Odds= .10148 513 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. where, -1.641 is the new intercept coefiScient; - .6669 is the unstandardized B coefficient for education and .97 is the mean value for the education variable in 1986. Using the mean value for 1996 in the 1986 equation would yield an infant mortality odds value of .0777 in 1996. Log odds = - 1.641 - .6669 X (1.37) = -2.5547 Odds= .0777 Transforming these odds values into probabilities would yield predicted probabilities of .919 in 1986 and of .0721 in 1996. P / 9 = P / (1-p) = .10148 P / q = P / (1-p) * • 0777 p = .0919 p = .0721 92/1000 is the original predicted probability of dying in the 1986 sample provided that education is the only mcplanatoiy variable. 72/1000 is the predicted probability of dying in the 1996 sample if the only accountable temporal changes relate to the gains in the educational attainment of the mother. 58/1000 is the actual probability of dying in the 1996 survey. The difference in the probability value between 1986 and 1996 (using the 1986 equation given the gains in education between 1986 and 1996) is .0198. 514 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. p (1986) - p ( 1996) = .0919 - .0721 = .0198. (A ), the change in infant mortahty attributable to maternal educational guns. Dividing this value by the difference between the probability for 1986 having education as the only ^cplanatoiy variable and the actual probability in 1996 will indicate the maximum relatwe estimated impact of education on infant mortality gains between 1986 and 1996. p ( 1986 ) - actual p ( 1996) = .0919 - .0582 = .0327 (B), the actual change in infant mortality. A/B = .0198 / .0327 = .6055 or 60.6%. The m ay iT T iu T n estimated impact of maternal educational gains on infant mortality decline m. Brazil’ s Northeast between 1986 and 1996 is 60.6%. Thus, almost 2/3 of the decline in infant mortality would be explained by the improvements m maternal education. If the other independent variables (excluding the ones not present in the 1986 data sets) are also included m the regression model we will arrive at the minimum impact of the educational gains on in fan t mortality dedine between 1986 and 1996. The regression model including these additional 9 variables would entail a log odds value of -2.4499 and an odds value of .086. Converting this value into probability would yield a p value of .0792 or an infant m ortali^ of 79/000. Subtracting this value from .0919 would yield a value of 0.117. 515 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. p (1986) - p ( 1996) = .0919 - .0792 = .0117. (C ), the change in infant mortality attributable to maternal educational gams, given that other independent variables are also included, or given that the other variables correlated with education ( goods in particular) are causally independent. Dividing this value (C ) by .0327 wül indicate the minimum contribution of improvements in maternal education attainment to gains m infant mortality in Brazil’ s Northeast between 1986 and 1996. C/B = .0117/ .0327 = .3578 or 36%. Hence, the relative impact of maternal education gains on infant mortality decline would range hrom a maximum of 60% to a minimum of 36%. In sum, the multivariate regression results indicate that the variable goods had a statistically significant impact on the dependent variable in 1986 and 1991 and that this effect has also increased in recent years. In addition, months of breastfeeding has been a determinant factor in all surveys, even though it cannot be concluded from the data that breastfeeding levels have been increasing. The prenatal care by a doctor variable is also an important factor in determinant the odds of survival for the infant in both 1991 and 1996, in spite of the fact that it is not absolutely clear fiom the data whether this independent variable explain the temporal changes in the dependent variable between 1986 and 1991. It can be concluded fiom this research study that breastfeeding levels, education, goods and prenatal care are all important factors affecting mfant 516 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. mortality. Furthermore, the acceleration in the rate of decline in the 1990*8 has been deeply influenced by improvements in income levels and, more importantly, by gains in the mother's educational levels. The multivariate repression results indicate that the recent improvements in education had the strongest impact on infant survival (accounting fiom 36 to 60% of the total improvement), as far as explaining the acceleration in the rate of decline during the 1986-1996 period. This conclusion indicating that maternal education is the most important single explanatory variable influencing infant mortality is in agreement with studies developed by Mosley & Chen Mosley 2 8», Behm 2 8 9 ^ Da Vanzo 2 9 0^ Van de Walle & Mbacke 2 9 1, and PaUoni 2 9 2. The main implication of these fTodings is that abating infant mortality in the Northeast would be and has been most efSciently promoted by educating women, either through schooling or th ro u ^ breastfeeding campaigns and policies. However, further studies must be developed to integrate these policy proposals and research findings to a thorough and continuous analysis of the involved costs as well as to the cost of implementing alternative policies. ^ H. Mosley & L Chen, Ibid. 1984. ^ H. Mosley. Ibid., 1985. 289 I H. Behm, Ibid, 1979. J. Da Vanzo. Ibid., 1 E. Van de Walle & C ^ Alberto Palloni. Ibid., 1981. ^ J. Da Vanzo, Ibid., 1984. ^ E. Van de Walle & C. Mbacke. Ibid, 1992. 517 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. BIBLIOGRAPHY Arriaga, Eduardo. New Life Tables far Latin American Popuiatian in the Nineteenth and Twentieth Centuries. Berkeley: In stitu te for International Studies, 1968. Arriaga, Eduardo and K. Davis. P a ttern s o f M ortality C h a n g e in L a tin A m erica. Demography, No.6, 1969. Arriaga, Eduardo and KingslQr Davis. T h e P attern o f M o rta lity C h a n g e in L a tm A m erica. Demography, Vol. 6, No. 3., 1969. Arriaga, Eduardo. M ortality D eclin e a n d its D em ographic E ffe c ts in L atin A m erica. B e rk e le y . Institute for International Studies, 1970. Arruda, Jose Maria, Naomi Rutenherg, Leo Morris and EHisaheth Anhel Ferraz. P esq u isa NadoncU S a b re S a u d e M atem o -Infan til e P U m ejam ento FamUUtr PNSMIPF - B ra sü , 1 9 8 6 . Rio de Janeiro, RJ: Sociedade Civil Bem-Estar Familiar no Brasfl. - B E ^M F A M , December 1987. Bartlett, F.R.S. T h e A n a ly sis o f B m a ry D ata. L ondon a n d C o lch ester. Printed in Great Britain by Spotdswood, Ballantyne & Co Ltd, 1970. Badiani, Rita and Ines Quental Ferrena. P esq u isa N a cio n a l S ob re D em ografia e S a u d e 1 9 9 6 . Calverton, M D: Rio de Janeiro, RJ: Sociedade Civil Bem- Estar Fam iliar no Brasü - BEM FAM , March 1997. Behm, Hugo and D. Primante. M ortalidad e n L o s P rim eras A flo s d e V ida e n La A m erica L a tin a . Notas de Poblaciôn Vbl.6, No. 16, 1978. Behm, Hugo. Socioeconom ic D eterm in a n ts o f M o rtality in L a tin A m erica. Proceedings of the Meeting on Socioeconomic Determinants and Consequences of Mortality, UN and W HO, Mexico City, 1979. Behn, Hugo. Socioeconomic Determinants of Mortali^ in Latin America. Proceedings of the Meetings on Socioeconomic Determinants and Consequences of Mortality, Mexico City, June 1979. Behm, H. and E. Vargas. O uatem ala: D iferen d a s S o d o eco n â n d ca s d e L a M ortalidad d e L o s M en o res d e D os A fio s 1 9 6 8 -1 9 7 6 . SenA, No.1044, San José, 1984. Bercy, W.D. and S. Feldman. M u ltip le R eg ressio n in P ractice U nderstanding Regression A ssu m p tio n s. Beverly Hills, C A : Sage University Papers no. 50,1985. 518 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. B erry, W .D , U n d ersta n d in g R eg ressm n A ssu m p tio n s. Newbuiy Park, C A : Sage University Papers no. 9 2 , 1993. Breen, Richard. R eg ressio n M odels, C ensored, S a m p lerS elected , o r T run ca ted D ata. London: Sage Publications, 1996. Bums, E. Bradford. A H isto ry ofB rasdL New York: Columbia Uhivarsity Press, 1970. Caldwell, J. E d u ca tio n a s a F actor in M ortality D eciine: A n E xxm dnatiori o f N igerian D ata. Population Studies 33, no. 3, 1979. Caldwell, J. E d u ca tio n a s a F actor in M ortality D ecH ne: A n E xxaninatiort o f N igerian D ata. Population Studies 33, no. 3, 1979. Caldwell, J., P. Caldwell and P. Reddy. T he S o cia l C om ponent o f M ortality D ecline: A n R w estig a tio n in S o u th In d ia em p lo yin g A tte m a tiv e M ethodologies. Population Studies 37, no. 2, July 1983. Caldwell, J. R o u te s to L ow M ortality. Population and Development Review, V ol. 12, No. 4, 1986. Carvalho, Jose Alberto de. A n a ly sis o f R egional T ren d s in F ertility, M ortality a n d M igration in B ra zil - 19 5 0 -1 9 7 0 . Ph.D. Dissertation, London School of Economics, 1973. Carvalho, Jose Alberto de. F ecun d ida d e e M ortaU dade no B ra sü . Ford Foundation Research Report. Belo Horizonte, 1977. Chen, SeeXue. b rfa n t M o rta lity a n d its D eterm in a n ts in L o s A n g e le s C ounty, 1 9 8 9 -1 9 9 1 . P.h.D. dissertation, U.S.C., 1997. .C hild M ortality in D eveloping C ountries: Socio-econom ic D ifferen tia ls, T rends and Im p lica tio n s. Department of International Ekzonomic and Social ASahs. United Nations, N ew York, 1991. Collett, D. M odeling B in a ry D ata. London, U K : Chapman & Hall, 1991. Correa, Hector and Mohamed Ali El Torlty. T he BiologU xd D eterm in a n ts o f T he D em ographic T ra n sitio n . Washington, D C.: University Press of America, Inc., 1982. Cox, D.R. R eg ressio n M o d els a n d L ife T ables. Journal of the Royal Statistical Society, Series B, 34 (2), 1971. 519 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Curtis, S., I. Diamond and J. McDonald. B irth In terva l a n d F a m ily E ffe c ts o n P ùstn eo n atrd M ortality in B ra ziL Demography 30, n o .l, February 1993. DaVanzo, J. A H o u seh o ld S u rv ey o f C h ild D eterm im ards in M a la ysia . Population and Development Review, Child Research- Strategies for Research, Supplement to Vol.10, 1984. DaVanzo, Julie, h ifa n t M ortality a n d S od o econ o m ic D evelo p m en t: E vid en c e fro m M a la ysia n H o u seh o ld D ata. Demography 25, no. 4, 1988. Davis, k. T he Am evzing D ecline o f M o rta lity in U n d erd evelo p ed A rea s. American Economic Review, No. 46, 1956. Diaz-Biiquets, Sergio. T he H e a lth R ev o lu tio n in C uba. Austin, Texas: library of Congress Cataloging in Publication Data by University of T^cas, 1983. Easterlin, Richard and Eileen Crimmins. T he F ertility R evo lu tio n : A S u p p ly - D em and A n a ly sis. Chicago: University of Chicago Press, 1985. Ffimindidnd*» anticoncepcao e m ortaU dade m fa n tiL P e sq u isa so b re S a u d e F am iliar no N o rd este, 1 9 9 1 . BEM FAM Sociedade Civil Bem-E^tar Familiar no Brasil. Rio de Janeiro, 1994. Foster, S. Im m u n iza tio n a n d R esp ira to ry D ise a se s a n d C h ild M ortality, Population and Development Review, Child Research- Strategies for Research, Supplement to Vol. 10, 1984. Furtado, Celso. Form acao E conontica d o B ra sü . Rio de Janeiro: Editora Fundo de Cultura, 1959. Furtado, Celso. U n d erd evelo p m en t a n d D ependence: University of Cambridge, 1964. Furtado, Celso. E conom ic D evelo p m en t in L a tin A m erica: A S u rv e y fro m C olonial H m es to th e C uban R evo lu tio n . Cambridge: Cambridge University Press, 1970. Furtado, Celso. D evelo p m en t a n d U n d erd evelo pm en t. A S u rv e y fio m ColonUil T im es to th e C uban R evo lu tio n . Cambridge University Press, 1970. Furtado, Celso. O b sta cles to D evelo p m en t in L a tin A m erica . Garden City, N Y : Anchor Books Doubleday & Company, Inc., 1970. Geer, John P. Van de. A h d tiva ria te A ru d y sis o f C ategorical D ata: Theory. London: Sage Publications, 1993. 520 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. G w a iià xi,D , tid ic a to r s o f C hcm ge in D evelo p in g C o u n tries M ortaHtg T ren d s: T h e E n d o f a n E ra?. Population and Development Review, V61.6. No. 4, 1980. Hamilton, Lawrence C. R eg ressio n iv ith G ra p h ics. Pacific Grove, CA: Cole Publishing Company, 1992. Hirschman, Albert O. T h e S tra teg y o f E conom ic D euelopm enL New Haven and London, Yale University, 1958. Hirschman, Albert O. D evelo pm ent P ro jects O b served . Washington, DC: The Brooking^ Institution, 1967. Hirschman, Albert O. J o u rn e y s Tow ard P ro g ress, S tu d ie s o f E conom ic P olicy- M aking in L a tin A m erica. New York, N Y : Greemwood Press, Publishers, 1968. Hobcraft, J., J. McDonald, J. and S. Rutstein. D em ographic D e term in a n ts o f In fa n t a n d E a rly C h ild M ortality: A C om petrative A rm lysis. Population Studies 39, 1985. Hobcraft J. W o m en 's E d ucation , C hild W elfa re cm d C hild Survived: A R ev ie w o f th e E vid en ce. Health Transition Review 3. no. 2, 1993. Inkeles, A . M aking M en M odem : O n th e C a u se s a n d C o n seq u en ces o f Indiiridued C hange in S ix D evelop in g C ountries. American Journal of Sociology, 75, N o. 2, 1969. Jaguacibe, Helio. In tro d u ca o ao D esen vo lvim en to Soded. Rio de Janeiro: Paz e Terra, 1978. Kuhn, Thomas S. T h e S tru ctu re o f S c ie n tific R e v o lu tio n s. London: The University of Chicago, Ltd., 1970. Kuznets, Simon. M o d em E conom ic G row th. New Haven: Yale University Press, 1966. Kuznets, Sûnon. P o p u la tio n T rends a n d M o d em E conom ic G row th: N a tes Tow ards a H tstoriccd P erspective. Population Debate: Dimensions and Perspectives, New York: World Population Council, 1975. Kuznets, Simon. P o p u la tio n T rends a n d M o d em E conom ic G row th: N o tes Tow ards a IS sto ru x d P ersp ective. Population Debate: Dimensions and Perspectives, New York: World Population Council, 1975. 521 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Lewîs-Beck, Michael S. A pplied. R eg ressio n , A n P ro d u c tio n . Newbiuy Park, C A : Sage Publications, Die., 1980. Long, Scott. R eg ressio n M od els fo r C ategorical a n d L im ited D e p en d e n t V ariables. Thousand Oaks, C A : Sage Publications, 1997. Maddison, Angus. D ynam ic F o rces in CcqpitaHst D evelopm ent. New York: Oxford University Press, 1991. Marcotte, John. The b npatd o fP ÿ re d W ater o n C hild M ortality a n d B re a stfe ed in g in E cuador cm d BrasaL Ph.D. dissertation. University of Wisconsin- Madison, 1988. McKeown, T. The M odem R ise o f P opulation. New York: Academic Press, 1976. Merrick, Thomas and Douglas Graham. P opulation cm d E conom ic D evelop m en t in B ra zil- 1 8 0 0 to th e P resen t. Baltùnore: The John Hopkins University Press, 1979. Merrick, Thomas. T he E ffec t o f P ip ed W ater on E a rly C h ild h o o d M o rta lity in U rban B razil, 1 9 7 0 to 1 9 7 6 . World Bank Staff Working Papers 594, Washington, D.C., 1983. Morse, Richard. The U rban D evelo p m en t o f L a tin A m erica - 1 7 5 0 -1 9 2 0 . Center for Latin American Studies, Stanford University, 1971. .MnrtaHtif a n d H ea lth P olicy. P roceedings o f th e G roup o n M ortality a n d H e a lth P olicy R om e, 3 0 M ay to 3 J u n e 1 9 8 3. Department of International Economic and Social Affairs. United Nations. New York, 1984. .M nrtfiH hj o f C h ild ren u n d e r a g e 5 . W orld E stim a te s a n d P rojections, 1950- 2 0 2 5 . Department of International Ekx>nom ic and Social Affairs. United Nations. New York, 1988. Mortara, Giorgio. E stu d o s so b re a U tûizacao do C enso D em ogrctfico p a ra a R econstrucao d e E sta tistic a s d o M ovim ento d a P opulacao d o B rasiL V I. Sinopse da Dinamica da Populacao do Brasil nos Ultimos Cem Anos, Revista Brasileira de Estadstica, Rio de Janeiro, 1941. Mortara, G io r^o . M etho d s o f U sing C e n su s Sfohsdcs fo r th e C a lcu la tio n o f l ife T a b les a n d O ther D em ographic M easure (urith A p p lica tio n to th e P opulation o f B ra zil ). Population Studies No.7, New York, 1949. Mortara, Giorgio. C ontribm coes p a ra o E stu d o d e D em ogm fia no B rasiL IBGE, 1970. 5 2 2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Mosley, Henry and L. Chen. A n A n a lytica l F ram ew ork: J a r th e S tu d y o f C h ild S u rviva l in D evelo p in g C ountries. Population and Development Review, Child Research- Strategies for Research, Supplement to Vol.10, 1984. Mosley, Henry. C h ild Survived: R esearch, a n d P olicy, Population and Development Review, Child Research- Strategies for Research. Supplement to Vol. 10, 1984. Mosley, Henry. L e s S o in s d e S a n té P rim aires P e u v e n t-ils R éd u ire L a M ortalité In fa n tile? B ila n C ritique d e Q u elq u es Program m e s A frica in s e t A sia tiq u e s. In La Lutte Contre La Mort, J. Vallin and A. Lopez (eds). Travaux et Documents No. 108, 101-136, INED-PUF, Paris, 1985. Nelder, J.A. and R.W. Wedderbum. G en era lized L in e a r M oeJels. Journal of Üie Royal Statistics Society, A 135, 1972. Newland, Kathleen, b fa n t M ortality a n d th e H e a lth o f S o cieties. Worldwatch Paper 47, December 1981. Norusis, Marija J. S P S S 6 .1 G uide to D ata A n a ly sis. Englewood Cliffs, New Jersey: Prentice Had, 1995. Ortiz, L . and C. Simôes. A M ortaU dade In fa n tü . no Brasü. n o s A n o s 8 0 . IBGE/DPE, Vol. 1, No.7, Rio de Janeiro, 1988. Palloni, Alberto. M o rtality D ecU ne in L a tin A m erica . Paper presented at the Annual Meeting of the Population Association of America. Washington D.C., 1979. Palloni, Alberto. M o rtality D e d in e in L a tin A m erica . Paper presented at the Annual Meeting of the Population Association of America, 1979. Palloni, Alberto. M o rtality in L a tin A m erica: E m erg in g P a ttern s. Population and Development Review, V01.7, No. 4, 1981. Palloni, Alberto and S. MOlman. E ffe c ts o f U rter-B irth In te rv a ls a n d B re a stfeed in g o n U nfant a n d E a rly C h ild h o o d M ortality. Population Studies 40, 1986. Palloni, Alberto. H e a lth L e v e ls a n d C are in L a tin A m erica: T he C a se o fb tfa n t M ortality 1 9 0 0 -1 9 8 5 . Health Transition, Chapter 10, Vol.l. Camberra: The National Library of Australia, 1989. Palloni, Alberto. H e a lth L e v e ls a n d Care in L a tin A m erica: T he C a se o fh fa n t M ortality 1 9 0 0 -1 9 8 5 a n d H ea lth T ra n sitio n . Canberra: National Library of Australia, 1989. 523 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Palloni, Alberto. H eaU h L e v els a n d C a re in L a tin A m erica: T h e C a s e o fB tfa n t M o rta lity 1 9 0 0 -1 9 8 5 . Health Transition, Chapter 10, V oL l. Camberra: The National Library of Australia, 1989. Palloni, Alberto. H e a lth L evels a n d C a re in L a tin A m erica: T h e C a se o fL rfa n t M o rta lity 1 9 0 0 -1 9 8 5 . Health Transition, Chapter 10, Vbl.1., Camberra: The National Library of Australia, 1989. Pool, L I s N ew Z e a la n d a H ea lth y C ou ntry?. New Zealand Population Review, VoL8. No.2, 1982. .P e sq id sa N a a o n a l so b re D em ografia e S a u d e 1 9 9 6 . R elatorio PreUndnar. Sociedade Crvü Bem-Estar Familiar no Brasil, BEM FANf. Programa de Pesquisas de Demografia e Saude (DHS)/Macro Intemacional, hic., Rio de Janeiro, 1996. .Pesquisa sobre Saude Familiar no Nordeste Brasfl. 1991. Sociedade Cânl Bem-E^tar Familiar no Brasfl, BEMFAM. Macro hitemational Inc. Rio de Janeiro, 1992. Press, S. James. C hoosing B e tw e e n L o g istic R e g re ssio n A n d D iscrim inant A n a ly sis. Santa Monica, C A : The Rand Corporation, 1979. Preston, Samuel. T he C hanging R e la tio n B e tw e e n M o rta lity a n d L evel o f E conom ic D evelopm ent. Population Studies, No. 2, 1975. Preston, Samuel. Aforta/dy, M orbidity a n d D evelo pm ent. P a p er p res/en ted to th e Seminar on Population and Development in the ECWA Region, Sept. 1978. Preston, Samuel. T he E ffec ts o f In fa n t a n d C h ild M ortality o n F ertility. New York, N Y : Academic Press, Inc., 1978. Preston, Samuel. T he C hanging R e la tio n s B e tw e e n M ortality a n d th e L evel o f E conom ic D evelopm ent. Population Studies, Vol. 29, No. 2, 1980. Preston, Samuel. Causes a n d C o n se q u e n ce s o f M ortality D e clin es in LDC D uring th e T w en tie th C entury" in "P opulation a n d E conom ic C hange in D evelo p in g C ountries", edited by R. Easterlin. Chicago: University of Chicago Press, 1980. Preston, Samuel and J. TrusselL E stim a tin g th e C o va ria tes o f C hildhood M ortahtg jro m R etro sp ective R e p o rts o f M others. Methodologies for the Collection and Analysis of Mortality Data, edited by J. Vallin, J. Pollard, and L. Heligman, 1984. 524 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Preston, Samuel. M o rta lity in C hüdhood: L e s s o n s fio m W HS. R ep ro d u ctive C h an g e in D evelo p in g C ountries. J. Cleland and J. Hobcraft (eds), Oxford University Press, 1985- Preston, Samuel and N G cfaael Haines. F a ta l Y ea rs: C h tid M ortality in L a te T w en tie th C e n tu ry A m erica. N ew Jersey: Princeton University Press, 1991. Preston, Samuel Preston and Michael Haines. F a ta l Y ears. Princeton University Press, 1991. .Q F IV E - U nited. N a tio n s Program Jb r C h ild M ortality E stim a tio n . A m icrocom puter p rogra m to accom pany th e S tep -b y-S tep G uide to th e E stim a tio n o f C h ild M ortality. Department of International Economic and Social Affairs. United Nations. New York, 1990. Ramanathan, Ramu. S ta tistic a l M eth o d s In E conom etrics. La Jolla, C A : Academic Press, Inc., 1993. Robock, Stefan H. B ra zil's D eveloping N o rth est: A S tu d y o f R egional P la n n in g a n d F oreign A id . Washington, DC: The Brookings Institution, 1963. Rosero, Luis and H. Camaano. T ablas d e V ida d e C o sta Rica, 1 9 0 0 -1 9 8 0 . Associacion Demografica Costarricense. San Jose, 1984. Rutstein, S., J. Sullivan and G. Bicego. ta fa n t a n d C h ild M ortality. DHS Comparative Study 15, Calverton, June 1994. Ruzicka, Lado. P ro b lem s a n d Is s u e s in th e S tu d y o f M ortality Itiffe r e n tid ls . Differential M ortali^ Methodological Issues and Biosocial Factors, L. Ruzicka & G. Wunsch & P. Kane, (eds.), Oxford: Clarendon Press, 1989. Ruzicka, Lado, Guillaume Wunsch and Penny Kane. D ifferentU d M ortality: M ethodological Is s u e s a n d B iosocial F a cto res. Clarendon Press, Oxford, 1989. Saboia, Ana Lucia. C ria n ca s & A d o lescen tes^ h u S c a d o res S o d a is. Rio de Janeiro, RJ: Fundacao Instituto Brasüeiro de Geografîa e Estatistica - IBGE/DEDIT/CDDEI, 1991. Sastiy, Narayan, N. Goldman and L. Moreno. T h e R ela tio n sh ip b etw een P la ce o f R e sid e n c e a n d C h tid S u rviva l in B ra zti. hitemational Population Conference, no. 3, lUSSP, Liège, 1993- Sastry, Narayan. F a m ily-L evel C lu sterin g o f C h ild h o o d M ortality R is k in N o rth ea st B ra ziL Pbpulation Studies 51, Nov. 1997. 525 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Schroeder, L. D., D. L. Sjoquist and P.E. Stephan. U ndersU m ding R eg ressio n A n a ly sis: A n B ntroductory G uide. Beverfy HOIs, CA: Sage Universily Papers no. 57, 1986. Serrano, R. and C. Puffer. C a ra cteristica s d e la M ortaH dad e n la N itiez. Pan American Health Association, Washington, D.C., 1973. Shorter, F. andT. Belgin. D e term in a n ts o f C h ild M ortality: A S tu d y o f S q u a tter S e ttle m e n ts in Jo rd a n . In Child Survival - S tra t^ e s for Research, edited by Henry Mosley and L. Chen, Population and Development Review 10, 1985. Simôes, Celso and D. Vetter. Aoesso a o s Serviços d e S a n ea m en to B ûsico e M ortalidade. Revista Brasileira de E^tatistica, Vol.42, No. 169, 1982. Simôes, Celso. A sp e c to s M etodologicos d a s E stim a tiva s d e M ortalidade h fa rü ü no B rasiL Rio de Janeiro; IBGE, 1990. Simôes, Celso and lüri Leite. PadrOo R eprodutivo, S erviço s d e S a û d e e M ortaH dade h fa n tü -N o rd e ste . Pesquisa sobre Saûde Familiar no Nordeste, 1991. Simôes, C. A S a û d e Irfa n til no B ra sü n o s A n o s 90. In fû n d a B ra sileira n o s A n o s 90. UNICEF/IBGE, Brasilia, 1996. Simôes, Celso. A S a û d e h fa n tU n o B rasU n o s A n o s 9 0 a n d Irfû n c ia B rasileira n o s A n o s 9 0 . UNICEF/IBGE, Brasilia, 1996. Simôes, Celso. A M ortalidade In fà n til n a Transicao d a M ortalidade no B rasü: Um E^studo C om parativo e n tre o N o rd e ste e o S u d e ste . UFMG/ Cedeplar, Belo Horizonte, 1997. Simôes, Celso. A M ortalidade ù ^ a n tü n a Transicao d a M ortalidade no B rasü: Um E stu d o C om parativo e n tre o N o rd este e o S u d e ste . Ph D. Dissertation in Demography. Belo Horizonte; Cedeplar, 1997. Simonsen, Roberto. H istoria E conom ica do B ra sü -1 5 0 0 -1 8 2 0 . Sao Paulo; Companhia Editora N adonal, 1969. Skidmore, Thomas. Uma H isto ria d o B ra siL Sao Paulo; Paz e Terra, 1999. .Sodo-E conondc D ^ e r e n tia ls in C h ü d M ortality in D evelop in g C ountries. Department of International Economic and Social Affairs. United Nations. New York, 1985. 526 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. .S t^ }-b y -S te p G u id e to th e E stim ation, o f C h ü d M ortality. Department of International Ekxmomic and Social Afiaûs. United Nations. New York, 1990. Stolnitz, G. R e c e n t M orta lity T ren d s in L a tin A m erica , A sia a n d A fric a Population Studies, Vol. 19, 1965. S to h d tz, G . In terru ttio n a l M ortcdity T ren d s: S o m e M ain T rends a n d Im p lica tio n s. UN, The Population Debate: Dimensions and Perspectives, Papers of the World Population Conference, Vol. 2, Bucharest, 1974. Strat&eld, K., I. Diamond and M . Singarùnbum. M a tem a l E du ca tio n a n d C h ild Im m u n iza tio a Demography 27, no. 3, 1990. Suzigan, Wilson, h id u stria liza tio n cm d E conom ic P o lity in H isttjritxd F ersp & rive. Rio de Janeiro: 1P E IA /IN P E ÎS , 1976. T Y x om as, D., J. Strauss and M . Henriques. C h ü d Survived, H eig h t fo r A g e a n d H o u seh o ld C h a ra cteristics in B ra zü . Journal of Development E)conomics 33, no. 2, 1990. Trussel, J. and Samuel Preston. E stim a tin g th e C ovariates o f C h ü d h o o d M ortality fro m R etro sp ective R ep o rts o f M others. Methodologies for the Collection of Mortality Data by J Vallin, & J. Pollard & L. Heligan (eds).Belgium: Ordina Editions, lUSSP, 1984. Trussel, J. and C. Hammerslough. A H txzards-M odel A n a ly sis o f th e C o va ria tes o f In fa n t a n d C h ild M ortality in S ri L a n k a Demography 20, no. 1, Feb. 1993. Vallin, Jacques, Stan D’ Souza and Alberto Palloni. M ea su rem en t a n d A n a ly s is o f M ortality. New York: Clarendon Press, Oxford, 1990. Victora, C.G., P. Smith and J.P. Vaughan. Sotticd a n d Erw ironm entcd In flu e n c e s o n C h ü d M o rta lity in B razü: lo g is tic R e g re ssio n A n ed ysis fr o m C e n su s P iles. Journal of Biosocial Science 18, no. 1, 1986. Walle, E. Van de and C. Mbacké. Sottio-E conontic F acto rs cm d H e a lth S e rv ic e U se in M ortcdity a n d S o ciety in S u b -S a h a ra n A fr ic a Oxford: Clarendon Press, 1992. Wetherill, Barrie G. R eg ressio n A n a ly sis tv ith A p p lica tio n s. New York, N Y : Chapman and Hall Ltd, 1986. Wood, Charles and Jose Carvalho. T he D em ography o f In eq u a lity in B raziL Cambridge: Cambridge University Press, 1988. 527 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. APPENDIX Table 192 Correlation. Matrix, Levels of Correlation Among Independent Variables, 1996 Logistic Regression Data Set (1 of 2) Birth Cohort Variable Urban X Rural Drinking Water Modem Sewage Household Crowding Age Risk for Mother Ethnicity Birth Cohort Variable 1.000 -.128 -.077 -.099 .150 .533 -.044 Urban X Rural -128 1.000 .470 .158 -.069 .005 .059 Drinking Water -.077 .470 1.000 .166 -.033 .030 -.012 Modem Sewage -.099 .158 .166 1.000 -.036 -.037 .054 Household Crowding .150 -.069 -.033 -.036 1.000 -.057 -.098 Age Risk for Mother .533 .005 .030 -.037 -.057 1.000 -.011 Ethnicity -.044 .059 -.012 .054 -.098 -.011 1.000 Mother’s Education .004 .345 .245 .149 -.156 .106 .145 Goods (Additive o f radio, tv, and car) -.190 .354 261 .175 -.111 -.058 .181 Birth Order Risk -.007 .129 .080 .071 -.205 .138 .073 Prenatal Care By Doctor .565 .057 .017 -.029 .035 .375 .008 Sex of Chüd -.028 -.010 -.022 .004 -.010 -.026 .004 DPT immuniz. (1 ,2 and 3) 3TJ -.029 -.018 -.046 .053 .379 -.010 528 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 193 Correlatioii Matrix, Levels of C orrelatioii Am ong Independent Variables, 1996 Logistic R egression D ata S et (2 of 2) Mother’s Education Goods (Additive o f radio, tv, and car) Birth Order Risk Prenatal Care By Doctor Sex of Child DPT immuniz (1 ,2 and 3) Birth Cohort Variable .004 -.190 -.007 .565 -.028 J 77 Urban X Rural .345 J54 .129 .057 -.010 -.029 Drinking Water .245 J267 .080 .017 -.022 -.018 Modem Sewage .149 .175 .071 -.029 .004 -.046 Household Crowding -.156 -.111 -J05 .035 -.010 .053 Age Risk for Mother .106 -.058 .138 .375 -.026 .379 Ethnicity .145 .181 .073 .008 .004 -.010 Mother’s Education 1.000 .434 .217 .145 -.009 .064 Goods (Additive o f radio, tv, and car) .434 1.000 .163 -.018 -.002 -.073 Birth Order Risk .217 .163 1.000 .020 -.001 .006 Prenatal Care By Doctor .145 -018 .020 1.000 -.028 .763 Sex of Child .009 -.002 -.001 -.028 1.000 -.028 DPT imminiiV (1, 2 and 3) .064 -.073 .006 .763 -.028 1.000 529 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 194 Correlation Matrix, Levels o f Correlation Among b idependent Variables, 1991 Logistic R egression D ata Set (1 o f 2) Birth Cohort Variable Urban X Rural Drinking Water Modem Sewage Household Crowding A ge Risk for Mother Ethnicity Birth Cohort Variable 1.000 -.146 -.140 -.181 .139 .461 -.019 Urban X Rural -.146 1.000 .657 .569 -.097 .014 .030 Drinking Water -.140 .657 1.000 J33 -.075 .010 .041 Modem Sewage -.181 .569 .533 1.000 -085 -.044 .107 Household Crowding .139 -.097 -.075 -.085 1.000 -125 -.059 Age Risk for Mother .461 .014 .010 -.044 -.125 1.000 -.011 Ethnicity -.019 .030 .041 .107 -059 -.011 1.000 Mother’s Education .014 252 250 .383 -.106 .139 .178 Goods (Additive o f radio, tv, and car) -.201 .396 .397 .560 -.051 -.034 .136 Birth Order Risk -.022 .121 .113 .166 -.250 201 .030 Prenatal Care By Doctor .459 .107 .078 .112 -.006 261 .026 Sex o f Chüd .011 -.001 .008 .008 -.039 .016 -.020 DPT im m uniz (1 ,2 and 3) .490 -.037 -.009 -.057 .022 359 .009 530 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 195 Correlation. M atrix, Levels o f Correlation A m ong Independent Variables, 1991 Logistic R egression D ata S et (2 o f 2) Mother’s Education Goods (Additive o f radio, tv, and car) Birth Order Risk Prenatal Care By Doctor Sex o f child DPT fmmuniz (1 ,2 and 3) Biith Cohort Variable .014 -201 -.002 .459 -.011 .490 Urban X Rural .252 296 .121 .107 -.001 -.037 Drinking Water .250 2 97 .113 .078 .008 -.009 Modem Sewage .383 .560 .166 .112 .008 -.057 Household Crowding -.106 -.051 -250 -.006 -.039 .022 Age Risk for Mother .139 -.034 201 261 .016 .359 Ethnicity .178 .136 .030 .026 -.020 .009 Mother’s Education 1.000 .417 209 .172 .004 .077 Goods (Additive o f radio, tv, and car) .417 1.000 .136 .024 -.024 -.091 Birth Order Risk .209 .136 1.000 .065 .017 .025 Prenatal Care By Doctor .172 .024 .065 1.000 -.013 .676 Sex of Chfld .004 -.024 .017 -.013 1.000 .Oil DPT inununiz. (1, 2 and 3) .077 -.091 .025 .676 .011 1.000 531 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 196 Correlation Matrix, L evels o f Correlation Among Independent Variables, 1986 L ogistic R egression D ata Set (1 o f 2) Birth Cohort Variable U ibanX Rural Drinking Water Modem Sewage Household Crowding Age Risk for Mother Mother’s Education Birth Cohort Variable L O G O -.153 -.108 -.044 .132 .330 -.015 Urban X Rural -.153 1.000 .604 .216 -.057 .036 .231 Drinking Water -.108 .604 1.000 .244 -.065 .039 .256 Modem Sewage -.044 .216 .244 1.000 -.066 .028 .232 Household Crowding .132 -.057 -.065 -.066 1.000 -JZ15 -.127 Age Risk for Mother .330 .036 .039 .028 -.215 1.000 .142 Mother’s Education -.015 .231 .256 .232 -.127 .142 1.000 Goods (Additive of radio, tv, and car) -.151 .340 .344 .298 -.017 -.037 347 Birth Order Risk .011 .062 .106 .099 -.289 .258 .182 Sex of Child -.017 -.023 -.012 .014 .003 -.032 -.026 532 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 197 Correlatioii Matrix, Levels o f Correlation Am ong Independent V ariables, 1986 Logistic R egression D ata S et (2 o f 2) Goods (Additive o f radio, tv, and car) Birth Older Risk Sex of Child Birtii Cohott VanaUe -.151 .011 -.017 U itanX Rural 340 .062 -.023 Drinking Water .344 .106 -.012 Modem Sewage 398 .099 .014 Household Crowding -.017 -389 .033 Age Risk for Mother -.037 .258 -.032 Mother’s Education .347 .182 -.026 Goods (Additive o f radio, tv, and car) 1.000 111 -.027 Birth Older Risk 111 1.000 -.035 Sex o f Child -.027 -.035 1.000 533 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Context is everything: An empirical study of institutions, instability and economic growth

PDF

Before and after reform: The effects of structural adjustment on women's employment in Latin America

PDF

Life-cycle demographic decisions and the economic well-being of single parents in the United States, 1964-1989

PDF

Two essays in political economy of education and economics of immigration

PDF

The determinants of electoral participation in the United States, 1952-1992: A theoretical and empirical examination

PDF

An analysis of the developmental state: The case of the Vietnamese textile and garment industries

PDF

Weight variations among preschoolers: An analysis of evidence from rural Tamilnadu

PDF

Compatibility standards, market power, and antitrust policy: Proprietary interfaces and their alternatives

PDF

After protection: The impact of import barriers on corporate equity values and profit expectations

PDF

An interregional demoeconomic simulation model of population distributation for California and Arizona

PDF

Some new perspectives on the issue of countercyclical, U.S. fertility

PDF

The drivers of economic change: Technology, globalization, and demography

PDF

Economic transition, micro- and small -scale enterprise in Zimbabwe, and the significance of participatory development to institutional economics

PDF

Dependence on foreign labor, quality of education and unemployment in the GCC countries: In search of solutions

PDF

Foreign capital inflows and trade policy change: The automotive and consumer electronics industries in Brazil and China

PDF

A Developing Pattern Of Unenumerated Individual Rights In The United States Constitution

PDF

She Works Hard For The Money: Consumerism And The Movement Of Housewives Into Wage Work

PDF

A comparative examination of the impact of business -government relations on labor market reform in Egypt and Mexico, 1975--1995

PDF

Special interest groups and the Lebanese conflict: A survey of attitudes

PDF

Comparative analysis of the transition to deliberate fertility control and its links to modernization in seven Asian nations

Asset Metadata

Creator Prata, Fernando Veiga (author)

Core Title A theoretical and empirical examination of infant mortality decline in Brazil's northeast region, 1986--1996

Degree Doctor of Philosophy

Degree Program Political Economy

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Economics, General,OAI-PMH Harvest,sociology, demography

Language English

Contributor Digitized by ProQuest (provenance)

Advisor Easterlin, Richard (committee chair), Heer, David (committee member), Lowenthal, Abraham (committee member)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c17-596787

Unique identifier UC11353823

Identifier 9955033.pdf (filename),usctheses-c17-596787 (legacy record id)

Legacy Identifier 9955033.pdf

Dmrecord 596787

Document Type Dissertation

Rights Prata, Fernando Veiga

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA