Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Armed conflict, education and the marriage market: evidence from Tajikistan
(USC Thesis Other)
Armed conflict, education and the marriage market: evidence from Tajikistan
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
ARMED CONFLICT, EDUCATION AND THE MARRIAGE MARKET: EVIDENCE FROM TAJIKISTAN by Olga N. Shemyakina A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (ECONOMICS) August 2007 Copyright 2007 Olga N. Shemyakina ii DEDICATION To the loving memory of my grandparents: baba Masha (Maria Petrovna Shemyakina) baba Dusya (Evdokiya Fedorovna Zuyeva) deda Kesha (Vikentii Frantsevich Romanovskii) and the only grand-grandparent I knew personally: baba Tanya (Tatyana Stepanovna Vasil’chenko) To my parents iii ACKNOWLEDGEMENTS I would like to thank my advisor John Strauss for guiding me in my research throughout the years, for his numerous helpful comments and encouragement. He has been an excellent mentor to me who introduced me to the fascinating world of micro-level development economics. I appreciate the insightful and constructive suggestions that Lynne Casper, Richard A. Easterlin, Jeffrey B. Nugent, and Geert Ridder provided me with. My advisors have been incredibly kind and patient. I also received helpful advice from John Ham and many seminar and conference participants. The World Bank, the State Statistical Committee and the Ministry of Education of Tajikistan made this research possible by granting permission to use their data. Elena Vasil’evna Budnikova of the State Statistical Committee of Tajikistan contributed valuable insight in the state of education in Tajikistan and helped me track down valuable statistical information and publications. Saidmomim Kamolov helped me establish contacts with the Ministries of Health and Education and the State Statistical Committee of Tajikistan. Naoko Hosaka introduced me to valuable contacts at the UNICEF Tajikistan. Dmitry Tkachenko maintained close email contact with me while I was in Dushanbe and ensured that I had a safe and comfortable accommodation. I gratefully acknowledge financial support from the University of Southern California (USC), the USC Urban Initiative and the Institute for Social Research/William Davidson Institute at the University of Michigan. I have enjoyed the friendship and support of Pouyan Mashayekh-Ahangarani, Ladan Masoudie, Subha Mani, Tomas Tencer, Rubina Verma, Engin Volkan and Anke Zimmermann who have been great friends and colleagues at USC. Marguerite Webley was a fun and supportive roommate during my last year at USC. iv I owe thanks to my family. My husband, Harikumar V. Iyer, has always encouraged me, been my best friend and often, editor. He is the best thing that happened to me. My family has offered continuous emotional support, encouragement, love and phenomenal patience. I owe to my parents; Nikolay Savel’evich Shemyakin and Lyudmila Vikent’evna Shemyakina, for helping me develop my love for reading and allowing me to pursue an education of my choice. My brother, Vladislav Shemyakin, has always been there for me, and kept me abreast of the news at home. My grandfather, Savelii Prohorovich Shemyakin, has set for me an example of sincerity and perseverance. My in-laws, R. Vijayan and Jamuna, have provided me with endless jokes, delicious food and recipes. Sudhakar Murthy and Rebecca Smith have become for me a family away from home in South Pasadena and have never failed to cheer me up with some wonderful fusion food creations and Hollywood gossip. Last, but not the least, I would like to thank people I met in Tajikistan for their help and hospitality. v TABLE OF CONTENTS DEDICATION ................................................................................................................... ii LIST OF TABLES............................................................................................................ vii LIST OF FIGURES ........................................................................................................... ix ABSTRACT ....................................................................................................................... x CHAPTER 1: OVERVIEW OF ARMED CONFLICT IN TAJIKISTAN: 1992-1998.......... 1 1. INTRODUCTION....................................................................................................... 1 2. BACKGROUND......................................................................................................... 2 2.1 Overview of Tajikistan........................................................................................... 2 2.2 The 1992-1998 armed conflict in Tajikistan............................................................ 3 2.3 Economic situation in Tajikistan after its independence .......................................... 5 CHAPTER 2: THE EFFECT OF ARMED CONFLICT ON ACCUMULATION OF SCHOOLING..................................................................................................................... 9 1. INTRODUCTION....................................................................................................... 9 2. EVIDENCE FROM OTHER COUNTRIES................................................................13 3. BACKGROUND........................................................................................................19 3.1 General education and enrollment rates in Tajikistan .............................................20 4. DATA ........................................................................................................................26 4.1 Exposure to conflict, school enrollment and grade attainment ................................26 4.1.1 Measures of exposure to armed conflict ..............................................................27 4.1.2 School enrollment and grade attainment by conflict exposure..........................28 4.2 Robustness ............................................................................................................34 5. IDENTIFICATION AND EMPIRICAL SPECIFICATION ........................................38 5.1 Empirical strategy 1: school enrollment .................................................................38 5.2 Empirical strategy 2: completion of mandatory schooling ......................................41 6. RESULTS ..................................................................................................................44 6.1 School enrollment .................................................................................................44 6.2 Completion of mandatory schooling ......................................................................56 7. DISCUSSION AND CONCLUSION .........................................................................59 CHAPTER 3: THE EFFECT OF ARMED CONFLICT ON THE MARRIAGE MARKET AND FEMALE REPRODUCTIVE BEHAVIOR...............................................................63 1. INTRODUCTION......................................................................................................63 2. LESSONS FROM OTHER COUNTRIES AND BACKGROUND .............................66 2.1 The marriage market .............................................................................................66 2.1.1 Economic shocks and the marriage market ......................................................67 2.1.2 Physical security concerns and marriage..........................................................69 2.1.3 Marriage squeeze............................................................................................69 2.2 Fertility outcomes and violent conflict...................................................................74 2.3 Marriage and family in Tajikistan..........................................................................76 vi 3. DATA ........................................................................................................................82 3.1 Descriptive statistics..............................................................................................84 3.2 Duration data.........................................................................................................87 4. ESTIMATION ...........................................................................................................90 4.1 Identification.........................................................................................................90 4.2 Sex ratio calculation ..............................................................................................92 4.3 Methodology.........................................................................................................96 5. RESULTS ................................................................................................................101 5.1 Marriage analysis ................................................................................................101 5.1.1 Descriptive analysis of age at first marriage ..................................................101 5.1.2 Marriage hazard analysis...............................................................................111 5.2 Fertility analysis..................................................................................................116 5.1.1 Descriptive analysis of first births .................................................................116 5.2.2 Hazard analysis of first births........................................................................117 5.3 Age differences between husbands and wives ......................................................127 6. DISCUSSION AND CONCLUSION .......................................................................130 BIBLIOGRAPHY............................................................................................................134 APPENDICES.................................................................................................................144 Appendix A - Measures of conflict exposure ................................................................144 Appendix B – Schooling variables ................................................................................146 Appendix C – TLSS data and sample construction: analysis of education......................147 Appendix D – Basic statistical framework: analysis of duration data.............................161 vii LIST OF TABLES Table 1 - Enrollment by gender and damage to household dwelling, ages 7-16. .................... 11 Table 2 - Enrollment by gender and community damage dwellings, 1999 vs. 2003. .............. 12 Table 3 - Enrollment rates by age, gender and household dwelling damage, 1999................. 29 Table 4 - Enrollment rates by age, gender and community damage dwelling, 1999. .............. 30 Table 5 - Enrollment rates by age, gender and community damage dwelling: 2003 ............... 30 Table 6 - Mean school grades completed by age, gender and household damage dwelling, 1999 ..................................................................................................................... 33 Table 7 - Mean school grades completed by age, gender and community damage dwelling: 1999 ..................................................................................................................... 33 Table 8 - Mean school grades completed by age, gender and community damage dwelling: 2003 ..................................................................................................................... 34 Table 9 - Enrollment by gender and reports of conflict activity (RCA).................................. 34 Table 10 - Enrollment by age, gender and reports of conflict activity (RCA). ....................... 36 Table 11 - Grades completed by age, gender and reports of conflict activity (RCA).............. 37 Table 12 - HDD and determinants of school enrollment, by gender, ages 7-15...................... 46 Table 13 - HDD and determinants of school enrollment: by age group. ................................ 50 Table 14 - Enrollment and students/teacher ratio in the 1997-1998 academic year. ............... 52 Table 15 - Enrollment and students/teacher ratio in an academic year of child’s initial enrollment............................................................................................................. 53 Table 16 - Interaction of damage dwelling variable with selected covariates, ages 7-15, by gender................................................................................................................... 55 Table 17 - Probability of completing nine grades of schooling and exposure to the armed conflict. ................................................................................................................ 56 Table 18 - Probability of completing nine grades of schooling: individual year dummies...... 58 Table 19 - Descriptive statistics for the sample of women: ages 15-49.................................. 86 Table 20 - Descriptive statistics: sample of married women.................................................. 87 viii Table 21 - Outline of the analysis of the transition data......................................................... 88 Table 22 - Three year birth cohorts: selected demographic data. ........................................... 92 Table 23 - Sex ratios by birth cohort and region of residence................................................ 94 Table 24 - Effect of regional conflict variable on sex ratio.................................................... 95 Table 25 - Marriage age by birth cohort.............................................................................. 102 Table 26 - OLS Regressions. Married by 18. Ages 18-37 in 2003....................................... 105 Table 27 - OLS Regressions. Married by age 20. Ages 20-37 in 2003................................. 107 Table 28 - OLS Regressions. Married by age 23. Ages 23-37 in 2003................................. 109 Table 29 - Semi-parametric marriage hazard regressions (Cox proportional hazard). .......... 114 Table 30 - Cumulative distribution of age at first birth, by 3-year birth cohort. ................... 116 Table 31 - Years between marriage and first birth, by number of years married. ................. 116 Table 32 - Semi-parametric conditional fertility hazard regressions (Cox proportional hazard)................................................................................................................ 121 Table 33 - Semi-parametric unconditional fertility hazard regressions (Cox proportional hazard)................................................................................................................ 123 Table 34 - Age difference (husband-wife). OLS regressions. .............................................. 128 ix LIST OF FIGURES Figure 1 - Central Asia: basic education gross enrollment rates, 1989-2003........................20 Figure 2 - Tajikistan: Enrollment trends, 1989-2003...........................................................22 Figure 3 - Students per teacher, by academic year and reports of conflict activity ...............25 Figure 4 - Boys: mean school grades completed (up to 9) by year of birth and conflict exposure ............................................................................................................................31 Figure 5 - Girls: mean school grades completed (up to 9) by year of birth and conflict exposure. ...........................................................................................................................32 Figure 6 - Tajikistan: mortality trends among children and young adults, 1989-1999 ..........65 x ABSTRACT This dissertation examines the effect of the 1992-1998 armed conflict in Tajikistan, a transition economy, on the micro-level behavior of individuals and households. The empirical strategy exploits the variation in regional differences in the extent and intensity of war-related events and the timing of the individual’s exposure to the civil war. Does exposure to armed conflict affect education? If it does, how long does the effect last and what populations are most vulnerable? The findings suggest that exposure to the conflict had a significant, negative and persistent impact on school enrollment and completion of the mandatory nine grades of schooling by girls who were of school age during the war and who lived in regions severely affected by armed conflict. While the enrollment rates in Tajikistan started to rise soon after the end of the war, school enrollments in the conflict affected areas took a longer time to catch up to their pre-war levels. Has the economic shock measured by conflict exposure also induced the population to postpone marriage and child-bearing? The second essay of my dissertation explores the effects of the conflict on the marriage market for women and timing of first births. The findings suggest that the exposure to conflict had a negative impact on the rate of entry into first marriages by women who were of marriageable age during the war and had virtually no effect on the interval between female age at first marriage and first birth. To what extent did the decrease in the number of men of marriageable age due to the war affect the marriage market for women? The results indicate that the sex ratio of men to women had a significant negative effect on difference in age between husbands and wives for women in the treatment group, however, no effect was found on entry into their first marriages by females and timing of first births. 1 CHAPTER 1: OVERVIEW OF ARMED CONFLICT IN TAJIKISTAN: 1992-1998 1. INTRODUCTION Civil wars and armed conflicts are common in less developed countries, and their detrimental effects are widely recognized. Most research on civil wars has been focused on their onset, development and end. 1 Also, existing research focuses primarily on African countries that experienced violent turmoil after their decolonization in the post World War II period (Collier et al. 2003). However, very few researchers have addressed the impact of civil wars on households and individuals. One possible reason is that large-scale, high quality household level data for developing countries affected by civil war are generally not available. Second, even when such data are available, it is difficult to identify whether the household coping behavior is induced by war or by economic conditions, as poor economic conditions often accompany or precede the conflict. For example, Miguel, Satyanath and Sergenti (2004), who study armed conflicts in 1981-1999 in 41 African countries, find that the likelihood of conflict increases by 50 percent in the year following a five percentage point negative growth shock. Third, detailed measures of conflict, and the destruction associated with it are often not available. Such information may be difficult to collect in countries that have emerged or are emerging from an armed conflict. This section of my dissertation provides a brief overview of the 1992-1998 armed conflict in Tajikistan and the economic and social conditions in Tajikistan after its independence. 1 Humphreys (2003) provides a recent review of the literature on this subject. 2 2. BACKGROUND 2.1 Overview of Tajikistan Tajikistan is a landlocked, mostly mountainous country in Central Asia. It borders Uzbekistan to the west and north, Kyrgyzstan to the north, China to the east and Afghanistan to the south. The country has a total land area of 143,000 square kilometers (slightly less than the state of Wisconsin in the USA). Only 6.52 percent of this land is arable as most of the territory is covered by mountains. Tajikistan is populated by approximately 7.3 million people as of 2006 (Central Intelligence Agency 2006). Almost 2.85 million of its inhabitants are under age 14 (World Bank 2006). Prior to it’s independence in 1991, Tajikistan was the poorest of the Former Soviet Union republics. The Tajik Soviet Socialist republic was largely supported by subsidies from the central Soviet government based in Moscow. Unfortunately, the legacy of poverty continues, and, fifteen years past independence, Tajikistan still occupies the bottom poverty spot in the FSU region. Almost 66 percent of Tajik population lives on less than USD2.15 a day (WFP 2006). The Republic of Tajikistan is administratively divided into four territorial regions: Sogdian oblast (Sugd, former Leninobod), Khatlon oblast, Gorno-Badagakshan Autonomous Oblast (GBAO) and Raions of Republican Subordination (RRS) also known as the Direct Rule Districts (Map 1). The first three territorial units, also known as veloyats or oblasts, are united administratively. The RRS region consists of 13 raions (districts). Each district is directly subordinate to the Central government in Dushanbe. The capital city Dushanbe is geographically situated in the RRS region. It is the largest city in Tajikistan. For analysis purposes Dushanbe is defined as a separate region in this dissertation. 3 Raions (districts) that were significantly exposed to war are primarily located in Dushanbe, Khatlon and the RRS regions. Some of the districts in the GBAO were also affected by the fighting and thus included in the definition of the war-affected regions. 2.2 The 1992-1998 armed conflict in Tajikistan One can draw parallel between the violent armed conflicts in the 1990s in the Former Soviet Union (FSU) region and the eruption of civil wars in some of African countries (Collier 2003). Both the collapse of the Soviet Union and decolonization in Africa created many small low-income economies that had difficulty in sustaining themselves. In contrast to the African post-colonial countries, the transition economies of the FSU inherited a well- developed infrastructure and highly trained human capital (Collier 2003). Those factors may have influenced the nature and length of the post-independence conflicts in the FSU region. Many of the low-level conflicts in the transition economies did not last as long as African conflicts. 2 In Tajikistan, old grievances and new political and economic problems led to a violent civil war that started in early 1992 and was followed by a prolonged armed conflict ending in 1998. In June of 1997, the United Nations facilitated a peace accord between the Tajikistan's government and the United Tajik Opposition (UTO). However, this ceasefire agreement was often broken in various parts of the country, with some districts serving as centers of insurgency. The reconciliation was complicated by a lack of a coherent national identity in Tajikistan, where warring sides were often associated with particular regions and 2 See Gleditsch et al (2002) on the length and intensity of civil wars in Africa and the FSU region. 4 ethnic groups. 3 The conflict ended in November, 1998 when government forces forced the remnants of opposition groups out of Tajikistan into Afghanistan. 4 The main parties to the conflict were the post-Communist Tajik government and several opposition groups, represented by the United Tajik Opposition (UTO). 5 Political affiliations in Tajikistan were based on regional sub-divisions and kinship groups, or clans. Thus, the conflict divided the country into geographical regions with opposing political interests. 6 The Tajik post-Communist party was comprised of members of the Kulob (Khatlon region) and Khodjent (Sugd region) clans, who had traditionally held power in the republic, and residents of the Uzbek-dominated Hissar region west of Dushanbe. 7,8 The opposition found regional support in the southern (Khatlon) and eastern (GBAO) districts and was supported by the members of Gharmi (Garmi) and Pamiri clans, who live in parts of Khatlon, RRS and most of the GBAO regions. The opposition represented interests of regionally based clans that lacked access to power during the Soviet times (Capisani 2000). Thus, the conflict could have been driven by the inequalities among ethnic groups or horizontal inequalities as outlined by Stewart (2002). 9 3 Tajikistan is inhabited by approximately 79.9% Tajiks and 15.3% Uzbeks (The World Factboook, CIA 2006 (Accessed: August 8, 2006).). The Tajik group also includes Pamiri people who speak a different language and belong to a different branch of Islam than the majority of the population (Roy 2000; Walker 2006). 4 University of Uppsala Conflict Database. (http://www.pcr.uu.se/database/conflictSummary.php?bcID=205 (Accessed: August 8, 2006).) 5 For detailed accounts of this conflict see Atkin (1997a), Nourzhanov (2005) and Walker (2006). 6 Allegiance to clans, also defined as patron-client networks (Atkin 1997b) or elite groups (Fridman 1994) is traditional in Central Asia, where the regional or kinship affiliations often overrule the sense of national identity (Roy 2000, Collins 2004). 7 Since the 1930s, members of those clans had occupied many important government positions and had managed state enterprises (Capisani 2000). 8 http://www.globalsecurity.org/military/world/war/tajikistan.htm. (Accessed: May 10th, 2005) and Capisani (2000: 168). 9 Stewart (2002) argues that civil war frequently occur when various ethnic groups mobilize against each other. The leaders of conflicting groups use religion, ethnicity or other identifiable characteristics to bring a group together. The mobilization becomes easy when there are horizontal inequalities among the above mentioned groups. Some groups may enjoy significant advantage in their access to economic, political and social resources, while the rest of the population may be denied access to the privileges due to the ethnic or religious differences. 5 As frequently happens, civilians also got involved in the conflict. Some civilians organized into small military units (Narodnii Front or People’s Front) to protect themselves and their communities from intruders. Many people were forced to leave their communities and houses. They found refuge in Dushanbe, Leninobod (Sugd) and GBAO or moved on to Uzbekistan and even Afghanistan. Others, in particular those of non-Tajik origin, left the country for good. The war led to a significant destruction of the state and people’s property. The disorder led to the formation of military groups who were interested in protracting the conflict. Crime escalated across the country. The capital of Tajikistan, Dushanbe, and southern regions such as Khatlon and Regions of Republican Subordination (RRS) were severely affected by the terror. Assassinations, hostage-taking, rapes, murders and robberies during the daylight became common. 10 The government was not able to contain the conflict independently and negotiated for outside political and military assistance, which was provided by Russia and Uzbekistan from 1992 to 1999. While the opposition did not receive any express financial or political support from any country in the region, it obtained arms from the bordering territories of Afghanistan and Iran and used these territories for their bases (Panarin 1994). 2.3 Economic situation in Tajikistan after its independence The 1992-1998 armed conflict took a significant toll on the country's physical infrastructure and destroyed much of its human and social capital. The first year of fighting brought the most damage. According to government official sources, 80 percent of the country's industry was destroyed by the end of 1992. The regional damage was felt even more in the 10 Based on the “Vechernii Dushanbe” and “Narodnaya Gazeta” news material for 1991-1999. 6 south, where 100 percent of industry was destroyed. 11 Agriculture was also severely affected. The human costs of the conflict were substantial for the 6.4 million inhabitants of Tajikistan. Approximately 40 percent of population was affected directly during the conflict. The conflict displaced at least 600,000 people internally. In addition, about 60,000 found temporary refuge in the neighboring states and 500,000 left the country for good. The conflict claimed the lives of at least 50,000 men, orphaned 55,000 children and widowed 20,000 women (Falkingham 2000). The conflict exacerbated economic problems that Tajikistan had experienced immediately after the dissolution of the FSU in 1991. Two post-transition problems were the deterioration of economic ties with business partners located in other FSU countries and the loss of subsidies from the Soviet central government in Moscow. The Soviet Union economic space was devised as a close-knit system, with some regions and republics specializing in particular products or industries. Tajikistan specialized in aluminum refining and cotton. The country’s independence in 1991 undermined both industries. Tajikistan's ties with its aluminum suppliers in Azerbaijan and Russia were severed. As a result, the Tajik aluminum smelter, the largest in Central Asia, was working below its operating capacity due to a decrease in the supply of its major raw material. Also, production and distribution of cotton were severely disrupted after the war. Two factors contributed to the temporary decline of the cotton industry during that period. First, similar to aluminum industry, the dissipation of industrial ties had left the cotton industry without a sufficient demand for its product; during the Soviet times almost 90 percent of cotton production was shipped outside of Tajikistan (McLean and Greene 1998). Second, shortage of laborers and 11 Nezavisimaja Gazeta, December 23, 1992 (as quoted in Fridman, 1994). 7 war damage in the southern cotton-growing regions led to the sharp decline in production of this major cash-crop. Over the course of the conflict, various military warlords and the government fought over control of the important agricultural and industrial centers, many of which are located in the southern part of Tajikistan. This fighting led to destruction of infrastructure, disruption of communication and transportation. For example, the railroad operated with severe disruptions. Bridges and rails were bombed and damaged. Fuel and labor were in short supply. The mass displacement of people during the first years of the war affected agricultural and industrial production in the south of Tajikistan and later led to insufficient labor supply in these areas. People who stayed at home in those regions were too scared to leave their houses to report for work. As a result of the labor shortage, fighting, fuel shortages and other disruptions, only 32.3 percent of the planned 820,261 tons of cotton collection was completed by the end of 1992. The war affected areas in the south, such as Kurgan-Tube, Bohtar, Tursunzade, Hissar, Gozimalik, Kumsangir raions, attained even smaller percentage of the planned output. Only 10 percent of the planned 360 thousand tons were collected in those raions. 12 Although labor was temporarily in short supply in some areas of Tajikistan, in other regions many employment opportunities disappeared as a result of war and the accompanying economic crisis. The situation was particularly harsh for some groups of the population. For example, after the war, many members of the Gharmi and Pamiri clans allegedly experienced employment discrimination when members of the Kulobi and Khodjenti clans, who were the "winners" of the conflict, made the jobs available (often for 12 Narodnaya Gazeta Nov. 14, 1992. 8 money) mainly to the members of their own clans (McLean and Greene). Thus, the conflict imposed significant hardships on some populations and enriched others. 9 CHAPTER 2: THE EFFECT OF ARMED CONFLICT ON ACCUMULATION OF SCHOOLING 1. INTRODUCTION The purpose of this paper is to examine the micro-economic impacts of civil conflicts, and in particular, to understand a link between armed conflict and accumulation of education. The findings reported in this paper indicate that while strong educational institutions and popular support for education matter for countries affected by civil strife, exposure to conflict may create or intensify regional and generational inequalities in educational attainment by decreasing educational opportunities for cohorts of individuals who were of school age during the war and who lived in the conflict affected regions. Previous research on the impacts of war on education examined cross-country (Stewart, Huang and Wang 2001; Ichino and Winter-Ebmer 2004) and regional differences (Merrouche 2006) in exposure to armed conflict and its impact on educational attainment. This study adds another dimension to the growing literature by examining the effect of the armed conflict in Tajikistan, a transition economy, on individuals' school attainment and enrollment by assessing differences across regions and cohorts in exposure to the conflict. The data on the events of the 1992-1998 Tajik armed conflict and two large household surveys, the 1999 and 2003 Tajik Living Standards Measurement Surveys, can be used to identify groups of individuals who were significantly exposed to the conflict. The educational attainment of such individuals is then compared to the attainment by individuals who did not suffer as much. Two empirical strategies are used for this comparison. The first strategy analyzes determinants of school enrollment and assumes random placement of individuals in affected and non-affected regions while controlling for observable characteristics. The second strategy employs a difference in differences approach to 10 determine the impact of regional and year of birth exposure to the conflict on the educational attainment of adults. If the effect of conflict on education is transient, we should observe enrollment rates rebounding to their pre-conflict levels and people catching up on years of education lost during the war. Conversely, if the effect of conflict is long-lived, we should expect a continuing downward trend or stagnation in average enrollment rates, in particular, in regions significantly affected by the conflict. To analyze the impact of the conflict on schooling, I use the 1999 and 2003 Tajikistan Living Standards Surveys (TLSS) collected by the State Statistical Agency of Tajikistan (Goskomstat RT) in collaboration with several international agencies. 13 The 1999 survey contains data on the past damage to homes during the civil war as reported in the answers to the household questionnaire. While only 6.8 percent of the 2000 households in the 1999 TLSS survey reported damage to their dwelling during the war, almost 40 percent of households surveyed in 1999 lived in a community where one or more dwellings were damaged. 14,15 These data, the household damage dwelling reports from the 1999 survey and household geographical location in high and low conflict intensity zones, described in detail in Appendix A, are used to identify individuals who lived in the conflict affected region. I assume that the past damage to the household's dwelling, other dwellings in a community, or residence in a high conflict intensity area reflect the household's exposure to economic and social hardships as a result of war. 13 The TLSS 1999 and 2003 data are publicly available for download at www.worldbank.org/lsms. 14 Appendix Tables 1 and 2. 15 The IMF (1998) reports that at least 36,000 homes were destroyed in the Tajik conflict. 11 Table 1 - Enrollment by gender and damage to household dwelling, ages 7-16. Household Damage Dwelling (HDD) Boys Girls Total Conflict affected region (HDD=1) 0.90 0.71 0.80 s.e. (0.03) (0.04) (0.02) % of sub-sample 7 8 8 Lesser affected region (HDD=1) 0.91 0.84 0.87 s.e. (0.01) (0.01) (0.01) % of sub-sample 93 92 92 Enrollment rate by gender 0.91 0.83 0.87 s.e. (0.01) (0.01) (0.01) N observations 1,806 1,815 3,621 Source: TLSS (1999). Author’s calculations My first estimation strategy is to evaluate the effect of exposure to conflict on school enrollment of children in the mandatory school age group, ages 7 to 15 in 1999 and ages 8 to 16 in 2003. The dependent variable is child's enrollment status in the academic year corresponding to the year of survey. The construction of the variables is discussed in detail in Appendix B. I use a rich set of observable characteristics to control for the potential selection effects. The findings suggest that there is a strong negative association between current school enrollment by girls and past damage to the household dwelling (Tables 1 and 2). This effect is particularly large for older girls, ages 12-15, who are 13 percent less likely to be enrolled if their household reported dwelling damage, but no effect is observed for boys. This finding implies that households affected by conflict played "safe" and invested more in schooling of boys. It is also possible that older girls were viewed by their families as more vulnerable to risks associated with conflict activity (such as rape and harassment), and therefore kept at home. High enrollment rates among younger children, ages 8-12, suggest that during the conflict, households attempted to protect the education of younger children by allowing them to complete at least primary school. 12 Among the household characteristics, educational attainments of parents, household per capita expenditure and mother’s widow status interacted with household damage dwelling have the greatest impact on child enrollment. Table 2 - Enrollment by gender and community damage dwellings, 1999 vs. 2003. TLSS 1999 sample (age 7-16) TLSS 2003 sample (age 8-16) Community damage dwelling (CDD) Boys Girls Total Boys Girls Total Conflict affected region (CDD=1) 0.90 0.79 0.84 0.90 0.85 0.88 s.e. (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) % of sub-sample 43 43 43 58 57 57 Lesser affected region (CDD=0) 0.92 0.86 0.89 0.92 0.89 0.91 s.e. (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) % of sub-sample 57 57 57 34 35 35 Missing CDD information: enrollment rate - - - 0.95 0.91 0.93 s.e. - - - (0.01) (0.01) (0.01) % of sub-sample - - - 8 8 8 N observations 1,806 1,815 3,621 3,403 3,263 6,666 Source: TLSS (1999), TLSS (2003). Author’s calculations. Since the datasets used in this paper are two separate cross-sections, I can not observe the same individuals or households across years. Nor do the data allow me to study the dynamics of drop-outs and re-enrollments by the same individuals and their progress towards completion of mandatory education during and after the conflict. My second estimation strategy is to investigate the impact of the Tajik war on the completion of the mandatory nine years of schooling. I focus on the educational attainment by adults who reached age 17 by June-July 2003 and thus, should have completed nine years of schooling by the time of the 2003 survey. This second estimation strategy allows me to use a difference in differences approach, observe the medium to long-term impact of the war on schooling and extend my analysis to individuals of younger ages who would have just completed their mandatory schooling by the time of the 2003 survey. All information on 13 adults is linked to the district (raion) level data on the regional exposure to the conflict during their schooling years. The findings suggest that individuals who were of school age during the war were significantly less likely to complete their mandatory education by age 17 than individuals who had an opportunity to complete this education level before the start of the conflict in 1992. The probability of completing 9 grades is 4 and 7 percent lower for boys and girls respectively. Further, the probability decreases by another 5% for girls, born in 1978-1986, who also lived in regions affected by the conflict during their schooling years. Potential migration during the war should not affect my estimates as 99.5 percent of individuals in the analytical dataset report that they lived in the region of survey continuously since birth. 16 The rest of the paper is organized as follows. Section 2 discusses relevant literature. Section 3 describes general trends in enrollment rates in Tajikistan. Section 4 focuses on data and descriptive statistics on school enrollment and grade attainment by conflict exposure. Section 5 discuses empirical approaches used in this paper. The results are presented in Section 6 and conclusion in Section 7. 2. EVIDENCE FROM OTHER COUNTRIES There are very few academic studies that document the effect of armed conflict on education. Most of such studies use cross-country data and analyze trends in aggregate enrollment rates as a part of larger projects on social impacts of conflicts. However, the country-level data are not suitable for estimating an impact of armed conflicts on particular regions and population groups within a country. 16 There is no significant difference in the migration rates of individuals from regions affected and non-affected by the conflict (TLSS 2003). 14 This study uses individual and household level data to explore the effect of armed conflict on individual human capital accumulation, as measured by school enrollment and completion of mandatory nine grades of education. The study aims to answer three questions. First, is there a strong (presumably negative) association between armed conflict and individual’s school attainment or enrollment? Second, if there is such a relationship, then does conflict have a long-lasting or a temporary impact on school attainment? Third, what are the characteristics of the individuals who were affected the most? To give answers to those questions, we need to understand the channels through which armed conflict can affect accumulation of human capital. In theory, armed conflict can affect schooling of individuals through the following four channels. First, civil wars and conflicts may reduce expected returns to schooling. In particular, returns may fall significantly for some elements of the populations. Lower returns to schooling may motivate decisions to stop attending school either temporarily or permanently. Second, armed conflicts reduce resources available to many households. An unexpected decrease in income may induce households to withdraw their children from school in an attempt to maintain current levels of consumption. Third, infrastructure is often destroyed in the course of internal wars, and schools and educational facilities are often targeted by militants. Some communities may be affected significantly. Children from such communities would have to travel to schools elsewhere or stop attending school altogether. If schools are not rebuilt within a reasonable period of time, we may observe some communities falling behind the rest of the country in their educational attainments. Fourth, civilians are often terrorized by armed forces and militias. Violence and feelings of insecurity may induce households to keep children away from public places, go into hiding or relocate. 15 Let’s explore these channels in more detail. First, conflict-induced societal and economic changes may alter the lifestyle of various populations. Due to destruction of industries, job opportunities for skilled labor may become scarce. If a conflict were based on religious grounds, winners may establish a regime that prevents women from working (example, Taliban). Assuming that households are rational and forward looking, they would redistribute their resources away from investments with lower returns. Second, exposure to conflict often unexpectedly reduces the financial resources available to many households. Jacoby and Skoufias (1997) and Thomas et al (2004) examined the impact of income shocks on household decisions with respect to schooling in peaceful environments. In India, agricultural households use seasonal school non-attendance by children and child labor as a form of self-insurance in the lean times (Jacoby and Skoufias). Similarly, in Indonesia, many households had to decrease their spending on education after the 1998 financial crisis (Thomas et al.). While the overall decrease in spending was small, it was large for the poor households with a high proportion of young children. Poor households, as authors conjecture, favored education of older children who were close to graduation. These older kids were able to finish their education at the expense of the younger ones. These studies assume that households would like to invest in the education of their children, but, when facing an unexpected income shock have to trade-off future for present consumption. When the shock hits, households withdraw children from school and send them to work to maintain current consumption levels. Once the crisis is over, households may expect to return to their previous state and re-enroll children in school. Thus, income uncertainty may adversely affect the quality and quantity of children's education. In some situations, households may have difficulty in reallocating resources because they have so little already. Such families may have to reduce consumption of 16 foodstuffs for all and children who were very young during the conflict may suffer from cognition problems as adults. Thus, the conflict may have a long legacy. The social legacy of the conflict becomes even more profound when we remember that data often allow us to observe information on only those individuals who survive the conflict. Armed conflicts are commonly accompanied by an increase in mortality and morbidity. For example, Hoeffler and Reynal-Querol (2003) find that infant mortality increases by 13 percent above baseline during a typical five year war and remains 11 percent higher in the first five years after the conflict. Ghobarah, Huth and Russett (2003) observe from cross-country data that the incidence of some infectious diseases and conditions increases as well. The authors attribute this increase to two factors. First, civilians find it more difficult to maintain health in the poor after-war conditions. Second, governments often have little money to spend on the public health and infrastructure, such as safe water supply. Poor individual health or loss of family may create serious restraints on access to schooling. There is a growing body of literature studying the impacts of losing parents on child well-being. In one such study, Evans and Miguel (2007) find that young children in rural Kenya are more likely to drop out of school after the parent's death and that effect is particularly strong for children who lost their mothers. Third, during civil wars households and communities have to deal with physical destruction of property and infrastructure. Internal wars force people to flee for safety. Refugees or internally displaced persons may find themselves in remote areas without any educational facilities. In more developed areas, massive influxes of refugees usually overcrowd local health and schooling facilities. Schools often serve as temporary shelters for refugees and internally displaced persons thus preventing students from accessing those facilities. Access to schooling may also be cumbersome for refugee and internally displaced 17 children if they lack the documents and residency permits required for enrollment. 17 Even if people do not relocate, schools are often specifically targeted for destruction or used as living quarters by parties in armed conflicts and refugees. The availability and quality of school facilities has been linked to student attendance and achievement (Glewwe 1994). Fourth, internal wars can have an impact by increasing fears of violence and insecurity. Violence against civilians has become a modern tool of warfare. 18 "Ethnic cleansing", torture, kidnappings and targeted killings are very common in conflict-affected countries. In the atmosphere of terror, households may attempt to protect their most vulnerable members, by keeping them at home or sending them away to relatives in more secure places. Children exposed to conflict may experience severe psychological effects that continue long after the war is over. They may become depressed and socially withdrawn. 19 Their school performance may be adversely affected and they may have to leave their studies prematurely. However, the effect of exposure to the conflict on children may be felt stronger in areas other than psychological health. Blattman (2006), who studies former child soldiers in Northern Uganda, finds that the labor market productivity and educational outcomes of children-abductees suffered as they missed out on acquiring labor market skills and education during the years they spent with the rebel forces. 17 See Mooney and French (2005) for the discussion of barriers of education for internally displaced persons (IDP) children and country examples. 18 Azam and Hoeffler (2002) theorize that civilians in internal wars are terrorized for two reasons. First, soldiers may terrorize civilians because they need to loot to augment their resources. Second, violence against civilians may play a direct military role. Large fractions of civilians fleeing for safety will leave local fighting forces without support. Thus militants would have difficulty in securing shelter in de-populated areas and would be easier to spot. 19 Many children affected by conflicts report high levels of distress, which are associated with serious disorders such as post-traumatic stress disorder (PTSD), depression or anxiety. The level of distress is directly linked to the severity of trauma or event children experienced during the war. While the levels of distress decline over time, many children remain severely affected for a long period of time (Yule et al. 2003). See Turner, Yuksel and Silove (2003) for a more general discussion of psychological responses to violence and torture in wars and conflicts. 18 Further, the conflict may have specific gender impacts. Girls may be withdrawn from school much earlier and married off to lift the burden from their families. Also girls may stay at home to avoid sexual assaults and harassment on their way to school (Machel 1996). Even if girls complete education, they may not be able to work outside their households, either because fewer opportunities are available in general or because society starts to frown at families who let their women engage in outside employment. For those reasons, we should expect a certain redistribution of household resources towards their more mobile and employable members –boys. The arguments above support the hypothesis that internal armed conflicts can lead to a decrease in school enrollment and school attainment, and in particular affect the education of vulnerable groups. So, can something be done to prevent or control the damage to education caused by internal wars? Recent research by Miguel and Roland (2004) and Stewart, Huang and Wang (2001) suggests that countries with strong institutions are able to maintain their enrollment rates during a conflict or recover to their pre-conflict enrollments once the conflict is over. A study by Stewart et al (2001) of African countries affected by internal armed conflicts finds that primary school enrollments decreased only in three out of eighteen countries, but improved in five during civil conflicts. On average, girls fared better than boys, which is not surprising given that boys often serve in the army. Lopez and Wodon (2005) report that school enrollment rates in Rwanda that decreased during the conflict returned to their pre-conflict levels within five years after the end of the conflict. A similar effect is found by Miguel and Roland who study the long-term impact of US bombing on the economic development in Vietnam and by Thomas et al. (2004) during the economic crisis in Indonesia. Despite such recoveries, conflicts and crises may still have an effect on children who were pulled out of school during calamities and did not have an opportunity to resume their studies. 19 In the next two sections, I will explore the relationship between armed conflict and school enrollments in Tajikistan. 3. BACKGROUND As I mentioned in the first chapter of my dissertation, the labor market in Tajikistan was affected by the conflict significantly. The few opportunities that were available before the conflict started, disappeared during the war due to economic difficulties, absence of fuel and communications and due to institutional constraints that prevented those who did not win in the conflict from access to good jobs later on. When the job market does not look good, in many developed countries and some transition economies (for example, Kyrgyzstan) many students return to school or stay in school for longer. For example, the recent economic downturns in the U.S. in 2002 and in 1990s coincided with high increases in enrollments in higher education (Callan 2002). However, in Tajikistan the enrollment rates at all levels of education were below their 1991 level over the course of the war and declined below the levels observed in most of the neighboring countries in Central Asia (see Figures 1 and 2). This decrease in enrollment by Tajik students could be attributed to both, political instability and economic concerns. The trends in enrollment are discussed in Section 3. 20 Figure 1 - Central Asia: basic education gross enrollment rates, 1989-2003 75.0 85.0 95.0 105.0 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 Enrollment Rate (% of ages 7-15) Kazakhstan Kyrgyzstan Tajikistan Uzbekistan Turkmenistan Source: UNICEF (2005). 3.1 General education and enrollment rates in Tajikistan I begin with the description of the trends in enrollment rates since the independence of Tajikistan in 1991 and proceed with a discussion of trends in school enrollments and number of years of schooling completed using the 1999 and 2003 TLSS data. 20 I focus on the distribution of enrollment rates across ages, gender and exposure to conflict for children of age 7 to 15. After the dissolution of the Soviet Union in 1991, Tajikistan inherited a strong school system. Like other FSU countries, Tajikistan mandates nine required grades of schooling for children age 7 to 15 and provides schooling up to grade eleven free of charge. Strong support for education is reflected in high secondary and primary enrollment rates that 20 The data used in this analysis are discussed in Appendix C. 21 are comparable to enrollments in other Central Asian countries (Figure 1). Despite the strong educational system and the popular support for education, enrollment rates in Tajikistan started to decline soon after it's independence in 1991. While we observe decline in overall enrollment rates across all countries in the ex-Soviet Central Asia in 1991-1998 periods, Tajikistan is the only country that was affected by the armed conflict in this region. Further, Tajikistan was one of few countries in the Eastern Europe and Former Soviet union regions where enrollment of girls decreased. Survey (LSMS and DHS) data suggest that the ratio of females to males in higher education was below one, in Kosovo (0.75), Uzbekistan (0.85) and Tajikistan (0.37). Similarly, enrollment ratios of girls' to boys' in the secondary education (ages 15-17) were also low in this set of countries: 0.78, 0.90 and 0.76 for Kosovo, Uzbekistan and Tajikistan respectively (Paci 2002). Returning back to Figure 1 we can observe that in Tajikistan the non-enrollment rate of children ages 7-15 increased by 10 percentage points between 1991 and 1993. 21 Over the same period, enrollment by young adults, ages 15-18, in the general secondary education levels dropped by 10 percentage points and continued to decline until 1998, with an overall decline of 20.9 percentage points between 1991 and 1998 (Figure 2). The initial decline in enrollments coincided with the first and most brutal years of the Tajik civil war. The enrollment rates at all levels of education began to recover in 1997-1998. This recovery may be partially explained by the Peace Agreement between the Tajik government and the United Tajik Opposition (UTO) that was signed in June 1997. The decrease in the enrollment rates over the 1991-1998 years can be further decomposed into regional and gender differences. 21 The enrollment rates are for September 1 st of the respective year. UNICEF (2003) 22 Figure 2 - Tajikistan: Enrollment trends, 1989-2003. 0 10 20 30 40 50 60 70 80 90 100 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 Enrollment (% of relevant age group) General Secondary (15-18) Vocational/Technical (15-18) Higher education(19-24) Basic education ( 7- 15 year olds) Civil war: 1992-1998 Peace Agreement June 1997 Source: Based on data from UNICEF (2005). First, during the war, some regions of Tajikistan, such as Khatlon and Regions of Republican Subordination (RRS) and the country capital Dushanbe were significantly affected by the conflict, while other regions, such as Leninobod and GBAO enjoyed relative stability because they were geographically isolated from the conflict affected areas. 22 The war, fighting and surge in criminal activity disrupted schooling and affected children’s experiences. Here are three examples. First, in the city of Kurgan-Tube and Kurgan-Tube raion in Khatlon region the official start of the academic year 1992-1993 was delayed by as much as 2 months. Even when the schools were officially open in November of 1993, parents did not send their children for education as they were concerned about their safety. 22 Leninobod region is connected to the rest of Tajikistan by a narrow road that is easy to block. The pass was blocked during the war. Gorno-Badakshon region is located in a mountainous area which is difficult to access. 23 Second, in Dushanbe, the government sent students of professional technical institutions for early winter holidays from November 13, 1992 to January 4, 1993. 23 The holidays were further extended to February 1 st , 1993. This forced school holiday was motivated by the low attendance of students and teachers due to unstable situation in the republic capital. 24 Third, schools and professional technical institutions were repeatedly targeted by military formations for hostage taking. Two separate incidents of such attempts occurred in Dushanbe in October of 1992. 25 In other regions, with less media coverage, many such events may have gone unreported. Second, within the affected regions there was a variation in the exposure to conflict across communities. Some people lived on the front lines while others enjoyed relative stability. Out of 136 households that reported damage dwelling in the TLSS 1999 data (6.8 percent of the sample of 2,000 households), 63 were located in Khatlon, 54 in RRS, 11 in Dushanbe and 8 in Leninobod (Sugd) regions. This amounts to 8.95, 12.5, 6.25 and 1.32 percent of households surveyed in each of these regions respectively. Correspondingly, the enrollment rates for children ages 7-15 are lower for areas with the most damage dwelling reports and stand at 88.27, 86.61, 83.78, 91.85 and 95.56 percent for Khatlon, RRS, Dushanbe, Leninobod and GBAO respectively. Third, the lower enrollment in the conflict affected areas can be partially explained by the school quality and damage to school facilities. Here is one example of such damage to a school in Lenin Yuri, Kurgan-Tube region (Khatlon): "In the gymnasium of Lenin Yuri's school, the concrete beams are fully exposed; the wooden floor they supported was ripped and carried away. The basketball backboards were plundered... All desks, tables and blackboards were stolen…" (Bonners 1993). 23 Narodnaya Gazeta, Oct. 26, 1992. 24 Narodnaya Gazeta, Nov. 13, 1992 and Jan. 23. 1993. 25 Narodnaya Gazeta, Oct. 15, 1992 and Oct. 16, 1992. 24 It is possible that students were likely to drop out if their classes were overcrowded and if quality of school facilities was very poor. The IMF (1998) estimates that approximately 20 percent of schools in Tajikistan were destroyed beyond repair in the conflict and that many teachers fled war affected areas. From Figure 3 we can observe that student per teacher ratio is increasing over the 1991-2002 academic years across Tajikistan. Decomposing this ratio, we find that, in the conflict affected areas, the mean (average by raion) number of students increased by 22 percent from 23,930 to 30,850 students over the 1991-2003 period, while the average number of teachers declined by almost 23 percent from 1,850 to 1,440 thousands over the 1991-1996 period. The number of teachers never approached its pre-war level during that time. In the contrast to that, in the lesser affected regions, both, the number of students and teachers, did not change as much over the 1991- 2002 academic years. Further, these differences in enrollment rates across regions can be explained by a greater danger in the conflict areas to the children. For example, parents from Gharm (RRS region) community were concerned that older girls would be harassed or abused by soldiers at checkpoints on their way to school. This concern was legitimate as rape was often used by warring parties in the conflict (U.S. Department of State 1994). Similarly, in Western Khatlon children of Gharmi and Pamiri origins reported fears of physical violence and of being beaten by other children as the main reason for skipping school (Falkingham 2000). 26 26 Pamiri and Gharmi ethnic groups or clans were strongly associated with supporting opposition forces. During the war, adults, whose passports indicated that they were born in Pamir or Garm regions, were killed or taken away by Narodnii Front or government associated militias and disappeared. Human Rights Watch (1994) reports that in the late December of 1992 in Dushanbe, Narodnii Front militias killed 300 and taken away hundreds of people (unfortunately the data used in this paper does not allow for identification of various ethnic groups and clans in Tajikistan). 25 Figure 3 - Students per teacher, by academic year and reports of conflict activity 0 2 4 6 8 10 12 14 16 18 20 1991-92 1992-93 1993-94 1994-95 1995-96 1996-97 1997-98 1998-99 1999-00 2000-01 2001-02 students per teacher Lesser affected region Conflict affected region Academic year Source: Based on data from State Statistical Committee of Tajikistan (2001, 2004) Notes: general education schools. Forth, the impact of war varied across households living in the same communities and by region. In the south, where the war was ravaging communities, it was common for Tajiks who belong to different clans to live in the same town or village. When the war started, residents of such communities commonly took different sides in the conflict. Members of Gharmi and Pamiri clans supported the opposition, while Kulobi and Khodzhenti (or Leninobodi) people backed up the Tajik government forces. There are numerous reports of houses owned by Gharmi and Pamiri people burned and destroyed and houses of Kulobi left standing in the Kulobi communities and the other way around. Gharmi clans during the Soviet time were relatively prosperous as compared to other Tajiks in the same region. This prosperity may have bred resentment and led to Gharmies being targeted during the war, their villages burned and houses destroyed (Walker 2006). Targeted 26 destruction of property of non-majority community members was very common during the Tajik civil war (McLean and Green 2003; Human Rights Watch 1993). Thus, decreased family incomes, low school quality and surrounding violence may have induced many children to leave school. To further our understanding of the impact of armed conflict in Tajikistan on schooling, we should compare enrollment rates and number of grades completed among children across and within regions, by age and gender. I explore these differences in the following section. 4. DATA This study is based on the 1999 and 2003 Tajik Living Standards Surveys (TLSS). Both surveys are a part of the Living Standards Measurement Survey Data Collection Project. The 1999 survey is the first nationally representative household survey conducted in Tajikistan after its independence in 1991. The survey was carried out only six months after the end of the armed conflict in November 1998 and, therefore, allows us to examine the short-term impact of exposure to conflict on school enrollment. From the 2003 data, I can make inferences about a longer term, cumulative impact of exposure to conflict on schooling of young adults who were exposed to the conflict during their schooling years. 4.1 Exposure to conflict, school enrollment and grade attainment This section explores the effect of differential exposure to conflict on school enrollment of and number of grades completed by individuals using the 1999 and 2003 Tajik Living Standards surveys. Prior to examining the schooling attainment by individuals by region, gender, age and exposure to the conflict, I outline the measures of conflict exposure used in this paper. 27 For the detailed description of the construction of conflict variables and the wording of the related survey questions refer to Appendix A. 4.1.1 Measures of exposure to armed conflict Three variables are used to compare the extent of conflict exposure on school enrollment and school grades completed by individuals. The first variable indicates direct effect of the conflict on an individual household. This measure is based on households' reports of damage to their own dwellings (henceforth, household damage dwelling or HDD) in the 1999 TLSS. 27 As I mentioned above, 136 households or 6.8 percent of households interviewed for the 1999 TLSS reported that their dwelling was damaged during the recent civil unrest. The second variable is recorded at the primary sampling unit level for the 1999 and at the raion level for the 2003 data. It is an indicator variable that is assigned a value of one for all households in a primary sampling unit in the 1999 TLSS if one or more households from this unit reported that their dwelling was damaged during the war (henceforth, community damage dwelling or CDD). In the construction of this variable, I assume that households who lived in the same locality as households that reported damage dwelling were affected by the conflict in the similar way. The third conflict variable is available only at the raion level for both 1999 and 2003 datasets. This variable indicates that a raion (district) was severely affected by a various conflict events. Such incidents include high levels of conflict and insurgent activities, episodes of violence and atrocities against the civilian population in Tajikistan between 1991 and 1998 years that are recorded at the raion level (henceforth, reports of conflict activity or 27 Using the 1999 TLSS household data, I found no apparent relationship between the type of construction material used in dwelling construction and reports of damage to household dwelling. 28 RCA). The variable is based on the news reports in local Tajik newspapers, reports of the UN agencies, the U.S. Department of State, human rights organizations and other literature on the Tajik civil war. This measure identifies raions within Tajikistan that sustained significant damage during the war and timing of such damage. 28 The above three measures of exposure to conflict are used to compare school enrollment rates and number of grades completed by individuals across communities that were impacted by the conflict to a different extent. 4.1.2 School enrollment and grade attainment by conflict exposure Tables 1 and 2 present enrollment rates by gender and degree of conflict exposure for children in the mandatory age groups using the 1999 and 2003 TLSS data. 29 Three observations emerge from Table 1. First, on average boys were 8 percent more likely to be enrolled than girls. 30 Second, only 71 percent of girls from households that reported household damage dwelling were enrolled as compared to 84 percent of girls from households that did not report such damage. Third, boys from households that reported damage dwelling were almost as likely to be enrolled as boys from households that did not reports such damage. A similar effect is observed when the community damage dwelling dummy is used to compare enrollment rates across sub-groups of children. From Table 2 we can observe that girls from communities exposed to conflict were 7 and 6 percent (significant at 1 percent level) less likely to be enrolled in 1999 and 2003 as compared to girls from communities where residents did not report damage dwelling during the conflict. 28 A possible limitation of this variable is that it may not include all communities that were affected during the war because the accounts of conflict activity may not discuss smaller incidents or less known communities. For example, the correlation coefficient between CDD and RCA variables is equal to 0.46. 29 Refer to Appendix B for the detailed definition of enrollment status. 30 t-statistic= - 6.856, p (2, 3621)=0.000. 29 Further exploring enrollment rates across different age groups, gender and degree of conflict exposure in Tables 3, 4, 5 we can make another four observations. First, younger children are more likely to be enrolled than older ones. Second, at all ages boys have higher enrollment rates than girls. Third, children from households and communities significantly affected by the conflict are less likely to attend school. The impact of an individual conflict shock as measured by damage dwelling variable is particularly high for older girls, ages 12- 15, who were 20 to 31 percent less likely (significant at 5 percent level) to be enrolled than girls from households that did not experience dwelling damage (Table 3). Fourth, we can also observe an increase in the enrollment rates by all children, and a large recovery in enrollment rate by girls in 2003 as compared to 1999. Table 3 - Enrollment rates by age, gender and household dwelling damage, 1999. Girls Boys Dwelling Damage Dwelling Damage age Reported Not p-value Reported Not p-value 7 0.80 0.78 0.90 0.80 0.77 0.87 8 0.79 0.90 0.16 1.00 0.93 0.32 9 0.77 0.92 0.06 0.90 0.95 0.48 10 0.94 0.92 0.78 0.89 0.94 0.45 11 0.82 0.91 0.23 1.00 0.93 0.40 12 0.70 0.90 0.01 1.00 0.94 0.33 13 0.62 0.85 0.03 0.93 0.91 0.79 14 0.60 0.85 0.02 0.80 0.91 0.25 15 0.42 0.73 0.01 0.81 0.89 0.33 16 0.73 0.55 0.27 0.84 0.79 0.59 N 149 1666 131 1675 Source: TLSS (1999). 30 Table 4 - Enrollment rates by age, gender and community damage dwelling, 1999. Girls Boys CDD CDD age Reported Not p-value Reported Not p-value 7 0.79 0.77 0.83 0.66 0.88 0.04 8 0.86 0.91 0.31 0.94 0.94 0.90 9 0.93 0.90 0.44 0.93 0.96 0.31 10 0.94 0.91 0.50 0.91 0.95 0.28 11 0.89 0.92 0.49 0.93 0.93 1.00 12 0.80 0.93 0.00 0.98 0.93 0.10 13 0.79 0.86 0.16 0.89 0.92 0.45 14 0.74 0.91 0.00 0.88 0.92 0.42 15 0.58 0.77 0.00 0.86 0.90 0.35 16 0.51 0.61 0.20 0.77 0.81 0.52 N 777 1038 770 1036 Source: TLSS (1999). Table 5 - Enrollment rates by age, gender and community damage dwelling: 2003 Girls Boys CDD (raion level) CDD (raion level) age Reported Not p-value Reported Not p-value 8 0.93 0.94 0.66 0.93 0.96 0.32 9 0.96 0.96 0.90 0.97 0.96 0.75 10 0.95 0.99 0.09 0.94 0.99 0.03 11 0.94 0.95 0.75 0.96 0.98 0.27 12 0.94 0.98 0.06 0.96 1.00 0.02 13 0.84 0.92 0.03 0.95 0.93 0.51 14 0.86 0.93 0.06 0.95 0.97 0.39 15 0.80 0.91 0.01 0.90 0.93 0.36 16 0.67 0.81 0.01 0.85 0.91 0.15 N 1856 1150 1964 1165 Source: TLSS (2003). Thus, the data suggest a negative association between conflict exposure and school enrollment, for girls, ages 12-15. The impact of conflict on enrollment is decreasing with time as shown by higher enrollment rates for 2003. This could be explained by more stable situations in the regions and reconstruction of schools by the government, international and non-governmental organizations (Rashidov c.2000). 31 As enrollment rates are measured at a point in time, or the time of the survey, they do not provide us with information of whether children, who were not in school at the time of the survey, are also behind in school as measured by their grade attainment. Figure 4 - Boys: mean school grades completed (up to 9) by year of birth and conflict exposure 8.00 8.20 8.40 8.60 8.80 9.00 9.20 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 Mean years of schooling completed (0-9) lesser affected region (CDD=0) conflict affected region (CDD=1) Should have completed 9 grades before the war (by Sept. 1991) born in 1966-1973 In school (ages 7-15) during the war: born in 1976-1986 Omitted group: born in 1974-1975 Notes: Year of birth: 1966-1986. CDD - community damage dwelling at the raion level. Source: TLSS (2003). Author’s calculations. Patterns in grade attainment by children ages 7-15 are very similar to those found in the enrollment data (Tables 6 and 7). First, in 1999 children from war-affected households and communities completed fewer grades of schooling than the rest of the sample and this gap increased with age. By 2003 (Table 8), boys of age 15 and older from conflict affected areas completed 0.23 (significant at 5% level) fewer grades of schooling than boys of the same age from regions not affected by the conflict. Similarly, girls from conflict affected regions completed 0.26 fewer grades of schooling than girls from non-affected areas (significant at 5% level). Third, in each age group boys consistently complete more years of education 32 than girls, which should not be surprising as we observed that boys are more likely to be enrolled than girls. Figure 5 - Girls: mean school grades completed (up to 9) by year of birth and conflict exposure. 8.00 8.20 8.40 8.60 8.80 9.00 9.20 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 Mean years of schooling completed (0-9) lesser affected region (CDD=0) conflict affected region (CDD=1) Should have completed 9 grades before the war (by Sept. 1991) born in 1966-1973 In school (ages 7-15) during the war: born in 1976-1986 Omitted group: born in 1974-1975 Notes: Year of birth: 1966-1986. CDD - community damage dwelling at the raion level. Source: TLSS (2003). Author’s calculations. Looking at the longer-term impact of the conflict measured by the mean grade attainment by adults (grades 0 to 9) born in 1966-1986 (Figures 4 and 5), we observe a downward trend in the grades attained by the cohort of women who were of school age during the conflict (born in 1976-1986). While the largest drop in grade attainment as compared to older cohorts is observed for those born in 1974-1975, there is an increasing gap in grade attainment between those who live in conflict affected regions and those who live in lesser affected areas. However, we do not find the same effect for men. 33 Table 6 - Mean school grades completed by age, gender and household damage dwelling, 1999 Girls Boys Damage dwelling Damage dwelling age Reported Not p-value Reported Not p-value 7 0.60 0.82 0.47 0.60 0.71 0.63 8 1.32 1.27 0.84 1.21 1.35 0.65 9 1.54 2.08 0.07 1.60 2.17 0.06 10 3.24 3.11 0.64 2.72 3.06 0.20 11 3.71 3.84 0.67 3.78 4.00 0.59 12 4.50 4.96 0.13 5.13 5.02 0.73 13 4.31 6.10 0.00 5.21 6.05 0.04 14 5.53 7.13 0.00 6.90 7.08 0.73 15 7.06 8.12 0.00 7.06 8.12 0.00 16 8.64 8.57 0.90 8.95 8.88 0.83 N 149 1666 131 1675 Source: TLSS (1999). Author’ calculations. Note: ages 7-16. Table 7 - Mean school grades completed by age, gender and community damage dwelling: 1999 Girls Boys CDD CDD age Yes No p-value Yes No p-value 7 0.75 0.83 0.63 0.55 0.84 0.02 8 1.24 1.30 0.72 1.31 1.36 0.72 9 2.00 2.08 0.60 2.03 2.23 0.14 10 3.16 3.10 0.68 2.91 3.14 0.13 11 3.71 3.92 0.23 3.83 4.09 0.16 12 4.85 4.96 0.55 4.98 5.07 0.58 13 5.76 6.15 0.08 5.81 6.10 0.17 14 6.76 7.23 0.05 6.76 7.32 0.02 15 7.28 7.86 0.04 7.78 8.21 0.04 16 8.52 8.61 0.75 8.70 9.04 0.06 N 777 1038 770 1036 Source: TLSS (1999). Author’ calculations. Note: ages 7-16. CDD – Community Damage Dwelling at the primary sampling unit level. 34 Table 8 - Mean school grades completed by age, gender and community damage dwelling: 2003 Girls Boys CDD CDD age Reported Not p-value Reported Not p-value 8 1.58 1.58 0.25 1.59 1.53 0.63 9 2.37 2.42 0.67 2.38 2.60 0.11 10 3.30 3.36 0.56 3.45 3.41 0.86 11 4.15 4.23 0.58 4.31 4.28 0.84 12 5.30 5.26 0.75 5.44 5.28 0.14 13 5.80 6.10 0.10 6.22 6.24 0.87 14 6.75 7.05 0.16 7.11 7.22 0.53 15 7.58 8.02 0.03 7.80 8.21 0.04 16 8.30 8.67 0.14 8.72 9.87 0.06 N 1855 1150 1964 1165 Source: TLSS (2003). Author’ calculations. Note: ages 7-16. CDD – Community Damage Dwelling at the raion level. 4.2 Robustness In this section the reports of conflict activity (RCA) variable is used to compare child school enrollment rates and number of school years completed for children across communities. This measure allows us to keep all observations for 1999 and 2003, as compared to community damage dwelling variable where we loose about 8 percent of the 2003 sample because it is not possible to match all raions surveyed in 1999 to raions surveyed in 2003. Table 9 - Enrollment by gender and reports of conflict activity (RCA). TLSS 1999 sample (ages 7-16) TLSS 2003 sample (ages 8-16) Enrollment rates by Conflict Activity (RCA) Boys Girls Total Boys Girls Total Conflict activity reported (RCA=1) 0.90 0.81 0.85 0.94 0.88 0.90 s.e. (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) % of sub-sample 48 50 49 51 51 51 No records (RCA=0) 0.92 0.85 0.88 0.95 0.93 0.94 s.e. (0.01) (0.01) (0.01) (0.00) (0.01) (0.00) % of sub-sample 52 50 51 49 49 49 N observations 1,806 1,817 3,623 3,098 2,957 6,055 Source: author's calculations. TLSS 1999 and 2003. 35 This third measure of conflict exposure provides results comparable to those found for the household and community damage dwelling variables. The results are provided in Tables 9, 10, and 11. Again, the differences in enrollment rates between affected and non- affected regions are large and statistically significant at 5 percent level for girls ages 14-15. Enrollment rates for both, boys and girls, noticeably improved in four years that passed between the surveys. For example, 85 percent of 14 year old girls from conflict affected regions were enrolled in 2003 as compared to only 77 percent of girls age 14 in 1999. Girls from less affected regions were more likely to be enrolled in 2003 as compared to 1999. As expected the improvement in enrollment rates is higher for the conflict affected areas because the lesser affected regions already had rather high enrollment rates in 1999. The next section describes identification and regression framework used to determine if a negative relationship between the exposure to conflict and enrollment rates holds when we control for individual, household and community characteristics. 36 Table 10 - Enrollment by age, gender and reports of conflict activity (RCA). Panel A: Children ages 7-15, TLSS 1999 Girls Boys Conflict activity Conflict activity age Reported Not p-value Reported Not p-value 7 0.77 0.78 0.94 0.74 0.80 0.60 8 0.87 0.90 0.49 0.92 0.96 0.35 9 0.95 0.87 0.03 0.93 0.98 0.11 10 0.92 0.92 1.00 0.92 0.95 0.29 11 0.90 0.91 0.67 0.94 0.92 0.64 12 0.88 0.88 0.87 0.95 0.94 0.87 13 0.82 0.84 0.76 0.91 0.91 1.00 14 0.77 0.89 0.02 0.90 0.91 0.86 15 0.58 0.79 0.00 0.89 0.89 1.00 16 0.55 0.58 0.69 0.78 0.82 0.49 N 901 916 933 873 Panel B: Children ages 8-16, TLSS 2003 Girls Boys Conflict activity Conflict activity age Reported Not p-value Reported Not p-value 7 8 0.92 0.94 0.55 0.93 0.95 0.58 9 0.95 0.98 0.35 0.95 0.99 0.09 10 0.96 0.98 0.20 0.95 0.98 0.10 11 0.93 0.97 0.17 0.96 0.97 0.65 12 0.97 0.95 0.17 0.97 0.98 0.47 13 0.86 0.91 0.12 0.95 0.94 0.73 14 0.85 0.93 0.01 0.94 0.97 0.23 15 0.79 0.91 0.00 0.90 0.94 0.17 16 0.66 0.81 0.00 0.90 0.87 0.39 N 1518 1439 1580 1518 Source: TLSS (1999), TLSS (2003). 37 Table 11 - Grades completed by age, gender and reports of conflict activity (RCA). Panel A: Children ages 7-15, TLSS 1999 Girls Boys Conflict activity Conflict activity age Reported Not p-value Reported Not p-value 7 0.81 0.78 0.89 0.65 0.77 0.34 8 1.29 1.26 0.84 1.17 1.53 0.02 9 2.10 1.98 0.39 2.04 2.28 0.07 10 3.03 3.23 0.18 2.79 3.31 0.00 11 3.69 3.95 0.13 3.98 4.00 0.91 12 4.88 4.95 0.70 4.99 5.06 0.64 13 5.87 6.10 0.29 5.68 6.26 0.01 14 6.83 7.18 0.14 7.08 7.07 0.97 15 7.37 7.87 0.07 7.89 8.19 0.15 16 8.55 8.59 0.87 8.65 9.17 0.00 N 901 916 933 873 Panel B: Children ages 8-16, TLSS 2003 Girls Boys Conflict activity Conflict activity age Reported Not p-value Reported Not p-value 7 8 1.47 1.57 0.18 1.64 1.54 0.15 9 2.29 2.51 0.16 2.39 2.55 0.19 10 3.34 3.33 0.46 3.46 3.40 0.06 11 4.13 4.21 0.55 4.22 4.37 0.36 12 5.30 5.22 0.22 5.42 5.33 0.32 13 5.82 6.04 0.13 6.31 6.12 0.11 14 6.59 7.19 0.01 7.62 7.19 0.54 15 7.62 7.95 0.02 7.76 8.26 0.00 16 8.30 8.63 0.07 8.90 9.43 0.63 N 1518 1439 1580 1518 Source: TLSS (1999), TLSS (2003). 38 5. IDENTIFICATION AND EMPIRICAL SPECIFICATION My study extends the previous research on the determinants of school enrollment in low- income countries and contributes to the literature on household and individual behavior in countries affected by internal armed conflicts. For the purposes of this paper, I rely primarily on detailed information on household and individual demographic characteristics, expenditure, school enrollment and community characteristics. In order to identify the effect of the Tajik armed conflict on school enrollment and completion of the mandatory 9 grades of schooling, this paper takes on two empirical approaches. In both approaches my identification strategy exploits variation in the conflict intensity over the regions and time. 5.1 Empirical strategy 1: school enrollment My first empirical strategy involves the estimation of the impact of exposure to conflict on school enrollment of children in the mandatory school age group. 31 My analytical datasets contain information on 3,284 children of ages 7-15 living in 1,435 households from the 1999 TLSS and on 6,055 children ages 8 to 16 living in 2,838 households from the 2003 TLSS. All variables used in this analysis are based on the information from the interview of the reporting adult in the household. To measure the impact of exposure to conflict in a regression framework, I estimate equation (1) below. The estimation framework controls for individual and household characteristics as well as year of birth effects. The decision on whether to enroll in a particular year has been modeled as a dummy variable that is equal to one if a child is enrolled in school and zero otherwise. In the model, 31 Ages 7 to 15 as of May 1999. See Appendix C for the details on the construction of the analytical samples, lists of variables and comparison of sample means between sub-samples of individuals residing in conflict affected and lesser affected households and communities. 39 I assume that households maximize their utility subject to a budget constraint and that the maximized utility value could be observed from household choices (Becker 1975). 32 Results are reported separately for boys and girls. As parents may have a differential preference for their sons' and daughters' education and market return to education may be different for boys and girls, education of boys and girls can be determined by different production functions (Rosenzweig and Schultz 1982; Strauss and Thomas 1995). 33 (1) ijk j i i k ijk u C D M c E + + + + + = 3 2 1 1 1 η η η β where E ijk is a binary variable indicating whether an individual i born in region j in year k was enrolled in school during the respective academic year at the time of the survey; D i is a dummy indicating whether the individual's dwelling was damaged during the war (I also use community damage dwelling variable to estimate the impact of exposure to the conflict at the community level); β 1k - age fixed effects; M i is a vector of individual and household specific socioeconomic characteristics (such as education of parents, household monthly expenditure per capita, value of household assets); C i is a vector of community of residence specific characteristics (such as availability of employment and access to education). To control for unobserved correlation of observations within localities, I estimate equation (1) with fixed effects at two community levels. First, I specify fixed effects at the primary sampling unit. Second, I compare these results with specification of fixed effects at the raion (district) level of the child's current residence. In both specifications, I estimate regressions with robust standard errors to control for the effect of unobserved heterogeneity on variance. The fixed-effects model purges all observed and unobserved community 32 Refer to Becker (1975) for the full details of this model. 33 This assumption is tested by estimating a regression equation on the pooled data for boys and girls. In this specification, all control and explanatory variables are interacted with an indicator for child’s gender. The hypothesis that estimated parameters are identical for boys and girls is rejected (F(16, 124) = 3.66, p-value = 0.000). 40 characteristics that are constant across individuals from the same community, thus removing the bias in the estimation of enrollment that is caused by child-invariant community characteristics. It is also possible that local characteristics are correlated with the war damage variable, and the fixed effects specification helps us to control for this correlation. Since the enrollment variable is a binary measure, it would be appropriate to estimate a probit or logit specification of the model (1). To compare the estimates from the reduced form OLS model, I re-estimate the model (1) above using probit specification with clustering at the raion and population point levels and robust standard errors. The results are qualitatively similar to the results received from the OLS specifications (results not reported). The regression specification (1) allows us to examine determinants of schooling and the impact of war damage variable. However, the results may not capture the long-term impact of exposure to conflict on final school attainment by children as we study the determinants of the enrollment rates at the particular point in time or May 1999 or June 2003, when the surveys were conducted. Some children may have left the school for good, while others returned to school after the conflict was over. The second empirical approach allows us to examine the long-term impact of the conflict on education by studying the completion of mandatory nine grades of schooling by adults. 41 5.2 Empirical strategy 2: completion of mandatory schooling My second specification examines the effect of exposure to conflict on the completion of mandatory schooling by adults and uses a difference-in-difference estimation strategy. To identify an individual's exposure to conflict I use a sample of men and women born between 1966 and 1986 and link an adult's education with the district (raion) data on the exposure to the conflict in the region where individual went to school. Exposure to the conflict is examined at the regional and cohort levels, and is determined by the date of birth and the region of residence during schooling. Using a difference-in-differences strategy I compare educational attainment by adults who should have completed their mandatory schooling before the war to the educational attainments by adults who were of school age during the war. Following Duflo (2001), equation 2 is specified as follows: (2) ijk j i k ijk K P c E ε γ β + + + = 2 1 1 ) ( where S ijk is a binary variable, =1 if individual completed nine grades of schooling, and zero otherwise. 34 Subscripts on the dependent variable denote individual i residing in the region j and born in year k. K i is a dummy indicating whether the individual i belongs to a young "exposed" cohort in the sub-sample. α 1j is a fixed effect for the individual’s region of residence during schooling. β 1k is a cohort of birth fixed effect. P j is the intensity of the conflict in the district of residence during schooling (measured by community damage dwelling or reports of conflict activity). 35 I did not include controls for the household characteristics in my regression specification as characteristics of the household of origin for each individual are not available from the TLSSs. The surveys include information only on 34 Most of the variation in the number of grades completed (0-9) is between 8 and 9 grades. Therefore, using a binary variable is appropriate. 35 Only 0.5 percent in the sample reported that they did not live continuously in the current region of residence, so migration should not have an effect on our estimates. I tested this assumption by including a dummy for “migrant” in specification (2). I also estimated equation with interactions of this dummy with the exposed cohort and war damage dummies. Both, the migrant dummy itself and the interaction terms were not significant. 42 the current household of residence. However, this information on current living environment may not accurately reflect the conditions experienced by adults in their childhood since many individuals by the time of the survey had formed their own families. Families of grown-up sons often stay together with parents while girls usually move-in with families of their husbands. Thus we would not have information on the main characteristics of the household of origin for married women as they are likely to live with their in-laws. To estimate parameters in equation (2), I use the sample of adults, born between 1966 and 1986 or adults ages 17 and above from the 2003 TLSS. In this difference-in- differences framework, I focus on two groups of individuals, the treatment and control groups as defined below. In this dataset, young adults exposed to war were born between 1975 and 1986. The sample of individuals just below and just above the cutoff year of birth 1975 would be very similar. Comparing an average value of S ijk of those who are just above and just below the cutoff point would provide us with an estimate of the average treatment effect (Cameron and Trivedi 2005). I exclude the sample of individuals born between 1974 and 1975 as some of those individuals can belong either to the “treatment” or to “control” group as defined below and their inclusion in the analytical sample may bias estimates. The treatment group is a subset of individuals from the young or “exposed” group who were of school age during the conflict. This subset consists of individuals who were of school age during the war and who also lived in the regions affected by the conflict. Based on the age criteria alone, this group of individuals should have had an opportunity to complete the mandatory nine grades of schooling by June 2002. Those individuals faced the decision to enroll or continue attending school in the academic years during the period of war, starting with the academic year 1991-1992, and ending with the academic year 1998- 1999. Respectively, this group was born between 1976 and 1986. Further, based on the 43 residence criteria, the individuals should have lived in the conflict affected regions from February 1992 until January 1999. The control group is a group of individuals that should not have been affected by the conflict significantly during their schooling years. This control group includes two subgroups. The first subgroup contains individuals born between 1976 and 1986 who lived in the lesser conflict affected regions. The second subgroup includes individuals who were born between 1966 and 1975 or pre-war cohort. This older cohort, born in 1966 to 1975, had an opportunity to complete their mandatory education of 9 grades (usually completed by age 16) by the start of the conflict in 1991. The cohort born is older cohort was not exposed to the conflict at all during its mandatory schooling years. Also this cohort is relatively young and should have been exposed to a similar grade school system 36 as the cohort born after 1975. I trim the pre-war cohort data by 2 years and exclude those born in 1974-1975. Individuals born in 1974-1975 could belong either to “war” or “pre-war” cohort based on their month of birth and age when they started schooling (however, the dataset does not include information on the day and month of birth). Therefore, erroneously including them in one of the groups would bias the estimates of the difference-in-differences estimator. This strategy also allows us to exclude those individuals who may have left school in anticipation of the conflict. 37 Coding of variables and summary statistics for the sample are reported in Appendix C. 36 This assumption needs further investigation. 37 Trimming the “young” group by two years or omitting those born in 1976-1977 in addition to the group born in 1974-1975 does not change the results significantly. 44 6. RESULTS In this section I discuss the results from the reduced form regressions that examine the impact of conflict exposure on enrollment by school age children and completion of mandatory schooling by adults. For all regressions reported in this paper, I use the linear probability (LP) model with fixed effects specified at raion level with robust standard errors to control for heteroscedasticity. 38 The advantage of the LP model is that coefficients are easy to interpret in terms of probability of enrollment by children or completion of mandatory schooling by adults. 39 6.1 School enrollment First, I run several LP regressions for the sub-samples of boys and girls ages 7-15 as of May 1999 and ages 8-16 as of June-July 2003. Then, I divide samples of boys and girls further into primary and general education age groups, respectively ages 7-11 and 12-15, and re- estimate my initial regression specifications. Table 12 presents the first set of regressions where I sequentially introduce various controls for household and community characteristics. Observed household characteristics include number of male and female adults (ages 17-65) in a household, parental level of education, household monthly expenditure per capita and land ownership. Table 13 reports regression results separately for primary and general education age groups and includes the household damage dwelling as a main independent variable of interest. The first set of specifications in columns 1-3 in Panels A and B of Table 12 includes child and household characteristics and the household damage dwelling variable. Results 38 I also run regressions with fixed effects specified at the primary sampling unit level and the results were consistent with fixed effects at the raion level. I specified fixed effects at the raion level to maintain consistency between regressions of child enrollment and completion of mandatory schooling by adults. 39 As a robustness check, I also completed all analysis using the probit model. The results do not differ qualitatively (not reported). 45 across regressions in Table 12 indicate that the household damage dwelling is strongly and negatively associated with the enrollment of girls. Girls are 11 to 12 percent (significant at 1% level) less likely to be enrolled in the 1998-1999 academic year if their household reported damage to the dwelling. Point estimates are relatively stable and significant when I introduce various controls. This result relies on the assumption that there are no omitted time-varying and region specific effects correlated with the damage dwelling variable. Since some community attributes could be correlated with the community damage dwelling variable, using fixed effects allows me to exclude all time-invariant community characteristics. 46 Table 12 - HDD and determinants of school enrollment, by gender, ages 7-15. Panel A: Boys Panel B: Girls Variables and Controls (1) (2) (3) (1) (2) (3) Household damage dwelling (HDD) 0.028 0.030 0.028 -0.117*** -0.122*** -0.123*** [0.028] [0.029] [0.029] [0.040] [0.041] [0.041] HH owns > 0.1 hectare of land -0.055** -0.052* 0.062* 0.063* [0.028] [0.028] [0.036] [0.037] Distance to school -0.010* -0.013 [0.005] [0.008] Parent's education (years) Mother 0.008** 0.008** 0.009** 0.020*** 0.020*** 0.020*** [0.004] [0.004] [0.004] [0.004] [0.004] [0.004] Father 0.009*** 0.009*** 0.010*** 0.020*** 0.020*** 0.020*** [0.003] [0.003] [0.003] [0.004] [0.004] [0.004] N adults (ages 17-65) Females 0.007 0.008 0.008 0.013 0.013 0.014 [0.008] [0.008] [0.008] [0.010] [0.010] [0.010] Males -0.016** -0.016** -0.016** -0.016 -0.016 -0.016 [0.008] [0.008] [0.008] [0.010] [0.010] [0.010] Ln spline of household expenditure per capita (pce) (at median) ≤12,842 rubles 0.127*** 0.127*** 0.127*** 0.023 0.021 0.022 [0.029] [0.029] [0.029] [0.029] [0.029] [0.029] >12, 842 rubles -0.002 -0.002 -0.002 0.007*** 0.007*** 0.007*** [0.002] [0.002] [0.002] [0.002] [0.002] [0.002] Rural 0.044 0.069** 0.067** 0.042 0.017 0.016 [0.031] [0.033] [0.033] [0.030] [0.033] [0.033] Child's age 8 0.148*** 0.144*** 0.145*** 0.129** 0.128** 0.126** [0.055] [0.055] [0.055] [0.061] [0.061] [0.060] 9 0.173*** 0.169*** 0.170*** 0.167*** 0.167*** 0.168*** [0.053] [0.054] [0.054] [0.060] [0.060] [0.060] 10 0.151*** 0.146*** 0.147*** 0.171*** 0.172*** 0.171*** [0.053] [0.054] [0.053] [0.060] [0.060] [0.060] 11 0.160*** 0.156*** 0.157*** 0.143** 0.143** 0.144** [0.055] [0.055] [0.055] [0.060] [0.060] [0.060] 12 0.174*** 0.171*** 0.172*** 0.133** 0.133** 0.133** [0.054] [0.054] [0.054] [0.061] [0.060] [0.060] 13 0.147*** 0.144*** 0.149*** 0.075 0.076 0.076 [0.055] [0.055] [0.055] [0.062] [0.062] [0.062] 14 0.132** 0.128** 0.129** 0.074 0.075 0.074 [0.056] [0.056] [0.056] [0.062] [0.062] [0.062] 15 0.114** 0.112** 0.114** -0.044 -0.042 -0.041 [0.055] [0.055] [0.055] [0.065] [0.065] [0.065] 47 Table 12 (continued) - HDD and determinants of school enrollment, by gender, ages 7-15. Constant -0.596** -0.559** -0.559** 0.089 0.068 0.064 [0.268] [0.267] [0.266] [0.270] [0.270] [0.270] Observations 1580 1580 1580 1626 1626 1626 Number of raions (fe) 56 56 56 56 56 56 R-squared 0.08 0.08 0.08 0.11 0.11 0.11 Notes: Columns represent OLS coefficients. All regressions contain community fixed effects at the raion (district) level and controls for missing information on parents. Robust standard errors are in brackets. Reference group is "age 7". Source: TLSS (1999). * significant at 10%; ** significant at 5%; *** significant at 1%. With respect to socioeconomic status, the levels of education received by child's parents are positively and significantly associated with enrollment. For each additional year of educational attainment by girl's mother, the probability of child’s enrollment increases by approximately 2 percent. The effect of parental education is almost two times lower for boys. Coefficients on land ownership (owning above 0.1 hectare of land) have opposite signs for boys and girls. It appears that boys’ enrollment is negatively related to land ownership while girls are more likely to be enrolled if their family owns land, possibly reflecting that boys are the ones who work on the land. The coefficient on distance to school has an expected negative sign, but it is significant only for boys (at 10 percent level). We can infer that for each additional kilometer walked to school, the probability of school enrollment declines by one percent for boys and by 1.3 percent for girls. At the community level, residence in urban areas, where schools are usually better equipped, is negatively related to school enrollment in post-war Tajikistan. Thus, lower attendance may reflect a higher risk or fear of being accosted or harassed by the militants congregating in urban areas, in particular for girls. Alternatively, families in rural areas may have better access to financial and other resources for consumption smoothing. Such 48 families may engage in subsistence agriculture, and rely less on the outside income in comparison to urban families. With respect to age groups, Table 13 illustrates that the household damage dwelling variable is negatively associated with enrollment by girls in both primary and general education age groups. The coefficient on “household damage dwelling” variable remains negative and is significant at 5 percent level. The impact appears to be greater for older girls. Girls ages 12-15 are 14.2% less likely to be enrolled if they come from a household that suffered damage to it’s dwelling as compared to only 9.3% decline in enrollments for girls ages 7-11. This greater effect of the household damage dwelling on enrollment of older girls may be partially attributed to several factors. First, public safety situation in Tajikistan significantly deteriorated over the years. Thus, parents were reluctant to allow girls to attend school because of the concern that girls might be harassed by soldiers or militants on their way to school, especially during the Tajik war (Gomart 2003; Falkingham 2000). Second, those girls, ages 12-15, were of school age during the first, most violent and turbulent years of the Tajik armed conflict. It is possible that they, they may have experienced a greater disruption to their education than younger children (ages 7-11) who reached school age by 1995-1998, when the conflict intensity decreased. Third, older girls may be withdrawn from school to help their families, while younger girls remain in school for a few years. The coefficient on the natural logarithm of spline of the per capita household expenditure (spline at median of 12,846 rubles) suggests that older girls from families with income higher than median are 1.2 percent more likely to be enrolled. The coefficient on distance to school variable has an expected negative sign. With each kilometer walked to school the probability of enrollment decreases by 2.3 and 1.6 percent respectively for boys and girls, ages 7-11 (significant at 10% level). Surprisingly, distance to school has a negligible effect on school attendance by older girls. 49 Coefficients on parental education are almost two times larger for older children. Each year of education completed by mother increases probability of enrollment of younger girls by 1.6 percent as compared to 2.6 percent for girls ages 12-15. Thus, teenage daughters of a woman who completed 10 grades of education are 26 percent more likely to be enrolled than daughters of a woman without schooling. There are no significant changes in the sign or size of other regression coefficients. 40 Tables 14 and 15 report results of OLS regressions where I add community schooling (at the raion level) characteristics interacted with age dummies. 41 This addition does not change significantly sign and size of other regression coefficients. 40 I have also completed analysis in Tables 12 and 13 using the 2003 data for the sample of children ages 8-16. The results suggest that community damage dwelling and reports of conflict activity were significantly negatively associated with enrollment of girls ages 12-15 in the 2002-2003 academic year. However, the relationship between enrollment and conflict exposure variables becomes insignificant when I add community variables such as distance to school in each primary sampling unit and number of students per school in each raion in the 2002- 2003 academic year. 41 Unfortunately I could not include level schooling characteristics themselves as I would like to keep in the regression equations controls for fixed effects at raion (district) level. 50 Table 13 - HDD and determinants of school enrollment: by age group. Panel A: Boys Panel B: Girls age 7-11 age 12-15 age 7-11 age 12-15 Variables and Controls (1) (2) (1) (2) Household damage dwelling 0.06 0.01 -0.093** -0.142** [0.040] [0.047] [0.046] [0.068] HH owns > 0.1 hectare of land -0.049 -0.05 0.042 0.072 [0.039] [0.044] [0.044] [0.062] Distance to school -0.023** -0.004 -0.016* -0.010 [0.011] [0.004] [0.009] [0.014] Parent's education (years) Mother 0.012** 0.003 0.012** 0.026*** [0.006] [0.006] [0.006] [0.006] Father 0.007 0.014*** 0.016*** 0.022*** [0.004] [0.005] [0.005] [0.006] N adults ages 17-65 Females 0.006 0.010 0.013 0.014 [0.012] [0.011] [0.013] [0.017] Males -0.024* -0.006 -0.019 -0.017 [0.014] [0.009] [0.015] [0.015] Ln spline of household expenditure per capita (pce) (at median) ≤12,842 rubles 0.106*** 0.138*** 0.036 0.005 [0.041] [0.042] [0.037] [0.045] >12, 842 rubles 0.000 -0.003 0.003 0.012*** [0.003] [0.003] [0.003] [0.004] Rural 0.034 0.109** -0.007 0.037 [0.051] [0.043] [0.041] [0.052] Child's age 7 reference reference 8 0.145*** 0.114* [0.056] [0.059] 9 0.162*** 0.174*** [0.053] [0.059] 10 0.140*** 0.170*** [0.053] [0.059] 11 0.155*** 0.149** [0.055] [0.059] 51 Table 13 (continued) - Household damage dwelling and determinants of school enrollment, by gender and age Panel A: Boys Panel B: Girls age 7-11 age 12-15 age 7-11 age 12-15 Variables and Controls (1) (2) (1) (2) Child's age 12 reference reference 13 -0.031 -0.063* [0.026] [0.035] 14 -0.048* -0.066* [0.028] [0.034] 15 -0.059** -0.185*** [0.027] [0.038] Constant -0.323 -0.556 0.124 0.222 [0.376] [0.374] [0.339] [0.408] Observations 834 746 840 786 Number of raions (fe) 56 56 56 56 R-squared 0.09 0.07 0.07 0.14 Notes: Primary education - ages 7-11; general education - ages 12-15. Columns represent OLS coefficients. All regressions contain community fixed effects at the raion (district) level and controls for missing information on parents. Robust standard errors are in brackets. Primary education - ages 7-11; general education - ages 12-15. “HDD”- household damage dwelling. Source: TLSS (1999). * significant at 10%; ** significant at 5%; *** significant at 1%. In the first set of regressions reported in Table 14, age dummies are interacted with number of students per teacher in the 1997-1998 academic year. This set of schooling variables allows me to measure the impact of school quality in the year prior to the current year of enrollment. In the second set of regressions in Table 15, age dummies are interacted with number of students per teacher in the academic year when a child was supposed to enter school if this child were to start attending school at age 7. For example, children who were age 9 in 1999 should have started attending a primary school in the academic year 1997- 1998. The main purpose of including these interactions is to add community controls to the regressions that are region and time specific and evaluate the impact of the conflict dummy on school enrollment. 52 Table 14 - Enrollment and students/teacher ratio in the 1997-1998 academic year. Dependent variable: Child school enrollment in the 1998-1999 academic year ages 7-15 ages 7-11 ages 12-15 Boys Girls Boys Girls Boys Girls Covariates (1) (2) (3) (4) (5) (6) 0.024 -0.129*** 0.057 -0.099** 0.001 -0.148** Household damage dwelling [0.029] [0.041] [0.039] [0.047] [0.049] [0.070] Interactions of age dummies with number of students per teacher in the 1997-1998 academic year age 8 X spt97 0.008 -0.011 0.007 -0.005 [0.015] [0.016] [0.015] [0.016] age 9 X spt97 0.009 -0.007 0.012 0.000 [0.015] [0.015] [0.015] [0.015] age 10 X spt97 0.019 -0.011 0.020 -0.003 [0.015] [0.016] [0.015] [0.016] age 11 X spt97 0.021 -0.019 0.023 -0.009 [0.016] [0.016] [0.016] [0.016] age 12 X spt97 0.018 -0.011 [0.015] [0.016] age 13 X spt97 0.008 -0.020 -0.013* -0.008 [0.016] [0.017] [0.008] [0.009] age 14 X spt97 0.003 -0.027 -0.017** -0.015 [0.016] [0.017] [0.008] [0.009] age 15 X spt97 0.018 -0.033* -0.002 -0.020** [0.016] [0.017] [0.007] [0.010] Constant -0.559** 0.009 -0.356 0.105 -0.534 0.091 [0.270] [0.273] [0.375] [0.344] [0.380] [0.415] Observations 1562 1604 824 829 738 775 Number of raions (fe) 55 55 55 55 55 55 R-squared 0.09 0.12 0.10 0.07 0.08 0.14 Joint F test: interactions=0 Prob > F 0.053 0.094 0.026 0.838 0.089 0.161 Notes: “sptXX” – number of students per teacher in 19xx-19(xx+1) academic year. Columns represent OLS coefficients. All regressions contain community fixed effects at the raion (district) level. Robust standard errors are in brackets. Reference group is "age 7" for ages "7-15" and "ages 7-11". "Age 12" is the reference group for "ages 12-15". All regressions include age dummies and controls for education of parents, missing information on parents, log spline household expenditure per capita, a dummy for owing more land the median of 0.1 hectare, number of male and female adults ages 17-65 in a household; rural location and distance to school. Source: TLSS (1999). * significant at 10%; ** significant at 5%; *** significant at 1%. 53 Table 15 - Enrollment and students/teacher ratio in an academic year of child’s initial enrollment. Dependent variable: Child school enrollment in the 1998-1999 academic year ages 7-15 ages 7-11 ages 12-15 Boys Girls Boys Girls Boys Girls Covariates (1) (2) (3) (4) (5) (6) 0.015 -0.115*** 0.052 -0.104** -0.003 -0.132* Household damage dwelling [0.031] [0.043] [0.042] [0.049] [0.050] [0.074] Interactions of age dummies with N students per teacher in the year when student was eligible for enrollment age 8 X spt98 -0.006 0.014** -0.014* -0.002 [0.005] [0.007] [0.007] [0.009] age 9 X spt97 -0.003 0.019*** -0.009 0.000 [0.006] [0.006] [0.007] [0.009] age 10 X spt96 0.003 0.013* -0.003 -0.004 [0.006] [0.007] [0.007] [0.010] age 11 X spt95 0.007 0.003 0.003 -0.008 [0.005] [0.006] [0.006] [0.008] age 12 X spt94 0.004 0.013** [0.005] [0.006] age 13 X spt93 -0.005 0.002 -0.009 -0.005 [0.007] [0.007] [0.007] [0.009] age 14 X spt92 -0.003 0.006 -0.008 -0.003 [0.006] [0.008] [0.007] [0.008] age 15 X spt91 0.001 -0.003 -0.003 -0.010 [0.005] [0.008] [0.007] [0.008] Constant -0.529* 0.107 -0.483 0.126 -0.504 0.124 [0.287] [0.287] [0.398] [0.373] [0.401] [0.420] Observations 1442 1463 782 776 685 708 Number of raions (fe) 51 51 52 52 51 51 R-squared 0.07 0.12 0.11 0.08 0.06 0.14 Joint F test: interactions=0 Prob > F 0.252 0.069 0.042 0.716 0.513 0.658 Notes: “sptXX” – number of students per teacher in 19xx-19(xx+1) academic year. Columns represent OLS coefficients. All regressions contain community fixed effects at the raion (district) level. Robust standard errors are in brackets. Reference group is "age 7" for ages "7-15" and "ages 7-11". "Age 12" is the reference group for "ages 12-15". All regressions include age dummies and controls for education of parents, missing information on parents, log spline household expenditure per capita, a dummy for owing more land the median of 0.1 hectare, number of male and female adults ages 17-65 in a household; rural location and distance to school. Source: TLSS (1999). * significant at 10%; ** significant at 5%; *** significant at 1%. The negative sign of the regression coefficients for the interactive terms in columns 3-6 in Tables 14 and 15 suggest that overcrowded schools (higher number of students per 54 teacher) had a negative impact on enrollment of both, boys and girls in the primary and general education school groups. In Table 15, the effect of initial school conditions on school enrollment is significant for the pooled sample of girls, ages 7-15 (at 10% level) and for both, boys and girls ages 7-11 (at 5% level). The effect of “damage dwelling” variable on school enrollment remains to be significant and rather large when we control for the interactions between school quality and age dummies. Again the “damage dwelling” variable has a negative impact only on enrollment of girls. Analyzing the impact of interactions of selected household variables with “damage dwelling” variable (Table 16, Panel B, column 3), we observe that girls, ages 7-15, were 22.8 (significant at 5% level) percent less to be enrolled in school if their mother was a widow and their household dwelling was damaged during the war. 42 A widow may have to assume a full-time job once her husband is dead and withdraw girls from school so they can help with chores. As in Tajikistan most of the housework chores are considered to be in women’s domain, keeping boys out of school for the household chores may be out of the question. The individual impact of the damage dwelling variable on enrollment decreases to 8% from 13% (for girls) after adding to the regression a variable constructed by interacting widow status of a child’s mother with “damage dwelling” dummy. Further, studying the impact of other interactions between household damage dwelling and family characteristics, we find that the regression coefficient on the interaction between damage dwelling and mother’s education is positive (0.022) and significant at 5% level. This suggests that mothers who completed more grades of education make a better use of available resources 42 In addition to the interactions given in Table 13 I also estimate regressions with interactions of damage dwelling variable with household expenditure per capita, number of adults in a household, distance to school and residence in the rural area. The estimated coefficients for those interaction terms are not significant. 55 even under difficult circumstances and supports the hypothesis for the intergenerational transmission of human capital. Table 16 - Interaction of damage dwelling variable with selected covariates, ages 7-15, by gender. Dependent variable: Child school enrollment in the 1998-1999 academic year Panel A: boys Panel B: girls Covariates (1) (2) (3) (1) (2) (3) 0.021 -0.179 -0.176 -0.080* -0.328*** -0.235** Household damage dwelling (HDD) [0.031] [0.163] [0.163] [0.044] [0.114] [0.120] 0.022 0.021 0.022** 0.016 Education of mother * HDD [0.016] [0.016] [0.011] [0.011] Mother is a widow 0.017 0.016 -0.028 -0.027 [0.056] [0.056] [0.063] [0.063] Mother is a widow * HDD 0.070 0.051 -0.228** -0.207** [0.056] [0.059] [0.103] [0.104] Constant -0.556** -0.553** -0.550** 0.017 0.054 0.013 [0.267] [0.266] [0.267] [0.271] [0.270] [0.271] Observations 1580 1580 1580 1626 1626 1626 Number of raions (fe) 56 56 56 56 56 56 R-squared 0.08 0.08 0.08 0.12 0.11 0.12 F test: interaction with HDD =0 Prob > F 0.209 0.162 0.279 0.026 0.048 0.011 Notes: Columns represent OLS coefficients. All regressions contain community fixed effects at the raion (district) level. Robust standard errors are in brackets. Reference group is "age 7". All regressions include age dummies and controls for education of parents, missing information on parents, log spline household expenditure per capita, a dummy for owing more land the median of 0.1 hectare, number of male and female adults ages 17- 65 in a household; rural location and distance to school. Source: TLSS (1999). * significant at 10%; ** significant at 5%; *** significant at 1%. Overall, the regression results reported in this section confirm that there is indeed a significant negative relationship between individual household damage dwelling and school enrollment of girls. This effect is strong and non-trivial. Further, girls ages 12-15 are less likely to be enrolled if their mother is a widow and their dwelling was damaged during the war. These results suggest that households who were individually affected during the conflict are more likely to pull older girls out of school to conserve resources or to ensure their safety. 56 6.2 Completion of mandatory schooling The second part of my analysis examines whether the conflict had a long-lasting impact on educational attainment by adults who were of school age during the armed conflict in Tajikistan. In this analysis I use a subset of data on education of adults from the 2003 TLSS. The 2003 TLSS provides a relatively complete set of individuals who were in the school age group during the conflict and who were old enough to finish their general education of nine grades by the time of the survey. I estimate equation (2) with raion level fixed effects using data on adults age 17 to 37 from the 2003 TLSS and present my results in Tables 17 and 18. Table 17 - Probability of completing nine grades of schooling and exposure to the armed conflict. Men: Panel A Women: Panel B (1) (2) (1) (2) CDD *(Born in 1976- 1986) -0.017 -0.034 [0.018] [0.021] RCA *(Born in 1976- 1986) 0.018 -0.054** [0.017] [0.022] Born in 1976-1986 -0.040*** -0.052*** -0.075*** -0.069*** [0.013] [0.012] [0.014] [0.012] Constant 0.953*** 0.954*** 0.920*** 0.925*** [0.007] [0.007] [0.008] [0.008] Observations 3211 3533 3522 3868 Number of raions (fe) 55 63 55 63 R-squared 0.01 0.01 0.02 0.02 F test: interaction term=0 Prob > F 0.345 0.299 0.114 0.012 Note: Coefficients of the interactions between cohort dummies and individual's exposure to the conflict in the region of schooling, by gender. Columns represent OLS coefficients. All regressions contain community fixed effects at the raion (district) level. Robust standard errors are in brackets. Reference group is "born in 1966- 1973". Born in 1974-1975 is excluded from the regression sample. Source: TLSS 2003. * significant at 10%; ** significant at 5%; *** significant at 1%. Two observations emerge from the analysis of Table 17. First, coefficients on the interaction terms in the linear probability regressions (“year of birth 1976-1986” interacted 57 with the residence in the conflict affected region variable) are negative and significant for girls. Thus, a girl who was of school age during the war and who attended school in a high conflict region was approximately 12% less likely to complete the mandatory nine grades of schooling than a girl who completed her schooling before the conflict started. Second, coefficients on the cohort term (born in 1976-1986 or of school age during the conflict) are negative and significant for both boys and girls. Clearly, individuals who were of school age during the conflict were less likely to complete 9 grades of schooling. The effect is greater for girls. The probability of completing nine grades of schooling is approximately 4.5 and 6.9 percent lower for boys and girls born in 1976-1986 as compared to the older cohorts. In Table 18 the analysis is repeated with the year of birth dummies. The point estimates for the interaction terms (regional exposure to conflict*of school age during the war) remain significant and are rather stable. The above results can be interpreted as a causal effect of the exposure to war, under an assumption that in the absence of the conflict activity in the exposed regions, the probability of completion of nine grades would not have been systematically different between the regions with high and low exposure to the conflict. This result depends on the assumption that there are no omitted time-varying and region specific effects correlated with the regional conflict measures. The cross-sectional TLSS 2003 does not contain individual background variables such as parental education, household of origin expenditure and assets. Time-series information on regional characteristics for the pre-independence period is also not available. Otherwise, it would have been informative to study the effects of interaction of the conflict variables with parental education, and stratify the sample by education of mother and/or household’s of origin assets and income. 58 Table 18 - Probability of completing nine grades of schooling: individual year dummies. Men: Panel A Women: Panel B (1) (2) (1) (2) CDD *(Born in 1976- 1986) -0.016 -0.034 [0.019] [0.021] RCA *(Born in 1976- 1986) 0.017 -0.054** [0.017] [0.022] Year of birth 1967 -0.008 -0.013 0.006 0.001 [0.033] [0.031] [0.035] [0.033] 1968 0.034 0.033 -0.007 -0.003 [0.031] [0.027] [0.037] [0.035] 1969 0.015 0.015 -0.011 -0.014 [0.032] [0.029] [0.035] [0.034] 1970 0.014 0.008 -0.003 0.005 [0.031] [0.029] [0.031] [0.030] 1971 -0.009 -0.018 -0.021 -0.008 [0.031] [0.030] [0.036] [0.033] 1972 0.03 0.023 0.016 0.024 [0.027] [0.026] [0.032] [0.031] 1973 0.024 0.019 0.004 0.006 [0.027] [0.024] [0.031] [0.029] 1976 0.007 -0.011 -0.030 -0.050 [0.030] [0.028] [0.035] [0.034] 1977 -0.043 -0.059* -0.083** -0.069* [0.034] [0.031] [0.039] [0.036] 1978 -0.046 -0.061** -0.075** -0.062* [0.034] [0.031] [0.036] [0.034] 1979 -0.014 -0.028 -0.099*** -0.086** [0.034] [0.031] [0.038] [0.036] 1980 -0.019 -0.032 -0.063* -0.048 [0.031] [0.028] [0.034] [0.032] 1981 -0.050 -0.067** -0.075** -0.065* [0.033] [0.031] [0.036] [0.034] 1982 -0.040 -0.054* -0.136*** -0.123*** [0.034] [0.030] [0.037] [0.035] 1983 -0.012 -0.030 -0.073** -0.064** [0.030] [0.027] [0.033] [0.031] 1984 -0.055 -0.068** -0.094*** -0.078** [0.033] [0.030] [0.034] [0.032] 1985 -0.020 -0.033 -0.033 -0.023 [0.029] [0.026] [0.032] [0.030] 1986 -0.022 -0.041 -0.085** -0.084*** [0.029] [0.027] [0.033] [0.031] 59 Table 18 (continued) - Probability of completing nine grades of schooling: individual year dummies. Men: Panel A Women: Panel B (1) (2) (1) (2) Constant 0.940*** 0.945*** 0.922*** 0.923*** [0.023] [0.021] [0.024] [0.023] Observations 3211 3533 3522 3868 Number of raions (fe) 55 63 55 63 R-squared 0.01 0.01 0.02 0.02 Notes: Columns represent OLS coefficients. All regressions contain community fixed effects at the raion level. Robust standard errors are in brackets. Reference group is "born in 1966". Born in 1974-1975 is excluded from the regression sample. Source: TLSS (2003). * significant at 10%; ** significant at 5%; *** significant at 1%. 7. DISCUSSION AND CONCLUSION My study takes a step towards understanding the impact of internal civil wars on school enrollment and completion of mandatory schooling by individuals. Two empirical approaches are employed to investigate this question. I use two separate cross-sectional datasets from Tajikistan. My first empirical strategy evaluates the impact of the conflict on school enrollment by children of ages 7-15. The second empirical strategy employs a difference-in-differences approach to determine whether the exposure to conflict affected the probability of completion of mandatory schooling by adults who were of school age during the conflict. The results suggest that there is a strong negative relationship between the exposure to conflict and school enrollment of girls. In particular, older girls, ages 12-15 were significantly less likely to be enrolled if their dwelling was damaged during the war. The point estimates for damage dwelling variable remain stable and significant when controlling for community fixed effects, community variables and important household and individual characteristics. The enrollment rates among boys are not significantly negatively associated with household measures of exposure to the conflict. 60 As we know from Section 2, armed conflict can lower the economic returns to schooling, reduce household resources, inflict damage to schools infrastructure and expose individuals to threats of physical violence. As a result, armed conflict can adversely affect school enrollment. The findings in this paper may be explained on the basis of the above mentioned factors. First, households facing uncertainty in the wake of the war were more inclined to investing in the education of boys and withdrawing girls from school. It is possible that the expected returns to investment in education of girls were lower in areas affected by conflict. For example, Gomart (2003) who worked on a poverty assessment project in Tajikistan in 1996 indicates that women and unskilled workers in rural areas and small cities were severely affected by unemployment. Many factories that employed primarily women closed their rural and regional branches, leaving women looking for jobs in agricultural sector with minimum wages. Second, families optimized their reduced resources by keeping younger children in school and rationed access to education for older children (World Bank 2005). Third, the results of regressions that include controls for one of the measures of the educational quality, namely, the number of students per teacher in the raion of residence, indicate that while more crowded schools detracted children from enrollment, the effects of damage to individual household dwelling remained strong and significant. Fourth, some children were exposed to the conflict for a longer period of time. For example, schooling years of children born in 1985-1987 significantly overlapped with the years of the conflict. Those children were of school age (age 7 and above) at the peak of fighting in 1992-1993. Thus, the long period of exposure and severity of the first years of war should have disrupted their education to a greater degree than the education of younger or older children. 61 The results from the difference-in-differences regressions reveal that while exposure to conflict reduced the probability of completion of mandatory schooling for all individuals in the cohort that was of school age during the conflict, the effect was greater for girls from conflict affected regions. Since we do not observe the same result for boys, unavailability or destruction of schools and other education related infrastructure in the regions affected by conflict may not explain this result entirely. Also, the regression analysis does not allow us to determine with certainty whether this difference in educational attainment by girls from regions affected and not affected by the conflict has always been present, or if it has emerged in response to economic and social shocks associated with conflict because time-series data on schooling availability and quality in Tajikistan are not available for the pre-independence years. However, lower level of school enrollment and completion by older girls maybe partially explained by the heightened level of violence and crime activity in Tajikistan, in particular, rape and bride-kidnapping of young girls (as discussed in the second paper of this dissertation). Let me repeat the questions I raised earlier. Who is affected by the conflict? Is it a long-term effect? What can be done? The results indicate that the exposure to conflict had a large, negative and lasting effect on the education of girls who were of school age during the conflict and lived in the conflict affected areas. Such girls were less likely to be enrolled and had a lower probability of completing the mandatory nine grades of schooling. These findings suggest that there are long-term effects of the conflict on education. While the enrollment rates in Tajikistan started to rise soon after the end of the war, the enrollments in the conflict affected areas were still below those in the less affected areas. What can Tajik government do to increase school enrollment by girls? - Anecdotal evidence indicates that older girls in Tajikistan are less likely to attend school if classes are 62 taught by male teachers and/or if there are few other girls of their age attending school. 43 Thus, it may be advisable to establish schools only for girls at the higher levels of education and make school attendance by older girls more acceptable in some conservative communities. To increase school enrollment and keep children in school until graduation, the Tajik government should strive to restore stability as quickly as possible, thereby addressing the war-induced instability and distress that society may still have as a result of the conflict. Given that this instability was possibly instrumental in restricting access to education, particularly for girls, the approach suggested above may bring girls back to school. It is also advisable to provide specific incentives for girls to attend schools. Such incentives could include improving the quality and safety of school facilities and quality of education, and providing secure transportation. 44 43 From a conversation with the education specialists of the UNDP Tajikistan (September 2006). 44 Many schools in Tajikistan do not have separate bathrooms for boys and girls. School bathrooms are often of very poor quality as well. Many schools still operate in 2-3 shifts and for those shifts many classes may end late at night. 63 CHAPTER 3: THE EFFECT OF ARMED CONFLICT ON THE MARRIAGE MARKET AND FEMALE REPRODUCTIVE BEHAVIOR 1. INTRODUCTION Crises and civil wars inflict large burdens on country’s population. War-related mortality and morbidity patterns (Hoeffler and Reynal-Querol 2003; Ghobarah, Huth, Russett 2003) and the relationship between violent conflict, child mortality and female reproductive behavior (Hoeffler and Reynal-Querol; Verwimp and van Bavel 2005; Lindstrom and Berhanu 1999; McGinn 2000) have been addressed by researchers. This study adds another dimension to the literature by exploring the link between violent conflict and the marriage market and female reproductive behavior. This paper combines a unique dataset on the events during the 1992-1998 armed conflict in Tajikistan with individual and household data from the 2003 Tajik Living Standards Measurement Survey. 45 The conflict data were collected from the centrally published Tajik daily and weekly newspapers, reports on the status of human rights in Tajikistan and studies of the post-independence period by academics and non-governmental organizations (such as Human Rights Watch, World Bank, United Nations, U.S. Department of State and others). These survey and conflict data are used to examine the impact of the 1992-1998 armed conflict on the marriage market for and reproductive choices made by women in Tajikistan. Demographic and regional data from the State Statistical Agency of Tajikistan are also extensively used in this paper. 45 www.worldbank.org/lsms. The surveys were collected by the World Bank and the State Statistical Agency of the Republic of Tajikistan. 64 In 1991 crude marriage rates started to decline in all five former Soviet Union countries in Central Asia (TransMONEE 2007). 46 In most of the region the decline stopped by 1994. In Tajikistan the marriage rates continued to decrease until 1998. Further, the crude marriage and age-specific marriage rates declined by much more than in the other four neighboring countries. 47 The rates reached their lows in 1998, with the crude and age specific marriage rates dropping to only 38 percent of the 1990 levels of 22.4 and 9.5 marriages per 1,000 of population respectively. The marriage rates subsequently increased in 2001-2005, but were still well below the 1990s levels. In addition to severe economic hardships associated with the conflict, the war also had a significant demographic impact on the population. More than 50,000 men perished in the Tajik armed conflict. During the first years of war, the mortality rate due to injuries among young adults, ages 15-19, increased by 225 percent compared to the 1989 levels (Figure 6). Also the mortality among boys in the age group 5-14 increased substantially during the same period and was higher than that of females of the same age. The combined effects of war-related mortality among adults and natural population growth may have led to a shortage of men in the prime marriageable age group (as defined further). In this paper I attempt to address an important set of questions, such as the extent to which the economic, social and demographic shocks affected the demographic processes such age at first marriage and first birth. Related questions include the following. Have poverty and devastation in the regions affected by war induced the population to postpone marriage and child-bearing? To what extent did the decrease in the number of men of marriageable age affect the marriage market for women? Was there a differential response 46 Crude marriage rate is number of marriages per 1,000 mid-year population. 47 An age-specific marriage rate is equal to the number of marriages per 1,000 mid-year population ages 15-44. 65 in the demographic processes due to the differential regional and temporary impact of the conflict? What other factors affected marriage and first birth decisions during that period? Figure 6 - Tajikistan: mortality trends among children and young adults, 1989-1999 0 20 40 60 80 100 120 140 160 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 Mortality Rate (per 100,000 relevant population) Due to injuries (aged 15-19) Mortality (aged 15-19) Females (aged 5-14) Males (aged 5-14) Source: Based on data from UNICEF (2005). Ages 5-19. This paper is concerned with estimating these relationships, with an emphasis on the traditions and newly emerging trends in family and marriage institutions in Tajikistan. Marriage remains a central part of Tajik culture. Majority of population enter marriage by age 30. In 2000, 20 percent of women ages 20-29 were not married as compared to 15 percent in 1989 (State Statistical Committee 2002). The proportion of unmarried men aged 20-29 was 42 and 32 percent in 2000 and 1989 respectively. However, only three percent of women and men ages 30-39 were not married in 2000. The mean age at first marriage increased for men from 24.3 to 26.7 years and for women from 21.5 to 23.0 years between 1989 and 2005 (TransMONEE 2007). The results in this paper are consistent 66 with an increased proportion of unmarried men and women ages 20-29 reported above. The findings point towards delayed marriage for women from the younger birth cohorts, in particular for those who live in the conflict affected regions. The proportion of women who had their first child by age 18 increased significantly for women born in 1975-1983 as compared to women born in 1966-1974. The analysis of transition data indicates that most of the differences in age at first birth are accounted for by the above mentioned cohort differences in age at first marriage. The sex ratio of men to women in the prime marriage age groups did not have a significant effect on the rate of entry into first marriage by women. This result is attributed to the limited relevance of the population based sex ratios in Tajikistan, where many marriages are concluded between members of the same lineage or traditionally defined regional group. However, the sex ratio variable has a significant negative effect on the age differences between husbands and wives. The rest of the paper is organized as follows. Section 2 discusses lessons from other countries and the social and economic background on the marriage market in Tajikistan. Section 3 describes the data. Section 4 presents empirical estimation. Section 5 summarizes findings. Section 6 concludes and discusses future work. 2. LESSONS FROM OTHER COUNTRIES AND BACKGROUND 2.1 The marriage market I continue with discussion of three competing but not unrelated theories that may illustrate the potential impact of civil wars and the related economic, social and demographic changes on the marriage market for women and their reproductive behavior. 67 2.1.1 Economic shocks and the marriage market Armed conflicts inevitably lead to a decrease in the resources available to many households for consumption. The effect of decrease in household resources on entry into marriage is not clear-cut. There are multiple factors that may contribute to increased or decreased entry into marriage during the hard economic times. First, if economic crises occur periodically, households may devise ex-ante strategies that would help them to smooth consumption over the time. Benefits to marriage due to specialization in labor and household production that were highlighted by Becker (1973) may become particularly valuable if these benefits allow to smooth consumption when there is insufficient or non-existent access to credit (Kotlikoff and Spivak 1981; Rosenzweig and Stark 1989). Thus, marriage may be used to secure access to networks of well-to-do relatives who may provide assistance in the hard times. However, the ex-ante arrangements may not be available for unexpected shocks such as wars and armed conflicts. Marriage of daughters may be also used as a self-insurance strategy. In developing countries, poor families may use unmarried daughters as assets and “cash” them in during the crisis (Hoogeveen et al. 2004). Hoogeveen et al. study the impact of shocks to household income on entry into marriage in Zimbabwe. They find that the marriage rate was higher for girls from poorer households and those poorer households had a higher rate of time preference with regard to the settlement of bride-wealth payments. In Zimbabwe, where the bride-wealth for the wife is paid over many years, during the difficult times poorer families chose to receive a significantly higher amount of bride-wealth upfront at the time of daughter’s marriage. 68 Second, if real wages or labor opportunities for women decrease to a larger extent than those for men, labor specialization in home and market production may become increasingly important. Rukumnuaykit (2003) finds an increased entry into marriage by females in Indonesia where the economic crisis led to drastic decline in female wages relative to males. Nobles and Buttenheim (2006) further pinpoint that an increased entry into marriage by women occurred primarily in the communities that were especially hard hit by the economic crisis in Indonesia. Entry into marriage may be also delayed during the crisis. The circumstances leading to delayed marriages are similar to those described above. For example, fewer economic opportunities and tight labor market may make marriages costlier, as families may not be able to afford costs of ceremonies and dowries. Caldwell, Reddy and Caldwell (1986) estimate that in South India during the drought of 1980-1983 the number of marriages declined by approximately 15%. They report that the proportion of marriages deferred was positively related to the amount of land owned by households. While wealthier households postponed the marriages to defer the payment of large dowries and wedding expenses, families of the lower castes accelerated marriages of their daughters to reduce number of mouths to feed. Palloni, Hill and Aguirre (1996) also observed postponement of marriages in Latin America, during and immediately after the economic crisis. However, after the recovery, the marriage rates started to rise. 69 2.1.2 Physical security concerns and marriage Societal conflict and instability may increase real or perceived threats of assault or ill- treatment of young women. The threat could be associated with a potential rape, abduction, harassment or other form of dishonor. In the traditional societies virgin brides are highly valued by potential marriage partners and family’s honor is very important. Thus, at the time of political and social uncertainty during an extended civil strife, parents may expedite marriages of their adult and adolescent daughters, in an attempt to shift their responsibility for maintaining family honor to sons-in-law and their families. Further, families in the conflict affected areas may attempt to find grooms for their daughters far away from the conflict zone (as daughters usually move-in with in-laws in traditional societies), thus possibly trading off their preference to live in the proximity for a better survival chances in the far-away land. Families may achieve two goals by following such a strategy. They improve the safety for their children and mitigate consumption risks by acquiring extended family members who may reside in the safer and less destroyed regions. 2.1.3 Marriage squeeze The marriage market for women is defined by the quality and quantity of men eligible to serve as partners and spouses. The deficit of desirable marriage partners for men or women in the relevant age and regional groups may motivate changes in the societal family practices and marriage institutions. Such changes may include an increase in the number of female- headed and polygamous households and informal unions, a decrease in the stability of unions (Becker 1973; Greene and Rao 1994; Olimova and Bosc 2003; Harris 2006), and a reallocation of relative gains from marriage to brides and grooms. The last may be reflected in changes in size, type and designation of the primary beneficiaries of marriage gifts and transfers of wealth between families of grooms and brides (Rao 1993). Allocation of income, labor force participation by women and distribution of income in divorce 70 settlements may be also affected by changes in sex ratios (Angrist 2002; Chiappori, Fortin and Lacroix 2002). It is well known that armed conflicts take a heavy toll on the population of men (Newth 1964; Das Gupta and Shuzhuo 1999; Roberts et al. 2004; Burnham et al. 2006). Newth estimates that, while the Soviet Union lost one out of five citizens during the World War II (WWII), about 30 percent of the Soviet men aged 15-59 perished during the WWII as compared to six percent of women aged 15-55. Roberts et al. and Burnham et al. compare death toll in Iraq in the pre- and post-2003 invasion periods. Burnham et al. find that the male-to-female ratio of post-invasion deaths significantly increased as compared to the pre- invasion years. The ratio was 3.4 for all deaths and 9.8 for violent deaths. Most deaths due to violence occurred to men aged 15-44. Thus, during or following a war, young and able-bodied men disproportionately “disappear” from country’s population. Men can be recruited in the military forces by the official government or opposition groups. They can also be kidnapped and forced to join insurgents (Blattman 2006). Even after peace is reached, young men who were associated with insurgent or losing groups may be placed in prison or mass executed as they may be perceived as a future security threat by the winning forces. 48 Since the death toll among young men exceeds that of women, such deficit of men may prevent women of marriageable age from finding suitable marriage partners. Thus, a decrease in the population of men in the marriageable age group may affect the marriage market equilibrium conditions. These 48 During the Tajik armed conflict, the Pamiri people who populate the GBAO, some parts of Khatlon, Dushanbe and RRS were associated with the opposition forces. During the conflict of 1992-1998, young Pamiri men were often targeted by the government forces and government affiliated military groups such as Narodnii Front as potentially dangerous militants and were treated accordingly. Multiple cases of mass execution and forced “disappearance” of men in and around Dushanbe were reported during the war. Several mass graves containing bodies with gunshot wounds were found near Dushanbe soon after the war ended (U.S. Department of State 1996, 1997; Human Rights Watch 1993). 71 adjustments in the marriage market are largely determined by country-specific institutional structure and cultural norms. The mechanics of sex ratios may affect rate of entry into marriage, relative characteristics of brides and grooms, size of dowries (bride-prices), quality and type of acceptable marriage unions. Following a large scale armed conflict, an age of entry into marriage may be affected in either direction. Many women may not be able to get married as early as prior to the war as fewer men are available and women (or their families) have to spend more time searching for a suitable candidate. The increased time spent searching will lead to an increase in the average age at marriage for the cohort of women exposed to war. At the same time, if there are more available women for each available man, this may lead to a decrease in the average age at marriage as younger and younger women enter the marriage market with an intention to capitalize on the value of their youth. Such increase in the supply of younger brides may crowd out marriage opportunities for slightly older women who reached the peak of their marriage age and leave a higher proportion of such women unmarried. Second, differences in age and education level between husbands and wives may change in the post-war period as compared to the pre-war years. Women may have to marry men who are older or younger than them or with education level or status lower than customary. For example, a study of trends in age at marriage in the postwar Ireland by Walsh (1972), finds a greater equality in the ages of brides and grooms, where the median age at marriage increased for women and decreased for men. The decrease in the age difference between brides and grooms may suggest an existence of the marriage squeeze on women (Schoen 1983; Greene and Rao 1995; Das Gupta and Shuzhuo 1999). 72 Third, a change in the sex ratio may influence customary marriage gift-giving practices and affect sizes of dowries or bride-prices. Since during wars, men are more likely to die in fighting than women, an equilibrium bride’s price (dowry) may fall (increase) because men in the relevant age group have become more scarce and therefore, more valuable. A relative scarcity of grooms in South India, allowed grooms to demand higher dowry payments and increased the average age at marriage for brides (Caldwell et al. 1983; Rao 1993). Fourth, a decrease in sex ratios may lead to a change in the acceptable marriage practices that are determined by the cultural settings. In societies where polygamy is or was common in the past, such as Sub-Saharan Africa and Central Asia, a marriage squeeze may lead to an increase in the number of polygamous marriages. If such polygamous unions also help to accommodate young childless widows in war-torn areas, polygamy may be tolerated by society as a necessary evil (Falkingham and ADB 2003). In societies where informal unions are fairly common such as Latin America, we may observe an increase in the number of consensual unions as compared to formal marriages (Greene and Rao). Thus, changes in sex ratios may affect many factors that contribute to the quality of marriage. These factors, as described above, can be adjusted either ex-ante or ex-post. The ex-post changes in the sex ratios may affect already existing marriages, while ex-ante arrangements may change the pre-marriage expectations, spouse search strategies and negotiations. However, the overall effect of sex ratios may be ambiguous if the relevant marriage pool is traditionally very narrow and marriage outside of the caste, social strata, region or age group is deemed inappropriate. 73 Summary The theories reviewed above suggest that there are multiple ways in which the decrease in the number of available marriage partners combined with exposure to violent conflict and economic and social instability associated with it may affect the marriage market for women. The combined effects of poverty and low prospects of finding marriage partners due to a marriage squeeze may force young unmarried women to settle for less. Such women may enter polygamous marriages as second and third wives, enter prostitution or accept a status of concubines or temporary wives for hire. Engagement in transactional sex places women at an increased risk for HIV, and is associated with gender-based violence, socio- economic disadvantage and use of illicit drugs (Dunckle et al. 2004). A marriage squeeze and poor economic prospects may reduce the well-being of women in formal marriages as well. Some women become more susceptible and even tolerant to domestic violence and mistreatment by their husbands and in-laws as women may be concerned that their husband may divorce them, or take a second wife. Further, women may be pressured to have more children than they desire or restricted in their use of the contraceptives and family planning methods. 74 2.2 Fertility outcomes and violent conflict The discussion below outlines several theories that elaborate on the impact of civil wars on fertility outcomes. First, according to the development economics literature, there is the old-age security motive for having children (Nugent 1985). The motivation behind this argument is the absence of society-provided retirement and disability systems. In such societies, children serve as insurance against old age and disability when household income is unstable and cannot be insured. During civil wars, physical assets of a household may not be considered a good investment as such assets may be looted or destroyed. In such situations, the real or perceived value of having many children may increase as well as the birth rate. Parents may view their children as assets with a more certain return than livestock or equipment who can provide them with certain income once parents grow old. Second, during wars and other calamities, household incomes are limited. To maintain their consumption, families may postpone having children until better times come. Thus, birth rates during wars and armed conflicts may decrease. However, in agricultural societies the demand for labor may increase as people may live of production at their plots. Third, a prolonged absence of men as they may be at war or dead may explain the decrease in birth rates. When large numbers of men return from war, there may be a baby boom as happened in many countries of the world after the end of World War II. 75 Fourth, medical facilities are often looted and destroyed in conflict. In such regions, home child deliveries or deliveries without attendance of qualified medical personnel are likely to increase. If sanitation conditions at home are poor and complications arise, infant and maternal mortality are likely to increase. For example, in Tajikistan, the already high maternal and under-five child mortality significantly increased during the 1991-1998 conflict. According to UNICEF (2005) maternal mortality almost doubled in the first three years of war in the 1992-1994 period and under-five child mortality increased by 11 to 26 percent in the 1992-1993 period as compared to the-war 1989 data. The maternal mortality rate increased from 53.2 deaths per 100,000 in 1991 to 74.0 in 1993 and 1994. The mortality rate among children under five increased from 57.7 deaths per 1,000 to 72.4 and 82.5 in 1992 and 1993 respectively. However, it is unclear how such mortality changes would affect fertility. It is possible that some women would delay childbirth until better times come, while others may attempt to have more children since some of the children may die very young. Fifth, wars or extended periods of hardship may affect societal practices and lead to more permanent changes in birth rates. For example, Lindstrom and Berhanu (1999) find that in Ethiopia during the years of famine and political instability, marital fertility declined and that this decrease was sustained after the war ended. Sixth, the decline in fertility during economic and social upheavals may be attributed to natural factors affecting reproduction, such as age of entry into marriage, separation of spouses during war, involuntary or voluntary abstinence from intercourse; use of contraception methods, either medical or due to lactation; and fecundity (Davis and Blake 1956). 76 In this paper, I primarily focus on the reproductive behavior as determined by the exposure to conception (e.g. female age at marriage) and economic and social shocks as measured by exposure to armed conflict. The following sub-section reviews marriage, family traditions and practices in Tajikistan. 2.3 Marriage and family in Tajikistan Marriage and family In Tajikistan, as in many traditional societies, men and elderly stand at the highest steps of the social and family ladders. In traditional families, young men and women are raised and taught to respect and obey the elderly in an implicit exchange of the promise of equal power once they reach an older age themselves (Harris 2006). Traditional Tajik families, and there is a rather high proportion of such families in both, urban and rural areas as urban Tajiks maintain close connections with their villages of origin, are patrilocal in nature (Harris 2006: p. 18). Young married women usually move in with their husband’s families where they occupy the lowest step of the social ladder. Kelins (daughters-in-law) are expected to assume a large amount of work, show obedience and respect to their new family, and in particular, their mother-in-law (Tett 1994; Falkingham 2000: 19; Harris 2006, 2004). New kelins usually do not appear in public for several months after marriage, however, such restrictions do not apply to older or younger women who can work, shop or engage in other activities outside their homes (Tett). Traditionally, in Tajik families, a mother of a young man who has reached the marriageable age and who is ready to get married financially, sets out to search for a kelin to suit her needs and preferences. The Tajik families for a long time preferred to marry 77 children to their cousins. 49,50 It is also common to marry someone from the same large extended family - avlod - especially if the avlod has a high social standing in the society (Bushkov and Mikulskii 1997). 51 However, a marriage proposal may be extended to a bride from a less important avlod if a groom is not attractive for marriage, for example, suffers from chronic health problems. Further, if the pool of eligible cousins and other distant relatives is exhausted and or if a family does not want to have a son married to someone in the extended family, the search for a suitable marriage partner is conducted among family friends and acquaintances. Once a girl is selected by the groom’s family, the girl’s parents are approached with a marriage proposal. While many brides and grooms have only a nominal choice in accepting or refusing the proposed match, some of them manage to manipulate or beg their parents into letting them get married to someone of their liking (Tett 1994; Harris 2004, 2006). There is a duality in views on the traditional arranged marriages in the Tajik society: while the majority of women support the continuation of this tradition for their children, the same women also indicate that they were able to exercise significant discretion in the choice of their husbands. 52 At least 58.3% of marriages are determined by parents (Bushkov and Mikulskii 1997). Once the marriage proposal is accepted, families of the bride and groom tear a large piece of white cloth into several headscarves and break several Tajik flat breads 49 It is not clear if a close cousin or a distant one is preferred as a marriage partner. However, an example provided by Harris (2004: 105) suggests brothers and sisters persuade each other to marry their children. She also notes that women who are married into their extended families report a large number of birth defects in their children. They blame for those defects many generations of such marriages. 50 Kuz’menko (1991) reports that there are many marriages among close relatives in Tajikistan as such marriages are cheaper. 51 In Tajik society, an avlod is a patriarchal community of blood relatives who have a common ancestor and common interests. Members of avlod can also have shared property, land and means of production. Often in extended families household budgets are consolidated or shared among members of an avlod (Bushkov and Mikulskii (1997); Abdullaev [n.d.], Olimova and Bosc (2003: 56)). 52 In the 2000 WHO survey on violence against women, 69 percent of respondents (a sample of 900) said that they will decide on the marriage of their children together with their husbands. Only one percent said that their children will also have a voice in that decision. However, 68 percent of the married respondents (sample of 840) said that they got married of their own free will and were able to choose their husbands (WHO 2000). 78 into pieces. Then they wrap pieces of bread into the scarves and distribute the bread and the scarves among the relatives of the bride and groom. This procedure symbolizes an acceptance of the union and indicates that the bride is spoken for (Usmanova 2006). At the time of engagement and marriage families also exchange a large number of gifts. The gifts commonly include sets of expensive dress material for traditional Tajik dresses, household goods, linens and bedding, sacks of rice and food for the marriage related feasts. 53 Traditionally, the groom’s family transfers a larger amount of wealth, or kalym (bride price), to the bride’s family. 54 Both, the amount and quality of kalym are carefully scrutinized. An insufficient amount of kalym or gifts of poor quality may be used as an excuse to break the engagement. The value of kalym is reportedly higher for girls without tertiary education, especially for those who left school after grade eight or earlier (Harris). This price differential is attributed to the ease with which control could be exercised over an uneducated woman by her new family. The marriage ceremony is usually conducted by a civil servant at the local civil office that registers births, deaths and marriages (ZAGS). Tajik family law defines the age of consent as 17 years old for both men and women. The court can reduce the age at marriage by maximum one year at the written request of persons entering marriage. 55 Tajik women tend to marry at a younger age, succumbing to pressure from their families and the society which looks down upon families who have unmarried but eligible daughters. Most men wait until they can accumulate enough money for the marriage 53 During the war and the post-war period community mullahs (Gomart 2003) and local governments often put restrictions on the size of the marriage feasts and number of invited guests thus putting restraints on the lavish weddings that were not appropriate due to general impoverishment of the population. 54 The existing evidence on kalym (bride-price) in Tajikistan is fragmentary and is limited to stories and anecdotes. Kalym, once a part of each marital transaction, was made illegal and punishable by fine and time in prison by the Soviets in 1920s. The authorities were trying to prevent forced marriages of minor girls. However, in the years following the passing of the above mentioned laws, the transgressors were often able to avoid the punishment by referring to kalym as marriage gifts that were not associated with a sale of a girl and especially a minor girl (Kamp 2006). 55 Semein’i’i kodeks: Chapter 3. Article 13. Dushanbe, November 13, 1998. # 683. 79 expenses and new family. 56 These traditions are reflected in the age at which most Tajiks marry, creating on average three to four years age gap between brides and grooms. The tabulations of age at marriage of males and females that are based on the 1989 Census data, indicate that 84.1 percent of the officially registered marriages (a total of 47,616) were officiated between women aged 18-24 and men aged 20-29 (Goskomstat 1990: 81). Single year tabulations based on the 2003 TLSS data for women ages 15-49 indicate that the majority of women and men were married when they were 17 to 22 and 18 to 27 years old respectively. Thus the median age at marriage for women was 19 years old and for men 22 years old. In the post-independence, the religious marriage ceremony – nikoh, has grown in popularity. The nikoh ceremony is not recognized by law and thus it does not provide spouses with legal and property rights in a case of divorce. 57 However, it may be chosen for religious, economic and social reasons. Nikoh often costs less than an official ZAGS registration and allows “officiating” polygamous marriages and marriage of minors. 58 Both types of marriages are not allowed by the Tajik family law. 59 Such marriages, especially polygamy, are more prevalent in the rural areas of Tajikistan. The 2000 WHO pilot survey on violence against women in Tajikistan reports that 21% of married women in a sample of 840 had polygamous husbands. Informal and polygamous unions are becoming more common in Tajikistan as women face a shortage of men due to civil war and labor migration of men to the more affluent countries in the Former Soviet Union region (Olimova and Bosc 56 "If you want to marry, you'll need so much money that even the sale of a house won't be enough. Parents often go into debt to finance a son' wedding. So there's no chance of getting married. Besides the wedding itself, one has to think about other aspects of family life: how to feed your family and provide clothes and shoes. It is not right to be a burden to your parents. You have to earn money first and then think about marriage." Construction worker, 21, Isfara. (from Olimova and Bosc (2003: p. 26)). 57 Many NGOs in Tajikistan have been trying to address this issue and lobby for the protection of the legal rights of the second wives and their children. 58 Although quantitative data on the polygamous marriages and marriages of minors is very scarce, such cases are frequently encountered in anecdotes, newspaper articles and reports produced by various NGOs in Tajikistan. 59 Semein’i’i kodeks [Family Law]: Chapter 3, Articles 13 and 14. 80 2003). Among women aged 15-49 surveyed for the TLSS 2003 data less than one percent reported that they were in polygamous marriages. The difference across surveys in the proportion of women in polygamous marriages may be attributed to differences in the methodologies and the sampling frames of those surveys. 60 Fertility Soon after marriage young women are pressured to produce a child, preferably a son. Having a child improves a woman’s status in her new family and signifies the procreation capacity of her husband. Large families in Tajikistan are very common and respected. The total fertility rate among Tajik women has been traditionally high. High fertility rates are often explained by high infant mortality rates after the World War II and the general preference to have large families (Falkingham). According to the 1989 census, ethnic Tajiks had the highest total fertility rate (TFR) of 5.9 lifetime births in the newly independent Central Asia. The second highest total fertility rate of 4.1 lifetime births was recorded among ethnic Turkmen in Turkmenistan (Turner 1993; UN 2003). Uzbekistan is the close third in TFR. The TFR in Tajikistan peaked in the 1970-1980 period, reaching 6.83 births per woman (DESA 2003). The rate averaged 6.41 births per woman between 1950 and 1975, falling slightly below 6.00 in 1975-1980. The TFR continued to decline in the 1980s and by 2000 it was projected to reach 3.72 births per woman. Several factors may have contributed to the decline in the TFR in the 1980s. First, it may have been a response to a steady 60 The 2003 TLSS used a two-stage random sampling process to draw the sample surveyed in each region (see Section 3 for more details). The questions about current marital status of the household members are answered by household’s head or a person who is most knowledgeable about the household affairs. The 2000 WHO’s pilot survey data is based on the detailed interviews of individual women. The WHO’s 2000 survey used two methods of selecting observations. The first method was to randomly select households (and one woman within a household) in various neighborhoods of towns and villages that were considered representative of the geographical areas selected for sampling. The second method was to interview women in academic and medical institutions, factories and farms. 81 decrease in the infant mortality during the same period, which declined from 110 in the 1950-1955 period to 57 deaths per 1,000 live births in the 1995-2000 period (DESA 2003). Second, the continuing decline in TFR in the 1990-2000 may be also related to the separation of spouses due to labor migration and war. Third, the decline may be attributed to the economic hardship after the country’s independence. The relative cost of having children increased since 1980s as the state support in the form of kindergartens (child-care) and subsidies for families with many children ceased to exist. In Tajikistan, relatively few ethnic Tajiks use contraception as compared to the rest of the post-Soviet Central Asia (Turner 1993). In 1985 only 14-18 percent of married Tajik women controlled their fertility as compared to 39-40 percent among ethnic Kazakh and 27- 32% among ethnic Uzbek women (Turner). Based on the 1991 estimates reported by Turner, only 3% of ethnic Tajiks used a modern contraceptive method as compared to 26% in Kazakhstan and 6% in Uzbekistan. However, with time, contraception use in Tajikistan has been increasing. In 2000 some form of contraception was used by 33.9% of women who were married or in a relationship (UNICEF 2000). The most common contraception method is an internal uterine device (IUD), used by 25 percent of married women ages 15 to 49 in Tajikistan. 61 Contraceptive use is the highest in GBAO at 63 percent and Sugd at 51 percent. Less than one quarter of the married women in Khatlon and RRS use contraception. The use of contraception increases with age. Among the currently married and in-union women who were surveyed for the 2000 Multiple Indicator Cluster Survey (MICS 2000) in Tajikistan, nine percent of women aged 15-19 used contraception as compared to 18 percent of women aged 20-24 and 40 percent among older women (UNICEF 2001). The higher use of contraception among older women indicates that they may have completed their families, as 61 Turner (1993) reports that the pill is difficult to procure. It is also viewed with suspicion by women and their doctors. 82 compared to younger women. Newly married young women are expected to become pregnant in the first year of their marriage. A failure to do so may result in a bride being returned to her parents (Harris 2004: 109). A desire to have children (67.86%) and resistance by husband or partner towards contraception (15.84%) are two main reasons for not using birth control by married women (TLSS 2003). 62 3. DATA In the analysis of transition data this study employs the household and individual data from the 2003 Tajik Living Standards Survey (TLSS 2003), conducted by the World Bank and the State Statistical Committee of Tajikistan. The survey includes 4,160 households with a total of 26,141 individuals in three oblasts (regions) and one autonomous region – the GBAO. A two-stage random sampling process was used to draw the sample surveyed in each oblast. The sample was stratified according to oblast and urban/rural settlements. The share of each stratum in the overall sample was proportional to its share in the 2000 Census. The 2003 TLSS was over-sampled by 40 percent in Dushanbe, by 300 percent in rural GBAO and by 600 percent in urban GBAO (World Bank 2005). 208 primary sampling units were surveyed in 2003 as compared to 108 units surveyed in the 1999 TLSS. The survey questionnaire provides extensive information on each household and individual. This study uses data on marriage and reproductive history from the female questionnaire in the 2003 TLSS. The questionnaire was filled by women aged 15 to 49 as of 2003. 63 The data on education, age and other socio-demographic characteristics of all 62 2003 TLSS data, sample of 2,285 married women ages 15-45. 63 In a case when a woman could not read or write, the survey was filled out for her by the interviewer. I found some mistakes in coding of dates of children’ birth and the identification number of mothers and their children. I investigated and corrected some of those mistakes. I reconciled the data from the female questionnaire with the data from the household roster. Specifically, I compared individual identification numbers of the household members, their age, sex and relationship to the household’s head. 83 married and unmarried women are obtained from the main 2003 TLSS questionnaire. The main questionnaire contains data on all household members who are eating and living under the same roof and who were not absent for more than 12 months from the household. The data on the characteristics of spouses and partners of the currently married women are obtained from the same main TLSS questionnaire. Unfortunately, the 2003 TLSS does not provide complete marital histories for both men and women and therefore precludes us from measuring transitions in and out of marriages and widowhood. The female questionnaire contains only information on female age at first marriage. Using a woman’s age at first marriage I calculate age at marriage for her (current) husband. In this calculation I assume that a woman did not divorce or was not widowed, and remarried. In comparison to marriage data, the questionnaire asks rather detailed questions about the birth history and reproductive health for all women ages 15-49 who had their first menses. Each birth history includes the full date of birth (day, month and year of birth) for all children born to a woman. The data also include information on whether a child is currently alive and lives in the household. In the case of child’s death, the surveyors recorded information on how long the child lived. The length is recorded in days for children who died before being one month old, in months for children under two years old, and in years for older children. While the survey includes an indirect question on a possible miscarriage: “Have you ever been pregnant, even if you had a pregnancy that lasted only a few weeks?” the survey does not contain any information on the date of miscarriage or unsuccessful pregnancy. Thus when I refer to the date of first birth, or age at first birth, this date is based on the year of birth of a woman’s first child, or the date of birth that was a result of a woman’s first successful pregnancy. 84 I also calculate sex ratios as a measure of marriage squeeze. In order to estimate the sex ratios, I use the 1989 and 2000 Tajik Census data on the distribution of population by raion of residence and age group. I describe the construction of sex ratios in detail in Section 4.2. Variables measuring exposure to conflict are described in Appendix A. In this paper I focus my analysis on an individual’s residence in the conflict affected region (RCA) during the period of exposure, be it pre-marriage or pre-first birth years. The RCA variable is a dummy variable, where “1” indicates exposure to conflict in the raion of residence, and “0” stands for the residence in the lesser affected region. I discuss the questions related to the assignment of residence in the section 3.2 below. In the estimation of factors contributing to entry into marriage and first birth that follows in Section 5 I link individual and household data, sex ratios and conflict exposure by region to review and compare trends in marriage and reproductive behavior of women born in 1966-1986. 3.1 Descriptive statistics The survey contains information on 6,182 women aged 15-49 (Table 19). The mean age of women in the sample is 28.56. On average, women completed 10.06 years of education. The mean ages at first marriage and first birth are 19.56 and 21.54 years respectively. On average there are 3.49 children born to women aged 15-49. I expect this number to be higher if the sample was restricted to women who have completed their reproductive cycles. 85 59% of women aged 15-49 report to be in a formal marriage at the time of the survey. 33% have never been married, four percent are widowed and two percent are divorced. While several studies (World Health Organization 2000; Olimova and Bosc 2003; Harris 2004, 2006) suggest that since 1991 there was a significant increase in the number of informal and polygamous marriages in Tajikistan, less than one percent of women surveyed in TLSS 2003 report to be in such relationships. Thus, my analysis of entry in marriage is focused on the transition from the never married into married state. The sample statistics for the analytical sample of husbands and wives are reported in Table 20. The average age gap between spouses is 3.69 years, with a standard deviation of 4.10 years. The average gap in years of education completed is 1.26 years, with a standard deviation of 2.78 years. 86 Table 19 - Descriptive statistics for the sample of women: ages 15-49. Variable N Mean Std. Dev. Min Max Personal Characteristics Age 6182 28.56 9.62 15 49 Year of birth 6182 1974.4 9.6 1954 1988 Educational Characteristics Indicator for enrollment in educational institution 6168 0.13 0.33 0 1 Indicator for "ever attended school" 6167 0.98 0.12 0 1 Years of education completed (total) 6167 10.06 2.32 0 21 Reproductive history characteristics Age at first period 5876 15.34 1.17 12 27 Age at first marriage 4084 19.56 2.32 9 42 N of children born (total) 3853 3.49 2.02 0 12 N of children died 3739 0.24 0.67 0 7 Age at first birth 3730 21.54 3.31 12 44 Age at second birth 3239 23.75 3.43 12 40 Indicator for Marital Status Married 6182 0.59 0.49 0 1 In a polygamous union 6182 0.00 0.07 0 1 Divorced 6182 0.02 0.15 0 1 In informal union 6182 0.00 0.04 0 1 Widowed 6182 0.04 0.21 0 1 Single 6182 0.33 0.47 0 1 Indicator for spouse present in the same household 3711 0.96 0.20 0 1 N months spouse has been absent from the household 155 0.55 2.87 0 24 Community characteristics Rural 6182 0.70 0.46 0 1 Reports of conflict activity (RCA) 6167 0.69 0.46 0 1 Notes: (1) 6 observations with age at first birth' values below 12 were omitted for ages 15-49 (of age 17, 19, 20, 24, 25 and 32). (2) 5 observations with age at second birth' values below age 12 were omitted for ages 15-49 (of ages 17, 19, 24, 25 and 30). Source: TLSS (2003). 87 Table 20 - Descriptive statistics: sample of married women. Variable Obs Mean Std. Dev. Min Max Wife's age 4796 37.84 13.04 15 90 Husband's age 4410 42.36 14.29 15 109 Wife's year of birth 4796 1965.2 13.0 1913 1988 Husband's year birth 4410 1960.6 14.3 1894 1988 Year of marriage 3640 1989.7 8.1 1970 2003 Wife's age at first marriage (years) 3645 19.53 2.22 9 42 Husband's age at marriage (years) 3328 23.23 4.40 10 75 Difference in age (years) 4410 4.03 4.38 -27 57 Years of education completed (wife) 4796 9.77 2.69 0 21 Years of education completed (husband) 4410 11.14 3.13 0 27 Difference in years of education completed 4410 1.39 3.03 -11 23 Indicator for residence in Conflict area (RCA) 4796 0.67 0.47 0 1 Rural area 4796 0.71 0.45 0 1 GBAO 5352 0.10 0.29 0 1 Sugd 5352 0.28 0.45 0 1 Khatlon 5352 0.26 0.44 0 1 Dushanbe 5352 0.08 0.27 0 1 RRS 5352 0.18 0.39 0 1 Notes: All differences are calculated as: “husband”- “wife”. Source: TLSS (2003). 3.2 Duration data In the analysis of duration data it is common to specify an onset of exposure, or time when individuals become exposed to the risk of failure. In this paper I consider transitions between following states: single/married; (married – no child)/ (married- given birth to first child); no child / have first child (see Table 21). In the analysis of marriage data, I follow Rukumnuaykit (2003), and specify the time of entry or time of exposure to being married at age 11. For each year between age 11 and the reported age at first marriage, women contribute one observation to the analysis. Ever- married women exit the analysis when they enter their first marriages. For women who were never married at the time of the 2003 survey, the data is censored at the time of their 88 interview. I assume such censoring to be exogenous (a conventional practice in the analysis of the duration data that assumes that censoring is not based on the outcome variable). Table 21 - Outline of the analysis of the transition data. Type of analysis Transition Age of exposure Place of exposure Censoring Age at first marriage single ⇒ married 11 residence at age 12 Age at first birth a) conditional married, no child ⇒ married, have a child age at first marriage b) unconditional no child ⇒ have a child age 14 residence at the time of the survey to proxy for post-marital residence time of the interview In the conditional fertility analysis, females enter the duration analysis at the age of their first marriage. I specify the failure time to be the age when the first child was born. The data does not contain information of pregnancies that ended prematurely, such as miscarriages and abortions. Women in the sample contribute one observation for each year that passed between their marriage and a pregnancy resulting in a birth. For example, a woman who got married at age 20 and gave birth to her first child at age 23 contributes to the duration analysis three years that passed between her marriage and the birth of her first child. Any conflict exposure after this reproductive event is discarded from the analysis. In the unconditional analysis of first births, I use age 14 as age of entry into analysis. Thus, a woman contributes a number of years passed between age 14 and her first birth to the analysis. Since we do not observe pregnancies or non-pregnancies of women who reported to be never pregnant in 2003, the observations on such women are censored. The data are right-censored at the time of the interview. For example, if a woman was married at the age 18 in the year 1995 and never been pregnant between 1995 and 2003, this woman contributes 8 years of exposure to the “risk” of being pregnant to the conditional analysis. 89 In the marriage analysis, the residence of a woman when she was 12 years old is used as a location of exposure to the armed conflict. In the fertility analysis, woman’s residence prior to when she had her first child is used as a location of her exposure to the conflict. Thus, the same woman who moved between age 12 and age when she had her first child is assigned to two different exposure groups at the place of her residences. For example, if a woman lived in Dushanbe between her birth and age 12, Dushanbe is used as a place of exposure in the analysis of entry into marriage. If at age 13, this woman moved to Khorog city in the GBAO region, Khorog city is used as a place of exposure in the fertility analysis. In addition to data on marriage and birth history, I combine the data from the individual, household, and community questionnaires. I control for the following socio- economic characteristics: education, raion and oblast and type (rural/urban) of residence at the time of exposure to marriage and fertility. The education level is potentially endogenous to marriage as the marriage hazard increases significantly upon completion of education, especially for younger people (Winship 1986; Brien and Lillard 1994). To tackle this potential endogeneity of education to marriage, I use completion of mandatory nine grades of schooling as an indicator for the education level. Assuming an individual does not repeat grades and enters school at age 7, nine grades of education should be completed the latest by age 16 or prior to the official marriage age of 17 years. The TLSS 2003 data have information on the district of origin only for those individuals who moved between 1990 and 2003. Such information is not available for those who moved before 1990. Thus, the raion (district) of residence at age 11 (for marriage hazards) or 14 (for unconditional fertility hazards) is established from the individual’s migration data for those who moved between 1990 and 2003. 90 The residence of a woman before she had her first child is assumed to be related to the factors that are correlated with fertility decisions, such as prices and availability of services, including those affecting fertility choices, availability of health care and local community practices (Rukumnuaykit 2003). A woman’s most recent place of residence is used in the analysis of first births. 4. ESTIMATION 4.1 Identification To identify an individual’s exposure to the war, I explore two sources of variation in the exposure to the armed conflict of 1992-1998. The first source of variation comes from the regional differences in the extent and intensity of war-related events, such as the destruction of infrastructure and industries, the degree of fighting and displacement during the conflict. Although the Tajik society was affected by violence and instability for a relatively long period of time, 1992-1998, the intensity of hostilities varied over the time. The first two years of the war, 1992-1993, were the most brutal and severe. 64 They were characterized by large displacements of population and high death toll due to the fighting. The intensity of fighting became much lower and more localized after the post-Communist Tajik government managed to take back power in 1993. Tajikistan is administratively divided into four regions such as Gorno-Badakshon Autonomous Oblast (GBAO), Khatlon, Raions of Republican Subordination (RRS) including Dushanbe, and Sugd. Hostilities, especially those of 1992-1993, did not affect all parts of the country equally. Some regions, in particular, the northern and eastern provinces 64 Some sources date the end of hostilities to as late as 2000 (Rashid 2003). 91 of the Sugd and Gorno-Badakhshon regions, remained relatively unaffected by the major fighting. The Sugd region was especially isolated from the conflict during the first years of the war as the region is separated from the rest of Tajikistan by a mountain range. The only railroad that connects the northern and southern parts of Tajikistan was deliberately destroyed during the war to reduce access to Sugd by militias (Rashid 2002). While relatively stable during the years of the war, the Sugd region experienced a shock in November 1998. At that time Colonel Makhmud Khudoberdiyev attacked Khodjent and the surrounding areas from the border of Uzbekistan. Sugd region was relatively unaffected during the most of the war, in 1992-1997. However, both regions, especially GBAO, were indirectly affected by war, through the general impoverishment of the population, their isolation from the rest of the country and severe deficits of foodstuffs in the first years of the conflict (Gomart 2003). 65 Parts of the GBAO experienced frequent attacks of insurgents who tried multiple times to cross the border (those areas of GBAO were coded as war affected regions). In the southern regions of Tajikistan, especially eastern Khatlon, RRS and Dushanbe, the impacts of war were particularly strong. The second source of variation is determined by the timing of the exposure to the civil war. Thus, women who attained the prime marriage age (as defined in Section 4.2) before the conflict may be affected differently from the cohort of women who reached the prime marriage age during the conflict. The marriage prospects of the younger women may have been compromised by their exposure to the armed conflict and by the deficit of men in the relevant marriage age group due to war, labor migration and costs considerations. The 65 GBAO was subjected to an economic blockade during the first few years of the 1992-1998 Tajik armed conflict, as the only land road that connects GBAO to other parts of Tajikistan was controlled by the opposition groups. The region experienced severe food shortages at that time. The provision of foodstuffs from Dushanbe and other regions to GBAO was suspended for presumably political reasons as Pamiri people who populate GBAO were linked to the opposition forces. 92 reproductive decisions of those young women may also have been affected by the war as first births often closely follow first marriages. Table 22 presents data for 12 three-year birth cohorts of women born between 1954 and 1988. The proportion of married women in each three year birth cohort is negatively related to woman' year of birth. While 79.67% of women born in 1969-1971 were married by 1992, less than 26% of women born in 1972-1974 were married by 1992. Further, less than one percent of women born in 1975-1977 and 1978-1980 were married as of 1992. Thus their marriage decisions may have been influenced by women's exposure to the conflict during their prime marriage years. Table 22 - Three year birth cohorts: selected demographic data. Birth cohort Age in 1992 Age in 1998 Age in 2003 % of the cohort married by 1991 (inclusive of 1991) % of the cohort who had first child by 1991 N 1954-1956 36-38 42-44 47-49 97.20 94.76 191 1957-1959 33-35 39-41 44-46 96.78 92.69 342 1960-1962 30-32 36-38 41-43 97.12 93.73 415 1963-1965 27-29 33-35 38-40 93.99 85.08 449 1966-1968 24-26 30-32 35-37 92.52 80.18 439 1969-1971 21-23 27-29 32-34 79.67 52.81 481 1972-1974 18-20 24-26 29-31 25.78 8.61 511 1975-1977 15-17 21-23 26-28 0.18 1.10 545 1978-1980 12-14 18-20 23-25 0.00 0.15 656 1981-1983 9-11 15-17 20-22 0.00 0.00 774 Total: 9-38 15-44 20-49 4,803 Note: This table allows us to identify birth cohorts whose decisions about the first marriage and first birth may have been influenced by the exposure to the Tajik armed conflict of 1992-1998 during their prime marriageable years (ages 17-22). 4.2 Sex ratio calculation Prime marriage age group and measures of marriage squeeze A ratio of males to females in the prime marriage age groups was used by Akers (1967), Keeley (1979) and Greene and Rao (1995) to measure a disproportion between sexes. I follow this practice in my study and use this ratio to control for the availability of men as 93 compared to women of marriageable age. I use the ratio of men aged 20-25 to women aged 15-25 in 1989 and 2000. These age groups are chosen on the basis of ages of brides and grooms in 1989 reported by Goskomstat (1990) and the reports of age at first marriage by females in the 2003 TLSS conducted by the World Bank in cooperation with local agencies. To calculate raion level sex ratios I use the 1989 and 2000 State Statistical Committee of Tajikistan population data by raion, sex and age group. The population data are reported by five-year age categories for the population ages 10 to 84, for example, ages 10 to 14, 15 to 19 and so forth. Thus, although the prime marriage age groups in Tajikistan appear to be women ages 17-22 and men ages 18-27, the above mentioned five-year age intervals in Census data restrict me to using either five- or ten-year age groups for the construction of the sex ratios. 66 I chose to use sex ratios based on the ten-year age groups such as men ages 20-29 and women ages 15-24 in 1989 and 2003 as those groups allow a better coverage of the potential marriage age partners, as compared to using five-year age groups. I assign the 1989 and 2003 sex ratios to the corresponding birth cohort as defined in Table 23. Each birth cohorts is assigned a raion level sex ratio. For each point in time a woman was at risk of being married, defined as age 11, she was assigned the closest ratio available for her age group. 66 88.92 percent of women ages 15-49 (a sample of 4,084 of married women) reported that they were married between age 17 and 22. Only 2.17 percent indicated to be married prior to reaching age 17 and 8.22% were married between age 23 and 29. The male age at marriage was calculated from the age at first marriage reported by ever married women. In the 2002 survey, I was able to link to each other spouses in 3,328 couples. 88.55% of husbands (a sample of 3,328) were married between age 18 and 27. Only 1.53% was married before reaching age 18; 6.46 percent of men were married between age 28 and 31. 94 Table 23 - Sex ratios by birth cohort and region of residence. Birth cohort Age in 1989 Age in 2000 Age in 2003 Sex Ratio=(men age 20-29)/ (women aged 15-24) 1966-1974 15-23 26-33 29-37 1989 data 1975-1986 3-14 14-25 17-28 2000 data Trends in sex ratios The sex ratios (Sex Ratio) are calculated for each raion and major city for 1989 and 2000. The comparison of sex ratios between 1989 and 2003 indicates that the sex ratios decreased between 1989 and 2000 in 51 out of the 66 regional units. In 1989 the ratios vary from 640 men per 1,000 women in Darvaz raion to 1,290 men per 1,000 women in Taboshar city. In 2003, the variation is slightly less, from 610 men per 1,000 women in Khovaling raion to 1,020 in Dushanbe city. The largest decreases are in Rogun (RRS), Taboshar and Chkalovsk (Sugd). To find out whether the declines in the district level sex ratios between 1989 and 2000 are explained by conflict activity in those regions, I estimate two OLS regressions. In these regressions, the raion-level sex ratios for 1989 and 2000 are separate dependent variables and the regional conflict exposure is an independent variable. The regression results (reported in Table 24) do not support the hypothesis of the negative impact of conflict on the sex ratio in the 2000 data. 95 Table 24 - Effect of regional conflict variable on sex ratio. Dependent variable: sex ratio Men ages 20- 29 to women ages 15-24 in 1989 Men ages 20- 29 to women ages 15-24 in 2000 (1) (2) RCA -0.009 0.021 [0.032] [0.021] Constant 0.889*** 0.810*** [0.026] [0.017] Observations 69 69 R-squared 0.00 0.01 Notes: OLS regressions. Standard errors in brackets. Source: population data – State Statistical Committee of Tajikistan (2002), conflict data – author’s calculations. * significant at 10%; ** significant at 5%; *** significant at 1%. The regression results for age at first marriage and first birth that include the sex ratio as an explanatory variable are reported in Section 5. The district level sex ratios based on the population data may not be crucial in determining the marriage market conditions, as many Tajik families prefer to marry their children to someone of their own kin. Thus, broad sex ratios that include all men and women available for marriage in the region of their current or even past residence may not measure the marriage squeeze phenomena adequately. Some regional groups and members of various clans who live in different regions may marry each other due to tradition. For example, some inhabitants of village Galadzor in Vose raion originated from Khovaling – a city to the north. Their ancestors moved from there to Vose several generations ago. The descendants’ ties to ancestral place of residence remained very strong. Many of them still marry people from Khovaling and thus do not have strong extended family ties in Vose, where they currently reside (Gomart 2003: p. 63). 96 Thus several generations may pass and people continue to maintain close ties with their relatives left behind in ancestral regions. 67 Therefore sex ratios that are based on the current place of residence may not adequately measure the appropriate pool of marriageable partners as defined by the ancestral places of residence and clan lineage. Further, the history of forced resettlement during the Soviet times makes tracking such family ties and relations more complex as one cannot assume that a group living in a particular geographical area of Tajikistan has long-standing traditional and recognized rights to the land and ties to the region of residence. Many rural settlements – kolkhozes - in Khatlon consist of smaller ethnic groups. These groups were forcefully resettled to Khatlon area from other regions of Tajikistan, such as Pamir (Gorno-Badakshon), Gharm and Kulob to cultivate the land for cotton (Bushkov and Mikulskii 1997; Gomart 2003). Unfortunately, it is not possible to identify the appropriate marriage pool for each regional group from the quantitative data such as TLSS 2003 as there is no information on where the ancestors of each family came from. 68 4.3 Methodology This paper focuses on two demographic behavioral processes – age at marriage and age at first birth. In this section I discuss the estimation techniques and hypotheses I consider. The basic statistical framework for the analysis of duration data is presented in Appendix D. 67 Bushkov and Mikulskii (1997) describe the pattern and composition of settlements in different areas of Tajikistan. They further sub-divide the regional based groups and ethnicities into smaller units. They argue that the high level of division of Tajikistan by ethnicity and sub-ethnic groups may have been one of the causes of the civil war in Tajikistan in 1992-1997. 68 To my knowledge there are no databases, maps or other significant attempts (apart of brief mention in the qualitative studies) of precise mapping of residences of clans and regional groups that are so important in Tajikistan. The commonly used generalization is that Pamiri clans live in the GBAO district and parts of Gharm, Gharmi (clans from Gharm) in Gharm, Kulobi in Kulob (Khatlon region), Khodjenti in Sugd region. However, such generalizations are incomplete and too broad. See Bushkov and Mikulskii (1997) and Gomart (2003) for more detailed discussion. 97 I employ hazard functions to compare, at same ages, the odds of becoming married for women who were exposed to the war and those who were not. The odds of having a first child are computed in a similar way. Hazard functions are a convenient way to interpret and describe the process that generates statistical failures, such as transition from a single state into being married. The hazard is defined as an unobservable or latent function that controls for the occurrence or non-occurrence of the event, and also for the length of time to the occurrence of an event (Kalbfleisch and Prentice 1980; Allison 1982). The empirical hazard regression techniques allow us to estimate the effect of covariates on such transitions. In the analysis of the duration data the dependent variables of interest are the odds of an event occurring. In this paper, I estimate the odds of the following events: 1) that a woman becomes married in year t given that she is unmarried at the start of the year t; 2) that a married woman has her first child after t years of marriage given that she did not have a child at the start of year t after marriage. I assume that the age of marriage is exogenous when studying the odds of first births. 3) that a woman has her first child in year t given that she did not have a child at the start of the year t. The main model to be estimated empirically is equation (3) below: (3) ijk j jk i j k j ijk C X P t t ε ν μ δ η β α λ λ + + + + + + + = 1 1 0 ) ( ln ) ( ln where λ ijk is a hazard rate. Subscripts on the dependent variable denote individual i residing in the district j at her time of exposure to conflict and born in year k. α 1j is a fixed effect for the individual’s district of residence at the time of exposure. β 1k is a cohort of birth fixed effect. P j is the intensity of the conflict in the raion of residence at the time of exposure to 98 marriage or first conception. i X is a set of individual characteristics (education level). jk C is a set of regional characteristics that are specific to a birth cohort. v j is a district specific random effect. 69 The district specific random effects do not correct for any possible correlation between the error term and the right-hand side variables. The random effects (i.e. frailties in survival analysis) are assumed to follow gamma distribution. The frailty variance θ is estimated from the data and measures variability of the frailty among groups of observations. Hypothesis The hypothesis that is being tested is the significance of the coefficients estimated on conflict and sex ratio variables and the effect of those variables on entry into marriage and first birth once other individual and regional characteristics are accounted for. The conflict exposure may have the following effects on marriage and reproductive behavior. Assuming that residence in the conflict affected area proxies for an economic hardship, the effect can be two-fold. First, if grooms postpone marriages because of high costs, the effects of conflict on entry into marriage for women is expected to be negative. Second, brides’ families may be willing to accept lower bride-prices and have less lavish weddings. If so, the effect of the conflict on entry into marriage may be very small with second factor offsetting the relative increase in marriage costs. Theoretically, the coefficient on the sex ratio variable should have a positive sign suggesting that higher sex ratios increase the woman’s entry into marriage. The effect of sex ratios on the probability of conception in any given time period should be positive as well, suggesting that an increase in the number of potentially available partners increases the probability of conception. 69 Model specification with random effects in the hazard regression framework is discussed in Appendix D. 99 Difference-in-differences estimation I extend model 1 to estimate it in a difference-in-differences framework. The difference-in- difference approach is used to compare the odds of being married or having a first child for women who entered marriageable age during the period of the conflict with of women those who were either already married by 1992 or who reached marriageable age after the end of the conflict in 1998. The model is defined as follows: (4) ijk j jk i j j k j ijk C X cohort war P P t t ε ν μ δ γ η β α λ λ + + + + + + + + = ) ( ) ( ln ) ( ln 1 1 0 where ‘war cohort’ is a dummy variable indicating whether individual i belongs to one of the three-year birth cohorts exposed to the conflict, such as born in 1975-1977, 1978-1980 or 1981-1983 (i.e. born between 1975 and 1983). To estimate parameters in equation (4), I use the sample of women, born between 1966 and 1986 (ages 17 to 37 in 2003). In the difference-in-differences framework specified above, I focus on two groups of individuals, the treatment and control groups as defined below. The individuals who turned 17 years old between 1992 and 2000 and who lived in the conflict-affected areas comprise the treatment group. Based on their year of birth, those women reached an age when they could officially marry during or immediately after the Tajik civil war. Thus those women were significantly exposed to the economic and demographic shocks associated with the conflict. 100 The control group is a group of individuals whose marriage prospects should not have been significantly affected by the conflict. The control group includes two subgroups. The first subgroup contains individuals born between 1975 and 1986 who lived in the regions less affected by conflict. The second subgroup includes individuals who were born between 1966 and 1974. This older cohort was likely to be married before the start of the conflict in 1992 and their marriage prospects should not have been affected by the conflict. Hypothesis The main hypothesis that is being tested is whether the estimated coefficient γ is equal to zero, thus suggesting that the conflict did not have a significant impact on age at marriage and/or age at first birth. Dataset The data set used in all regressions includes data from the 2003 TLSS on women born in 1966-1986. The analysis includes observations on women born in 1966-1986 (i.e. women who were age 17 to 37 in 2003). 70 Thus I compare several 3-year birth cohorts (born in 1975-1983) to the cohorts born in 1966-1974. The estimation excludes from the analysis two sets of observations. The first set is the sample of women who were too young to be married as of 2003, or women who were age 16 and below (only 2 out of 533 women ages 15-16 were married in 2003). The second set is a sample of women born in 1954-1965, or women who would be less comparable to the cohorts of interest. 71 70 I also performed regressions that included cohorts born in 1954-1965 and those born in 1987-1988). The results do not differ significantly from these reported in this paper. 71 I also estimated regressions that include omitted groups and the results (size, sign and significance of the estimated coefficients) are not very different from the regression results for the 1966-1986 sample reported in section 5. 101 5. RESULTS The first part of my analysis is focused on the transition from unmarried into married state. The estimation does not consider competing risks, such as entering informal or polygamous relationship as there are very few women in the survey who are in such relationships. The second part considers factors that affect female reproductive behavior measured by first births. 5.1 Marriage analysis 5.1.1 Descriptive analysis of age at first marriage Table 25 presents the cumulative probability of being married by age 18, 20 and 23 by three- year birth cohort for women born in 1954-1983. Note that younger women, born in 1981- 1983 would not have been exposed to the possibility of being married by age 23 in 2003. From Table 25 we can observe that the proportion of women married by age 18 increased by six percentage points for the cohort born in 1975-1977 as compared to women in the birth cohorts preceding and following this cohort. The cohort born in 1975-1977 reached age 18 between 1993 and 1995, or during the first harsh years of the Tajik civil war. Thus, a relatively large increase in the proportion of women married by age 18 may indicate that some families rushed to marry their daughters early at the start of the war. Further, starting with the cohorts born in 1972-1974 we can observe a significant decline in the proportion of women married by age 20 as compared to the older cohorts. Next, the proportion of women not married by age 23, the age an unmarried girl is considered a spinster in Tajikistan, has significantly increased among those born in 1972-1980. 102 To test whether there are significant differences in age at first marriage among women living in the conflict affected and lesser affected regions, I estimate a set of ordinary least squares (linear probability) models. In these models a dependent variable is a dummy variable defined as one if a woman was married by age 18, 20 and 23 and the main independent variable of interest is an individual’s residence in the conflict affected area interacted with a dummy variable indicating whether individual was born between 1975 and 1983, i.e. was of marriageable age during the war. The estimation results are reported in Tables 26, 27 and 28. All regressions are estimated with clustering at the raion (district level). This procedure assumes that instead of N independent observations located in M regions, we have “M” independent groups of observations. Robust standard errors are estimated with the White-Huber sandwich estimator of variance and appear in brackets. Table 25 - Marriage age by birth cohort. Age first married (%) 3 year birth cohort Median marriage age 18 and below 20 and below 23 and below Not married by 23 N observations 1954-1956 19 36.5 76.6 91.1 8.8 192 1957-1959 19 32.5 71.4 91.2 8.8 342 1960-1962 20 29.7 74.9 91.8 8.2 415 1963-1965 19 27.9 70.7 87.3 12.7 448 1966-1968 19 28.9 73.6 90.0 10.0 440 1969-1971 20 26.9 72.0 87.9 12.1 479 1972-1974 19 29.7 68.6 83.0 17.0 512 1975-1977 19 36.0 64.9 80.3 19.7 542 1978-1980 19 29.3 55.1 68.2 31.7 655 1981-1983 18 21.9 42.2 - - 775 Total 19 29.0 64.1 - - 4,800 Source: TLSS (2003). Author’s calculations. Women born in 1954-1983. 103 Interactions between birth cohort dummies and residence in the conflict area In the subset of regressions reported in Tables 26, 27 and 28 (columns 3-8), I allowed the coefficients for the cohort terms to vary with the residence of women prior to their marriage for women who were of marriageable age during the war. 72 The main variable of interest is the interaction of woman’s residence in the conflict affected area and her being of marriageable age during the war, i.e. born between 1975 and 1983. One can then test the null hypothesis that estimated coefficient on this interaction term (treatment effect) is not statistically different from null. If this hypothesis is true, then the coefficients on the birth cohorts do not vary by women’s residence in the conflict affected area. I perform a set of F-tests, testing whether the estimated coefficient for the interactive term is equal to zero in all specifications. The tests indicate that the treatment effect: being born in 1975-1983 and residing in the conflict affected region before age 12, has a negative and statistically significant effect on the probability of being married by age 18 and age 20 (6.7 and 10.2 percent respectively). Effect of residence in the conflict affected area From Col. 1 of Tables 27 and 28 we can see that residence in the conflict affected area has a significant negative impact on the probability of being married by age 20 and 23. Women in the conflict affected regions are 11.1 and 9.2 percent less likely to be married by age 20 and 23 respectively as compared to women in the lesser affected regions. Birth cohort effects Most of the estimated coefficients for the birth cohort dummies are statistically significant and different from each other, indicating that there are differential trends in age at first marriage for by birth cohort. The coefficients indicate that women born in 1975-1977 were 72 I also tested if there is a differential trend across all birth cohorts by their place of residence. The results shown that the difference existed primarily for the cohort of women born in 1975-1983 (not reported). 104 6.7 percent more likely to enter marriage by age 18 as compared to cohorts born in 1966- 1968, while cohorts born in 1981-1983 and 1984-1986 were 7.3 and 16.7 percent less likely to be married by age 18. Further, results from Col. 1 in Tables 27 and 28 indicate that the propensity to enter marriage by age 20 or age 23 is lower for the younger birth cohorts. Other effects Next I include in the regressions additional variables such as dummies for region of residence, completion of nine grades of schooling, sex ratio and regional variables. The set of regional characteristics includes percentage of population in a raion employed, crime rate and number of doctors per 1,000 people in a raion. The completion of nine grades of schooling reduces the probability of a woman to be married by age 18, 20 or 23 by 10.5 (significant at 5%), 8.5 and 9.8 (significant at 1%) percent respectively (Column 3 in Tables 26, 27 and 28). Among the regional characteristics, only the coefficient on the proportion of district population employed is statistically significant. The variable has a negative impact on the probability of being married by age 18, 20 or 23. This result shows that if a woman resided in a district with high level of employment prior to reaching official minimum marriage age (age 17) then she was less likely to be married early. One explanation for a lower hazard of marriage in the region with high employment rates may be then the presence of opportunities other than marriage for young women. The probability of getting married declines by 2.5 to 5.1 percent for each 10 percentage point increase in proportion of population employed. The effect of employment on probability of being married by age 18, 20 and 23 declines once controls for individual’s residence in Dushanbe, RRS, Khatlon or GBAO are added to the regressions, indicating that there are significant differences in the employment levels across the regions. The results of the F-tests indicate that region of residence has a significant impact on the dependent variable of interest (Column 8 in Tables 26, 27 and 28). 105 Table 26 - OLS Regressions. Married by 18. Ages 18-37 in 2003. (1) (2) (3) (4) (5) RCA -0.027 0.008 0.008 0.008 0.006 [0.038] [0.032] [0.032] [0.032] [0.032] Birth Cohort 1969-1971 -0.018 -0.019 -0.016 -0.019 -0.015 [0.031] [0.031] [0.032] [0.031] [0.031] 1972-1974 0.006 0.005 0.011 0.005 0.009 [0.030] [0.030] [0.031] [0.030] [0.031] 1975-1977 0.069** 0.115*** 0.116*** 0.114*** 0.128*** [0.033] [0.038] [0.040] [0.040] [0.040] 1978-1980 0.000 0.046 0.044 0.045 0.059 [0.037] [0.041] [0.043] [0.043] [0.042] 1981-1983 -0.073** -0.027 -0.027 -0.027 -0.031 [0.030] [0.038] [0.040] [0.041] [0.042] 1984-1986 -0.167*** -0.168*** -0.178*** -0.169*** -0.191*** [0.027] [0.027] [0.029] [0.029] [0.036] -0.067* -0.075** -0.068** -0.082** (Born in 1975-1983)* RCA [0.034] [0.034] [0.033] [0.035] -0.105*** -0.098*** Completed 9 grades [0.023] [0.022] Regional controls Sex ratio -0.017 [0.126] -0.284** % raion population employed [0.137] N doctors per 1,000 raion population N crimes per 1,000 raion population Region of residence controls no no no no no Constant 0.285*** 0.262*** 0.359*** 0.278** 0.445*** [0.039] [0.037] [0.043] [0.123] [0.067] Observations 3959 3959 3895 3959 3895 R-squared 0.03 0.03 0.03 0.03 0.04 F-tests, p-values (Born in 1975-1983) *RCA 0.050 0.032 0.045 0.024 (Born in 1975-1983)*RCA, RCA 0.131 0.094 0.119 0.075 Residence controls 106 Table 26 (continued) - OLS Regressions. Married by 18. Ages 18-37 in 2003. (6) (7) (8) RCA 0.000 0.000 -0.010 [0.031] [0.032] [0.051] Birth Cohort 1969-1971 -0.016 -0.018 -0.016 [0.031] [0.031] [0.032] 1972-1974 0.002 -0.004 0.003 [0.030] [0.030] [0.028] 1975-1977 0.126*** 0.120*** 0.116*** [0.041] [0.039] [0.038] 1978-1980 0.061 0.068 0.061 [0.043] [0.043] [0.041] 1981-1983 -0.035 -0.03 -0.031 [0.041] [0.042] [0.040] 1984-1986 -0.184*** -0.185*** -0.182*** [0.032] [0.035] [0.031] -0.072* -0.077** -0.081** (Born in 1975-1983)* RCA [0.037] [0.037] [0.035] Completed 9 grades Regional controls Sex ratio -0.353** -0.401*** -0.228** % raion population employed [0.138] [0.137] [0.107] 0.003 N doctors per 1,000 raion population [0.009] 0.006* 0.006 N crimes per 1,000 raion population [0.003] [0.005] Region of residence controls no no yes Constant 0.365*** 0.358*** 0.318*** [0.066] [0.062] [0.057] Observations 3959 3959 3959 R-squared 0.04 0.04 0.05 F-tests, p-values (Born in 1975-1983) *RCA 0.055 0.038 0.023 (Born in 1975-1983)*RCA, RCA 0.156 0.114 0.073 Residence controls 0.000 Notes: Columns represent OLS coefficients. Standard errors (in brackets) are corrected for heteroscedasticity and are robust to clustered residuals across individuals who resided in the same raion at age 11. Reference categories: born in 1966-1968, “Resident of Sugd”. Sample: born in 1966-1986 (age 18 to 37 in 2003). “Married by 18” – is an indicator variable equal to one if a woman was married at age 18 or earlier, the variable is equal to zero otherwise. Regressions also include dummy variable controls for missing information on the regional control variable (when included). All regressions include a dummy variable controlling for residence in the rural area. *significant at 10%; ** significant at 5%; *** significant at 1%. 107 Table 27 - OLS Regressions. Married by age 20. Ages 20-37 in 2003. (1) (2) (3) (4) (5) RCA -0.112** -0.051 -0.052 -0.051 -0.029 [0.051] [0.047] [0.047] [0.047] [0.056] Birth Cohort 1969-1971 -0.010 -0.013 -0.012 -0.013 -0.012 [0.031] [0.031] [0.033] [0.031] [0.032] 1972-1974 -0.056* -0.057* -0.050* -0.056* -0.055* [0.029] [0.029] [0.030] [0.029] [0.030] 1975-1977 -0.092*** -0.022 -0.017 -0.025 0.004 [0.034] [0.043] [0.044] [0.046] [0.048] 1978-1980 -0.195*** -0.126*** -0.126*** -0.129*** -0.100** [0.036] [0.044] [0.045] [0.046] [0.047] 1981-1983 -0.321*** -0.251*** -0.249*** -0.254*** -0.253*** [0.029] [0.040] [0.040] [0.042] [0.045] (1975-1983)*RCA -0.102** -0.109** -0.104** -0.146*** [0.044] [0.044] [0.043] [0.052] -0.085** -0.076** Completed 9 grades [0.035] [0.034] Regional controls Sex ratio -0.075 [0.165] -0.443** % raion population employed [0.220] N doctors per 1,000 raion population N crimes per 1,000 raion population Region of residence controls no no no no no Constant 0.767*** 0.725*** 0.805*** 0.795*** 0.938*** [0.049] [0.047] [0.053] [0.156] [0.083] Observations 3403 3403 3346 3403 3346 R-squared 0.07 0.08 0.08 0.08 0.09 F-tests, p-values (Born in 1975-1983)*RCA 0.023 0.015 0.018 0.007 (Born in 1975-1983)*RCA, RCA 0.033 0.019 0.026 0.006 Residence controls 108 Table 27 (continued) - OLS Regressions. Married by age 20. Ages 20-37 in 2003. (6) (7) (8) RCA -0.038 -0.031 0.013 [0.052] [0.055] [0.071] Birth Cohort 1969-1971 -0.010 -0.012 -0.012 [0.031] [0.031] [0.033] 1972-1974 -0.061** -0.067** -0.054** [0.028] [0.030] [0.026] 1975-1977 -0.007 -0.004 -0.008 [0.051] [0.048] [0.045] 1978-1980 -0.108** -0.093* -0.102** [0.049] [0.049] [0.045] 1981-1983 -0.261*** -0.256*** -0.257*** [0.046] [0.047] [0.042] (1975-1983)*RCA -0.129** -0.145** -0.132*** [0.054] [0.056] [0.047] Completed 9 grades Regional controls Sex ratio -0.384** -0.530** -0.253* % raion population employed [0.161] [0.246] [0.139] -0.006 N doctors per 1,000 raion population [0.014] 0.004 0.003 N crimes per 1,000 raion population [0.005] [0.005] Region of residence controls no no yes Constant 0.884*** 0.876*** 0.817*** [0.077] [0.072] [0.054] Observations 3403 3403 3403 R-squared 0.09 0.09 0.13 F-tests, p-values (Born in 1975-1983)*RCA 0.020 0.011 0.006 (Born in 1975-1983)*RCA, RCA 0.017 0.013 0.023 Residence controls 0.000 Notes: Columns represent OLS coefficients. Standard errors (in brackets) are corrected for heteroscedasticity and are robust to clustered residuals across individuals who resided in the same raion at age 11. Reference categories: born in 1966-1968, “Resident of Sugd”. Sample: born in 1966-1983 (age 20 to 37 in 2003). “Married by 20” – is an indicator variable equal to one if a woman was married at age 20 or earlier, the variable is equal to zero otherwise. Regressions also include dummy variable controls for missing information on the regional control variable (when included). All regressions include a dummy variable controlling for residence in the rural area. * significant at 10%; ** significant at 5%; *** significant at 1%. 109 Table 28 - OLS Regressions. Married by age 23. Ages 23-37 in 2003. (1) (2) (3) (4) (5) RCA -0.092** -0.057* -0.060* -0.056* -0.039 [0.044] [0.031] [0.031] [0.032] [0.039] Birth Cohort 1969-1971 -0.016 -0.018 -0.017 -0.018 -0.018 [0.018] [0.018] [0.018] [0.018] [0.018] 1972-1974 -0.074*** -0.075*** -0.071** -0.074*** -0.075*** [0.027] [0.027] [0.027] [0.027] [0.028] 1975-1977 -0.101*** -0.05 -0.052 -0.054 -0.035 [0.032] [0.048] [0.046] [0.049] [0.048] 1978-1980 -0.225*** -0.174*** -0.185*** -0.178*** -0.163*** [0.034] [0.046] [0.045] [0.048] [0.045] (1975-1983)*RCA -0.076 -0.078* -0.078* -0.108** [0.046] [0.044] [0.045] [0.050] Completed 9 grades -0.098*** -0.088*** [0.032] [0.029] Regional controls Sex ratio -0.097 [0.141] -0.425** % raion population employed [0.203] N doctors per 1,000 raion population N crimes per 1,000 raion population Region of residence controls no no no no no Constant 0.926*** 0.901*** 0.998*** 0.992*** 1.129*** [0.039] [0.036] [0.040] [0.136] [0.067] Observations 2628 2628 2589 2628 2589 R-squared 0.06 0.06 0.07 0.06 0.08 F-tests, p-values (Born in 1975-1980)*RCA 0.106 0.083 0.086 0.034 (Born in 1975-1980)*RCA, RCA 0.120 0.090 0.116 0.060 Residence controls 110 Table 28 (continued) - OLS Regressions. Married by age 23. Ages 23-37 in 2003. (6) (7) (8) RCA -0.039 -0.035 0.030 [0.039] [0.040] [0.045] Birth Cohort 1969-1971 -0.018 -0.018 -0.020 [0.018] [0.018] [0.020] 1972-1974 -0.077*** -0.083*** -0.064*** [0.025] [0.028] [0.022] 1975-1977 -0.039 -0.035 -0.033 [0.050] [0.050] [0.043] 1978-1980 -0.160*** -0.149*** -0.163*** [0.048] [0.048] [0.046] (1975-1983)*RCA -0.099** -0.111** -0.082* [0.049] [0.054] [0.048] Completed 9 grades Regional controls Sex ratio -0.317* -0.499** -0.109 % raion population employed [0.174] [0.233] [0.118] -0.009 N doctors per 1,000 raion population [0.016] 0.003 -0.004 N crimes per 1,000 raion population [0.004] [0.004] Region of residence controls no no yes Constant 1.054*** 1.050*** 0.970*** [0.055] [0.055] [0.034] Observations 2628 2628 2628 R-squared 0.07 0.07 0.15 F-tests, p-values (Born in 1975-1980)*RCA 0.046 0.045 0.091 (Born in 1975-1980)*RCA, RCA 0.082 0.083 0.169 Residence controls 0.000 Notes: Columns represent OLS coefficients. Standard errors (in brackets) are corrected for heteroscedasticity and are robust to clustered residuals across individuals who resided in the same raion at age 11. Reference categories: born in 1966-1968, “Resident of Sugd”. Sample: born in 1966-1980 (age 23 to 37 in 2003). “Married by 23” – is an indicator variable equal to one if a woman was married at age 23 or earlier, the variable is equal to zero otherwise. Regressions also include dummy variable controls for missing information on the regional control variable (when included). All regressions include a dummy variable controlling for residence in the rural area. * significant at 10%; ** significant at 5%; *** significant at 1%. 111 5.1.2 Marriage hazard analysis I continue with the semi-parametric analysis to estimate the impact of various covariates on entry (hazard rate) into first marriages by females. The hazard analysis is more appropriate for the duration data, such as time to first marriage, as it accounts for censoring of the observations and exit of individuals from the analysis upon marriage at different age. The PH model is used to estimate the hazard ratios (parameter effects). The aim of this model is to establish the relationship between an individual, cohort and region specific covariates and entry into first marriage. Table 29 reports estimation results from seven different specifications of the PH model. All regressions are specified with random effects (frailty) at the raion level to control for unobserved heterogeneity. All tables report estimated regression coefficients in the exponential form (hazard rates). Standard errors are in brackets. The standard errors are conditional on the estimated intra-group varianceθ. The standard errors are treated as conditional on θ fixed at its optimal value, as estimated from the data. The main variable of interest in the marriage hazard regressions is the coefficient on the interaction term between the dummy variable for being of in the prime marriage age during the war and an individual’s residence in the conflict affected area. 112 Interaction terms Column 2 of Table 29 reports regressions results that include an interaction term, that is constructed by multiplying the dummy variable for being born in 1975-1983 by the dummy indicating that an individual lived in the region affected by the conflict (RCA). The addition of this interaction term to the regression reduces the direct effect of the residence in the conflict zone on entry into marriage. The χ 2 -test reported in Column 2 indicates that residence in the conflict affected regions had a significant and negative impact on the marriage hazard for women born in 1975-1983. The results are significant at 1 percent level. I also perform a set of several joint-χ 2 tests evaluating the significance of the interaction terms once other regional and individual characteristics are added to the regressions (Columns 4-8). Residence in the conflict affected area The first reported regression in Column 1 of Table 29 shows a negative (significant at 5% level) contribution of an individual’s residence in the conflict affected region to the hazard of entry into marriage. The estimated hazard rate of almost 0.80 for the residence in the conflict region dummy should be interpreted in the following way: the hazard for women in the conflict affected areas is about 20 percent less than the hazard rate for women in the less affected areas. Birth cohort effects The estimated effect of the time trend on entry into marriage is significant assuming either linear (including a control for the birth year only, not reported) or non-linear (cohort dummies) effects (Column 1). Further, the estimated coefficients on birth cohort dummies are significantly different from each other for the cohorts born in 1975-1986. The 113 coefficients follow a downward pattern with older birth cohorts being more likely to be married as compared to younger birth cohorts. Education The number of years of education completed reduces entry into marriage (results not reported). The estimated hazard rate for “completed nine grades of education” variable indicates a negative effect of education on entry into marriage (not significant) (Column 3). This relatively weak result can be explained by the fact that the majority of women (88.2%) in the analytical sample have completed this level of education. Sex Ratio The estimated hazard rate associated with the Sex Ratio variable suggests that an increase in the sex ratio decreases entry into marriage (Column 4). This is a surprising result as we would have expected that high sex ratios increase entry in marriage due to the increased pool of marriage age partners, and thus a better choice. However, the coefficient on the sex ratio variable is not statistically significant. This weak result may be associated with the prevalence of customs in Tajikistan when people marry a member of an extended family or someone from a larger group that shares similar ancestral roots or place of residence. Thus, the sex ratio variable based on the Tajik Census data may not adequately measure the relevant pool of marriageable partners for some individuals. Regional Controls Columns 5-8 of Table 29 include a set of regional controls to test for the robustness of the treatment effect (i.e. ‘Born in 1975-1983’*RCA). None of the regional controls is statistically significant. Their inclusion in the regression slightly decreases the stand-alone effect of the treatment variable (from 0.709 in Col. 2 to 0.671 in Col. 8) but does not have an effect on the statistical significance of this term. 114 Table 29 - Semi-parametric marriage hazard regressions (Cox proportional hazard). (1) (2) (3) (4) RCA 0.940 0.933 0.941 0.943 [0.102] [0.102] [0.102] [0.104] Birth cohort 1969-1971 0.978 0.980 0.978 0.979 [0.067] [0.067] [0.067] [0.067] 1972-1974 0.905 0.903 0.905 0.897 [0.062] [0.062] [0.062] [0.062] 1975-1977 1.117 1.115 1.105 1.136 [0.097] [0.098] [0.098] [0.101] 1978-1980 0.836** 0.822** 0.826** 0.838** [0.072] [0.072] [0.073] [0.074] 1981-1983 0.566*** 0.566*** 0.559*** 0.564*** [0.052] [0.052] [0.052] [0.052] 1984-1986 0.323*** 0.308*** 0.319*** 0.299*** [0.037] [0.036] [0.037] [0.036] (1975-1983)*RCA 0.709*** 0.713*** 0.708*** 0.693*** [0.059] [0.059] [0.059] [0.063] Completed 9 grades 0.907 0.911 [0.058] [0.059] Regional controls Sex ratio 0.788 [0.273] 0.664 % raion population employed [0.208] N doctors per 1,000 raion population N crimes per 1,000 raion population Region of residence controls no no no no Observations 4223 4153 4223 4153 Number of groups 67 67 67 67 Log likelihood -20488.4 -20185.6 -20488.2 -20184.7 Chi2-tests, p-values (1975-1983)*RCA 0.000 0.000 0.000 0.000 (1975-1983)*RCA, RCA 0.000 0.000 0.000 0.000 Residence controls 115 Table 29 (continued) - Semi-parametric marriage hazard regressions (Cox proportional hazard). (5) (6) (7) RCA 0.983 0.957 1.063 [0.110] [0.105] [0.131] Birth cohort 1969-1971 0.970 0.977 0.972 [0.066] [0.067] [0.066] 1972-1974 0.899 0.897 0.899 [0.062] [0.063] [0.063] 1975-1977 1.150 1.142 1.136 [0.102] [0.101] [0.101] 1978-1980 0.863* 0.854* 0.846* [0.076] [0.076] [0.075] 1981-1983 0.567*** 0.562*** 0.558*** [0.052] [0.052] [0.051] 1984-1986 0.308*** 0.309*** 0.306*** [0.037] [0.037] [0.036] (1975-1983)*RCA 0.664*** 0.682*** 0.671*** [0.060] [0.061] [0.060] Completed 9 grades Regional controls Sex ratio 0.590 0.597 0.666 % raion population employed [0.213] [0.202] [0.215] 1.002 N doctors per 1,000 raion population [0.026] 1.000 0.997 N crimes per 1,000 raion population [0.011] [0.013] Region of residence controls no no yes Observations 4223 4223 4223 Number of groups 67 67 67 Log likelihood -20485.1 -20487.0 -20470.2 Chi2-tests, p-values (1975-1983)*RCA 0.000 0.000 0.000 (1975-1983)*RCA, RCA 0.000 0.000 0.000 Residence controls 0.000 Notes: Women born in 1966-1986 (ages 17-37 in 2003). Columns represent hazard ratios. Standard errors are in brackets. All regressions include controls for heterogeneity at the raion level (frailty effects) specified with gamma distribution. Reference categories: born in 1966-1968, “Resident of Sugd”. Subjects enter analysis at age 11. Sample: born in 1966-1986. Regressions also include dummy variable controls for missing information on the regional control variable (when included). All regressions include a dummy variable controlling for residence in the rural area. * significant at 10%; ** significant at 5%; *** significant at 1%. 116 5.2 Fertility analysis 5.1.1 Descriptive analysis of first births In this section I discuss results from the conditional and unconditional analyses of first births. I start with the descriptive analysis of the data on first births by women surveyed in the 2003 TLSS. Table 30 presents cumulative probability of having first child by age 18, 20 and 23 for women born in 1954-1983 by three-year birth cohort. Table 30 - Cumulative distribution of age at first birth, by 3-year birth cohort. Age at first birth (%) 3 year birth cohort Median age at first birth 18 and below 20 and below 23 and below No first birth by age 23 N obs 1954-1956 19 3.1 29.2 59.4 40.6 192 1957-1959 19 4.7 28.1 61.4 38.6 342 1960-1962 20 4.3 26.5 68.2 31.8 415 1963-1965 19 4.7 28.4 69.9 30.1 448 1966-1968 19 5.0 33.2 73.9 26.1 440 1969-1971 20 7.7 38.0 72.7 27.4 479 1972-1974 19 7.4 37.1 70.5 29.5 512 1975-1977 19 14.9 43.7 70.1 29.9 542 1978-1980 19 9.8 34.7 57.4 42.6 655 1981-1983 18 11.0 25.3 - - 775 Total 19 8.1 32.6 - - 4,800 Note: women born in 1954-1983. Source: TLSS 2003. Author’s calculations. Table 31 - Years between marriage and first birth, by number of years married. Years between marriage and first birth Years married 0 to 1 <=2 <=3 <=4 <=5 <=6 No child N 0 9.3 - - - - - 90.7 43 1 29.1 - - - - - 70.9 103 2 63.0 76.5 - - - - 23.5 119 3 59.8 74.6 80.3 - - - 19.7 122 4 64.5 79.4 86.5 90.8 - - 9.2 141 5 49.0 74.8 83.9 86.7 89.5 - 10.5 143 6 plus 46.4 68.2 78.8 85.3 88.9 95.3 4.7 1,744 N 1,153 472 214 124 67 111 274 2,415 Note: Sample: ever married women born in 1966-1983. First births given after the date of first marriage. Source: TLSS 2003. Author’s calculations. 117 The data in Table 30 point towards a trend of decreasing age at first birth, in particular for the cohorts born in 1975-1983. 14.9 and 43.7 percent of women born in 1975-1977 had their first child by age 18 and 20 respectively, as compared to 7.4 and 37.1 percent of women born in 1972-1974. Women born in 1975-1977 were of age 15-17 when the war started in 1992. From the previous section, we know that almost 36% of women in this birth cohort were also married by age 18 as compared to about 30% for the preceding and following birth cohorts. Further, tabulations in Table 30 indicate that about 10 to 11 percent of women born in 1978-1980 and 1981-1983 respectively had their first child by age 18. Table 31 tabulates the numbers on proportion of women who had their first child within N years of their marriage by the number of years those women were married. From Table 11 we can observe that among women who were married for at least two to five years, almost three fourths reported having their first child within two years of their marriage. At least 50 percent of those women also reported having their first child within the first year of their marriage. Those results are not surprising as contraception in Tajikistan is not widespread as it was discussed in Section 2.3 above. In 2000, among currently married women aged 15-19 and 20-24, less than one in ten and fewer than one in five women respectively reported to be using contraception in 2000 (UNICEF 2001). Thus, an earlier exposure to a chance of conception due to an earlier age at first marriage must be a likely explanation for an earlier age at first birth for some women born in 1975-1977. 5.2.2 Hazard analysis of first births I would like to further explore the birth cohort and regional patterns in the timings of first births in the regression framework, as in the preceding section we learned that there were differential trends across regions and birth cohorts in entry into marriage. For this, I turn to 118 the hazard analysis of first births for women born in 1966-1986. These women were at least 17 years old in 2003 and thus were eligible to be married under the Tajik family law. In the analysis of duration data I focus on two transition processes. In the first part of my analysis, I evaluate factors explaining differential trends in first marital births. The transition occurs in year t when a married women who did not have a child at the start of year t, gives birth to her first child in year t. In this analysis exposure to conception is defined as age at which a woman was first married. This estimation is performed on the sample of ever married women born in 1966-1986. The sample excludes from the analysis women who had their first child prior to the date of their first marriage or who were never married. The observation on a woman is censored at the date of the 2003 TLSS survey if a woman was married by that date but did not have children. The second part of my analysis is focused on the transition from “no child” status at the start of year t to “have first child” at the end of year t. The exposure to first conception is defined as age 14. Thus, the analytical sample includes all women born in 1966-1986 as opposed to the analysis in this section that is focused only on ever married women. As in the analysis of marriage hazards, the main variable of interest is the interaction of woman’s residence in the conflict affected area and her being of marriageable age during the war, i.e. born between 1975 and 1983. First marital births Results from the CPH regressions of first birth conditional on being married for women born in 1966-1986 are presented in Table 32 that reports hazard rates estimated for all independent variables. Robust standard errors account for clustered residuals across each raion and are reported in brackets. 119 Interaction terms The results reported in Column 3 of Table 32 include the estimated hazard rate for the interactive term that controls for the regional and year of birth exposure to the conflict for the cohorts who were of marriageable age during the war (i.e. born in 1975-1983). As in the previous section on marriage hazard, I perform a set of χ 2 tests of the estimated hazard rates for the interaction terms across various specifications in Table 32. The χ 2 tests consistently indicate that the estimated hazard rates for this treatment effect are not statistically different from one. Thus there is no significant difference in the hazard rates of first birth across the birth cohort coefficients by women’s residence in the conflict affected area. The above results are not affected by the addition of women’ education levels and regional controls to the regressions. The effects of those variables on age at first birth for ever married women are reported below. Residence in the conflict affected area The estimated hazard rate for the residence in the conflict affected area variable is less than one (Column 1). Thus women who lived in more affected areas were less likely to have their first child as soon as women in lesser affected areas. However, this result is not significant in all of the estimated models. 120 Birth cohort dummies The hazard rates estimated for the birth cohort dummies indicate that women born in 1975- 1977 and 1978-1980 were 12 percent more likely to have their first child sooner as compared to the older cohorts (significant at 10 percent level). Women born in 1984-1986 were 53 percent less likely to have their first child soon after their marriage (significant at 5% level). The estimated hazard rates for other cohort dummies are not significantly different from one, indicating a rather uniform child-bearing behavior for cohorts born in 1966-1968 (reference), 1969-1971, 1972-1974 and 1981-1983. Regional Controls The uniformity of reproductive behavior of married women across regions in Tajikistan is further confirmed by the regression results (Col. 8) that include controls for residence in one of the regions in Tajikistan. Further, none of the regional controls, such as proportion of population employed, crime rate and number of doctors per 1,000 raion population have a significant impact on the risk of conception as measured by first birth (Columns 4-8). 121 Table 32 - Semi-parametric conditional fertility hazard regressions (Cox proportional hazard). (1) (2) (3) (4) (5) RCA 0.985 0.941 0.922 0.94 0.924 [0.040] [0.048] [0.047] [0.055] [0.059] Birth cohort 1969-1971 1.080 1.082 1.075 1.081 1.086 [0.067] [0.067] [0.068] [0.067] [0.069] 1972-1974 1.020 1.020 1.003 1.018 1.022 [0.065] [0.066] [0.065] [0.067] [0.067] 1975-1977 1.120* 1.046 1.010 1.057 1.049 [0.077] [0.072] [0.069] [0.075] [0.074] 1978-1980 1.122* 1.050 1.015 1.057 1.050 [0.072] [0.065] [0.067] [0.067] [0.065] 1981-1983 1.074 1.006 0.969 1.006 1.006 [0.092] [0.078] [0.074] [0.080] [0.080] 1984-1986 0.466** 0.466** 0.401*** 0.462** 0.473** [0.141] [0.141] [0.134] [0.144] [0.147] (1975-1983)*RCA 1.108 1.133* 1.102 1.128 [0.077] [0.080] [0.088] [0.095] 0.970 Completed 9 grades [0.055] Regional controls Sex ratio 0.850 1.033 % raion population employed [0.215] [0.352] 0.986 N doctors per 1,000 raion population [0.014] N crimes per 1,000 raion population Region of residence controls no no no no no Observations 2031 2031 2005 2031 2031 Log likelihood -12290.9 -12290.3 -12125.1 -12289.9 -12289.4 Prob>Wald Chi2 0.005 0.005 0.014 0.016 0.040 Chi2-tests, p-values Interactions (1975-1983)*RCA 0.138 0.079 0.221 0.151 (1975-1983)*RCA, RCA 0.305 0.173 0.451 0.343 Residence controls 122 Table 32 (continued) - Semi-parametric conditional fertility hazard regressions (Cox proportional hazard). (6) (7) (8) RCA 0.941 0.94 0.968 [0.057] [0.053] [0.060] Birth cohort 1969-1971 1.081 1.083 1.083 [0.068] [0.067] [0.068] 1972-1974 1.020 1.014 1.021 [0.067] [0.066] [0.067] 1975-1977 1.060 1.068 1.054 [0.076] [0.077] [0.075] 1978-1980 1.056 1.071 1.052 [0.067] [0.070] [0.063] 1981-1983 1.005 1.017 1.000 [0.080] [0.081] [0.079] 1984-1986 0.463** 0.467** 0.457** [0.144] [0.146] [0.143] (1975-1983)*RCA 1.103 1.104 1.101 [0.089] [0.088] [0.089] Completed 9 grades Regional controls Sex ratio 1.349 [0.303] 0.872 0.809 0.949 % raion population employed [0.213] [0.215] [0.260] N doctors per 1,000 raion population 0.998 N crimes per 1,000 raion population [0.007] Region of residence controls no no yes Observations 2031 2031 2031 Log likelihood -12289.9 -12289.3 -12289.2 Prob>Wald Chi2 0.024 0.022 0.046 Chi2-tests, p-values Interactions (1975-1983)*RCA 0.222 0.218 0.234 (1975-1983)*RCA, RCA 0.461 0.434 0.485 Residence controls 0.642 Notes: Women born in 1966-1986. Columns represent hazard ratios. Standard errors (in brackets) are corrected for heteroscedasticity and are robust to clustered residuals across individuals residing in the same raion. Reference categories: born in 1966-1968, “Resident of Sugd”. Subjects enter analysis at the age when they were first married. Sample: born in 1966 -1986. Regressions also include dummy variable controls for missing information on the regional control variable (when included). All regressions include a dummy variable controlling for residence n the rural area. * significant at 10%; ** significant at 5%; *** significant at 1%. 123 Table 33 - Semi-parametric unconditional fertility hazard regressions (Cox proportional hazard). (1) (2) (3) (4) RCA 0.835* 0.923 0.905 0.961 [0.083] [0.067] [0.068] [0.079] Birth cohort 1969-1971 1.033 1.030 1.036 1.042 [0.054] [0.054] [0.056] [0.053] 1972-1974 0.895 0.894 0.881* 0.891* [0.065] [0.066] [0.066] [0.062] 1975-1977 0.935 1.081 1.055 1.131 [0.081] [0.113] [0.104] [0.130] 1978-1980 0.708*** 0.816** 0.789** 0.856 [0.062] [0.076] [0.075] [0.088] 1981-1983 0.530*** 0.613*** 0.597*** 0.605*** [0.048] [0.068] [0.065] [0.073] 1984-1986 0.226*** 0.226*** 0.207*** 0.210*** [0.041] [0.041] [0.039] [0.041] (1975-1983)*RCA 0.805** 0.817** 0.754** [0.078] [0.073] [0.084] 0.802** Completed 9 grades [0.070] Regional controls Sex ratio 0.404** % raion population employed [0.145] N doctors per 1,000 raion population 0.990 N crimes per 1,000 raion population [0.038] Region of residence controls no no no no Observations 4235 4235 4164 4235 Log likelihood -17739.3 -17736.3 -17487.2 -17724.4 Prob>Wald Chi2 0.000 0.000 0.000 0.000 Chi2-tests, p-values Interactions (1975-1983)*RCA 0.024 0.025 0.012 (1975-1983)*RCA, RCA 0.076 0.068 0.034 Residence controls 124 Table 33 (continued) - Semi-parametric unconditional fertility hazard regressions (Cox proportional hazard). (5) (6) (7) RCA 0.968 0.984 1.056 [0.084] [0.084] [0.110] Birth cohort 1969-1971 1.039 1.036 1.043 [0.052] [0.052] [0.052] 1972-1974 0.873* 0.885 0.922 [0.065] [0.066] [0.056] 1975-1977 1.124 1.162 1.156 [0.126] [0.144] [0.103] 1978-1980 0.885 0.882 0.867* [0.091] [0.096] [0.071] 1981-1983 0.612*** 0.618*** 0.616*** [0.074] [0.079] [0.062] 1984-1986 0.206*** 0.210*** 0.220*** [0.043] [0.044] [0.041] (1975-1983)*RCA 0.727** 0.734** 0.753*** [0.092] [0.091] [0.079] Completed 9 grades Regional controls Sex ratio 1.661 [0.659] 0.281*** 0.316** 0.690 % raion population employed [0.133] [0.156] [0.178] 1.015* N doctors per 1,000 raion population [0.009] N crimes per 1,000 raion population Region of residence controls no no yes Observations 4235 4235 4235 Log likelihood -17722.7 -17723.1 -17656.1 Prob>Wald Chi2 0.000 0.000 0.000 Chi2-tests, p-values Interactions (1975-1983)*RCA 0.012 0.012 0.007 (1975-1983)*RCA, RCA 0.036 0.041 0.024 Residence controls 0.000 Notes: Women born in 1966-1986 (age 17-37 in 2003). Columns represent hazard ratios. Standard errors (in brackets) are corrected for heteroscedasticity and are robust to clustered residuals across individuals residing in the same raion. Reference categories: born in 1966-1968, “Resident of Sugd”. Subjects enter analysis at age 14. Regressions also include dummy variable controls for missing information on the regional control variable (when included). All regressions include a dummy variable controlling for residence in the rural area. * significant at 10%; ** significant at 5%; *** significant at 1%. 125 Unconditional Fertility Hazard Estimation The analysis of the age of first births in the preceding section takes age at first marriage as exogenous. The estimation of the hazard rates of first births is conditional on the woman’s marriage. The results of those estimations may be biased if female age at first marriage is affected by the exposure to the armed conflict. The results from the analysis of conditional birth hazards discussed above indicate that the first birth hazard, conditional on being married, increased for cohorts born in 1975- 1980. However, some of the first births may have occurred out of the wedlock. 73 Thus, using only a conditional on marriage first birth hazard function may underestimate the overall impact of the time trend and regional differences on timing of first births. In attempt to correct for this omission, I estimate unconditional first birth hazard functions. The start of exposure is defined as age 14. In this estimation, the dependent variable is the odds of a woman having her first child in the year t given that she did not have children at the start of the year t. Table 33 reports unconditional first birth hazard rates. Robust standard errors account for clustered residuals across raions and are reported in brackets. Interaction effects Column 2 reports the hazard rate estimated for the interaction term between the dummy variable that is equal to one if an individual was born in 1975-1983 and woman’s residence in the conflict affected area. The results suggest that women in the conflict affected areas who were of marriageable age during the war had about 20 percent lower (significant at 5 percent level) hazard of first birth as compared to the rest of the sample. Comparing this result to the hazard rate estimated for the interactive term in Table 32, we may infer the 73 In fact, in the TLSS 2003 about 9.5 percent of women (out of 3,405) reported to have their first child before they were married. 126 delayed marriage of younger women in the regions affected by war accounts for the differences in the unconditional first birth hazards. Residence in the conflict affected area The results suggest a negative effect of the residence in the conflict affected region on age at first birth. The hazard rate (HR) of first birth is about 17 percent lower for women in the areas affected by the conflict (HR=0.835, significant at 10% level) (Column 1). Cohort Effects Column 1 also includes birth cohort dummies to control for time trend effects. The coefficients on birth cohort dummies for women born in 1978-1986 are significantly different from each other, with younger cohorts having a lower risk of conception. The estimated hazard rates for women born in 1978-1980 and 1981-1983 are 0.708 and 0.530 respectively, indicating that women are 30=100*(1-0.71) and 47=100*(1-0.53) percent less likely to have children as soon as women born in 1966-1969 (the reference group). Regional controls Among the regional controls, the proportion of employed population in a raion of residence has a significant negative impact on first births possibly indicating that women who have opportunities other than marriage are less likely to have their first child early (Columns 4-6). The estimated coefficient becomes insignificant once controls for the region of residence are added to the regression (Column 7). The estimated hazard rates for the individual’s region of residence are jointly statistically significant, indicating that there are significant regional differences in the patterns of first births. Sex ratio The effect of sex ratio on unconditional hazard of first birth is positive, with the estimated hazard rate of to 1.661, but not significant (Column 6). 127 Thus, the results in Table 32 suggest that the hazard of first birth for married women born in 1975-1983 who resided in the conflict affected regions was comparable to this of women in the rest of the sample. The results in Table 33 indicate that hazard of first birth is lower for women in the treatment group when I do not control for the effect of marriage on the birth hazard. Thus, the conflict had an effect on the rate of entry into first marriage by women who were of marriageable age during the war and who lived in the conflict affected regions. Regional characteristics do not have significant effect on the conditional or unconditional hazard of first births. Traditionally in Tajikistan childbirth closely follows marriage. Thus, delayed marriage is the main channel in which conflict had an impact on the marriage market and reproductive behavior by women who were of marriageable age during the war. 5.3 Age differences between husbands and wives In this section I explore factors affecting the age differences between husbands and wives. Table 34 presents results from a set of linear regressions where a dependent variable is a difference in age between husbands and wives and the independent variables are birth cohort dummies, interactions between birth cohort dummies and residence in the conflict affected region, and regional controls. The coefficient on the interactive term in Column 3 indicates that women who were born in 1975-1983 and who also lived in the conflict affected regions were married to men who on average were 0.74 years older to them (significant at 10 percent level). Sex ratio had a negative impact on the age difference between husbands and wives. An increase in the sex ratio by 10 percentage points would decrease age difference between husbands and wives by 0.14 years (significant at 10 percent level). Both, higher sex ratios and employment levels reduce age differences between husbands and wives. 128 The birth cohort dummies are not significantly different from zero with an exception of women from the 1984-1986 birth cohort who were more likely to get married to men 0.81 years older than them. Table 34 - Age difference (husband-wife). OLS regressions. (1) (2) (3) (4) RCA 1.017*** 1.003*** 0.634*** 0.621*** [0.204] [0.202] [0.213] [0.203] Rural -0.381* -0.487** -0.392** -0.498** [0.196] [0.220] [0.194] [0.220] Birth cohort 1969-1971 -0.119 -0.116 -0.108 -0.105 [0.273] [0.273] [0.274] [0.273] 1972-1974 -0.229 -0.210 -0.229 -0.210 [0.288] [0.290] [0.287] [0.289] 1975-1977 0.148 0.082 -0.349 -0.413 [0.268] [0.274] [0.316] [0.323] 1978-1980 0.075 -0.003 -0.407 -0.484 [0.284] [0.285] [0.325] [0.325] 1981-1983 0.990*** 0.919** 0.518 0.448 [0.351] [0.358] [0.356] [0.362] 1984-1986 0.832* 0.746* 0.818* 0.732* [0.418] [0.431] [0.414] [0.426] (born 1975-1983)*RCA 0.744** 0.742** [0.320] [0.321] Regional controls Sex ratio -1.436* -1.431* [0.847] [0.854] % raion population employed Region of residence controls no no no no Constant 3.161*** 4.516*** 3.430*** 4.779*** [0.226] [0.891] [0.213] [0.885] Observations 2316 2316 2316 2316 R-squared 0.03 0.03 0.03 0.03 F-tests, p-values (born 1975-1983)*RCA 0.024 0.024 (born 1975-1983)*RCA, RCA 0.000 0.000 Residence controls 129 Table 34 (continued) - Age difference (husband-wife). OLS regressions. (5) (6) RCA 0.643*** 0.187 [0.212] [0.280] Rural -0.599*** -0.437* [0.221] [0.238] Birth cohort 1969-1971 -0.11 -0.086 [0.275] [0.271] 1972-1974 -0.249 -0.207 [0.283] [0.283] 1975-1977 -0.286 -0.33 [0.320] [0.313] 1978-1980 -0.339 -0.383 [0.337] [0.322] 1981-1983 0.505 0.559 [0.371] [0.353] 1984-1986 0.753* 0.827* [0.432] [0.417] (born 1975-1983)*RCA 0.683* 0.717** [0.348] [0.322] Regional controls Sex ratio -1.555* % raion population employed [0.827] Region of residence controls no yes Constant 3.934*** 3.374*** [0.343] [0.241] Observations 2316 2316 R-squared 0.03 0.03 F-tests, p-values (born 1975-1983)*RCA 0.054 0.030 (born 1975-1983)*RCA, RCA 0.000 0.019 Residence controls 0.325 Notes: Born in 1966-1986. Intact couples as of 2003. Columns represent OLS coefficients. Standard errors (in brackets) are corrected for heteroscedasticity and are robust to clustered residuals across individuals who resided in the same raion at age 12. Reference categories: born in 1966-1968, “Resident of Sugd”. Regressions also include a dummy variable control for missing information on the regional control variable (when included). * significant at 10%; ** significant at 5%; *** significant at 1%. 130 6. DISCUSSION AND CONCLUSION This paper investigated relationship between the temporal and spatial exposure to the 1992- 1998 armed conflict in Tajikistan and two demographic processes: female age at first marriage and age at first birth. It also examined the effect of sex ratio on the marriage market for the pre- and post-war cohorts of women. To summarize my main findings, I find substantial, highly significant and robust negative impact of temporal and regional exposure to armed conflict on entry into their first marriages by females in Tajikistan. Women born in 1975-1983, who also lived in the conflict affected areas were about 30 percent less likely to enter marriage than women of the same age from the lesser affected regions. Those results are robust to inclusion of regional dummies, and raion-level characteristics, such as employment level, crime rate and number of doctors per 1,000 residents. The estimated delay in marriage for younger cohorts in the conflict affected regions may suggest that households affected by civil war and economic crisis decided to conserve their scarce resources and focus on investing in areas other than marriages and celebrations. It is possible that high costs of getting married induced families to postpone marriages of their children as the traditional aspirations of what constitutes a well-provided for marriage were well above their means. Assuming that exposure to armed conflict measures the economic hardship of the families, the estimated effects of war on the marriage market for women in Tajikistan are similar to those in other countries affected by economic crises, for example, see Palloni, Hill and Aguirre (1996). 131 The effect of the sex ratio of men to women, a variable that was shown to have a significant impact on the marriage market for women in India, U.S.A., China and Malaysia, does not appear to have a significant effect on the demographic outcomes of interest in this paper. Rao (1993) who finds a significant effect of sex ratio on the dowry sizes in India also controls for the caste of the bride and groom in his analysis. Unfortunately, information on individual’s social status or regional affiliation, such as belonging to a particular avlod or clan is not available for Tajikistan. Several qualitative studies on Tajikistan suggest that many marriages occur between cousins, members of extended families and regionally based clan networks (see sections 2.3 and 4.1 for more detailed discussion). Such regional groups are often defined historically and the available data do not allow us to construct the appropriate measures of such marriage pools. While sex ratios did not have a significant effect on the rate of entry into first marriage by women, the sex ratios had a significant negative effect on the age differences between husbands and wives. Further, I find that virtually all noted differences in the decreased age at first birth among birth cohorts born in 1966-1986 are accounted for by the cohort differences in age at first marriage. This result is attributed to the uniformly short waiting period between marriage and first birth and relatively low non-marital childbearing in Tajikistan. Brien and Lillard (1994) find a similar result for Malaysia. 132 Since non-marital fertility is not common in Tajikistan, the delayed marriage is the main mechanism affecting fertility rates. A low effect of the exposure to mass violent conflict on age at first births for married women, suggests that the Tajik marriage and family institutions are resilient to change, even when a change is induced by a strong force such as armed conflict or an economic crisis. This resilience of the Tajik society has been earlier manifested by the continued adherence of Tajiks to religious practices and beliefs that were persecuted by the Soviets. Tajik society devised a way to maintain the practices by removing them from public into private (home) sphere. In this private sphere, women, who traditionally led more secluded and private lives, were “appointed” to serve as “guardians of the faith” (Tett 1994). Women became almost solely responsible for maintaining the family status of “good Muslims” and family engagement in relevant religious practices. Thus, I conclude by saying that the exposure to conflict had a negative impact on the rate of entry into first marriages by women who were of marriageable age during the war and virtually no effect on the interval between female age at first marriage and first birth. The sex ratio variable had a significant negative effect on difference in age between husbands and wives for women in the treatment group, but no effect of sex ratio was found for age at first marriage and first birth. Due to data limitations this paper left unanswered questions about the changes in dowry and bride-prices and relative characteristics of families of brides and grooms. Further research is needed to establish the long-term effects of economic, demographic and social changes on the distribution of bargaining power in households, female labor force participation and well-being of children. 133 An increased age at first marriage for women is more likely to be the good news, as children of more mature mothers usually have better chances of survival. However, the increased age at first marriage for women points towards a longer period of time they had to rely on the natal family resources. While the government is limited in its ability to influence family practices, import marriage age men or provide sizeable kalym and dowries for all wedding couples, the government should focus on persuading women to complete secondary education, invest into education for women beyond secondary school, and create job opportunities for young women. 74 These strategies may allow young women to accumulate the necessary job market skills, contribute income to their households and reduce their dependence on the income of their parents, husbands or brothers. 74 In September 2006, the Tajik government paid for weddings of about 250 disadvantaged couples, who were either orphans, financially disadvantaged or from broken homes. Each couple also received a refrigerator, kitchenware sets and washing machines as wedding gifts (United Press International 2006). 134 BIBLIOGRAPHY Abdullaev, Kamoludin. “Current Local Government Policy Situation in Tajikistan.” [n.d.] Akers, Donald S. "On Measuring the Marriage Squeeze." Demography, 1967, 4, pp. 907-24. Akiner, Shirin. "History and Politics In Central Asia: Change and Continuity." in Health care in Central Asia, edited by Martin McKee, Judith Healy, Jane Falkingham, and European Observatory on Health Care Systems. (Eds.). Buckingham: Open University Press. 2002. Allison, Paul D. “Discrete-Time Methods for the Analysis of Event Histories”, Social Methodology, 1982, Vol. 13, pp. 61-98. Angrist, Joshua. “How Do Sex Ratios Affect Marriage and Labor Markets? Evidence from America’s Second Generation.” Quarterly Journal of Economics, August 2002, pp. 997- 1038. Atkin, Muriel. "Tajikistan." in Kazakhstan, Kyrgyzstan, Tajikistan, Turkmenistan, and Uzbekistan: country studies, edited by Glenn E Curtis. Washington, D.C.: Federal Research Division, Library of Congress. 1997a. ----------------. "Thwarted democratization in Tajikistan." in Conflict, cleavage, and change in Central Asia and the Caucasus, edited by Karen Dawisha and Bruce Parrot. Cambridge, U.K.: Cambridge University Press. 1997b. Azam, Jean-Paul, and Anke Hoeffler. "Violence against civilians in civil wars: looting or terror?" Journal of Peace Research 39, 2002, (4):461-486. Becker, Gary. “A Theory of Marriage: Part I.” Journal of Political Economy, 1973, Vol. 81 (4), pp. 813-846. Becker, Gary. A Treatise on the Family. Cambridge, MA: Harvard University Press. 1981. Becker, Gary Stanley. Human capital: a theoretical and empirical analysis, with special reference to education. New York: National Bureau of Economic Research: distributed by Columbia University Press. 1975. Blattman, Christopher. "The Consequences of Child Soldiering." NEUDC 2006 Conference Presentation. 2006. Bonners, Raymond. "Tajik Civil War Fades, but the Brutality Goes On," New York Times. New York, Nov. 26, 1993, A3. Brien, Michael J.; Lee A. Lillard. “Education, Marriage and First Conception in Malaysia”, Journal of Human Resources, Vol. 29, No. 4 (Autumn, 1994), pp. 1167-1204. 135 Brzezinski, Zbigniew, and Paige Sullivan. Russia and the Commonwealth of Independent States: documents, data, and analysis. Armonk, N.Y.: M.E. Sharpe. 1997. Burnham, Gilbert, Riyadh Lafta, Shannon Doocy, Les Roberts. . "Mortality after the 2003 Invasion of Iraq: A Cross-Sectional Cluster Sample Survey " Lancet, 2006, 368(9545), pp. 1421-29. Bushkov, V.I. and D.V. Mikulskii. Anatomiia Grazhdanskoi Voiny V Tadzhikistane: Etno- Sotsial'nye Protsessy I Politicheskaia Bor'ba, 1992-1996. [The Anatomy of the Civil War in Tajikistan: Ethnic and Social Processes and Political Struggle, 1992-1996]. Moscow: Institut etnologii i antropologii im. Miklukho-Maklaia RAN [Institute of ethnic studies and anthropology named after Mikluho-Maklai. Russian Academy of Sciences]. 1997. Caldwell, John C.; Reddy, Palli Hanumantha and Caldwell, Pat. "The Causes of Marriage Change in South India." Population Studies, 1983, 37(3): 343-361. Caldwell, John C.; P.H. Reddy; Pat Caldwell. "Periodic High Risk as a Cause of Fertility Decline in a Changing Rural Environment: Survival Strategies in the 1980-1983 South Indian Drought." Economic Development and Cultural Change, 1986, 34(4), pp. 677-701. Callan, Patrick M. Copying with Recession: Public Policy, Economic Downturns and Higher Education. The National Center for Public Policy and Higher Education. 2002. Cameron, A. Colin, and Pravin K. Trivedi. Microeconometrics: Methods and Applications. Cambridge: Cambridge University Press, 2005. Capisani, Giampaolo R. The handbook of Central Asia: a comprehensive survey of the new republics. London; New York: I.B. Tauris. 2000. Central Intelligence Agency. The World Factbook: Tajikistan. (http://www.cia.gov/cia/publications/factbook/index.html Accessed: March 20th, 2005.) Chiappori, Pierre-Andre; Bernard Fortin and Guy Lacroix. "Marriage Market, Divorce Legislation, and Household Labor Supply." Journal of Political Economy, 2002, Vol. 110(11). Cleves, Mario, William Gould, and Roberto Gutierrez. An Introduction to Survival Analysis Using Stata. College Station, Texas: Stata Press, 2004. Collier, Paul, Lance Elliot, Håvard Hegre, Anke Hoeffler, Marta Reynal-Querol, and Nicholas Sambanis. Breaking the conflict trap: civil war and development policy. Washington, D.C., New York: World Bank; Oxford University Press. 2003. Collier, Paul, and Anke Hoeffler. "Aid, policy and growth in post-conflict societies." European Economic Review, 2004, 48:1125-45. Collins, Kathleen. "The logic of clan politics: evidence from the Central Asian trajectories." World Politics, 2004, 56 (2):224. 136 Das Gupta, Monica and Li Shuzhuo. Gender Bias in China, the Republic of Korea, and India 1920-90: Effects of War, Famine, and Fertility Decline. SSRN, 1999. Davis, Kingsley; Judith Blake. “Social Structure and Fertility: An Analytic Framework”, Economic Development and Cultural Change, 1956, 4(3), pp. 211-235. Department of Economic and Social Affairs (DESA), Population Division. "World Population Prospects: The 2002 Revision. Volume I: Comprehensive Tables," New York: United Nations, 2003. Djalili, Mohammad Reza, Frédéric Grare and Shirin Akiner. Tajikistan: the trials of independence. New York: St. Martin's Press. 1997 Duflo, Esther. "Schooling and labor market consequences of school construction in Indonesia: evidence from an unusual policy experiment." American Economic Review, 2001, 91, pp. 795-813. Dunckle, Kristin L., Rachel K. Jewkes, Hather C. Brown, Glenda E. Gray, James A. McIntryre, Sioban D. Harlow. "Transactional Sex among Women in Soweto, South Africa: Prevalence, Risk Factors and Association with HIV Infection." Social Science and Medicine, 2004, 59(8), pp. 1581-92. Falkingham, Jane. Women and gender relations in Tajikistan. [Manila]: Asian Development Bank Programs Dept. East and Office of Environment and Social Development. 2000. Fridman, Leonid A. "Economic crisis as a factor of building up socio-political and ethnonational tensions in the countries of Central Asia and Transcaucasia." in Central Asia and Transcaucasia: ethnicity and conflict, edited by Vitalii V. Naumkin. Westport, Conn.: Greenwood Press. 1994. Evans, David K., and Edward Miguel. "Orphans and Schooling in Africa: A Longitudinal Analysis." Demography, 2007, 44 (1), pp: 35-57. Ghobarah, H. A., P. Huth, and B Russett. "Civil wars kill and maim people long after shooting stops." American Political Science Review, 2003, 97. Gleditsch, N.P., P. Wallensteen, M. Eriksson, M. Sollenberg, and H. Strand. "Armed Conflict 1946-2001: A New Dataset." Journal of Peace Research, 2002, pp: 615-637. Glewwe, Paul. "Schools and skills in developing countries: education policies and socioeconomic outcomes." Journal of Economic Literature, 2002, 40:436-82. Global Security. “Tajikistan Civil War”. (http://www.globalsecurity.org/military/world/war/tajikistan.htm Accessed: May 10th, 2005). 2005. 137 Gomart, Elizabeth. "Between Civil War and Land Reform: Among the Poorest of the Poor in Tajikistan." in When things fall apart: qualitative studies of poverty in the former Soviet Union, edited by Nora Dudwick, Elizabeth Gomart, and Alexandre Marc. Washington, D.C.: World Bank. 2003. Goskomstat Tajik SSR [State Statistical Committee of the Tajik Soviet Socialist Republic]. Naselenie Tajik SSR. [Population of the Tajik SSR.] Dushanbe. 1990. Greenberg, Ilan. “After a Century, Public Polygamy Is Re-Emerging in Tajikistan.” New York Times. New York, November 13, 2006, Section A. p. 10. Greene, Margaret E,; Rao, Vijayendra. "The Marriage Squeeze and the Rise of Informal Marriage in Brazil." Social Biology, 1995, 42(1-2), pp. 65. Harris, Colette. Control and Subversion: Gender Relations in Tajikistan. London: Pluto Press, 2004. Harris, Colette. Muslim Youth: Tensions and Transitions in Tajikistan. Boulder: Westview Press, 2006. Hoeffler, Anke, and Martha Reynal-Querol. "Measuring the Cost of Conflict." Unpublished manuscript, Centre for the Study of African Economies, Oxford University, Oxford, U.K. 2003. Hoogeveen, J.G.M. (Hans), van der Klaauw, Bas and van Lomwel, A. Gijsbert C. "On the Timing of Marriage, Cattle and Weather Shocks," Tinbergen Institute Discussion Papers. June 2004. Human Rights Watch. Human Rights in Tajikistan. Helsinki, Finland: Helsinki Watch: a Division of Human Rights Watch. 1993. Humphreys, Macartan. "Economics and Violent Conflict." Unpublished manuscript, Program on Humanitarian Policy and Conflict Research, Harvard University, Cambridge, MA. 2003. Ichino, Andrea and Rudolf Winter-Ebmer. "The Long-Run Educational Cost of World War Ii." Journal of Labor Economics, 2004, 22(1), pp. 57-86. International Monetary Fund. Republic of Tajikistan: Recent Economic Developments. Washington, D.C.: International Monetary Fund. 1998. International Monetary Fund. Republic of Tajikistan: Recent Economic Developments and Statistical Appendix. Washington, D.C.: International Monetary Fund. 2005. Jacoby, Hanan G., and Emmanuel Skoufias. "Risk, financial markets, and human capital in a developing country." Review of Economic Studies, 1997, 64(3):311-35. Kalbfleisch, J. D. and R. L. Prentice, The Statistical Analysis of Failure Time Data, New York: Wiley. 1980. 138 Kamp, Marianne. The New Woman in Uzbekistan: Islam, Modernity, and Unveiling under Communism. Seattle: University of Washington Press, 2006. Keeley, Michael C. “An Analysis of the Age Pattern of First Marriage.” International Economic Review, 1979, 20 (2): 527-544 Kotlikoff, Laurence and A. Spivak. “The Family as an Incomplete Annuities Market.” Journal of Political Economy, 1981, 89 (2): 372 – 391. Kuz’menko, T. “E net spaseniya dushe…” Narodnaya Gazeta, Dushanbe: December 13, 1991, p. 6. Lindstrom, D. P. and B. Berhanu. “The Impact of War, Famine, and Economic Decline on Marital Fertility in Ethiopia.” Demography, 1999, 36: 247-261. Lopez, J. Humberto, and Quentin Wodon. "The economic impact of armed conflict in Rwanda." Journal of African Economies, 2005, 14: 586-602. Machel, Graca. Impact of Armed Conflict on Children. UNICEF. 1996. McGinn, Therese. “Reproductive Health of War-Affected Populations: What do we know?” International Family Planning Perspectives, 2000, 26(4): 174-180. McLean, Jennifer, and Thomas Greene. "Turmoil in Tajikistan: addressing the crisis of internal displacement." in The Forsaken People: Case Studies of the Internally Displaced, edited by Roberta Cohen and Francis Mading Deng. Washington, D.C.: Brookings Institution Press. 1998. Merrouche, Ouarda. "The Human Capital Cost of Landmine Contamination in Cambodia," HiCN Working Paper 25. 2006. Miguel, Edward; Shanker Satyanath and Ernest Sergenti. "Economic Shocks and Civil Conflict: An Instrumental Variables Approach." Journal of Political Economy, 2004. 112(4). Miguel, Edward, and Gerard Roland. "The long run impact of bombing Vietnam." National Bureau of Economic Research. Working Paper Series No. 11954:1-46. 2006. Mooney, Erin, and Colleen French. "Barriers and bridges: access to education for internally displaced children." Working Paper, Washington, D.C.: Brookings Institution. 2005. Monaghan, John and Peter Just. Social and Cultural Anthropology: A Very Short Introduction, Oxford, UK. Oxford University Press. 2000. Narodnaya Gazeta [People’s Newspaper], Oct. 15, 1992. Dushanbe, Tajikistan. - - - - - - - - - - - - - - - - - - - - - - - - - - - - , Oct. 16, 1992. Dushanbe, Tajikistan. - - - - - - - - - - - - - - - - - - - - - - - - - - - - , Nov. 13, 1992. Dushanbe, Tajikistan. 139 - - - - - - - - - - - - - - - - - - - - - - - - - - - - , Nov. 14, 1992. Dushanbe, Tajikistan. - - - - - - - - - - - - - - - - - - - - - - - - - - - - , Jan. 23, 1993. Dushanbe, Tajikistan. Newth, J.A. "The Soviet Population: Wartime Loses and the Postwar Recovery." Soviet Studies, 1964, 15(3), pp. 345-51. Nobles, Jenna and Alison Buttenheim. "Marriage in Periods of Crisis: Evidence from Indonesia," California Center for Population Research On-Line Working Paper Series. Los Angeles: University of California, Los Angeles, 2006. Nourzhanov, Kirill. "Saviours of the nation or robber barons? Warlord politics in Tajikistan." Central Asia Survey, 2005, 24 (2):109-130. Nugent, J.B., (1985). “The Old Age Security Motive for Fertility,” Population and Development Review, 1985, 11:75-98. Olimova, Saodat and Igor Bosc. "Labour Migration from Tajikistan." International Organization for Migration in Cooperation with the Sharq Scientific Research Center. 2003. Paci, Pierella. “Gender in Transition.” World Bank. May 2002. (Accessed: June 28, 2007). http://lnweb18.worldbank.org/ECA/eca.nsf/General/ Palloni, A., Hill, K., and Aguirre, G. P. . "Economic Swings and Demographic Changes in the History of Latin America" Population Studies, 1996, 50: 105-32. Panarin, Sergei A. "Political Dynamics of the "New" East." in Central Asia and Transcaucasia: ethnicity and conflict, edited by Vitalii V. Naumkin. Westport, Conn.: Greenwood Press. 1994. Porter, Maria. “The Effects of Demographic Change in China on Marriage Markets and Household Bargaining” NEUDC Conference Presentation. 2006. Rao, Vijayendra. "The Rising Price of Husbands: A Hedonic Analysis of Dowry Increases in Rural India." Journal of Political Economy, 1993, 101(4). Rashidov, Abdulbashir. Obrazovanie dlya Vseh: Otsenka na 2000 god po Respublike Tajikistan. [Education for all: evaluation as of 2000 for the Republic of Tajikistan.] [Dushanbe, Tajikistan]: n.p. [c.2000]. Roberts, Les; Riyadh Lafta; Richard Garfield; Jamal Khudhari; Gilbert Burnham. "Mortality before and after the 2003 Invasion of Iraq: Cluster Sample Survey." Lancet, 2004, 364: 1857-64. Rosenzweig, Mark, and T. Paul Schultz. "Market opportunities, genetic endowments and intra-family resource distribution: child survival in rural India." Journal of Development Economics, 1982, 72 (4):808-812. 140 Rosenzweig, Mark R. and Oded Stark. "Consumption Smoothing, Migration and Marriage: Evidence from rural India." Harvard University. Migration and Development Program. Discussion Paper, No. 32, pp. 1-27. 1987. Roy, Olivier. The New Central Asia: the Creation of Nations. New York, NY: New York University Press. 2000. Rukumnuaykit, Pungpond. “Economic Crisis and Demographic Outcomes: Evidence from Indonesia.” Ph.D. Dissertation. Michigan State University. 2003. Schoen, Robert. "Measuring the Tightness of a Marriage Squeeze." Demography, 1983, 20, pp. 61-78. Semein’i’i kodeks Respubliki Tajkistan. [Family Law of the Republic of Tajikistan]. Dushanbe, November 13, 1998. # 683. State Statistical Committee of Tajikistan. Regioni Respubliki Tajikistan 2001. [Regions of Republic of Tajikistan 2001.] Dushanbe: State Statistical Committee of Tajikistan. 2001. State Statistical Committee of Republic of Tajikistan. "Vozrastnoi Sostav Naseleniya Respubliki Tajikistan: Po Dannum Vseobshei Perepisi Naseleniya 2000 Goda. [Age Structure of the Population of the Republic of Tajikistan: According to the 2000 Census Data.] " Edited by O. A. Djabbarov, Dushanbe: State Statistical Committee of Republic of Tajikistan, 2002. - - - - - - - - - - - - - - - - - - - - - - - - - - - - . Regioni Respubliki Tajikistan 2003. [Regions of Republic of Tajikistan 2003.] Dushanbe: State Statistical Committee of Tajikistan. 2004. - - - - - - - - - - - - - - - - - - - - - - - - - - - - . Obrazovanie v Respublike Tajikistan. [Education in Republic of Tajikistan] Dushanbe: State Statistical Committee of Tajikistan. 2006. Stewart, Frances. "Horizontal Inequalities: A Neglected Dimension of Development", Queen Elizabeth House, Working Paper Series No. 81. 2002. Stewart, Frances, and Valpy Fitzgerald. War and underdevelopment. Oxford; New York: Oxford University Press. 2001. Stewart, Frances, Cindy Huang, and Michael Wang. "Internal wars in developing countries: an empirical overview of economic and social consequences." in War and underdevelopment, edited by Frances Stewart, and Valpy Fitzgerald. Oxford; New York: Oxford University Press. 2001. Strauss, John, and Duncan Thomas. "Human resources: empirical modeling of household and family decisions." in Handbook of Development Economics, Volume 3A, edited by J. Berhman and T.N. Srinivasan. Amsterdam: North Holland. 1995. 141 Tett, Gillian. “Guardians of the faith: gender and religion in the (ex)Soviet Tajik village”, in Muslim women’s choices: religious belief and social reality. Edited by Camilla Fawzi El- Solh and Judy Mabro. (Cross-Cultural Perspectives on Women, Vol. 12), Oxford: Berg Publishers, 1994. Thomas, Duncan, Kathleen Beegle, Elizabeth Frankenberg, Bondan Sikoki, John Strauss and Graciela Teruel. "Education in a crisis." Journal of Development Economics, 2004, 74 (1):53-85. 2004. TransMONEE 2007 Database, UNICEF IRC: Florence, Italy, 2007. Turner, R. “Tajiks Have the Highest Fertility Rates in the Newly Independent Central Asia”, Family Planning Perspectives, 1993, 25 (3). Turner, Stuart, Sahika Yuksel, and Derrick Silove. "Survivors of Mass Violence and Torture." in Trauma Interventions in War and Peace: Prevention, Practice, and Policy, edited by Bonnie L. Green. Hingham, MA, USA: Kluwer Academic Publishers. 2003. United Nations. Map of United Nations Mission of Observers in Tajikistan. (http://www.un.org/Depts/DPKO/Missions/unmot/Unmot.htm. Accessed: March 30, 2005). 2005. UNDP. Human Development Report. New York: Human Development Report Office, United Nations Development Programme. 2003. UNDP. “Central Asia Development Report.” Bratislava, Slovak Republic. UNDP Regional Bureau for Europe and the Commonwealth of Independent States. 2005. UNICEF, Innocenti Research Center. "Transmonee 2005," Statistical Tables. UNICEF Regional Office for Central and Eastern Europe and the Commonwealth of Independent States, 2005. UNICEF, Innocenti Research Center. "Transmonee 2007," Statistical Tables. UNICEF Regional Office for Central and Eastern Europe and the Commonwealth of Independent States, 2007. UNICEF. “The Status of Women and Children: Tajikistan, 2000. Multiple Cluster Indicator Survey”. Regional Office for Central and Eastern Europe and the Commonwealth of Independent States. 2001. United Press International. “Tajikistan Stages Mass Wedding for Poor.” News Daily [United Press International], September 4, 2006. http://www.newsdaily.com/Quirks/UPI-1- 20060904-16273100-bc-tajikistan-weddings.xml (Accessed: June 14, 2007.) U.S. Department of State. Tajikistan Human Rights Practices, 1993. 1994. (http://dosfan.lib.uic.edu/ERC/democracy/1993_hrp_report/93hrp_report_eur/Tajikistan.htm l (Accessed: February 15, 2007.) ------------------------------. "Tajikistan Report on Human Rights Practices for 1996," The Bureau of Democracy, Human Rights, and Labor, 1996. 142 ------------------------------. "Tajikistan Report on Human Rights Practices for 1997," The Bureau of Democracy, Human Rights, and Labor. 1997. Uppsala Universitet, Department of Peace and Conflict Research. "Conflict Database: Tajikistan," (http://www.pcr.uu.se/ Accessed: May 20th 2005). 2005. Usmanova, Gulchehra. “Ah, svad’ba pela i plyasala…” [A wedding was singling and dancing], Vechernii Dushanbe [Evening Dushanbe], N. 36(459) June 6, 2006. pp. 8-9. Vechenii Dushanbe. [Evening Dushanbe]. [Dushanbe, Tajikistan] (newspaper) 1991-1999 years. Verwimp, Philip and Jan van Bavel. “Child Survival and Fertility of Refugees in Rwanda.” European Journal of Population, 2005, 21: 271-290 Walker, Edward. W. “Ethnic War, Holy War, War O. War: Does The Adjective Matter In Explaining Collective Political Violence?” Berkeley Program in Soviet and Post-Soviet Studies Working Paper Series. (Accessed on February 15, 2007.) http://socrates.berkeley.edu/~bsp/publications/2006_01-walk.pdf. 2006. Walsh, Brendan M.. "Trends in Age at Marriage in Postwar Ireland." Demography, 1972, 9(2), pp. 187-2002. Weatherford, Jack. “Blood on the Steppes: Ethnicity, Power, and Conflict” in Conformity and Conflict: Readings in Cultural Anthropology. James Spadley and David W. McCurdy, editors. 10 th Edition. Needham Heights, MA: Allyn and Bacon. 2000. Winship, Christopher. “Heterogeneity and Interdependence: A Test Using Survival Models”, Sociological Methodology, 1986, Vol. 16, pp. 250-282. World Health Organization. "Violence against Women: Report on the 1999 Who Pilot Survey in Tajikistan," Copenhagen: WHO Regional Office for Europe, 2000. World Bank. Tajikistan Living Standards Measurement Survey 1999. (www.worldbank.org/lsms). 1999. ---------------. Tajikistan Living Standards Measurement Survey 2003. (www.worldbank.org/lsms). 2003. ---------------. Republic of Tajikistan: Poverty Assessment Update. Report No.30853. Washington, D.C.: World Bank. 2005. ---------------. World Development Report 2007: Development and the Next Generation. Washington, D.C.: The International Bank for Reconstruction and Development/ World Bank. 2006. World Food Programme. “World Hunger - Tajikistan.” 2006. 143 World Health Organization. "Violence against Women: Report on the 1999 Who Pilot Survey in Tajikistan," Copenhagen: WHO Regional Office for Europe, 2000. Yule, William, Rune Stuvland, Florence K. Baingana, and Patrick Smith. "Children in Armed Conflict." in Trauma Interventions in War and Peace: Prevention, Practice, and Policy, edited by Bonnie L. Green. Hingham, MA, USA: Kluwer Academic Publishers. 2003. 144 APPENDICES Appendix A - Measures of conflict exposure From 1992 to 1998, Tajikistan was embroiled in one of the most violent and long armed conflicts on the territory of the former Soviet Union. The most violent bout of war ravaged the southern regions of the country in 1992-1993. Conflict facilitated access to prominence and power to some of local warlords who supported the government forces. This created militia groups who benefited from the war tremendously and were determined to keep the conflict going. Some regions of Tajikistan were affected significantly as the conflict had an ethnic character where regionally based clans supported either government forces or opposition (McLean and Greene 1998, Capisani 2000, Nourzhanov 2005). In order to identify regions and communities that suffered the most during the war, I use three variables. Those variables allow me to compare the extent of conflict exposure across households and communities. The first measure is based on the households' reports of damage to their dwellings (henceforth, household damage dwelling or HDD) in the 1999 TLSS. Approximately 6.8 percent of the 2000 households surveyed in 1999 reported that their dwelling suffered damage during the armed conflict. The person interviewed was the head of household or a person who could give information on all household members. This damage dwelling measure is available only for the 1999 TLSS data. The question is worded in the following way: - “WAS YOUR DWELLING DAMAGED DURING THE RECENT CIVIL UNREST?” (Question 12 from section 2 “Dwelling”, Household Questionnaire, Tajikistan Survey of Living Standards 1999). My second conflict variable reflects damage in a community (henceforth, community damage dwelling or CDD). It is a dummy variable that is assigned a value of one for all households in a primary sampling unit if one or more households from this unit reported that their dwelling was damaged during the war. For example, if 16 households were selected in a primary sampling unit and one of them reported damage dwelling during the war, then I assign a value of one to each household in this primary sampling unit in the 1999 TLSS data. The 2003 TLSS data does not have a similar question on the damage household dwelling. So, in order to compare communities from the 1999 TLSS data to the 2003 TLSS, I match raions surveyed in 1999 to those in 2003 data. For that, I calculate the CDD measure for each raion, which is a collection of several primary sampling units, in the 1999 survey. Then I match raion names between the 1999 and 2003 surveys and assign to each raion in the 2003 dataset a corresponding value of CDD from the 1999 data. 75 75 It was not possible to match individual primary sampling units between 1999 and 2003 datasets, as the 2003 survey data does not contain the names of the primary sampling units selected for the survey. 145 The above two measures of the conflict have several drawbacks. First, there is a chance that households migrated during and after the war and their current community of residence is different from the one where their dwelling was damaged. Second, households or communities that suffered the most during the war may not have been selected for the TLSS interviews. Households could have relocated during the war and communities could have been destroyed to the ground, or still occupied by militants. For example, the 1999 TLSS research team was not able to secure permission from the United Tajik Opposition to survey households in some parts of the Gharm area in the RRS region. Third, the community damage dwelling data is available for all primary sampling units in the 1999 TLSS data, as compared to only 92 percent of observations with community damage dwelling data available for 2003. Also the community damage dwelling data is available only at the raion (district) level for the 2003 data and at the primary sampling unit level data for the 1999 sample. Thus, I use a third, also dummy, variable that is based on the reports of conflict activity (henceforth, RCA) and/or atrocities against the civilian population in Tajikistan during the civil war. I used multiple data sources to cross-check the type and degree of damage to villages and towns. The report of conflict activity (RCA) variable is used to identify residents of communities that sustained a significant damage during the war. Often, residents were terrorized and their houses looted by armed forces stationed in or passing through those communities. 76 Two good detailed chronological sources on the events of the Tajik civil war are Brzezinski and Sullivan (1997) and Djalili, Grare and Akiner (1997). Communities included in the RCA measure I constructed a geographic mapping of the civil conflict using references to fighting in literature. I collected names raions, villages, towns and cities where fighting or other conflict related event occurred from two central newspapers published in Dushanbe, Tajikistan between 1991 and 1999. I also added to this list names of the communities mentioned as associated with conflict from the reports by human rights organizations, international not-for profit organizations that involved in the monitoring of armed conflicts around the world and publications on the Tajik civil war. 77 To match villages, towns and cities to raion data I used the detailed map of the deployment of the United Nations Mission of Observers in Tajikistan (UNMOT). 78 Further, I matched the list of community names that were identified to be highly exposed to the conflict to the list of primary sampling units (PSU) that were surveyed for the TLSS 1999 and raions in the 2003 TLSSs. The list of communities is available on request. 76 The possible disadvantage of using the RCA measure of conflict activity as a primary conflict variable is that it allows us to use only raion/district level data in 2003. 77 I did not include in the RCA measure the attack by Colonel Makhmud Khudoberdiyev in Leninabad region in November of 1998. This attack was an individual conflict event reported for Leninabad (Sugd) region over the 1991-1998 period. The attack lasted for less than a month as compared to the minimum of several months of conflict exposure in the RRS, Dushanbe and Khatlon regions. 78 Available at http://www.un.org/Depts/DPKO/Missions/unmot/Unmot.htm. (Accessed 03/30/2005.) 146 Appendix B – Schooling variables Questions and response categories for enrollment variables and number of grades completed 1. ENROLLED - There is a difference in the wording of this question between the 1999 and 2003 surveys. In the 1999 TLSS (collected in May of 1999) the question is worded as: - "Is [NAME] currently studying?" In the 2003 TLSS (collected in June-July 2003) the question is worded as - "Did you enroll in school last academic year (2002-2003)?" In this paper, both questions are coded as 1 for "yes" and as 0 for "no" response. The questions are comparable because both are used as a filter. Questions regarding school quality, school related expenses and frequency of attendance are asked only of those individuals who indicated that they are currently enrolled or studying, while questions regarding school non-attendance are asked only to those who responded that they are not enrolled/attending school. 2. COMPLETION OF 9 GRADES - coded as 1 if a person responds that he or she completed 9 grades and above, and zero otherwise. 3. GRADES COMPLETED - number of school grades completed. 147 Appendix C – TLSS data and sample construction: analysis of education Description and Comparison of the 1999 and 2003 Tajik Living Standards Surveys The 1999 and 2003 surveys include 3 modules: a household questionnaire, a community level questionnaire and a female questionnaire. Both, 1999 and 2003, samples were stratified by oblast (region) and rural and urban areas. While the sample designs were very similar, there are several differences between the two surveys (World Bank 2005) 79 . First, the sample size in the 2003 survey is much larger than the sample surveyed in 1999. The 1999 sample size is 2,000 households (14,142 individuals) as compared to 4,160 households (26,141 individuals) surveyed in 2003. Second, the 2003 survey over sampled households from Dushanbe by 40%, rural GBAO by 300% and by 600% in urban areas of GBAO, while the 1999 TLSS sample was designed to be nationally representative. The 2003 survey is representative at regional and urban/rural levels. Third, the 2003 TLSS sample was based on the 2000 Census shares of each strata, while the 1999 TLSS stratification was based on the "best estimates" from community registers as the survey was conducted prior to the Census. More information on the surveys and the sample design can be obtained from the 1999 TLSS survey documents available online at www.worldbank.org/lsms/ and "Republic of Tajikistan: Poverty Assessment Update" (World Bank 2005). Strategy 1: enrollment sample construction To analyze the impact of the conflict on school enrollment, I identify individuals who were in age groups eligible for enrollment in the academic years corresponding to the information in the 1999 and 2003 surveys. For this, I use a rule established by the Tajik law on education. First, I identify the youngest children who were eligible to be enrolled in school during the academic year under consideration. 80 Since by law only children age seven and above are eligible to be enrolled in school, I include in my analytical samples children who reached age 7 by the end of September 1998 (for 1999 data) or 2002 (for 2003 data). 81 Thus, I avoid i) erroneously overstating non-enrollment rates among six- and seven- year olds; and ii) mitigate the selection bias problem as young children who enroll in school before reaching the official enrollment age may have a particularly high ability or very motivated parents. On the upper age bound I limit my regression analysis to 15 year olds in the 1999 data and to 16 year olds in the 2003 dataset. Coding of variables and summary statistics for the 1999 and 2003 samples are reported in Tables C.1 and C.2. Comparison of sample means of the schooling outcome variables and schooling covariates by conflict exposure is provided in Tables C.3 - C.5. 79 Those differences will not affect the results presented in this paper. 80 Since the 1999 TLSS was collected in May, all questions regarding student enrollment in grade school refer to the 1998-1999 academic year. In the 2003 survey, all questions refer to the 2002-2003 academic year. 81 September 1st is the official start of a school year in Tajikistan and other former Soviet Union countries. School year ends around May 25-27. 148 Table C.1 - Description of Key Variables in the Enrollment Dataset Variable Description Conflict Exposure Variables HDD Indicator (=1 if household reported damage dwelling) CDD Indicator (=1 if at least one household in the primary sampling unit reported damage dwelling) RCA Indicator (=1 if there are records of high conflict activity in the community) Educational Outcomes Indicator of School Enrollment Indicator (=1 if a respondent was enrolled/ studying in the academic year 1998-1999 (2002-2003)) Total years of education completed (years) Total number of years of education completed (in years) Currently enrolled students only Missed school for 2 weeks Indicator (=1 if a respondent missed school for more than two weeks in the academic year 1998-1999) Missed school for 4 weeks Indicator (=1 if a respondent missed school for more than four weeks in the academic year 2002-2003) Hours missed school last week Hours absent from school per week outside of vacation 1 Time traveled to school Time traveled to school (one way, fraction of an hour) Individual Variables Age Respondent's age (in years) Year of birth Year of birth Female Indicator (= 1 if female) Household Characteristics Parent's education (years) Mother Mother's education (years) Father Father's education (years) Parent's age (years) Mother Mother's age (years) Father Father's age (years) Indicator variables controlling for missing information for Mother _miss Mother's information is not available Father _miss Father's information is not available N adults ages 17-65 Females Number of females ages 17-65 in a household Males Number of males ages 17-65 in a household Household size Household size Per capita household expenditure (rubles) Per capita household expenditure (rubles) for 1999 data Per capita household expenditure (somoni) Per capita household expenditure (somoni) for 2003 data HH owns > 0.1 hectare of land (=1) Household owns more than 0.1 hectares of land 149 Table C.1 (continued) - Description of Key Variables in the Enrollment Dataset Variable Description Community Characteristics Distance to school Distance to school (km) Rural Resident of rural area Regions: Dushanbe Resident of Dushanbe RRS Resident of Raions of Republican Subordination Khatlon Resident of Khatlon GBAO Resident of GBAO Sugd Resident of Sugd 150 Table C.2 - Descriptive Statistics of Key Variables in the Enrollment Dataset. Children, ages 7-15, TLSS 1999 data Panel A Variable # of Obs. Sample Mean Standard Deviation Min Max Conflict Exposure Variables HDD 3285 0.08 (0.27) 0 1 CDD 3285 0.43 (0.49) 0 1 RCA 3285 0.51 (0.50) 0 1 Educational Outcomes Indicator of School Enrollment 3285 0.89 (0.32) 0 1 Total years of education completed (years) 3285 4.37 (2.59) 0 11 Currently enrolled students only Missed school for 2 weeks 2915 0.37 (0.48) Hours missed school last week 2910 2.49 6.46 0 42 Time traveled to school 2915 0.26 (0.17) 0 1.5 Missed school for 4 weeks Individual Variables Age 3285 11.32 (2.38) 7 15 Year of birth 3284 87.15 (2.41) 82 93 Female 3285 0.51 (0.50) Household Characteristics Parent's education (years) Mother 3285 9.96 (2.00) 0 15 Father 3285 11.63 (2.31) 0 18 Parent's age (years) Mother 3285 37.89 (4.88) 17 54 Father 3285 41.88 (6.20) 24 74 Indicator variables controlling for missing information for Mother 3285 0.15 (0.35) Father 3285 0.25 (0.43) N adults ages 17-65 Females 3285 1.68 (1.16) 0 7 Males 3285 1.76 (1.07) 0 7 Household size 3285 8.20 (3.17) 2 27 Per capita household expenditure (rubles) 3285 13709 (8841) 903 128398 Per capita household expenditure (somoni) na Per capita household expenditure (somoni in rubles) 2 na HH owns > 0.1 hectare of land (mean for 1999) 3285 0.87 (0.34) HH owns > mean amount of land (mean for 2003) na 151 Table C.2 (continued) - Descriptive Statistics for Key Variables in the Enrollment Dataset. Children, ages 7-15, TLSS 1999 data Panel A Variable # of Obs. Sample Mean Standard Deviation Min Max Community Characteristics Distance to school 3285 0.76 (1.00) 0 30 Distance to school (0 to 10 km) 3 Rural 3285 0.79 (0.40) Regions: Dushanbe 3285 0.06 (0.23) RRS 3285 0.24 (0.43) Khatlon 3285 0.43 (0.49) GBAO 3285 0.04 (0.20) Sugd 3285 0.24 (0.42) 152 Table C.2 (continued) - Descriptive Statistics of Key Variables in the Enrollment Dataset. Children, ages 8-16, TLSS 2003 data Panel B Variable # of Obs. Sample Mean Standard Deviation Min Max Conflict Exposure Variables HDD na CDD 5562 0.62 (0.48) 0 1 RCA 6055 0.51 (0.50) 0 1 Educational Outcomes Indicator of School Enrollment 6055 0.93 (0.26) 0 1 Total years of education completed (years) 6047 5.19 (2.63) 0 11 Currently enrolled students only Missed school for 2 weeks na Hours missed school last week na Time traveled to school 5472 0.25 (0.18) 0 2.67 Missed school for 4 weeks 5605 0.05 (0.23) Individual Variables Age 6055 12.03 2.53 8 16 Year of birth 6055 90.97 2.53 87 95 Female 6055 0.49 0.50 0 1 Household Characteristics Parent's education (years) Mother 6055 10.12 (1.80) 0 21 Father 6055 11.62 (2.11) 0 27 Parent's age (years) Mother 6055 37.75 (4.68) 17 49 Father 6055 41.24 (4.96) 20 73 Indicator variables controlling for missing information for Mother 6055 0.21 (0.41) Father 6055 0.35 (0.48) N adults ages 17-65 Females 6055 1.80 (1.09) 0 8 Males 6055 1.73 (1.16) 0 7 Household size 6055 7.69 (3.17) 1 31 Per capita household expenditure (rubles) Per capita household expenditure (somoni) 6055 45.7 (30.11) 1 380 Per capita household expenditure (somoni in rubles) 2 6055 35131 (23131) 988 291644 HH owns > 0.1 hectare of land (mean for 1999) 6055 0.996 (0.06) HH owns > mean amount of land (mean for 2003) 6055 0.221 (0.42) 153 Table C.2 (continued) - Descriptive Statistics of Key Variables in the Enrollment Dataset. Children, ages 8-16, TLSS 2003 data Panel B Variable # of Obs. Sample Mean Standard Deviation Min Max Community Characteristics Distance to school 6055 1.05 (6.47) 0 384 Distance to school (0 to 10 km) 3 6027 0.83 (0.90) 0 10 Rural 6055 0.72 (0.45) Regions: Dushanbe 6055 0.11 (0.32) RRS 6055 0.21 (0.41) Khatlon 6055 0.32 (0.47) GBAO 6055 0.10 (0.30) Sugd 6055 0.25 (0.44) Notes: (1) 5 observations with values higher than 45 hours per week were omitted. (2) Calculated using exchange rates from IMF (2003). (3) 8 observations with values higher than 10 km were omitted. Sample of children ages 7-15. Source: TLSS (1999), TLSS (2003). 154 Table C.3 - Covariates of schooling outcomes by household and community damage dwelling variables. Sample means Sample means HDD=0 HDD=1 Differen- ce CDD=0 CDD=1 Differen- ce Covariate (1) (2) (3) (4) (5) (6) Age 11.32 11.34 -0.02 11.39 11.28 0.11 (0.04) (0.15) (0.16) (0.07) (0.05) (0.09) Female 0.50 0.55 -0.05 0.51 0.50 0.00 (0.01) (0.03) (0.03) (0.01) (0.01) (0.02) Parent's education (years) Mother 10.02 9.25 0.76 ** * 10.21 11.56 -1.35 ** * (0.04) (0.16) (0.13) (0.05) (0.05) (0.08) Father 11.66 11.27 0.39 ** * 11.78 11.56 0.22 (0.04) (0.14) (0.15) (0.07) (0.05) (0.08) Parent's age (years) Mother 37.88 38.04 -0.16 38.00 37.84 0.16 (0.09) (0.29) (0.32) (0.15) (0.10) (0.18) Father 41.85 42.22 -0.37 41.78 41.93 -0.15 (0.11) (0.32) (0.41) (0.19) (0.13) (0.23) N adults ages 17-65 Females 1.76 1.83 -0.08 1.69 1.80 -0.11 (0.02) (0.08) (0.07) (0.03) (0.02) (0.04) Males 1.69 1.60 0.09 1.58 1.73 -0.15 ** * (0.02) (0.07) (0.08) (0.03) (0.03) (0.00) Household size 8.16 8.64 -0.48 ** 7.84 8.39 -0.55 ** * (0.06) (0.21) (0.21) (0.08) (0.07) (0.12) 13838 12132 1706 ** * 12256 14459 -2203 ** * Per capita household expenditure (rubles) (162) (501) (581) (197) (209) (323) 0.86 0.94 -0.07 ** * 0.90 0.85 0.05 ** * HH owns > 0.1 hectare of land (=1) (0.01) (0.02) (0.02) (0.01) (0.01) (0.01) Distance to school 0.77 0.72 0.05 ** * 0.73 0.78 -0.05 (0.00) (0.00) (0.00) (0.02) (0.02) (0.04) Rural 0.79 0.85 -0.06 ** 0.82 0.78 0.05 (0.01) (0.02) (0.03) (0.01) (0.01) (0.01) Notes: Sample of children, ages 7-15. Source: TLSS (1999). * significant at 10%; ** significant at 5%; *** significant at 1%. 155 Table C.4 - Comparison of Schooling Outcomes. Panel A - Sample of children, ages 7-15. TLSS 1999. Sample means Sample means HDD=0 HDD=1 Differenc e CDD=0 CDD=1 Differenc e Schooling variables (1) (2) (3) (4) (5) (6) Enrolled in school 0.89 0.80 0.09 ** * 0.91 0.87 0.04 ** * (0.01) (0.03) (0.02) (0.01) (0.01) (0.01) 4.41 3.84 0.57 ** * 4.52 4.29 0.22 ** Total years of education completed (years) (0.05) (0.16) (0.17) (0.08) (0.06) (0.10) Currently enrolled students only Missed school for 2 weeks 0.36 0.51 -0.15 ** * 0.35 0.39 -0.04 * (0.01) (0.04) (0.00) (0.01) (0.01) (0.02) 2.36 4.30 -1.94 ** * 2.40 2.61 -0.21 Hours missed school last week (0.01) (0.04) (0.00) (0.24) (0.18) (0.30) Panel B - Sample of children, ages 8-16. TLSS 2003. Sample means Sample means CDD=0 CDD=1 Differenc e RCA=0 RCA=1 Differenc e Schooling variables (1) (2) (3) (4) (5) (6) Enrolled in school 0.95 0.91 0.04 ** * 0.94 0.91 0.03 ** * (0.00) (0.00) (0.01) (0.00) (0.01) (0.01) 5.32 5.10 0.22 ** * 5.30 5.09 0.20 ** * Total years of education completed (years) (0.06) (0.04) (0.07) (0.05) (0.05) (0.07) Currently enrolled students only 0.06 0.05 0.01 0.05 0.06 -0.02 ** * Missed school for 4 weeks (0.01) (0.04) (0.01) (0.00) (0.00) (0.01) Note: Standard errors are in parenthesis. Group sample means by exposure to conflict variables. * significant at 10%; ** significant at 5%; *** significant at 1%. 156 Table C.5 - Comparison of Group Means by Exposure to Conflict Variables. Panel A: Sample Means and Differences for the sample of children, ages 7-15, TLSS 1999 Sample means Sample means HDD=0 HDD=1 Difference CDD=0 CDD=1 Difference Covariate (1) (2) (3) (4) (5) (6) Age 11.32 11.34 -0.02 11.39 11.28 0.11 (0.04) (0.15) (0.16) (0.07) (0.05) (0.09) Female 0.50 0.55 -0.05 0.51 0.50 0.00 (0.01) (0.03) (0.03) (0.01) (0.01) (0.02) Parent's education (years) Mother 10.02 9.25 0.76 *** 10.21 11.56 -1.35 *** (0.04) (0.16) (0.13) (0.05) (0.05) (0.08) Father 11.66 11.27 0.39 *** 11.78 11.56 0.22 (0.04) (0.14) (0.15) (0.07) (0.05) (0.08) Parent's age (years) Mother 37.88 38.04 -0.16 38.00 37.84 0.16 (0.09) (0.29) (0.32) (0.15) (0.10) (0.18) Father 41.85 42.22 -0.37 41.78 41.93 -0.15 (0.11) (0.32) (0.41) (0.19) (0.13) (0.23) N adults ages 17-65 Females 1.76 1.83 -0.08 1.69 1.80 -0.11 (0.02) (0.08) (0.07) (0.03) (0.02) (0.04) Males 1.69 1.60 0.09 1.58 1.73 -0.15 *** (0.02) (0.07) (0.08) (0.03) (0.03) (0.00) Household size 8.16 8.64 -0.48 ** 7.84 8.39 -0.55 *** (0.06) (0.21) (0.21) (0.08) (0.07) (0.12) Per capita household expenditure (rubles) 13838.39 12132.34 1706.05 *** 12255.81 14459.07 -2203.26 *** (161.57) (501.03) (581.04) (197.22) (208.80) (323.25) 157 Table C.5 (continued) - Comparison of Group Means by Exposure to Conflict Variables. Panel A: Sample Means and Differences for the sample of children, ages 7-15, TLSS 1999 Sample means Sample means HDD=0 HDD=1 Difference CDD=0 CDD=1 Difference Covariate (1) (2) (3) (4) (5) (6) Per capita household expenditure (somoni) na na na na na na HH owns > 0.1 hectare of land (=1) 0.86 0.94 -0.07 *** 0.90 0.85 0.05 *** (0.01) (0.02) (0.02) (0.01) (0.01) (0.01) Distance to school 0.77 0.72 0.05 *** 0.73 0.78 -0.05 (0.00) (0.00) (0.00) (0.02) (0.02) (0.04) Distance to school (<11 km) 1 Rural 0.79 0.85 -0.06 ** 0.82 0.78 0.05 (0.01) (0.02) (0.03) (0.01) (0.01) (0.01) 158 Table C.5 (continued) - Comparison of Group Means by Exposure to Conflict Variables. Panel B: Sample Means and Differences for the sample of children, ages 8-16, TLSS 2003 Sample means Sample means CDD=0 CDD=1 Difference RCA=0 RCA=1 Difference Covariate (1) (2) (3) (4) (5) (6) Age 12.08 11.99 0.09 12.07 11.99 0.08 (0.05) (0.04) (0.07) (0.05) (0.05) (0.06) Female 0.49 0.49 0.00 0.49 0.49 0.00 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Parent's education (years) Mother 10.31 9.99 0.32 *** 10.28 9.98 0.30 *** (0.04) (0.03) (0.05) (0.03) (0.03) (0.05) Father 11.57 11.58 -0.01 11.55 11.68 -0.13 ** (0.04) (0.04) (0.06) (0.04) (0.04) (0.08) Parent's age (years) Mother 37.80 37.74 0.07 37.70 37.79 -0.09 (0.09) (0.29) (0.13) (0.09) (0.08) (0.12) Father 41.15 41.32 -0.17 41.08 41.39 -0.31 ** (0.11) (0.08) (0.14) (0.09) (0.09) (0.13) N adults ages 17-65 Females 1.74 1.82 -0.08 *** 1.71 1.89 -0.19 *** (0.02) (0.02) (0.03) (0.02) (0.02) (0.03) Males 1.75 1.70 0.04 1.69 1.77 -0.08 ** (0.02) (0.02) (0.03) (0.02) (0.02) (0.03) Household size 7.44 7.82 -0.39 *** 7.23 8.13 -0.90 *** (0.06) (0.06) (0.09) (0.05) (0.06) (0.08) 159 Table C.5 (continued) - Comparison of Group Means by Exposure to Conflict Variables. Panel B: Sample Means and Differences for the sample of children, ages 8-16, TLSS 2003 Sample means Sample means CDD=0 CDD=1 Difference RCA=0 RCA=1 Difference Covariate (1) (2) (3) (4) (5) (6) Per capita household expenditure (somoni) 40.85 48.44 -7.59 *** 42.41 48.89 -6.47 *** (0.51) (0.57) (0.83) (0.49) (0.59) (0.77) HH owns > 0.1 hectare of land (=1) 0.22 0.21 0.01 0.20 0.25 -0.05 *** (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Distance to school 1.21 0.89 0.32 *** 1.09 1.02 0.07 (0.15) (0.03) (0.12) (0.14) (0.10) (0.17) Distance to school (<11 km) 1 0.92 0.78 0.13 *** 0.87 0.79 0.08 *** Rural 0.80 0.65 0.14 *** 0.79 0.65 0.14 *** (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Notes: Standard errors are in parenthesis. (1) 28 observations with values above 10 km are omitted. Sample of children: 1999 data - ages 7-15 (TLSS 1999), 2003 data - ages 8-16 (TLSS 2003). Source: TLSS (1999), TLSS (2003). * significant at 10%; ** significant at 5%; *** significant at 1%. Strategy 2: Completion of mandatory schooling In the difference in differences regression framework that is used to analyze the completion of mandatory nine grades of schooling in essay on the effect of armed conflict on education, I use the sample of adults, born between 1966 and 1986 or adults ages 17 and above from the 2003 TLSS. Table C.6 below contains definitions of variables and their sample means. 160 Table C.6 - Description of Key Variables in the Completion of Mandatory Schooling Dataset. Variable Description # of Obs. Sample Mean Standard Deviation Conflict Exposure Variables CDD Indicator equaling to one if at least one household in the primary sampling unit reported damage dwelling 6733 0.58 (0.49) RCA Indicator equaling to one if there are records of high conflict activity in the community 7401 0.46 (0.50) Educational Outcomes Indicator of Mandatory School Completion Indicator equaling to one if a respondent completed nine grades of education 7401 0.89 (0.31) Years of education completed (0-9) Years of education completed (between 0 and 9) 7401 8.73 (1.17) Indicator for school attendance Indicator equaling to one if a respondent ever attended school 7401 0.99 (0.11) Total years of education completed (years) Total number of years of education completed (in years) 7401 10.48 (2.28) Individual Variables Age Respondent's age (in years) 7401 24.9 (6.20) Year of birth Year of birth 7401 1978.1 (6.20) Born in 1966-1973 Indicator equaling to one if born in 1966-1973 2194 0.3 Born in 1976-1986 Indicator equaling to one if born in 1976-1986 5207 0.7 Moved since January of 1990 Indicator equaling to one if a respondent moved since January 1990 7400 0.00 (0.07) Female Indicator equaling to one if female 7401 0.52 (0.50) Notes: Sample: adults, born in 1966-1973 and 1976-1986. Source: TLSS (2003). 161 Appendix D – Basic statistical framework: analysis of duration data 82 In this section I present the basic statistical framework for the analysis of the duration data on which are based the estimates of the parameters of the hazard models. Let T > 0 be the N of years a woman “survived”, e.g. remained single, in years. Assume that T has some distribution over population. Let F (t; X) be a conditional c.d.f of T, where X is a vector of time-invariant covariates. (D.1) 0 ), ; ( ) ; ( ≥ ≥ = t X t T P X t F The survivor function is defined as in (1) above. (D.2) ) ; ( ) ( 1 ) ; ( X t T P X F X t S > = − = The probability of leaving the initial state is the time interval (t; t+m) is ) , | ( X t T m t T t P ≥ + ≤ ≤ for 0 > m The hazard function is defined as: (D.3) ) ; | ( ) ; ( X t T m t T t P X t ≥ + < ≤ = λ ) ; ( 1 ) ; ( ) ; ( ) ; ( ) ( X t F X t F X m t F X t T P m t T t P − − + = ≥ + < ≤ = If c.d.f is differentiable, then take the limit of the right-hand-side and divide the function by m, as m → 0. (D.4) ) ( 1 1 . ) ; ( ) ; ( lim ) ; ( 0 X F h X t F X m t F X t h − − + = → λ ) ; ( ) ; ( ) ; ( 1 ) ; ( X t S X t f X t F X t f = − = Then all probabilities can be computed using the above hazard function. For example, the probability of exit from time a to time b, where a<b can be computed using equation (D.5) below: (D.5) ∫ − − = ≥ < ≤ b a ds X s X a T b T a P ] ) ; ( exp[ 1 ) ; | ( λ In this paper I use discrete time hazard functions as I do not exactly at what point in time the event of interest has occurred. From the TLSS data we know that a marriage occurred in a particular year, same applies to the majority of first birth dates. 82 This section heavily draws on Cleves, Gould and Gutierrez (2004) and Cameron and Trivedi (2005). 162 The discrete time hazard function is the probability of transition at discrete time j t , j = 1, 2, … given survival to time j t . (D.6) ) ( ) ( ] | [ Pr − = ≥ = = j d j d j j j t S t f t T t T λ where superscript d stands for discrete, ) ( lim ) ( j d a t d a t S a S − → − = The discrete-time cumulative hazard function is then: (D.7) ∑ ≤ = Λ t t j j d j t | ) ( λ In this paper I use a combination of non-parametric and semi-parametric methods. Below I outline models used in this paper. Non-parametric The Kaplan-Meier (KM) cumulative survival and Nelson-Aalen hazard functions are used to estimate discrete yearly hazard rates. The KM estimate is a non-parametric estimate of the survivor function S(t) or the probability of surviving past time t. For a dataset with observed failure times, t 1 … t k , the KM estimate at any time can be expressed as (D.8) ^ ) (t S = ∏ ≤ − t t j j j j j n d n | where n j is the number of individuals at risk at time t j , and d t is the number of failures at time t j . The product is taken over all observed failure times less than or equal to t. The KM estimate provides us with the probability of survival beyond each time t. In the analysis of entry into marriage I set the minimum age of marrying at 11. Thus, subjects enter analysis at age 11 and exit at the age of first marriage. Some women are unmarried at the time of the survey. Thus the spells of being unmarried are right-censored at the date survey and timing of such censoring is assumed to be exogenous. In the analysis of reproductive decisions the age of entry is set as age of first marriage in the conditional hazard estimation, and age 14 in the unconditional analysis. Proportional hazards estimation (semi-parametric) In addition to non-parametric estimation, I use the Cox proportional hazards (PH) regression model to estimate how hazard (of being married or having first birth) changes over time. The PH model also measures effect of various covariates on the hazard rate. In the estimation of the Cox proportional hazard model, the rate of entry into marriage is set to co- vary with birth cohort, region of residence, cohort and region specific sex ratio, women’s education level and a set of regional characteristics. Including interactions of the regional conflict exposure with the cohort of birth for younger women allows me to estimate the combined regional and temporal effect of exposure to conflict on the marriage and reproductive behavior. 163 The proportional hazards model is defined below: 83 (D.9) ) , ( ) , ( ) , ( β φ α λ λ x t x t = where ) , ( 0 α λ t is an unspecified function of time. It is also called baseline hazard that is the same for all subjects in the analysis. ) , ( β φ x – is a function of x, a vector of individual specific covariates. The commonly used specification of φ is ) exp( ) , ( x x β β φ ′ = . The model is called proportional hazards model because the ratio of the hazard rates is for any two individuals at any point in time is constant over time. The timing of the event of interest is assumed to follow a failure-time process which incorporates various individual and regional characteristics and heterogeneity common to districts (raions). Proportional hazard model with heterogeneity component In the analysis of age at first marriage I report results of the PH models specified with shared frailty effects. 84 In the analysis of duration data shared frailty models are analogous to random effects specification. In such models, observations within the same group are assumed to be correlated because they share the same frailty. A frailty is a latent random component that enters multiplicatively on the hazard functions. In the PH model, the data are organized as j = 1, …, n groups with i = 1, …, n i subjects in group j. For the ith subject in the jth group, the hazard is then specified as follows: (D.10) ) exp( ) , ( ) , ( x w t x t j ji β α λ λ ′ = where j ω is the group level frailty, an unobserved random quantity with mean one and variance θ (theta) to be estimated from the data. If we let i i w v log = , then the log-hazard function can be expressed as (D.11) β ji j ji x v t h t h + + = ) ( ln ) ( ln 0 in (D.11) j v are log-frailties and, thus, they are similar to random effects in linear models. 83 This definition is based on Allison (1982). 84 The likelihood ratio tests showed that the estimated parameter θ that measures correlation within the same district in Tajikistan is significantly different from zero in the analysis of marriage data.
Abstract (if available)
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Essays on human capital accumulation -- health and education
PDF
Vertical integration and two-sided market pricing: evidence from the video game industry
PDF
Community forest management in Nepal: saving the forest at the expense of the poorest
PDF
Essays in the economics of education and conflict
PDF
Essays on family planning policies
PDF
Essays on development and health economics: social media and education policy
PDF
Market reforms, foreign direct investment and national identity: Non-national identity of Kazakhstan
PDF
Student, staff, and parent perceptions of the reasons for ethnic conflict between Armenian and Latino students
PDF
The political economy of Brazilian trade policy: domestic determinants, world and regional strategies
PDF
Are there hidden costs associated with teacher layoffs? The impact of job insecurity on teacher effectiveness in the Los Angeles Unified School District
PDF
CEO reputation: who benefits -- the firm and the CEO?
PDF
An empirical analysis of the quality of primary education across countries and over time
PDF
The Catholic church in Latin America: an evaluation of the institutional and political impacts of progressive church reforms
PDF
Teaching in the context of NCLB: a qualitative study of the impact on teachers' work, morale, and professional support
PDF
The impact of resource allocation on professional development for the improvement of teaching and student learning within a site-based managed elementary school: a case study
PDF
A qualitative look at what parents of infants aged birth -- 1 year old want and need in a parent training program.
PDF
Three essays on cooperation, social interactions, and religion
PDF
A world of education: the influence of culture on instructional style and perceived teacher effectiveness
PDF
Processes, effects, and the implementation of market-based environmental policy: southern California's experiences with emissions trading
PDF
Does fair value accounting exacerbate the pro-cyclicality of bank lending?
Asset Metadata
Creator
Shemyakina, Olga N.
(author)
Core Title
Armed conflict, education and the marriage market: evidence from Tajikistan
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Economics
Publication Date
07/20/2007
Defense Date
04/27/2007
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
armed conflict,education,household,Marriage,OAI-PMH Harvest,Tajikistan
Place Name
Tajikistan
(countries)
Language
English
Advisor
Strauss, John A. (
committee chair
), Casper, Lynne M. (
committee member
), Easterlin, Richard A. (
committee member
), Nugent, Jeffrey B. (
committee member
), Ridder, Geert (
committee member
)
Creator Email
shemyaki@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-m631
Unique identifier
UC1433953
Identifier
etd-Shemyakina-20070720 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-530401 (legacy record id),usctheses-m631 (legacy record id)
Legacy Identifier
etd-Shemyakina-20070720.pdf
Dmrecord
530401
Document Type
Dissertation
Rights
Shemyakina, Olga N.
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Repository Name
Libraries, University of Southern California
Repository Location
Los Angeles, California
Repository Email
cisadmin@lib.usc.edu
Tags
armed conflict
education