University of Southern California Dissertations and Theses

Essays on education: from Indonesia to Los Angeles, CA

(USC Thesis Other)

Essays on education: from Indonesia to Los Angeles, CA

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ESSAYS ON EDUCATION: FROM INDONESIA TO LOS ANGELES, CA by Will Kwon 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 2014 Copyright 2014 Will Kwon to Audrey ii Acknowledgments It has been 12 long years since I first entered the doors of the Kaprielian Hall at Univer- sity of Southern California. The long journey will always be cherished because of the people who made it possible. First and foremost, I would like to thank my advisor Professor John Strauss for his guidance. I would not have been able to start nor complete this dissertation without his support. His desire for excellence will always push me to be the best that I can be, in whatever path of life that will follow. He was the best teacher I could have asked for, and it has been a privilege to have been his student. I am grateful for the support given by Professor Jeff Nugent. From my acceptance letter to the final days of graduation, he has been the gentle giant of endless support. His keen intellect and shrewd suggestions shaped my passion for research from the earliest years of the Ph.D. program. I have become a full-blooded Trojan, and he has been the cornerstone of my loyalty to USC. I am grateful for the support given by Professor Tatiana Melguizo from the School of Education. Working as one of her research assistants for the Los Angeles Community College District project, I was fortunate enough to have met and worked with other researchers who were passionate about education. Her guidance and encouragement were vital to the completion of this dissertation. I am especially thankful to my colleague and co-author Federick Ngo whom I met through the LACCD project. iii I thank the economics department at USC for dealing with all the administrative mat- ters from my I-20s, leave of absence then returning, and change of citizenship. Young Miller has been the caregiver that quietly supported not only me, but everyone in the de- partment. I especially thank Morgan Ponder for his professionalism and for our personal friendship. I thank all my friends and colleagues throughout the years: Ken Changun Jung, Huseyin Gunay, Rubina Verma, Subha Mani, Olga Shemyakina, Farideh Motamedi, Babara Currier, Arya Gaduh, Saurabh Singhal, Jay Gwon, Bo Min Kim, Holly Kosiewicz, Kristen Fong, Fei Wang, Jaime Meza, Maggie Switek, and Qiankun Zhou. I learned from each and every one of them, and I am grateful for their companionship. Last, but certainly not least, I am indebted to my wife Audrey for her love and support throughout the entire journey. She allowed me to pursue my academic dream, never once wavering despite the rising student loans. She took care of me, then she took care of our son so that I may complete this dissertation. I would not be who I am without her, and I cannot be who I desire to be without her love. iv Contents Dedication ii Acknowledgments iii List of Tables vii List of Figures ix Abstract x 1 Introduction 1 1.1 Indonesia ................................. 1 1.2 Los Angeles, CA ............................. 3 2 Junior Secondary Transitional Dropouts in Indonesia 6 2.1 Introduction ................................ 6 2.2 Background & Literature Review..................... 7 2.3 Data .................................... 9 2.3.1 Indonesia Family Life Survey .................. 9 2.3.2 Children and Young Adults.................... 11 2.3.3 Dropouts ............................. 12 2.3.4 Variables of Interest ....................... 15 2.3.5 Empirical Strategy ........................ 20 2.4 Primary-Junior Secondary Transitional Dropouts ............ 22 2.4.1 Descriptive Statistics ....................... 22 2.4.2 Regressions ............................ 25 2.5 Junior-Senior Secondary Transitional Dropouts ............. 32 2.5.1 Descriptive Statistics ....................... 34 2.5.2 Regressions ............................ 36 2.6 Children Aged 13-14 ........................... 41 2.6.1 All Dropouts: Mid-school and Transitional ........... 42 2.6.2 Additional Variables ....................... 43 v 2.6.3 Descriptive Statistics ....................... 44 2.6.4 Regressions ............................ 47 2.7 What Happens to the Dropouts? ..................... 51 2.7.1 Outcomes in IFLS4........................ 52 2.8 Conclusion and Future Research ..................... 58 Appendix 2 ................................... 60 3 Using Multiple Measures to Make Course Placement Decisions 1 68 3.1 Introduction ................................ 68 3.2 Background ................................ 72 3.2.1 Using Multiple Measures .................... 74 3.3 Literature Review ............................. 74 3.3.1 Placement Tests ......................... 75 3.3.2 High School Information .................... 76 3.3.3 Noncognitive Measures ..................... 78 3.3.4 Using Multiple Measures for Course Placement ......... 79 3.4 Conceptual Framework .......................... 80 3.4.1 Validation ............................ 80 3.4.2 Placement Decisions ....................... 81 3.4.3 Placement Using Multiple Measures .............. 82 3.5 Multiple Measures in the LACCD .................... 84 3.5.1 Variation in Placement Policies ................. 85 3.6 Data & Methods ............................. 87 3.6.1 Multivariate Regression ..................... 88 3.6.2 Two Focus Colleges ....................... 89 3.6.3 Comparison Groups ....................... 90 3.6.4 Potential Selection Bias ..................... 91 3.7 Findings ................................. 92 3.7.1 Multiple Measures and Access to Higher-Level Courses .... 92 3.7.2 Multiple Measures and Student Success ............. 97 3.7.3 Cautions ............................. 100 3.8 Discussion ................................ 102 3.8.1 Increasing Placement Accuracy ................. 102 3.8.2 Promoting Access and Success ................. 104 3.9 Conclusion and Future Research ..................... 105 Appendix 3 ................................... 107 Appendix 116 Bibliography 117 1 This chapter is co-authored with Federick Ngo. vi List of Tables 2.1 Repeated/Failed a Grade ......................... 11 2.2 Age Distribution in IFLS ......................... 12 2.3 Education Level for 15-24 Year Olds ................... 13 2.4 Age Distribution for 13-24 Year Olds in IFLS Waves Combined .... 16 2.5 Summary Statistics for Primary School Graduates among 15-24 Year Olds 23 2.6 Linear Probability Regression for Primary School Graduates among 15- 24 Year Olds ............................... 27 2.6 Linear Probability Regression for Primary School Graduates among 15- 24 Year Olds (cont.)............................ 28 2.7 Oaxaca Decomposition between IFLS4 & IFLS1 Primary-Junior Sec- ondary Transitional Dropouts ....................... 31 2.8 Summary Statistics for Junior Secondary School Graduates among 15- 24 Year Olds ............................... 35 2.9 Linear Probability Regression for Junior Secondary School Graduates among 15-24 Year Olds .......................... 37 2.9 Linear Probability Regression for Junior Secondary School Graduates among 15-24 Year Olds (cont.) ...................... 38 2.10 Oaxaca Decomposition between IFLS4 & IFLS1 Junior-Senior Sec- ondary Transitional Dropouts ....................... 40 2.11 Education Level for 13-14 Year Olds ................... 42 2.12 Summary Statistics for All 13-14 Year Olds ............... 45 2.13 Linear Probability Regression for All 13-14 Year Olds ......... 48 2.14 Oaxaca Decomposition between IFLS4 & IFLS1 13-14 Year Olds . . . 50 2.15 Marriage reported in IFLS4........................ 53 2.16 Pregnancy reported in IFLS4 ....................... 54 2.17 Work and Earnings reported in IFLS4 .................. 55 2.18 Subjective Wellbeing reported in IFLS4 ................. 57 2.19 Subjective Wellbeing reported in IFLS3 ................. 58 A.2.1IFLS .................................... 62 A.2.2Pooled Linear Probability Regression for Primary School Graduates among 15-24 Year Olds .............................. 63 vii A.2.3With and Without Raven’s Test - Linear Probability Regression for Pri- mary School Graduates among 15-24 Year Olds ............. 64 A.2.4Pooled Linear Probability Regression for Junior Secondary School Grad- uates among 15-24 Year Olds....................... 65 A.2.5With and Without Raven’s Test - Linear Probability Regression for All 13-14 Year Olds .............................. 66 A.2.6Pooled Linear Probability Regression for All 13-14 Year Olds ..... 67 3.1 Multiple Measure Points for Math Placement .............. 86 3.2 Boosted and Enrolled Students ...................... 93 3.3 Students Receiving Multiple Measure Boost into Higher-Level Courses 94 3.4 Placement by Level within African-American and Latino Students . . . 95 3.5 Placement by Ethnicity within Pre-Algebra ............... 96 3.6 Regression Results, College A ...................... 99 3.7 Regression Results, College D ...................... 100 A.3.1Grade Distribution ............................ 107 A.3.2Placements by Ethnicity within Level .................. 108 A.3.3Placements by Level within Ethnicity .................. 109 A.3.4Regression Results, College A, Pass First Math ............. 110 A.3.5Regression Results, College A, Degree Credits ............. 111 A.3.6Regression Results, College A, Transfer Credits ............. 112 A.3.7Regression Results, College D, Pass First Math ............. 113 A.3.8Regression Results College D, Degree Credits .............. 114 A.3.9Regression Results College D, Transfer Credits ............. 115 viii List of Figures 2.1 Survival Estimate for Years of Education................. 14 2.2 Primary-Jr Transitional Dropouts among 15-24 Year Olds (%) ..... 24 2.3 Jr-Sr Transitional Dropouts among 15-24 Year Olds (%) ........ 33 2.4 Dropouts among All 13-14 Year Olds (%) ................ 43 A.2.1Female to Male Enrollment Ratio (%) .................. 60 A.2.2Literacy Rate (%) ............................. 60 A.2.3Indonesia Family Life Survey Coverage Map .............. 61 A.2.4Grade Repetition in Primary School ................... 61 A.1 Enrollment ................................ 116 ix Abstract This dissertation is a two-part essay on access to education and educational outcomes in two very different parts of the world: Indonesia and the United States. While the two countries are at different stages of educational development, we focus on the young pop- ulation who may be disadvantaged from achieving socially desired levels of education. First, from the developing country of Indonesia, young children who fail to obtain minimum level of education are studied. In 1994, the government of Indonesia extended universal education from primary to junior secondary school. Enrollment at junior sec- ondary schools remain below universal, with the largest dropouts coming from the tran- sition between primary school and junior secondary school. This essay examines the primary-junior secondary school transition for young adults from Indonesia using the the four waves of the Indonesia Family Life Survey from 1993 to 2007. I find the socio-economic-status of the family plays a significant role in the students dropping out after primary school, with parental education having the largest effect. A similar pattern is found for the junior-senior secondary school transition and for a younger cohort of teenagers. When the primary-junior secondary transitional dropouts are tracked from the previous waves to the latest wave, the dropouts are more likely to have been mar- ried, pregnant, working in agriculture, earning less, and rate themselves lower in general subjective wellbeing. x Second, from the city of Los Angeles, CA, community college students are studied with respect to their placements in mathematics courses. All students must satisfy a mathematics requirement prior to earning an associate degree or transferring to a 4-year institution. Based on the results of a single assessment, most community college stu- dents are placed in developmental math courses. However, concerns about placement accuracy have led states and colleges to consider using multiple measures to inform placement decisions. While student background measures such as high school GPA, prior math course-taking, and noncognitive traits are known to be predictive of college outcomes, there is limited evidence of their usefulness for course placement. We pro- vide evidence from California, where community colleges are required to use multiple measures, and examine whether the use of multiple measures increases access and suc- cess in college-level courses. We find that students who were placed into higher-level math only due to the use of multiple measures, particularly via prior math background and via high school GPA, performed no differently from their higher-scoring peers in terms of passing rates and long-term credit completion. xi Chapter 1 Introduction This dissertation is a two-part essay on access to education and educational outcomes in two very different parts of the world: Indonesia and the United States. Economically, Indonesia is the 15th largest economy in the world, but only at 7.7% of the United States in terms of GDP. 1 Indonesia’s GDP per capita ranks at 124th in the world, at 9.8% of the United States. Educationally, Indonesia’s mandatory education is 9 years, and the gross enrollment ratio in tertiary education is 27.2%. 2 United States’ mandatory education is 12 years, and the gross enrollment ratio in tertiary education is 95.3%. While the two countries are at very different stages of educational development, we focus on the young population who may be disadvantaged from achieving socially desired levels of education. 1.1 Indonesia In many developing countries, access to universal education remains an issue. Despite mandatory years of schooling offered by the government, the youngest and poorest have traditionally been physically and financially restricted from primary schools. Indonesia is in the midst of educational transition from developing to developed, as is its economy. Enrollment at all levels of schooling have steadily increased over the last two decades, and has surpassed world averages for both primary and secondary 1 International Monetary Fund, World Economics Outlook Database, April 2014 2 UNESCO Institute for Statistics, Data Centre, April 2014 1 enrollment. 3 While the enrollment rates have yet to reach the levels of OECD countries, Indonesia has lifted the level of education from below the world average to above it. Accordingly, the focus has shifted to higher levels of schooling. In 1994, the government of Indonesia expanded universal education from primary school to junior secondary school. Using the Indonesia Family Life Survey (IFLS), one of the largest micro-level data sets available in the world, I track the improvements in school enrollment particularly at the junior secondary level since the mid-1990s. The corollary to enrollment is dropping out. In an attempt to better identify the the students at the risk of dropping out, I dis- tinguish between different types of dropouts. A dropout is a student who drops out of school prior to completing mandatory years of schooling; however, a student dropping out after 1 year of schooling should not be considered the same as a student who drops out after 8 years of schooling. Naturally, not entering school and having zero years of formal education is the worst, and the lack of education gets progressively better or less worse with additional years of schooling. For Indonesia, dropping out of primary school has become less common across the years. Dropping out of junior secondary, once enrolled, has also become less common across the years. The largest instance of dropping out occurs during the transition from primary to junior secondary school. Over half of all dropouts prior to 9 years of manda- tory schooling drop out after completing primary school. In this essay, I focus mainly on the junior secondary transitional dropouts, the students who complete primary schools but then choosing to not enroll in junior secondary schools. I find that the percentage of students who do not continue onto junior secondary school after graduating from primary school has decreased from 1993 to 2007 from 3 Refer to Appenix Figure A.1 2 36% to 12%. These dropouts are coming from lower socio-economic-status house- holds, particularly with lower parental education. Father’s education level appears to be a stronger predictor of school continuation than mother’s education. I find a similar pattern for transitional dropouts at the next level between junior and senior secondary schools, as well as for younger children aged 13-14. The community school characteristics do not appear to affect the probability of drop- ping out during the transition within each of the waves. Yet, a majority of the change in the probability of dropping out from the first wave of the IFLS to the last wave ap- pears to be largely attributed to the changes in school characteristics. The significance of changes at the community school level may highlight the social improvements in school infrastructure during this time period. When the respondents from the first three waves are tracked in the latest wave, the dropouts appear to exhibit a very different pattern than the non-dropouts. The dropouts are more likely to have married, and subsequently the female dropouts are more likely to have been pregnant. The dropouts are more likely to work in labor intensive agri- cultural/forestry/fishing industries, while learning less income. The dropouts are more likely to rate themselves lower in the subjective wellbeing measures. The Indonesian dropouts who were unable to obtain the minimum level of education are characteristi- cally different from the rest of society at the time of schooling transition as well as later on in life. 1.2 Los Angeles, CA In the United States, dropping out of senior secondary or high schools is still preva- lent, but the choice is voluntary. The most cited reasons for dropping out are dislike for 3 school, not finding school interesting, missing too many days, and failing school (Roter- mund, 2007). With the widely accepted general educational development (GED) test, obtaining high school equivalent certification is readily available for the dropouts. Ac- cording to the U.S. Department of Education (2013), only 7% of young American aged 16-24 have not earned a high school credential as of 2011. Of high school graduates, near 70% continue onto 2- or 4-year colleges. More than half of these colleges students attend 2-year community colleges. Community colleges serve a broad population and purpose. 4 Community colleges are well suited for more career-oriented students, looking to obtain technical training. The community college can also serve as a pathway to a 4-year college, taking courses that can be transferred to the destination college, and thereby save on tuition costs and living expenses since most students choose to attend campuses near their homes. 5 Fur- thermore, the community colleges serve as a medium for students to prove themselves academically for their desired colleges, especially for late bloomers. Particularly for those wishing to transfer from community colleges to 4-year col- leges, students must meet minimum requirements in English and in mathematics. To be placed in the mathematics sequence, students undertake a placement test. While the validity of placement tests have been questioned in the education literature, our research addresses one aspect of the placement policy that may improve accuracy of placements in the short-run. 6 In addition to the raw test score from the placement test, students receive additional points for certain answers on the demographics and personal history 4 As of 2013, the largest community college in the country is the Miami Dade College with about 175,000 enrolled students. The largest university in the country is Arizona State University with about 60,000 enrolled students. 5 As of 2014, the per-unit cost of tuition at University of Southern California is $1,536. The per-unit cost of tuition at Santa Monica Community College is $46 for in-state residents and $315 for out-of-state residents including international students. 6 Long-run changes made at the institutional level would require full assessment of the validity of the placement tests and guidelines. Should the tests and guidelines prove to be unsuitable, then alternative 4 survey. For example, students with good GPA in high school or students who completed pre-calculus in high school are awarded extra points. These multiple measure points are intended to be used in conjunction with the raw test score to better place students in the appropriate classes. In this chapter, co-authored with Federick Ngo from the USC Rossier School of Ed- ucation, we use the data from the Los Angeles Community College District to examine the impact of the multiple measures. 7 In an attempt to identify the validity of specific multiple measures, we focus on two colleges with only a single multiple measure cri- teria: College A awards points only for prior math background and College D awards points only for high school GPA. We find that the multiple measure points awarded in these colleges only marginally affected the overall distribution of placements. Only the African-American and Latino students from the lowest levels are boosted up signifi- cantly to the higher-level with the additional multiple measure points. Since the boosted students in the higher-level course are now the lowest-scoring students, we would be concerned if the multiple measures points placed in a course too challenging for them to succeed. We find that the boosted students perform equally successfully in the higher-level courses, when compared to all the students in the higher-level courses, as well as other lower-scoring students who made it into the higher-level courses without assist from the multiple measure points. We believe that multiple measure points can be an effective way to increase the accuracy of placement for students, and we believe further inves- tigation is required to better identify which multiple measures are good predictors of student success in community college mathematics courses. solutions must be proposed and examined. Rather than analyzing such large-scale policy implications, we focus on how minor additions to test scores may impact students. 7 Federick Ngo, federick.ngo@usc.edu 5 Chapter 2 Junior Secondary Transitional Dropouts in Indonesia 2.1 Introduction This chapter is about the young children of Indonesia who do not attain junior sec- ondary education. Since 1994, mandatory education in Indonesia has been set at junior secondary level; yet, enrollment at junior secondary schools remain far below universal level even to this day. I aim to examine the trends of low enrollment in the junior sec- ondary schools via the primary-junior secondary transitional dropouts from Indonesia using the four waves of the Indonesia Family Life Survey (IFLS) from 1993 to 2007. I define a transitional dropout as someone who chooses not to pursue the next level of ed- ucation after completing the previous level; a primary-junior transitional dropout is thus a student deciding not to pursue junior secondary school after graduating from primary school. Treating the waves as both independent cross-sections and panel, I examine the characteristics of the dropouts, then track them to the latest wave to see if there are any short-run and medium-run impacts of dropping out. I find the following results. Indonesia’s overall education level increases from the first wave of IFLS in 1993 to the fourth wave in 2007, and the percentage of primary- junior secondary transitional dropouts decreases from 36% to 12%. The transitional 6 dropouts are more likely to come from families with lower socio-economic-status, par- ticularly from rural households with lower per capita expenditure levels. Parental edu- cation plays an important role in the decision to transition, possibly more so in times of economic distress. The dropouts are subsequently much more likely to have been mar- ried, more likely to have given birth at earlier ages, more likely to be working in agri- culture/forestry/fishing and earning less salary, and consider themselves to be worse-off than non-dropouts. 2.2 Background & Literature Review Returns to education are large for developing countries. The simple empirical approxi- mation of human capital theoretical framework in the form of a log earnings equation, stemming from the seminal benchmark model by Jacob Mincer (1974), is often used to estimate such returns. Psacharopoulos and Patrinos (2004) estimate an average rate of return as high as 10% per additional year of schooling across all years and levels of schooling. However, returns to education are larger for developing countries partly because en- rolling and retaining young children in primary and secondary schools have been a chal- lenge. The World Bank Human Development Network (2009) reports that substantial proportions of children in poor developing countries never enroll in schools or attend for only a year or two, and poverty and shortage of schools are the reasons. Glewwe and Jacoby (1995) estimate that an average delay of two years into primary school in Ghana costs about 6% of life-time wealth. In places where gender bias has traditionally been prevalent, being a girl meant even less education. Parish and Willis (1993) report the early born females in Taiwan receive less education, often marrying early. Lloyd et al. (2000) find that girls’ enrollment decreases upon entering teenage years in Kenya. 7 Connelly and Zheng (2003) find that rural girls are less likely to enroll and graduate in China. Skoufias and Parker (2006) find that teenage girls in Mexico are more likely to be drop out when household head faces unemployment. This paper will demonstrate that gender bias does not appear to be an issue for Indonesia, but that enrollment still is. 1 Indonesia has been active in promoting and protecting education. Duflo (2001) stud- ies the impact of primary school construction program in Indonesia from 1973 to 1978 - building of over 61,000 primary schools - resulting in increases in the average years of schooling and wages. Behrman (1987) cites Indonesia as being in the top quartile of developing countries in schooling achievement relative to levels predicted by per capita income in both 1960 and 1981. Deolalikar (1993) reports that females in Indonesia ac- quired secondary and tertiary education in “relatively larger numbers” than males. 2 The World Development Indicator confirms Indonesia’s higher literacy rate compared to the least developed countries and the world from 1990 to present. 3 With the onset of the Asian Financial Crisis of 1997/98 (AFC), there were fears of school enrollment decreasing as was the case in the mid-1980s during the decline in oil prices (Cameron, 2000). Thomas et al. (2004) examine the short-run impacts of the crisis on education enrollment based on just before and after snapshots from IFLS to capture the immediate impacts of the crisis. The authors find that the increasing trend of school enrollment regressed during the crisis, and the schooling of younger, poorer children suffered. Yet, other studies find that the crisis did not affect Indonesia’s education as negatively as feared. Hartono and Ehrmann (2001) report that enrollment 1 Refer to Appendix Figure A.2.1 for female to male ratio for Indonesia using the World Development Indicator. 2 Deolalikar’s enrollment data was for 1987. Female enrollment in tertiary education remained low at 16%, but was a substantial growth from 1% in 1960 and 5.9% in 1980. 3 Refer to Appendix Figure A.2.2. 8 did not drop noticeably for primary and junior secondary schools in years 1997-2000 from the SUSENAS (Indonesian census) data. The authors further report that enrollment in senior secondary schools even increased. Sparrow (2007) using the SUSENAS data confirms their findings. Cameron (2001), using the 100 Villages Survey focusing on rural and poor areas of Indonesia, finds that school enrollment dropped only slightly at the onset of the crisis and rebounded quickly by May of 1999. Even with the IFLS data, using the 1997 and 2000 waves, Strauss et al. (2004) finds enrollment rates of primary and junior secondary school-aged children are slightly higher in 2000, suggesting a quick recovery from the crisis. Both Cameron and Sparrow, along with Jones and Hagul (2001), attribute the Jaring Pengamanan Sosial (JPS) or the social safety net program offered by the Indonesian government, providing direct cash payments to students and to schools, for the quick recovery. Using the Indonesia Family Life Survey, I contribute to the literature on Indonesia’s educational trends in two ways. First, I focus on the dropouts between primary-junior secondary school transition where the dropping out is the largest. Second, using IFLS as repeated cross-sectional data, I outline the changes in the characteristics of the tran- sitional dropouts from 1993 to 2007. 2.3 Data 2.3.1 Indonesia Family Life Survey The data set used is the Indonesia Family Life Survey, conducted by the RAND Cor- poration. 4 The IFLS is a large-scale socio-economic and health survey. IFLS acquires extensive information at the individual, household, and community levels. The first 4 Refer to Appendix Figure A.2.3 for a map of Indonesia and IFLS coverage. 9 wave of IFLS began in 1993 (IFLS1), second wave in 1997 (IFLS2), third wave in 2000 (IFLS3), and the fourth wave in 2007 (IFLS4). One of the major strengths of IFLS is that the respondent households and members have been tracked with a remarkably low attrition rate. From the original 7,224 house- holds of IFLS1, 6,596 households (91%) remain in IFLS4. When a member splits from the household, the split-offs are followed, thus resulting the initial 7,224 households to increase to 13,536 households by IFLS4. In the process, the number of responding individuals increased from 22,169 in IFLS1 to 43,619 in IFLS4. 5 Overall, 87.6% of responding household members have been tracked in IFLS4 in 2007. Considering some 17,000 islands that make up the Indonesian archipelago, IFLS does an excellent job of representing the Indonesian population across the 13 of the 27 provinces and over 312 enumeration areas in the IFLS surveys. 6,7 With respect to education, the major advantage the IFLS offers is its rich retrospec- tive educational component. Rather than merely the years of completed education, I am able to create the entire educational history of the respondent including repetition and gaps between grades. Behrman and Deolalikar (1991) demonstrate that using completed years of schooling, rather than years of schooling including repetition, in estimating re- turns to education can lead to heavily biased estimates. 8 5 Refer to Appendix Table A.2.1. 6 In IFLS1, there were 27 provinces. Since then, East Timor gained independence while 7 new provinces have been created. 7 Total enumeration areas equal 321, but 9 are twin areas. 8 While I do not estimate returns to schooling in a Mincerian framework, I was able to compare the Kaplan-Meier survival functions using years of schooling versus completed years of schooling. I found that the largest exit between the transition of schools was better illustrated using the completed years of schooling. 10 2.3.2 Children and Young Adults Indonesia primary school requires children to enroll by age 7. The primary school is 6 years, followed by 3 years of junior secondary school. Since this paper is about young adults or particularly those facing schooling decisions, I select respondents aged 13- 14 from Book5 for children under 15 years of age and respondents aged 15-24 from Book3 for adults over 15 years of age. When focusing on primary-junior secondary transition, sufficient time needs to be allowed for the respondents to complete primary school. Grade repetition is prevalent in Indonesia, as is the case in many developing countries, and repeating or failing a grade would delay a student from entering junior secondary school. 9 Table 2.1 shows that nearly 20% of students report having repeated a grade in primary school. Still, about 95% of the primary students who ever repeated, repeat only once or twice. Thus at least for respondents aged 15 and older, there should have been enough time to complete primary and reach the junior secondary transitional period even with repetition. 10 For the remaining 13-14 year olds, discussed further in detail in Section 2.6.1, all types of dropouts will be considered. Table 2.1: Repeated/Failed a Grade (a) Ever IFLS1 IFLS2 IFLS3 IFLS4 Primary 0.197 0.210 0.199 0.169 Jr 0.0426 0.0165 0.0167 0.0137 Sr 0.0308 0.0142 0.0142 0.0112 N 7282 16558 17869 47983 Ages 15-49, conditional on ever enrolling in school (b) Total Repeats in Primary (%) IFLS1 IFLS2 IFLS3 IFLS4 1 84.30 86.34 82.84 81.29 2 12.92 10.55 13.16 14.24 ≥ 3 2.784 3.110 4.003 4.470 N 898 1640 1474 3758 Conditional on ever repeating/failing a grade 9 Refer to Appendix Figure A.2.4 for comparison of grade repetition in primary school for Indonesia and the world. Indonesia has reduced grade repetition below the world average and edging closer to OECD average. 10 By IFLS4, less than 2% of all 15 year olds are still enrolled in primary school. 11 All the IFLS waves combined, the initial unconditional sample size was 5,926 for 13-14 year olds and 21,460 for the 15-24 year olds as shown in Table 2.2. The 13-14 year olds represent the oldest cohort of children, as defined by IFLS. The age distribution among children is evenly divided for children up to 14 years of age. The 15-24 year olds represent the youngest cohort of adults, specifically, the youngest quartile of the adult population. With age restricted to 13-24, the most relevant population of children and young adults should be included in the sample. Table 2.2: Age Distribution in IFLS (a) Children IFLS1 IFLS2 IFLS3 IFLS4 Total 6 & Under 3203 (41.38) 4398 (42.23) 5574 (47.48) 6951 (50.92) 20126 (46.22) 7-8 1028 (13.28) 1415 (13.59) 1517 (12.92) 1958 (14.34) 5918 (13.59) 9-10 1205 (15.57) 1396 (13.40) 1546 (13.17) 1571 (11.51) 5718 (13.13) 11-12 1164 (15.04) 1507 (14.47) 1523 (12.97) 1665 (12.20) 5859 (13.45) 13-14 1141 (14.74) 1699 (16.31) 1579 (13.45) 1507 (11.04) 5926 (13.61) N 7741 10415 11739 13652 43547 (b) Adults IFLS1 IFLS2 IFLS3 IFLS4 Total 15-24 1343 (9.316) 5443 (27.34) 7590 (29.78) 7084 (24.38) 21460 (24.15) 25-34 3573 (24.78) 4451 (22.36) 6002 (23.55) 7991 (27.50) 22017 (24.78) 35-44 3474 (24.10) 3883 (19.50) 4702 (18.45) 5667 (19.50) 17726 (19.95) 45-54 2489 (17.27) 2581 (12.96) 3019 (11.85) 4005 (13.78) 12094 (13.61) 55 & Over 3537 (24.54) 3551 (17.84) 4170 (16.36) 4310 (14.83) 15568 (17.52) N 14416 19909 25483 29057 88865 column pct in parenthesis 2.3.3 Dropouts In defining the term dropout, I distinguish two types of dropouts. Mid-school dropout When a student does not complete a particular level of schooling after he/she began. Transitional dropout When a student does not continue to the next level of schooling after completing the previous level. 12 The mid-school dropouts are likely to be different from the transitional dropouts; a child who drops out shortly after entering the first grade should not be considered the same as another who finishes sixth grade but chooses not to continue. Table 2.3 shows that the proportion of those not completing primary level is relatively small at only 6% by IFLS4 but is much larger at around 15% in IFLS1. Table 2.3 also illustrates the educational attainment in Indonesia from 1993 to 2007. In IFLS1, females had lower educational attainment across all levels. By IFLS4, females have not only caught up but surpassed the males marginally across all levels. In particular, junior secondary education shows the largest change: in IFLS1, only 46.4% of females completed ju- nior secondary school or higher versus 57.4% for males. By IFLS4, 74.3% of females completed junior secondary or higher versus 69.9% for males. Table 2.3: Education Level for 15-24 Year Olds Female Male IFLS1 IFLS2 IFLS3 IFLS4 IFLS1 IFLS2 IFLS3 IFLS4 No school 0.069 0.027 0.022 0.013 0.036 0.021 0.016 0.010 Completed primary or higher 0.827 0.923 0.920 0.945 0.883 0.905 0.914 0.936 Completed junior or higher 0.464 0.616 0.636 0.743 0.574 0.591 0.630 0.699 Completed senior or higher 0.237 0.293 0.290 0.361 0.313 0.255 0.283 0.349 N 873 2896 4023 3804 470 2547 3567 3280 Unconditional: all 15-24 year olds Figure 2.1 illustrates the Kaplan-Meier survival function for completing each grade. 11 Figure 2.1 shows two educational trends in Indonesia among the young adults aged 15-24: that overall educational attainment increases from IFLS1 to IFLS4 and that the transitional dropouts dominate the mid-school dropouts. The proportions of those who complete both primary and junior secondary schools are visibly higher in IFLS4 than in IFLS1. In IFLS1, the largest proportion of dropouts is between primary and junior secondary. By IFLS4, the largest proportion of dropouts is between junior and 11 The product limit method of Kaplan and Meier (1958) is estimated via Stata command sts graph, survival. 13 senior secondary, as well as between senior secondary and college. The cumulative per- centage of mid-school dropouts is not trivial: in IFLS1, mid-school dropout rates during primary school years 0-5 sum to nearly 20%. Still, the transitional dropout rate after primary school is noticeably higher at around 30%. Figure 2.1: Survival Estimate for Years of Education 0.00 0.25 0.50 0.75 1.00 0 5 10 15 20 Years of completed education Kaplan−Meier survival estimate (a) IFLS1 0.00 0.25 0.50 0.75 1.00 0 5 10 15 20 Years of completed education Kaplan−Meier survival estimate (b) IFLS2 0.00 0.25 0.50 0.75 1.00 0 5 10 15 20 Years of completed education Kaplan−Meier survival estimate (c) IFLS3 0.00 0.25 0.50 0.75 1.00 0 5 10 15 Years of completed education Kaplan−Meier survival estimate (d) IFLS4 For ages 15-24. Considered to exit upon the final year of education obtained; duration equals the completed years of education. Recognizing the difference between a mid-school and transitional dropout, I focus mainly on the transitional dropouts; initially the primary-junior secondary transitional 14 dropouts who have completed primary education in Section 2.4 and then the junior- senior secondary transitional dropouts who have completed junior secondary education in Section 2.5. Narrowing the focus to the time of transition, excluding the choices made prior to reaching the completion of a school level, may better identify the characteristics of the dropouts who contribute to what has been the largest bottlenecks in educational attainment in Indonesia. I additionally look at both types of dropouts for 13-14 year olds in Section 2.6. Table 2.4 shows the breakdown of the final sample by age for the IFLS waves com- bined. The proportion of each age cohort is roughly equal at 50% for the 13-14 year olds and at 10% for the 15-24 year olds. The smaller number of junior secondary grad- uates is expected for the 15 year olds given the possibility of delay in entry and via grade repetition. When the final sample is separated by gender, more females appear to be graduating from primary and junior secondary school, consistent with the pattern of rising female education in Table 2.3. 2.3.4 Variables of Interest The main outcome of interest is the transitioning of schooling from one level to next; or in the case of 13-14 year olds, whether the student is currently enrolled in primary or ju- nior secondary school. The explanatory variables are categorized into personal, house- hold, and community characteristics. Personal characteristics such as age and gender are at the individual level. Household characteristics such as per capita expenditures (PCE) are shared across all individuals within the same household. Community charac- teristics such as number of schools and average distance to schools are shared across all households living within the same community. But while personal, household, and com- munity characteristics are collected and reported in the descriptive summary statistics, not all variables are included in all of the regression analysis. 15 Table 2.4: Age Distribution for 13-24 Year Olds in IFLS Waves Combined (a) Combined Unconditional Sample Selected 13-14 15-24 13-14 Primary Junior Secondary 13 2948 (49.75) 2901 (49.55) 14 2978 (50.25) 2954 (50.45) 15 2259 (10.53) 2089 (10.69) 832 (6.155) 16 2309 (10.76) 2151 (11.01) 1440 (10.65) 17 2355 (10.97) 2165 (11.08) 1649 (12.20) 18 2290 (10.67) 2128 (10.89) 1607 (11.89) 19 2004 (9.338) 1839 (9.414) 1414 (10.46) 20 2092 (9.748) 1854 (9.491) 1334 (9.869) 21 1982 (9.236) 1803 (9.230) 1305 (9.655) 22 2036 (9.487) 1803 (9.230) 1272 (9.410) 23 2067 (9.632) 1863 (9.537) 1359 (10.05) 24 2066 (9.627) 1839 (9.414) 1305 (9.655) N 5926 21460 5855 19534 13517 column pct in parenthesis (b) Sample Selected 13-14 Primary Junior Secondary Female Male Female Male Female Male 13 1449 (49.95) 1452 (50.05) 14 1491 (50.47) 1463 (49.53) 15 1047 (50.12) 1042 (49.88) 458 (55.05) 374 (44.95) 16 1103 (51.28) 1048 (48.72) 760 (52.78) 680 (47.22) 17 1133 (52.33) 1032 (47.67) 873 (52.94) 776 (47.06) 18 1145 (53.81) 983 (46.19) 857 (53.33) 750 (46.67) 19 994 (54.05) 845 (45.95) 752 (53.18) 662 (46.82) 20 986 (53.18) 868 (46.82) 692 (51.87) 642 (48.13) 21 1027 (56.96) 776 (43.04) 716 (54.87) 589 (45.13) 22 1024 (56.79) 779 (43.21) 711 (55.90) 561 (44.10) 23 1034 (55.50) 829 (44.50) 757 (55.70) 602 (44.30) 24 1015 (55.19) 824 (44.81) 707 (54.18) 598 (45.82) N 5855 19534 13517 row pct in parenthesis Because the decision to transition to junior secondary school is made at the end of primary school, some of the variables may not necessarily reflect the environment at which the respondent faced at the time of school enrollment. A household’s level of wealth may be a determining factor for schooling at the time of transition; thus, a lagged value of household’s wealth may be a better predictor than the current value of wealth for the relatively older respondents. However, since lagged values come only from the 16 previous waves, the available lagged values may not necessarily reflect the conditions at the transitional age (roughly around 12-14) any better than the current value: for a 20 year old in IFLS2, the lagged values come from when the respondent was 16 years old in IFLS1. These issues remain for the junior-senior secondary school transition. Considering the issue of pinning down the variables at the time of transition for both the primary-junior secondary transition and junior-senior secondary transition, the regression analysis will have two models: first with parental education and geographical indicators only and second with additional community characteristics at the time of the survey (though not necessarily at the time of schooling decision). The geographical indicators are the location of the respondents at age 12, and not at the time of survey. IFLS tracks migration history of the respondents, including the location at birth and location at age 12. The inclusion of additional community variables in the second model is to break- down the geographical effects. Since the main outcome of interest are schooling deci- sions, community characteristics are the school environment and infrastructure available to respondents. Although the community characteristics are at the time of survey, they are likely to persist over time. Number of schools, distance to schools or number of students are likely to change slowly since such institutional changes are made at the community level and not at the individual or household level. 2.3.4.1 Personal Characteristics Personal characteristics include gender, religion (Muslim vs. non-Muslim), age, and Raven’s test scores if available. Indonesia is one of the most Islamic countries in the world, but other religions are openly accepted as one of the 5 philosophical foundations of the nation, the Pancasila. 17 Although religion may be based on geographical location, being a non-Muslim may represent different attitudes towards education. 12 In IFLS3, the Raven’s progressive matrices test was introduced to respondents aged 15-24 to measure reasoning ability of the respondents. The Raven’s test does not re- quire any formal education and is a measure of fluid intelligence that is deemed essen- tial to successfully carry out school activities such as reading, writing, and arithmetic (Ferrer and McArdle, 2004). In an environment where education is relatively costly, student’s ability and performance at school would influence parents’ decision to finan- cially support their child’s education. Primary school in Indonesia concludes with a national exam known as the EBTANAS, which is often used as an entrance exam to junior secondary schools. Ideally, both the EBTANAS and Raven’s test scores should be considered when examining the determinants of school continuation. However, the EBTANAS scores were recorded by only about 60% of the respondents, with only a small proportion reporting failing scores; thus instead, only the Raven’s test scores are utilized when possible. 13 The Raven’s test was administered for 7-24 year olds in IFLS4, but across two age groups: 7-14 and 15-24. The Raven’s test scores are normalized for compatibility purposes. 2.3.4.2 Household Characteristics Household characteristics include a rural indicator, per capita expenditures, an indicator for experiencing crop-loss, an indicator for having dirt floors in the house, the value of land, an indicator for a farming household, and parental education. The per capita expenditures are used as a measure of long-run resources, whereas land value and having 12 About 3% of the Indonesian population are Hindu, but most of the Indonesian Hindus are Balinese. 13 The correlation between between EBTANAS score for mathematics and the Raven’s score is about 0.53 for IFLS3 and IFLS4. 18 dirt floors are used as proxies for wealth. 14 Experiencing crop-loss is used as a proxy for external shocks. In the first model, only the rural indicator, the value of land, and parental education are used. Although the value of land is at the time of survey, it is used as a proxy for wealth. Land sale, particularly by farming households, is relatively rare in developing countries. The land values are measured in quartic-roots rather than logs because of observations with zeroes. Parental education is categorized into levels, with senior sec- ondary and tertiary education grouped together. Parental education is matched at the individual level when possible, dropped if not matched. 15 2.3.4.3 Community Characteristics Community primary school characteristics include number of schools, uniform expen- ditures, distance to schools, EBTANAS scores, number of students, and student/teacher ratio. In the first model, none of the characteristics are used. Instead, I rely on geograph- ical indicators at age 12 only. For the second model, several school characteristics are obtained from the com- munity section of IFLS, the COMFAS. The COMFAS includes a section for schools, whereby up to 3 schools in each of the primary, junior secondary, and senior secondary levels are surveyed per enumeration area. I create community averages of school charac- teristics, then the community averages are merged with individual data at the sub-district or kecamatan level when possible. If not, district or province averages are assigned for regression purposes. Although the individuals are tracked with a remarkably low at- trition rate, mainly because the IFLS does extensive work in locating the movers, a 14 The aggregated PCE data was downloaded from the RAND Corporation. 15 About 6% of the available sample is dropped in IFLS1 due to missing parental education informa- tion. The latter waves of the IFLS contained more detailed surveys on parental education, including the deceased. The proportion of IFSL4 respondents without parental education information is around 1%. 19 detailed COMFAS was not surveyed at the community level if the movers move outside the enumeration areas. 16 Despite being averages at the sub-district level, some com- munity characteristics showed extremely skewed tails: in IFLS1, mean student/teacher ratio was 42.4 with a standard deviation of 216.5. Such outliers were identified and excluded from averages using the Hadi (1992) multivariate outlier method. 17 For the primary-junior secondary transition, community characteristics of primary and junior secondary schools are included. The community school characteristics, par- ticularly at the next level may represent the general educational infrastructure and avail- ability for students to consider during the transition. The number of schools, distance to schools, and number of students per school help represent the community in which the respondent lived. Uniform expenditures, EBTANAS scores, and the student/teacher ratio may reflect the quality of the schools within the community. Uniform expenditures are also measured in quartic-roots due to observations with zeroes. Similarly, for junior- senior secondary transition, community characteristics of junior and senior secondary schools are included. 2.3.5 Empirical Strategy All estimates in the paper are from linear probability regressions. I use the linear prob- ability model as the results are not sensitive to using logit or probit models. The ex- planatory regressors are vectors of personal (P ), household (H), and community (C) variables. 16 Starting in IFLS3, a shortened community survey version known as the mini-CFS was asked to the leaders of the new village; however, since the community characteristics of interest come from the partic- ular school modules, some respondents remained unmatched. In IFLS4, about 5% of the respondents are unmatched for at the sub-district level for community characteristics. 17 Via Stata command hadimvo, following the method outlined in Hadi (1992). 20 y i = α+βP +γH +ηC +δD +ε i (2.1) All personal characteristics are binary, including age indicators. The omitted age category is the youngest age indicator within the reference group. All household char- acteristics are binary, except for log PCE and land value. Highest parental education is separated into 4 levels: no schooling, primary, junior secondary, and senior secondary or higher, with No schooling being the omitted category. All community characteris- tics other than number of schools are sub-district primary and junior secondary school averages. Additional indicator variablesD are included as controls in the regression. Observa- tions missing personal and household characteristics are excluded from the regression. Observations missing community characteristics are replaced with district or province averages for regression purposes, and indicators for such missing observations are in- cluded. The geographical indicators at the district or kabupaten level are included. Since any districts with all 0 or 1 for the dependent variable would be excluded from the regression, I assign the nearest non-collinear district for such collinear districts to retain them in the regression. The standard errors are robust and clustered at the district level. As noted, the regression analysis for the primary-junior secondary transition and junior-senior secondary transition comprise two models; one with geographical indica- tors only and one with community characteristics. The first model includes only gender, religion, age, rural indicator, land value, parental education, and geographical indica- tors. The second model adds community characteristics at the primary and junior sec- ondary levels. In the first regression, the geographical indicators are intended to capture all community effects, hence, indicators at the sub-district or kecamatan level would 21 be ideal. Then in the second regression, district indicators can be used in conjunction with community characteristics at the sub-district level. However, there are excessive number of collinear sub-districts. In IFLS1, the number of distinct sub-districts is 361, of which 312 are collinear; 205 sub-districts had all the respondents drop out and 105 sub-districts had all the respondents continue schooling. With such an overwhelming majority of collinear sub-districts, the assigning of nearest non-collinear sub-district be- came unfeasible. As a result, only district indicators could be used in both of the models. 2.4 Primary-Junior Secondary Transitional Dropouts The primary-junior secondary transitional dropouts across the waves are shown in Fig- ure 2.2. The proportion of dropouts decreases from 41% in IFLS1 to 13% in IFLS4 for females and from 28% to 11% for males. Between the waves, the decrease is noticeably large from IFLS1 to IFLS2 then again from IFLS3 to IFLS4. The overall decrease for females represents the increased educational opportunities for girls in Indonesia. 2.4.1 Descriptive Statistics Separating the non-dropouts from the dropouts yields the descriptive statistics shown in Table 2.5. Within each wave, the non-dropouts and dropouts clearly display differ- ent characteristics. Simple tests of means reveal that vast majority of the variables are significantly different at the 0.1% level. The dropouts are more likely to be Muslim and have lower standardized Raven’s test scores. The Raven’s test z-scores are not only lower for the dropouts, but the averages for the dropouts are negative while the averages for the non-dropouts are positive, both in IFLS3 and IFLS4. The dropouts are more likely to have parents, both father and 22 Table 2.5: Summary Statistics for Primary School Graduates among 15-24 Year Olds IFLS1 IFLS2 IFLS3 IFLS4 ND Dropout Diff ND Dropout Diff ND Dropout Diff ND Dropout Diff Personal: Male 0.417 0.291 0.126 ∗∗∗ 0.486 0.414 0.0717 ∗∗∗ 0.484 0.416 0.0684 ∗∗∗ 0.468 0.411 0.0577 ∗∗ Non-muslim 0.155 0.0812 0.0739 ∗∗∗ 0.128 0.0627 0.0657 ∗∗∗ 0.125 0.0652 0.0600 ∗∗∗ 0.114 0.0588 0.0557 ∗∗∗ Raven’s test (z-score) 0.196 -0.447 0.642 ∗∗∗ 0.184 -0.358 0.542 ∗∗∗ Household: Rural 0.374 0.641 -0.268 ∗∗∗ 0.381 0.655 -0.274 ∗∗∗ 0.389 0.677 -0.288 ∗∗∗ 0.420 0.657 -0.237 ∗∗∗ Log PCE 12.98 12.68 0.295 ∗∗∗ 12.91 12.48 0.433 ∗∗∗ 12.95 12.62 0.330 ∗∗∗ 13.04 12.60 0.445 ∗∗∗ Faced crop loss 0.0547 0.0759 -0.0213 0.0948 0.130 -0.0348 ∗∗∗ 0.0836 0.159 -0.0752 ∗∗∗ 0.0496 0.0526 -0.00293 Dirt floor 0.0369 0.0838 -0.0468 ∗∗ 0.0614 0.240 -0.179 ∗∗∗ 0.0548 0.190 -0.135 ∗∗∗ 0.0422 0.134 -0.0917 ∗∗∗ Land value 28.87 30.76 -1.895 34.43 33.20 1.239 32.17 30.31 1.864 27.35 22.78 4.571 ∗∗ Farming household 0.286 0.408 -0.123 ∗∗∗ 0.283 0.433 -0.150 ∗∗∗ 0.295 0.381 -0.0852 ∗∗∗ 0.270 0.305 -0.0352 ∗ Father’s edu: None 0.0899 0.356 -0.266 ∗∗∗ 0.0626 0.245 -0.182 ∗∗∗ 0.0618 0.253 -0.192 ∗∗∗ 0.0536 0.219 -0.166 ∗∗∗ Father’s edu: Primary 0.544 0.595 -0.0504 0.516 0.692 -0.176 ∗∗∗ 0.520 0.692 -0.172 ∗∗∗ 0.510 0.721 -0.211 ∗∗∗ Father’s edu: Jr 0.123 0.0301 0.0929 ∗∗∗ 0.166 0.0440 0.122 ∗∗∗ 0.157 0.0374 0.120 ∗∗∗ 0.153 0.0494 0.103 ∗∗∗ Father’s edu: Sr/College 0.243 0.0192 0.224 ∗∗∗ 0.256 0.0195 0.236 ∗∗∗ 0.262 0.0176 0.244 ∗∗∗ 0.283 0.0104 0.273 ∗∗∗ Mother’s edu: None 0.198 0.501 -0.303 ∗∗∗ 0.129 0.380 -0.251 ∗∗∗ 0.118 0.379 -0.261 ∗∗∗ 0.0962 0.314 -0.218 ∗∗∗ Mother’s edu: Primary 0.562 0.471 0.0905 ∗∗ 0.579 0.594 -0.0147 0.592 0.601 -0.00836 0.573 0.650 -0.0773 ∗∗∗ Mother’s edu: Jr 0.114 0.0193 0.0946 ∗∗∗ 0.144 0.0218 0.122 ∗∗∗ 0.139 0.0160 0.123 ∗∗∗ 0.151 0.0295 0.121 ∗∗∗ Mother’s edu: Sr/College 0.126 0.00826 0.118 ∗∗∗ 0.147 0.00418 0.143 ∗∗∗ 0.151 0.00436 0.147 ∗∗∗ 0.180 0.00641 0.174 ∗∗∗ Community - Primary: Number of schools 3.206 3.145 0.0610 6.096 5.035 1.061 ∗∗∗ 4.185 3.592 0.593 ∗∗∗ 4.559 4.446 0.112 Uniform expenditures 4.838 4.082 0.755 ∗ 14.88 15.25 -0.365 17.85 18.22 -0.367 ∗∗∗ 18.92 18.31 0.609 ∗∗∗ Distance to schools (km) 0.876 1.115 -0.239 ∗∗∗ 1.398 1.291 0.107 ∗∗ 1.692 1.839 -0.147 ∗∗∗ 2.880 2.799 0.0804 EBTANAS Indonesian 6.675 6.423 0.251 ∗∗∗ 6.781 6.435 0.347 ∗∗∗ 6.942 6.762 0.180 ∗∗∗ 7.119 7.018 0.101 ∗∗∗ EBTANAS Math 6.362 6.106 0.256 ∗∗∗ 6.043 5.621 0.422 ∗∗∗ 6.540 6.319 0.221 ∗∗∗ 6.595 6.595 0.000439 Number of students 234.2 218.2 15.97 ∗∗ 219.8 203.5 16.30 ∗∗∗ 234.5 217.6 16.88 ∗∗∗ 263.4 265.9 -2.508 Student/teacher ratio 24.32 27.28 -2.962 ∗∗∗ 15.92 16.33 -0.409 24.68 26.64 -1.962 ∗∗∗ 19.21 21.36 -2.157 ∗∗∗ Community - Junior: Number of schools 2.337 2.269 0.0681 5.625 4.982 0.643 ∗∗∗ 3.722 3.223 0.498 ∗∗∗ 3.747 3.457 0.290 ∗∗∗ Uniform expenditures 10.35 10.13 0.212 18.39 18.25 0.138 19.60 19.83 -0.229 ∗ 20.79 19.78 1.003 ∗∗∗ Distance to schools (km) 1.843 2.537 -0.694 ∗∗∗ 3.481 4.181 -0.700 ∗∗∗ 3.916 4.863 -0.947 ∗∗∗ 4.449 4.574 -0.125 EBTANAS Indonesian 6.842 6.725 0.116 ∗∗ 6.763 6.638 0.124 ∗∗∗ 5.389 5.298 0.0912 ∗∗∗ 7.373 7.272 0.101 ∗∗∗ EBTANAS Math 4.539 4.390 0.149 ∗∗ 5.193 5.091 0.101 ∗∗∗ 5.241 5.092 0.150 ∗∗∗ 6.909 6.850 0.0593 Number of students 465.8 450.4 15.39 561.9 537.4 24.48 ∗∗∗ 553.8 517.6 36.18 ∗∗∗ 523.8 504.8 18.97 ∗ Student/teacher ratio 15.27 15.45 -0.181 11.12 10.76 0.363 ∗∗ 16.38 16.43 -0.0454 14.29 15.32 -1.029 ∗∗∗ N 677 382 3623 1213 5421 1398 5782 799 Transitional: conditional on graduating primary school Land value and uniform expenditures: quartic rooted Community variables: merged at sub-district level, missing observations replaced with district/province averages 23 Figure 2.2: Primary-Jr Transitional Dropouts among 15-24 Year Olds (%) 40.7 25.8 22.6 13.3 28.2 19.1 18.1 10.8 10.0 20.0 30.0 40.0 Dropout Rate (%) 1 2 3 4 IFLS Waves Female Male Transitional: conditional on graduating primary school mother, with lower levels of education. The dropouts are also more likely to come from rural households. At the time of survey, the dropouts are more likely to have lower per capita expen- ditures, to have faced crop-loss, and to be living in a farming household with dirt floors. The dropouts are more likely to be living in communities with fewer number of schools, the distance to the schools being farther, fewer number of students per school, lower EBTANAS scores, and higher student/teacher ratios. Looking across the waves in the descriptive summary statistics, the overall changes in personal, household, and community characteristics paint a shifting picture of the dropouts over time. The gender gap appears to be closing. The proportion of females in dropouts falls from 71% in IFLS1 to 59% in IFLS4 and is similar in proportions to non-dropouts. Difference in religion is also decreasing over this time period. Compared with IFLS1, the IFLS4 dropouts appear to be in families with lower socio- economic-status (SES). The differences between the non-dropouts and dropouts have 24 widened in terms of per capita expenditures, having dirt floors, and both father’s and mother’s education levels. The huge gap between the non-dropouts and dropouts in parental education remains or is possibly even increasing; the difference in Father’s edu: Primary is at 5% (and not significantly different) in in IFLS1 but at 21% in IFLS4, and the difference in Mother’s edu: Sr/College is at 12% in IFLS1 but at 17% in IFLS4. These noticeable household characteristics are contrasted by the inconsistent differ- ences among the community variables. Characteristics such as distance to schools and EBTANAS Math have become non-significant over time for both primary and junior sec- ondary schools. Characteristics such as uniform expenditures and number of schools have become significant over time for junior secondary schools. 2.4.2 Regressions The linear probability regression results from IFLS1 to IFLS4 are shown in Table 2.6, showing the marginal effects on being a primary-junior secondary transitional dropout conditional on completing primary school. 18 Table 2.6.a represents the first model with district indicators only, and Table 2.6.b represents the second model with inclusion of community variables. The inclusion of community variables increases the R 2 of the regressions only marginally; the largest increase is in IFLS2 from 0.259 to 0.269. More importantly, the overall pattern and significance are robust to the expansion. For com- parison purposes with the latter junior-senior secondary transition and the 13-14 year olds, I will focus mainly on the second model with community characteristics in Table 2.6.b. The regressions suggest that personal and household characteristics affect the likeli- hood of dropping out, but that community characteristics do not. Being female, Muslim, 18 In a pooled regression for all four waves shown in Appendix Table A.2.2, the test of pooling for the estimates being equal across the waves is strongly rejected. 25 and being older all increase the likelihood of dropping out. 19 For consistency purposes across the waves, the Raven’s test scores are excluded in Table 2.6 since the scores are only available from IFLS3 onwards. 20 Being a member of a rural household with little land and less educated parents in- creases the likelihood of dropping out. In particular, Father’s edu: Primary coefficient estimates suggest that when compared to having fathers with no formal education, hav- ing father with even primary education decreases the probability of dropping out by 0.153 in IFLS1, 0.125 in IFL2, 0.129 in IFLS3, and 0.130 in IFLS4. Mother’s edu: Primary coefficient estimates suggest similar decreases in the probability of dropout out for having mothers with just primary education versus no formal education: 0.158 in IFLS1, 0.111 in ILFS2, 0.142 in IFLS3, and 0.114 in IFLS4. The pattern is consistent with the established findings on mother’s education positively affecting child’s educa- tion (Behrman and Wolfe, 1984; Thomas et al., 1996; Behrman et al., 1999; Behrman and Rosenzweig, 2002). The probability of dropping out decreases even further in mag- nitude with highly educated parents such that having a father with senior secondary or college education decreases the probability of dropping out by 0.329 in IFLS1, 0.271 in IFLS2, 0.236 in IFLS3, and 0.213 in IFLS4. All indicators for parental education across the waves are negative and significant at 95% confidence level. 21 19 Statistically significant in at least 2 of the four waves. 20 Using the panel structure, I attempted to backtrack and match the Raven’s scores from IFLS3 and IFLS4 to the respondents in IFLS1 and IFLS2. However, the success rates of matching were only 17% in IFLS1 and 62% in IFLS2. Appendix Table A.2.3 shows the regression results for IFLS3 and IFLS4, with and without the Raven’s test scores. As expected, the Raven’s test scores contribute to dropping out when included, but the signs and significance levels remain unchanged for the most part. 21 While coefficient estimates for Father’s edu: Jr and Father’s edu: Sr/College are larger than for Father’s edu: Primary, I focus on the partial effect change from Father’s edu: Noneto Father’s edu: Primary because these two groups represent the vast majority of father’s education level for dropouts. For IFLS1 dropouts, 35.9 % had fathers with no education and 59.6% had fathers with only primary education. For IFLS4 dropouts, 22% had fathers with no education and 72% had fathers with only primary education. 26 Table 2.6: Linear Probability Regression for Primary School Graduates among 15-24 Year Olds y i = 1 if drops out after primary school 0 if continues to junior secondary school (a) w/ District Indicators Only IFLS1 IFLS2 IFLS3 IFLS4 Personal: Male -0.100 ∗∗∗ (-3.51) -0.0467 ∗∗ (-3.03) -0.0379 ∗∗∗ (-4.07) -0.00853 (-1.32) Non-muslim -0.169 ∗∗ (-2.99) -0.0604 ∗ (-2.16) -0.0379 ∗ (-2.09) -0.0279 ∗ (-2.46) Age=16 -0.107 (-1.46) -0.00114 (-0.05) -0.0274 (-1.66) 0.0365 ∗ (2.49) Age=17 -0.0103 (-0.15) 0.0147 (0.62) -0.00327 (-0.23) 0.0317 (1.94) Age=18 -0.0371 (-0.49) 0.0142 (0.65) 0.00437 (0.25) 0.0514 ∗∗ (3.06) Age=19 -0.0373 (-0.47) 0.0454 (1.80) 0.0231 (1.27) 0.0476 ∗∗ (2.70) Age=20 -0.0310 (-0.47) 0.0782 ∗∗ (2.64) 0.0789 ∗∗∗ (4.27) 0.0366 ∗ (2.51) Age=21 -0.0284 (-0.43) 0.0746 ∗∗ (2.96) 0.102 ∗∗∗ (5.03) 0.0644 ∗∗∗ (3.63) Age=22 -0.0124 (-0.20) 0.0796 ∗∗ (2.72) 0.113 ∗∗∗ (5.09) 0.0649 ∗∗∗ (3.99) Age=23 -0.0297 (-0.45) 0.0226 (0.79) 0.0991 ∗∗∗ (5.16) 0.0653 ∗∗∗ (3.56) Age=24 -0.0647 (-1.04) 0.0690 ∗ (2.02) 0.125 ∗∗∗ (5.86) 0.0766 ∗∗∗ (3.89) Household: Rural 0.175 ∗∗∗ (4.03) 0.0917 ∗∗ (3.11) 0.0842 ∗∗∗ (4.08) 0.0618 ∗∗∗ (4.24) Land value -0.000772 (-1.92) -0.000685 ∗∗∗ (-4.14) -0.000493 ∗∗∗ (-4.19) -0.000383 ∗∗ (-2.84) Father’s edu: Primary -0.165 ∗∗∗ (-3.56) -0.131 ∗∗∗ (-4.21) -0.134 ∗∗∗ (-5.54) -0.129 ∗∗∗ (-4.94) Father’s edu: Jr -0.294 ∗∗∗ (-5.08) -0.246 ∗∗∗ (-6.12) -0.227 ∗∗∗ (-8.78) -0.196 ∗∗∗ (-6.88) Father’s edu: Sr/College -0.372 ∗∗∗ (-5.82) -0.277 ∗∗∗ (-7.08) -0.245 ∗∗∗ (-9.25) -0.210 ∗∗∗ (-7.57) Mother’s edu: Primary -0.172 ∗∗∗ (-3.73) -0.118 ∗∗∗ (-6.55) -0.140 ∗∗∗ (-7.79) -0.114 ∗∗∗ (-5.36) Mother’s edu: Jr -0.185 ∗∗ (-2.97) -0.168 ∗∗∗ (-8.31) -0.181 ∗∗∗ (-8.91) -0.149 ∗∗∗ (-7.11) Mother’s edu: Sr/College -0.170 ∗∗ (-2.65) -0.160 ∗∗∗ (-7.26) -0.174 ∗∗∗ (-8.68) -0.134 ∗∗∗ (-6.46) District indicators Yes Yes Yes Yes Observations 990 4719 6672 6417 AdjustedR 2 0.303 0.259 0.261 0.167 Joint test for age 0.933 0.014 0.000 0.003 Joint test for father’s edu 0.000 0.000 0.000 0.000 Joint test for mother’s edu 0.003 0.000 0.000 0.000 Dropout rate 0.361 0.251 0.205 0.121 t statistics in parentheses Robust, cluster(district) ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 From the community characteristics, a variable possibly worth noting is the stu- dent/teacher ratio; it is non-significant at primary level and only significant at the junior secondary level in IFLS4. Hanushek (1986) demonstrates that student/teacher ratio is generally not associated with learning or student outcomes. Although student/teacher ratio is a topic of much debate and discussion in the education literature, its impact on helping primary students to transition to the next level may be limited in Indonesia. 27 Table 2.6: Linear Probability Regression for Primary School Graduates among 15-24 Year Olds (cont.) y i = 1 if drops out after primary school 0 if continues to junior secondary school (b) w/ Community Characteristics IFLS1 IFLS2 IFLS3 IFLS4 Personal: Male -0.111 ∗∗∗ (-3.52) -0.0489 ∗∗ (-3.21) -0.0365 ∗∗∗ (-3.99) -0.00853 (-1.36) Non-muslim -0.183 ∗∗ (-2.62) -0.0419 (-1.84) -0.0326 ∗ (-1.99) -0.0290 ∗ (-2.39) Age=16 -0.0902 (-1.22) 0.00275 (0.13) -0.0295 (-1.80) 0.0356 ∗ (2.43) Age=17 -0.00406 (-0.06) 0.0119 (0.48) -0.00236 (-0.16) 0.0297 (1.81) Age=18 0.000406 (0.01) 0.0163 (0.72) 0.00381 (0.22) 0.0503 ∗∗ (3.05) Age=19 -0.0210 (-0.25) 0.0447 (1.75) 0.0221 (1.21) 0.0495 ∗∗ (2.87) Age=20 -0.0526 (-0.76) 0.0762 ∗ (2.52) 0.0791 ∗∗∗ (4.10) 0.0395 ∗∗ (2.69) Age=21 -0.0237 (-0.36) 0.0717 ∗∗ (2.72) 0.104 ∗∗∗ (5.17) 0.0649 ∗∗∗ (3.71) Age=22 -0.0253 (-0.38) 0.0758 ∗∗ (2.70) 0.117 ∗∗∗ (5.27) 0.0655 ∗∗∗ (3.98) Age=23 -0.0306 (-0.44) 0.0224 (0.76) 0.0988 ∗∗∗ (5.18) 0.0633 ∗∗∗ (3.53) Age=24 -0.0512 (-0.80) 0.0678 ∗ (2.08) 0.123 ∗∗∗ (5.77) 0.0738 ∗∗∗ (3.56) Household: Rural 0.139 ∗∗ (2.83) 0.0530 (1.81) 0.0664 ∗∗ (2.78) 0.0579 ∗∗∗ (4.21) Land value -0.000921 ∗ (-2.17) -0.000707 ∗∗∗ (-4.58) -0.000555 ∗∗∗ (-4.63) -0.000363 ∗ (-2.60) Father’s edu: Primary -0.153 ∗∗ (-3.16) -0.125 ∗∗∗ (-4.15) -0.129 ∗∗∗ (-5.34) -0.130 ∗∗∗ (-5.01) Father’s edu: Jr -0.271 ∗∗∗ (-4.19) -0.240 ∗∗∗ (-6.36) -0.223 ∗∗∗ (-8.67) -0.197 ∗∗∗ (-6.95) Father’s edu: Sr/College -0.329 ∗∗∗ (-4.93) -0.271 ∗∗∗ (-7.11) -0.236 ∗∗∗ (-9.07) -0.213 ∗∗∗ (-7.77) Mother’s edu: Primary -0.158 ∗∗ (-3.23) -0.111 ∗∗∗ (-6.02) -0.142 ∗∗∗ (-7.85) -0.114 ∗∗∗ (-5.41) Mother’s edu: Jr -0.200 ∗∗ (-2.76) -0.158 ∗∗∗ (-7.61) -0.179 ∗∗∗ (-8.86) -0.146 ∗∗∗ (-6.64) Mother’s edu: Sr/College -0.127 (-1.82) -0.153 ∗∗∗ (-6.89) -0.175 ∗∗∗ (-8.76) -0.129 ∗∗∗ (-5.87) Community - Primary: Number of schools -0.00411 (-0.23) 0.00265 (0.58) -0.00260 (-0.53) 0.0113 ∗∗ (3.06) Uniform expenditures -0.00334 (-0.68) -0.00209 (-0.82) 0.000194 (0.05) -0.0000911 (-0.07) Distance to schools (km) 0.0280 (1.13) -0.00558 (-0.94) 0.01000 (1.60) -0.000109 (-0.05) EBTANAS Indonesian -0.0583 (-0.98) -0.0296 ∗∗∗ (-4.76) -0.0142 (-0.57) 0.0135 (1.20) EBTANAS Math 0.0552 (1.53) -0.0129 (-1.11) -0.0123 (-0.73) 0.00191 (0.21) Number of students -0.000307 (-0.95) -0.000117 (-0.95) -0.0000272 (-0.24) -0.0000425 (-0.30) Student/teacher ratio 0.00658 (1.81) 0.00142 (0.97) 0.00150 (1.00) 0.00288 (1.56) Community - Junior: Number of schools -0.00424 (-0.19) -0.0176 ∗∗∗ (-3.73) -0.00629 (-0.96) -0.0119 ∗ (-2.29) Uniform expenditures -0.00434 (-0.93) 0.00113 (0.52) 0.00292 (0.98) 0.000426 (0.25) Distance to schools (km) 0.00291 (0.16) 0.00714 (1.44) 0.00256 (0.65) -0.00112 (-0.51) EBTANAS Indonesian -0.0410 (-0.78) -0.0274 (-1.11) 0.00240 (0.09) -0.0146 (-1.35) EBTANAS Math 0.0280 (0.75) 0.0100 (0.49) 0.0317 (1.00) 0.00655 (0.59) Number of students 0.0000328 (0.21) 0.0000455 (0.48) -0.000125 ∗ (-2.24) -0.0000757 ∗ (-2.27) Student/teacher ratio -0.00158 (-0.22) -0.00663 (-1.34) -0.00249 (-0.85) 0.00572 ∗∗ (2.98) District indicators Yes Yes Yes Yes Missing indicators Yes Yes Yes Yes Observations 909 4583 6620 6326 AdjustedR 2 0.297 0.269 0.269 0.171 Joint test for age 0.946 0.028 0.000 0.003 Joint test for father’s edu 0.000 0.000 0.000 0.000 Joint test for mother’s edu 0.011 0.000 0.000 0.000 Joint test for community: Pr 0.194 0.000 0.167 0.001 Joint test for community: Jr 0.923 0.004 0.003 0.010 Joint test for community 0.397 0.000 0.000 0.000 Dropout rate 0.361 0.251 0.205 0.121 t statistics in parentheses Robust, cluster(district) ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 28 2.4.2.1 Changes from IFLS1 to IFLS4 Looking across the waves in Table 2.6.b, the magnitudes of the marginal effects are de- creasing from IFLS1 to IFLS4. The male coefficient estimates dwindle from -0.111 in IFLS1 to -0.0489 in IFLS2 to -0.0365 in IFLS3 to -0.00853 in IFLS4, but more impor- tantly, the estimate is no longer significant in IFLS4. By IFLS4, gender does not matter in the decision to transition to junior secondary school. Other personal and household variables show similar drastic reductions in their effects: non-muslim reduces from - 0.183 in IFLS1 to -0.00290 in IFLS4 and rural from 0.139 to 0.0579. The magnitudes of the coefficient estimates for parental education are decreasing but to a smaller degree than personal variables: the largest reduction is about 35% for Father’s edu: Sr/College from -0.329 in IFLS1 to -0.213 in IFLS4. The negative significance for rural indicator and land value suggest that if born into a rural household with less wealth, particularly with low parental education, the child would be at greater risk of not advancing to lower secondary school. The implica- tion corresponds with the vast literature on educational attainment and socio-economic- status, particularly of parental education and income, in both education and economic literature such as Rumberger (1987), Chuang (1994), and Brown (2006), as well as the findings from IFLS2 & IFLS3 in Strauss et al. (2004). 2.4.2.2 Oaxaca-Blinder Decomposition While the regression analysis using the IFLS as independent cross sections cannot iden- tify a time trend, I believe the society’s educational infrastructure may be contributing to the overall decrease in transitional dropouts from 1993 to 2007. Although the com- munity characteristics, neither at the primary nor the junior secondary level individually, show much significance, all of the community characteristics jointly are significant from IFLS2 onward. 29 I check to see what proportion of the difference in dropout rates across the waves can be captured by differences in the socio-economic-status characteristics across IFLS1 and IFLS4. The standard three-fold Oaxaca-Blinder decomposition method is applied to the linear probability model in Table 2.6.b (Oaxaca, 1973; Blinder, 1973). 22 Table 2.7 shows the summary decompositions of the gap in the dropout rates from IFLS1 to IFLS4 for personal, household, and community characteristics. Of household characteristics, rural, and parental education (levels jointly) are highlighted. Of the 25.7 percentage point difference in predicted dropout rates from IFLS1 and IFLS4, about 92% of the difference can be explained by the characteristics or endow- ments: should IFLS1 have had IFLS4 endowments, the dropout rates in IFLS1 would decrease by 23.6 percentage points. The characteristics or endowments are theexplained part in the two-fold decomposition; since 92% of the difference is considered to be ex- plained by endowments, the decomposition suggests that the dropouts in IFLS1 and IFLS4 are characteristically different. Breakdown of the endowments indicates that parental education plays a large role in the explainable differences: father’s and mother’s education combined account for 35% (-0.0828/-0.236) of the explainable differences. An unexpected outcome of the endowment decomposition is the importance of com- munity characteristics. From the regressions in Table 2.6, the community school char- acteristics, neither at the primary or junior secondary, seemed to have effect on the de- cision to transition. However, the joint test for community for both primary and junior secondary schools did seem to have an overall effect throughout the waves. While there is no way to identify the extent of the effect in the regressions, the decompositions sug- gests that change in dropout rates from IFLS1 to IFLS4 is largely influenced by changes in community characteristics. Should IFLS1 have had IFLS4 community primary and junior secondary school characteristics, the dropout rate would change 13.5 percentage 22 Via Stata command oaxaca, following the method outlined in Jann (2008). 30 Table 2.7: Oaxaca Decomposition between IFLS4 & IFLS1 Primary-Junior Secondary Transitional Dropouts Expanded Differential Prediction 1 0.118 (0.00407) Prediction 2 0.375 (0.0163) Difference -0.257 (0.0168) Endowments Personal -0.00949 (0.00500) Rural -0.00379 (0.00253) Household other 0.00320 (0.00203) Father edu -0.0546 (0.00923) Mother edu -0.0282 (0.0100) Community -0.135 (0.0842) Missing -0.00768 (0.00814) Total -0.236 (0.0846) Coefficients Personal 0.116 (0.0514) Rural -0.00486 (0.0215) Household other 0.0101 (0.0123) Father edu 0.0783 (0.0374) Mother edu 0.0154 (0.0276) Community -0.0509 (0.277) Missing -0.00708 (0.0201) Constant -0.303 (0.295) Total -0.146 (0.0294) Interaction Personal 0.00711 (0.00501) Rural 0.000506 (0.00224) Household other -0.00152 (0.00191) Father edu 0.0240 (0.00859) Mother edu 0.00373 (0.0102) Community 0.0879 (0.0879) Missing 0.00254 (0.00841) Total 0.124 (0.0880) Observations 7235 Endowments/Difference (%) 91.69 Coefficients/Difference (%) 56.66 Prediction 1=IFLS4, Prediction 2=IFLS1 Personal: male, non-muslim, age indicators Household other: land value Father edu: primary, jr, sr/college Mother edu: primary, jr, sr/college Community: number of schools, uniform expenditures, distance to school, EBTANAS, number of students, student/teacher ratio (at primary and junior secondary levels) Missing: missing indicators for community characteristics points or would account for 57% (-0.135/-0.236) of explainable differences via endow- ments. One possible explanation that could be consistent with the overall educational progress in Indonesia may simply be that the availability of schools, both in quantity 31 and quality, have improved greatly for the Indonesian population. More schools were accessible, more teachers, fewer students per teacher, better test scores, and so forth all may have contributed to more students attending junior secondary school. Similarly, of the 25.7 percentage point difference in dropout rates from IFLS1 to IFLS4, about 57% of the difference can be explained by the coefficients: should IFLS1 have had IFLS4 coefficients, the dropout rates in IFLS1 would decrease by 15 percent- age points. Breakdown of the coefficients however reveals that for most of decompo- sition estimates are positive, suggesting that for most of the independent variables in IFLS1, applying IFLS4 coefficients would increase the dropout rate. The large nega- tive comes from the constant term, suggesting that the overall average dropout level in IFLS4 is significantly lower. 2.5 Junior-Senior Secondary Transitional Dropouts Figure 2.1 illustrates that junior-senior secondary transitional dropouts are increasing over time, and this trend may strengthen the case to examine junior-senior secondary transition in more detail. However, Table 2.3 shows an alarmingly low percentage of respondents complete junior secondary or higher in IFLS1: below 40% for females and below 50% for males. Thus, any results from the junior-senior secondary transition are likely to be heavily biased from selection. More importantly, the junior-senior secondary transition should not be considered directly comparable to the primary-junior secondary transition. Cost of attending senior secondary school is even greater in both time and money than attending junior secondary school. Young men are able to work in the labor market rather than merely helping to farm as children; young women are able to be mar- ried off or to work full time within the household. With the mandate of junior secondary education, increasing enrollment at senior secondary and tertiary levels are certainly 32 intended, even if not directly targeted. The overall increase in educational attainment from 1997-2007 is evident via IFLS as previously shown in Table 2.3. However, since the increase in junior secondary enrollment caused by the mandate cannot be separately identified from the time trend, the increase in senior secondary enrollment caused by the mandate can only be estimated with even less certainty. Still, with all the differences between the primary-junior secondary transition and the junior-senior secondary transition, the junior-senior secondary transitional dropouts are analyzed cautiously in this section because the pattern for junior-senior secondary transitional dropouts is worth comparing to that of primary-junior secondary transitional dropouts. 23 Figure 2.3: Jr-Sr Transitional Dropouts among 15-24 Year Olds (%) 36.4 26.1 27.0 25.4 33.3 23.9 23.6 19.3 20.0 25.0 30.0 35.0 40.0 Dropout Rate (%) 1 2 3 4 IFLS Waves Female Male Transitional: conditional on graduating junior secondary school 23 I proceed with caution noting that the conditions for junior-senior secondary transition needs to be more carefully considered in order to be directly compared with the primary-junior secondary transition. Endogenous selection aside, other independent variables may have to be adjusted for the junior secondary school graduates: age distribution may have to be moved up by 3 years or geographical indicators may have to be the location at age 15. 33 The junior-senior secondary transitional dropouts across the waves are shown in Figure 2.3. The proportion of dropouts decreases from 36% in IFLS1 to 25% in IFLS4 for females and from 33% to 19% for males. Between IFLS1 (1993) and IFLS2 (1997), there is almost a 10 percentage point drop for both males and females, but the proportion of female junior secondary school graduates who pursue senior secondary education remains stagnant from 1997 to 2007. 2.5.1 Descriptive Statistics Separating the non-dropouts from the dropouts yields the descriptive statistics shown in Table 2.8. Within the waves, the non-dropouts again display clearly different character- istics than the dropouts. Among the personal characteristics, all the personal characteristics are strongly sig- nificantly different by IFLS4. By IFLS4, the pattern is identical to that of primary-junior secondary dropouts; essentially all the variables are statistically different between the non-dropouts and the dropouts. Looking across the waves in the descriptive summary statistics, the differences be- tween the non-dropouts and dropouts are increasing at larger magnitudes than in the primary-junior secondary summary statistics in Table 2.5. The differences in parental education are particularly noteworthy. In IFLS1, the differences between non-dropouts and dropouts for Father’s edu: Primary and Mother’s edu: Primary are 19% and 8%, respectively; by IFLS4, the differences are 30% and 23%. At the time of survey, the gap between the non-dropouts and dropouts are also in- creasing from IFLS1 to IFLS4 for log PCE and dirt floors. As for community character- istics, the gap between the non-dropouts are dropouts are increasing overall. For junior secondary schools, only EBTANAS Indonesian is significantly different in IFLS1. By 34 Table 2.8: Summary Statistics for Junior Secondary School Graduates among 15-24 Year Olds IFLS1 IFLS2 IFLS3 IFLS4 ND Dropout Diff ND Dropout Diff ND Dropout Diff ND Dropout Diff Personal: Male 0.411 0.379 0.0322 0.485 0.455 0.0299 0.482 0.437 0.0448 ∗∗ 0.468 0.382 0.0861 ∗∗∗ Non-muslim 0.173 0.131 0.0413 0.157 0.0931 0.0643 ∗∗∗ 0.153 0.0782 0.0748 ∗∗∗ 0.132 0.0621 0.0699 ∗∗∗ Raven’s test (z-score) 0.333 -0.0311 0.364 ∗∗∗ 0.297 -0.0568 0.354 ∗∗∗ Household: Rural 0.271 0.449 -0.178 ∗∗∗ 0.279 0.517 -0.238 ∗∗∗ 0.305 0.557 -0.252 ∗∗∗ 0.350 0.573 -0.223 ∗∗∗ Log PCE 13.14 12.83 0.306 ∗∗∗ 13.06 12.68 0.379 ∗∗∗ 13.08 12.74 0.332 ∗∗∗ 13.18 12.76 0.418 ∗∗∗ Faced crop loss 0.0329 0.0808 -0.0479 ∗ 0.0681 0.138 -0.0696 ∗∗∗ 0.0615 0.131 -0.0694 ∗∗∗ 0.0389 0.0735 -0.0346 ∗∗∗ Dirt floor 0.0192 0.0606 -0.0414 ∗∗ 0.0274 0.112 -0.0843 ∗∗∗ 0.0249 0.117 -0.0926 ∗∗∗ 0.0215 0.0936 -0.0721 ∗∗∗ Land value 25.24 31.13 -5.897 32.26 36.66 -4.401 ∗ 31.94 32.90 -0.967 27.68 25.44 2.240 Farming household 0.217 0.338 -0.121 ∗∗ 0.207 0.400 -0.193 ∗∗∗ 0.259 0.352 -0.0923 ∗∗∗ 0.251 0.288 -0.0367 ∗ Father’s edu: None 0.0294 0.173 -0.144 ∗∗∗ 0.0314 0.114 -0.0824 ∗∗∗ 0.0326 0.124 -0.0916 ∗∗∗ 0.0324 0.110 -0.0778 ∗∗∗ Father’s edu: Primary 0.450 0.643 -0.193 ∗∗∗ 0.429 0.673 -0.243 ∗∗∗ 0.430 0.692 -0.262 ∗∗∗ 0.427 0.729 -0.302 ∗∗∗ Father’s edu: Jr 0.159 0.0865 0.0723 ∗ 0.183 0.131 0.0519 ∗∗∗ 0.177 0.111 0.0664 ∗∗∗ 0.169 0.103 0.0658 ∗∗∗ Father’s edu: Sr/College 0.362 0.0973 0.264 ∗∗∗ 0.356 0.0822 0.274 ∗∗∗ 0.360 0.0732 0.287 ∗∗∗ 0.372 0.0573 0.314 ∗∗∗ Mother’s edu: None 0.130 0.290 -0.160 ∗∗∗ 0.0855 0.215 -0.130 ∗∗∗ 0.0676 0.230 -0.163 ∗∗∗ 0.0654 0.177 -0.112 ∗∗∗ Mother’s edu: Primary 0.522 0.602 -0.0804 0.520 0.681 -0.161 ∗∗∗ 0.536 0.688 -0.152 ∗∗∗ 0.511 0.736 -0.225 ∗∗∗ Mother’s edu: Jr 0.162 0.0538 0.109 ∗∗∗ 0.183 0.0760 0.107 ∗∗∗ 0.179 0.0567 0.122 ∗∗∗ 0.177 0.0665 0.110 ∗∗∗ Mother’s edu: Sr/College 0.186 0.0538 0.132 ∗∗∗ 0.212 0.0279 0.184 ∗∗∗ 0.217 0.0246 0.193 ∗∗∗ 0.247 0.0195 0.227 ∗∗∗ Community - Junior: Number of schools 2.306 2.318 -0.0119 5.670 5.475 0.195 ∗∗ 3.892 3.724 0.169 ∗∗∗ 3.815 3.648 0.167 ∗∗∗ Uniform expenditures 10.99 10.63 0.362 18.72 18.71 0.0174 19.90 19.31 0.590 ∗∗∗ 20.90 20.32 0.585 ∗∗∗ Distance to schools (km) 1.712 1.911 -0.199 3.195 3.715 -0.520 ∗∗∗ 3.638 3.934 -0.295 ∗∗∗ 4.175 4.218 -0.0433 EBTANAS Indonesian 6.926 6.821 0.104 ∗ 6.802 6.726 0.0758 ∗∗∗ 5.425 5.394 0.0306 7.431 7.341 0.0896 ∗∗∗ EBTANAS Math 4.590 4.475 0.116 5.206 5.169 0.0370 5.295 5.201 0.0943 ∗∗∗ 6.982 6.862 0.121 ∗∗∗ Number of students 477.0 465.8 11.20 568.2 550.7 17.51 ∗ 565.1 554.1 11.06 537.8 508.7 29.14 ∗∗∗ Student/teacher ratio 15.48 15.34 0.144 11.35 11.04 0.308 ∗∗ 16.47 16.61 -0.138 14.17 14.61 -0.436 ∗∗∗ Community - Senior: Number of schools 2.019 1.948 0.0712 5.469 5.012 0.456 ∗∗∗ 4.556 4.233 0.323 ∗∗∗ 4.320 4.025 0.296 ∗∗∗ Uniform expenditures 12.24 11.86 0.389 19.77 19.52 0.246 21.01 20.58 0.437 ∗∗∗ 22.12 21.04 1.081 ∗∗∗ Distance to schools (km) 2.094 2.583 -0.488 ∗∗ 4.472 4.925 -0.453 ∗∗∗ 4.952 5.628 -0.676 ∗∗∗ 5.243 5.782 -0.539 ∗∗∗ EBTANAS Indonesian 6.279 6.174 0.104 6.547 6.457 0.0899 5.200 5.107 0.0932 ∗∗∗ 7.078 7.030 0.0479 ∗ EBTANAS Math 3.916 3.644 0.272 ∗ 3.789 3.862 -0.0732 3.504 3.405 0.0991 ∗∗∗ 6.973 6.790 0.183 ∗∗∗ Number of students 520.5 500.8 19.71 599.9 573.5 26.43 ∗∗ 570.6 555.1 15.42 ∗ 566.5 521.7 44.89 ∗∗∗ Student/teacher ratio 12.75 12.29 0.458 10.48 10.16 0.316 ∗ 16.44 16.50 -0.0638 12.26 12.45 -0.199 N 365 198 2409 806 3495 1192 3909 1143 Transitional: conditional on graduating junior secondary school Land value and uniform expenditures: quartic rooted Community variables: merged at sub-district level, missing observations replaced with district/province averages 35 IFLS4, all but distance to schools are significantly different. Similarly for senior sec- ondary schools, only EBTANAS Math and distance to schools are significantly different in IFLS1. By IFLS4, all but student/teacher ratio are significantly different. 2.5.2 Regressions The linear probability regression results from IFLS1 to IFLS4 are shown in Table 2.9, showing the marginal effects on being a junior-senior secondary transitional dropout conditional on completing junior secondary school. 24 Similar to the regression results for primary-junior secondary transition in Table 2.6, the first model with district indica- tors only in Table 2.9.a and the second model with community characteristics in Table 2.9.b are similar in overall pattern and significance. Hence, I will again mainly discuss the second model in Table 2.9.b. The personal and household characteristics affect the likelihood of dropping out much more so than community characteristics. Females have a lower probability of continuing to senior secondary after junior secondary than males conditional on com- pleting junior secondary school. Children with better educated parents are less likely to drop out. These results are consistent with the education literature on post-compulsory dropouts in the developed countries. Bradley and Lenton (2007) report that having par- ents with manegeraial/professional occupation increases the likelihood of continuing onto higher education in the United Kingdom. Similary, in the United States, hav- ing higher educated mothers increases the likelihood of enrolling in colleges (McElroy, 1996). 24 In the pooled regression for all four waves in Appendix Table A.2.4, the test of pooling for the estimates being equal across the waves is again strongly rejected. 36 Table 2.9: Linear Probability Regression for Junior Secondary School Graduates among 15-24 Year Olds y i = 1 if drops out after junior secondary school 0 if continues to senior secondary school (a) w/ District Indicators Only IFLS1 IFLS2 IFLS3 IFLS4 Personal: Male -0.0680 (-1.62) -0.0289 (-1.82) -0.0284 ∗ (-2.09) -0.0423 ∗∗∗ (-3.71) Non-muslim 0.0105 (0.14) -0.00922 (-0.45) -0.0582 ∗∗ (-2.96) -0.0492 ∗∗ (-2.97) Age=16 -0.349 ∗ (-2.20) -0.102 ∗∗ (-2.84) -0.0477 (-1.70) -0.0656 ∗ (-2.13) Age=17 -0.467 ∗∗ (-3.17) -0.130 ∗∗∗ (-4.38) -0.0255 (-0.83) -0.00647 (-0.24) Age=18 -0.483 ∗∗ (-3.30) -0.110 ∗∗ (-3.31) -0.0176 (-0.66) 0.0177 (0.67) Age=19 -0.457 ∗∗∗ (-3.41) -0.0958 ∗∗ (-2.85) -0.0196 (-0.71) 0.0130 (0.48) Age=20 -0.264 (-1.79) -0.0414 (-1.13) -0.0111 (-0.37) 0.0189 (0.68) Age=21 -0.469 ∗∗ (-3.19) -0.0950 ∗∗ (-2.64) -0.0410 (-1.25) 0.00774 (0.28) Age=22 -0.573 ∗∗ (-3.41) -0.0596 (-1.42) -0.0226 (-0.68) 0.0422 (1.39) Age=23 -0.359 ∗ (-2.54) -0.0207 (-0.50) 0.00405 (0.13) 0.0549 (1.97) Age=24 -0.555 ∗∗∗ (-3.94) 0.00786 (0.22) -0.0279 (-0.89) 0.0897 ∗∗ (2.91) Household: Rural 0.0917 (0.98) 0.104 ∗∗∗ (3.93) 0.0746 ∗∗∗ (3.43) 0.106 ∗∗∗ (3.45) Land value 0.000145 (0.22) -0.000141 (-0.79) -0.000383 ∗ (-2.45) -0.000544 ∗∗∗ (-3.51) Father’s edu: Primary -0.175 (-1.68) -0.114 ∗ (-2.25) -0.0932 ∗∗ (-2.66) -0.104 ∗ (-2.33) Father’s edu: Jr -0.330 ∗∗ (-2.64) -0.191 ∗∗∗ (-3.94) -0.183 ∗∗∗ (-4.52) -0.219 ∗∗∗ (-4.27) Father’s edu: Sr/College -0.475 ∗∗∗ (-3.66) -0.258 ∗∗∗ (-5.25) -0.229 ∗∗∗ (-5.93) -0.255 ∗∗∗ (-4.86) Mother’s edu: Primary -0.0569 (-0.71) -0.0706 ∗ (-1.99) -0.162 ∗∗∗ (-6.54) -0.117 ∗∗∗ (-4.61) Mother’s edu: Jr -0.127 (-1.27) -0.155 ∗∗ (-3.20) -0.241 ∗∗∗ (-8.06) -0.211 ∗∗∗ (-6.71) Mother’s edu: Sr/College -0.108 (-1.04) -0.188 ∗∗∗ (-4.21) -0.259 ∗∗∗ (-8.58) -0.227 ∗∗∗ (-7.28) District indicators Yes Yes Yes Yes Observations 521 3122 4617 4948 AdjustedR 2 0.172 0.185 0.216 0.186 Joint test for age 0.002 0.000 0.557 0.000 Joint test for father’s edu 0.000 0.000 0.000 0.000 Joint test for mother’s edu 0.645 0.000 0.000 0.000 Dropout rate 0.352 0.251 0.254 0.226 t statistics in parentheses Robust, cluster(district) ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 2.5.2.1 Changes from IFLS1 to IFLS4 Looking across the waves in Table 2.9.b, the changes from IFLS1 to IFLS4 appear to be visibly different than the changes of the primary-junior secondary dropouts in Section 2.4.2. In IFLS1, age appears to stand out as an important determinant in dropping out at the junior-senior secondary transition; by IFLS4, most age indicators have lost their significance. In contrast, age seems to matter in IFLS4, but not in IFLS1, for the 37 Table 2.9: Linear Probability Regression for Junior Secondary School Graduates among 15-24 Year Olds (cont.) y i = 1 if drops out after junior secondary school 0 if continues to senior secondary school (b) w/ Community Characteristics IFLS1 IFLS2 IFLS3 IFLS4 Personal: Male -0.0936 ∗ (-2.01) -0.0186 (-1.07) -0.0292 ∗ (-2.21) -0.0399 ∗∗∗ (-3.44) Non-muslim 0.00278 (0.03) -0.0144 (-0.58) -0.0563 ∗∗ (-3.08) -0.0522 ∗∗ (-2.93) Age=16 -0.295 (-1.80) -0.105 ∗∗ (-2.79) -0.0421 (-1.50) -0.0670 ∗ (-2.18) Age=17 -0.512 ∗∗∗ (-3.43) -0.161 ∗∗∗ (-5.01) -0.0175 (-0.56) -0.00835 (-0.31) Age=18 -0.471 ∗∗ (-3.09) -0.130 ∗∗∗ (-3.67) -0.0121 (-0.44) 0.0149 (0.56) Age=19 -0.532 ∗∗∗ (-3.87) -0.109 ∗∗ (-3.12) -0.0103 (-0.37) 0.0105 (0.38) Age=20 -0.332 ∗ (-2.21) -0.0400 (-1.00) -0.00591 (-0.19) 0.0201 (0.72) Age=21 -0.520 ∗∗ (-3.37) -0.109 ∗∗ (-2.67) -0.0331 (-0.98) 0.0106 (0.38) Age=22 -0.611 ∗∗∗ (-3.52) -0.0508 (-1.18) -0.0138 (-0.41) 0.0446 (1.47) Age=23 -0.392 ∗∗ (-2.73) -0.0236 (-0.54) 0.0101 (0.31) 0.0534 (1.91) Age=24 -0.613 ∗∗∗ (-4.29) -0.0185 (-0.45) -0.0182 (-0.56) 0.0878 ∗∗ (2.85) Household: Rural 0.114 (1.08) 0.107 ∗∗ (2.96) 0.0760 ∗∗ (3.25) 0.0833 ∗∗ (3.14) Land value 0.000155 (0.23) -0.000236 (-1.16) -0.000393 ∗ (-2.50) -0.000615 ∗∗∗ (-3.83) Father’s edu: Primary -0.215 (-1.98) -0.104 (-1.70) -0.0939 ∗∗ (-2.68) -0.0974 ∗ (-2.36) Father’s edu: Jr -0.415 ∗∗ (-2.97) -0.180 ∗∗ (-2.98) -0.185 ∗∗∗ (-4.54) -0.214 ∗∗∗ (-4.42) Father’s edu: Sr/College -0.516 ∗∗∗ (-3.63) -0.247 ∗∗∗ (-4.13) -0.228 ∗∗∗ (-5.82) -0.251 ∗∗∗ (-5.12) Mother’s edu: Primary -0.00888 (-0.09) -0.0801 ∗ (-2.02) -0.161 ∗∗∗ (-6.53) -0.114 ∗∗∗ (-4.78) Mother’s edu: Jr -0.142 (-1.29) -0.174 ∗∗∗ (-3.48) -0.241 ∗∗∗ (-7.54) -0.207 ∗∗∗ (-6.90) Mother’s edu: Sr/College -0.0641 (-0.54) -0.207 ∗∗∗ (-4.66) -0.255 ∗∗∗ (-8.10) -0.220 ∗∗∗ (-7.01) Community - Junior: Number of schools 0.0218 (0.60) -0.00406 (-0.43) 0.0137 (1.11) -0.000400 (-0.05) Uniform expenditures 0.0106 (1.39) 0.00194 (0.97) -0.00495 (-1.12) 0.00432 (1.01) Distance to schools (km) -0.0639 (-1.52) -0.00476 (-0.66) 0.000583 (0.11) -0.00157 (-0.64) EBTANAS Indonesian 0.0872 (1.07) -0.0293 (-0.76) -0.0399 (-1.17) -0.0136 (-0.69) EBTANAS Math -0.0649 (-1.11) 0.00350 (0.13) -0.000332 (-0.01) 0.00403 (0.34) Number of students 0.000362 (1.52) -0.0000422 (-0.40) -0.000135 (-0.65) -0.0000427 (-0.92) Student/teacher ratio -0.0169 (-1.17) 0.000740 (0.15) 0.0149 (0.95) 0.00461 (1.40) Community - Senior: Number of schools -0.00766 (-0.10) -0.00932 (-1.32) -0.0179 ∗ (-2.19) -0.00814 (-1.37) Uniform expenditures -0.00313 (-0.57) -0.00255 (-1.40) 0.000933 (0.20) -0.00372 ∗ (-2.38) Distance to schools (km) 0.0216 (0.70) -0.000609 (-0.08) -0.00436 (-1.19) 0.00558 ∗ (2.48) EBTANAS Indonesian -0.0720 (-1.02) -0.000111 (-0.04) -0.00223 (-0.11) 0.00529 (0.36) EBTANAS Math -0.0124 (-0.30) 0.0104 (0.94) 0.00409 (0.26) -0.0175 ∗ (-2.21) Number of students -0.000169 (-1.21) -0.00000860 (-0.12) 0.000174 (0.95) -0.0000180 (-0.38) Student/teacher ratio 0.00229 (0.24) -0.000285 (-0.05) -0.0156 (-1.07) 0.000746 (0.33) District indicators Yes Yes Yes Yes Missing indicators Yes Yes Yes Yes Observations 463 2652 4576 4877 AdjustedR 2 0.173 0.193 0.219 0.190 Joint test for age 0.000 0.000 0.613 0.000 Joint test for father’s edu 0.001 0.000 0.000 0.000 Joint test for mother’s edu 0.357 0.000 0.000 0.000 Joint test for community: Jr 0.313 0.915 0.276 0.546 Joint test for community: Sr 0.373 0.290 0.158 0.037 Joint test for community 0.398 0.052 0.305 0.013 Dropout rate 0.352 0.251 0.254 0.226 t statistics in parentheses Robust, cluster(district) ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 38 primary-junior secondary transition. However, the different role of age may be expected since the mandate of junior secondary education occurs in 1994. Older cohort simply may have graduated primary school before the mandate could be implemented. Being female and from households with little land continue to be negatively affect- ing the junior-senior secondary transition as in the primary-junior secondary transition, again with the magnitudes of the estimates decreasing over time. Most interesting is the role of mother’s education. In IFLS1, mother’s education does not contribute to the junior-senior secondary transition; by IFLS4, mother’s education most strongly affects the decision to transition. Father’s education still continues to play a very strong and significant role, but the decrease in magnitudes of the estimates suggest that father’s education is becoming less important. Furthermore, as mother’s education becomes more important in IFLS3 and IFLS4, the coefficient estimates are larger in magnitude than those for primary-junior secondary transition. In the junior- senior secondary transition, Mother’s edu: Sr/College estimates are -0.255 and -0.220 in IFLS3 and IFLS4, respectively. In the primary-junior secondary transition, Mother’s edu: Sr/College estimates are -0.175 and -0.129. The level of mother’s education may affect her child’s schooling more depending on the child’s level of schooling. 2.5.2.2 Oaxaca-Blinder Decomposition Table 2.10 shows the summary decomposition for the gap in the dropout rates from IFLS1 to IFLS4 for the junior secondary school graduates. Of the 14.5 percentage point difference in dropouts rates, essentially all of the difference is explained by differences in endowments. Community characteristics continue to play the most significant role in the decomposition, but now explaining 89% (-0.280/-0.313) the difference in dropout rates through endowments versus 57% in the primary-junior secondary transition. 39 Table 2.10: Oaxaca Decomposition between IFLS4 & IFLS1 Junior-Senior Secondary Transitional Dropouts IFLS4-IFLS1 Differential Prediction 1 0.224 (0.00599) Prediction 2 0.369 (0.0231) Difference -0.145 (0.0239) Endowments Personal 0.00243 (0.00888) Rural 0.00462 (0.00345) Household other 0.0000875 (0.00171) Father edu -0.0295 (0.00946) Mother edu -0.0112 (0.00872) Community -0.280 (0.149) Missing 0.000495 (0.00839) Total -0.313 (0.151) Coefficients Personal 0.391 (0.101) Rural -0.0109 (0.0202) Household other -0.0146 (0.0170) Father edu 0.183 (0.0834) Mother edu -0.0580 (0.0542) Community 0.187 (0.395) Missing 0.00210 (0.0299) Constant -0.707 (0.424) Total -0.0274 (0.0407) Interaction Personal -0.00710 (0.00933) Rural -0.00127 (0.00246) Household other 0.00153 (0.00203) Father edu 0.0149 (0.00645) Mother edu -0.00900 (0.00892) Community 0.195 (0.153) Missing 0.00174 (0.00868) Total 0.196 (0.154) Observations 5340 Endowments/Difference (%) 216.1 Coefficients/Difference (%) 18.88 Prediction 1=IFLS4, Prediction 2=IFLS1 Personal: male, non-muslim, age indicators Household other: land value Father edu: primary, jr, sr/college Mother edu: primary, jr, sr/college Community: number of schools, uniform expenditures, distance to school, EBTANAS, number of students, student/teacher ratio (at junior and senior secondary levels) Missing: missing indicators for community characteristics In contrast, parental education plays a much lesser role in the explainable endow- ments: father’s and mother’s education combined contribute only 13% (-0.0407/-0.313). Considering the widening of parental educational differences between the non-dropouts 40 and dropouts from the summary statistics and the persistent (and possibly increasing im- portance of mother’s education), parental education accounting for only small percent- ages may suggest that there may be a different mechanism affecting the junior-senior secondary transition than the primary-junior secondary transition. One possible mecha- nism is that although parents in general have higher education by IFLS4, the decision to transition to senior secondary school may be more dependent on availability of schools such as distance to school or cost of attending school. The coefficients have a smaller impact in explaining the differences in junior-senior secondary transitional dropout rates from IFLS1 to IFLS4; should IFLS4 coefficients be applied to IFLS1 characteristics, the dropout rate would only decrease by 2.7 percentage points. 2.6 Children Aged 13-14 The previous sections looked at transitional dropouts. However, since graduating from primary school is a requirement to be selected in the sample for the primary-junior sec- ondary transition analysis, the restriction may include selection bias in the estimates. Although Table 2.3 shows that only about 7% of 15-24 year olds do not complete pri- mary school in IFLS4, about 20% of 15-24 year olds do not complete primary school in IFLS1. The issue is likely to be amplified for the junior-senior secondary transitional analysis. For females in IFLS1, less than 50% of 15-24 year olds graduate from junior secondary schools. To address the possible selection bias, I look at all respondents aged 13-14 without requiring the respondents to graduate from primary schools. Restricting the ages to 13- 14 helps to provide sufficient time for most respondents to graduate from primary school while limiting the maximum educational attainment to junior secondary school. 41 2.6.1 All Dropouts: Mid-school and Transitional Because of age and possible grade repetition, many 13-14 year olds simply may not have had the opportunity to complete primary school. Table 2.11 shows that only 78% of boys aged 13-14 have completed primary school in IFLS4 and only 66% in IFLS1. For males aged 15-24, the primary school completion rates are higher at 93% in IFLS4 and 84% in IFLS1. Table 2.11: Education Level for 13-14 Year Olds Female Male IFLS1 IFLS2 IFLS3 IFLS4 IFLS1 IFLS2 IFLS3 IFLS4 No school 0.0121 0.0120 0.0101 0.00940 0.0195 0.00716 0.0128 0.00937 Completed primary or higher 0.694 0.773 0.830 0.877 0.663 0.716 0.769 0.784 Completed jr or higher 0.0138 0.0216 0.0404 0.0295 0.0124 0.0215 0.0321 0.0107 N 578 832 793 745 563 838 779 747 Unconditional: all 13-14 year olds Hence for 13-14 year olds, being a dropout is initially defined as not currently en- rolled in school. I drop those who never enrolled in school, but as Table 2.11 shows, only 1% of the 13-14 year olds never enrolled in school across the waves. I change the status for those who are currently not enrolled in school because they already completed ju- nior secondary school as a non-dropout. Table 2.11 shows that about 3% of females and 1% of males in IFLS4 complete junior secondary school by age 14. The dropouts then should be equivalent to primary-junior secondary transitional dropouts and mid-school dropouts from both primary school and junior secondary school, and the dropout rates are shown in Figure 2.4. 25 The proportion of dropouts decreases from 23% in IFLS1 to 10% in IFLS4 for males and from 21% to 12% for males. The similar percentages for males and females would again represent further (equal) educational opportunities for younger girls in Indonesia. 25 Table 2.11 does reveal a very small proportion of students completing junior-secondary school by age 14, at around 3% for females and 1% for males in IFLS4. 42 Figure 2.4: Dropouts among All 13-14 Year Olds (%) 22.6 17.3 19.8 9.8 20.7 16.3 15.7 11.8 10.0 15.0 20.0 25.0 Dropout Rate (%) 1 2 3 4 IFLS Waves Female Male Unconditional: all children 13−14 years old 2.6.2 Additional Variables Limiting the age distribution helps to include few important variables that may affect schooling decision; the raven’s test score and per capita expenditures, as well as other proxies for wealth. 2.6.2.1 Raven’s Test Scores In the previous analysis involving the 15-24 year olds, the Raven’s test scores were only available for IFLS3 and IFLS4 and thus excluded from the regressions involving all four waves. However, with a much narrower age cohort, one of the benefits is the backward matching of the Raven’s test scores. A 13-14 year old in IFLS1 should be 20-21 in IFLS3, and a 13-14 year old in IFLS2 should be 16-17 in IFLS3. About 78% of IFLS1 respondents and 91% of IFLS2 respondents are able to be matched backward 43 from the Raven’s test scores from IFLS3 and IFLS4. 26 Regressions can now include the Raven’s test scores, excluding those unable to be matched. Raven’s scores are a good proxy of the respondent’s cognitive ability The importance of congnitve ability for schooling and eventually for income have been extensively studied (Boissiere et al., 1985; Hanushek, 1986; Glewwe and Jacoby, 1994; Behrman et al., 1996; Hanushek and Woessmann, 2008), and thus should be included when possible, even with the loss in sample size. Additionally, since the Raven’s test is administered to all respondents aged 15-24 regardless of educational background, those who never finished primary school can be compared with the primary school graduates. 27 2.6.2.2 Proxies for Wealth In the previous analysis involving the 15-24 year olds, proxies for wealth could not be included in the regression as they were measures at the time of survey and not at the time of transition despite being important components of socio-economic-status. With an age cohort close to or at the primary-junior secondary transition, per capita expenditures, an indicator for dirt floor, an indicator for farming household, and an indicator for facing crop-loss could all be included as part of household characteristics. 2.6.3 Descriptive Statistics Separating the non-dropouts from dropouts for the 13-14 year olds yields the descriptive statistics shown in Table 2.12. Similar to Table 2.5 for 15-24 year olds, the non-dropouts and dropouts clearly display different characteristics. 26 In IFLS1, 869/1121 successfully matched. In IFLS2, 1514/1670 successfully matched. 27 Appendix Table A.2.5 shows the regression results for IFLS1 and IFLS2, with and without the Raven’s test scores. Despite the reduced sample size, the estimates are comparable and the adjusted R 2 increases by 11% with the inclusion of the Raven’s test scores for IFLS1 from 0.202 to 0.224 and by 25% for IFLS2 from 0.153 to 0.192. 44 Table 2.12: Summary Statistics for All 13-14 Year Olds IFLS1 IFLS2 IFLS3 IFLS4 ND Dropout Diff ND Dropout Diff ND Dropout Diff ND Dropout Diff Personal: Male 0.498 0.469 0.0286 0.505 0.488 0.0171 0.508 0.437 0.0708 ∗ 0.495 0.547 -0.0515 Non-muslim 0.187 0.0905 0.0963 ∗∗∗ 0.127 0.0534 0.0740 ∗∗∗ 0.125 0.0538 0.0717 ∗∗∗ 0.106 0.0497 0.0562 ∗ Raven’s test (z-score) 0.101 -0.431 0.532 ∗∗∗ 0.170 -0.601 0.771 ∗∗∗ 0.411 -0.100 0.511 ∗∗∗ 0.227 -0.259 0.486 ∗∗∗ Household: Rural 0.490 0.671 -0.181 ∗∗∗ 0.511 0.719 -0.208 ∗∗∗ 0.548 0.728 -0.180 ∗∗∗ 0.516 0.609 -0.0925 ∗ Log PCE 12.81 12.45 0.354 ∗∗∗ 12.72 12.31 0.407 ∗∗∗ 12.75 12.43 0.313 ∗∗∗ 12.79 12.43 0.366 ∗∗∗ Faced crop loss 0.107 0.107 0.0000656 0.120 0.164 -0.0435 ∗ 0.118 0.168 -0.0509 ∗ 0.0541 0.0745 -0.0204 Dirt floor 0.0285 0.119 -0.0909 ∗∗∗ 0.121 0.263 -0.142 ∗∗∗ 0.103 0.215 -0.112 ∗∗∗ 0.0759 0.0994 -0.0235 Land value 38.72 34.44 4.276 33.71 28.70 5.009 ∗ 35.86 29.50 6.360 ∗ 32.69 22.35 10.34 ∗∗ Farming household 0.406 0.550 -0.143 ∗∗∗ 0.335 0.470 -0.134 ∗∗∗ 0.400 0.419 -0.0190 0.353 0.248 0.105 ∗∗ Father’s edu: None 0.0748 0.269 -0.194 ∗∗∗ 0.0801 0.265 -0.185 ∗∗∗ 0.0758 0.221 -0.146 ∗∗∗ 0.0410 0.163 -0.121 ∗∗∗ Father’s edu: Primary 0.513 0.639 -0.126 ∗∗∗ 0.531 0.655 -0.124 ∗∗∗ 0.544 0.672 -0.127 ∗∗∗ 0.479 0.700 -0.221 ∗∗∗ Father’s edu: Jr 0.178 0.0485 0.130 ∗∗∗ 0.156 0.0442 0.112 ∗∗∗ 0.137 0.0627 0.0747 ∗∗∗ 0.149 0.0813 0.0677 ∗ Father’s edu: Sr/College 0.234 0.0441 0.190 ∗∗∗ 0.234 0.0361 0.197 ∗∗∗ 0.242 0.0443 0.198 ∗∗∗ 0.331 0.0563 0.275 ∗∗∗ Mother’s edu: None 0.131 0.403 -0.272 ∗∗∗ 0.133 0.369 -0.236 ∗∗∗ 0.127 0.339 -0.212 ∗∗∗ 0.0773 0.157 -0.0799 ∗∗∗ Mother’s edu: Primary 0.578 0.550 0.0284 0.592 0.590 0.00157 0.582 0.602 -0.0202 0.507 0.717 -0.210 ∗∗∗ Mother’s edu: Jr 0.144 0.0216 0.122 ∗∗∗ 0.132 0.0321 0.100 ∗∗∗ 0.131 0.0438 0.0871 ∗∗∗ 0.182 0.0818 0.100 ∗∗ Mother’s edu: Sr/College 0.147 0.0260 0.121 ∗∗∗ 0.143 0.00803 0.135 ∗∗∗ 0.160 0.0146 0.145 ∗∗∗ 0.234 0.0440 0.189 ∗∗∗ Community - Primary: Number of schools 3.277 3.161 0.116 5.662 5.183 0.479 ∗∗ 4.066 3.851 0.216 ∗ 4.635 4.611 0.0241 Uniform expenditures 4.386 4.469 -0.0826 14.29 14.43 -0.141 17.89 18.05 -0.165 18.57 18.48 0.0905 Distance to schools (km) 0.952 1.158 -0.206 ∗∗ 1.453 1.314 0.139 1.938 1.956 -0.0182 2.840 3.071 -0.231 EBTANAS Indonesian 6.605 6.242 0.364 ∗∗∗ 6.630 6.365 0.265 ∗∗∗ 6.807 6.668 0.139 ∗∗ 7.095 6.939 0.156 ∗∗∗ EBTANAS Math 6.224 5.892 0.331 ∗∗∗ 5.867 5.632 0.235 ∗∗∗ 6.402 6.268 0.134 ∗ 6.612 6.564 0.0478 Number of students 226.6 209.1 17.51 ∗ 214.2 206.4 7.849 224.3 217.2 7.183 254.0 267.3 -13.31 Student/teacher ratio 24.77 26.85 -2.079 ∗∗∗ 16.09 18.02 -1.923 ∗∗∗ 25.19 27.61 -2.424 ∗∗∗ 19.18 21.20 -2.015 ∗∗ Community - Junior: Number of schools 2.262 2.287 -0.0241 5.438 4.910 0.528 ∗∗∗ 3.594 3.333 0.260 ∗∗∗ 3.701 3.540 0.161 Uniform expenditures 10.86 10.58 0.274 18.48 18.16 0.320 19.46 19.30 0.161 20.50 19.94 0.562 ∗ Distance to schools (km) 1.977 2.461 -0.484 ∗∗∗ 3.719 4.067 -0.348 ∗∗ 4.266 4.887 -0.621 ∗∗∗ 4.345 4.432 -0.0867 EBTANAS Indonesian 6.833 6.756 0.0771 ∗ 6.737 6.570 0.166 ∗∗∗ 5.379 5.284 0.0945 ∗∗ 7.345 7.306 0.0393 EBTANAS Math 4.467 4.420 0.0473 5.171 5.031 0.140 ∗∗∗ 5.216 5.075 0.140 ∗∗∗ 6.906 6.949 -0.0431 Number of students 463.8 459.5 4.288 548.6 528.8 19.81 527.1 506.7 20.40 504.2 496.0 8.279 Student/teacher ratio 15.65 15.76 -0.112 10.93 10.63 0.298 16.34 16.15 0.189 14.15 14.44 -0.292 N 878 243 1389 281 1293 279 1331 161 Transitional: conditional on graduating primary school Land value and uniform expenditures: quartic rooted Community variables: merged at sub-district level, missing observations replaced with district/province averages 45 All personal characteristics except for gender are significantly different between non-dropouts and dropouts. 28 Dropouts are more likely to be Muslim and have lower Raven’s test scores. Again worth noting, the average standardized Raven’s z-score is positive for non-dropouts and is negative for dropouts. Household characteristics con- tributing to dropping out are those indicating lower SES. Dropouts are more likely to come from rural households with lower per capita expenditures, dirt floors, and hav- ing faced crop-loss. The dropouts are also more likely to come from households with lower parental education, particularly of parents without any formal education. Among community characteristics, onlyEBTANASIndonesian andstudent/teacherratio are per- sistently different between the non-dropouts and dropouts across the waves. Dropouts are more likely to live in communities with higher EBTANAS scores and higher stu- dent/teacher ratios. Dropouts live in communities where distance is farther but only in IFLS1. Dropouts also live in communities with fewer primary and junior secondary schools in IFLS2 and IFLS3. Looking across the waves in the descriptive summary statistics, the differences be- tween the non-dropouts and dropouts persist and even increase at the personal and household levels while decreasing at the community level. While differences remain, the gap is not as wide as was the case for 15-24 year olds; the 18% difference in rural households is reduced to 9%. On the other hand, the differences in Raven’s test scores, per capita expenditures, and parental education persist from IFLS1 to IFLS4. All pri- mary school characteristics but for student/teacher ratio decrease over time. Only the difference in uniform expenditures at the junior secondary schools have increased for the non-dropouts and dropouts. 28 The percentage of males for non-dropouts is weakly higher in IFLS3 only. 46 2.6.4 Regressions The linear probability regression results from IFLS1 to IFLS4 for all the 13-14 year olds are shown in Table 2.13, showing the marginal effects on not being enrolled in school. Raven’s test score is a strong predictor for dropping out throughout the waves. A low Raven’s score and being older increases the likelihood of dropping out in IFLS4. The results are consistent with studies of parental education and scholastic ability. Lam et al. (2011) find that parental education and ability as measured by literacy and nu- meracy evaluation help to explain post-secondary school enrollment differences across children from different socio-economic-status in South Africa. The log PCE estimates are consistently negatively significant. In IFLS4, mother’s education appear to not af- fect the likelihood of being in or out of school while father’s education affects being a dropout. By IFLS4, the young children not enrolled are likely to be academically poor- performing students from financially disadvantaged households, but especially with less educated fathers. Community characteristics do not contribute to dropping out at all for the 13-14 year olds. Individual coefficient estimates are insignificant for the most part across the waves, and the joint test for community characteristics is significant in the first three waves. 29 2.6.4.1 Changes from IFLS1 to IFLS4 Looking across the waves in Table 2.13, in contrast to many of the reductions in magni- tudes of the coefficient estimates for the 15-24 year olds, many of the variables increase 29 The pooled regression for all four waves is shown in Appendix Table A.2.6. The test of pooling for the estimates being equal across the waves is strongly rejected. However, an interesting results is the non-significance of IFLS2 and IFLS3 indicators. In 1997 and 2003, the likelihood of a 13-14 year old dropping out of school or not enrolled in school is just as likely as in 1993, despite the substantial increase in overall educational attainment in Indonesia during this time. The pooled regression for 15-24 year old primary-junior secondary transitional dropouts in Appendix Table A.2.2 shows wave indicator estimates of -0.106, -0.129, and -0.201 for IFLS2, IFLS3, and IFLS4, respectively. 47 Table 2.13: Linear Probability Regression for All 13-14 Year Olds y i = 1 if not currently attending school 0 if currently attending school IFLS1 IFLS2 IFLS3 IFLS4 Personal: Male -0.00252 (-0.10) 0.00769 (0.33) -0.0353 ∗ (-2.30) 0.0122 (0.80) Non-muslim -0.0426 (-1.07) -0.102 ∗∗ (-2.82) -0.0324 (-0.82) -0.0158 (-0.59) Raven’s test (z-score) -0.0325 ∗ (-2.56) -0.0665 ∗∗∗ (-6.52) -0.0516 ∗∗∗ (-5.68) -0.0418 ∗∗∗ (-4.44) Age=14 0.0508 (1.77) 0.0534 ∗∗∗ (3.84) 0.0491 ∗∗ (2.76) 0.0673 ∗∗∗ (4.74) Household: Rural -0.0544 (-1.71) 0.0598 ∗∗∗ (3.92) 0.0563 ∗ (2.61) 0.00446 (0.25) Log PCE -0.0412 (-1.99) -0.0570 ∗∗∗ (-5.13) -0.0531 ∗∗∗ (-3.71) -0.0572 ∗∗ (-2.78) Faced crop loss -0.0634 (-1.64) -0.0526 (-1.18) 0.0227 (0.48) -0.0172 (-0.61) Dirt floor 0.203 (1.75) 0.0793 ∗∗ (2.93) 0.0329 (1.02) 0.00480 (0.11) Land value -0.000577 (-1.70) -0.000304 (-1.26) -0.000665 ∗ (-2.12) 0.0000130 (0.04) Farming household 0.132 ∗ (2.06) 0.0984 (1.76) -0.0200 (-0.49) -0.106 ∗ (-2.11) Land value x Farming HH -0.000653 (-0.83) -0.00100 (-1.56) 0.000497 (0.87) 0.00105 (1.86) Father’s edu: Primary -0.118 (-1.67) -0.0817 ∗ (-2.24) -0.109 (-1.93) -0.159 ∗ (-2.22) Father’s edu: Jr -0.171 ∗ (-2.44) -0.129 ∗∗∗ (-3.75) -0.156 ∗ (-2.57) -0.235 ∗∗ (-3.21) Father’s edu: Sr/College -0.180 ∗∗ (-2.71) -0.112 ∗∗∗ (-3.48) -0.165 ∗ (-2.62) -0.233 ∗∗ (-2.87) Mother’s edu: Primary -0.0898 ∗ (-2.15) -0.0624 ∗ (-2.38) -0.107 (-1.67) -0.00498 (-0.15) Mother’s edu: Jr -0.115 ∗ (-2.48) -0.0799 ∗∗ (-3.10) -0.150 ∗ (-2.17) -0.0317 (-0.98) Mother’s edu: Sr/College -0.0716 (-1.35) -0.0618 ∗ (-2.42) -0.124 ∗ (-2.01) -0.0178 (-0.49) Community - Primary: Number of schools -0.0405 ∗ (-2.16) 0.00599 (0.83) -0.00262 (-0.25) 0.0149 (1.74) Uniform expenditures 0.00143 (0.55) 0.00544 (1.95) 0.00694 ∗ (2.05) 0.00158 (0.66) Distance to schools (km) 0.0164 (0.86) -0.00495 (-0.66) -0.0131 ∗ (-2.08) 0.00599 (1.86) EBTANAS Indonesian -0.112 ∗∗∗ (-4.17) -0.0194 (-1.69) -0.0309 (-0.75) -0.00780 (-0.41) EBTANAS Math 0.0246 (1.56) 0.00444 (0.27) 0.0291 (1.02) -0.0161 (-1.29) Number of students -0.000310 (-1.37) -0.000196 ∗ (-2.18) -0.000283 (-1.69) 0.0000232 (0.41) Student/teacher ratio 0.00620 ∗∗∗ (4.03) 0.00357 ∗∗ (3.27) 0.00370 (1.58) 0.00169 (0.72) Community - Junior: Number of schools 0.00396 (0.15) -0.0170 (-1.41) 0.00724 (0.40) -0.0202 (-1.58) Uniform expenditures 0.00530 (1.72) -0.000311 (-0.05) -0.00240 (-0.51) -0.00380 (-1.12) Distance to schools (km) 0.0103 (0.77) -0.0112 (-1.10) -0.00525 (-0.82) -0.00482 (-1.21) EBTANAS Indonesian 0.00799 (0.18) -0.0105 (-0.27) 0.0119 (0.25) 0.0122 (0.53) EBTANAS Math -0.0209 (-0.64) 0.0150 (0.37) -0.0839 (-1.25) -0.00818 (-0.64) Number of students 0.0000114 (0.08) 0.000250 (1.38) 0.0000144 (0.11) 0.0000445 (0.70) Student/teacher ratio 0.00884 (1.25) -0.0141 (-1.67) -0.000179 (-0.04) -0.000499 (-0.15) Geographical indicators Yes Yes Yes Yes Missing indicators Yes Yes Yes Yes Observations 859 1480 1472 1403 AdjustedR 2 0.224 0.192 0.153 0.173 Joint test for father’s edu 0.021 0.000 0.012 0.000 Joint test for mother’s edu 0.046 0.021 0.087 0.514 Joint test for community: Pr 0.000 0.000 0.001 0.520 Joint test for community: Jr 0.245 0.219 0.771 0.306 Joint test for community 0.000 0.000 0.008 0.543 Dropout rate 0.217 0.168 0.177 0.108 t statistics in parentheses Robust, cluster(district) ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 48 in both magnitude and significance. Particularly worth noting are the changes in Raven’s test scores and parental education. In IFLS1, Raven’s test score does not significantly contribute to dropping out; but becomes very significant from IFLS2 to IFLS4. Mother’s education is weakly significant in IFLS1, with p-value for the joint test of mother’s ed- ucation being just over 5%. But by IFLS4, mother’s education does not seem to affect the decision to transition. From IFLS2, father’s education becomes very significant in explaining dropouts and increasingly more so with time. On the other hand, mother’s education in IFLS2 and IFLS4 are not significant at all while being just barely signif- icant at 4.8%. However, estimates from IFLS1 (and IFLS2 to a smaller degree) need to interpreted with additional caution since I dropped about 20% of the sample when reversely matching the Raven’s test score. As mentioned in the introduction, much of the literature estimate the impact of the Asian Financial Crisis on enrollment was not as severe as feared. Had negative impact on education persisted, the 13-14 year old sample would have been the most likely to illustrate any lingering effects. The regression for IFLS3 in Table 2.13 do not indicate noticeable differences from other waves, implying that any negative shock on enrollment was short lived and consistent with the findings by Hartono and Ehrmann (2001) and Strauss et al. (2004). 2.6.4.2 Oaxaca-Blinder Decomposition Table 2.14 shows the summary decomposition for the gap in the dropout rates from IFLS1 to IFLS4 for all 13-14 year olds. Of the 10.9 percentage point difference in dropout rates from IFLS1 to IFLS4, 64% is explained by the differences in endowments; should IFLS1 have had IFLS4 endowments, the dropout rates in IFLS1 would decrease by 7 percentage points. 49 Table 2.14: Oaxaca Decomposition between IFLS4 & IFLS1 13-14 Year Olds IFLS4-IFLS1 Differential Prediction 1 0.0948 (0.00790) Prediction 2 0.204 (0.0140) Difference -0.109 (0.0160) Endowments Personal 0.00429 (0.00256) Raven’s test (z-score) -0.00890 (0.00318) Rural -0.0000732 (0.000370) Log PCE -0.00161 (0.00175) Household other 0.00719 (0.00561) Father edu -0.0139 (0.00578) Mother edu -0.0169 (0.00772) Community -0.0501 (0.0717) Missing 0.00983 (0.00790) Total -0.0702 (0.0732) Coefficients Personal 0.0166 (0.0209) Raven’s test (z-score) -0.0000361 (0.000227) Rural 0.00899 (0.0218) Log PCE -0.0183 (0.307) Household other -0.0334 (0.0201) Father edu -0.0295 (0.0551) Mother edu 0.128 (0.0404) Community 0.226 (0.286) Missing 0.0289 (0.0198) Constant -0.414 (0.413) Total -0.0876 (0.0542) Interaction Personal 0.00108 (0.00263) Raven’s test (z-score) 0.00121 (0.00319) Rural 0.0000852 (0.000427) Log PCE -0.0000423 (0.000712) Household other -0.000957 (0.00714) Father edu -0.00378 (0.00620) Mother edu 0.0177 (0.00902) Community 0.0493 (0.0888) Missing -0.0157 (0.00960) Total 0.0489 (0.0897) Observations 2262 Endowments/Difference (%) 64.48 Coefficients/Difference (%) 80.45 Prediction 1=IFLS4, Prediction 2=IFLS1 Personal: male, non-muslim, age indicator Household other: faced crop loss, dirt floor, land value, farming household, land value× farming household Father edu: primary, jr, sr/college Mother edu: primary, jr, sr/college Community: number of schools, uniform expenditures, distance to school, EBTANAS, number of students, student/teacher ratio (at primary and junior secondary levels) Missing: missing indicators for community characteristics 50 Similar to the Oaxaca decomposition for primary-junior secondary and junior-senior secondary transitional dropouts for 15-24 year olds, the community characteristics are surprisingly large in magnitude, accounting for 71% (-0.0501/-0.0702) of explainable differences via endowments. The individual non-significance of community character- istics in the regressions yet large collective contribution in the decomposition is con- trasted with the Raven’s test scores - Raven’s test scores are significant across the waves yet only contribute marginally in the decomposition. This suggests that individual de- cisions to transition to junior secondary schools may have been dependent upon one’s ability; however, the overall ability of the students do not change drastically from IFLS1 to IFLS4. Once again, while the community characteristics do not appear to affect the individual decisions to transition to junior secondary schools, the communities in which the individuals live in IFLS1 and in IFLS4 are characteristically different in promoting or supporting junior secondary education. 2.7 What Happens to the Dropouts? In countries like the United States, issues associated with dropouts have been researched extensively in the education literature. Wolman et al. (1989) argue that while finishing high school may not guarantee success, not finishing high school may result in nega- tive outcomes for both the individual and society. Rumberger (1987) and Pallas (1987) demonstrate higher levels of unemployment, lower income, and lower lifetime earnings as explicit examples of such individual consequences. Hahn et al. (1987) and Stacey (1998) argue that dropouts contribute negatively to society via loss in tax revenues, in- crease in welfare and unemployment expenditures, and increase in crime prevention expenditures. More recently, overall private and public external benefits from education are becoming evident in other areas beyond the labor market. Cutler and Lleras-Muney 51 (2006) and Grossman (2006) find that more education is negatively associated with poor health conditions. Lochner (2011) summarizes many of the developments in crime re- duction along with health improvements as non-production benefits of education in his Handbook of the Economics of Education chapter. This section discusses some of the outcomes when tracking the respondents to IFLS4. I take the 15-24 year old primary school graduates from IFLS1, IFLS2, and IFLS3 - the sample selected in Section 2.4 - and I match them in IFLS4. Hence I look at the 3 overlapping age cohorts in IFLS4: 29-38 year olds (from IFLS1), 25-34 year olds (from IFLS2), and 22-31 year olds (from IFLS3). For simplicity and consistency sake, other possible trackings such as IFLS1 to IFLS2 are excluded. 2.7.1 Outcomes in IFLS4 I survey some of the outcomes that may be more directly related with dropping out of school at a young age: marriage, pregnancy, work, and subjective well-being. What I consider to be less directly related for the sake of the paper are secondary outcomes such as crime rates and health in latter years. Strauss (1990) and Thomas et al. (1991) have established that better educated mothers have positive influences on their child’s health. While the primary-junior secondary transitional dropouts would surely be considered as less educated, since the effect is inter-generational, I do not track their health out- comes here. For these four outcomes, I acknowledge the possibility of confoundedness and even reverse causality. Through simple tests of means, the outcome variables of marriage, pregnancy, work, and subjective wellbeing all indicate significant differences between non-dropouts and dropouts by the time the panel respondents reach IFLS4. 52 Table 2.15: Marriage reported in IFLS4 y i = 1 if ever married 0 if never married Female Male Non-dropout Dropout Difference Non-dropout Dropout Difference Ever married: IFLS1 respondents 0.949 0.972 -0.0228 0.847 0.921 -0.0732 IFLS2 respondents 0.801 0.928 -0.127 ∗∗∗ 0.642 0.713 -0.0712 ∗ IFLS3 respondents 0.703 0.884 -0.181 ∗∗∗ 0.489 0.583 -0.0947 ∗∗∗ Transitional: conditional on graduating primary school 2.7.1.1 Marriage and Pregnancy Table 2.15 shows that by the time respondents reach IFLS4, the dropouts are much more likely to have ever been married at younger ages. The differences between non-dropouts and dropouts are smallest for IFLS1 respondents and largest for IFLS3 respondents; the pattern is naturally expected as older people are simply more likely to be married since the original IFLS1 respondents are oldest respondents between ages of 29-38 in IFLS4. The differences are significant for IFLS2 respondents between ages of 25-34 in IFLS4, and the even more so in magnitude for the youngest IFLS3 respondents between ages of 22-31 in IFLS4. The differences in marriage rates between females and males, particularly at younger cohorts, would suggest that females are more likely to be married at earlier ages than males. Table 2.16 shows the the proportion of females who have ever given birth by IFLS4. Similar to the pattern in Table 2.15, the difference is smallest for oldest IFLS1 respon- dents and largest for the youngest IFLS3 respondents. Table 2.15 and Table 2.16 com- bined suggest that the female dropouts are not only more likely to get married but also have children at younger ages. Note that marriage or pregnancy is not the likely cause of dropping out at earlier ages. In the United States, Levine and Painter (2003) report very strong correlation between fertility and low levels of schooling for teenage mothers. However, Eloundou-Enyegue (2004) finds that only 20% of sub-Saharan girls dropping 53 out of school before completing junior secondary is attributed to marriage and/or preg- nancy. Table 2.16: Pregnancy reported in IFLS4 y i = 1 if ever given birth 0 if never given birth Non-dropout Dropout Difference IFLS1 respondents 0.960 0.936 0.0238 IFLS2 respondents 0.905 0.948 -0.0429 ∗ IFLS3 respondents 0.874 0.913 -0.0394 ∗ Females only The expedited marriage and pregnancy patterns are consistent with the long stand- ing literature on education and fertility. Schultz (1973) find that increase in schooling for females in Taiwan led to lower teenage births. Rosenzweig (1982) finds similar pattern in India, Lam and Duryea (1999) in Brazil. Brien and Lillard (1994) find that increased level of education for Malaysian females explain the delay in marriage and in first conception. Field and Ambrus (2008) estimate that one year of delay in marriage is associated with 0.22 years of additional schooling in Bangladesh. Osili and Long (2008) estimate that one additional year of schooling reduces fertility by 0.26 births. 2.7.1.2 Work and Earnings Table 2.17 shows primary work status and activity for the respondents. For males, the dropouts are equally like to be working, but are much more likely to be involved in agri- culture/forestry/fishing than the non-dropouts. 30 The proportion of dropouts involved in the agriculture/forestry/fishing are consistent around 42% for males. Meanwhile, the proportion of non-dropouts involved are decreasing from 22.2% to 18.3% to 17.0% 30 This classification of work is from the work (TK) section of Book 3A in IFLS4, TK1818Ab=01. 54 Table 2.17: Work and Earnings reported in IFLS4 Female Male Non-dropout Dropout Difference Non-dropout Dropout Difference Primary activity last week: Work IFLS1 respondents 0.498 0.493 0.00519 0.917 0.940 -0.0231 IFLS2 respondents 0.456 0.399 0.0568 ∗ 0.894 0.906 -0.0118 IFLS3 respondents 0.447 0.356 0.0911 ∗∗∗ 0.846 0.874 -0.0281 Primary work in Agriculture/Forestry/Fishing IFLS1 respondents 0.183 0.301 -0.118 ∗∗ 0.222 0.412 -0.190 ∗∗ IFLS2 respondents 0.124 0.294 -0.170 ∗∗∗ 0.183 0.422 -0.239 ∗∗∗ IFLS3 respondents 0.109 0.393 -0.284 ∗∗∗ 0.170 0.416 -0.247 ∗∗∗ Salary earned last year (10,000 Indonesian Rupiah) IFLS1 respondents 1333.8 459.5 874.3 ∗∗∗ 1599.6 1009.1 590.5 ∗ IFLS2 respondents 925.3 433.9 491.4 ∗∗∗ 1123.6 540.6 583.0 ∗∗∗ IFLS3 respondents 883.6 363.1 520.5 ∗∗∗ 1026.3 548.7 477.6 ∗∗∗ Primary activity last week: Housekeeping IFLS1 respondents 0.491 0.502 -0.0109 IFLS2 respondents 0.524 0.587 -0.0626 ∗ IFLS3 respondents 0.518 0.630 -0.112 ∗∗∗ for the IFLS1, IFLS2, and IFLS3 respondents, respectively. For females, the propor- tion involved in agriculture/forestry/fishing is decreasing even more from 18.3%, 12.4% to 10.9%, possibly suggesting that the occupational choices of the non-dropouts are moving away from agriculture/forestry/fishing. Likely to be linked with the type of oc- cupation, the salaries between the non-dropouts and dropouts from their primary and secondary occupations are shown in Table 2.17. 31 The dropouts are much more likely to earn less than the non-dropouts, with the difference being the largest for the oldest IFLS1 respondents. For female dropouts not working in the labor market, youngest IFLS3 respondents are more likely to be engaged in housekeeping as their primary ac- tivity. Given that female dropouts are more likely to be married and have given birth at younger ages, the statistical difference seems consistent. While there are plenty of studies linking education and employment (Psacharopou- los, 1973; Duflo, 2001), there are other studies supporting employment opportunities for the less educated. Eckstein and Wolpin (1999) suggest that high school dropouts in the 31 As noted in Table 2.17, the units are per 10,000 Indonesia Rupiah. In 2007, the exchange rate per US dollar was about 9,000 Indonesian Rupiah per 1 US dollar. 55 United States are inherently different from those staying in school. The authors suggest that the dropouts indeed have a comparative advantage for the jobs that they acquire or qualify for than the school graduates. Pitt et al. (2012) support the idea that returns to schooling differ by occupation in rural Bangladesh. Given the more labor-intensive labor market conditions of rural Indonesia, the primary-junior secondary transitional dropouts may have decided to leave school with similar comparative advantages. 2.7.1.3 Subjective Wellbeing For the youngest respondents, the 15 year olds from IFLS3, the male respondents may not have entered the workforce while the female respondents may not have entered marriage or experienced pregnancy. Thus, a subjective wellbeing measure may be a good proxy for overall life satisfaction. The subjective wellbeing section was introduced to IFLS in the third wave, and the questions are as follows. Please imagine a six-step ladder where on the bottom (the first step), stand the poorest people, and on the highest step (the sixth step), stand the richest people. On which are you? On which step were you five years ago? 32 On which step do you expect to find five years from now? While the question is focused on economic wellbeing, classifying the “poorest” ver- sus the “richest”, the question would reveal the respondent’s subjective assessment of oneself among others. Table 2.18 shows that overall economic life satisfaction of non- dropouts is greater than dropouts in IFLS4 whether the respondents answered for the past, present or future. The differences between non-dropouts and dropouts are all sig- nificant across the waves. The past, present, and future differences depict the changes in subjective recollection and expectation within each wave. For females, the gap between the non-dropouts and dropouts increase over time, suggesting not only do the female 32 In IFLS3, the question for past was for 1 year ago; in IFLS4, 5 years ago. 56 dropouts always feel as though they are in the lower steps in the ladder of life, the gap increases for older cohorts. Table 2.18: Subjective Wellbeing reported in IFLS4 Female Male Non-dropout Dropout Difference Non-dropout Dropout Difference IFLS1 respondents: Past 2.800 2.701 0.0990 2.778 2.360 0.418 ∗∗ Present 3.187 2.948 0.238 ∗ 2.967 2.580 0.387 ∗∗ Future 4.021 3.651 0.370 ∗∗ 3.865 3.292 0.574 ∗∗ IFLS2 respondents: Past 2.832 2.672 0.160 ∗∗ 2.785 2.533 0.253 ∗∗∗ Present 3.137 2.939 0.199 ∗∗∗ 3.029 2.786 0.243 ∗∗∗ Future 3.982 3.735 0.247 ∗∗∗ 3.926 3.648 0.278 ∗∗∗ IFLS3 respondents: Past 2.822 2.560 0.262 ∗∗∗ 2.759 2.483 0.276 ∗∗∗ Present 3.116 2.894 0.223 ∗∗∗ 3.008 2.721 0.287 ∗∗∗ Future 3.980 3.686 0.294 ∗∗∗ 3.926 3.614 0.312 ∗∗∗ Past: 5 years ago Future: 5-year ahead expectation I also check the possibility of dropping out of school having an short-run impact on subjective wellbeing: by looking at the subjective wellbeing of IFLS4 dropouts in IFLS3 through a reverse tracking. 33 Table 2.19 shows that IFLS4 dropouts report significantly lower wellbeing measure than the non-dropouts in IFLS3. The reverse tracking is not a check for reverse causality. A 15-17 year old would have already faced primary-junior secondary school transition, and his/her dropout status would be the same in IFLS3 and in IFLS4. Since the subjective well being question is only asked to adults and only from IFLS3, I cannot see if un-happier respondent were more likely to dropout. Additionally worth noting, the IFLS3 dropouts report significantly lower wellbeing measures than the non-dropouts within IFLS3, such that the dropouts are simply less satisfied with life at all times than their counterparts overall. 33 Note that the likelihood of matching when tracking backward is very limited given the age restriction of 15-24, especially since the subjective wellbeing is asked to adults only. With the 7-year gap between IFLS3 and IFLS4, the only eligible overlapping age cohort in IFLS3 would be the 15-17 year olds. 57 Table 2.19: Subjective Wellbeing reported in IFLS3 (a) IFLS4 respondents Female Male Non-dropout Dropout Difference Non-dropout Dropout Difference Past 3.159 2.898 0.261 ∗∗∗ 3.156 2.730 0.425 ∗∗∗ Present 3.126 2.870 0.256 ∗∗∗ 3.106 2.757 0.349 ∗∗∗ Future 3.506 3.161 0.345 ∗∗∗ 3.451 3.162 0.288 ∗∗ (b) IFLS3 respondents Female Male Non-dropout Dropout Difference Non-dropout Dropout Difference Past 3.203 2.893 0.310 ∗∗∗ 3.136 2.830 0.306 ∗∗∗ Present 3.173 2.921 0.252 ∗∗∗ 3.093 2.862 0.231 ∗∗∗ Future 3.609 3.380 0.229 ∗∗∗ 3.508 3.295 0.212 ∗∗∗ Past: 1 year ago Although a direct relationship between level of education and subjective wellbeing may not have been formed, Clark and Oswald (1994) and Oswald (1997) have long es- tablished a strong correlation between employment status and happiness. Should the primary-junior secondary dropouts be at a disadvantage in sustaining a continuous em- ployment status in the labor market, the dropouts may continue to express less satisfac- tion. 2.8 Conclusion and Future Research This paper aimed to focus on the young children of Indonesia who drop out of school between transitions. Despite Indonesia’s educational growth in the last four decades, the transition from primary to junior secondary school, then from junior to senior secondary school, remain a challenge for many young children from disadvantaged backgrounds. Across the three relatively different groups - the primary-junior secondary transitional students among 15-24 year olds, the junior-senior secondary transitional students among 15-24 year olds, and all 13-14 year olds - a few results are consistent. First, female 58 educational attainment have not only caught up but marginally surpassing the males. Second, being from a family with lower SES characteristics hampers the likelihood of schooling, particularly if parents have low levels of education. Third, personal and household characteristics influence the likelihood of schooling more than community primary and junior secondary school characteristics. When the primary-junior secondary transitional dropouts are tracked to the lastest wave of the IFLS, the dropouts appear to get married and give birth at younger ages, to work in agricultural/forestry/fishing industry, to earn less income, and to be less satisfied in their general subjective well-being. When the data for the next wave of IFLS becomes available, the analysis suggested in this paper can be re-examined and extended. By tracking the original IFLS1 respondents for 20 years, possible long-term impacts may be predicted and/or observed. 59 Appendix 2 Figure A.2.1: Female to Male Enrollment Ratio (%) (a) Primary 60 70 80 90 100 Female to Male Ratio 1970 1980 1990 2000 2010 Year Indonesia OECD members World Least developed countries Source: World Development Indicator 2013 (b) Secondary 40 60 80 100 Female to Male Ratio 1970 1980 1990 2000 2010 Year Indonesia OECD members World Least developed countries Source: World Development Indicator 2013 Figure A.2.2: Literacy Rate (%) (a) Adults, 15 & older 50 60 70 80 90 Literacy Rate 1990 1995 2000 2005 2010 Year Indonesia Least developed countries World Source: World Development Indicator 2013 (b) Youth, 15-24 60 70 80 90 100 Literacy Rate 1990 1995 2000 2005 2010 Year Indonesia Least developed countries World Source: World Development Indicator 2013 60 Figure A.2.3: Indonesia Family Life Survey Coverage Map Source: http://www.rand.org/labor/FLS/IFLS.html Figure A.2.4: Grade Repetition in Primary School (a) Female 0 5 10 15 Repeaters (% of enrollement) 1990 1995 2000 2005 2010 Year Indonesia OECD members World Least developed countries Source: World Development Indicator 2013 (b) Male 0 5 10 15 Repeaters (% of enrollement) 1990 1995 2000 2005 2010 Year Indonesia OECD members World Least developed countries Source: World Development Indicator 2013 61 Table A.2.1: IFLS (a) Households IFLS1 IFLS2 IFLS3 IFLS4 Urban 3436 3505 5025 7386 (47.56) (45.91) (48.16) (54.57) Rural 3788 4129 5410 6150 (52.44) (54.09) (51.84) (45.43) Observations 7224 7634 10435 13536 Source: Book K Control (b) Adults IFLS1 IFLS2 IFLS3 IFLS4 Female 7863 11075 13583 15633 (54.54) (53.95) (52.59) (52.17) Male 6555 9454 12246 14334 (45.46) (46.05) (47.41) (47.83) Observations 14418 20529 25829 29967 Source: Book 3A Adults (c) Children IFLS1 IFLS2 IFLS3 IFLS4 Female 3775 5152 5736 6618 (48.70) (49.47) (48.86) (48.48) Male 3976 5263 6003 7034 (51.30) (50.53) (51.14) (51.52) Observations 7751 10415 11739 13652 Source: Book 5 Children 62 Table A.2.2: Pooled Linear Probability Regression for Primary School Graduates among 15-24 Year Olds yi = 1 if drops out after primary school 0 if continues to junior secondary school w/ District Indicators w/ Community Characteristics Personal: Male -0.0315 ∗∗∗ (-5.11) -0.0313 ∗∗∗ (-5.02) Non-muslim -0.0482 ∗∗∗ (-3.94) -0.0441 ∗∗∗ (-3.79) Age=16 -0.00298 (-0.29) -0.00452 (-0.45) Age=17 0.0113 (1.24) 0.00973 (1.05) Age=18 0.0230 ∗ (2.28) 0.0221 ∗ (2.16) Age=19 0.0367 ∗∗∗ (3.37) 0.0360 ∗∗∗ (3.34) Age=20 0.0604 ∗∗∗ (5.71) 0.0609 ∗∗∗ (5.65) Age=21 0.0770 ∗∗∗ (6.05) 0.0771 ∗∗∗ (6.04) Age=22 0.0815 ∗∗∗ (6.40) 0.0806 ∗∗∗ (6.40) Age=23 0.0690 ∗∗∗ (5.38) 0.0665 ∗∗∗ (5.24) Age=24 0.0852 ∗∗∗ (6.41) 0.0813 ∗∗∗ (5.99) Household: Rural 0.0840 ∗∗∗ (5.99) 0.0707 ∗∗∗ (4.78) Land value -0.000501 ∗∗∗ (-6.08) -0.000528 ∗∗∗ (-6.38) Father’s edu: Primary -0.141 ∗∗∗ (-8.72) -0.138 ∗∗∗ (-8.48) Father’s edu: Jr -0.233 ∗∗∗ (-13.10) -0.230 ∗∗∗ (-12.88) Father’s edu: Sr/College -0.250 ∗∗∗ (-13.59) -0.247 ∗∗∗ (-13.36) Mother’s edu: Primary -0.133 ∗∗∗ (-10.39) -0.128 ∗∗∗ (-9.79) Mother’s edu: Jr -0.174 ∗∗∗ (-13.09) -0.166 ∗∗∗ (-12.34) Mother’s edu: Sr/College -0.157 ∗∗∗ (-11.87) -0.148 ∗∗∗ (-11.15) Community - Primary: Number of schools 0.0000335 (0.01) Uniform expenditures -0.000672 (-0.60) Distance to schools (km) 0.000619 (0.30) EBTANAS Indonesian -0.0202 ∗∗∗ (-4.48) EBTANAS Math -0.0106 ∗ (-1.98) Number of students 0.0000303 (0.38) Student/teacher ratio 0.00176 ∗ (2.11) Community - Junior: Number of schools -0.00683 ∗ (-2.21) Uniform expenditures 0.000886 (0.74) Distance to schools (km) 0.00327 (1.94) EBTANAS Indonesian -0.00409 (-0.51) EBTANAS Math 0.0108 (1.42) Number of students -0.0000351 (-1.25) Student/teacher ratio -0.000907 (-0.56) Waves: IFLS2 -0.0612 ∗∗ (-3.33) -0.0413 (-1.45) IFLS3 -0.111 ∗∗∗ (-6.62) -0.114 ∗∗∗ (-3.82) IFLS4 -0.197 ∗∗∗ (-10.43) -0.199 ∗∗∗ (-6.13) Geographical indicators Yes Yes Missing indicators Yes Yes Observations 18798 18438 AdjustedR 2 0.247 0.251 Test of pooling 0.000 0.000 Joint test for age 0.000 0.000 Joint test for father’s edu 0.000 0.000 Joint test for mother’s edu 0.000 0.000 Joint test for community: Pr 0.000 Joint test for community: Jr 0.071 Joint test for community 0.000 Dropout rate 0.197 0.197 t statistics in parentheses Robust, cluster(district) ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 63 Table A.2.3: With and Without Raven’s Test - Linear Probability Regression for Primary School Graduates among 15-24 Year Olds y i = 1 if drops out after primary school 0 if continues to junior secondary school IFLS3 IFLS4 w/o Raven’s w/ Raven’s w/o Raven’s w/ Raven’s Personal: Male -0.0365 ∗∗∗ (-3.99) -0.0182 (-1.96) -0.00853 (-1.36) -0.00147 (-0.23) Non-muslim -0.0326 ∗ (-1.99) -0.0242 (-1.48) -0.0290 ∗ (-2.39) -0.0235 ∗ (-2.08) Raven’s test (z-score) -0.0686 ∗∗∗ (-11.43) -0.0446 ∗∗∗ (-6.47) Age=16 -0.0295 (-1.80) -0.0285 (-1.77) 0.0356 ∗ (2.43) 0.0343 ∗ (2.33) Age=17 -0.00236 (-0.16) -0.00134 (-0.10) 0.0297 (1.81) 0.0306 (1.84) Age=18 0.00381 (0.22) 0.00389 (0.24) 0.0503 ∗∗ (3.05) 0.0535 ∗∗ (3.18) Age=19 0.0221 (1.21) 0.0178 (0.96) 0.0495 ∗∗ (2.87) 0.0483 ∗∗ (2.85) Age=20 0.0791 ∗∗∗ (4.10) 0.0765 ∗∗∗ (3.86) 0.0395 ∗∗ (2.69) 0.0357 ∗ (2.51) Age=21 0.104 ∗∗∗ (5.17) 0.0955 ∗∗∗ (4.92) 0.0649 ∗∗∗ (3.71) 0.0631 ∗∗∗ (3.73) Age=22 0.117 ∗∗∗ (5.27) 0.102 ∗∗∗ (4.75) 0.0655 ∗∗∗ (3.98) 0.0604 ∗∗∗ (3.70) Age=23 0.0988 ∗∗∗ (5.18) 0.0889 ∗∗∗ (4.51) 0.0633 ∗∗∗ (3.53) 0.0559 ∗∗ (3.15) Age=24 0.123 ∗∗∗ (5.77) 0.113 ∗∗∗ (5.36) 0.0738 ∗∗∗ (3.56) 0.0664 ∗∗ (3.17) Household: Rural 0.0664 ∗∗ (2.78) 0.0536 ∗ (2.41) 0.0579 ∗∗∗ (4.21) 0.0543 ∗∗∗ (4.11) Land value -0.000555 ∗∗∗ (-4.63) -0.000516 ∗∗∗ (-4.33) -0.000363 ∗ (-2.60) -0.000357 ∗ (-2.59) Father’s edu: Primary -0.129 ∗∗∗ (-5.34) -0.132 ∗∗∗ (-5.53) -0.130 ∗∗∗ (-5.01) -0.122 ∗∗∗ (-4.68) Father’s edu: Jr -0.223 ∗∗∗ (-8.67) -0.217 ∗∗∗ (-8.71) -0.197 ∗∗∗ (-6.95) -0.185 ∗∗∗ (-6.58) Father’s edu: Sr/College -0.236 ∗∗∗ (-9.07) -0.226 ∗∗∗ (-9.07) -0.213 ∗∗∗ (-7.77) -0.196 ∗∗∗ (-7.13) Mother’s edu: Primary -0.142 ∗∗∗ (-7.85) -0.131 ∗∗∗ (-7.32) -0.114 ∗∗∗ (-5.41) -0.106 ∗∗∗ (-5.30) Mother’s edu: Jr -0.179 ∗∗∗ (-8.86) -0.161 ∗∗∗ (-8.42) -0.146 ∗∗∗ (-6.64) -0.136 ∗∗∗ (-6.41) Mother’s edu: Sr/College -0.175 ∗∗∗ (-8.76) -0.143 ∗∗∗ (-7.34) -0.129 ∗∗∗ (-5.87) -0.113 ∗∗∗ (-5.25) Community - Primary: Number of schools -0.00260 (-0.53) -0.00205 (-0.42) 0.0113 ∗∗ (3.06) 0.0107 ∗∗ (2.96) Uniform expenditures 0.000194 (0.05) -0.000169 (-0.05) -0.0000911 (-0.07) -0.000225 (-0.17) Distance to schools (km) 0.01000 (1.60) 0.0141 ∗ (2.42) -0.000109 (-0.05) -0.0000398 (-0.02) EBTANAS Indonesian -0.0142 (-0.57) -0.0177 (-0.76) 0.0135 (1.20) 0.0106 (0.94) EBTANAS Math -0.0123 (-0.73) -0.00562 (-0.34) 0.00191 (0.21) 0.00261 (0.28) Number of students -0.0000272 (-0.24) -0.0000210 (-0.20) -0.0000425 (-0.30) -0.0000170 (-0.12) Student/teacher ratio 0.00150 (1.00) 0.00128 (0.89) 0.00288 (1.56) 0.00288 (1.61) Community - Junior: Number of schools -0.00629 (-0.96) -0.00308 (-0.47) -0.0119 ∗ (-2.29) -0.0101 (-1.96) Uniform expenditures 0.00292 (0.98) 0.00223 (0.76) 0.000426 (0.25) 0.000666 (0.40) Distance to schools (km) 0.00256 (0.65) 0.00143 (0.40) -0.00112 (-0.51) -0.00143 (-0.69) EBTANAS Indonesian 0.00240 (0.09) 0.00672 (0.25) -0.0146 (-1.35) -0.00791 (-0.70) EBTANAS Math 0.0317 (1.00) 0.0330 (1.06) 0.00655 (0.59) 0.00353 (0.32) Number of students -0.000125 ∗ (-2.24) -0.000146 ∗∗ (-2.82) -0.0000757 ∗ (-2.27) -0.0000759 ∗ (-2.10) Student/teacher ratio -0.00249 (-0.85) -0.00118 (-0.44) 0.00572 ∗∗ (2.98) 0.00532 ∗∗ (2.81) Geographical indicators Yes Yes Yes Yes Missing indicators Yes Yes Yes Yes Observations 6620 6516 6326 6270 AdjustedR 2 0.269 0.290 0.171 0.184 Dropout rate 0.205 0.205 0.121 0.121 t statistics in parentheses Robust, cluster(district) ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 64 Table A.2.4: Pooled Linear Probability Regression for Junior Secondary School Gradu- ates among 15-24 Year Olds yi = 1 if drops out after junior secondary school 0 if continues to senior secondary school w/ District Indicators w/ Community Characteristics Personal: Male -0.0335 ∗∗∗ (-3.88) -0.0335 ∗∗∗ (-3.88) Non-muslim -0.0472 ∗∗∗ (-3.98) -0.0472 ∗∗∗ (-3.98) Age=16 -0.0784 ∗∗∗ (-4.06) -0.0784 ∗∗∗ (-4.06) Age=17 -0.0652 ∗∗∗ (-3.62) -0.0652 ∗∗∗ (-3.62) Age=18 -0.0469 ∗∗ (-2.61) -0.0469 ∗∗ (-2.61) Age=19 -0.0386 ∗ (-2.26) -0.0386 ∗ (-2.26) Age=20 -0.0181 (-1.01) -0.0181 (-1.01) Age=21 -0.0539 ∗∗ (-2.61) -0.0539 ∗∗ (-2.61) Age=22 -0.0264 (-1.28) -0.0264 (-1.28) Age=23 -0.00862 (-0.44) -0.00862 (-0.44) Age=24 -0.00978 (-0.50) -0.00978 (-0.50) Household: Rural 0.0842 ∗∗∗ (4.42) 0.0842 ∗∗∗ (4.42) Land value -0.000376 ∗∗∗ (-3.72) -0.000376 ∗∗∗ (-3.72) Father’s edu: Primary -0.113 ∗∗∗ (-4.97) -0.113 ∗∗∗ (-4.97) Father’s edu: Jr -0.211 ∗∗∗ (-7.74) -0.211 ∗∗∗ (-7.74) Father’s edu: Sr/College -0.259 ∗∗∗ (-9.81) -0.259 ∗∗∗ (-9.81) Mother’s edu: Primary -0.120 ∗∗∗ (-8.25) -0.120 ∗∗∗ (-8.25) Mother’s edu: Jr -0.214 ∗∗∗ (-13.23) -0.214 ∗∗∗ (-13.23) Mother’s edu: Sr/College -0.226 ∗∗∗ (-13.13) -0.226 ∗∗∗ (-13.13) Community - Junior: Number of schools 0.00812 (1.48) 0.00812 (1.48) Uniform expenditures -0.000555 (-0.37) -0.000555 (-0.37) Distance to schools (km) -0.00179 (-0.87) -0.00179 (-0.87) EBTANAS Indonesian -0.00431 (-0.35) -0.00431 (-0.35) EBTANAS Math -0.00174 (-0.20) -0.00174 (-0.20) Number of students -0.0000153 (-0.39) -0.0000153 (-0.39) Student/teacher ratio 0.00270 (1.14) 0.00270 (1.14) Community - Senior: Number of schools -0.0105 ∗∗ (-2.76) -0.0105 ∗∗ (-2.76) Uniform expenditures -0.00285 ∗ (-2.42) -0.00285 ∗ (-2.42) Distance to schools (km) 0.00390 ∗ (2.27) 0.00390 ∗ (2.27) EBTANAS Indonesian -0.00250 (-0.86) -0.00250 (-0.86) EBTANAS Math 0.00292 (0.56) 0.00292 (0.56) Number of students 0.0000292 (0.83) 0.0000292 (0.83) Student/teacher ratio -0.00275 (-1.38) -0.00275 (-1.38) Waves: IFLS2 -0.0229 (-0.69) -0.0229 (-0.69) IFLS3 -0.0381 (-1.13) -0.0381 (-1.13) IFLS4 -0.0779 ∗ (-2.02) -0.0779 ∗ (-2.02) Geographical indicators Yes Yes Missing indicators Yes Yes Observations 12568 12568 AdjustedR 2 0.194 0.194 Test of pooling 0.000 0.000 Joint test for age 0.000 0.000 Joint test for father’s edu 0.000 0.000 Joint test for mother’s edu 0.000 0.000 Joint test for community: Jr 0.411 Joint test for community: Sr 0.003 Joint test for community 0.032 Dropout rate 0.247 0.247 t statistics in parentheses Robust, cluster(district) ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 65 Table A.2.5: With and Without Raven’s Test - Linear Probability Regression for All 13-14 Year Olds y i = 1 if not currently attending school 0 if currently attending school IFLS1 IFLS2 w/ Raven’s w/o Raven’s w/ Raven’s w/o Raven’s Personal: Male -0.00252 (-0.10) -0.00109 (-0.05) 0.00769 (0.33) 0.00232 (0.12) Non-muslim -0.0426 (-1.07) -0.0751 ∗ (-2.25) -0.102 ∗∗ (-2.82) -0.0974 ∗∗ (-3.11) Raven’s test (z-score) -0.0325 ∗ (-2.56) -0.0665 ∗∗∗ (-6.52) Age=14 0.0508 (1.77) 0.0701 ∗∗∗ (3.55) 0.0534 ∗∗∗ (3.84) 0.0495 ∗∗∗ (3.50) Household: Rural -0.0544 (-1.71) -0.0333 (-1.08) 0.0598 ∗∗∗ (3.92) 0.0603 ∗∗ (3.14) Log PCE -0.0412 (-1.99) -0.0559 ∗∗ (-3.17) -0.0570 ∗∗∗ (-5.13) -0.0652 ∗∗∗ (-4.27) Faced crop loss -0.0634 (-1.64) -0.0723 (-1.94) -0.0526 (-1.18) -0.0385 (-1.01) Dirt floor 0.203 (1.75) 0.183 ∗ (2.16) 0.0793 ∗∗ (2.93) 0.0858 ∗∗ (3.03) Land value -0.000577 (-1.70) -0.000622 (-1.61) -0.000304 (-1.26) -0.000317 (-1.25) Farming household 0.132 ∗ (2.06) 0.0953 (1.82) 0.0984 (1.76) 0.0940 (1.80) Land value x Farming HH -0.000653 (-0.83) -0.000471 (-0.69) -0.00100 (-1.56) -0.000996 (-1.48) Father’s edu: Primary -0.118 (-1.67) -0.0967 (-1.30) -0.0817 ∗ (-2.24) -0.0834 ∗ (-2.56) Father’s edu: Jr -0.171 ∗ (-2.44) -0.182 ∗ (-2.38) -0.129 ∗∗∗ (-3.75) -0.144 ∗∗∗ (-4.58) Father’s edu: Sr/College -0.180 ∗∗ (-2.71) -0.177 ∗ (-2.58) -0.112 ∗∗∗ (-3.48) -0.118 ∗∗∗ (-3.47) Mother’s edu: Primary -0.0898 ∗ (-2.15) -0.111 ∗ (-2.58) -0.0624 ∗ (-2.38) -0.0893 ∗∗ (-2.90) Mother’s edu: Jr -0.115 ∗ (-2.48) -0.154 ∗∗ (-3.18) -0.0799 ∗∗ (-3.10) -0.113 ∗∗∗ (-3.89) Mother’s edu: Sr/College -0.0716 (-1.35) -0.0884 (-1.69) -0.0618 ∗ (-2.42) -0.110 ∗∗ (-3.42) Community - Primary: Number of schools -0.0405 ∗ (-2.16) -0.0348 (-1.86) 0.00599 (0.83) 0.000251 (0.03) Uniform expenditures 0.00143 (0.55) 0.00157 (0.55) 0.00544 (1.95) 0.00444 (1.53) Distance to schools (km) 0.0164 (0.86) 0.0282 (1.61) -0.00495 (-0.66) -0.00511 (-0.64) EBTANAS Indonesian -0.112 ∗∗∗ (-4.17) -0.110 ∗∗∗ (-3.62) -0.0194 (-1.69) -0.0226 ∗ (-2.14) EBTANAS Math 0.0246 (1.56) 0.0168 (1.03) 0.00444 (0.27) 0.00336 (0.18) Number of students -0.000310 (-1.37) -0.000266 (-1.33) -0.000196 ∗ (-2.18) -0.000284 ∗∗ (-2.69) Student/teacher ratio 0.00620 ∗∗∗ (4.03) 0.00627 ∗∗ (2.89) 0.00357 ∗∗ (3.27) 0.00419 ∗∗∗ (3.88) Community - Junior: Number of schools 0.00396 (0.15) 0.0125 (0.52) -0.0170 (-1.41) -0.0157 (-1.13) Uniform expenditures 0.00530 (1.72) 0.00580 ∗ (2.26) -0.000311 (-0.05) -0.000884 (-0.15) Distance to schools (km) 0.0103 (0.77) 0.00708 (0.57) -0.0112 (-1.10) -0.0200 ∗ (-2.41) EBTANAS Indonesian 0.00799 (0.18) 0.0288 (0.83) -0.0105 (-0.27) -0.00320 (-0.08) EBTANAS Math -0.0209 (-0.64) -0.0248 (-0.87) 0.0150 (0.37) 0.00784 (0.19) Number of students 0.0000114 (0.08) 0.000101 (0.68) 0.000250 (1.38) 0.000224 (1.16) Student/teacher ratio 0.00884 (1.25) 0.00200 (0.35) -0.0141 (-1.67) -0.0136 (-1.53) Geographical indicators Yes Yes Yes Yes Missing indicators Yes Yes Yes Yes Observations 859 1061 1480 1547 AdjustedR 2 0.224 0.202 0.192 0.153 Joint test for community: Pr 0.000 0.000 0.000 0.000 Joint test for community: Jr 0.245 0.099 0.219 0.172 Joint test for community 0.000 0.000 0.000 0.000 Dropout rate 0.217 0.217 0.168 0.168 t statistics in parentheses Robust, cluster(district) ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 66 Table A.2.6: Pooled Linear Probability Regression for All 13-14 Year Olds yi = 1 if drops out after junior secondary school 0 if continues to senior secondary school dropoutall Personal: Male -0.00591 (-0.55) Non-muslim -0.0669 ∗∗∗ (-3.57) Raven’s test (z-score) -0.0520 ∗∗∗ (-7.55) Age=14 0.0552 ∗∗∗ (6.39) Household: Rural 0.0309 ∗ (2.14) Log PCE -0.0546 ∗∗∗ (-6.74) Faced crop loss -0.0121 (-0.41) Dirt floor 0.0464 ∗ (2.22) Land value -0.000372 ∗ (-2.42) Farming household 0.0342 (0.96) Land value x Farming HH -0.000289 (-0.67) Father’s edu: Primary -0.118 ∗∗∗ (-3.63) Father’s edu: Jr -0.176 ∗∗∗ (-5.39) Father’s edu: Sr/College -0.168 ∗∗∗ (-5.06) Mother’s edu: Primary -0.0810 ∗∗ (-3.10) Mother’s edu: Jr -0.107 ∗∗∗ (-3.72) Mother’s edu: Sr/College -0.0837 ∗∗ (-2.86) Community - Primary: Number of schools 0.00422 (0.94) Uniform expenditures 0.00494 ∗∗∗ (3.58) Distance to schools (km) 0.00114 (0.38) EBTANAS Indonesian -0.0212 (-1.88) EBTANAS Math 0.00237 (0.26) Number of students -0.000144 (-1.70) Student/teacher ratio 0.00396 ∗∗∗ (4.30) Community - Junior: Number of schools -0.00440 (-0.67) Uniform expenditures -0.000571 (-0.35) Distance to schools (km) -0.00404 (-1.50) EBTANAS Indonesian 0.000746 (0.05) EBTANAS Math -0.00320 (-0.30) Number of students 0.0000139 (0.28) Student/teacher ratio 0.000259 (0.12) Waves: IFLS2 -0.0548 (-1.43) IFLS3 -0.0489 (-1.27) IFLS4 -0.104 ∗ (-2.46) Geographical indicators Yes Missing indicators Yes Observations 5214 AdjustedR 2 0.184 Test of pooling 0.000 Joint test for father’s edu 0.000 Joint test for mother’s edu 0.002 Joint test for community: Pr 0.000 Joint test for community: Jr 0.882 Joint test for community 0.000 Dropout rate 0.165 t statistics in parentheses Robust, cluster(district) ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 67 Chapter 3 Using Multiple Measures to Make Course Placement Decisions 1 3.1 Introduction This chapter is about the community college students of Los Angeles, CA, and their mathematic course placements. An examination of math assessment and course place- ment in community colleges shows that many students are deemed unprepared for the demands of college-level work. It is estimated that over 60 percent of community college students nationally are placed in at least one postsecondary remedial or developmental course upon entry (NCPPHE & SREB, 2010; Bailey, 2009). 2 Although developmental courses can serve as necessary and helpful stepping-stones to college success, they can also delay access to critical gateway courses necessary for degree attainment or transfer to four-year colleges. In some colleges, students are placed into a sequence of develop- mental math courses that can consist of three or even four courses that are prerequisites for college- and transfer-level work (Bailey et al. 2010; Melguizo et al., forthcoming). This is of concern because recent descriptive research shows that only a small proportion of students placed in lower levels of developmental math sequences enroll in and pass the subsequent math courses needed to attain an associate’s degree or transfer (Bailey et 1 This chapter is co-authored with Federick Ngo. 2 The terms remedial, developmental, basic skills, and preparatory are often used interchangeably in reference to the set of courses that precede college-level courses. We prefer to use the term developmental. 68 al., 2010; Fong et al., 2013). Given the large number of students identified as needing remediation upon entry into community colleges, it is critical to accurately assess and place students into the courses where they are most likely to succeed while not unnec- essarily extending their time towards degree completion or transfer. Placement tests are commonly used in community colleges across the country to make these initial course placement decisions. While practices vary by state and even at the local college level, an increasing number of states have mandated placement testing and the use of common assessments, seeing placement policies as a potential lever for increasing student success (Collins, 2008). At the same time, studies have provided ev- idence that placement tests have low predictive validity and are only weakly correlated with students’ college outcomes, such as college GPA or credit completion (Armstrong, 2000), and that they may inaccurately place as many as one-quarter of community col- lege students in math courses (Belfield and Crosta, 2012; Scott-Clayton, 2012). These same studies suggest that using other measures, such as information from high school transcripts, may be more accurate for placing students than using placement tests alone. Amidst these concerns, states such as Connecticut, Florida, North Carolina, and Texas are considering using multiple measures in their assessment and placement policies for developmental math (Burdman, 2012). While there is knowledge of measures that are predictive of student success in college, there is limited understanding of which mea- sures are useful for making course placement decisions to guide these policy efforts. Community colleges in California have been required to use multiple measures to make course placement determinations since the early 1990s. Prior to this, community colleges were given the mandate to use state-approved tests to assess students during the matriculation process. However, advocacy groups challenged the fairness and ac- curacy of placement tests based on evidence that underrepresented minority students 69 were being disproportionately placed into remedial courses (Perry et al., 2010). A re- vision to Title 5 of the California Code of Regulations in 1992 prohibited community colleges from using “any single assessment instrument, method or procedure, by itself, for placement” (55521). The goal of the policy was to mitigate the disproportionate impact of remediation on underrepresented minority students and to increase access to college-level courses. Yet in line with California’s decentralized educational governance structure, these Title 5 regulations neither formalize a specific statewide assessment and placement pro- cess nor delineate the specific multiple measures to be used (Brewer and Smith, 2008). This shifts responsibility and accountability for implementation to local districts and colleges. Being open access institutions that serve a wide array of students, the colleges must also ensure that students are placed in courses where they can be successful. While using multiple measures in addition to placement test scores can increase the number of students referred to higher-level courses, they may simultaneously result in students be- ing placed into courses for which they are not academically prepared. The colleges thus have the dual goals of increasing access and ensuring success. In this study, we examine the extent to which using multiple measures for course placement can achieve these dual goals of access and success. We analyze evidence from the Los Angeles Community College District (LACCD), the largest community college district in California, and one of the largest in the country. During the ma- triculation process in LACCD, students provide additional information regarding their educational background or college plans in addition to taking a math placement test. In most of the LACCD colleges, this multiple measure information is used to determine whether students should receive points in addition to their placement test score, which can sometimes result in a student being placed into the next higher-level course. We call this a multiple measure boost. Using district-level administrative and transcript data 70 from 2005-2008, we examine the impact of the multiple measure boost on access and success in developmental mathematics. Two research questions guide our study: 1. Does using multiple measures increase access to higher-level math courses and reduce the disproportionate impact of remediation on un- derrepresented minority students? 2. How do students who receive a multiple measure boost into a higher- level math course perform in comparison to their peers? Given that a number of states are considering using multiple measures to inform placement decisions in developmental math, the evidence from the LACCD are timely. The multiple measures policy in California offers a unique context to examine student outcomes and validate the use of multiple measures in making course placement de- cisions. Individual college policies in LACCD allow us to focus on two measures in particular: high school GPA and prior math achievement, each of which is used singu- larly in two LACCD colleges. We begin with a brief overview of the policy background leading up to the man- date to use multiple measures for course placement in California community colleges. We then review the literature on measures that are commonly used to identify college readiness. Modern conceptions of validation provide the framework that we use to ex- amine the usefulness of multiple measures for making placement decisions. Following this theoretical discussion, we describe the implementation of the multiple measures policy in the LACCD and the data and methods used for the study. We then use multi- variate linear regression techniques to compare the outcomes of students who received a multiple measure boost into a higher-level course with those of their higher-scoring peers. Our findings indicate that while using multiple measures does increase access to higher-level courses, the racial composition of courses remains largely unchanged. 71 However, students who received a multiple measure boost based on prior math or high school GPA performed no differently from their peers in terms of course passing rates as well as longer-term credit completion. We conclude by discussing the implications of our findings for assessment and placement policies in developmental math, namely that multiple measures can be used to promote access and success for community college students. 3.2 Background Under the provisions of the Matriculation Act of 1986, the California Community Col- leges Chancellor’s Office (CCCCO) was charged with the responsibility for overseeing that colleges provided core educational services including assessment, orientation, coun- seling, and advising for entering community college students. These services sought to optimize community college students’ opportunities for success and assist them in iden- tifying realistic educational goals upon enrollment, particularly for underrepresented minority students (Wiseley, 2006). Specifically, the Matriculation Act authorized the use of state-approved assessments to determine “student competency in computational and language skills” (78212[b]), and to inform placement in initial math and English courses. Despite the Matriculation Act’s intention to smoothen the transition to postsec- ondary education, the use of placement testing in community colleges became a source of public consternation. Although assessments had to be authorized and were to be used for advisory purposes only, advocates quickly decried how the assessment pol- icy as implemented in community colleges resulted in a large number of Latino and other minority students being placed into remedial classes (Cage, 1991). In 1988, the Mexican-American Legal Defense and Educational Fund (MALDEF) filed a lawsuit 72 charging the CCCCO with failing to adequately monitor the use of assessment tests in its system of colleges. 3 MALDEF argued that the placement tests were biased and be- ing used to “exclude Latino students from college-level courses and to track them into a series of remedial courses” (Venezia et al., 2010). The lawsuit was settled outside of court but succeeded in raising concern about is- sues of equity in matriculation. The State Board of Education subsequently filed legis- lation in Title 5 of the California Code of Regulations in 1992 that explicitly prohibits practices in matriculation that result in disproportionate impact. According to the code, disproportionate impact occurs when “. . . the percentage of persons from a particular racial, ethnic, gender, age or disability group who are directed to a particular service or placement based on an assessment instrument, method, or procedure is significantly different from the representation of that group in the population of persons being assessed, and that discrepancy is not justified by empirical evidence” (55502[d]). According to Stone (2002), disproportionate impact can be demonstrated when there is evidence that “the results of a selection process were highly unlikely to occur by chance” (p.201). 4 Plaintiffs do not need to prove a policy’s intent to discriminate, just discriminatory application or effects. 3 MALDEF filed a lawsuit with the Superior Court of Sacramento County on behalf of a student plaintiff, Christopher Romero-Frias, against the California Community College Chancellor’s Office and then-chancellor David Mertes (Romero-Frias, et al. v. Mertes, et al., 1988/1991). Romero-Frias and other student plaintiffs were students at four-year universities who, upon taking placement tests at community colleges, were unexpectedly placed in remedial courses and barred from taking college-level courses (Wiseley, 2006). 4 Disproportionate impact was first codified in the area of employment discrimination in Title VII of the Civil Rights Act of 1964, and then again in a Supreme Court case, Griggs v. Duke Power Co., 1971. The concept was also applied to litigation in fair housing under the Civil Rights Act of 1968 as well as in voting rights. 73 3.2.1 Using Multiple Measures The goal of the Title 5 revisions is therefore to mitigate disproportionate impact on access to college-level courses for underrepresented student populations through the use of multiple measures. The California Code of Regulations now prohibits community colleges from using “any single assessment instrument, method or procedure, by itself, for placement” (55521). In addition to standardized test scores, multiple measures can include measures of a student’s prior academic achievement and other noncognitive attributes, such as educational goals or motivation. The information can be gathered through writing samples, performance-based assessments, surveys and questionnaires, student self-evaluations, counseling interviews during the enrollment period, or other processes (CCCCO, 2011; Melguizo et al., forthcoming). The Title 5 regulations mandate the use of multiple measures but allow for consid- erable local autonomy in implementation. They do not formalize a specific statewide assessment and placement process, nor do they delineate the specific multiple measures to be used; the selection of multiple measures in California’s community colleges is en- tirely determined by the local institution or district. Even within one district, such as the LACCD, there is a wide variety of measures used and little standardization across col- leges. Researchers, policymakers, and practitioners alike are therefore concerned with the selection and validation of measures that are useful for making placement decisions. 3.3 Literature Review Absent alignment between the K-12 and higher education systems, community colleges need some means of identifying students’ preparedness for college-level work. How- ever, with neither a common definition of college readiness nor a common approach to remediation, a variety of measures are utilized to identify student skill level and college 74 preparedness (Merisotis and Phipps, 2000). These measures often include standardized placement test scores and information from high school transcripts, as well as informa- tion gleaned from student interviews with counselors. An important task for researchers has been to identify and validate measures that are predictive of college success. Validation has generally involved testing a group of subjects for a certain construct, and then comparing them with results obtained at some point in the future, such as college persistence, grades, or completion of a college credential (American Educational Research Association et al., 1999). This approach is known as predictive validity, which is the ability of a measure to predict future outcomes given present information. Here, we review the literature on the predictive validity of common measures used to identify readiness for college-level work. 3.3.1 Placement Tests Standardized placement tests are the most common instruments that community colleges use to assess students and deem them college-ready or place them in developmental math courses. These placement tests, many of which are now computerized, can be less time-consuming and resource-intensive than interviews or reviews of individual applications and transcripts. The computerized format can also enable colleges to assess many students and provide course placement results more quickly. There is considerable variation in the types of tests used across colleges, but ACCUPLACER and COMPASS, two commercially produced tests, are among the most common (Hughes and Scott- Clayton, 2011). Commercially-produced tests, such as ACCUPLACER, generally provide predictive validity estimates for their products (Mattern and Packman, 2009). However, individual colleges are advised to conduct validations within their own settings and with respect to their uses of the assessments. In an examination of validation practices across the 75 United States, Fulton (2012) found that colleges vary in terms of how they validate their placement tests, with only a handful of states or college systems having validation requirements. Research studies have provided some evidence that placement tests have low predic- tive validity, finding weak correlations between placement tests and students’ course passing rates and college grades (Armstrong, 2000; Medhanie et al., 2011; Scott- Clayton, 2012). After looking at the predictive validity of placement tests across the Virginia Community College System, Jenkins et al. (2009) found only weak correla- tions between placement test scores and student pass rates for both developmental and college-level courses. These findings may reflect the fact that college readiness is a function of several academic and non-academic factors that placement tests do not ad- equately capture. In fact, Belfield and Crosta (2012) found that the positive but weak association between placement test scores and college GPA disappeared after control- ling for high school GPA, suggesting that high school information may offer more useful measures for course placement. 3.3.2 High School Information While standardized placements tests are the most common instruments that commu- nity colleges use to assess and place students in developmental math courses, there is growing interest in incorporating high school information into the placement decision. High school transcripts can provide information about academic competence, effort, and college-readiness that placement tests do not measure. High school grades have been found to better predict student achievement in college than typical admissions tests do (Geiser and Santelices, 2007; Geiser and Studley, 2003), and this relationship may be even more pronounced in institutions with lower selectivity and academic achievement 76 (Sawyer, 2013). This may stem from the ability of report card grades to assess compe- tencies associated with students’ self-control, which can help students study, complete homework, and have successful classroom behaviors (Duckworth et al., 2012). In the community college setting, measures of prior math course-taking, such as the number of high school math courses, grades in high school math courses, and highest level of math taken have been found to be better predictors of achievement than place- ment test score alone (Lewallen, 1994). Adelman (2006) demonstrated that a composite of student performance (i.e., GPA or class rank and course-taking), what he referred to as students’ “academic resources,” can be useful information for identifying readi- ness for college-level work and can be highly predictive of college success. DesJardins and Lindsay (2007) confirmed these findings in subsequent analyses. Similar work in California demonstrates that scores on the California High School Exit Exam and high school transcript information are also predictive of math readiness (Jaffe, 2012; Long Beach Promise Group (LBPG), 2008). This type of evidence has led some community colleges to partner with local school districts and experiment with using high school information in developmental course placement (Fain, 2013; LBPG, 2008). Hesitation to use high school background information for placement purposes may be due to concerns about the consistency of these measures. High school graduation, for example, is not widely accepted as evidence of college readiness because of the wide variability in the quality of high school experiences (Sommerville and Yi, 2002). Also, there is no common metric or meaning across all high schools in regards to student performance and course-taking (Porter and Polikoff, 2012). Grades and summative as- sessments from high school vary both in rigor and breadth of content, making them more difficult for colleges to use systematically as college readiness indicators (Maruyama, 2012). 77 Nonetheless, the empirical evidence described above points towards certain com- binations of measures that may be the strongest predictors of college performance. Belfield and Crosta (2012) finding that such measures as prior math background in con- junction with high school GPA are strongly associated with college outcomes, hypothe- sized that “the optimal decision rule may be to combine information from a placement test with a high school transcript,” (p.4). Similarly, Noble and Sawyer (2004) argued that test scores, high school grades, and other measures could be used jointly to identify students who are ready for college-level work. 3.3.3 Noncognitive Measures Research in educational psychology further suggests that an array of factors beyond cognitive intelligence and skills are predictive of college success and future outcomes. Heckman et al. (2006) argue that noncognitive measures of adjustment, motivation, and perception are strong predictors of success, particularly for under-represented minority students. In a longitudinal study of community college students, Porchea et al. (2010) found an integration of psychosocial, academic, situational, and socio-demographic fac- tors to be predictive of persistence and attainment, with motivation being among the strongest predictors of future achievement. This may be due to the ability of these vari- ables to capture the effect of unobserved student characteristics associated with success, such as the importance of college to a student, preference and perseverance towards long-term goals, effort, and self-control (Duckworth et al., 2007, 2012). Given these findings, there is increasing interest in and advocacy for using noncog- nitive measures for course placement, which may provide colleges with a vital source of holistic student information (Boylan, 2009; Hodara et al., 2012). The ACT and ETS, for example, have developed noncognitive assessments such as the ACT ENGAGE assess- ments and the ETS Personal Potential Index (ACT, 2012; ETS, 2013), which identify 78 noncognitive attributes associated with student success in college and are predictive of student performance and persistence (Allen et al., 2008; Robbins et al., 2006). In prac- tice however, very few institutions use noncognitive measures for placement purposes (Gerlaugh et al., 2007; Hughes and Scott-Clayton, 2011). This may be due to faculty perceptions that self-reported student information is inaccurate or irrelevant (Melguizo et al., forthcoming; Perin, 2006), or to the lack of evidence about their ability to improve placement decisions. 3.3.4 Using Multiple Measures for Course Placement This scan of the literature reveals that while researchers have identified cognitive and noncognitive measures that are strongly associated with and predictive of student out- comes, there is relatively scant evidence showing that using these measures to make course placement decisions would be beneficial. This is an important distinction be- cause even though there may be a strong positive correlation between a measure such as high school GPA and passing the course in which a student enrolled (predictive valid- ity), we cannot conclude that the same relationship would hold if that student was placed into a course under a decision rule that incorporated GPA as a placement measure. Scott-Clayton et al. (2012) examined both district- and state-wide community col- lege data and estimated that placement using high school GPA instead of tests would significantly reduce the rate of severe placement errors in both developmental math and English courses. Aside from these prediction-based estimates, the only empirical evi- dence on actual placement decisions has come from institutional research, such as one experimental study that utilized a randomized design to determine the impact of differ- ent placement schemes. Marwick (2004) found that Latino students in one community college who were placed into higher-level courses due to the use of multiple measures (high school preparation and prior math coursework) achieved equal and sometimes 79 greater outcomes than when only placement test scores were considered. Another re- port of an on-going study by the Long Beach Promise Group (2008) shows that students who were placed in courses via a “predictive placement” scheme based on high school grades instead of test scores spent less time in developmental courses and were more likely to complete college-level English and math courses. Overall, there is limited use of multiple measures during assessment and placement for developmental math, and this may stem from a lack of evidence about their ability to improve placement decisions. Furthermore, qualitative research has found that faculty and staff often do not feel supported in the identification and validation of measures that can be incorporated into placement rules, while others perceive measures besides test scores to be insignificant (Melguizo et al., forthcoming). Given the numerous studies demonstrating the predictive validity of these other measures, it is important to gather evidence on the usefulness of measures for making course placement decisions. This involves a process of validation, which is described next. 3.4 Conceptual Framework 3.4.1 Validation The Title 5 multiple measures mandate in California provides a unique opportunity to validate measures in terms of their usefulness for making course placement decisions. This approach is in line with modern conceptions of validation, which emphasize not just accurate predictions, but actual success. Under this theory of validation, the validity of a measure such as a placement test is not a property of the test itself but a property of an inference or proposed use of the test (American Educational Research Association et al., 1999; Kane, 2006; Sawyer, 2007). A validation argument thus considers the goals and uses of a measure to be more important than its predictive properties. 80 This approach to validation emphasizes the examination of outcomes that result from proposed uses. In seeking to justify the use of a measure, it is necessary to demonstrate that the positive consequences of use outweigh any negative consequences. If the in- tended goals are achieved, then policies can be considered as successes. If goals are not achieved, then polices would be considered as failures (Kane, 2006). The measures used to make course placement decisions in developmental math would therefore be evaluated in terms of student outcomes – placement and success in the highest-level course possible, and the frequency with which these accurate place- ments occur (Sawyer, 1996). Following this validation approach, measures used for placement would be considered helpful if they place students in a level of coursework where they are likely to be successful, and harmful if students are placed in a level where they are unlikely to be successful. We next expand this validation argument to consider the use of multiple measures in conjunction with test scores to make course placement decisions. 3.4.2 Placement Decisions Assume that a math assessment enables us to make inferences about the academic prepa- ration of a math student. Students who receive low scores have low academic prepara- tion and students with high scores have high academic preparation. A typical placement policy would sort students into various math levels based on scores from this math as- sessment. For a simple model, let s L = Student with low academic preparation,C L = Low-level course, s H = Student with high academic preparation,C H = High-level course. 81 Let P be the probability of successfully passing the course. Given the two types of students and two types of courses, we would expect two sets of inequalities to hold. First, P(s L C L ) ≥ P(s L C H ) andP(s H C L ) ≥ P(s H C H ), (3.1) such that the probability of passing a low-level course is greater than the probability of passing a high-level course, for both types of students separately. Additionally, P(s H C L ) ≥ P(s L C L ) andP(s H C H ) ≥ P(s L C H ), (3.2) such that the probability of passing a given course is higher for a high academic prepara- tion student than for a low academic preparation student. Transitivity should predict that P(s H C L ) ≥ P(s L C H ). As a result, there are only two possible monotonic distributions: P(s L C H ) ≤ P(s L C L ) ≤ P(s H C H ) ≤ P(s H C L ) and (3.3) P(s L C H ) ≤ P(s H C H ) ≤ P(s L C L ) ≤ P(s H C L ) (3.4) If the raw assessment test score correctly places students in the appropriate math courses (i.e., cutoff scores are correct), every low academic preparation student should be placed into the low-level course and every high academic preparation student should be placed into the high level course. The placements(s H C L ) and(s L C H ) should not occur. 3.4.3 Placement Using Multiple Measures Including multiple measures can be thought of as increasing collateral information, which should improve the accuracy of placement decisions (van der Linden, 1998). 82 Consider a decision in which other relevant information from multiple measures is in- cluded and students can earn additional points which are added to the raw test score. In some cases, students identified as low academic preparation by the raw test score may be placed higher if the total score with additional points surpasses the cutoff score. This multiple measure boost thus places the low academic preparation student into the higher-level course. The boost is equivalent to (s L C H ). The boosted students would have had the highest scores on the placement test had they remained in the lower level. As a result of the multiple measure boost, they are now the lowest-scoring students in the higher-level course. The question of interest is whether the boosted students are equally likely to succeed when compared with other students in the higher-level course despite having lower raw placement test scores. Following the approach to validation suggested by Kane (2006), the multiple measure boost can be considered as helpful if boosted students are at least as likely to pass the higher-level course as their comparable peers. Should the boost be helpful, then there is an increase in placement accuracy. 5 The boost is harmful if the boosted students are less likely to pass the high-level course than their peers. In this case, the student would be better served if placed in the lower-level course. Empirically, the comparison of probabilities is between P(s L C H ) and P(s H C H ), where the boosted student is compared with other non-boosted students in the high-level course. Hence, the multiple measure boost can be considered as helpful if P(s L C H ) ≈ P(s H C H ) and harmful if P(s L C H )<P(s H C H ). 6 5 Although the boosted students can have a higher probability of passing the high-level course than the low-level course, we would have to expand the simple model to account for unobservable factors such as easiness of grading or grade inflation at the classroom level to attainP(s L C H )>P(s L C L ). 6 While it is possible for the boosted students to have a greater probability of passing the high-level course than the high academic preparation students, we would have to expand the simple model to include unobservables such as diligence to alter the direction of the inequality toP(s L C H )>P(s H C H ). 83 We use this validation argument to proceed with our analysis of student outcomes in the Los Angeles Community College District, a context where multiple measures are used in conjunction with test scores to inform placement decisions in developmental math. 3.5 Multiple Measures in the LACCD The LACCD is composed of nine community colleges serving nearly 250,000 students annually, making it the largest community college district in California and one of the largest in the country. According to our calculations, nearly 80 percent of students en- tering the LACCD each year are placed in developmental math courses. In most of the colleges, the developmental math sequence is comprised of four courses and includes arithmetic, pre-algebra, algebra, and intermediate algebra. This means an entering stu- dent can be placed several levels below college-level, extending time towards degree or certificate attainment. Students seeking to enroll in degree-applicable or transfer-level math courses in one of the LACCD colleges take an assessment test to determine course placement. The LACCD colleges have opted to use the ACCUPLACER, COMPASS, or Mathematics Diagnostic Testing Program (MDTP) to assess and place students. The ACCUPLACER and COMPASS are computer-adaptive standardized tests developed by College Board and ACT, respectively. The MDTP, a joint project of the California State University and the University of California, is a set of math diagnostics designed to measure stu- dent readiness for mathematics. During the period of this study, 2005-2008, five of the LACCD colleges used the ACCUPLACER, two of the colleges used COMPASS, and two colleges used the MDTP to make course placement decisions. 84 Each of the three placement tests are comprised of sub-tests, separated by levels of difficulty. The ACCUPLACER has arithmetic (AR), elementary algebra (EA), and college level math (CLM) sub-tests. When a student starts out in the AR sub-test and performs well, the computer-adaptive branching mechanism refers the student up to the EA sub-test. Conversely, when a student starts out in the CLM sub-test and performs poorly, the branching mechanism refers the student down to the EA sub-test. 3.5.1 Variation in Placement Policies Per the regulations of the Title 5 revisions to the Matriculation Act, California com- munity colleges must use multiple measures instead of a sole assessment instrument to place students in mathematics coursework. However, colleges are afforded autonomy in determining which measures to consider, provided the measures are not discrimina- tory (i.e., based on race, ethnicity, or gender). Some manuals provide guidance on how to appropriately select and validate measures at the institutional level (CCCCO, 1998, 2011; Lagunoff et al., 2012), but the devolved autonomy has resulted in considerable variation in the multiple measures utilized across the LACCD (Melguizo et al., forth- coming). Most often, information is collected through a survey taken before or after the assessment test and points are rewarded or even deducted for various responses. These are combined with the student’s placement test score and result in a final score used to make a course placement recommendation based on each college’s set of cutoff scores. Table 3.1 shows the multiple measures used to supplement student placement test scores in eight of the nine LACCD colleges for which multiple measures information was avail- able. As Table 3.1 shows, each college has also chosen to utilize a different combination of multiple measures. For example, College C awards a varying amount of points for 85 Table 3.1: Multiple Measure Points for Math Placement College Point Range Academic Background College Plans Motivation Diploma HS GPA Math English A 0to4 + B 0to3 + C 0to3 + + + + + D 0to4 + E -2 to 5 + +/- +/- F 0to2 + G -2 to 2 +/- +/- Notes: (+) indicates measures for which points are added, and (-) indicates measures for which points are subtracted. College plans include hours planned to attend class, hours of planned employment, and time out of formal education. Motivation includes importance of college and importance of mathematics. Multiple measure information not available for two of the nine LACCD colleges. college plans, high school GPA, previous math courses taken, and prior English. Fur- thermore, while most of the schools add multiple measure points to the test score, two schools in LACCD subtract points for selected responses. College G gives points for college plans (which include the number of units a student plans to take and the number of hours they plan to work while taking college classes), and the degree to which college is important to the student (motivation), an example of a noncognitive measure. It also deducts points if the student is a returning student but has not been enrolled for several years. It is important to note that at no time during the assessment process are students made aware of the college’s cut scores or the formula used for placement. Given these assessment and placement rules in the LACCD, the addition or sub- traction of multiple measure points can sometimes be the determining factor in course placement. The multiple measure points awarded can be enough to place students into a higher-level course or place them into a lower-level course. As described earlier, stu- dents are considered to have received a multiple measure boost if the additional multiple measure points placed the students in a math course one level higher than they otherwise 86 would have been by raw test score alone. Although there are two colleges that use mul- tiple measure information to subtract points and drop students down into a lower-level course, this does not happen frequently enough to warrant further investigation. 7 3.6 Data & Methods We obtained the data used for the study through a restricted-use agreement with the LACCD. We examined the assessment and enrollment information for all students who took a placement test between the 2005/06 and 2007/08 academic years. Transcripts provided outcome data through the spring of 2012, which resulted in seven years of outcome data available for the 05/06 cohort, six years for the 06/07 cohort, and five years for the 07/08 cohort. For the access analysis, we restrict the sample to five colleges which either use the ACCUPLACER or MDTP as placement tests. The full sample of assessed students for these five colleges between 2005 and 2008 includes 63,173 students. The rich assessment data enable us to identify each student whose raw test score was below the cutoff score at the institution in which they took the placement test, but whose multiple measure points resulted in an adjusted test score that was above the cutoff score. Students who met these criteria were coded as having received the multiple measure boost. This enabled us to determine the total number of students who received a multiple measure boost in each college between 2005 and 2008, as well as examine the number of boosted students by college and level of developmental math. 7 In College E, only 27 out of 4,303 students earned negative multiple measure points, and of those, only 2 were placed in a lower-level course. 87 3.6.1 Multivariate Regression To estimate the association between multiple measures and student success outcomes, we used linear probability regression models to compare the outcomes of students who were boosted into a higher-level course due to added multiple measure points with stu- dents whose higher test scores placed them directly into a course. The short-range out- come of interest is a dichotomous variable indicating whether or not the student passed the first enrolled math course withaCor better. Scott-Clayton et al. (2012) noted the potential controversy of using earningaCasan outcome since developmental educators and policy-makers may think of getting C as a mediocre achievement. However, since students who earn a C complete the prerequisite and can move on to the next course, we believe that earning a C is an appropriate short-term outcome for examining placement accuracy. 8 The transcript data also allow us to examine two important longer-term outcomes for community college students—total of number of degree-applicable units completed and total number of transfer-level units completed. Degree-applicable units are those which can be applied towards an associate’s degree, and transfer-level units are those which would be accepted at a California four-year university. The linear probability regression model is: y i = α+βBOOST i +γMMPOINTS i +ηX i + i (3.5) where y i is the outcome of interest. The treatment variable of interest is BOOST i ,a dichotomous variable indicating whether or not the student received multiple measure points that resulted in a boost to the next highest level math course. MMPOINTS i is 8 The grade distribution in our two focus colleges are reported in Appendix Table A.3.1. We believe that earningaCor better is still a worthy accomplishment since nearly 50% of the students do not achieve C. A large proportion of students choose to withdraw from the class after enrollment. 88 the number of multiple measure points a student received, and X i is a vector of student information including age, race, sex, raw test score, and assessment cohort. We control for age, race, sex, assessment cohort, and the number of total multiple measure points to obtain a more precise estimate of the multiple measure boost. We also include raw test score and test type to control for any variation that may be associated with the placement test itself. 3.6.2 Two Focus Colleges We focus on the effect of the multiple measure boost in two LACCD colleges: College A, which awards multiple measure points based solely on a student’s prior math back- ground, and College D, which awards multiple measure points based solely on a stu- dent’s self-reported high school GPA. Since the multiple measure boost is determined by a single measure in addition to the placement test score, we can determine the effec- tiveness of that specific measure in increasing placement accuracy. College A awards one point for each of the following prior math background measures: the highest level of math previously taken with a grade of C or better (+1 for trigonometry or higher), the number of years of math taken in high school (+1 for three years or more), the length of time since math was last taken (+1 if less than one year), and whether or not the student has taken algebra (+1). Students who take the placement ACCUPLACER test at College A can score a maximum of 120 points and earn a maximum of four multiple measure points. College D awards two additional points for a high school GPA in the B to B- range, and four additional points for a high school GPA in the A to A- range. Students who take the MDTP placement test at College D can score between 40 and 50 points depending on the last sub-test that they take. 89 3.6.3 Comparison Groups We run two linear probability regression models for each level of developmental math in the two colleges. First, we compare boosted students to other students whose raw test scores are similar to the adjusted test scores of the boosted students. In the second model, we include all students within a given course level. To illustrate this, consider College A, for which the cut score for placement in pre-algebra is 35 on the ACCU- PLACER AR sub-test. Students who attain a score of 35 and above are placed in pre- algebra (three levels below transfer) while students scoring below 35 are placed in arith- metic (four levels below transfer). The multiple measure boost could have pushed a stu- dent from arithmetic to pre-algebra if the addition of multiple measure points pushed the adjusted ACCUPLACER score (raw score + multiple measure points) to 35 or above. For these boosted students, the range of raw AR scores is 31 ≤ AR r ≤ 34.9 with a maximum of four multiple measure points. Their resulting adjusted AR score is 35 ≤ AR a ≤ 38.9. In the first regression model (Similar), we compare the boosted students with 35 ≤ AR a ≤ 38.9 to the nonboosted students whose raw AR tests scores were in the range 35 ≤ AR r ≤ 38.9. In the second regression model (Entire), we compare the boosted students to the entire range of students in the same course level. We recognize the weaknesses in both of the models. By restricting the the sample in the first (Similar) model via an endogenous BOOST variable, the estimates may be biased. While we can attempt to instrument for BOOST status, the only relevant student demographic information available are gender and ethnicity. By including all students in the entire level in the second model, we are also likely to be comparing a very het- erogeneous group of students. In College A, students can get placed into pre-algebra with a score of 35 ≤ AR a < 65. This spread of 30 points is an extremely wide range, considering the average AR score is 32.5 with a standard deviation of 16.3 points. Thus a student scoring 64.9 points on the AR sub-test is highly unlikely to have the same set 90 of math skills as a student scoring 35 points, and those students with higher test scores within a given level may have a higher probability of passing the course regardless of their multiple measure points. 3.6.4 Potential Selection Bias A separate issue too broad in scope to discuss in this study is the non-enrollment of test- takers. Non-enrollment after assessment may account for much of the low completion and persistence rates in developmental math sequences. The proportion of students who are assessed but do not attempt courses can even be higher than those who attempt and complete courses (Fong et al., 2013). Among the students in College A, only 72 percent of all the students who take the ACCUPLACER enroll in a math course. Only 65 percent enroll in a math course within a year of assessment, and slightly less than 63 percent enroll in the math course in which they are placed. Similarly, in College D, only 78 percent of all the students who take the MDPT enroll in a math course, and only 73 percent enroll in the math course in which they are placed. In terms of this study, the attrition is of concern if there is selection bias. If those students who do not choose to enroll in a class are systematically different from those who do enroll, then any estimates using only the enrolled students may be biased. Since the treatment in our analysis is BOOST , we should be concerned about attrition if enrollment is unbalanced between the nonboosted and boosted students. Table 3.2 shows the differences in attempting the placed math course among the non- boosted and boosted students, along with the differences in receiving the boost among nonenrolled and enrolled students. In both colleges, there is no statistical difference in either of the groups. This is the desired result since bias would be introduced if students enrolled or attempted the placed math course based on receiving a multiple measure boost. There are several significant differences in terms of demographic characteristics 91 between the students who enroll in their placed math course versus those who do not, as well as between those who are boosted versus those who are not. However, there appears to be no relationship between receiving a multiple measure boost and enrolling in a math course. We therefore proceed with the enrolled students only. In particular, we restrict the sample to those who enrolled within a year of assessment and attempted the math course in which they were placed. This provides an estimate of the treatment on the treated effect. The one year time window is intended to allow sufficient time for students to enroll in courses as well as to limit the change in the mathematical knowledge of the students since the assessment. Rather than assigning a value of zero for passing the attempted math course for those who never enrolled and calculating the intent to treat effect, we reasoned that students being unaware of the placement process and whether or not a boost was received would help to limit selection bias. Students are not informed of the placement criteria or placement rules of the college. After the placement test, the student simply receives a summary of their score and a course recommendation. Importantly, the student does not know if placement into a particular level was the result of a multiple measure point boost. 3.7 Findings 3.7.1 Multiple Measures and Access to Higher-Level Courses Table 3.3 shows the percentage and the number of students boosted into a higher-level course due to the multiple measure reward structure at five LACCD colleges that used either MDTP or ACCUPLACER for assessment and placement. Overall, about 5.2 per- cent of all students in this sample were boosted to the next level course between 2005/06 and 2007/08 academic years. That is, although their raw test score would have placed 92 Table 3.2: Boosted and Enrolled Students (a) College A Non-boosted vs. Boosted Non-enrolled vs. Enrolled Non-boosted Boosted Difference Non-enrolled Enrolled Difference Enrolled: Attempted Placed Math First 0.636 0.610 0.0261 Boosted: Boosted Up 0.0654 0.0589 0.00646 Demographics: Male 0.473 0.478 -0.00576 0.502 0.457 0.0451 ∗∗∗ Asian 0.200 0.202 -0.00153 0.237 0.179 0.0578 ∗∗∗ African-American 0.128 0.125 0.00237 0.146 0.117 0.0281 ∗∗∗ Latino/a 0.436 0.469 -0.0322 0.369 0.478 -0.109 ∗∗∗ White 0.164 0.147 0.0172 0.172 0.158 0.0142 Other 0.0710 0.0569 0.0142 0.0759 0.0669 0.00902 20-30 0.392 0.325 0.0663 ∗∗ 0.417 0.371 0.0460 ∗∗∗ 30-40 0.127 0.0882 0.0391 ∗∗ 0.159 0.105 0.0542 ∗∗∗ Over 40 0.0864 0.0784 0.00796 0.115 0.0693 0.0453 ∗∗∗ English at home 0.568 0.549 0.0194 0.570 0.565 0.00485 Spanish at home 0.208 0.241 -0.0327 0.172 0.233 -0.0605 ∗∗∗ US Citizen / PR 0.852 0.835 0.0164 0.842 0.856 -0.0142 Observations 7813 510 8323 3044 5279 8323 (b) College D Non-boosted vs. Boosted Non-enrolled vs. Enrolled Non-boosted Boosted Difference Non-enrolled Enrolled Difference Enrolled: Attempted Placed Math First 0.733 0.691 0.0422 Boosted: Boosted Up 0.0306 0.0251 0.00556 Demographics: Male 0.481 0.367 0.114 ∗∗∗ 0.506 0.468 0.0384 ∗∗∗ Asian 0.187 0.149 0.0381 0.161 0.195 -0.0337 ∗∗∗ African-American 0.0648 0.0509 0.0139 0.0926 0.0541 0.0385 ∗∗∗ Latino/a 0.344 0.320 0.0236 0.338 0.345 -0.00655 White 0.291 0.367 -0.0764 ∗∗ 0.295 0.292 0.00323 Other 0.114 0.113 0.000832 0.112 0.114 -0.00145 Under 20 0.685 0.669 0.0158 0.639 0.701 -0.0621 ∗∗∗ 20-30 0.221 0.222 -0.000831 0.237 0.215 0.0217 ∗ 30-40 0.0525 0.0582 -0.00567 0.0635 0.0487 0.0148 ∗∗ Over 40 0.0416 0.0509 -0.00931 0.0606 0.0350 0.0256 ∗∗∗ English at home 0.724 0.702 0.0218 0.744 0.715 0.0292 ∗∗ Spanish at home 0.109 0.131 -0.0214 0.114 0.109 0.00527 US Citizen / PR 0.907 0.905 0.00124 0.915 0.903 0.0118 Observations 10074 275 10349 2774 7575 10349 mean coefficients ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 them in the lower course, the addition of multiple measure points caused them to surpass 93 the cutoff score and be placed into a higher-level course. Although the percentages vary by college, very few students overall are moved to higher-level courses. Table 3.3: Students Receiving Multiple Measure Boost into Higher-Level Courses (a) % AR to PA PA to EA EA to IA IA to CLM Total A 5.60 7.86 9.15 13.20 6.13 C 0.00 3.95 2.93 3.26 0.87 D 1.97 1.71 2.43 6.09 2.66 E 27.82 20.10 15.44 20.00 13.76 F 3.50 6.29 5.49 4.23 3.80 Total 5.71 7.66 6.08 6.20 5.17 Observations 16580 16227 12743 4858 63173 (b) Number AR to PA PA to EA EA to IA IA to CLM All Boosted Total Boosted Total Boosted Total Boosted Total Boosted Total A 360 6434 278 3536 292 3192 90 682 1020 16646 C 5722 48 1216 24 818 6 184 78 8940 D 28 1420 65 3803 79 3245 103 1692 275 10349 E 518 1862 538 2676 122 790 6 30 1184 8606 F 40 1142 314 4996 258 4698 96 2270 708 18632 N 946 16580 1243 16227 775 12743 301 4858 3265 63173 Arithmetic (AR); Pre-Algebra (PA); Elementary Algebra (EA); Intermediate Algebra (IA); College Level Math (CLM) One explicit goal of the Title 5 revisions was to mitigate disproportionate impact on the number of underrepresented minority students being placed into remediation. To examine disproportionate impact, we calculated math placement rates for each racial subgroup using the adjusted test scores including multiple measure points. We then simulated counterfactual course placements for each student by using unadjusted test scores without multiple measure points. Placement rates are provided for two colleges, A and D, which are also the subject of our multivariate analyses described below. These two colleges each use one type of additional measure for course placement: prior math course-taking in College A and self-reported high school GPA in College D. 94 We present the disproportionate impact results in two ways. First we looked at the overall course placements by race. Then, we show the distribution of students by race within each level of developmental math. Comparing the actual placements with the simulated counterfactual placements both ways enabled us to determine the extent to which the use of multiple measures mitigated disproportionate impact of remediation by racial subgroup. Table 3.4 shows the results of the simulated placements without multiple measure points and actual placement with multiple measure points by level of developmental math for Latino and African-American students. Table 3.5 shows placement by racial subgroups within pre-algebra. 9 Table 3.4: Placement by Level within African-American and Latino Students (a) African-American College A College D w/o MMP w/ MMP Difference w/o MMP w/ MMP Difference Arithmetic 0.337 0.303 0.0339 ∗ 0.0568 0.0502 0.00655 Pre-Algebra 0.403 0.426 -0.0235 0.225 0.225 0 Algebra 0.151 0.150 0.00188 0.410 0.415 -0.00437 Intermediate Algebra 0.0988 0.107 -0.00847 0.253 0.255 -0.00218 College Level Math 0.0103 0.0141 -0.00376 0.0546 0.0546 0 Observations 1063 1063 458 458 Pearson (p-value) 0.461 0.995 (b) Latino/a College A College D w/o MMP w/ MMP Difference w/o MMP w/ MMP Difference Arithmetic 0.213 0.189 0.0241 ∗∗ 0.0237 0.0198 0.00395 Pre-Algebra 0.449 0.451 -0.00192 0.176 0.174 0.00216 Algebra 0.227 0.231 -0.00466 0.412 0.412 0.000359 Intermediate Algebra 0.101 0.116 -0.0151 ∗∗ 0.316 0.315 0.000359 College Level Math 0.0101 0.0126 -0.00247 0.0722 0.0790 -0.00682 Observations 3649 3649 2784 2784 Pearson (p-value) 0.0363 0.755 mean coefficients ∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.01 9 Refer to Appendix Tables A.3.2 and A.3.3 for complete distributions by ethnicity and by level. 95 Table 3.5: Placement by Ethnicity within Pre-Algebra College A College D w/o MMP w/ MMP Difference w/o MMP w/ MMP Difference Asian 0.114 0.113 0.00166 0.139 0.143 -0.00411 African-American 0.135 0.141 -0.00618 0.0899 0.0920 -0.00209 Latino/a 0.516 0.512 0.00391 0.428 0.432 -0.00457 White 0.164 0.163 0.00101 0.235 0.224 0.0106 Other 0.0714 0.0718 -0.000400 0.109 0.109 0.000146 Observations 3179 3217 1146 1120 80% Majority Ethnicity 0.412 0.409 0.342 0.346 80% White 0.131 0.131 0.188 0.179 Pearson (p-value) 0.969 0.982 mean coefficients ∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.01 The results indicate that the use of multiple measures as currently operationalized in California only marginally increases the number of underrepresented minority students being placed in higher levels of math. For example, in Table 3.4 we see that under College A’s multiple measure policy, about 3.4 percent fewer African-American and 2.4 percent fewer Latino students were referred to arithmetic, the lowest-level course in the developmental math sequence. There was also a 1.5 percent increase in the number of Latino students being placed in Intermediate Algebra, the highest-level course in the developmental math sequence. Although the use of multiple measure points increased access to higher-level courses for African-American and Latino students, the results in Table 3.5 show that the overall racial composition of math classes remains largely unchanged, with no statistical dif- ference even at the 10% level. We only present the distribution of students by racial subgroups within pre-algebra, but the results are similar for all math levels. This evi- dence suggests that despite the current use of multiple measures in the LACCD colleges, there continues to be disproportionate impact in assignment to remediation. 96 3.7.2 Multiple Measures and Student Success While this descriptive analysis offers some insight into the efficacy of multiple mea- sures in increasing access to higher-level math courses, one of the goals of the Title 5 revisions, it is also important for community colleges to design and use assessment and placement policies that promote student success. Students should be placed into courses where they are likely to succeed given their level of college readiness and math skills. A summary of our linear probability regression estimates examining student outcomes are presented in Tables 3.6 and 3.7. 10,11 Two sets of regression results are provided. In the “Similar” columns, we restricted the analytical sample to those students whose test scores were similar to the scores of the boosted students who took the same sub-test. 12 In the “Entire” columns, we included all students within the course level who took the same sub-test. 3.7.2.1 Prior math Table 3.6 shows that, all else constant, lower-scoring students in College A who received a multiple measure boost (based on prior math) that placed them in a higher-level course performed no differently from their peers in terms of passing the first math course they enrolled in. They also showed no difference in the total number of degree-applicable and transfer-level credits they completed through spring 2012. While some of the mag- nitudes of the differences are negative, the estimates are not statistically significant. We 10 The full regression results with coefficients for all predictors are provided in the Appendix Tables A.3.4-A.3.9. In these summarized tables, we show differences in outcomes between boosted students and their higher-scoring peers at each level of developmental math. 11 We did not examine outcomes for students boosted to college-level math since students have a range of course options available to them if they are not placed in developmental math and are allowed to take college-level math courses (e.g. pre-calculus, calculus, or statistics). 12 In some LACCD colleges, including Colleges A and D, students who take different ACCUPLACER or MDTP sub-tests may end up being placed in the same course level. 97 thus fail to reject the null hypothesis that there is no difference between the outcomes of boosted and non-boosted students. The results are consistent for all levels of develop- mental math and all placement sub-tests. 3.7.2.2 High school GPA Students in College D are awarded multiple measure points based on their self-reported high school GPA. The results in Table 3.7 indicate that, with respect to the linear prob- ability of passing the first enrolled math course and eventual credit accumulation, most of the differences between boosted students and their peers were not statistically sig- nificant. However, students who were placed into intermediate algebra by the MDTP Elementary Algebra sub-test had a 14.4 percent higher likelihood of passing the course than their higher-scoring peers (p<.05). These students also appear to have better long- term outcomes. Students who received the multiple measure boost and the opportunity to take intermediate algebra instead of elementary algebra, completed about 10.5 more degree-applicable and 9.5 more transfer-level credits than their higher-scoring peers in the same cohort (p<.05). It is interesting to note that results from both colleges are fairly consistent when comparing the boosted students to higher-scoring peers from two different test score ranges. Students receiving the multiple measure boost in both colleges neither have lower passing rates or credit completion than students who have similar test scores to their own, nor lower rates or credits than all the students placed in their same level of math. In fact, students placed in intermediate algebra instead of elementary algebra in College D on average outperformed all students with higher test scores with respect to both short-term and long-term outcomes. 98 Table 3.6: Regression Results, College A (a) Pass First Math Similar Entire Test=AR Test=EA Test=CLM Test=AR Test=EA Test=CLM PA PA EA IA IA PA PA EA IA IA Boosted Up -0.364 -0.127 0.152 -0.0497 -0.216 -0.155 -0.105 0.0631 -0.0892 -0.142 (0.209) (0.0720) (0.123) (0.136) (0.146) (0.102) (0.0537) (0.0604) (0.0624) (0.134) Observations 65 427 250 191 83 275 1877 1217 812 201 AdjustedR 2 -0.032 0.077 0.035 0.055 0.001 0.069 0.074 0.053 0.083 0.080 (b) Degree Credits Similar Entire Test=AR Test=EA Test=CLM Test=AR Test=EA Test=CLM PA PA EA IA IA PA PA EA IA IA Boosted Up -7.555 -0.592 1.338 2.756 -9.858 0.555 0.631 3.989 -1.601 -4.883 (11.82) (4.414) (7.328) (8.503) (9.562) (5.814) (3.130) (3.803) (3.998) (8.516) Observations 65 427 250 191 83 275 1877 1217 812 201 AdjustedR 2 -0.022 0.013 0.031 0.049 -0.032 0.027 0.039 0.036 0.039 0.083 (c) Transfer Credits Similar Entire Test=AR Test=EA Test=CLM Test=AR Test=EA Test=CLM PA PA EA IA IA PA PA EA IA IA Boosted Up -7.926 -1.981 1.501 1.852 -9.393 1.394 -1.451 2.531 -1.920 -3.890 (10.89) (3.497) (6.100) (7.638) (9.046) (5.218) (2.366) (3.163) (3.596) (8.097) Observations 65 427 250 191 83 275 1877 1217 812 201 AdjustedR 2 -0.019 0.004 0.028 0.041 0.022 0.018 0.037 0.031 0.025 0.076 se in parentheses ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 99 Table 3.7: Regression Results, College D (a) Pass First Math Similar Entire Test=AR Test=EA Test=AR Test=EA PA EA IA PA EA IA Boosted Up 0.0660 0.0397 0.108 -0.0517 0.00609 0.144 ∗ (0.161) (0.112) (0.0908) (0.117) (0.0783) (0.0622) Observations 321 333 349 1060 712 605 AdjustedR 2 0.019 0.058 0.039 0.057 0.082 0.052 (b) Degree Credits Similar Entire Test=AR Test=EA Test=AR Test=EA PA EA IA PA EA IA Boosted Up 17.61 -4.626 8.342 10.48 -3.843 10.80 ∗ (10.03) (6.820) (5.970) (8.045) (5.106) (4.583) Observations 321 333 349 1060 712 605 AdjustedR 2 0.051 0.015 0.077 0.028 0.037 0.056 (c) Transfer Credits Similar Entire Test=AR Test=EA Test=AR Test=EA PA EA IA PA EA IA Boosted Up 13.32 -3.192 7.110 8.596 -3.032 9.691 ∗ (8.375) (5.726) (5.465) (6.843) (4.313) (4.221) Observations 321 333 349 1060 712 605 AdjustedR 2 0.051 0.033 0.077 0.028 0.046 0.058 se in parentheses ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 3.7.3 Cautions Two cautions in interpretation must be noted. First, these findings may be sensitive to the correctness of each college’s specified cutoff scores. A possible interpretation of the results is that placement test cut scores are too low; any low-scoring student given the opportunity to take the next highest class would be just as successful as their higher- scoring peers, irrespective of multiple measure points. This would bias our estimate of the difference between boosted students and their peers. Studies using causal methods may be more efficacious in answering the question of correctly set cut scores (Melguizo 100 et al., 2013)). Second, variation in instructors’ grading practices may explain some of the similarities in student outcomes (Armstrong, 2000). Math faculty may adjust their instruction or grading in order to meet the needs of students, which would bias the estimate of the relationship between multiple measure boosts and student outcomes. To address these concerns, we make the assumption that instructors do not adjust their practices much in response to student academic preparation, and this is reasonable given that a relatively small percentage of students in any given course would have received the boost. 13 Furthermore, the non-significant estimates we obtained in this analysis are consistent across nearly all levels of developmental math, as well as in both colleges, which suggests that it is unlikely that cut scores are incorrect across the board or that instructor practices are systematically forgiving to students. Finally, our analytic approach consisted of two models – one in which we compared boosted students to students with similar placement test scores, and another in which we compared boosted students to all other students in their level within a given institution. These results demonstrated that boosted students were not only as successful as the students within a similar score range, but also as successful as students within the entire level. Given that students at a given level of developmental math can have differences as large as 30 points on their placement tests scores, we remain confident that our estimates of the effect of receiving a multiple measure boost are consistent and reliable. 13 As shown in Table 3.3, only 5.17% of students overall received the boost to the higher-level courses. 101 3.8 Discussion 3.8.1 Increasing Placement Accuracy This evidence from the LACCD is timely given the changing landscape of placement testing for developmental education. Several states have already or are in the process of revising developmental education assessment and placement policies to incorporate multiple measures. In Connecticut, new legislation requires colleges and universities to use multiple measures to determine whether students require college-readiness support (SB-40). In Florida, Senate Bill 1720 indicates that other “documented student achieve- ments such as grade point averages, work history,..., career interests, degree major declaration, or any combination of such achievements...” may be used in conjunction with placement test scores (Florida Senate Bill 1720, 1008.30 6(a)1). Likewise, North Carolina is reconsidering the use of placement testing given research findings about the ability of standardized placement tests to make accurate placements. The community college system is working with the College Board to develop a customized placement assessment that includes gathering information from multiple measures, such as high school grades and noncognitive measures (Burdman, 2012). The new Texas Success Initiative includes revised assessment and cut score standards, including the recom- mendation that additional multiple measures such as high school GPA, work hours, or noncognitive measures be considered in conjunction with assessment test scores (Texas Higher Education Coordinating Board (THECB), 2012). The underlying goal of these efforts is to improve placement accuracy in develop- mental math. In the context of math course placement, placement accuracy is maximized when students who are likely to pass a higher-level course are placed into the higher- level course, and students who are not likely to pass the higher-level course are placed 102 into the lower-level course. Recent studies of placement accuracy find that nearly one- quarter of students may be inaccurately placed by placement test scores in community college developmental math courses (Belfield and Crosta, 2012; Scott-Clayton et al., 2012). Finding that information from high school transcripts are strong predictors of college achievement, these studies hypothesize that using additional measures in con- junction with test scores may reduce the severity of placement errors. While these findings provide some evidence regarding the placement accuracy and error rates of placement tests and additional measures, the studies did not examine ac- tual placement policies that use multiple measures to make placement decisions. Our findings from California thus add to this evidence base. The results suggest that commu- nity colleges can increase placement accuracy by using multiple measure information in conjunction with placement test scores. Student placement and outcome data at the two LACCD colleges demonstrate that those students who were placed into higher-level de- velopmental math courses using multiple measures performed no differently from their higher-scoring peers. Since these students were given the opportunity to take a higher- level course and performed at least as well as their higher-scoring peers, these students were more accurately placed than they would have been by placement test score alone. Furthermore, the findings of this study are important because there is limited un- derstanding of what measures can be validly used to make course placement decisions. Qualitative research with community college faculty and staff has shown that practition- ers do not feel supported in measure selection and validation, and that they sometimes perceive measures to be insignificant (Melguizo et al., forthcoming). Even though co- ordinating entities such as the CCCCO and the THECB provide guides for multiple measure use, there is limited evidence and validation of measures that can be used to inform placement decisions. 103 This evidence from the LACCD provides validation for two specific measures – prior math background and high school GPA. Even though these measures are known to be predictive of college outcomes, current conceptions of validation highlight the need to examine actual outcomes in contexts where measures are used to make placement decisions. The findings from Colleges A and D indicate that these two measures can be systematically used to improve course placement decisions. Using them in conjunction with test scores can increase placement accuracy and may be, as Belfield and Crosta (2012) suggest, closer to the optimal decision rule for placement in developmental math. 3.8.2 Promoting Access and Success This examination of community college assessment and placement policies also high- lights the underlying tension between the goals of access and success when making placement decisions. Indeed, promoting progression versus maintaining standards is one of the “opposing forces” that community colleges often operate under (Jaggars and Hodara, 2011; Perin, 2006). Community colleges have the responsibility to place stu- dents in courses in which they are most likely to succeed given their math skills, while simultaneously promoting progression towards completion and attainment. The results of the study demonstrate that multiple measures can be utilized to achieve both of these goals. Based on this evidence from two LACCD colleges, students who received a multiple measure boost based on prior math and high school GPA took higher- level courses and succeeded in them at rates no different from their higher-scoring peers. Using these additional student background measures in conjunction with test scores to make course placement decisions may therefore achieve the goals of increasing access and ensuring student success. Nonetheless, while boosted students are just as likely to be successful as their peers, our analyses also show that the goals of mitigating dispropor- tionate impact in remediation are not being fully realized. The Title 5 policy explicitly 104 states that assessment practices should not result in disproportionate impact on any un- derrepresented minority group. Our simulated placements with and without multiple measures show that the use of these particular multiple measures only marginally in- creased access to higher level math courses for African-American and Latino students. Community colleges should therefore continue to explore other ways to improve assess- ment and placement such that disproportionate placement in remediation is mitigated while the likelihood of student success is maximized. Some research shows that noncognitive measures may be useful for identifying col- lege readiness and promoting access and success in college, particularly for underrepre- sented minority students (Sedlacek, 2004). The validity of these measures has yet to be explored in the context of developmental education. Even though some of the LACCD colleges’ placement rules include noncognitive measures such as college plans, edu- cational goals, availability of social supports, and motivation, these are often used in conjunction with other measures. We were thus unable to identify the singular effect of using these to make placement decisions. In addition, colleges that used these mea- sures also weighted multiple measures in such a way that, relative to Colleges A and D, very few students received a multiple measure boost. Further research should focus on validating other cognitive and noncognitive measures that can be useful for identi- fying incoming student readiness, specifically those that increase access to higher-level courses for underrepresented student populations. 3.9 Conclusion and Future Research The multiple measures policy in California provides an opportunity to validate measures in terms of their usefulness for course placement. The results of this study indicate that students who were placed into higher-level courses using information from multiple 105 measures, in this case high school GPA and prior math course-taking, performed no dif- ferently from their peers who earned higher test scores. This suggests that community colleges can systematically improve placement accuracy by using student background information in addition to assessment data to make initial course placement decisions. Such policies would increase access to higher-level math without decreasing students’ chances of success in the first math course in which they enroll or eventual credit accu- mulation. We do recognize that these data may be unavailable for non-traditional stu- dents or international students whom community colleges serve in substantial numbers. Further research should be done to identify and validate a broader range of measures that can improve placement accuracy for all types of students. Still, using multiple sources of information about incoming community college students not only increases access to higher-level math courses, but can also increase placement accuracy and ensure that students are placed at a level where they are likely to be successful. This ultimately can promote equity and efficiency in the assessment and placement process, accelerate college completion, and reduce the financial and academic burdens of postsecondary remediation. 106 Appendix 3 Table A.3.1: Grade Distribution (a) College A AR PA EA IA CLM ALL F or W 772 (55.38) 1903 (49.79) 1887 (44.59) 1694 (42.79) 1186 (36.57) 7442 (44.70) D 150 (10.76) 387 (10.13) 494 (11.67) 391 (9.876) 265 (8.171) 1687 (10.13) C 244 (17.50) 641 (16.77) 777 (18.36) 760 (19.20) 533 (16.44) 2955 (17.75) B 117 (8.393) 469 (12.27) 597 (14.11) 598 (15.10) 555 (17.11) 2336 (14.03) A 111 (7.963) 422 (11.04) 477 (11.27) 516 (13.03) 704 (21.71) 2230 (13.39) N 1394 3822 4232 3959 3243 16650 (b) College D AR PA EA IA CLM ALL F or W 72 (43.11) 551 (31.45) 1995 (39.41) 2230 (34.83) 2362 (32.78) 7228 (35.07) D 11 (6.587) 178 (10.16) 544 (10.75) 712 (11.12) 545 (7.564) 1990 (9.656) C 26 (15.57) 331 (18.89) 1007 (19.89) 1471 (22.97) 1282 (17.79) 4117 (19.98) B 28 (16.77) 318 (18.15) 795 (15.71) 1034 (16.15) 1450 (20.12) 3626 (17.59) A 30 (17.96) 374 (21.35) 721 (14.24) 956 (14.93) 1566 (21.73) 3648 (17.70) N 167 1752 5062 6403 7205 20609 column pct in parenthesis 107 Table A.3.2: Placements by Ethnicity within Level (a) Arithmetic College A College D w/o MMP w/ MMP Difference w/o MMP w/ MMP Difference Asian 0.0916 0.0864 0.00523 0.115 0.110 0.00436 African-American 0.226 0.230 -0.00368 0.166 0.169 -0.00351 Latino/a 0.490 0.491 -0.000869 0.420 0.404 0.0160 White 0.125 0.125 0.000168 0.178 0.191 -0.0128 Other 0.0670 0.0678 -0.000847 0.121 0.125 -0.00398 Observations 1583 1401 157 136 80% Majority Ethnicity 0.392 0.393 0.336 0.324 80% White 0.100 0.0999 0.143 0.153 Pearson (p-value) 0.991 0.998 (b) Pre-Algebra College A College D w/o MMP w/ MMP Difference w/o MMP w/ MMP Difference Asian 0.114 0.113 0.00166 0.139 0.143 -0.00411 African-American 0.135 0.141 -0.00618 0.0899 0.0920 -0.00209 Latino/a 0.516 0.512 0.00391 0.428 0.432 -0.00457 White 0.164 0.163 0.00101 0.235 0.224 0.0106 Other 0.0714 0.0718 -0.000400 0.109 0.109 0.000146 Observations 3179 3217 1146 1120 80% Majority Ethnicity 0.412 0.409 0.342 0.346 80% White 0.131 0.131 0.188 0.179 Pearson (p-value) 0.969 0.982 (c) Algebra College A College D w/o MMP w/ MMP Difference w/o MMP w/ MMP Difference Asian 0.214 0.209 0.00485 0.145 0.144 0.000813 African-American 0.0910 0.0899 0.00103 0.0636 0.0647 -0.00107 Latino/a 0.467 0.477 -0.0101 0.388 0.390 -0.00204 White 0.153 0.151 0.00209 0.296 0.295 0.00159 Other 0.0751 0.0730 0.00218 0.107 0.106 0.000709 Observations 1770 1768 2956 2938 80% Majority Ethnicity 0.374 0.382 0.311 0.312 80% White 0.122 0.121 0.237 0.236 Pearson (p-value) 0.984 0.999 (d) Intermediate Algebra College A College D w/o MMP w/ MMP Difference w/o MMP w/ MMP Difference Asian 0.399 0.380 0.0183 0.202 0.197 0.00473 African-American 0.0702 0.0714 -0.00119 0.0442 0.0448 -0.000602 Latino/a 0.247 0.266 -0.0188 0.335 0.336 -0.00128 White 0.216 0.212 0.00427 0.291 0.293 -0.00259 Other 0.0676 0.0702 -0.00262 0.129 0.129 -0.000257 Observations 1495 1596 2627 2614 80% Majority Ethnicity 0.319 0.304 0.268 0.269 80% White 0.173 0.169 0.233 0.235 Pearson (p-value) 0.757 0.996 (e) College Level Math College A College D w/o MMP w/ MMP Difference w/o MMP w/ MMP Difference Asian 0.632 0.616 0.0159 0.354 0.350 0.00418 African-American 0.0372 0.0440 -0.00683 0.0198 0.0186 0.00115 Latino/a 0.125 0.135 -0.00990 0.159 0.164 -0.00491 White 0.149 0.155 -0.00678 0.354 0.353 0.000452 Other 0.0574 0.0499 0.00758 0.113 0.114 -0.000871 Observations 296 341 1263 1341 80% Majority Ethnicity 0.505 0.493 0.283 0.280 80% White 0.119 0.124 0.283 0.283 Pearson (p-value) 0.965 0.996 mean coefficients ∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.01 108 Table A.3.3: Placements by Level within Ethnicity (a) Asian College A College D w/o MMP w/ MMP Difference w/o MMP w/ MMP Difference Arithmetic 0.0869 0.0725 0.0144 0.0114 0.00948 0.00190 Pre-Algebra 0.217 0.217 0.000599 0.100 0.101 -0.000632 Algebra 0.226 0.221 0.00539 0.271 0.268 0.00316 Intermediate Algebra 0.357 0.364 -0.00659 0.335 0.325 0.00948 College Level Math 0.112 0.126 -0.0138 0.282 0.296 -0.0139 Observations 1669 1669 1583 1583 Pearson (p-value) 0.447 0.902 (b) African-American College A College D w/o MMP w/ MMP Difference w/o MMP w/ MMP Difference Arithmetic 0.337 0.303 0.0339 ∗ 0.0568 0.0502 0.00655 Pre-Algebra 0.403 0.426 -0.0235 0.225 0.225 0 Algebra 0.151 0.150 0.00188 0.410 0.415 -0.00437 Intermediate Algebra 0.0988 0.107 -0.00847 0.253 0.255 -0.00218 College Level Math 0.0103 0.0141 -0.00376 0.0546 0.0546 0 Observations 1063 1063 458 458 Pearson (p-value) 0.461 0.995 (c) Latino/a College A College D w/o MMP w/ MMP Difference w/o MMP w/ MMP Difference Arithmetic 0.213 0.189 0.0241 ∗∗ 0.0237 0.0198 0.00395 Pre-Algebra 0.449 0.451 -0.00192 0.176 0.174 0.00216 Algebra 0.227 0.231 -0.00466 0.412 0.412 0.000359 Intermediate Algebra 0.101 0.116 -0.0151 ∗∗ 0.316 0.315 0.000359 College Level Math 0.0101 0.0126 -0.00247 0.0722 0.0790 -0.00682 Observations 3649 3649 2784 2784 Pearson (p-value) 0.0363 0.755 (d) White College A College D w/o MMP w/ MMP Difference w/o MMP w/ MMP Difference Arithmetic 0.146 0.129 0.0169 0.0117 0.0109 0.000839 Pre-Algebra 0.384 0.387 -0.00221 0.113 0.105 0.00755 Algebra 0.200 0.197 0.00295 0.367 0.363 0.00419 Intermediate Algebra 0.238 0.249 -0.0110 0.320 0.322 -0.00126 College Level Math 0.0324 0.0390 -0.00663 0.188 0.199 -0.0113 Observations 1358 1358 2384 2384 Pearson (p-value) 0.621 0.817 (e) Other College A College D w/o MMP w/ MMP Difference w/o MMP w/ MMP Difference Arithmetic 0.182 0.163 0.0188 0.0202 0.0181 0.00213 Pre-Algebra 0.389 0.396 -0.00685 0.133 0.130 0.00319 Algebra 0.228 0.221 0.00685 0.335 0.331 0.00426 Intermediate Algebra 0.173 0.192 -0.0188 0.360 0.359 0.00106 College Level Math 0.0291 0.0291 0 0.152 0.163 -0.0106 Observations 584 584 940 940 Pearson (p-value) 0.867 0.972 mean coefficients ∗ p< 0.10, ∗∗ p< 0.05, ∗∗∗ p< 0.01 109 Table A.3.4: Regression Results, College A, Pass First Math Similar Entire Test=AR Test=EA Test=CLM Test=AR Test=EA Test=CLM (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) PA PA EA IA IA PA PA EA IA IA Boosted Up -0.364 -0.127 0.152 -0.0497 -0.216 -0.155 -0.105 0.0631 -0.0892 -0.142 (0.209) (0.0720) (0.123) (0.136) (0.146) (0.102) (0.0537) (0.0604) (0.0624) (0.134) MMP -0.100 0.00121 0.0474 -0.0271 -0.00640 -0.00326 0.00168 0.00539 -0.0129 -0.0507 (0.0720) (0.0323) (0.0409) (0.0524) (0.0860) (0.0316) (0.0142) (0.0167) (0.0207) (0.0334) ACCU AR -0.141 ∗ 0.00459 (0.0659) (0.00243) ACCU EA 0.00676 0.0301 0.0163 0.0118 ∗∗∗ 0.00890 ∗∗∗ 0.00443 ∗∗ (0.0188) (0.0365) (0.0349) (0.00201) (0.00215) (0.00161) ACCU CLM 0.00811 0.00340 (0.0291) (0.00565) Male -0.0178 -0.141 ∗∗ -0.0418 0.0599 -0.0110 -0.0744 -0.0869 ∗∗∗ -0.0929 ∗∗∗ -0.0675 ∗ -0.114 (0.151) (0.0479) (0.0643) (0.0743) (0.126) (0.0591) (0.0223) (0.0280) (0.0327) (0.0641) Age -0.00956 0.00842 ∗∗ 0.0107 0.0114 -0.000135 0.00210 0.0109 ∗∗∗ 0.0103 ∗∗∗ 0.00526 ∗ -0.00236 (0.00782) (0.00290) (0.00578) (0.00670) (0.0115) (0.00301) (0.00147) (0.00211) (0.00228) (0.00561) Ethnicity Asian 0.0826 0.103 -0.135 -0.117 0.0499 0.171 0.0426 -0.0437 -0.0699 0.0588 (0.272) (0.121) (0.147) (0.186) (0.254) (0.148) (0.0556) (0.0617) (0.0679) (0.155) African-American 0.132 -0.169 -0.308 ∗ -0.163 -0.162 -0.0754 -0.202 ∗∗∗ -0.175 ∗ -0.0803 -0.0998 (0.302) (0.102) (0.146) (0.221) (0.384) (0.137) (0.0519) (0.0730) (0.0865) (0.227) Latino/a -0.294 0.0733 -0.162 -0.163 -0.206 -0.0224 -0.0528 -0.101 -0.155 ∗ -0.135 (0.264) (0.0990) (0.130) (0.179) (0.283) (0.129) (0.0466) (0.0603) (0.0719) (0.170) White 0.204 0.130 -0.0774 -0.0681 0.104 0.114 0.0507 0.0173 -0.0972 0.216 (0.271) (0.107) (0.141) (0.184) (0.268) (0.129) (0.0500) (0.0630) (0.0700) (0.155) At home English 0.0186 -0.156 0.0442 -0.249 ∗ -0.0578 -0.227 ∗ -0.0622 -0.0561 -0.142 ∗∗ -0.0599 (0.255) (0.0827) (0.0988) (0.0986) (0.119) (0.0996) (0.0381) (0.0397) (0.0431) (0.0682) Spanish 0.0197 -0.153 0.0242 -0.137 0.202 -0.101 -0.0449 0.0186 0.0245 0.122 (0.330) (0.105) (0.131) (0.147) (0.233) (0.131) (0.0477) (0.0549) (0.0730) (0.154) Cohort 2005 Fall 0.186 0.157 0.180 0.139 0.182 -0.253 0.0779 0.0323 0.0625 0.0726 (0.423) (0.0996) (0.150) (0.157) (0.221) (0.141) (0.0526) (0.0654) (0.0709) (0.124) 2006 Spring 0.445 0.146 0.225 0.302 ∗ 0.354 -0.352 ∗ 0.0786 0.0240 0.109 0.0712 (0.441) (0.103) (0.150) (0.152) (0.194) (0.142) (0.0549) (0.0666) (0.0724) (0.119) 2006 Fall 0.304 0.113 0.0134 0.164 0.0433 -0.310 ∗ 0.0675 0.0170 0.0734 0.0558 (0.428) (0.102) (0.144) (0.160) (0.220) (0.142) (0.0537) (0.0651) (0.0710) (0.118) 2007 Spring 0.106 0.150 0.203 0.276 0.436 ∗ -0.332 ∗ 0.0287 0.0343 0.171 ∗ 0.144 (0.416) (0.110) (0.160) (0.154) (0.174) (0.145) (0.0574) (0.0697) (0.0732) (0.110) 2007 Fall 0.196 0.0950 0.214 0.166 0.329 -0.311 ∗ 0.0309 0.0442 0.0251 0.0897 (0.409) (0.0987) (0.147) (0.164) (0.192) (0.144) (0.0527) (0.0641) (0.0721) (0.115) Constant 5.743 ∗ 0.134 -1.434 -0.787 0.220 0.712 ∗∗ -0.0351 -0.0698 0.294 0.755 (2.402) (0.586) (1.907) (2.749) (1.420) (0.238) (0.115) (0.170) (0.193) (0.403) Observations 65 427 250 191 83 275 1877 1217 812 201 AdjustedR 2 -0.032 0.077 0.035 0.055 0.001 0.069 0.074 0.053 0.083 0.080 se in parentheses Omitted ethnicity: other Omitted languages: other than English or Spanish Omitted cohort: 2008 Spring ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 110 Table A.3.5: Regression Results, College A, Degree Credits Similar Entire Test=AR Test=EA Test=CLM Test=AR Test=EA Test=CLM (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) PA PA EA IA IA PA PA EA IA IA Boosted Up -7.555 -0.592 1.338 2.756 -9.858 0.555 0.631 3.989 -1.601 -4.883 (11.82) (4.414) (7.328) (8.503) (9.562) (5.814) (3.130) (3.803) (3.998) (8.516) MMP -1.716 2.322 3.461 -1.765 6.647 3.438 1.701 ∗ 1.558 -0.277 -0.303 (4.286) (1.934) (2.271) (3.221) (4.359) (1.871) (0.834) (1.064) (1.439) (2.741) ACCU AR -4.857 0.113 (2.942) (0.129) ACCU EA 0.352 -1.060 0.938 0.428 ∗∗∗ 0.377 ∗∗ 0.00754 (1.194) (2.279) (2.104) (0.121) (0.129) (0.113) ACCU CLM 0.909 0.836 ∗ (1.522) (0.413) Male -7.959 -4.171 -2.388 1.681 -1.109 -6.528 ∗ -3.207 ∗ -5.796 ∗∗∗ -4.051 -6.672 (7.471) (2.679) (3.657) (4.661) (7.094) (3.252) (1.273) (1.736) (2.189) (4.376) Age -0.200 0.275 -0.326 -0.263 -0.323 0.271 0.169 -0.157 0.0218 -0.944 ∗ (0.315) (0.161) (0.252) (0.387) (0.639) (0.177) (0.0912) (0.133) (0.154) (0.398) Ethnicity Asian -5.279 -0.698 -4.562 -15.65 6.455 -2.341 0.878 0.128 -8.747 2.575 (13.97) (6.708) (8.155) (9.820) (13.52) (8.446) (3.240) (4.214) (4.506) (8.785) African-American -1.732 -5.216 -6.291 -31.32 ∗∗ -2.189 -8.812 -3.043 -6.827 -10.87 ∗ -9.025 (14.55) (5.630) (8.498) (11.14) (20.07) (6.783) (2.969) (4.574) (5.481) (11.33) Latino/a -7.378 1.010 -3.864 -10.71 -1.368 0.115 -1.867 -4.932 -4.735 -2.054 (14.74) (5.552) (6.852) (9.901) (15.49) (7.142) (2.641) (3.907) (4.895) (10.57) White 25.54 5.765 0.502 -15.39 14.21 1.708 7.050 ∗ 0.264 -10.57 ∗ 14.38 (15.24) (6.250) (7.585) (10.35) (14.25) (7.061) (2.983) (4.177) (4.603) (9.792) At home English 12.00 -7.551 -8.485 -21.76 ∗∗∗ -0.837 -7.811 -6.374 ∗∗ -8.260 ∗∗ -13.91 ∗∗∗ -4.867 (12.22) (5.442) (5.880) (6.277) (7.223) (6.313) (2.460) (2.636) (2.756) (4.756) Spanish 19.22 -8.246 -17.84 ∗ -15.26 5.241 -7.668 -7.241 ∗ -4.767 -6.011 -4.534 (16.66) (6.415) (7.877) (9.646) (11.88) (7.847) (2.888) (3.555) (4.787) (9.895) Cohort 2005 Fall -3.182 7.182 16.79 10.56 14.46 -9.606 7.734 ∗∗ 3.559 9.985 ∗ 9.140 (20.76) (4.701) (9.126) (10.30) (12.32) (9.890) (2.610) (3.796) (4.764) (7.767) 2006 Spring -6.093 8.231 11.88 11.03 22.60 ∗ -7.276 7.687 ∗∗ 4.757 7.817 2.052 (19.60) (5.077) (8.895) (9.096) (10.31) (10.13) (2.799) (3.826) (4.585) (7.223) 2006 Fall 4.172 4.983 -0.426 4.741 7.190 -7.983 7.322 ∗∗ 5.870 7.491 10.77 (20.76) (4.795) (8.727) (9.613) (11.14) (10.06) (2.723) (3.866) (4.599) (7.665) 2007 Spring 7.614 8.225 15.16 15.87 18.53 -6.343 7.046 ∗ 6.798 10.09 ∗ 15.23 (21.02) (5.592) (9.640) (9.592) (13.35) (10.35) (2.896) (4.109) (4.925) (8.559) 2007 Fall 12.07 8.217 12.51 16.19 17.47 -4.612 5.775 ∗ 3.608 10.65 ∗ 11.06 (20.75) (4.830) (8.839) (9.521) (10.53) (10.14) (2.566) (3.681) (4.627) (6.791) Constant 193.6 3.507 87.33 -6.666 -32.06 23.72 3.970 19.69 47.96 ∗∗∗ 19.53 (105.2) (36.02) (118.5) (163.5) (74.80) (13.15) (6.689) (10.75) (12.26) (26.99) Observations 65 427 250 191 83 275 1877 1217 812 201 AdjustedR 2 -0.022 0.013 0.031 0.049 -0.032 0.027 0.039 0.036 0.039 0.083 se in parentheses Omitted ethnicity: other Omitted languages: other than English or Spanish Omitted cohort: 2008 Spring ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 111 Table A.3.6: Regression Results, College A, Transfer Credits Similar Entire Test=AR Test=EA Test=CLM Test=AR Test=EA Test=CLM (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) PA PA EA IA IA PA PA EA IA IA Boosted Up -7.926 -1.981 1.501 1.852 -9.393 1.394 -1.451 2.531 -1.920 -3.890 (10.89) (3.497) (6.100) (7.638) (9.046) (5.218) (2.366) (3.163) (3.596) (8.097) MMP -1.562 1.503 2.520 -1.603 7.580 ∗ 3.344 ∗ 1.320 1.233 0.194 0.771 (3.827) (1.638) (1.884) (2.882) (3.788) (1.594) (0.698) (0.895) (1.317) (2.537) ACCU AR -4.901 0.0713 (2.800) (0.110) ACCU EA 0.332 -0.518 0.655 0.335 ∗∗∗ 0.266 ∗ -0.00875 (0.976) (1.896) (1.897) (0.101) (0.110) (0.103) ACCU CLM 0.737 0.775 (1.432) (0.393) Male -6.694 -4.515 ∗ -3.470 1.690 0.0641 -4.413 -3.241 ∗∗ -5.212 ∗∗∗ -3.371 -5.674 (6.505) (2.179) (3.035) (4.220) (6.190) (2.697) (1.058) (1.458) (1.989) (4.078) Age -0.112 0.156 -0.312 -0.332 -0.325 0.190 0.0645 -0.245 ∗ -0.129 -0.907 ∗ (0.296) (0.126) (0.204) (0.351) (0.584) (0.141) (0.0752) (0.108) (0.136) (0.365) Ethnicity Asian -3.906 -1.621 -6.434 -12.27 4.627 -2.262 -1.090 -0.866 -7.021 2.980 (11.62) (5.593) (6.937) (8.866) (12.41) (7.210) (2.715) (3.724) (4.193) (7.930) African-American 0.600 -4.453 -3.421 -27.33 ∗∗ -2.081 -6.865 -3.247 -6.027 -7.894 -7.123 (12.02) (4.911) (7.677) (9.913) (18.30) (5.700) (2.534) (3.994) (5.095) (10.12) Latino/a -3.982 -1.058 -3.327 -9.495 -0.662 -0.599 -2.870 -5.337 -3.436 0.0465 (12.03) (4.788) (5.785) (9.138) (14.56) (5.929) (2.255) (3.478) (4.566) (9.611) White 23.49 3.713 0.843 -12.54 12.03 1.837 5.432 ∗ -0.103 -7.393 13.55 (13.26) (5.364) (6.503) (9.391) (13.03) (5.976) (2.556) (3.706) (4.281) (8.923) At home English 8.723 -3.020 -8.615 -19.50 ∗∗∗ -0.939 -5.136 -3.600 -4.755 ∗ -11.46 ∗∗∗ -3.464 (9.974) (4.285) (4.998) (5.660) (6.442) (5.424) (2.002) (2.210) (2.504) (4.426) Spanish 14.18 -3.792 -16.44 ∗ -17.90 ∗ 3.889 -5.635 -4.896 ∗ -2.820 -8.380 -5.087 (13.72) (5.065) (6.600) (8.800) (11.23) (6.610) (2.350) (2.998) (4.270) (9.323) Cohort 2005 Fall -4.603 4.459 12.69 9.157 17.51 -7.755 5.859 ∗∗ 3.121 8.252 10.29 (18.10) (3.816) (7.802) (9.994) (10.77) (8.117) (2.156) (3.230) (4.374) (7.100) 2006 Spring -7.017 7.022 10.11 9.210 25.93 ∗∗ -4.513 6.168 ∗∗ 4.464 6.675 3.608 (17.12) (4.286) (7.618) (8.823) (9.005) (8.415) (2.307) (3.262) (4.196) (6.689) 2006 Fall 1.381 3.815 -0.173 5.591 10.89 -6.452 5.915 ∗∗ 4.937 5.994 11.47 (18.17) (4.066) (7.562) (9.392) (9.634) (8.304) (2.273) (3.304) (4.221) (7.047) 2007 Spring 4.318 6.900 12.90 12.45 20.99 -5.038 5.773 ∗ 5.699 8.007 16.17 ∗ (18.13) (4.735) (8.264) (9.248) (11.62) (8.606) (2.409) (3.512) (4.543) (7.987) 2007 Fall 8.667 5.261 10.01 13.12 19.91 ∗ -3.561 4.288 ∗ 2.576 8.442 ∗ 11.63 (18.07) (3.963) (7.695) (9.303) (9.040) (8.473) (2.117) (3.133) (4.246) (6.244) Constant 192.4 3.026 56.17 7.997 -36.55 18.74 5.390 18.77 ∗ 43.04 ∗∗∗ 8.256 (99.20) (29.19) (98.69) (147.4) (70.48) (10.87) (5.628) (9.175) (11.09) (24.87) Observations 65 427 250 191 83 275 1877 1217 812 201 AdjustedR 2 -0.019 0.004 0.028 0.041 0.022 0.018 0.037 0.031 0.025 0.076 se in parentheses Omitted ethnicity: other Omitted languages: other than English or Spanish Omitted cohort: 2008 Spring ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 112 Table A.3.7: Regression Results, College D, Pass First Math Similar Entire Test=AR Test=EA Test=AR Test=EA (1) (2) (3) (4) (5) (6) PA EA IA PA EA IA Boosted Up 0.0660 0.0397 0.108 -0.0517 0.00609 0.144 ∗ (0.161) (0.112) (0.0908) (0.117) (0.0783) (0.0622) MMP -0.0242 0.0382 ∗ 0.0224 0.00817 0.0331 ∗∗ 0.0241 (0.0282) (0.0190) (0.0221) (0.0142) (0.0122) (0.0130) MDTP AR 0.0183 0.0298 0.0163 ∗∗∗ 0.00851 ∗∗ (0.0202) (0.0176) (0.00373) (0.00265) MDTP EA 0.00112 0.0120 ∗∗∗ (0.0182) (0.00306) MDTP IA Male -0.0561 -0.0612 -0.0434 -0.0519 -0.0469 0.0142 (0.0663) (0.0521) (0.0497) (0.0304) (0.0324) (0.0341) Age 0.00232 0.00957 ∗∗∗ 0.00658 ∗ 0.00677 ∗∗∗ 0.00890 ∗∗∗ 0.00628 ∗∗ (0.00332) (0.00231) (0.00269) (0.00144) (0.00134) (0.00204) Ethnicity Asian -0.150 0.257 ∗ 0.0552 -0.0191 0.152 ∗ -0.0282 (0.106) (0.113) (0.0924) (0.0529) (0.0610) (0.0565) African-American -0.0678 0.202 -0.0749 -0.0687 0.0582 -0.145 (0.117) (0.114) (0.136) (0.0638) (0.0776) (0.111) Latino/a -0.126 0.233 ∗ 0.149 -0.0461 0.0798 0.0132 (0.0930) (0.0939) (0.0927) (0.0501) (0.0630) (0.0608) White -0.0765 0.155 0.0463 -0.0503 0.0660 -0.0228 (0.0976) (0.0928) (0.0812) (0.0504) (0.0551) (0.0529) At home English -0.162 ∗ -0.0469 -0.0934 -0.135 ∗∗∗ -0.0525 -0.0855 ∗ (0.0770) (0.0766) (0.0656) (0.0384) (0.0396) (0.0418) Spanish -0.156 -0.139 -0.236 ∗ -0.131 ∗ -0.133 -0.157 (0.112) (0.108) (0.113) (0.0578) (0.0771) (0.0865) Cohort 2005 Summer 0.0408 -0.129 -0.262 ∗ 0.145 ∗ -0.0642 -0.0423 (0.138) (0.131) (0.112) (0.0723) (0.0756) (0.0788) 2005 Fall -0.0536 -0.0147 -0.147 0.0450 -0.0293 -0.0342 (0.0887) (0.0795) (0.0786) (0.0493) (0.0508) (0.0585) 2006 Spring -0.236 ∗ -0.0402 -0.189 -0.0250 -0.0200 -0.00475 (0.108) (0.0945) (0.114) (0.0562) (0.0577) (0.0726) 2006 Summer -0.0935 0.000888 -0.314 ∗∗ 0.0260 -0.0195 -0.0316 (0.152) (0.146) (0.118) (0.0751) (0.0875) (0.0783) 2006 Fall -0.192 -0.00927 -0.306 ∗∗∗ -0.0815 -0.0559 -0.155 ∗ (0.0996) (0.0840) (0.0842) (0.0525) (0.0532) (0.0642) 2007 Spring -0.115 0.0623 -0.323 ∗ -0.0552 -0.0442 -0.0939 (0.117) (0.0986) (0.129) (0.0588) (0.0608) (0.0877) 2007 Summer 0.283 ∗∗ -0.0337 -0.236 0.0726 -0.0392 -0.0589 (0.0881) (0.125) (0.138) (0.0885) (0.0858) (0.0951) 2007 Fall -0.115 -0.0901 -0.189 ∗ -0.0209 -0.0566 -0.0413 (0.0910) (0.0812) (0.0843) (0.0490) (0.0537) (0.0621) Constant 0.693 ∗ -0.458 0.795 0.425 ∗∗∗ 0.263 ∗ 0.375 ∗∗ (0.339) (0.514) (0.550) (0.0969) (0.113) (0.144) Observations 321 333 349 1060 712 605 AdjustedR 2 0.019 0.058 0.039 0.057 0.082 0.052 se in parentheses Omitted ethnicity: other Omitted languages: other than English or Spanish Omitted cohort: 2008 Spring ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 113 Table A.3.8: Regression Results College D, Degree Credits Similar Entire Test=AR Test=EA Test=AR Test=EA (1) (2) (3) (4) (5) (6) PA EA IA PA EA IA Boosted Up 17.61 -4.626 8.342 10.48 -3.843 10.80 ∗ (10.03) (6.820) (5.970) (8.045) (5.106) (4.583) MMP -2.179 1.876 1.556 -0.307 2.955 ∗∗∗ 2.068 ∗ (1.327) (1.282) (1.387) (0.808) (0.837) (0.979) MDTP AR 0.434 -0.742 0.381 0.0269 (1.114) (1.034) (0.213) (0.187) MDTP EA -0.584 0.587 ∗ (1.187) (0.268) MDTP IA Male -7.894 ∗∗ -5.598 -1.874 -5.934 ∗∗∗ -1.092 2.739 (2.946) (3.301) (3.244) (1.682) (2.025) (2.419) Age -0.157 -0.173 -0.474 0.0629 -0.0921 -0.664 ∗∗ (0.165) (0.169) (0.309) (0.0916) (0.108) (0.229) Ethnicity Asian -4.938 6.162 -0.353 -1.654 1.340 -3.883 (5.422) (7.412) (5.898) (3.296) (4.057) (4.110) African-American -2.408 7.088 -12.93 -3.176 -0.182 -15.10 ∗ (6.246) (6.508) (7.609) (3.678) (4.525) (6.167) Latino/a -9.322 1.699 -4.069 -4.002 -0.346 -6.896 (4.768) (5.171) (6.297) (2.870) (3.424) (4.391) White -4.512 6.512 -6.192 -1.474 4.874 -6.528 (5.378) (5.336) (5.377) (3.070) (3.249) (3.813) At home English -1.365 -10.24 -5.348 -5.427 ∗ -6.329 -4.910 (4.271) (5.732) (4.724) (2.462) (3.304) (3.487) Spanish 7.713 -14.77 ∗ -5.135 -2.380 -11.84 ∗ -4.823 (6.333) (7.186) (7.985) (3.518) (4.849) (6.583) Cohort 2005 Summer -6.419 -8.849 4.857 5.094 1.171 8.996 (8.673) (7.189) (7.877) (5.540) (4.776) (5.436) 2005 Fall -1.941 2.317 14.76 ∗ 2.700 6.838 9.258 ∗ (4.619) (5.497) (6.666) (2.701) (3.550) (4.319) 2006 Spring -1.902 0.806 1.501 3.563 2.638 0.347 (5.821) (6.614) (8.985) (3.249) (3.945) (6.289) 2006 Summer 14.28 9.867 1.805 10.29 ∗ 15.16 ∗ 7.916 (8.680) (9.219) (7.919) (4.734) (6.162) (5.558) 2006 Fall -4.213 -1.264 -4.400 0.436 1.973 -0.944 (4.947) (5.414) (6.319) (2.837) (3.416) (4.166) 2007 Spring -2.483 -3.941 -1.567 -0.222 -2.972 6.683 (5.914) (6.322) (9.064) (3.159) (3.945) (6.437) 2007 Summer 34.77 ∗∗∗ 4.035 -2.417 19.51 ∗∗ 3.687 -0.340 (8.067) (7.639) (8.990) (6.202) (5.279) (6.120) 2007 Fall -1.446 2.930 -3.117 1.795 4.640 0.249 (4.687) (5.329) (6.484) (2.671) (3.434) (4.332) Constant 29.25 64.04 ∗ 71.42 ∗ 21.42 ∗∗∗ 34.15 ∗∗∗ 39.63 ∗∗∗ (17.95) (29.39) (35.60) (5.921) (7.946) (11.25) Observations 321 333 349 1060 712 605 AdjustedR 2 0.051 0.015 0.077 0.028 0.037 0.056 se in parentheses Omitted ethnicity: other Omitted languages: other than English or Spanish Omitted cohort: 2008 Spring ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 114 Table A.3.9: Regression Results College D, Transfer Credits Similar Entire Test=AR Test=EA Test=AR Test=EA (1) (2) (3) (4) (5) (6) PA EA IA PA EA IA Boosted Up 13.32 -3.192 7.110 8.596 -3.032 9.691 ∗ (8.375) (5.726) (5.465) (6.843) (4.313) (4.221) MMP -1.530 1.480 1.369 -0.182 2.523 ∗∗∗ 1.936 ∗ (1.080) (1.093) (1.258) (0.666) (0.738) (0.911) MDTP AR 0.111 -0.671 0.281 0.0129 (0.911) (0.884) (0.172) (0.167) MDTP EA -0.683 0.562 ∗ (1.088) (0.251) MDTP IA Male -6.336 ∗∗ -5.322 -1.846 -4.815 ∗∗∗ -1.003 2.726 (2.365) (2.789) (2.945) (1.369) (1.749) (2.236) Age -0.190 -0.202 -0.442 0.00632 -0.116 -0.629 ∗∗ (0.133) (0.149) (0.292) (0.0750) (0.0972) (0.213) Ethnicity Asian -3.911 5.243 -0.557 -0.923 1.208 -4.163 (4.456) (6.479) (5.357) (2.739) (3.575) (3.801) African-American -3.298 6.427 -11.63 -3.162 -0.305 -13.96 ∗ (4.948) (5.555) (6.979) (2.969) (3.897) (5.723) Latino/a -8.301 ∗ 1.517 -3.905 -3.168 -0.177 -7.067 (3.834) (4.392) (5.728) (2.357) (2.962) (4.046) White -5.970 6.842 -5.489 -1.186 4.812 -5.959 (4.302) (4.566) (4.857) (2.523) (2.840) (3.516) At home English -2.143 -11.24 ∗ -5.671 -4.768 ∗ -6.772 ∗ -5.276 (3.429) (5.109) (4.349) (2.032) (2.962) (3.244) Spanish 6.367 -15.38 ∗ -5.099 -1.791 -12.01 ∗∗ -4.798 (5.187) (6.201) (7.286) (2.922) (4.154) (6.065) Cohort 2005 Summer -5.249 -7.542 4.401 4.211 1.606 8.413 (6.912) (6.036) (6.979) (4.511) (4.025) (4.960) 2005 Fall -1.012 2.742 13.53 ∗ 3.007 7.033 ∗ 8.742 ∗ (3.713) (4.690) (5.982) (2.189) (3.103) (3.938) 2006 Spring -2.024 0.151 2.057 3.462 1.936 0.589 (4.702) (5.410) (8.198) (2.643) (3.372) (5.792) 2006 Summer 10.02 6.575 2.215 7.572 ∗ 13.08 ∗ 7.715 (6.778) (7.439) (7.102) (3.767) (5.189) (5.134) 2006 Fall -3.257 -0.661 -3.713 0.962 2.100 -0.219 (4.021) (4.564) (5.635) (2.302) (2.962) (3.777) 2007 Spring -2.571 -3.412 -0.334 -0.585 -2.148 6.850 (4.384) (5.372) (8.295) (2.548) (3.465) (5.942) 2007 Summer 26.97 ∗∗∗ 2.701 -2.833 16.24 ∗∗ 3.195 -0.287 (6.922) (6.476) (7.985) (5.127) (4.568) (5.522) 2007 Fall -1.889 2.498 -2.989 1.436 4.306 0.106 (3.752) (4.495) (5.822) (2.168) (2.989) (3.971) Constant 29.55 ∗ 55.26 ∗ 67.05 ∗ 18.20 ∗∗∗ 27.27 ∗∗∗ 33.31 ∗∗ (14.26) (25.14) (32.54) (4.838) (7.002) (10.41) Observations 321 333 349 1060 712 605 AdjustedR 2 0.051 0.033 0.077 0.028 0.046 0.058 se in parentheses Omitted ethnicity: other Omitted languages: other than English or Spanish Omitted cohort: 2008 Spring ∗ p< 0.05, ∗∗ p< 0.01, ∗∗∗ p< 0.001 115 Appendix Figure A.1: Enrollment (a) Primary (Net %) 20 40 60 80 100 Enrollment (% net) 1990 1995 2000 2005 2010 Year Indonesia OECD members World Least developed countries Source: World Development Indicator 2013 (b) Secondary (Net %) 20 40 60 80 100 Enrollment (% net) 1990 1995 2000 2005 2010 Year Indonesia OECD members World Least developed countries Source: World Development Indicator 2013 (c) Tertiary (Gross %) 0 20 40 60 80 Enrollment (% gross) 1990 1995 2000 2005 2010 Year Indonesia OECD members World Least developed countries Source: World Development Indicator 2013 116 Bibliography ACT, ENGAGE College User’s Guide, Iowa City, IA: ACT, Inc., 2012. Adelman, Clifford, “Paths to Degree Completion From High School Through College Paths to Degree Completion from High School Through College,” Technical Report, Office of V ocational and Adult Education, US Department of Eudcation, Washington, DC 2006. Allen, Jeff, Steven B. Robbins, Alex Casillas, and In-Sue Oh, “Third-year College Retention and Transfer: Effects of Academic Performance, Motivation, and Social Connectedness,” Research in Higher Education, April 2008, 49 (7), 647–664. American Educational Research Association, American Psychological Association, and National Council on Measurement in Education, Standards for educational and psychological testing, Washington, DC: AERA Publications Sales, 1999. Armstrong, William B., “The Association Among Student Success in Courses, Place- ment Test Scores, Student Background Data, and Instructor Grading Practices,” Com- munity College Journal of Research and Practice, September 2000, 24 (8), 681–695. Bailey, Thomas, “Challenge and opportunity: Rethinking the role and function of de- velopmental education in community college,” New Directions for Community Col- leges, 2009, (145), 11–30. , Dong Wook Jeong, and Sung-Woo Cho, “Referral, enrollment, and completion in developmental education sequences in community colleges,” Economics of Education Review, April 2010, 29 (2), 255–270. Behrman, Jere R., “Schooling in developing countries: Which countries are the Over- and underachievers and what is the schooling impact?,” Economics of Education Re- view, 1987, 6 (2), 111–127. and Anil B. Deolalikar, “School repetition, dropouts, and the rates of return to schooling: The case of Indonesia,” Oxford Bulletin of Economics and Statistics, May 1991, 53 (4), 467–480. 117 and Barbara L. Wolfe, “The socioeconomic impact of schooling in a developing country,” The Review of Economics and Statistics, 1984, 66 (2), 296–303. and Mark R. Rosenzweig, “Does increasing women’s schooling raise the schooling of the next generation? Reply,” The American Economic Review, 2002, 92 (1), 323– 334. ,AndrewD.Foster,MarkR.Rosenzweig,andPremVahishtha, “Women’s school- ing, home teaching, and economic growth,” Journal of Political Economy, 1999, 107 (4), 682–714. , Mark R. Rosenzweig, and Paul Taubman, “College choice and wages: Estimates using data on female twins,”...of Economics and Statistics, 1996, 78 (4), 672–685. Belfield, Clive R. and Peter M. Crosta, “Predicting Success in College: The Impor- tance of Placement Tests and High School Transcripts,” Community College Research Center, Columbia University, Working Paper 42, 2012. Blinder, Alan S., “Wage discrimination: reduced form and structural estimates,” Jour- nal of Human Resources, 1973, 8 (4), 436–455. Boissiere, M, JB Knight, and RH Sabot, “Earnings, schooling, ability, and cognitive skills,” The American Economic Review, 1985, 75 (5), 1016–1030. Boylan, Hunter R., “Targeted intervention for developmental education students (TIDES),” Journal of Developmental Education, 2009, 32 (3), 14–23. Bradley, Steve and Pam Lenton, “Dropping out of post-compulsory education in the UK: an analysis of determinants and outcomes,” Journal of Population Economics, 2007, 20 (2), 299–328. Brewer, Dominic J. and Joanna Smith, “A framework for understanding educational governance: The case of California,” Education and Finance Policy, 2008, 3 (1), 20– 40. Brien, Michael J. and Lee A. Lillard, “Education, marriage, and first conception in Malaysia,” Journal of Human Resources, 1994, 29 (4), 1167–1204. Brown, Philip H., “Parental education and investment in children’s human capital in rural China,” Economic Development and Cultural Change, 2006, 54 (4), 759–789. Burdman, Pamela, “Where to Begin? The Evolving Role of Placement Exams for Students Starting College.,” Technical Report, Jobs for the Future, Boston, MA 2012. Cage, Mary Crystal, “Cal. Community Colleges System Agrees to Change Role of Testing,” The Chronicle of Higher Education, 1991, June 12, A21. 118 California Code of Regulations, “Definitions,” 55502. , “Orientation,” 55521. California Community Colleges Chancellor’s Office, “Multiple measures and other sorrows: A guide for using student assessment information with or instead of test scores,” Technical Report, Sacramento, CA 1998. , California Community Colleges Matriculation Program Handbook, Sacramento, CA: Student Services and Special Programs Division, 2011. California Education Code, “Student Matriculation,” 78212. Cameron, Lisa, “An Analysis of the Role in Reducing School Drop-Out During the In- donesian Economic Crisis,” Innocenti Working Paper, 2000, (82), Florence: UNICEF Innocenti Research Centre. , “The impact of the Indonesian financial crisis on children: an analysis using the 100 villages data,” Bulletin of Indonesian Economic Studies, April 2001, 37 (1), 43–64. Chuang,Hwei-Lin, “An empirical study of re-enrolment behavior for male high-school students,” Applied Economics, 1994, 26 (11), 1071–1082. Clark, Andrew E. and Andrew J. Oswald, “Unhappiness and Unemployment,” The Economic Journal, 1994, 104 (424), 648–659. Collins, Michael Lawrence, “It’s not just about the cut score: Redesigning placement assessment policies to improve student success,” Technical Report, Jobs for the Fu- ture, Boston, MA 2008. Connelly,RachelandZhenzhenZheng, “Determinants of school enrollment and com- pletion of 10 to 18 year olds in China,” EconomicsofEducationReview, August 2003, 22 (4), 379–388. Cutler, David M. and Adriana Lleras-Muney, “Education and Health: Evaluating Theories and Evidence,” NBER Working Paper 12352, 2006. Deolalikar, Anil B., “Gender differences in the returns to schooling and in school en- rollment rates in Indonesia,” Journal of Human Resources, 1993, 28 (4), 899–932. DesJardins, Stephen L. and Nathan K. Lindsay, “Adding a Statistical Wrench to the ”Toolbox”,” Research in Higher Education, October 2007, 49 (2), 172–179. Duckworth, Angela L, Christopher Peterson, Michael D Matthews, and Dennis R Kelly, “Grit: perseverance and passion for long-term goals.,” Journal of personality and social psychology, June 2007, 92 (6), 1087–101. 119 , Patrick D Quinn, and Eli Tsukayama, “What No Child Left Behind Leaves Be- hind: The Roles of IQ and Self-Control in Predicting Standardized Achievement Test Scores and Report Card Grades.,” Journal of educational psychology, May 2012, 104 (2), 439–451. Duflo, Esther, “Schooling and labor market consequences of school construction in In- donesia: Evidence from an unusual policy experiment,” American Economic Review, 2001, 91 (4), 795–813. Eckstein, Zvi and KI Wolpin, “Why youths drop out of high school: The impact of preferences, opportunities, and abilities,” Econometrica, 1999, 67 (6), 1295–1339. Eloundou-Enyegue, PM, “Pregnancy-related dropouts and gender inequality in educa- tion: A life-table approach and application to Cameroon,” Demography, 2004, 41 (3), 509–528. ETS, “Personal Potential Index: Evaluator User’s Guide,” Technical Report July 2013, Princeton, NJ 2013. Fain, Paul, “Two community colleges get serious about working with K12,” February 2013. Ferrer, Emilio and John J McArdle, “An experimental analysis of dynamic hypothe- ses about cognitive abilities and achievement from childhood to early adulthood.,” Developmental psychology, November 2004, 40 (6), 935–52. Field, Erica and Attila Ambrus, “Early Marriage, Age of Menarche, and Female Schooling Attainment in Bangladesh,” Journal of Political Economy, October 2008, 116 (5), 881–930. Florida Senate, “Bill 1720,” 2013. Fong, Kristen, Tatiana Melguizo, Johannes M. Bos, and George Prather, “A differ- ent view on how we understand progression through the developmental math trajec- tory,” Technical Report, Rossier School of Education, University of Southern Califor- nia, Los Angeles, CA 2013. Fulton, Mary, “Using State Policies to Ensure Effective Assessment and Placement in Remedial Education,” Technical Report, Education Commission of the States, Den- ver, CO 2012. Geiser, Saul and MV Santelices, “Validiy of high-school grades in predicting student success beyond the freshman year: High-school record vs. standardized tests as indi- cators of four-year college outcomes,” Center for Studies in Higher Education, 2007. 120 andRogerStudley, “UC and the SAT: Predictive Validity and Differential Impact of the SAT I and SAT II at the University of California,” Educational Assessment, 2003, 8 (1), 1–26. Gerlaugh, Katherine, Lizette Thompson, Hunter Boylan, and Hildreth Davis, “Na- tional Study of Developmental Education II: Baseline data for community colleges,” Research in Deveopmental Education, 2007, 20 (4), 1–4. Glewwe, Paul and Hanan G. Jacoby, “Student achievement and schooling choice in low-income countries: Evidence from Ghana,” Journal of Human Resources, 1994, 29 (3), 843–864. and , “An economic analysis of delayed primary school enrollment in a low in- come country: the role of early childhood nutrition,” The review of Economics and Statistics, 1995, 77 (1), 156–169. Grossman, Michael, “Education and nonmarket outcomes,” in Eric A. Hanushek and F. Welch, eds., Handbook of the Economics of Education, V ol. 1, Elsevier B.V ., 2006, chapter 10, pp. 578–633. Hadi, Ali S., “Identifying multiple outliers in multivariate data,” Journal of the Royal Statistical Society, 1992, 54 (3), 761–771. Hahn, A., J Danzberger, and B Lefkowitz, “Dropouts In America: Enough Is Known for Action,” Institute for Edcuational Leadership, 1987. Hanushek, Eric A., “The economics of schooling: Production and efficiency in public schools,” Journal of Economic Literature, 1986, 24 (3), 1141–1177. and Ludger Woessmann, “The role of cognitive skills in economic development,” Journal of Economic Literature, 2008, 46 (3), 607–668. Hartono, D. and D. Ehrmann, “The Indonesian Economic Crisis and its Impact on Educational Enrolment and Quality,” Institute of Southeast Asian Studies, 2001, (7), 182. Heckman, James J, Jora Stixrud, and Sergio Urzua, “The Effects of Cognitive and Noncognitive Abilities on Labor Market Outcomes and Social Behavior,” Journal of Labor Economics, 2006, 24 (3). Hodara, Michelle, Shanna Smith Jaggars, and Melinda Mechur Karp, “Improv- ing Developmental Education Assessment and Placement: Lessons From Commu- nity Colleges Across the Country,” Community College Research Center, Columbia University, Working Paper No. 51, 2012. 121 Hughes, Katherine L. and Judith Scott-Clayton, “Assessing Developmental Assess- ment in Community Colleges,” Community College Review, October 2011, 39 (4), 327–351. Jaffe, Louise, “Mathematics from High School to Community College: Preparation, Ariculation, and College Un-readiness.” PhD dissertation, University of California Los Angeles 2012. Jaggars, Shanna Smith and Michelle Hodara, “The Opposing Forces that Shape De- velopmental Education: Assessment, Placement, and Progression at CUNY Commu- nity Colleges,” Community College Research Center, Columbia University, Working Paper No. 36, 2011. Jann, Ben, “A Stata implementation of the Blinder-Oaxaca decomposition,” The Stata Journal, 2008, 8 (4), 453–479. Jenkins, Davis, Shanna Smith Jaggars, and Josipa Roksa, “Promoting Gatekeeper Course Success Among Community College Students Needing Remediation Find- ings and Recommendations from a Virginia Study (Summary Report),” Community College Research Center, Columbia University, 2009. Jones, Gavin W and Peter Hagul, “Schooling in Indonesia: crisis-related and longer- term issues,” Bulletin of Indonesian Economic Studies, 2001, 37 (2), 207–231. Kane, Michael T., “Validation,” in R.L. Brennan, ed., Educational Measurement, 4 ed., Westport, CT: ACE/Praeger Publishers, 2006, pp. 17–64. Kaplan, EL and Paul Meier, “Nonparametric estimation from incomplete observa- tions,” Journal of the American statistical association, 1958, 53 (282), 457–481. Lagunoff, Rachel, Hillary Michaels, Patricia Morris, and Pamela Yeagley, A Framework for Evaluating the Technical Quality of Multiple Measures Used in Cali- fornia Community College Placement, San Francisco, CA: WestEd, February 2012. Lam, David and Suzanne Duryea, “Effects of schooling on fertility, labor supply, and investments in children, with evidence from Brazil,” Journal of Human Resources, 1999, 34 (1), 160–192. , Cally Ardington, and Murray Leibbrandt, “Schooling as a Lottery: Racial Dif- ferences in School Advancement in Urban South Africa.,” Journal of Development Economics, July 2011, 95 (2), 121–136. Levine, David I. and Gary Painter, “The schooling costs of teenage out-of-wedlock childbearing: analysis with a within-school propensity-score-matching estimator,” Review of Economics and Statistics, 2003, 85 (4), 884–900. 122 Lewallen, Willard Clark, Multiple Measures in Placement Recommendations: An Ex- amination of Variables Related to Course Success, Lancaster, CA: Antelope Valley College, 1994. Lloyd, Cynthia B., Barbara S. Mensch, and Wesley H. Clark, “The effects of pri- mary school quality on school dropout among Kenyan girls and boys,” Comparative Education Review, 2000, 44 (2), 113–147. Lochner, Lance, “Non-production benefits of education: Crime, health, and good citi- zenship,” in Eric A. Hanushek, Stephen Machin, and Ludger Woessmann, eds., Hand- book of the Economics of Education, V ol. 4, Amsterdam: North Holland, 2011, chap- ter 2, pp. 183–282. Long Beach College Promise, 5-Year Progress Report (2008-2013): A breakthrough in student achievement 2008. Maruyama, Geoffrey, “Assessing College Readiness: Should We Be Satisfied With ACT or Other Threshold Scores?,” Educational Researcher, September 2012, 41 (7), 252–261. Marwick, Judith D., “Charting a Path To Success: the Association Between Institu- tional Placement Policies and the Academic Success of Latino Students,” Community College Journal of Research and Practice, March 2004, 28 (3), 263–280. Mattern, Krista D. and Sheryl Packman, “Predictive validity of ACCUPLACER scores for course placement: A meta-analysis,” 2009. McElroy, Susan Williams, “Early childbearing, high school completion, and college enrollment: Evidence from 1980 high school sophomores,” Economics of Education Review, June 1996, 15 (3), 303–324. Medhanie, Amanuel G., Danieele N. Dupuis, Brandon LeBeau, Michael R. Har- well, and Thomas R. Post, “The Role of the ACCUPLACER Mathematics Place- ment Test on a Student’s First College Mathematics Course,” Technical Report 2 Au- gust 2011. Melguizo, Tatiana, Holly Kosiewicz, George Prather, and Johannes M. Bos, “How are community college students assessed and placed in developmental math: Ground- ing our understanding in reality,” Journal of Higher Education, Forthcomin. , Johannes M. Bos, and George Prather, “Using a regression discontinuity design to estimate the impact of placmeent decisions in developmental math in Los Angeles Community College District (LACCD),” Technical Report, Rossier School of Educa- tion, University of Southern California, Los Angeles, CA 2013. 123 Merisotis, Jamie P. and Ronald A. Phipps, “Remedial Education in Colleges and Universities: What’s Really Going On?,” The Review of Higher Education, 2000, 24 (1), 67–85. Mincer, Jacob, Schooling, Experience, and Earnings, New York, NY: National Bureau of Economic Research, 1974. National Center for Education Statistics, “The Condition of Education 2013,” Tech- nical Report, Institute of Education Sciences 2013. National Center for Public Policy and Higher Education and Southern Regional Education Board, “Beyond the rhetoric: Improving college readiness through co- herent state policy,” Technical Report June, National Center for Public Policy and Higher Edcuation & Southern Regional Education Board 2010. Oaxaca, Ronald, “Male-female wage differentials in urban labor markets,” Interna- tional economic review, 1973, 14 (3), 693–709. Osili, Una Okonkwo and Bridget Terry Long, “Does female schooling reduce fertil- ity? Evidence from Nigeria,” Journal of Development Economics, August 2008, 87 (1), 57–75. Oswald, Andrew J., “Happiness and Economic Performance,” The Economic Journal, 1997, 107 (445), 1815–1831. Pallas,AaronM., “School Dropouts in the United States.,” Center for Edcuation Statis- tics, 1987, (ED283119). Parish, William L. and Robert J Willis, “Daughters, education, and family budgets Taiwan experiences,” Journal of Human Resources, 1993, 28 (4), 863–898. Perin, Dolores, “Can Community Colleges Protect Both Access and Standards? The Problem of Remediation,” Teachers College Record, March 2006, 108 (3), 339–373. Perry, Mary, Matthew Rosin, Kathryn Morgan Woodward, and Trish Williams, Course-taking patterns, policies , and practices in developmental education in the California Community Colleges number June, Mountain View, CA: EdSource, 2010. Pitt, Mark M., Mark R. Rosenzweig, and Mohammad Nazmul Hassan, “Human capital investment and the gender division of labor in a brawn-based economy,” Amer- ican Economic Review, 2012, 102 (7), 3531–3560. Porchea, Sameano F., Jeff Allen, Steve Robbins, and Richard P. Phelps, “Predictors of long-term enrollment and degree outcomes for community college students: Inte- grating academic, psychosocial, socio-demographic, and situational,” The Journal of Higher Education, 2010, 81 (6), 750–778. 124 Porter, Andrew C. and Morgan S. Polikoff, “Measuring academic readiness for col- lege,” Educational Policy, 2012, 26 (3), 394–417. Psacharopoulos, George, Returns to education: An international comparison, San Francisco: Jossey-Bass, 1973. and Harry Anthony Patrinos, “Returns to investment in education: a further up- date,” Education Economics, August 2004, 12 (2), 111–134. Robbins, Steven B., Jeff Allen, Alex Casillas, Christina Hamme Peterson, and Huy Le, “Unraveling the differential effects of motivational and skills, social, and self- management measures from traditional predictors of college outcomes.,” Journal of Educational Psychology, 2006, 98 (3), 598–616. Rosenzweig, Mark R., “Educational subsidy, agricultural development, and fertility change,” The Quarterly Journal of Economics, 1982, 97 (1), 67–88. Rotermund, Susan, “Why Students Drop Out of High School Comparisons from Three National Surveys,” Technical Report 2, Sanata Barbara 2007. Rumberger, Russell W., “High School Dropouts: A Review of Issues and Evidence,” Review of Educational Research, January 1987, 57 (2), 101–121. Sawyer,Richard, “Decision theory models for validating course placement tests,” Jour- nal of Educational Measurement, 1996, 33 (3), 271–290. , “Indicators of Usefulness of Test Scores,” Applied Measurement in Education, July 2007, 20 (3), 255–271. , “Beyond Correlations: Usefulness of High School GPA and Test Scores in Making College Admissions Decisions,” Applied Measurement in Education, April 2013, 26 (2), 89–112. Schultz, T. Paul, “Explanation of birth rate changes over space and time: A study of Taiwan,” The Journal of Political Economy, 1973, 81 (2). Scott-Clayton, Judith, “Do High-Stakes Placement Exams Predict College Success?,” Community College Research Center, Columbia University, Working Paper 41, 2012. , Peter M. Crosta, and Clive R. Belfield, “Improving the Targeting of Treatment: Evidence from College Remediation,” National Bureau of Economic Research, Work- ing Paper 18457, 2012. Sedlacek, William E., Beyond the Big Test: Noncognitive Assessment in Higher Edu- cation, San Francisco, CA: Jossey-Bass, 2004. 125 Skoufias, Emmanuel and Susan W Parker, “Job Loss and Family Adjustments in Work and Schooling during the Mexican Peso Crisis,” Journal of Population Eco- nomics, 2006, 19 (1), 163–181. Sommerville,JanisandYunYi, “Aligning K-12 and postsecondary expectations: State policy in transition,” Technical Report, National Association of System Heads, Wash- ington, DC 2002. Sparrow, Robert, “Protecting Education for the Poor in Times of Crisis: An Evalua- tion of a Scholarship Programme in Indonesia,” Oxford Bulletin of Economics and Statistics, February 2007, 69 (1), 99–122. Stacey, Nevzer, “Social benefits of education,” Annals of the American Academy of Political and Social Science, 1998, 559 (The Changing Educational Quality of the Workforce), 54–63. Stone, D.A., Policy Paradox: The Art of Polictical Decision Making, New York, NY: W.W. Norton, 2002. Strauss, John, “Households, communities, and preschool children’s nutrition out- comes: Evidence from rural Cˆ ote d’Ivoire,” Economic Development and Cultural Change, 1990, 38 (2), 231–261. , Kathleen Beegle, Agus Dwiyanto, Yulia Herawati, Daan Pattinasarany, Elan Satriawan, Bondan Sikoki, Sukamdi, and Firman Witoelar, Indonesian Living Standards: Before and After the Financial Crisis, V ol. 137, Santa Monica, CA: RAND Corporation, 2004. Texas Higher Education Coordinating Board, “2012-2017 Statewide Developmental Education Plan,” Technical Report 2012. Thomas,Duncan,JohnStrauss,andMHHenriques, “How Does Mother’s Education Affect Child Height?,” Journal of Human Resources, 1991, 26 (2), 183–211. , K. Beegle, E. Frankenberg, B. Sikoki, J. Strauss, and G. Teruel, “Education in a Crisis,” Journal of Development Economics, June 2004, 74 (1), 53–85. , Robert F. Schoeni, and John Strauss, “Parental investments in schooling: The roles of gender and resources in urban Brazil,” RAND Working Paper Series 96-02, 1996. van der Linden, Wim J., “A Decision Theory Model for Course Placement,” Journal of Educational and Behavioral Statistics, January 1998, 23 (1), 18–34. 126 Venezia, Andrea, Kathy Reeves Bracco, and Thad Nodine, One-Shot Deal? Stu- dents’ Perceptions of Assessment and Course Placment in California’s Community Colleges, San Francisco, CA: WestEd, 2010. Wiseley, W. Charles, “Regulation, Interpretation, Assessment and Open Access in Cal- ifornia Community Colleges,” California Community Colleges Chancellor’s Office Working Paper, 2006. Wolman,Clara,RobertBruininks,andMarthaL.Thurlow, “Dropouts and Dropout Programs: Implications for Special Education,” Remedial and Special Education, September 1989, 10 (5), 6–20. World Bank Human Development Network, Averting a Human Crisis During the Global Downturn, conference ed., Washington, DC: World Bank, 2009. 127

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

A multi-perspective examination of developmental education: student progression, institutional assessment and placement policies, and statewide regulations

PDF

Essays on economics of education

PDF

How extending time in developmental math impacts persistence and success: evidence from a regression discontinuity in community colleges

PDF

Reforming developmental education in math: exploring the promise of self-placement and alternative delivery models

PDF

Three essays on the high school to community college STEM pathway

PDF

Essays on human capital accumulation -- health and education

PDF

Essays on the empirics of risk and time preferences in Indonesia

PDF

Ready or not? Unprepared for community college mathematics: an exploration into the impact remedial mathematics has on preparation, persistence and educational goal attainment for first-time Cali...

PDF

Math and the making of college opportunity: persistent problems and possibilities for reform

PDF

Essays on health economics

PDF

To what extent does being a former high school English learner predict success in college mathematics? Evidence of Latinx students’ duality as math achievers

PDF

Essays on economics of education and private tutoring

PDF

Three essays on cooperation, social interactions, and religion

PDF

Essays on family planning policies

PDF

Essays in the economics of education and conflict

PDF

Essays in political economy and mechanism design

PDF

Unprepared for college mathematics: an investigation into the attainment of best practices to increase preparation and persistence for first-time California community college freshmen in remedial...

PDF

Measuring the alignment of high school and community college math assessments

PDF

Oppression of remedial reading community college students and their academic success rates: student perspectives of the unquantified challenges faced

PDF

Three essays on health & aging

Asset Metadata

Creator Kwon, William Wookun (author)

Core Title Essays on education: from Indonesia to Los Angeles, CA

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Publication Date 08/18/2014

Defense Date 08/16/2014

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

Tag developmental mathematics,Education,IFLS,Indonesia,LACCD,multiple measures,OAI-PMH Harvest,placement testing

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Strauss, John A. (committee chair), Melguizo, Tatiana (committee member), Nugent, Jeffrey B. (committee member)

Creator Email wkwon@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-461172

Unique identifier UC11287066

Identifier etd-KwonWillia-2822.pdf (filename),usctheses-c3-461172 (legacy record id)

Legacy Identifier etd-KwonWillia-2822.pdf

Dmrecord 461172

Document Type Dissertation

Format application/pdf (imt)

Rights Kwon, William Wookun

Type texts

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

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

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA