Vietnam's Two-Child Policy: Impact on Maternal Employment & Family Size - Demographic Surv, Study notes of Law

Download Vietnam's Two-Child Policy: Impact on Maternal Employment & Family Size - Demographic Surv and more Study notes Law in PDF only on Docsity! Essays on Women’s Fertility, Education, and Female Labor Supply in Vietnam BY ANH P. NGO B.A., Coe College, 2013 M.A., University of Illinois at Chicago, 2017 THESIS Submitted as partial fulfillment of the requirements for the degree of Doctor of Philosophy in Economics in the Graduate College of the University of Illinois at Chicago, 2018 Chicago, Illinois Defense Committee: Robert Kaestner, Chair and Advisor Ben Ost Darren Lubotsky Javaeria Qureshi Neeraj Kaushal, Columbia University ii This thesis is dedicated to my family back home, especially my parents, Mr. Ngo Van Phu and Mrs. Nguyen Thi Tam. I am grateful for their unconditional love and support. Without them, I would have never finished this thesis and been where I am today. v TABLE OF CONTENTS CHAPTER I: EFFECTS OF VIETNAM’S TWO-CHILD POLICY ON FERTILITY, SON PREFERENCE, AND FEMALE LABOR SUPPLY ..................................................................... 1 A. INTRODUCTION ............................................................................................................ 1 B. THE TWO-CHILD POLICY IN VIETNAM ...................................................................... 4 C. CONCEPTUAL FRAMEWORK .................................................................................... 6 C.1. Demand for Children ....................................................................................................... 6 C.2. Son Preference and Son Targeting Fertility Behavior ..................................................... 8 C.3. The Effect of Children on Female Labor Supply ............................................................. 9 D. DATA ................................................................................................................................... 11 D.1. The Vietnam Population and Housing Censuses ........................................................... 11 D.2. Measures of family size, son preference, and maternal labor supply ............................ 12 D.3. Summary Statistics ........................................................................................................ 13 E. RESEARCH DESIGN .......................................................................................................... 13 E.1. The First Stage – The Effect of the Two-child Policy on Family Size and Son Preference .............................................................................................................................. 13 E.2. The Reduced Form – The Effect of the Two-child Policy on Maternal Labor Supply . 18 F. RESULTS ............................................................................................................................. 19 F.1. The Effect of Vietnam’s Two-child Policy on Family Size ........................................... 19 F.2. The Effect of Vietnam’s Two-child Policy on Son Preference ...................................... 22 F.3. The Effect of Vietnam’s Two-child Policy on Maternal Labor Supply ......................... 23 F.4. Instrumental Variables Estimates of the Effect of Fertility on Maternal Employment in Vietnam ................................................................................................................................. 25 G. FALSIFICATION TESTS ................................................................................................... 26 H. DISCUSSION & CONCLUSION ........................................................................................ 27 CITED LITERATURE ............................................................................................................. 28 FIGURES .................................................................................................................................. 31 TABLES .................................................................................................................................... 35 Appendix A ............................................................................................................................... 45 CHAPTER II. EFFECTS OF UNIVERSAL PRIMARY EDUCATION (UPE) LAW ON EDUCATIONAL ATTAINMENT AND FERTILITY: EVIDENCE FROM VIETNAM .......... 61 A.INTRODUCTION ................................................................................................................. 61 vi B. LITERATURE REVIEW ..................................................................................................... 63 C. CONCEPTUAL FRAMEWORK ......................................................................................... 66 C.1. The Effect of the UPE Policy on Education .................................................................. 66 C.2. Education and Fertility ...................................................................................................... 67 D. DATA ................................................................................................................................... 68 D.1. Vietnam Household Living Standards Surveys ............................................................. 68 D.2. Vietnam Population and Housing Census ..................................................................... 69 D.3. Descriptive Statistics ..................................................................................................... 70 E. RESEARCH DESIGN .......................................................................................................... 70 E.1. Defining Treatment ........................................................................................................ 70 E.2. Empirical Framework ..................................................................................................... 72 E.3. Validity ........................................................................................................................... 75 F. RESULTS ............................................................................................................................. 77 F.1. The Effect of the UPE Policy on Educational Attainment ............................................. 77 F.2. The Effect of the UPE Policy on Fertility ...................................................................... 79 G. DISCUSSION & CONCLUSION ........................................................................................ 82 CITED LITERATURE ............................................................................................................. 83 TABLES .................................................................................................................................... 86 FIGURES .................................................................................................................................. 93 APPENDIX B ........................................................................................................................... 97 CHAPTER III: CHILDREN AND MATERNAL EMPLOYMENT IN VIETNAM IN 1989, 1999, AND 2009 ..................................................................................................................... 101 A. INTRODUCTION ........................................................................................................ 101 B. CONCEPTUAL FRAMEWORK .................................................................................... 103 B.1. Sex Mix and Family Size ................................................................................................. 103 C. DATA ................................................................................................................................. 108 C.1. Vietnam Population and Housing Censuses ................................................................ 108 C.2. Measures of fertility and female labor supply ............................................................. 109 D. RESEARCH DESIGN ....................................................................................................... 109 E. RESULTS ........................................................................................................................... 113 E.1. The Effect of Sibling-Sex Composition on Family Size .............................................. 113 E.2. The Effect of Number of Children on Maternal Employment ..................................... 113 vii F. DISCUSSION AND CONCLUSION ................................................................................. 116 CITED LITERATURE ........................................................................................................... 117 TABLES .................................................................................................................................. 120 APPENDIX C ......................................................................................................................... 127 VITA ....................................................................................................................................... 128 x TABLE XX: THE ETIMATES OF THE EFFECT OF THE TWO CHILD POLICY ON PROPORTION OF SONS IN EACH FAMILY - RESULTS FROM SUBSAMPLE 2 ............. 57 TABLE XXI: THE ESTIMATES OF THE EFFECT OF THE TWO CHILD POLICY ON MATERNAL EMPLOYMENT - RESULTS FROM SUBSAMPLE 1 ....................................... 58 TABLE XXII: THE EFFECT OF THE TWO-CHILD POLICY ON FAMILY SIZE WITH FURTHER CONTROLS .............................................................................................................. 59 TABLE XXIII: THE EFFECT OF THE TWO-CHILD POLICY ON MATERNAL EMPLOYMENT WITH FURTHER CONTROLS ...................................................................... 59 TABLE XXIV: THE ESTIMATES OF THE EFFECT OF THE TWO CHILD POLICY ON OTHER HOUSEHOLD OUTCOMES ......................................................................................... 60 TABLE XXV: DESCRIPTIVE STATISTICS ............................................................................. 86 TABLE XXVI: MEAN OF EDUCATION BY BIRTH COHORTS AND HIGH/LOW BASELINE REGIONS/ETHNICITIES – RESULTS FROM VHLSSs ...................................... 87 TABLE XXVII: MEANS OF NUMBER OF LIVING CHILDREN BY BIRTH COHORTS AND HIGH/LOW REGIONS AND ETHNICITY: RESULTS FROM CENSUS 2009 ............. 88 TABLE XXVIII: THE EFFECT OF THE UPE POLICY ON YEARS OF EDUCATION ACROSS HIGH/LOW BASELINE REGIONS - RESULTS FROM VHLSSs ........................... 89 TABLE XXIX: THE EFFECT OF THE UPE POLICY ON YEARS OF EDUCATION ACROSS ETHNICITY GROUPS - RESULTS FROM VHLSSs ................................................ 90 TABLE XXX: THE EFFECT OF THE UPE POLICY ON FERTILITY ACROSS HIGH/LOW REGIONS - RESULTS FROM CENSUS 2009 ........................................................................... 91 TABLE XXXI: THE EFFECT OF THE UPE POLICY ON FERTILITY ACROSS ETHNICITY GROUPS RESULTS FROM CENSUS 2009 ............................................................................... 92 xi TABLE XXXII: MEANS OF EDUCATION BY BIRTH COHORTS AND HIGH/LOW REGIONS AND ETHNICITY GROUPS - RESULTS FROM CENSUS 2009 .......................... 98 TABLE XXXIII: THE EFFECT OF THE UPE POLICY ON YEARS OF EDUCATION ACROSS HIGH/LOW REGIONS - RESULTS FROM CENSUS 2009 ..................................... 99 TABLE XXXIV: THE EFFECT OF THE UPE POLICY ON YEARS OF EDUCATION ACROSS ETHNICITY GROUPS - RESULTS FROM CENSUS 2009.................................... 100 TABLE XXXV: FRACTION OF MOTHERS HAVING ALL CHILDREN LIVING WITH THEM ......................................................................................................................................... 120 TABLE XXXVI: SUMMARY STATISTICS FOR WOMEN AGED 21-35 IN THE SAMPLES ..................................................................................................................................................... 121 TABLE XXXVII: FIRST STAGE RESULTS ........................................................................... 122 TABLE XXXVIII: OLS ESTIMATES OF THE EFFECT OF FERTILITY ON MATERNAL EMPLOYMENT IN VIETNAM ................................................................................................ 123 TABLE XXXIX: 2SLS ESTIMATES OF THE EFFECT OF FERTILITY ON MATERNAL EMPLOYMENT IN VIETNAM ................................................................................................ 124 TABLE XL: OLS ESTIMATES OF THE EFFECT OF CHILDREN ON MATERNAL EMPLOYMENT ACROSS EDUCATION AND URBAN/RURAL AREAS .......................... 125 TABLE XLI: 2SLS ESTIMATES OF THE EFFECT OF CHILDREN ON MATERNAL EMPLOYMENT ACROSS EDUCATIONAL LEVELS AND URBAN/RURAL AREAS ..... 126 TABLE XLII: PROBABILITY OF HAVING THE FIRST TWO GIRLS/TWO BOYS .......... 127 xii LIST OF FIGURES FIGURE Page Figure 1: Coefficients of the interactions between dummy variables of women’s age in 1989 and the ethnic majority in the probability of having more than two children equation– Results from Sample 1........................................................................................................................................ 31 Figure 2: Coefficients of the interactions between dummy variables of women’s age in 1989 and the ethnic majority in the mothers’ age at first birth equation – Results from subsample 2 ........ 32 Figure 3: Coefficients of the interactions between dummy variables of women’s age in 1989 and the ethnic majority in the proportion of sons equation - Results from subsample 2 .................... 33 Figure 4: Coefficients of the interactions between dummy variables of women’s age in 1989 and the ethnic majority in the maternal employment equation - Results from sample 1 .................... 34 Figure 5: Coefficients of the interactions between dummy variables of women’s age in 1989 and the ethnic majority in the probability of having more than two children equation– Results from the subsample of women with less than primary education ......................................................... 45 Figure 6: Coefficients of the interactions between dummy variables of women’s age in 1989 and the ethnic majority in the probability of having more than two children equation – Results from the subsample of women with at least primary education ............................................................ 45 Figure 7: Coefficients of the interactions between dummy variables of women’s age in 1989 and the ethnic majority in the probability of having more than two children equation – Results from the subsample of women in rural areas ......................................................................................... 46 Figure 8: Coefficients of the interactions between dummy variables of women’s age in 1989 and the ethnic majority in the probability of having more than two children equation – Results from the subsample of women in urban areas ....................................................................................... 46 xv SUMMARY In 1988, facing a total fertility rate of over four births per woman, the Vietnamese government introduced a new policy that required parents to have no more than 2 children. In the first chapter, using data from the Vietnam Population and Housing Censuses from 1989, 1999, and 2009, I apply a differences-in-differences framework to assess the effects of this policy on family size, son preference, and maternal labor supply. There are three main findings. First, the policy decreased the probability that a woman has more than two children by 15 percentage points (50%) for women aged less than 30 in 1989 and by 7 percentage points (11.5%) for women aged 30-39 in 1989. The policy reduced the average number of living children by 0.2 births per woman (10%). Low-educated women and women in rural areas were more affected by the policy. The policy had no effects on mothers’ age at first birth. Second, the policy decreased the proportion of sons in each family by 1.2 percentage points (2.4%). Third, the policy increased women’s labor force participation by 1.3 percentage points (1.5%). In 1991, the Vietnamese government implemented the universal primary education (UPE) law, which required that children aged 6 must enroll in primary schools at the beginning of the school year and children attending public schools were exempt from tuition fees. In the second chapter, using data from the Vietnam Household Living Standards Surveys from 2008, 2010, 2012, and 2014 and the Vietnam Population and Housing Census from 2009, I apply a differences-in- differences framework to assess the impacts of this policy on educational attainment and fertility. There are two main findings. First, the policy led to an average increase of 0.46 to 0.87 years of education. Second, there is no consistent evidence that the policy decreased the average number of living children. xvi In the third chapter, using the gender composition of the first two children in families with at least two children as instrumental variables for family size, which is an approach that has been used in other contexts (e.g. Angrist and Evans, 1998), I examine the causal relationship between fertility and maternal employment in Vietnam at different points of time in 1989, 1999, and 2009. The Vietnam setting is quite advantageous because of the well-documented son preference that causes families with only girls to have more children. I found no evidence that parents use sex selective abortion to achieve their son preference. The IV estimates of the effect of fertility on maternal employment in 1989 and 1999 are negative, suggesting that having an additional child potentially decreased maternal employment by 3.4 and 1.8 percentage points respectively. The estimate for 2009 is -0.02 and statistically significant, suggesting that the presence of an additional child reduced maternal employment by 2 percentage points in 2009. 1 CHAPTER I: EFFECTS OF VIETNAM’S TWO-CHILD POLICY ON FERTILITY, SON PREFERENCE, AND FEMALE LABOR SUPPLY A. INTRODUCTION High fertility rates and low economic growth are prevalent problems in developing countries. Family planning policies are considered a solution to these problems. In both China and India, the two most populous countries in the world, the governments have relied on family planning policies to limit family size1. Despite a body of literature that evaluates the effect of the one-child policy in China, we know little about the effect of the more common two-child policy, which is less extreme and has been promoted in several countries such as Vietnam, Indonesia, Iran, Singapore, Hong Kong, United Kingdom, and currently China2. Understanding the effect of this policy on family size facilitates evaluating the effects of the policy on other substantive issues such as child quality, parental labor supply, and ultimately economic prosperity. Studying the consequences of the two-child policy also has relevance to understanding central issues in economics such as the tradeoff between child quality and quantity, and the causal relationship between fertility and labor supply. In addition, in developing countries where there is a strong preference for sons and where sons act as social security for parents in their old age, this policy may affect gender balance as well. In 1988, facing a total fertility rate of over four births per woman, the Vietnamese government introduced a new policy that required parents to have no more than two children. The 1 In 1979, China formally introduced its dramatic one-child policy that limited family size to one child per couple. 2 As of the start of 2016, China relaxed its one-child policy and expanded it to the two-child policy in the hope to bring back its fertility rates to the replacement level. 4 first one to examine the effect of the policy on other outcomes, such as maternal labor supply. In addition, I show that reductions in fertility caused by the two-child policy are associated with a decline in the ability of parents to achieve son preference, which lowers the ratio of male-to-female children and the sex imbalance in many developing countries. Second, my paper identifies a new instrumental variable—the two-child policy—that can be used to examine the tradeoff between child quality and quantity, and the causal relationship between fertility and labor supply. The policy provides a natural experiment to examine whether with fewer children, parents will have “higher” quality children and whether women will work more. Third, my findings have important policy implications for developing countries where the governments rely on family planning policies to curb population growth and promote economic development. My results suggest that while the policy was effective at reducing family size, it had smaller effects on female labor supply. B. THE TWO-CHILD POLICY IN VIETNAM The two-child policy was recommended by the Vietnamese government in 1981 and made law in 1988. The goal of the policy is to maintain national population growth at 2 percent (Council of Ministers, 1989). The policy applied to every family except for families of ethnic minorities. Families of ethnic minorities could have a third child if they desire (Council of Ministers, 1989). Couples who already have one child, but have twins or triplets in the second birth are not considered to violate the policy. The specific guidelines of the policy also included requirements on the minimum childbearing age and the birth spacing (Council of Ministers, 1989). For state employees and government officials, childbearing age should be 22 or older for women and 24 or older for men. For others, the childbearing age should be 19 or older for women and 21 or older 5 for men. The second child, if desired, should be spaced 3 to 5 years apart from the first one, except for women aged 30 and older (Council of Ministers, 1989)4. To promote the two-child policy, the Vietnamese government imposed fines and punishments on families that violated the policy. The government denied a third child a birth certificate (PRI Review, 1995). If families violated the policy, they would be fined about $80, which was equivalent to 10 months of income in 1995 (PRI Review, 1995). State employees and government officials would not get promoted or would be relegated to lower status jobs in smaller cities, or in some cases, would lose their jobs if they violated the policy (Nikkei Asian Review, 2017). The government subsidized the fees of housing, healthcare, and education for the first two children, but not for the third child. Families with more than two children had to pay extra fees for housing, education, and health care of the third child (Council of Ministers, 1989). With these fines and punishments, the policy imposed real costs on families that have a third child and there is a plausible expectation that the policy reduced fertility. Besides imposing fines, the government also engaged in public health strategies, such as posters and billboards that depicted happier families with fewer children. Television programs that provided information on family planning were shown several times per week and at prime time (Goodkin, 1995). The government offered a reward of $20 to women who had a hysterectomy, “a procedure that approximately half of all village women were subjected” (PRI Review, 1995). The government also supplied birth control devices and birth control pills at free of charge to eligible people and poor families that registered to practice family planning (Council of Ministers, 1989). 4 These features of the policy were not equally enforced as the restriction to have no more than two children since the government did not impose punishments or fines on violating these requirements. 6 C. CONCEPTUAL FRAMEWORK In this section, I discuss economic theories on demand for children, son preference, and the effect of children on labor supply to highlight mechanisms through which the two-child policy could affect family size, son preference, and maternal employment. C.1. Demand for Children In developing countries where agriculture plays important role in the economy and people are poor, parents demand children as children act as parents’ retirement investment. People in developing countries cannot rely on social security or retirement plans in their old age and have to depend on their children for future support (Priebe, 2010). Thus, having more children increases the probability that one of them may be successful (high quality) and take care of parents in their old age. Especially in rural areas where most economic activities are farm work, having more children implies that families have more labor to work in the farm in the future. However, with harsh punishments that the government imposed on families that violated the two-child policy, families face a higher cost of having a third child and thus would not want to have more than two children as they did before. In other words, the policy is likely to cause a reduction in family size. The policy may affect mothers’ age at first birth as well. Women growing up under the two-child policy know that they will have only two children. As a result, they may delay having their first birth at a later age. In addition, the policy requires the minimum childbearing age of 22 for women who are state employees and government officials, and of 19 for others. Thus, it is plausible to expect that the policy increases mothers’ age at first birth. In the standard quantity and quality model of fertility, parents derive utility from both the quantity and quality of children (Becker and Lewis, 1973; Becker and Tomes, 1976; Rosenzweig 9 and already has both of them at the first two attempts, she will stop at two children. However, if she has achieved only one son, she may continue to three children to obtain the second son. Vietnam’s two-child policy may affect the ability of parents to achieve son preference. The policy may decrease the proportion of sons in each family because it lowers the number of children that a woman may have and places an additional constraint on her ability to achieve the desired number of sons. On the other hand, families may try to mediate the impact of the two-child policy on son preference by using sex-selective abortions to assure that they have at least one son. With the availability of ultrasound in the late 1980s and its widespread availability in provincial hospitals in mid-1990s, families could turn to sex-selective abortions as an alternative, especially if abortions were affordable for them6. If so, then the policy may increase the proportion of sons that families have. C.3. The Effect of Children on Female Labor Supply In the standard labor leisure model augmented to include a desire for children, women decide their fertility and their labor supply at the same time. Children require mothers’ time and increase the value of household work. Therefore, the more children a woman has, the less likely she is going to work. Vietnam’s two-child policy is an exogenous change in fertility. With the presence of the policy, women are less likely to have more than two children. For those affected, the desired number of children is above the realized number. With fewer children, the demands of child bearing and rearing on women’s time may fall, which potentially gives them more time for increased market work. In addition, with fewer children, mothers’ productivity at home may 6“Abortion has been legal in Vietnam since 1954 and is currently legal until 22 weeks of pregnancy.” (Tran, 2018). 10 decline, thus lowering the value of non-market time. Therefore, the two-child policy is expected to increase maternal employment. The effect of fertility on labor supply may vary with mothers’ schooling (Angrist and Evans, 1998) and across urban and rural areas. Gronau (1986) pointed out a number of empirical studies, which found that more educated women were more responsive to a fertility shock than less educated women. Gronau (1973) also documented that as mothers obtain more education, children have a stronger effect on their mothers’ value of time. Therefore, I expect that changes in family size due to the two-child policy will have a larger effect on more educated women. Women in urban areas may be more responsive to a shock in family size than women in rural areas. Women in urban areas tend to have more job opportunities and thus face a higher marginal cost of having a third child (Ebenstein, 2009). On the other hand, women in rural areas work at home and have a relatively lower marginal cost of having an additional child (Ebenstein, 2009). These considerations suggest that the effect of Vietnam’s two-child policy on maternal labor supply may be heterogeneous and differ by rural/urban residence (Priebe, 20107; Caceres-Delpiano, 20128). In this circumstance, it is plausible to expect that the policy have a larger effect on maternal employment of high-educated women and women in urban areas. As discussed above, these women face higher opportunity costs of childbearing and thus they would be more responsive to the fertility shock caused by the two-child policy. 7 Priebe (2010) examined the effect of fertility on maternal employment in Indonesia and documented that less educated women and women in rural areas were more responsive to the presence of children. 8 Caceres-Delpiano (2012) investigated the impact of children on maternal employment in 40 developing countries and found that the impact of children is stronger among high-educated mothers and mothers in urban areas. 11 D. DATA D.1. The Vietnam Population and Housing Censuses The data used in this study are from the Vietnam Population and Housing Censuses from 1989, 1999, and 20099. The data are from the 5-percent, 3-percent, and 15-percent nationally representative samples of the population (Minnesota Population Center, 2017, published data). The surveys include information on the number of children ever born and the number of surviving children of women at the time of survey. The surveys also contain other relevant information on individuals’ characteristics such as age, ethnicity, marital status, educational attainment, work, and current place of residency (at provincial levels). The analyses are conducted using two samples drawn from these three surveys. To examine the effect of the policy on women’s fertility and maternal labor supply, I use what I refer to as sample 1, which includes women aged 10-49 in 1989. Since the surveys ask fertility questions for women at the childbearing age and the key measure of treatment (exposure) is age in 1989, I do not observe every birth cohort (age in 1989) in all of the three Censuses. Specifically, I only observe fertility of women aged 40-49 in 1989 in the survey year 1989. Similarly, I only observe fertility of women aged 30-39 in 1989 in the survey years 1989 and 1999 (aged 40-49) and fertility of women aged less than 15 in 1989 in the survey years 1999 and 2009. Finally, for women aged 15 to 29 in 1989, I observe their fertility in 1989, 1999 (aged 25 to 39), and 2009 (aged 35 to 49). To investigate the impact of the policy on the proportion of sons in each family, I use a different sample -what I refer to as subsample 2- because this analysis requires that I know the 9 Access to the Vietnam Population and Housing Census from 1979 is not publicly available. Thus, I did not use the 1979 census in this study. 14 1989. Younger (e.g., <30) women in 1989 should be affected more by the policy than older (e.g., >39) women who are closer to have completed fertility by that age. In practice, I use women’s age in 1989 as a continuous variable to determine the length of the exposure to the policy. As an alternative specification, I also group women in three different age groups (less than 30, 30-39, and 40+ in 1989). The average probabilities of having more than two children and the average number of living children for four groups (Yi) are derived as below: Ethnic majority women Ethnic minority women Younger women in 1989 E (Yi |Majority=1, Younger = 1) E (Yi | Majority= 0, Younger =1) Older women in 1989 E (Yi |Majority= 1, Younger = 0) E (Yi | Majority= 0, Younger=0) The change in family size for ethnic majorities is [E (Yi | Majority = 1, Younger = 1) – E (Yi | Majority = 1, Younger = 0)] = cohort effects majority + policy effects Similarly, the change in family size for ethnic minorities is [E (Yi | Majority = 0, Younger = 1) - E (Yi | Majority = 0, Younger = 0)] = cohort effects minority My differences-in-differences (DID) estimates are as follows: DID = [E (Yi| Majority = 1, Younger = 1) – E (Yi| Majority= 1, Younger = 0)] – [E (Yi| Majority = 0, Younger = 1) – E (Yi| Majority= 0, Younger = 0)] = (cohort effects majority + policy effects) – (cohort effects minority) = (cohort effects majority - cohort effects minority) + policy effects (1) 15 Under the assumption that cohort effects majority = cohort effects minority, my DID estimates will capture the causal effect of the two-child policy on family size. In practice, I use the following regression model to apply my differences-in-differences framework: Y(ijt)= b0 +b1 age (it) + b2 age2 (it) + b3 age(it) ˣ majority(i) + b4 age2 (it) ˣ majority(i) + b5j ∑ age in 1989(ij) 48 j=10 + b6 majority(i) + b7j ∑ age in 1989(ij) 48 𝑗=10 ˣ majority(i) + province (ijt)+ v(ijt) (2) in which i = 1,..., N (index of person) j=10,…, 49 (index of age in 1989) t= survey years 1989, 1999, 2009 Y(ijt) is the probability of having more than two children, the number of living children, mothers’ age at first birth, and the proportion of sons in each family. The omitted group is women aged 49 in 1989. To account for differences in parental preferences and the costs of raising children across provinces, I also include provincial dummies in equation (2). The dummy variables indicating women’s age in 1989 measure the length of the exposure to the policy and an individual’s birth cohort. The indicator of the ethnic majority defines the treatment and control groups. The coefficients of interest are thus on the interaction terms between dummy variables of women’s age in 1989 and the ethnic majority (b7j). Here, b7j captures the relative effect of the two-child policy on family size and the proportion of sons of each cohort. To interpret b7j as the causal effect of the policy, I need to assume that in the absence of the policy, cohort effects would have been the same for both ethnic groups. While this assumption is untestable, I indirectly test it by examining cohort effects of women aged 40 and older in 1989 and 16 cohort effects of women aged less than 20 in 1989. Since women aged 40 and older in 1989 are too old to be affected by the policy, the estimates of these women will capture only the differences in cohort effects of both ethnic groups. Thus, if these estimates are close to zero, they imply that in the pre-policy period, changes in family size by birth year cohort are the same for both ethnic groups. On the other hand, women aged less than 20 in 1989 are fully affected by the policy. Thus, if I observe no differences in cohort effects of these women, it implies that cohort effects are likely to be the same for both ethnic groups in the post policy period. Figure 1 illustrates the coefficients of the interactions in equation (2) and provides the first piece of evidence that cohort effects are the same for both ethnic groups. As seen from Figure 1, the estimates of women aged 40 and older in 1989 (relative to the omitted group – women aged 49 in 1989) are close to zero and not statistically significant. Since these estimates capture differences in cohort effects of both ethnic groups, this implies that changes in family size are the same for both groups in the pre-policy period. As Figure 1 further indicates, the estimates of women aged less than 20 in 1989 are constant across age. This suggests that cohort effects are likely to be the same for both groups in the post-policy period. Although the evidence above suggests that the required common trend (birth year cohort) assumption holds, my estimates may be biased if the policy has spillover effects on ethnic minorities. If ethnic minorities follow the fertility behavior of ethnic majorities and stop at two children as well, then my estimates would be biased downward. Since the policy does not apply to ethnic minorities, ethnic majorities may have an incentive to marry ethnic minorities, which would also lead to spillover effects on ethnic minorities. Even though there are inter-ethnic marriages, they may not be a big concern here. The number of these marriages is very small. In 2009, the 19 E.3. Instrumental Variables Estimates of the Effect of Fertility on Maternal Employment in Vietnam To obtain the instrumental variables estimates of the effect of the number of children on maternal labor supply, I use estimates from the first stage regression of the two-child policy on the number of children to construct the predicted number of children, which is the instrument. The regression model used for the instrumental variables procedure is as follows: Mother’s employment ijt = e0 + e1 age it + e2 age2 it + e3 age it ˣ majority i + e4 age2 it ˣ majority i + e5j ∑ age in 1989ij 48 𝑗=10 + e6 majority (i) + e7 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑐ℎ𝑖𝑙𝑑𝑟𝑒𝑛𝑖𝑗𝑡 ̂ + e8 province ijt + ε ijt (5) Equation (5) uses the predicted number of children from equation (2) instead of the actual number of children. As I show later, the two-child policy is a significant predictor of the number of children and therefore the instrument has good explanatory power in the first stage. The exclusion restriction of the instrumental variables approach is that fertility should be the only channel through which the policy could affect maternal employment. Although the policy can affect maternal employment through other channels (e.g. delayed marriage, increase in education), these effects operate through the fertility channel. Thus, the exclusion restriction is still likely to hold in this context. F. RESULTS F.1. The Effect of Vietnam’s Two-child Policy on Family Size Figure 1 shows the coefficients of the interaction terms between dummy variables of women’s age in 1989 and the ethnic majority in the probability of having more than 2 children for 20 women’s birth cohorts 1941-197913. As seen from Figure 1, the estimates of women aged 40 and older in 1989 are close to zero and not statistically significant, indicating that there are no differences in cohort effects of both ethnic groups in the pre-policy period. Women aged 30 to 39 in 1989 are partially affected by the policy. There is a monotonic increasing effect of the policy for them. The estimates of these women range from -0.05 to -0.13, suggesting an average effect of a 9-percentage point decrease in the probability of having more than two children. As Figure 1 further indicates, women aged less than 30 in 1989 are the most affected group. The estimates of these women are constant across age, indicating a 15-percentage point decrease in the probability of having more than two children, equivalent to a 50% reduction at means. Table II contains the estimates of the effect of Vietnam’s two-child policy on the probability of having more than two children and the number of living children for women of different age groups in 1989. In Column (1) and (3), instead of including dummy variables for each age in 1989, I include indicators of three age groups as mentioned above. As Table II indicates, the policy was more effective at reducing fertility of younger women in 1989. The policy decreased the probability that a woman has more than two children by 15 percentage points (50%) for women aged less than 30 in 1989 and by 7 percentage points (11.5%) for women aged 30-39 in 1989. This result is in line with my expectation. Since most women in Vietnam have children in their 20s, the policy should have a bigger effect on fertility of these women. On average, the policy decreased the number of living children by 0.2 births per woman (10%). The estimates are very similar with the inclusion of year effects. 13 The coefficients of the interaction terms in Figure 1 are also shown in Table XVIII, Appendix. 21 As discussed in section C.1, there might be different effects of the two-child policy by education and urban/rural status. Table III presents estimates of the effect of the two-child policy on women’s fertility by mothers’ schooling1415. As Table III indicates, low-educated women were more affected by the policy. The policy decreased the probability of having more than two children for these women by 16 percentage points (37%) and reduced their average number of living children by 0.58 births per woman (24%). The policy, on the other hand, had a small effect on more-educated women. It reduced the probability of having more than two children of these women by 5 percentage points (20%) but had no effect on the number of living children. Table IV shows heterogeneity in the effect of the policy on family size by urban and rural areas16. As Table IV indicates, the policy was more effective at reducing fertility in rural areas. Since women in rural areas tend to less educated than women in urban areas, this finding is in line with the results shown in Table III, which finds that less-educated women are more affected by the policy. The policy decreased the probability of having more than two children of women in rural areas by 13 percentage points (35%) for women aged less than 30 and by 5 percentage points (7%) for women aged 30-39. The policy reduced the average number of living children of both age groups by 0.2 births per woman (a 9-percentage decrease for younger women and a 5.6- percentage decrease for middle-aged women). The estimates are very similar with the inclusion of year effects. 14 I also show the coefficients of the interaction terms between dummy variables of women’s age in 1989 and the ethnic majority in the fertility equation for women with less than primary education in Figure 5, Appendix and for women with at least primary education in Figure 6, Appendix. 15 If the policy has an impact on educational attainment, then women with less than primary education may obtain more education and thus have at least primary education. 16 I show the coefficients of the interaction terms in the fertility equation for women in rural areas in Figure 7, Appendix and for women in urban areas in Figure 8, Appendix. 24 The estimates are similar with the inclusion of year effects. In contrast, the policy had no effect on the labor supply of women aged 30-39 in 1989. Table VIII shows estimates of the effect of the policy on maternal labor supply by mothers’ schooling. The estimates suggest that the policy increased the labor supply of low-educated women aged less than 30 in 1989 by 3 percentage points (1.7%). The policy had no effects on the labor supply of low-educated women aged 30-39 in 1989. On the other hand, the policy decreased the labor supply of more-educated women by 4 percentage points (3.5%). The estimates are similar with the inclusion of year effects. Table IX presents estimates of the effect of the policy on maternal employment by urban and rural areas. The estimates suggest that the policy had opposite effects on the labor supply of women in urban and rural areas. The policy increased the labor supply of younger women in rural areas by 4 percentage points (4.4%) and it had no effects on the labor supply of women aged 30-39 in 1989. In contrast, the policy decreased the labor supply of younger women in urban areas by 12 percentage points (17%) and reduced the labor supply of middle-aged women by 7 percentage points (9.3%). Even though the estimates in Tables VIII and IX suggest an effect of the policy on the labor supply of women with at least primary education and women in urban areas, the results shown in Tables III and IV indicate no significant effects of the policy on the number of living children of these women. Since the mechanism through which the policy can affect maternal employment is through a reduction in fertility, it is hard to conclude that the policy affected maternal employment of these women if it had no effects on their fertility. 25 F.4. Instrumental Variables Estimates of the Effect of Fertility on Maternal Employment in Vietnam Table X shows the OLS and 2SLS estimates of the causal effect of fertility on maternal employment in Vietnam. As mentioned above, the interactions between dummy variables of women’s age in 1989 and ethnic majorities serve as instruments for the number of living children in the maternal employment equation. The exclusion restriction would be violated if the policy can affect maternal employment through other channels such as education and marriage. Although the policy may facilitate women to obtain more education, delay marriage, and thus participate more in the labor force, the effects of the policy on these outcomes operate through the effect of the policy on fertility. Thus, despite the potential effects of the policy on women’s education and marriage, the exclusion restriction is still likely to hold in this context. The upper panel presents the OLS estimates of the impact of children on maternal employment. The point estimate is -0.008 and statistically significant. The estimate indicates a small negative effect of children on mothers’ labor supply, which is in line with the findings of a recent study (Aaronson et al., 2017, unpublished document). The lower panel shows the instrumental variables estimates in equation (5). The point estimate suggests that having an additional child decreased maternal employment by 15 percentage points (17.4%). Compared to the estimates of other studies, my estimates are larger. The US estimates of fertility on maternal employment reported by (Angrist and Evans, 1998) are -10.4 percentage points for 1980 and -8.4 percentage points for 1990. The estimates reported by (Cruces and Galiani, 2007) also range from 8.1 to 9.6 percentage points for Argentina and from 6.3 to 8.6 percentage points for Mexico. However, my estimates capture the average effect of children on the labor 26 supply of women who comply with the two-child policy. In contrast, the estimates of other studies capture the average effect of fertility on the labor supply of women who either have multiple births or prefer a mixed-sibling gender composition. Thus, it is possible that my estimates are larger than the estimates of others. G. FALSIFICATION TESTS In this section, I present the results of falsification tests in which I use the probability of having at least one child and the probability of getting married as alternative outcomes. The two- child policy should have no effects on the probability that women have at least one child since it only imposes fines and punishments on parents who have more than two children. Similarly, the policy should not affect the probability of getting married of women aged 30 and over in 1989. Most of these women should have gotten married by the time that the policy was in place. The policy may have a small effect on the probability of getting married of women aged less than 30 in 1989. Since these women know that they will have only two children, they may adjust their age of marriage accordingly. Table XXIV, Appendix shows the coefficients of interest of alternative outcomes. As the table indicates, the policy had no effect on the probability that a woman has at least one child. The policy also did not affect the probability of getting married of women aged 30 and over in 1989. The policy had a small feedback effect on the probability of getting married of women aged less than 30 in 1989. The point estimate is 0.007 and statistically significant. These results are consistent with my expectations and together they suggest the validity of my research design. 29 Caceres-Delpiano, J.: Can we still learn something from the relationship between fertility and mother's employment? Evidence from developing countries. Demography, 49 (1), 151–174, 2012. Council of Ministers: Vietnam’s New Fertility Policy. Population and Development Review, 15: 169-172, 1989. Clark, S.: Son Preference and Sex Composition of Children: Evidence from India. Demography 37:95–108, 2000. Cruces, G., and Galiani, S.: Fertility and female labor supply in Latin America: new causal evidence. Lab Econ, 14(3):565–573, 2007. Ebenstein, A.: When is the local average treatment close to the average? Evidence from fertility and labor supply. Journal of Human Resources, 44(4): 955-75, 2009. General Statistics Office of Vietnam (GSO). The 2009 Vietnam Population and Housing Census, 2010. Available at http://www.gso.gov.vn/default_en.aspx?tabid=515&idmid=5&ItemID=9813 Goodkind, D.: Vietnam's One-or-Two Child Policy in Action. Population and Development Review, 21:85-109, 1995. Gronau, R.: The Effect of Children on the Housewife's Value of Time. Journal of Political Economy, Pt. II, 81(2), pp. S168-99, 1973. Gronau, R.: Home Production-A Survey. In Handbook of labor economics, ed. Orley Ashenfelter and Richard Layard, Vol. 1. Amsterdam: North-Holland, pp. 273-304, 1986. Jensen, R.T.: Equal Treatment, Unequal Outcomes? Generating Sex Inequality through Fertility Behaviour. Unpublished document. JFK School of Government, Harvard University, 2002. Hoa, H. T., Toan, N. V., Johansson, A., Hoa, V.T., Hojer, B., & Persson, L. A.: Child spacing and two child policy in practice in rural Vietnam: Cross sectional survey. BMJ, 313, 1113–1116, 1996. Larsen, U., Chung, W., Gupta M.D.: Fertility and son preference in Korea. Population Studies, 52, 317-325, 1998. Li, H., Zhang, J., Zhu, Y.: The effect of the One-Child Policy on fertility in China: identification based on differences-in-differences, mimeo. Chinese University of Hong Kong, 2005. Minnesota Population Center: Integrated Public Use Microdata Series, International: Version 6.5 [dataset]. Minneapolis: University of Minnesota, 2017. Nikkei Asian Review. Vietnam’s declining birthrate spells end of two-child policy, 2017. Available at http://asia.nikkei.com/Politics-Economy/Policy-Politics/Vietnam-s-declining- birthrate-spells-end-of-two-child-policy Priebe, J.: Child Costs and the Causal Effect of Fertility on Female Labor Supply: An Investigation for Indonesia 1993-2008. CRC-PEG Discussion Paper No. 45, University of Göttingen, 2010. 30 Population Research Institute: PRI Review 5(5): September/October. San Francisco Chronicle: 1- 2, 1995. Rosenzweig, M. R., and Wolpin, K.I.: Testing the Quantity-Quality Fertility Model: The Use of Twins as a Natural Experiment. Econometrica, 48(1), pp. 227-40, 1980. Tran, M.H.: Coping with Sex-Selective Abortions in Vietnam: An Ethnographic Study of Selective Reproduction as Emotional Experience. In Selective Reproduction in the 21st Century, ed. A. Wahlberg, T.M. Gammeltoft, 2018. 31 FIGURES Figure 1: Coefficients of the interactions between dummy variables of women’s age in 1989 and the ethnic majority in the probability of having more than two children equation– Results from Sample 1 34 Figure 4: Coefficients of the interactions between dummy variables of women’s age in 1989 and the ethnic majority in the maternal employment equation - Results from sample 1 35 TABLES TABLE I: SUMMARY STATISTICS OF WOMEN’S BIRTH COHORTS 1940-1979 Variables Sample 1 Subsample 2 Mean or % SD Mean or % SD Number of living children 2.173 1.570 2.079 1.248 % Having No child 15.70 7.23 One child 15.04 23 2 children 34.91 42.31 >3 children 34.35 27.46 Probability (having more than 2 children) 0.344 0.475 0.275 0.446 Proportion of sons in the family 0.506 0.394 Probability (being employed) 0.856 0.351 0.872 0.334 Age 35.948 8.193 33.615 7.458 Mothers’ age at first birth 23.067 3.914 Rural 0.654 0.476 0.690 0.462 Less than primary education 0.340 0.474 0.327 0.469 Number of Obs. 3,197,622 1,852,725 Note: Sample 1 includes all women aged 10-49 in 1989 in three survey years. Subsample 2 includes women aged 10-19 in 1989 who satisfies the two following conditions: (1) they are household heads or wives of household heads and (2) they have all children living with them. 36 TABLE II: THE EFFECT OF THE TWO-CHILD POLICY ON THE PROBABILITY OF HAVING MORE THAN TWO CHILDREN AND THE NUMBER OF LIVING CHILDREN Dependent Variables Probability (more than 2 children) Number of living children Mean for women <30 in 1989 (SD) 0.30 (0.46) 2.01 (1.41) Mean for women 30-39 in 1989 (SD) 0.61 (0.49) 3.14 (1.90) (1) (2) (3) (4) age 89<30 × majority -0.149*** -0.152*** -0.183*** -0.192*** (0.005) (0.005) (0.031) (0.031) age in 1989 (30- 39) × majority -0.067*** -0.067*** -0.195*** -0.196*** (0.006) (0.006) (0.035) (0.035) year 1999 N Y N Y year 2009 N Y N Y Province FE Y Y Y Y N 3,197,622 3,197,622 3,197,622 3,197,622 Note: The numbers in parentheses are standard errors. The omitted group is those aged 40 and older in 1989. Other covariates included in the regressions are the dummy variables of women aged less than 30 in 1989, women aged 30-39 in 1989, the ethnic majority, age at the time of survey, age squared, age× majority, and age squared× majority. *p<0.05. **p<0.01, ***p<0.001. 39 TABLE V: THE EFFECT OF VIETNAM’S TWO CHILD POLICY ON MOTHERS’ AGE AT FIRST BIRTH Dependent Variable Mothers’ age at first birth Mean – for women < 30 in 1989 (SD) 23.283 (3.943) Mean – for women 30-39 in 1989 (SD) 24.223 (4.086) age less than 30 in 1989 × majority 0.132 0.130 (0.086) (0.086) age 30-39 in 1989 × majority 0.0283 0.031 (0.094) (0.094) year 1999 N Y year 2009 N Y Province FE Y Y N 1,768,536 1,768,536 Note: The numbers in parentheses are standard errors. The omitted group is those aged 40 and older in 1989. Other covariates included in the models are the indicators of women aged less than 30 in 1989, women aged 30-39 in 1989, the ethnic majority, age, age squared, age × majority, and age squared × majority. *p<0.05, **p<0.01, ***p<0.001. 40 TABLE VI: THE EFFECT OF VIETNAM’S TWO CHILD POLICY ON THE PROPORTION OF SONS IN EACH FAMILY Dependent Variable Proportion of sons in each family Mean –for women < 30 (SD) 0.499 (0.386) Mean - for women 30-39 in 1989 (SD) 0.506 (0.322) age in 1989 <30 × majority -0.0123* (0.005) -0.0051 (0.006) N N -0.0124* (0.005) -0.0052 (0.006) Y N age in 1989 (30-39) × majority year 1999 year 2009 Province FE Y Y N 1,852,725 1,852,725 Note: The numbers in parentheses are standard errors. The omitted group is those aged 40 and older in 1989. Other covariates included in the models are the dummy variables of women aged less than 30 in 1989, women aged 30-39 in 1989, and the ethnic majority, age, age squared, age × majority, and age squared × majority. *p<0.05, ** p<0.01, *** p<0.001. 41 TABLE VII: THE EFFECT OF VIETNAM’S TWO CHILD POLICY ON MATERNAL EMPLOYMENT Dependent Variable Probability (being employed) Mean for women < 30 in 1989 (SD) 0.86 (0.34) Mean for women 30-39 in 1989 (SD) 0.80 (0.40) (1) (2) age 89<30 × majority 0.012** 0.013** (0.004) (0.004) age 89 (30-39) × majority -0.002 -0.001 (0.005) (0.005) year 1999 N Y year 2009 N Y Province FE Y Y N 3,126,106 3,126,106 Note: The number in parentheses are standard errors. The omitted group is those aged 40 and older in 1989. Other covariates included in the models are the indicators of women aged less than 30 in 1989, women aged 30-39 in 1989, and the ethnic majority, age, age squared, age × majority, and age squared × majority. *p<0.05, **p<0.01***p<0.001. 44 TABLE X: THE EFFECT OF FERTILITY ON MATERNAL EMPLOYMENT IN VIETNAM Dependent variable: Probability of being employed Coefficient OLS Number of living children -0.008* (0.0003) 2SLS Coefficient Number of living children (Year of birth dummies × the ethnic majority) -0.15* (0.013) Number of Obs. 3,126,106 Note: The numbers in parentheses are standard errors. Year of birth dummies, province indicators, the ethnic majority indicator, age, age squared, age × majority, and age squared × majority are included in the regressions. * p<0.05. 45 Appendix A Figure 5: Coefficients of the interactions between dummy variables of women’s age in 1989 and the ethnic majority in the probability of having more than two children equation– Results from the subsample of women with less than primary education Figure 6: Coefficients of the interactions between dummy variables of women’s age in 1989 and the ethnic majority in the probability of having more than two children equation – Results from the subsample of women with at least primary education 46 Figure 7: Coefficients of the interactions between dummy variables of women’s age in 1989 and the ethnic majority in the probability of having more than two children equation – Results from the subsample of women in rural areas Figure 8: Coefficients of the interactions between dummy variables of women’s age in 1989 and the ethnic majority in the probability of having more than two children equation – Results from the subsample of women in urban areas 49 TABLE XII: SUMMARY STATISTICS FOR SAMPLE 1 OF WOMEN’S BIRTH COHORTS 1940-1979: ETHNIC MAJORITY VERSUS ETHNIC MINORITY Census 1989 Census 1999 Census 2009 Ethnic Majority Ethnic Minority Ethnic Majority Ethnic Minority Ethnic Majority Ethnic Minority Number of children 1.83 (2.05) 2.23 (2.31) 1.93 (1.63) 2.40 (1.93) 2.19 (1.19) 2.8 (1.60) % Having No child 38.62 33.69 22.2 18.31 7.79 5.41 One child 14.65 13.2 20.45 16.36 15.01 9.74 2 children 15.87 14.37 27.25 24.31 45.04 34.81 >3 children 30.86 38.74 30.1 41.02 32.16 50.04 Probability (≥ 2children) 0.31 (0.46) 0.39 (0.49) 0.3 (0.46) 0.41 (0.50) 0.32 (0.47) 0.5 (0.5) Percent employed 0.76 (0.42) 0.91 (0.29) 0.79 (0.41) 0.91 (0.29) 0.867 (0.33) 0.96 (0.20) Mothers’ age 25.35 24.69 35.36 34.86 45.12 44.41 (10.42) (10.58) (10.40) (10.55) (10.25) (10.41) Rural 0.55 0.8 0.45 0.69 0.68 0.88 (0.5) (0.4) (0.5) (0.46) (0.47) (0.33) < Primary Education 0.36 (0.48) 0.66 (0.47) 0.26 (0.44) 0.54 (0.50) 0.28 (0.45) 0.62 (0.48) N of Obs. 706,134 119,591 517,049 82,362 2,436,388 575,442 Note: The numbers in parentheses are standard deviations. 50 TABLE XIII: SUMMARY STATISTICS FOR SUBSAMPLE 2 OF WOMEN’S BIRTH COHORTS 1940-1979: ETHNIC MAJORITY VERSUS ETHNIC MINORITY Censuses Census 1989 Census 1999 Census 2009 Ethnic Majority Ethnic Minority Ethnic Majority Ethnic Minority Ethnic Majority Ethnic Minority Number of children 2.66 3.15 2.24 2.66 1.85 2.25 (1.69) (1.89) (1.33) (1.59) (1.00) (1.30) % Having No Child 5.93 4.60 5.79 5.07 8.54 4.98 One Child 19.92 14.88 22.56 17.40 24.53 21.32 2 Children 26.98 22.07 37.75 31.40 47.21 40.81 >=3 Children 47.17 58.45 33.90 46.13 19.72 32.89 Probability (>=2 children) 0.47 0.58 0.34 0.46 0.20 0.33 (0.50) (0.49) (0.47) (0.50) (0.40) (0.47) Proportion of sons 0.487 0.489 0.498 0.497 0.493 0.509 (0.357) (0.333) (0.373) (0.351) (0.46) (0.333) % employed 0.82 0.93 0.79 0.92 0.87 0.96 (0.39) (0.25) (0.41) (0.26) (0.34) (0.19) Mothers’ age 32.21 31.90 33.94 32.60 34.24 32.44 (6.93) (7.39) (7.13) (7.23) (7.47) (7.65) Mothers’ age of first birth 23.288 22.659 23.315 22.528 23.748 22.868 (3.744) (4.076) (3.806) (3.779) (4.102) (3.977) Rural 0.58 0.82 0.47 0.72 0.68 0.90 (0.49) (0.38) (0.50) (0.45) (0.47) (0.29) < Primary Education 0.36 0.67 0.25 0.54 0.23 0.63 (0.48) (0.47) (0.43) (0.50) (0.42) (0.48) N of Obs. 248,858 38,044 262,071 45,423 1,411,727 424,808 Note: The numbers in parentheses are standard deviations. 51 TABLE XIV: SUMMARY STATISTICS FOR THE SUBSAMPLE OF WOMEN WITH LESS THAN PRIMARY EDUCATION OF BIRTH COHORTS 1940-1979: ETHNIC MAJORITY VERSUS ETHNIC MINORITY Census 1989 Census 1999 Census 2009 Ethnic Majority Ethnic Minority Ethnic Majority Ethnic Minority Ethnic Majority Ethnic Minority Number of children 2.63 (2.47) 2.66 (2.50) 2.51 (2.04) 2.82 (2.14) 2.39 (1.44) 3.13 (1.72) Percent Having No child 29.07 28.35 19.43 16.22 9.57 5.21 One child 11.35 12.01 16.15 13.54 13.91 6.99 2 children 12.8 12.84 19.57 19.22 34.49 27.45 >=3 children 46.78 46.8 44.85 51.02 42.03 60.35 Probability (≥2 children) 0.47 (0.50) 0.47 (0.50) 0.45 (0.50) 0.51 (0.50) 0.42 (0.50) 0.60 (0.50) Percent employed 0.69 (0.46) 0.91 (0.28) 0.68 (0.47) 0.88 (0.31) 0.74 (0.44) 0.90 (0.30) Age 27.78 24.99 37.52 35.58 47.72 44.56 (11.47) (10.39) (11.42) (11.37) (11.21) (10.86) Rural 0.67 0.88 0.59 0.81 0.78 0.92 (0.47) (0.33) (0.49) (0.39) (0.41) (0.27) % of the sample 76.1 23.9 74.48 25.52 65.53 34.47 N of Obs. 484,806 152,288 251,931 86,343 1,328,975 699,169 Note: The number in parentheses are standard deviations. 54 TABLE XVII: SUMMARY STATISTICS FOR THE SUBSAMPLE OF WOMEN IN RURAL AREAS OF BIRTH COHORTS 1940-1979: ETHNIC MAJORITY VERSUS ETHNIC MINORITY Census 1989 Census 1999 Census 2009 Ethnic Majority Ethnic Minority Ethnic Majority Ethnic Minority Ethnic Majority Ethnic Minority Number of children 2.0 (2.15) 2.43 (2.37) 2.20 (1.78) 2.66 (2.0) 2.33 (1.22) 2.90 (1.60) Percent Having No child 36.66 30 19.77 15.24 6.47 4.45 1 child 13.2 13.19 17.83 14.35 12.53 8.57 2 children 15 14.29 24.71 22.92 43.20 33.79 >=3 children 35.14 42.52 37.69 47.49 37.80 53.19 Probability (≥2 children) 0.35 (0.48) 0.43 (0.50) 0.38 (0.48) 0.47 (0.50) 0.38 (0.48) 0.53 (0.50) Percent employed 0.82 (0.38) 0.96 (0.20) 0.82 (0.38) 0.94 (0.24) 0.86 (0.35) 0.93 (0.26) Age 24.98 24.31 35.10 34.57 45.17 44.31 (10.46) (10.63) (10.45) (10.58) (10.29) (10.42) Less than primary 0.48 (0.50) 0.76 (0.43) 0.39 (0.49) 0.68 (0.47) 0.38 (0.49) 0.71 (0.45) % of the sample 80.08 19.9 80.03 19.97 91.63 8.37 N of Obs. 391,849 95,335 231,344 56,735 1,662,459 504,801 Note: The numbers in parentheses are standard deviations. 55 TABLE XVIII: THE ESTIMATES OF THE EFFECT OF THE TWO CHILD POLICY ON THE PROBABILITY OF HAVING MORE THAN 2 CHILDREN - RESULTS FROM SAMPLE 1 Dependent Variable Probability that a woman has more than 2 children Birth year × Coefficient 95% CI majority Age in 1989 1941 48 0 -0.04 0.04 1942 47 -0.02 -0.07 0.02 1943 46 -0.01 -0.05 0.04 1944 45 -0.04 -0.08 0.01 1945 44 -0.02 -0.06 0.03 1946 43 -0.03 -0.07 0.01 1947 42 -0.01 -0.06 0.03 1948 41 0.01 -0.03 0.05 1949 40 0 -0.04 0.04 1950 39 -0.05 -0.09 -0.02 1951 38 -0.04 -0.08 -0.01 1952 37 -0.05 -0.09 -0.01 1953 36 -0.05 -0.09 -0.01 1954 35 -0.06 -0.1 -0.02 1955 34 -0.07 -0.1 -0.03 1956 33 -0.08 -0.12 -0.04 1957 32 -0.11 -0.15 -0.08 1958 31 -0.11 -0.15 -0.08 1959 30 -0.13 -0.17 -0.09 1960 29 -0.16 -0.19 -0.13 1961 28 -0.17 -0.2 -0.13 1962 27 -0.17 -0.2 -0.13 1963 26 -0.17 -0.2 -0.14 1964 25 -0.16 -0.2 -0.13 1965 24 -0.16 -0.19 -0.13 1966 23 -0.16 -0.19 -0.13 1967 22 -0.16 -0.19 -0.13 1968 21 -0.16 -0.2 -0.13 1969 20 -0.16 -0.2 -0.13 1970 19 -0.17 -0.2 -0.14 1971 18 -0.17 -0.2 -0.14 1972 17 -0.17 -0.2 -0.14 1973 16 -0.17 -0.2 -0.14 1974 15 -0.17 -0.2 -0.13 1975 14 -0.14 -0.18 -0.11 1976 13 -0.14 -0.17 -0.1 1977 12 -0.13 -0.16 -0.1 1978 11 -0.13 -0.16 -0.1 1979 10 -0.13 -0.16 -0.1 56 TABLE XIX: THE ESTIMATES OF THE EFFECT OF THE TWO CHILD POLICY ON MOTHERS’AGE AT FIRST BIRTH FOR WOMEN’S BIRTH COHORTS 1940- 1979 - RESULTS FROM SUBSAMPLE 2 Dependent Variable Mothers' age at first birth Birth year × majority age in 1989 Coefficient 95% CI 1941 48 0.152 -0.795 1.099 1942 47 -0.451 -1.371 0.468 1943 46 -0.673 -1.593 0.247 1944 45 -0.233 -1.128 0.662 1945 44 -0.743 -1.602 0.116 1946 43 -0.912 -1.755 -0.069 1947 42 -0.959 -1.846 -0.072 1948 41 -0.371 -1.178 0.436 1949 40 -0.910 -1.699 -0.120 1950 39 -0.600 -1.373 0.174 1951 38 -1.037 -1.804 -0.270 1952 37 -0.533 -1.303 0.238 1953 36 -0.738 -1.496 0.020 1954 35 -0.407 -1.160 0.346 1955 34 -0.436 -1.181 0.309 1956 33 -0.569 -1.312 0.174 1957 32 -0.380 -1.125 0.364 1958 31 -0.446 -1.179 0.287 1959 30 -0.526 -1.269 0.217 1960 29 -0.386 -1.103 0.331 1961 28 -0.453 -1.168 0.263 1962 27 -0.600 -1.319 0.119 1963 26 -0.506 -1.220 0.208 1964 25 -0.576 -1.291 0.139 1965 24 -0.447 -1.162 0.268 1966 23 -0.517 -1.233 0.198 1967 22 -0.522 -1.238 0.193 1968 21 -0.426 -1.141 0.290 1969 20 -0.498 -1.214 0.217 1970 19 -0.493 -1.207 0.222 1971 18 -0.504 -1.219 0.212 1972 17 -0.496 -1.211 0.220 1973 16 -0.470 -1.185 0.246 1974 15 -0.368 -1.083 0.346 1975 14 -0.342 -1.057 0.372 1976 13 -0.254 -0.968 0.460 1977 12 -0.157 -0.871 0.556 1978 11 -0.028 -0.742 0.685 1979 10 0.040 -0.674 0.754 59 TABLE XXII: THE EFFECT OF THE TWO-CHILD POLICY ON FAMILY SIZE WITH FURTHER CONTROLS Probability (having more than 2 children) Number of living children age in 1989<30× majority -0.152*** -0.144*** -0.133*** -0.192*** -0.166*** -0.123*** (0.005) (0.005) (0.005) (0.031) (0.031) (0.03) age in 1989 (30-39) × majority -0.067*** -0.067*** -0.063*** -0.196*** -0.196*** -0.180*** (0.006) (0.006) (0.006) (0.035) (0.035) (0.035) N 3,197,622 3,197,622 3,188,893 3,197,622 3,197,622 3,188,893 Controls Year indicators Y Y Y Y Y Y Rural Indicators N Y Y N Y Y Education indicators N N Y N N Y Note: The numbers in parentheses are standard errors. Other covariates included in the models are the indicators of women aged less than 30 in 1989, women aged 30-39 in 1989, the ethnic majority, age, age squared, age × majority, and age squared × majority, and province indicators. *p<0.05. **p<0.01, ***p<0.001. TABLE XXIII: THE EFFECT OF THE TWO-CHILD POLICY ON MATERNAL EMPLOYMENT WITH FURTHER CONTROLS Proportion of sons Probability of being employed age in 1989<30 × majority -0.013* -0.012* -0.011* 0.013** 0.019*** 0.015*** (0.005) (0.006) (0.006) (0.004) (0.004) (0.004) age in 1989 (30-39) × majority -0.0051 -0.005 -0.0048 -0.0008 -0.0008 -0.001 (0.006) (0.007) (0.007) (0.005) (0.005) (0.005) N 1,852,725 1,852,725 1,850,465 3,126,106 3,126,106 3,123,326 Controls Year indicators Y Y Y Y Y Y Rural indicators N Y Y N Y Y Education indicators N N Y N N Y Note: The numbers in parentheses are standard errors. Other covariates included in the models are the indicators of women aged less than 30 in 1989, women aged 30-39 in 1989, the ethnic majority, age, age squared, age × majority, age squared × majority, and province indicators. * p<0.05**p<0.01, ***p<0.001. 60 TABLE XXIV: THE ESTIMATES OF THE EFFECT OF THE TWO CHILD POLICY ON OTHER HOUSEHOLD OUTCOMES Probability of having at least one child Probability of getting married age in 1989 <30 × majority 0.0001 0.007** (0.003) (0.003) age in 1989 (30-39) × majority 0.003 0.004 (0.004) (0.004) N 3,197,622 3,196,708 Note: The numbers in parentheses are standard errors. Other covariates included in the models are the indicators of women aged less than 30 in 1989, women aged 30-39 in 1989, the ethnic majority, age, age squared, age × majority, age squared × majority, year indicators, and province indicators. * p<0.05, **p<0.01. 61 CHAPTER II. EFFECTS OF UNIVERSAL PRIMARY EDUCATION (UPE) LAW ON EDUCATIONAL ATTAINMENT AND FERTILITY: EVIDENCE FROM VIETNAM A.INTRODUCTION Education is considered one of the most important human capital investments and those with more education receive lifelong benefits manifested by greater income, better health and overall improvements in wellbeing. Previous research has shown that the labor market returns to education tend to be larger at lower levels of education (i.e. highest at primary education) and larger in developing countries than developed ones (Psacharopoulos, 1994). That is the primary reason why many developing countries aim to provide their people with at least some primary education. Indeed, the United Nations has considered universal primary education one of its millennium development goals (United Nations, 2013). Despite the world’s efforts to achieve universal primary education in 2015, 58 million (9%) children of the primary school age were still out of school in 2012 (UNESCO, 2015). Among them, 43% never entered school; 34% entered school late; and the rest (23%) attended school but dropped out early. The majority of these out-of-school children are those from the poorest families, children from rural areas, and children from ethnic minority groups (UNESCO, 2015). As estimated, children from the poorest families (lowest income quintile) are more than five times more likely to stay out of school than are children from the richest families (UNESCO, 2015). Similarly, in rural areas children are twice less likely to be in schools than in urban areas (UIS and UNICEF, 2015). One of the most common barriers that keep these children to be out of school is the cost of enrolling in schools. Especially, in developing countries where people are poor, parents may have difficulties to finance their children’s education, and the cost of enrolling in primary 64 attendance and reduced the inequalities in attendance across gender, income, and regions20. Using data from over 900 rural households in Uganda, another study found that the UPE policy helped to decrease delayed enrollment and increase grade completion rates, especially for poor households (Nishimura et al., 2008). Similarly, Osili and Long (2008) investigated the effects of the Nigerian UPE program and found that the program increased the education of the affected women by 1.54 years of education and reduced the number of children that women give births before age 25 by 1.09. Osili and Long (2008) further documented that a one-year increase in women’s education led to a reduction of 0.26 (11%) to 0.48 (19%) births per woman. Exploring exogenous variation in education caused by the UPE policies, previous studies also examine the causal relationship between education and other outcomes such as fertility and child mortality. Behrman (2015) utilized the exogenous variation caused by the UPE policies in Malawi, Uganda, and Ethiopia to investigate the impact of education on women’s self-reported, desired fertility. The author found that a one-year increase in education decreased women’s ideal family size by 0.34 in Malawi and Ethiopia and by 0.11 in Uganda. A one-year increase in education also reduced the probability that a woman desired a high number of children by 0.05- 0.11 percentage points in these countries. Likewise, Makate and Makate (2016) exploited the variation in education caused by Malawi’s 1994 universal primary education program to examine the impact of education on child mortality. The authors documented that a one-year increase in education reduced mortality for infants by 3.22 percentage points and mortality for children under age five by 6.48 percentage points. However, this one-year increase in education had no significant effects on neonatal survival. Recently, a study investigated the effects of Ethiopia’s nationwide 20 Uganda implemented its UPE program in 1997, which got rid of the fees of primary education for up to four children in households with at least two girls (Deininger, 2003). 65 free primary education program and documented that an additional year of education led to a reduction of 0.155 births per woman (Chicoine, 2017, unpublished document). Similarly, several studies that use exogenous variations in schooling caused by a school construction program and compulsory schooling laws also document the effects of women’s education on fertility and children’s health outcomes. Capitalizing on a massive school construction program in Indonesia between 1973 and 1978, Breierova and Duflo (2004) found that parental education had a strong negative effect on child mortality and that mothers’ education had a dominant effect on fertility decisions than fathers’ education. Chou et al (2010) examined the impact of Taiwan’s 1968 compulsory education reform that increased compulsory schooling from 6 to 9 years and documented the positive effects of an increase in parental education on infant health outcomes. Exploiting a policy change in Kenya in 1985, which required children to stay in primary schools by one more year, Chicoine (2012) also found that the reform increased education by 0.74 years, delayed women’s marriages by 6.2 percentage points, and reduced a woman’s fertility by almost a third of a child at the age of 25. Likewise, Dincer, Kaushal, and Grossman (2014) took advantage of Turkey’s compulsory education reform that raised compulsory schooling from 5 to 8 years and found a reduction of 0.11 births per woman due to a 10-percentage point increase in the fraction of married mothers with eight years of education. Overall, these studies indicate that, not surprisingly, the UPE policies have increased school attendance and educational attainment. In addition, there is consistent empirical evidence supporting that increased schooling likely decreases women’s desired fertility and child mortality. Thus, the UPE policies should reduce women’s fertility through both of the above-mentioned channels. 66 C. CONCEPTUAL FRAMEWORK C.1. The Effect of the UPE Policy on Education In Vietnam, like other developing countries, particularly in 1991, children often stayed home to help with the household chores or to work in a city or a farm for a small amount of income21. In short, children are an important source of household income. The UPE policy was targeted at changing this outcome. The main feature of Vietnam’s UPE policy is to make primary education free for children, although it also mandated enrollment. In Vietnam, people value education highly and education is considered a path out of poverty. Thus, even though there were no penalties for parents not complying with the policy, parents were still more likely to send their children to schools. In other words, it is plausible to expect that the policy will increase people’s educational attainment. The effect of the UPE policy on educational attainment may vary across families by income, region and ethnicity. The policy should have a stronger effect on children of low-income families and children of ethnic minorities. High-income families can afford primary education and they will send their children to schools even in the absence of the UPE policy. On the other hand, tuition fees of primary education may account for a significant portion of poor families’ income. Thus, in the absence of the policy, poor families cannot afford to send their children to schools. On the other hand, it is ambiguous to determine whether the effect of the policy would be stronger in rural or urban areas. Families in rural areas tend to be poorer than the others. In addition, 21 The child labor law is not well enforced in Vietnam. In 2012, 1.75 million Vietnamese children, with two out of every five of them under the age of 15 were still child laborers (ILO and GSO of Vietnam, 2014). 69 ages 22 and 28. All people in the treatment group are 22 or 23 and all people in the control group are 24 to 28. In 2014, the sample is between ages 25 to 34, and people in the treatment group are aged 25 to 29 and people in the control group are aged 30 to 34. The collinearity of age, period, and cohort is addressed in the research design. D.2. Vietnam Population and Housing Census Although I use the VHLSSs to examine the effect of the UPE policy on educational attainment, I do not use the surveys to investigate the effect of the policy on women’s fertility since the surveys do not ask women in the childbearing age about the number of children ever born and surviving23. Thus, I use the Vietnam Population and Housing Census from 2009 to examine the impact of the UPE law on fertility. The Census 2009 provides me with the information on the number of children ever born and survived of all women aged 15-49 at the time of survey (Minnesota Population Center, 2017, published data). The survey also includes information on individuals’ age, gender, ethnicity, marital status, educational levels, and current residency (at provincial levels). The fertility sample is restricted to individuals born between 1980 and 1989. However, the sample is not restricted to those aged 22 and over at the time of the survey because those in the treatment group were between the ages of 20 and 24 in 2009. All people in the control group are 25 to 29. Again, the collinearity between age, period, and cohort is addressed in the research design. 23 The Demographic Health Surveys (DHS) for Vietnam have full information on women’s fertility and their birth history. However, the latest DHS available is from 2005. At that time, all of people in the treatment and control groups were just between ages 16 and 25. 70 D.3. Descriptive Statistics Individuals’ years of education are derived from individuals’ self-reported highest educational levels. The standard years of education of each educational level in Vietnam are as follows: 5 years for compulsory primary education, 4 years for secondary school, and 3 years for high school. For individuals who do not complete primary/secondary/or high schools, their years of education are derived based on their self-reported years of schooling. Fertility is measured as the number of living children that women reported at the time of survey. Table XXV presents descriptive statistics for individuals in the education and fertility samples. Compared to individuals in the fertility sample, people in the education sample are older. The average age of individuals in the education sample and in the fertility sample is 26.8 and 23.7 respectively. As Table XXV further indicates, the average years of education in both samples are similar. The average years of education are 9.1 and 9.2 in both samples respectively. Approximately 22% of the people in the samples are ethnic minorities. The average number of living children in the fertility sample is 0.75. Approximately 71% of the people in the samples live in rural areas. E. RESEARCH DESIGN E.1. Defining Treatment Together, the year of birth, region of birth, and family income identify individuals’ exposure to the policy. The policy required that children aged 6 must enroll in primary schools at the beginning of school year. Thus, those aged 6 or less in 1991 (born in 1985 and after) were fully affected by the policy. The policy did not require that children out of school go back to schools. Thus, those aged 7 and older in 1991 (born in 1984 and before) were at most partially affected if 71 they decided to go back to schools due to the tuition fees exemption. For those who were out of schools and did not come back to schools, they would be not affected by the policy.24 Exploiting differences in the exposure to the policy across different cohorts, I treat the more affected cohorts - those born between 1985 and 1989 as the treatment group and the less affected cohorts - those born between 1980 and 1984 as the control group. Region of birth is another dimension of variation in the exposure to the policy. Individuals born in regions with high baseline primary school completion rates would not be affected by the policy as much because most children were going to primary schools anyway. On the other hand, children born in regions with low baseline primary school completion rates would be more likely affected. The baseline primary school completion rates are calculated as those of the 1976-1979 cohorts, who are just prior to the law. The national average primary school completion rates of the cohorts is 76.65%. High baseline regions are defined as provinces with at least 87.7% (75th percentile) baseline primary school completion rates. Low baseline regions, on the other hand, are provinces with less than 67% (25th percentile) baseline primary school completion rates. The average primary school completion rates of the cohorts in high and low baseline regions are 93.74% and 56.35% respectively. Similarly, children from families of ethnic minorities would be more affected by the UPE policy than children from families of ethnic majorities. Ethnic minorities tend to be poorer and more economically disadvantaged. Moreover, ethnic minorities have lower baseline primary school completion rates than ethnic majorities. While ethnic majorities had baseline primary 24 The policy may have some spillover effects on children who do not attend primary school. For example, children who attend primary school due to the UPE law may help their siblings to learn some basic math and reading at home. 74 Panel B of Table XXVII presents the DID estimates of ethnic minorities. While the affected cohorts of ethnic minorities have a reduction of 0.71 births per woman in the post-policy period, the affected cohorts of ethnic majorities experience a decrease of 0.68 births per woman. The DID estimates are negative and statistically significant, suggesting that the policy decreased the number of living children of ethnic minorities by 0.03 birth per woman. Overall, the simple DID approach in Table XXVII suggests a positive significant effect of the policy on women’s fertility in low baseline regions, which is contradictory with my expectation. Thus, I will further investigate this issue later in the chapter. On the other hand, the simple DID estimates suggest that the policy likely affected fertility of ethnic minorities. The simple DID approach shown in Tables XXVI and XXVII can be implemented using the following regression models. To be specific, using differences in the exposure to the policy by regions, I estimate the below model. Years of Educationijt= a1 + ∑ Birth yeark k=1989 k=1981 + a2 (Birth year ≥ 1985)i + a3 Low Regioni + γ1 (Birth year ≥ 1985i ×Low Regioni) + ∑ Age ij + ηijt (1) And for differences in the exposure to the policy by ethnicities, I estimate the model Years of Educationijt= b1 + ∑ birth yeark k=1989 k=1981 + b2 (Birth year ≥ 1985)i + b3 Ethnic minorityi + γ2 (Birth year ≥ 1985i×Ethnic minorityi) + ∑ Age ij + ηijt (2) In these regressions, I additionally control for birth year dummies and age dummies. Age is a key factor that determines people’s education and fertility. When people get older, they have more education and more children. Thus, by including age dummies in the regressions, I am able to observe individuals’ educational attainment and fertility over time. In addition, by controlling for birth year dummies, I further control for systematic differences across cohorts that may affect education and fertility. 75 The coefficients of interest are 𝛾1 andγ2. The estimates capture the relative effect of the UPE policy on people’s educational attainment across regions and ethnicities. Similar models are estimated for fertility using the sample drawn from the 2009 Census. The only difference is that instead of controlling for age dummies in the regressions, I control for age and age2 since I only have one year of survey for this analysis. Because ethnic minorities tend to live in rural areas and the effect of the policy may vary across cohorts in rural areas, I also show the robustness of the estimates by further controlling for the indicator of rural areas and the interactions between birth cohorts and rural areas in the regressions. E.3. Validity The differences-in-differences estimates would capture the causal effects of the policy on educational attainment and fertility if in the absence of the policy, the increase (decrease) in education (fertility) would have been the same across regions and ethnicity groups. As noted, this is a relative effect because of the inability to measure the exposure to the policy perfectly. To assess the likely validity of the parallel trends assumption, I first estimate an event study specification in which the proxies for treatment—being an ethnic minority or living in low-baseline regions—are interacted with year effects. The event-study regression frameworks are thus as follows. Outcome ijt = c1 + ∑ θk k=1989 k=1980 YR_Bk + c2 Low Regioni + ∑ ϑk1 𝑘1=1989 k1=1980 YR_Bk × Low Regioni +∑ Age ij + vijt (3) Outcome ijt = d1 + ∑ θk k=1989 k=1980 YR_Bk + d2 Ethnic Minorityi + ∑ ϑk1 𝑘1=1989 k1=1980 YR_Bk × Ethnic Minorityi +∑ Age ij + vijt (4) in which Outcome ijt is an individual i’s years of schooling and fertility at survey j and time t. ∑ YR_B𝑘 and ∑ Age ij are vectors of birth year dummies and age dummies respectively. The omitted 76 group is those born in 1984. The coefficients of interest are on the interaction terms. I present these results below. Second, I conduct a falsification analysis. Specifically, I re-estimate the differences-in- differences models, but use unaffected birth cohorts. For these analyses, I treat individuals born between 1980 and 1984 as the treatment group and individuals born between 1975 and 1979 as the control group. Although there may have been some spillover effects for the 1980-1984 cohorts if older children went back to school because of the law, neither group should be directly affected by the UPE law. The results of the falsification analysis are presented in Panel B of Tables XXVIII and XXIX. The DID estimates of the control experiment are indistinguishable from zero and not statistically significant across regions and ethnicities. The estimates are -0.08 for low regions and 0.07 for ethnic minorities. These results suggested that time trends in educational attainment did not differ between the treated and comparison groups, which is the evidence to support the validity of my research design. Migration is also a threat to the validity of my research design. Since I do not have data on individuals’ birth of place, I assume that their current residency is their birth of place. If the policy helps people from low baseline regions obtain more education and migrate to high baseline regions, I will categorize them wrongly. To test this, I re-estimate the effect of the UPE policy on educational attainment using the survey from 2014 - the only survey that contained data on people’s birth of place. In 2014, 88.13% of the people in the survey reported to live in the same region as their birthplace. In addition, instead of defining people into high and low baseline regions using their current residency, I use their birth of place and treat current residency as an alternative dependent variable. I find that the estimates obtained are similar to the ones that I had before and 79 Table XXIX presents the effect of the UPE policy on years of education of ethnic minorities. The results are estimated using the data from VHLSSs. Panel A of Table XXIX presents the estimates of the experiment of interest and suggests that an individual who was ethnic minority and young enough received on average 0.64-0.89 more years of education29. Panel B shows the estimates of the control experiment. The estimates are 0.06 and not statistically significant. Thus, my DID estimates likely capture the causal effects of the UPE policy on education of ethnic minorities. F.2. The Effect of the UPE Policy on Fertility Figure 12 displays the coefficients of the interactions between birth year dummies and the low baseline region dummy in the fertility equation across high and low baseline regions. The DID estimates for birth cohorts 1980-1983 are negative and significantly different from zero, suggesting that in the pre-policy period, women in low regions tend to have fewer children than women in high regions. This is in contrast with my expectation. Since prior to the law, people in low regions have fewer years of education than people in high regions, I expect that they should have more children than those in high regions. The estimates for birth cohorts 1986-1989 are 0.05, significantly different from zero, and constant across cohorts. The estimates suggest a positive, significant effect of the policy on fertility of these women, which is contradictory with my expectation. The estimate for birth cohort 1985 is positive but not statistically significant, implying that there is no significant increase in the number of living children of this cohort. Similarly, the estimate for birth cohort 1989 is 0.01 and not significantly different from zero. 29 Table XXXIV, Appendix shows the estimates of the effect of the UPE policy on educational attainment of ethnic minorities, estimated using the 2009 Census, and suggest the same results. 80 Table XXX shows the effect of the UPE policy on fertility of women in low baseline regions. Panel A shows the DID estimates of the experiment of interest. The estimates are positive and significantly different from zero. The estimates are 0.104 and 0.094 with and without the inclusion of further controls. The estimates suggest that in the post-policy period, ethnic minorities have a higher number of living children than ethnic majorities. Specifically, in this case they have approximately 0.1 more births per woman than ethnic majorities. Panel B shows the DID estimates of the control experiment and indicates a differential trend in the number of living of children between both regions prior to the law. The DID estimates are 0.11 and statistically significant. The estimates indicate that in the pre-policy period, people in low regions tend to have more children than people in high regions. Together, these results suggest that the identification assumption of the DID approach is not likely to hold in this context. Thus, it is hard to make any conclusions about the effect of the policy on fertility of women in low regions. Figure 13 shows the coefficients of interactions between birth year dummies and the ethnic minority dummy in the fertility equation across ethnicity groups. The analysis is performed using the data from Census 2009. The DID estimates of the 1980-1983 cohorts are negative and statistically significant for birth cohorts 1980 and 1982. The estimates range from -0.01 to -0.04, suggesting that in the pre-policy period, ethnic minorities tend to have more children than ethnic majorities. However, these estimates are very small—four and two percent of the mean of the baseline of the control group. Thus, while technically this result indicates that there may be systematic difference in the number of living children between both ethnic groups in the pre-policy period, the practical importance is very small. The estimates of the affected cohorts (1985-1989) are also negative and statistically significant, suggesting that younger cohorts have fewer children than older cohorts. Estimates are also sizeable, for example 20% of the mean. However, one 81 inconsistent result is that there is no big reduction in the number of living children of birth cohort 1985, compared to that of birth cohort 1984, which is in contrast with my expectation. Table XXXI presents the effect of the UPE policy on the number of living children of ethnic minorities. Panel A shows the estimates of the experiment of interest. The DID estimates are negative and statistically significant, suggesting that the policy decreased the number of living children of the affected ethnic minorities by 0.045 births per woman, which is a relatively small effect. When the indicator of rural areas and the interactions between birth cohort dummies and rural areas are further included in the regression, the estimate is still statistically significant. However, it gets smaller, decreasing to -0.013 births per woman. Panel B shows the estimates of the control experiment. The DID estimates of interest are positive and statistically significant. The estimates suggest that prior to the policy ethnic minorities had 0.06-0.07 more births than did ethnic majorities. This result implies that the identification assumption of the DID approach does not hold in this context. This is contradictory to the findings from Figure 13, which suggest that although there is a differential trend in fertility between both ethnic groups in the pre-policy period, the difference is very small and thus the DID approach is fine. Overall, the results suggest that there is no consistent evidence that the UPE policy had an effect on women’s fertility. The DID estimates of low baseline regions are positive and statistically significant, suggesting that in the post-policy period, people in low regions have more children than people in high regions. The DID estimates of ethnic minorities, on the other hand, are negative and statistically significant, indicating a potential effect of the policy on fertility of ethnic minorities. However, the magnitude of these estimates (0.013-0.045) is quite small. For instance,