Family Size, Sibling Rivalry and Migration Evidence from Mexico Mariapia Mendola (U Milan-Bicocca) joint with Massimiliano Bratti (U Milan) Simona Fiore (U Venice) Summer School in Development Economics Prato, June, 2017 1 / 34
Objective Investigate the within-family dimension of international migration choices in Mexico Isolate the causal effect of sibship size, birth order and sibling composition (by age and gender) on young adults migration outcomes Shed light on (i) the inter-household and (ii) intra-household allocation of migrants: Do migrants come from larger families? Is there any systematic role for the household structure? I.e. who migrates among sibilings within the family? 2 / 34
Motivation Migration ia a key human capital investment whose returns decline with age (Sjaastad, 1962; Dustmann and Glitz, 2012) Workers moving from a poor to a rich country can experience significant increases in earnings and welfare (Ashenfelter, 2012) It is typically a household-level decision mostly involving (only a few) young members International migrants aged 15-35 (15-24) account for over 30% (13%) of total migrants (UNDESA, 2001) In developing contexts this is a household resource allocation problem, which entails a number of trade-offs (Rosenzweig, 1988; Ghatak and Price, 1996) Little evidence on the impact of the childhood family (i.e. siblings) on offspring international migration 3 / 34
This paper We add to the literature on the determinants of migration by documenting the causal impact of the household enviroment on migration choices This is studied for the case of Mexico-US migration in the 1990s (when both migration and fertility rates were substantial) by using a large national demographic survey (ENADID). Identification of family size comes from exploiting exogenous variation in the number of born children per mother that occurs due to miscarriages and infertility shocks. We exploit differences across siblings to identify the impact of birth order and sibling composition on child s migration outcome We find no evidence that family size drives migration, while we show that the chances to migration are not equally distributed across siblings within the household. 4 / 34
Contribution By addressing the causal effect of parental fertility on migration outcomes, we show that family size per se is not push factor for migration first order effects of population size We shed light on (intended or unintended) consequences on migration of family-planning programs in developing countries Our findings hint to the fact that migration may matter for fertility decisions i.e. migration is endogenous. This is so as children provide support to the family in developing countries, even through migration. By showing that not all children in the household have the same chance to migrate, our results point to the existence of an intra-household selection process which may have significant implications in terms of migration impact. 5 / 34
Outline Background literature The context of Mexican migration Data Empirical strategy and identification Results Discussion and Conclusions 6 / 34
Background 1 Migration as an investment strategy for households in developing countries: High returns for the migrant in terms of higher productivity, income (Gibson and McKenzie 2012) Returns for the origin household, in terms of remittances, access to credit, risk-management, social security (Chen et al., 2003; Stark 1991) Extensive evidence on the support of migrants to their households of origin (e.g. Rapoport and Docquier, 2006; Ratha et a. 2013; Antman, 2012) Little evidence, though, on the household-level determinants of this decision problem 7 / 34
Background 2 The link btw household structure and investment in children s human capital has been influenced by the Q-Q trade-off theory (Becker 1960, Becker and Lewis 1973, Becker and Tomes, 1976) Resource dilution effects lead to a fall in child migration In developing countries though, parents use their children as a substitute for missing institutions/mkts (children as investment goods )- e.g. old-age social security (Cigno 1993, Ray 1998, Duflo and Banerjee, 2011). Resource-augmentation effect (from children to parents) such that more siblings increase the likelihood to migrate Diversification effect 8 / 34
Background 3 A related question is how family structure affects migration outcomes among children within the same family If migration is placed in a household resource allocation framework, birth order and sibling composition may influence the marginal utility from child migration In the human capital literature, birth order has been found to have a significant influence on health or education (Jayachandran and Pande 2017, Pande 2003) By affecting household resources, the composition of siblings (by age and gender) may influence child outcomes unevenly (sibling rivarly effect) (Morduch 1998) 9 / 34
The context Mexico One of the largest migrant-sending remittance-recipient countries in the world The approximate number of Mexicans moving to the U.S. annually with or without documents has increased steadily since the 1970s Between 1970 and 1980, according to U.S. Census data, Mexican immigrants went from the fourth largest to the largest immigrant group in the U.S. According to Mexican records, more than 5% of Mexican population moved to the U.S. in 1990 and more than 10% in 2005 A high-fertility country where the number of children per woman gradually declined from an average of 6 in the 70s to about 2.5 children in the 2000 (Cabrera, 1994). 10 / 34
Mexican migration Mexican migration patterns differ by age and gender, with a significant fraction of Mexican males migrating in the age between 15 and 30 Source: Hanson and McIntosh, 2010 11 / 34
Data Encuesta Nacional de la Dinamica Demografica (ENADID), pooled cross-section 1992, 1997 Big samples (50,000 hhs each wave), nationally representative, detailed data on both fertility and migration histories Fertility on all women aged 15-54 Migration of all household members, even absent (undercount of permanent migration) In the 1997 (1992) ENADID, 18% (15%) of households report having a member migrated in the US. We define individual migration episodes as non-tied migration, i.e. we exclude from the sample individuals migrated joint with their parents or other household members. 12 / 34
Figure: Mexican individual migration by age and gender 0.02.04.06.08.1.12 5 20 35 50 Age at migration Male Female Source: Our computations on ENADID, 1992 and 1997. 13 / 34
Sample selection We focus on children aged 15 to 25 The number of siblings is tallied as the number of each child s biological brothers and sisters currently alive In line with the medical definition of infertility, we restrict our sample to children of nonsterilized women who are not currently using contraceptives or who never did (about 80% of the sample) (Agüero and Marks 2011). Our final estimation sample is made of 26,743 Mexican youngsters in the age range 15-25 Average mothers age (45) and average birth spacing btw first and last child (13) ensure we measure completed fertility at the moment of child s migration. 14 / 34
Figure: Individual migration rate by family size 0.05.1.15 2 3 4 5 6 7 8 9 10 11 Family size Male Female Source: Our computations on ENADID, 1992 and 1997. 15 / 34
Descriptive stats Table: Sample characteristics by migration status Non-migrants Migrants p values (A) (B) (A)-(B) Individual-level characteristics Age 18.878 20.982 0.000 Female 0.458 0.250 0.000 N. of siblings 5.071 5.869 0.000 Birth order 1 0.181 0.192 0.300 Birth order 2 0.231 0.225 0.555 Birth order 3 0.178 0.178 0.978 Birth order 4 0.137 0.154 0.077 Birth order 5 0.102 0.102 0.993 Birth order 6 0.071 0.073 0.781 Birth order 7 0.046 0.041 0.343 Birth order 8 0.028 0.021 0.100 Birth order 9 0.014 0.009 0.121 Birth order 10+ 0.011 0.006 0.107 Household-level characteristics Mother s age 44.612 46.171 0.000 Mother s age at first pregnancy 20.030 19.699 0.182 Mother s years of schooling 4.091 3.452 0.010 Mother chronic illness 0.023 0.008 0.131 Single mother 0.185 0.188 0.896 Mother s labor income 600 862 0.037 Father s age 48.799 52.207 0.000 Father s years of schooling 4.931 3.789 0.059 Father s labor income 2,112 3,135 0.000 16 / 34
Empirical Strategy 1 Two steps procedure in order to control for both sibship size and birth order effects the two variables have little independent variation (Black et al. 2005, Bagger et al. 2013) I step: estimate birth order effects by exploiting within family variation (family FE estimator) M ij = α 0 + K α 1k bo ijk + α 2 X ij + u j + ɛ ij (1) k=1 X ij is a vector of individual level characteristics (gender, age, age square, birth cohort fe) II step: estimate family size effect exploiting cross-family variation (OLS and IV estimator) dependent variable is the netted migration (NM), where ˆ NM ij = M ij K ˆα 1k bo ijk k=1 17 / 34
Empirical Strategy 2 II step: sibship size effect ˆ NM ij = β 0 + β 1 S ij + β 2 X ij + β 3 W j + v ij (3) W ij is a vector of household level characteristics (mother s and father s age, education, maternal health, single mother), plus municipality FE we correct standard errors by weighting the estimation with the inverse of the standard error of ˆ NM ij Standard errors are clustered at the household level 18 / 34
Identification Strategy The estimation of (3) with OLS may deliver spurious results because of unobservables generating a bias For example, monetary or non-monetary factors that increase fertility choices may also affect the decision to have a child migrating to the US. IVs: we instrument sibship-size with biological fertility (subfecundity) and infertility shocks (miscarriage at first pregnancy) Biological infertility: indicator variable for whether women declare they never took contraception or stopped doing that because of infertility (Aguero and Marks, 2011) Miscarriages: women report miscarriage episode on their first pregnancy (Miller, 2011) 19 / 34
Relevance of IVs- infertility shock Figure: Average number of children by women s subfecundity status No. of live births 1 2 3 4 5 6 30 35 40 45 50 55 Women's age Infertile Fertile 20 / 34
Relevance of IVs- miscarriage at first birth Figure: Average number of children by women s miscarriage at first pregnancy No. of live births 2 3 4 5 6 30 35 40 45 50 55 Women's age Miscarriage No Miscarriage 21 / 34
Exogeneity of IVs The epidemiological literature has been finding few risk factors driving secondary infertility (Buck et al. 1997) In the socio-economic literature subfecundity has been assessed as virtually random (uncorrelated with female background characteristics) in both advanced and developing countries (Wilcox and Mosher, 1993; Aguero and Marks 2011; Markussen and Strom, 2015) Yet, in our estimates we do control for a number of key and predetermined characteristics, in particular mother s education (which is correlated with the cumulative level of nutrition and human capital) and father s educaiton (which is a proxy for household well-being). As RC we run regressions while controlling for municipality by year FEs and municipality by father s education (high/low) FEs 22 / 34
Results Birth order effects Variables (1) (2) female -0.036*** -0.035*** (0.003) (0.003) birth order -0.019*** (0.003) birth order 2-0.002 (0.005) birth order 3-0.021*** (0.007) birth order 4-0.038*** (0.010) birth order 5-0.068*** (0.013) birth order 6-0.086*** (0.016) birth order 7-0.112*** (0.019) birth order 8-0.136*** (0.022) birth order 9-0.161*** (0.026) birth order 10+ -0.199*** (0.030) age 0.020** 0.021** (0.009) (0.009) age squared 0.000 0.000 (0.000) (0.000) Year of birth indicators YES YES Family fixed effects YES YES Observations 26,743 26,743 23 / 34
Sibship size effect: WLS estimates Variables (1) (2) (3) (4) (5) (6) (7) N. siblings 0.011*** 0.011*** 0.014*** 0.010*** 0.013*** 0.010*** 0.013*** (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) N. siblings female -0.008*** -0.007*** -0.006*** (0.001) (0.001) (0.001) female -0.038*** -0.036*** -0.033*** -0.031*** -0.033*** -0.031*** (0.003) (0.006) (0.003) (0.006) (0.003) (0.003) Individual s controls YES YES YES YES YES YES Mother s controls YES YES YES YES Father s controls YES YES YES YES Municipality indicators YES YES Observations 26,743 26,743 26,743 26,743 26,743 26,743 26,743 24 / 34
Sibliship size effect: 2SLS estimates Variables (1) (2) (3) Second stage N. siblings 0.004-0.018-0.005 (0.014) (0.023) (0.012) female -0.033*** -0.033*** -0.033*** (0.003) (0.003) (0.003) IV: infertility miscarriage overidentified Anderson-Rubin F statistic 0.0734 0.686 0.389 [0.787] [0.407] [0.678] Hansen J statistic 0.737 [0.391] First stage N. siblings infertility -0.494*** -0.491*** (0.095) (0.095) miscarriage -0.437*** -0.433*** (0.10) (0.10) Angrist-Pischke F statistic instrument(s) 26.90 19.13 23.37 Individual s controls YES YES YES Mother s controls YES YES YES Father s controls YES YES YES Municipality indicators YES YES YES Observations 26,743 26,743 26,743 25 / 34
Child gender and sibliship size effect: 2SLS estimates Variables (1) (2) (3) Second stage N. siblings 0.005-0.065-0.007 (0.016) (0.048) (0.015) N. siblings female -0.005 0.112 0.005 (0.013) (0.079) (0.013) female -0.032*** -0.064*** -0.034*** (0.004) (0.022) (0.005) IV: infertility miscarriage overidentified Anderson-Rubin F statistic 0.074 2.210 1.150 [0.928] [0.110] [0.331] Hansen J statistic 4.399 [0.111] First stage N. siblings infertility -0.567*** -0.564*** (0.109) (0.108) infertility female 0.168 0.169 (0.115) (0.115) miscarriage -0.453*** -0.450*** (0.117) (0.117) miscarriage female 0.037 0.038 (0.106) (0.105) Angrist-Pischke F statistic instrument(s) 28.62 11.98 15.68 First stage N. siblings female infertility 0.125*** 0.125*** (0.038) (0.038) infertility female -0.694*** -0.691*** (0.131) (0.131) miscarriage -0.067-0.068 (0.044) (0.043) miscarriage female -0.261** -0.254* (0.131) (0.130) Angrist-Pischke F statistic instrument(s) 26.93 4.27 13.83 Observations 26,743 26,743 26,743 26 / 34
Robustness check: Household level analysis It allows us to test the robustness to both estimation strategy and estimation sample Household-level regressions disregard birth order effects and allow us to include one-child households in the estimation sample We estimate a specification as follows: m j = γ 0 + γ 1 n j + γ 2 W j + v j (4) where m j is the number of migrants and n j, is total fertility in household j. We estimate both OLS and 2SLS using the same instruments as in the individual-level models 27 / 34
HH-level results (1) (2) (3) (4) Variables OLS 2SLS 2SLS 2SLS Second stage N. children 0.012*** -0.001-0.014-0.005 (0.001) (0.013) (0.022) (0.011) IV: infertility miscarriage overidentified Anderson-Rubin F statistic 0.006 0.400 0.203 [0.940] [0.527] [0.816] Hansen J statistic 0.265 [0.607] First stage N. children infertility -0.643*** -0.640*** (0.063) (0.063) miscarriage -0.442*** -0.438*** (0.070) (0.070) Angrist-Pischke F statistic instrument(s) 103.87 39.81 72.40 Mother s controls YES YES YES YES Father s controls YES YES YES YES Municipality indicators YES YES YES YES Observations 17,544 17,544 17,544 17,544 28 / 34
Discussion So far we found no causal effect of fertility (family size) on migration outcomes, while controlling for birth order The latter result is robust to the use of different IVs, samples and empirical models. All else equal though, the invididual likelihood of migration is sistematically higher for boys and decreases with birth order The latter results is consistent with an optimal migraiton model where family s migration returns (i) are higher for boys and (ii) decrease with child parity It may be also explained by a pure preference for sons or for eldest children Hence we explore the interplay between child gender, b.o. and sibling composition in order to better pin down the role of household environment on migration outcomes. 29 / 34
female -0.032*** -0.031*** (0.006) (0.007) birth order -0.019*** (0.003) birth order x female -0.001 (0.001) birth order 2 0.002 (0.006) birth order 3-0.023*** (0.008) birth order 4-0.034*** (0.011) birth order 5-0.070*** (0.014) birth order 6-0.077*** (0.017) birth order 7-0.103*** (0.020) birth order 8-0.140*** (0.023) birth order 9-0.166*** (0.028) birth order 10+ -0.188*** (0.033) birth order 2, female -0.011 (0.009) birth order 3, female 0.005 (0.010) birth order 4, female -0.010 (0.010) birth order 5, female 0.006 (0.011) birth order 6, female -0.018 (0.012) birth order 7, female -0.017 (0.015) birth order 8, female 0.010 (0.018) birth order 9, female 0.012 (0.024) birth order 10+, female -0.022 (0.027) age 0.020** 0.021** (0.009) (0.009) age squared 0.000 0.000 (0.000) (0.000) Year of birth indicators YES YES Family fixed effects YES YES Observations 26,743 26,743 R-squared 0.050 0.053 Migration, gender and birth order Table 9: Birth order effects Variables (1) (2)
Discussion Low parity children are in general more likely to migrate but a first-born daughter is significantly less likely to migrate than a first-born son (by 3 p.p. which translates in a probability lower than the average by 60 percent) This is consistent with Mexican migration being a male dominated phenomenon due to higher labor market returns and lower opportunity costs (both in Mexico and the U.S.) for Mexican boys than for girls In order to better separate whether these results are due to a pro-eldest-son bias or to the intrahousehold allocation of resources, we next estimate the direct effect of sibiling composition, while controlling for family size, gender and birth order. 31 / 34
Migration as a function of sibling composition Variables (1) (2) (3) (4) N. older brothers -0.014*** -0.014*** -0.017*** -0.016*** (0.004) (0.004) (0.004) (0.005) female -0.028*** -0.026*** -0.022*** -0.016*** (0.003) (0.005) (0.005) (0.006) N. older brothers female -0.002-0.002-0.003 (0.002) (0.002) (0.002) Next brother -0.005 0.001 (0.004) (0.004) Next brother female -0.012** (0.006) Age, age squared YES YES YES YES Birth order fixed effects YES YES YES YES Year of birth indicators YES YES YES YES Family fixed effects YES YES YES YES 32 / 34
Discussion We find that sibling composition (by age and gender) matters for child s migration outcomes, due to intrahousehold allocation of resources. In particular, while having an older brother instead of an older sister decreases the likelhood to migrate for boys and girls alike, having a next-born brother instead of nex-born sister significanlty reduces the chances to migrate for girls only. Indeed, a low-parity Mexican boy in the 90s may be more valuable to send as a household migrant abroad than a girl. Plus, the opportunity cost of sending girls abroad is higher because of household productions duties and social norms. 33 / 34
Conclusions By using detailed individual-level data we provide novel evidence on the impact of siblings on migration of Mexican youngsters to the US. We find no causal effect of fertility- opposite to popular view but consistent with the literature on the determinants of human capital investment. On the other hand, the chances to migrate are not equally distributed across children within the same family: Older siblings, especially firstborn sons, are more likely to migrate, while having more brothers than sisters systematically decreases the chance of migration, especially among girls. Results are stable with respect to several robustness checks and are consistent with an optimal household migration strategy where private costs and returns of migration are shared among all siblings. 34 / 34