Proper interpretation of our estimates requires consideration of several features specific to our approach. First, not all children will begin school in the year predicted by school entry policies. The parents of a child born before the school entry date may hold their child back by a year, and the parents of a child born after the school entry date may petition for their child to start school a year before typically allowed, or may start their child in private school.3 For neither type of child will schooling progression be affected by school entry policies. This suggests that our estimates may disproportionately reflect the experience of women from low socio-economic backgrounds, whose parents are somewhat more likely to comply with school entry policies (Elder and Lubotsky 2006).
Second, even if school entry policies affect a woman’s schooling progression, they may not affect education at motherhood. For example, for a woman intent on obtaining a specific level of schooling before dropping out and beginning a family, education at motherhood is unaffected by the timing of school entry. School entry policies affect education at motherhood for two types of women: those still enrolled in school and those who have already completed schooling, whose school-leaving decision was age-dependent (i.e., not just schooling-dependent). For example, a woman who drops out of school at the earliest age allowed under a typical compulsory schooling law will have fewer years of education if she starts school late (Angrist and Krueger 1992).
This suggests that our estimates may be most relevant for women at risk of dropping out of school. Such women are likely to give birth at earlier ages than women intent on attaining a specific level of schooling, such as a college degree. Empirically, we find that school entry policies exert the greatest impact on the education of women giving birth at young ages. Thus, we stratify most of our analysis by age, focusing on women age 23 or younger, for whom our first stage relationship is strongest. Third, if the dropout decision is affected by both current age and schooling level, and if school completion is a binding constraint on fertility timing, then we should expect to see effects of school entry policies on two factors that potentially affect infant health—education at motherhood and age at motherhood. life without allergy
In general, this would lead to a failure of the order condition for instrumental variables (fewer instruments than endogenous regressors), preventing consistent estimation of the effect of female education on infant health without additional instruments. However, surprisingly, we document that school entry policies affect neither the probability of becoming a mother nor age at motherhood. This suggests that for women whose dropout decisions depend on schooling level, schooling completion is not a binding constraint on fertility timing. Importantly, this also implies that our approach identifies an education effect unconfounded by selection into motherhood and unconfounded by age at motherhood.