Panel Paper: Using the Timing of Pregnancy and Local Job Losses to Identify the Causal Effect of Marriage on Children’s School Achievement

Thursday, November 2, 2017
Stetson G (Hyatt Regency Chicago)

*Names in bold indicate Presenter

Elizabeth Ananat, Anna Gassman-Pines and Christina Gibson-Davis, Duke University

Policymakers fear that the growing gap in marriage between higher- and lower-income parents may exacerbate inequality by increasing income disparities in children’s academic achievement. Whether it does, however, depends on whether marriage actually causes improved child outcomes or whether marriage is only spuriously related to child well-being. We revisit this longstanding question by identifying quasi-experimental variation in whether parents who conceive non-maritally marry mid-pregnancy, as a function of whether bad economic news arrives during the pregnancy or not until after the birth. In doing so, we will answer the question: does being born to married parents lead to improved academic achievement among children?

This paper will establish that the timing of local job losses is a valid instrument for parental marriage with a non-marital conception. Assuming that the necessary conditions are satisfied, we will use the instrument to conduct a causal analysis of the effect of marriage on children’s educational outcomes.

We use data from North Carolina (NC) from four administrative data sources, covering the years 1990 to 2012. (1) Data on births are from the Detailed Birth Records, from the NC Department of Vital Statistics (NCDVS), which are compiled from questionnaires obtained at the time of birth certificate filing. (2) Data on marriages are from the NCDVS and contain the names of those marrying, the marriage date, and the county of marriage. (3) The Job Loss Dataset from the NC Employment Security Commission provides information on businesses that shut down or laid off workers, including the number of workers affected and the date of the job loss. (4) Information on public school students’ academic achievement test scores are from the NC Education Research Data Center.

We have combined the administrative data, allowing us to identify within each county the month job losses occurred, the month a birth was conceived, and the month a marriage took place. This information on the timing of job losses, pregnancies, and marriages will be used to conduct an instrumental variable analysis, which enables causal estimation in contexts in which a controlled experiment is not possible. Such an approach isolates a variable that only affects an outcome through its effect on the predictor of interest—in this case, job loss on marriage between biological parents. First, we show that job losses during pregnancy affect the probability that women with nonmarital conceptions marry before the baby is born (b = .85; SE = .26; p < .01), but that job losses after pregnancy do not (b = .20; SE = .22). Second, we compare those effects to effects on women with marital conceptions, and find that job losses do not affect them (b = .002; SE = .006). Finally, we will conduct a two-stage analysis: using the interaction of pregnancy timing and job loss timing to predict marriage at the time of birth, we will examine the effect of parental marriage on children’s schooling outcomes.