We also find that exiting the military in a period of high unemployment affects veteran’s marital status and living arrangements. Our estimations indicate that 1 percentage-point increase in the unemployment rate in the year of exit from active service leads to a 1.8 percentage-points reduction in the probability of ever being married, to a 1.3 percentage-points increase in the probability of divorce, and to a 2 percentage-point reduction in the probability of living in a couple household. In these models, the effect sizes are quite similar in the OLS and 2SLS models, suggesting that the decision to marry may be less influenced by the unemployment rate than the decision to attend college. We note that previous investigation on the effect of entering the labor market during weak economic conditions have neglected to study its effect on marriage and living arrangements. However, our findings are supported by findings in the recent study by Autor, Dorn and Hanson (2017) that links the decline in the prevalence of marriage among young adults to reduced labor-market opportunities faced by males, which make them less valuable as marital partners. It is interesting to note that veterans in our sample who enter the labor market in poor conditions are more likely than others to obtain post-secondary education; this should increase their potential earnings (and thus, their value as partners); nonetheless we find that they are still somewhat less likely to marry. It is possible that some individuals put off marriage in the face of initially negative labor market conditions; if those same individuals obtain additional education, doing so may further delay marriage and this temporal effect could explain at least part of the decreased rates of marriage.
To summarize, our analyses indicate that initial labor market conditions do impact other outcomes. Veterans who enter the civilian labor market during periods of relatively high unemployment achieve more post-secondary education than others (perhaps due to lower opportunity costs), but are less likely to marry and less likely to live in a coupled household.
In the next subsections, we investigate the effect of reentering the civilian life in years of high unemployment on earnings, wealth and retirement decisions. In these analyses, we do not control for education attainment, marital status or coupled status because (as just shown) they are also affected by the notional experiment at hand. Moreover, they are important determinants of earnings, wealth and retirement decisions. Thus, controlling for education, marital status and coupled status would bias our estimates of the average total effect of reentering the civilian life in periods of high unemployment. In other words, in this case education attainment, marital status and couple status are cases of “bad controls” (Angrist and Pischke, 2008).11Our methodology somewhat resembles that of Neal and Johnson (1996) who argue that controlling for educational attainment introduces endogeneity into wage regressions that are designed to measure discrimination. Instead, they opt to include only AFQT scores from the teen years and exclude educational attainment and other measures of skills; they interpret their results as a measure of the large role of premarket discrimination in labor market outcomes. The combination of the unemployment rate at the point veterans enter the civilian labor market and their educational attainment at that point could be viewed as a measure of their initial allotment of skills and luck. However, we do use the 2SLS methodology (discussed above) for correcting for endogeneity of veteran status.

\label{section}