New Instrumental-Variables Estimates of the Effects of Schooling and Military Service: Empirical Strategies Using Non-Public-Use Data

National Bureau Of Economic Research Inc, Cambridge MA

Investigators

Abstract

This proposal outlines four projects that rely on non-public use data from the 1990 and 2000 censuses. The first three projects involve record linkage with other data sources. All work is to be carried in the Census Bureau's Boston and Michigan Research Data Centers (RDCs). The purpose of these projects is to use instrumental variables (IV) methods to estimate four sets of causal effects: (1) The effect of schooling on labor force participation and disability status; (2) The effect of maternal education on fertility; (3) The effect of schooling on wage inequality; and (4) Long-term effects of Vietnam-era veteran status on a variety of outcomes, including earnings, labor force participation, and disability status. The veteran status project uses draft lottery numbers; the schooling projects use quarter-of-birth instruments and college proximity instruments. Quarter-of-birth instruments are available in recent Census files because of the presence of exact date-of-birth on the non-public use file for the 2000 Census and year of birth on internal files for the 1990 Census. Birthday information can therefore be used to construct indicators of quarter of birth and draft lottery eligibility status in 2000, and a first-quarter dummy in 1990. The project also plans to link the 1990 and 2000 Census files, thereby generating more detailed birthday information for the 1990 respondents who are observed in 2000 and allowing the investigators to create college-proximity instruments for the 2000 respondents who are observed in 1990. Finally, part of this agenda involves a strategy for birthday-based imputation of a 1980-style highest grade completed (HGC) variable for the 1990 and 2000 Censuses since the public use files include only categorical degree receipt. The imputation exploits an existing non-public-use link of the 1991 CPS to birthday information from the Social Security Administration. The 1991 CPS is especially useful because it can be used to produce a file (by linking rotation groups) that contains both HGC and categorical schooling variables. The causal effect of schooling on average wages is one of the most important questions in labor economics. At the same time, interest is growing in other effects of schooling, including effects on labor force participation, health, and features of the wage distribution beyond average earnings. For example, a number of researchers have tried to establish a causal link between education and longevity. This project uses quasi-experimental variation to look at the effects of education on the labor force participation and disability status of older Americans in the 2000 Census, which includes 6 disability measures, 5 of which are new. A second application focuses on the effects of education on fertility. A provocative recent finding is that there is no relation between quarter-of-birth and fertility timing, suggesting that schooling has no effect on fertility either. However, these estimates rely on non-random samples and data from only two states. This project augments this investigation using random samples for all states from the 2000 and earlier Censuses. A third application explores sharp changes in the link between education and income inequality since 1990. Descriptive analysis shows that higher education is associated with much more inequality in 2000 than 1990. The investigators plan to use semi-parametric and IV methods to determine whether this link is causal. Finally, the conflicts in Afghanistan and Iraq have renewed interest in the social and economic consequences of military service. The investigators plan to use draft lottery numbers to provide credible estimates of the labor force participation and disability effects of Vietnam-era service on men now in their fifties. Broader Impacts: This project will provide improved and intertemporally comparable schooling variables that can be used when working with 1980, 1990, and 2000 Census microdata. It will add Year of Birth to the internal RDC version of the 1990 Census. It will develop a linked version of the 1990 and 2000 Census microdata files. This file should prove useful to any researcher interested in longitudinal research with large samples, or who would like to combine variables across these important data sets.