Model Specifications for Estimating Labor Market Returns to Associate Degrees: How Robust Are Fixed Effects Estimates? A CAPSEE Working Paper

Recently, studies have adopted fixed effects modeling to identify the returns to college. This method has the advantage over ordinary least squares estimates in that unobservable, individual-level characteristics that may bias the estimated returns are differenced out. But the method requires extensive longitudinal data and involves complex specifications, raising the possibility that results are sensitive either to sample restrictions or to alternative specifications. Also, the extra requirements might not be justified if results from fixed effects models are broadly similar to those from conventional ordinary least squares models. In this paper we review results from fixed effects models of the earnings gains from completing an associate degree relative to non-completion for community college students. We examine both sampling restrictions and specification issues. We find results to be sensitive to assumptions about missing earnings data and to how time trend specifications are modeled. However, we find no substantively meaningful differences between estimates using fixed effects models and ordinary least squares methods. A main benefit of fixed effects models--controlling for unobservable student characteristics--should be weighed against the difficulty in interpreting coefficients and more intensive data requirements. On the other hand, a distinct advantage of fixed effects models is that they allow for analysis of earning profiles over the period from before to after college. Given the large fluctuations in earnings over this period, this advantage may be significant in yielding evidence on the full returns to college. The following two tables are appended: (1) Robustness Checks on Returns to Associate Degrees; and (2) Percentage Decline in Earnings From Two Quarters Prior to College Entry to Quarter of College Entry (Kentucky, Ohio).