Estimating the Effect of Job-Training Programs, Using Longitudinal Data: Ashenfelter's Findings Reconsidered

This paper is an examination of how autoregressive earnings models commonly used to evaluate job-training programs can produce badly biased estimates of both the magnitude and the temporal pattern of program impacts. Ashenfelter's results are used to illustrate this point, and a new, more appropriate model is used to reanalyze his data. Of particular importance is the finding that the decay in Ashenfelter's estimated training effect for men was produced by a time-varying bias in his model.