Dynamic Treatment Effects of On-the-Job Training

There has been an extensive focus in examining the role of job training for delivering successful labor market outcomes. In this paper, we estimate the dynamic returns to job training for a large scale program, taking... [ view full abstract ]

There has been an extensive focus in examining the role of job training for delivering successful labor market outcomes. In this paper, we estimate the dynamic returns to job training for a large scale program, taking advantage of administrative data on job training records,matched employee-employer data on wage and employment outcomes and pre-labor market ability measures. We first evaluate the approach followed by the literature, which estimates average treatment effects using an individual fixed-effects strategy. We carefully document the assumptions required for identification in this strategy, test and reject these assumptions. We instead present a dynamic discrete choice model that captures a variety of static and dynamic treatment effects, which allows for selection into training depending on observed characteristics as well as unobserved heterogeneity, which we interpret as initial stock of skills. We find that higher ability individuals self-select into training at each decision node. Furthermore, we exploit our dynamic model and estimate a number of different treatment effect parameters for a combination of counterfactual training histories. We find that the dynamic average treatment effect and the dynamic treated on the treated effect deliver different results than those estimated in a static setting.