Through the federal Workforce Investment Act (WIA) and state workforce programs, the public workforce system offers the principal means by which the unemployed can obtain the occupational skills training they need to advance economically. While a number of researchers have evaluated workforce training, the recent integration of longitudinal data across workforce programs has created the opportunity to conduct quasi-experimental evaluations of training that use evaluation designs and variables that have been used only rarely in the past. This paper uses linked administrative data from New Jersey to evaluate the effect of occupational skills training on the earnings of participants in WIA- and New Jersey Workforce Development Partnership Program-funded occupational skills training.

Recent evaluations of training programs primarily use individuals receiving only Employment Services (ES) under Wagner-Peyser as a comparison group. While these individuals are clearly similar to trainees in their job status and willingness to use government services to find employment, using ES participants as a comparison also has some shortcomings. In particular, ES participants may lack several characteristics that training participants possess, such as soft skills and professional ambition, which could artificially inflate estimates of training program impacts. Therefore, in addition to comparing training completers to ES participants, we also  compare Unemployment Insurance (UI) claimants who were eventually referred to, enrolled in, and completed occupational skills training to a comparison group of UI claimants who were referred to and enrolled in occupational skills training but did not complete it. This design allows us to match not only based on demographics and pre-training earnings, but also on UI claim date and the time between filing a claim and the start date of training.

To mitigate the effect of unobservable, individual-level factors on the estimate, we use a difference-in-differences design to compare the earnings growth of training completers to the earnings growth of non-completers. We select a comparison group of individuals using two matching methods. First is propensity score matching (PSM), by far the most common method used in evaluating training programs. Recent studies have demonstrated, however, that PSM can lead to biased estimates that misrepresent the true impact of training programs (King et. al. 2011, 2012). Thus, we will also estimate all models using coarsened exact matching (CEM), an alternative method not subject to the potential biases of PSM, and evaluate the robustness of our results across all models.

The paper finds that occupational skills training has a positive effect on the earnings of trainees, but that the effect is greater with the Employment Services comparison group than with the non-completers comparison group. The findings are suggestive that workforce counselors may engage in “creaming” when deciding whom to refer to training and that the completer versus non-completer comparison group may be a preferable design for estimating the treatment effect of training.