The Effects of Training Incidence and Duration on Labor Market Transitions

This paper develops a dynamic evaluation approach in discrete time to estimate the impact of training programs for the unemployed on employment transitions. Our framework accounts for the endogeneity of program incidence and duration and considers confounding caused by lagged outcomes and treatments, time-constant unobservables, as well as time-varying observed covariates. We specify a flexible bivariate random effects probit model for employment and training status that we estimate with Bayesian Markov Chain Monte Carlo (MCMC) techniques. Based on our estimates, we simulate different treatment effects of interest. Our estimation results imply positive effects of training on the employment probability of the treated, lying between 6 and 14 percentage points 2.5 years after program start. The effects are higher for women than for men and initially negative effects persist shorter in West Germany than in East Germany.