Varying Biases in the Matching Estimates: Evidence from Two Randomized Job Search Training Experiments

We compare various matching estimators to the results of two randomized field experiments that evaluate employment effects of job search training programs. We find that commonly used non-experimental matching estimators tend to over-estimate the program effects and that the bias differs between programs owing to different procedures used in selecting the program participants. The bias is larger when participation is voluntary than when caseworkers assign the applicants to training. JEL codes: C93