A nonexperimental evaluation of training programs for the unemployed in Sweden

This study uses a unique administrative database to evaluate employment training program effects on three samples of Swedish adult males. Alternative nonexperimental models suggest both positive and negative relationships between training and annual earnings. Model specification tests decisively reject the fixed-effect model, but not the random-growth model, which shows that specification tests can reject misspecified models. The chosen model predicts no effects or significantly negative earnings effects of the training program. One explanation for this result is that some participants may have enrolled in the program not for the purpose of making a human capital investment, but for the purpose of collecting unemployment insurance benefits. (C) 2002 Elsevier Science B.V. All rights reserved.