The Case for Evaluating Training Programs with Randomized Trials

Abstract

This brief paper presents the reasons that I have come to conclude that the evaluation of the economic benefits of training programs will be greatly enhanced by the use of classical experimental methods. In particular, I am convinced that some of these training programs should be operated so that control and experimental groups are selected by ran- dom assignment (randomized trials). It follows that a simple comparison of earnings, employment, and other outcomes as between control and experimental groups subsequent to participation in the experimental program will provide a simple and credible estimate of program success (or failure). The principal reason why randomized trials should be used in this field is that too much of the non-experimental estimation of the effects of training programs seems dependent on elements of model specification that cannot be subjected to powerful statistical tests. Moreover, these specification tests are merely necessary and not sufficient for the acceptability of a particular non—experimental estimation method, as an extensive example due to LaLonde demonstrates.