Comparison of earnings between control and experimental groups subsequent to participation in experimental training programs provides a credible estimate of program effectiveness. Too much of the nonexperimental estimation of the effects of training programs seems dependent on models that cannot be subjected to powerful statistical tests. (MLF)