John C. Ham, Robert J. Lalonde, The Effect of Sample Selection and Initial Conditions in Duration Models: Evidence from Experimental Data on Training, Econometrica, Vol. 64, No. 1 (Jan., 1996), pp. 175-205

The authors address two questions using experimental data on disadvantaged women. First, what is the impact of being offered Job Training Partnership Act classroom training on the duration of unemployment and employment? Second, what is the effect of actually participating in this training on the length of such spells? Belonging to the treatment group shortens unemployment spells but has no effect on employment spells. Actually participating in training has a larger positive effect on the exit rate from unemployment than the effect of simply being a member of the treatment group. Ignoring the endogeneity of actual training in estimation substantially underestimates its effect.

We estimate the effect of endogenous training participation on transitions in and out of employment for disadvantaged women in the Job Training Partnership Act (JTPA) study. Decomposing the effect of training on employment into its effects on transitions in and out of employment has the potential to develop more effective programs. We also consider a potentially serious identification problem that arises when individuals do not undertake training immediately, and we propose a test to shed light on this problem. We find that this problem is not important in our context. JTPA classroom training substantially reduced unemployment durations, and thus it complements programs that increase employment durations.