We propose a method for estimating the mean impact of an assigned social program when it is not feasible to do a pre-intervention baseline survey but it is feasible to track ex-participants. In our triple-difference estimator, measured outcome changes are compared between continuing participants...