A dynamic discrete choice model is set up to estimate the effects of grade retention in high school, both in the short- (end-of-year evaluation) and long-run (drop-out and delay). In contrast to regression discontinuity designs, this approach captures treatment heterogeneity and controls for...