Grade retention and unobserved heterogeneity

We study the treatment effect of grade retention using a panel of French junior high-school students, taking unobserved heterogeneity and the endogeneity of grade repetitions into account. We specify a multistage model of human-capital accumulation with a finite number of types representing unobserved individual characteristics. Class-size and latent student-performance indices are assumed to follow finite mixtures of normal distributions. Grade retention may increase or decrease the student's knowledge capital in a type-dependent way. Our estimation results show that the average treatment effect on the treated (ATT) of grade reten- tion on test scores is positive but small at the end of grade 9. Treatment effects are heterogeneous: we find that the ATT of grade retention is higher for the weak- est students. We also show that class size is endogenous and tends to increase with unobserved student ability. The average treatment effect of grade retention is neg- ative, again with the exception of the weakest group of students. Grade repetitions reduce the probability of access to grade 9 of all student types.