Stéphane Bonhomme Ulrich Sauder Accounting for unobservables in comparing selective and comprehensive schooling

We compare the effects of selective and non selective secondary education on children's test scores, using British data from the National Child Development Study (NCDS). Test scores are modelled as the output of an additive production function. Inputs include family and school characteristics, as well as the child's unobserved initial endowment, which may be correlated with the education system attended. In the model, the average effect of selective education can be estimated using semiparametric Difference-in-Difference (DID) methods. We generalize the DID approach and provide conditions under which the entire counterfactual distribution of potential outcomes is identified, and can be consistently estimated using a deconvolution-related approach. Descriptive statistics on the NCDS data show that children perform better in selective schools. Our results suggest that this is essentially due to differences in pupils' composition between selective and non selective schools. When correcting for these differences, we find that the effects of selective education are small and mostly insignificant.