Dobbelsteen, S., Levin, J. and Oosterbeek, H. (2002). The causal effect of class size on scholastic achievement: Distinguishing the pure class size effect from the effect of changes in class composition Oxford Bulletin of Economics and Statistics, 64(1):17--38.

This paper estimates the causal effect of class size on school performance in Dutch primary schools using an instrumental variable approach. The instrument used is based on Dutch regulations determining the numbers of teachers as a function of school enrollment. Without instrumenting class size the effect of class size on school performance is non-negative (smaller classes are not associated with higher performance). After instrumenting class size, the effects become larger (more positive). Hence the non-negative effect of class size cannot be attributed to endogeneity bias. Next we investigate an explanation for the non-negative class size effect advanced in the social cognitive learning literature. This literature strongly suggests that pupils' achievement may benefit from a larger number of classmates with a similar level of ability. As class size reduction also reduces the expected number of classmates with similar ability, this effect may neutralize the pure class size effect. To test this idea, we repeat the analyses including the number of classmates with similar IQ as an additional regressor. The results show that this new variable has a highly significant and positive effect on school performance. In addition, inclusion of this variable reduces the class size coefficient and in most cases produces the proper negative sign.