This paper is concerned with the relationship between class size and the student outcome length of time in post-compulsory schooling. Research on this topic has been problematic partly because omitted unobservables, like parents' incomes and education levels, are likely to be correlated with class size. Two potential ways to resolve this problem are to exploit either experimental or instrumental variation. In both cases, the methods require that the variation in both class size and the outcome should not be contaminated by other unobservable factors that affect the outcome - like family background. An alternative approach, which we pursue here, is to take advantage of variation in class size between siblings which allows unobservable family effects to be differenced out. Our aim is to provide estimates of the effect of class size and use these to conduct an evaluation of the costs and benefits of a reduction in class sizes.
class size regression discontinuity sibling differences