Selectivity and Heterogeneous Returns to Schooling in Italy: An Evaluation Based on Control Functions

The estimation of economic average returns to education is subject to a number of well-known econometric problems. The first is the so-called ‘ability bias’, responsible for the ‘endogeneity’ of education in earnings equations. The second is the so-called ‘return bias’, due to heterogeneous returns to education so that individuals can self-select themselves into given education levels according to specific unoservable gains. In many empirical applications, instrumental variables (IV) are used to tackle selectivity issues. A well-known limitation of IV is that they provide consistent estimates of average returns to schooling only if they are homogeneous in the population, but subject to endogeneity and the associated ‘ability bias’. If returns to education are heterogeneous, estimates by instrumental variables are able to identify only ‘local’ effects (LATE), i.e. the average return for the subpopulation of individuals affected by the instrument, making OLS and IV estimates not directly comparable. The aim of this paper is to overcome these limitations and to use control functions (CF) a la Heckman to get consistent estimates of returns to schooling in a model which allows for hetereogeneity in both ability and returns to education. At the cost of additional restrictions on the joint determination of schooling and education, the CF approach more flexible than IV since it can simultaneaously corrects for both endegenity and self-selection by means of specific corrections terms included in the earnings equation. To this purpose, we use data Bank of Italy data for Italian men in the 25-60 age interval. Similarly to other papers for Italy, identification is ensured by a major reform which in 1962 increased the number of years of compulsory schooling, and we use a set of cohort dummies as exclusion restrictions.