Evaluating the Impact of Education on Earnings in the UK: Models, Methods and Results from the NCDS. CEE DP 47

When looking at the relationship between individual earnings and schooling, there are potential sources of bias which arise due to individual education choices; individuals of higher unobserved ability or with higher unobserved payoffs from schooling may for instance invest more in education. This paper reviews alternative models and estimation methods meant to overcome these sources of bias and to thus recover the true causal effect of education on earnings. As to the specification of the model, the paper highlights the distinction between models which focus on the impact of a specific educational level, such as undertaking higher education compared to not doing so (single-treatment models), and models which allow for a number of sequential levels of schooling (multiple-treatment models). The latter framework is particularly attractive when interest lies in a wide range of educational qualifications with potentially very different returns. A second crucial choice as to model specification concerns the nature of the returns to education, in particular whether to allow returns to vary across individuals for the same educational qualification (homogeneous versus heterogeneous returns models). Once heterogeneity in returns is allowed, it becomes crucial to define the subpopulation for whom one is interested in estimating returns. The authors first provide a methodological comparison of four main estimation methods: least squares, matching methods, instrumental variable methods and control function methods. For the different model specifications, the paper highlights the assumptions needed for the validity of each method and investigates the estimators' properties. In particular, whilst the feasibility of these alternative methods clearly hinges on the nature of the available data, their implementation and properties differ according to whether the model is one of homogeneous or heterogeneous returns and whether education is represented through a single or a multiple measure. Appended are: (1) Summary statistics; (2) Classification of educational qualifications; (3) Covariate balancing indicators before and after matching (best specification); and (4) Difference in returns for the no-qualifications group. (Contains 2 tables.)