When looking at the relationship between individual earnings and schooling, there are a number of potential sources of bias which arise due to individual educational choices; individuals of higher unobserved ability or with higher unobserved payoffs from schooling may for instance invest more in education.

This work reviews alternative models (single- and multiple-treatment, with homogeneous or heterogeneous returns) and four popular estimation methods (least squares, matching, instrumental variable and control function methods) meant to overcome these sources of bias and to thus recover the true causal effect of education on earnings.

After having been extensively compared from a methodological point of view, these models and methods are applied to high quality data the British 1958 NCDS birth cohort to estimate the wage returns to different educational investments (O levels or equivalent, A levels or equivalent and higher education) and to compare their respective performance.

This application has highlighted that correcting for detailed ability and family background differences is crucial. We have found evidence of heterogeneity in the returns to higher education in terms of the (rich) observables in the NCDS. Overall, in our specific application matching on detailed individual characteristics appears to perform well for the average return to higher education reaped by graduates. The overall returns to educational qualifications at each stage of the educational process were found to remain sizeable and significant, even after allowing for selection and heterogeneity in returns.