Recent work has highlighted the differing returns that individuals from different social groups experience following a university degree. Combining these insights into the heterogeneity in returns to higher education with insights on the technology of skill formation from another important recent literature, we estimate the wage returns to a university degree along a fundamental dimension: prior human capital. Employing a novel methodology exploiting recent advances in the identification of mixture models, we group individuals by their prior human capital and identify and estimate the wage returns to a university degree by group. Applying our method to data from a UK cohort study, our findings reflect recent evidence that skills and ability are multidimensional.

Why do some people choose to attend university, and enjoy state-subsidised benefits, while others do not? We shed new light on this key issue by comparing and quantifying the roles of earnings, financial, and non-pecuniary factors in the educational decisions of young people in the UK. We investigate changes in these factors over time, and their implications for social mobility. We specify a model of educational choice, explicitly including expectations about earnings, financial, and non-pecuniary factors. Our estimation strategy exploits panel survey data on young people’s expectations about key outcomes both at, and after, university, linked to their realised outcomes. Income maximisation, despite its prevalent role in the literature, is only a small part of the story: other factors are four times as important as earnings in determining whether someone goes to university. Non-pecuniary factors also drive both the SES-gap in educational attainment, and the huge growth in degree attainment between the 1980s and 2010s.

We develop a finite-mixture framework for nonparametric difference-in-difference analysis with unobserved heterogeneity correlating treatment and outcome. Our framework includes an instrumental variable for the treatment, and we demonstrate that our method allows us to relax the no common trend restriction usually required in difference-in-difference analysis. We also show that outcomes can be Markovian provided there are multiple post-treatment observations. Our main theoretical contributions are the substitution of an instrument for the common-trends assumption, and a non-parametric identification proof. Empirically, we apply our framework to evaluate the effect of on-the-job/professional (re)training on wages, using novel French linked employee-employer data. Estimating our model using the EM-algorithm, we find small ATEs and ATTs on hourly wages of between 2% and 3%. However, we find larger effects on hours and annual wages with both ATEs and ATTs of over 5%. A simple extension to our model to include observed as well as unobserved heterogeneity produces very similar results..