Selection Bias in Returns to College Education in Europe

Information on economic returns to college education has indisputable value for policymakers and individuals, yet standard estimates hinge on unrealistic assumptions about the homogeneity of treatment effects across the population or do not take into account differences in the backgrounds of college-goers and non-college-goers. In this paper we address those issues by using the local instrumental variable method for heterogeneous treatment effects due to unobservables presented in general form in Heckman, Urzua, and Vytlacil (2006), which holds a number of methodological advantages over the approach more commonly known in sociology (Brand and Xie 2010). Estimates are performed for 28-to-38-year-olds in 29 EU countries participating in the 2011 EU-SILC survey, i.e. the most recent wave when data on parental background was also collected. We estimate returns for respondents who went to college (TT), unrealized returns for those who did not go to college (TUT), and mean returns (ATE) for men and women in all countries. Our results reveal the presence of negative ability bias – i.e. that Mincer-type returns to college strongly underestimate those returns compared to models utilizing the Heckman estimation procedure – particularly in countries with the lowest degrees of educational opportunity at the tertiary level. On the other hand, the degree of ability bias tends to decline as educational opportunity rises, leading to insignificant differences between OLS and MTE estimates for some countries.