Nonparametric Bounds on Returns to Schooling: Overcoming Ability and Selection Bias

Juergen Meinecke, Martine Mariotti
Australian National University, Australia

Our objective is to estimate the average treatment effect (ATE) of education on earnings for African men in South Africa. Estimation of the ATE in our data is difficult because of omitted ability bias and a high degree of sample selection due to low labor force participation. Manski and Pepper (2000) suggest is a promising nonparametric identification strategy but it only helps with the problem of omitted ability bias. We propose an extension of their identification strategy to deal with the sample selection problem. Accounting for ability and selection bias, we compute upper bounds on the ATE for the years 1995 and 2000. We estimate an upper bound of 12.64 percent in 1995 and 10.68 percent in 2000. Compared to parametric estimation our bounds are informative: The OLS returns to schooling equal 15.59 percent in 1995 and 15.31 percent in 2000. Our results suggest that many parametric estimates are severely upwards biased, which results from unobserved heterogeneity.

View full paper