Earnings Functions, Rates of Return and Treatment Effects: The Mincer Equation and Beyond

Numerous studies regress log earnings on schooling and report estimated coefficients as "Mincer rates of return". A more recent literature uses instrumental variables. This chapter considers the economic interpretation of these analyses and how the availability of repeated cross section and panel data improves the ability of analysts to estimate the rate of return. We consider under what conditions the Mincer model estimates an ex post rate of return. We test and reject the model on six cross sections of U.S. Census data. We present a general nonparametric approach for estimating marginal internal rates of return that takes into account tuition, income taxes and forms of uncertainty. We also contrast estimates based on a single cross-section of data, using the synthetic cohort approach, with estimates based on repeated cross-sections following actual cohorts. Cohort-based models fitted on repeated cross section data provide more reliable estimates of ex post returns. Accounting for uncertainty affects estimates of rates of return. Accounting for sequential revelation of information calls into question the validity of the internal rate of return as a tool for policy analysis. An alternative approach to computing economic rates of return that accounts for sequential revelation of information is proposed and the evidence is summarized. We distinguish ex ante from ex post returns. New panel data methods for estimating the uncertainty and psychic costs facing agents are reviewed. We report recent evidence that demonstrates that there are large psychic costs of schooling. This helps to explain why persons do not attend school even though the financial rewards for doing so are high. We present methods for computing distributions of returns ex ante and ex post. We review the literature on IV estimation. The link of the estimates to the economics is not strong. The traditional instruments are weak, and this literature has not produced decisive empirical estimates. We exposit new methods that interpret the economic content of different instruments within a unified framework.