The Efficacy of Ability Proxies for Estimating the Returns to Schooling: A Factor Model-Based Evaluation

Abstract

This paper revisits the issue of estimating the returns to schooling within a framework that allows multiple unobserved skills with potentially time-varying prices where both skills and prices are possibly correlated with schooling. A common approach to addressing the problem of ability bias emanating from the unobservability of skills is to augment the earnings-schooling regression with various measures of cognitive and non-cognitive ability as proxies for these skills. While the adequacy of such measures has long been questioned, a rigorous assessment of their effectiveness in estimating the magnitude of ability bias has received relatively less attention. This paper undertakes a formal evaluation of the proxy approach by adopting a methodology that models the skills and their prices using an unobserved factor structure in which the factor loadings represent the latent skills and the common factors their associated prices. Our factor model approach allows consistent estimation of the returns to schooling without the need to rely on proxies thereby providing a flexible test bed for quantifying the contribution of the proxies to the aggregate least squares bias. The factor model estimators may also be viewed as implicitly estimating the measurement error inherent in the proxies. Our empirical results using panel data from the NLSY79 show that the estimated ability bias lies between 39-51% for our most general specification. A bias decomposition analysis indicates that the commonly used proxies can explain very little of the estimated bias. Direct tests for the viability of the proxies further confirm their inadequacy in capturing the underlying latent skills.