Sample Selection Bias As a Specification Error (with an Application to the Estimation of Labor Supply Functions)

Abstract

In this paper, the bias that results from using nonrandomly selected samples to estimate behavioral relationships is shown to arise because of a missing data problem. In contrast with the standard omitted variable problem in econometrics, in which certain explanatory variables of a regression model are missing, the problem of sample selection bias arises because data are missing on the dependent variable of an analysis. Regressions estimated on the data available from the nonrandom sample will not, in general, enable the analyst to estimate parameters of direct interest to economists. Instead, such regression coefficients confound meaningful structural parameters with the parameters of the function determining the probability that an observation makes its way into the nonrandom sample.