Simple tests for sample selection bias in censored and discrete choice models

This paper presents two simple tests of sample selection bias for models where the primary equation of interest has a censored or discrete dependent variable. The first test is derived as a conditional moment test and can be implemented in a regression-based framework. The second test is an extension of the testing procedures proposed by Heckman (1979) and Vella (1993) and is a t-test on a constructed variable in an auxiliary equation. The utility of the tests is illustrated in a model determining the receipt of work conditioned nonwage labour income over a subsample of working women.