Selection Models in Accounting Research

This study explains pitfalls associated with the Heckman (1979) procedure to control for potential selectivity (self-selection) bias and assesses its application in accounting research. Despite a growing use of selection models, we find that accounting studies have implemented the technique in a fairly mechanical way with limited appreciation of the econometric issues surrounding its use. Economists have long stressed that the researcher must identify exogenous independent variables from the first stage choice model that can be validly excluded from the second stage regression in order to successfully implement the Heckman (1979) procedure. However, the importance of this appears to have fallen under the radar of the accounting literature, where studies fail to (a) impose exclusion restrictions in their second stage models, or (b) provide compelling grounds for the chosen restrictions. Consequently, the selection models are unlikely to provide robust results and are vulnerable to high multicollinearity. Using two empirical examples, we illustrate that selection models can provide inferences that are extremely fragile and that have severe multicollinearity problems. In fact, we demonstrate that the selection model can yield quite literally any possible outcome in the second stage estimation response to fairly minor changes in model specification. Our findings underscore the importance of assessing robustness, and we conclude with guidance on how researchers can implement selection models that will provide more convincing evidence on the presence (or absence) of selection bias.