Estimating Models with Binary Dependent Variables: Some Theoretical and Empirical Observations

Abstract

Many mathematically similar models are being used by business researchers to link binary dependent variables with a set of predictor variables. Typical research results indicate little difference between models in their ability to properly classify observations.

But, there appear to be major differences in the interpretation of coefficients resulting from the calibration of these competing models.

The empirical results in this article clearly show that when the assumptions underlying binary-dependent-variable techniques are violated, parameter estimates may be misleading. This can be true even when the goodness-of-fit statistics are not substantially affected.