Binary Regression Models: An Average Partial Effects Approach

Abstract

This chapter provides a comprehensive guide to binary regression modelling for management research, focusing on average partial effects (APEs) as the primary estimand. Using data on employment gaps and leadership emergence, I demonstrate how linear probability models and probit specifications can yield complementary insights when properly implemented within an APE-focused framework. The chapter emphasizes systematic approaches to interaction analysis in nonlinear models, highlighting how coefficient-based interpretation can mislead researchers and advocating for visualization-based methods combined with formal statistical testing of APEs. I then apply this APE approach to instrumental variable estimation using both two-stage least squares (2SLS) and control function methods, extending the systematic APE workflow to settings where credible causal inference requires addressing endogeneity. Through comprehensive sensitivity analysis, I demonstrate the importance of evaluating assumption validity and the challenges of detecting interaction effects in IV frameworks, showing how the APE approach maintains interpretability even in complex settings.