Fractional Regression Models in Strategic Management Research

Abstract

Many outcomes of interest to management and strategy researchers are fractional in nature. Fractional outcomes are attractive to researchers because they are easy to understand and facilitate interpretation of effect sizes that are meaningful and comparable across organizations. We review 10 years of research in five leading strategy and management journals. We find that approximately 5 percent of articles include statistical estimation of a fractional outcome. However, few studies use the best available statistical techniques. Using two empirical examples and a simulation, we demonstrate that linear regressions (incl. transformations) and Tobit models are inferior to fractional logit models and mixture models. The former are likely to produce biased estimates and unreliable significance tests. We present advice on how to estimate fractional outcomes in different situations.