Out of Control : The (Over)use of Controls in Accounting Research

The selection of controls in model design is essential to convincing empirical accounting research. However, we note that the theory for determining which factors should be “controlled” is often underappreciated, especially given the complexity of accounting data. In a sampling of articles from top accounting journals, we find that studies devote relatively little page space to the theoretical justification for selection and specification of control variables and note that studies often include control variables that could interfere with causal interpretations of the study's coefficient estimates. To guide future accounting research, we modify the widely-used “Libby Box” framework to incorporate control variable selection and construct to control variable mapping. Using this framework, we discuss several common issues with control variables that can lead to inappropriate model specification for estimating causal effects following Gow et al. (2016). We then use simulations and examples to demonstrate the implications of these issues. First, we illustrate that “more control” is not always better and that some control variables can introduce endogeneity. Next, we discuss issues with control variable measurement error, control variable selection for interactions, and complications associated with fixed effects controls. Lastly, we provide suggestions for future accounting research