Outliers and Robust Inference in Archival Accounting Research

Abstract

This study examines the nature of outliers in archival accounting research, and evaluates the merits and limitations of robust estimators in identifying and downweighting their influence. Using simulated and actual data, we demonstrate how outliers arise non-randomly from the data generating process, research design choices such as scaling, and model misspecification. We find that robust estimators lead to more precise inferences than OLS in common archival data, but these estimators can also bias inferences due their downweighting of substantial and non-random proportions of the data. In addition, we demonstrate that a failure to account for nonlinear relations can induce biases in robust estimators that are more severe than with OLS. We advise researchers to carefully evaluate the source and nonrandom nature of outliers in their samples, to be cautious in implementing and interpreting robust estimators, and to evaluate and report the sensitivity of these methods to critical design choices. We also highlight the usefulness of combining data visualizations with robust estimators to assess both the nature of outliers and their impact on inference.