Outliers and Robust Inference in Archival Accounting Research

This study examines the nature of outliers in archival research and evaluates the merits and limitations of robust regression estimators in identifying and downweighting their influence. Using simulated and actual data, we demonstrate how outliers arise from the data-generating process, research design choices such as scaling, and model misspecification. We next show that robust regression estimators can generate substantially more precise estimates than OLS in common archival data, but also conclude that these estimators can bias inferences because they downweight substantial and nonrandom subsets of the data. We recommend researchers to be cautious in implementing robust regression estimators and to evaluate and disclose the sensitivity of these estimators to alternative design choices. Whether robust estimators improve inference critically hinges on the nature of the observations that are downweighted.