Sample Heterogeneity and Measurement Error in Large Sample Accounting Studies: The Case of Earnings Management Research

I review the use of regression-based estimates of discretionary accruals and conclude that misspecification is so severe that the findings of most earnings management studies using this approach are most likely due to Type I errors. In a sample of firm-years spanning the period 1988-2016, I find that 71% of the observations have estimated abnormal accruals that exceed the magnitudes of even the most egregious cases of fraud. I also find that evidence of earnings management by AAER firms is most likely due to model-misspecification and not the underlying earnings management these firms engaged in. Finally, I demonstrate similar, if not worse, misspecification issues when researchers use the absolute value of discretionary accruals to draw inferences. I conclude by suggesting, 1) a much greater emphasize be placed on research that introduces or validates new research methods to assure subsequent research that applies such methods is built on a strong foundation, 2) researchers conduct more narrowly focused studies (e.g., specific industries) to improve internal validity.