Discontinuous Distribution of Test Statistics Around Significance Thresholds in Empirical Accounting Studies

Examining test statistics from articles in six leading accounting journals, we detect discontinuities in their distributions around conventional significance thresholds (p-values of 0.05 and 0.01) and find an unusual abundance of test statistics that are just significant. Further analysis reveals that these discontinuities are more prominent in studies with smaller samples and are more salient in experimental than in archival studies. The discontinuity discrepancy between experimental and archival studies relates to several proxies for researcher degrees of freedom. Nevertheless, this evidence does not imply that experimental research is more prone to questionable research practices than archival studies. Overall, our findings speak to the concern of whether accounting researchers could exercise undisclosed discretion to obtain and report statistically significant results. Based on our results, a healthy skepticism of some just-significant test statistics is warranted.