Publication Bias in Recent Empirical Accounting Research

Abstract

We analyze p-values in empirical papers published in the “top three” accounting journals in 2011. We estimate average Type I error and Type II error rates to be 11% and 22% respectively. The p-value distribution is strongly concentrated near zero and the frequencies of p-values just below conventional significance levels are higher than frequencies just above these levels. These p-values patterns are stronger and error rates higher for Test variables than Control variables and in rounded and directly reported p-values. The findings likely derive from selective reporting of statistically significant results or searching across research designs for p-values for Test variables below conventional significance levels.

Keywords: statistical significance, research methodology, t-statistic, meta-analysis