Publication Bias and Editorial Statement on Negative Findings

The interpretation relies on the identification condition that the distribution of test statistics in the treated and control journals would follow the same time trend in the absence of the treatment, i.e., editorial statement. [...] Furthermore, the share of tests significant at the 10% and 5% level did not changed from “During” to “After.” The rightward shift of the distribution of tests for the control journals possibly reflects a change in the preferences for negative findings from the editors and referees and/or more specification searching from authors. [...] The identification assumption is that the proportion of reported significant findings would have been the same in the absence of the editorial statement. [...] The year of submission determines whether the journal article is in the category “After 2015− 16” or “After 2017.” Similarly, we split the category “Before” and compare the “Pre Before” and “Before” time periods In the treated journals, approximately 43% (48%) of test statistics were significant at the 5% (10%) level for the category “During” in comparison to 44% (51%) for the category “After 2015. [...] More precisely, we check whether the leftward shift in the distribution of tests after the editorial statement is related to the number of authors and the presence of a theoretical model.