Conservative Tests under Satisficing Models of Publication Bias

• DOI: 10.1371/journal.pone.0149590
• Publication Date: 2016-02-22T18:51:17Z
• Authors: Justin McCrary; Garret Christensen; Daniele Fanelli
• Keywords: Satisficing; Null (SQL); Statistical power

Publication bias leads consumers of research to observe a selected sample statistical estimates calculated by producers research. We calculate critical values for significance that could help adjust after the fact distortions created this selection effect, assuming only source publication is file drawer bias. These adjusted are easy and differ from unadjusted approximately 50%-rather than rejecting null hypothesis when t-ratio exceeds 2, analysis suggests 3. Samples published social science indicate on average, across fields, 30% t-statistics fall between standard cutoffs.