Abstract
Rosenthal's (1979) Fail-Safe N (FSN) is an estimate of the number of unpublished studies (in "file drawers" of labs, p. 638) that would threaten the validity of significant combined results of a meta-analysis. However, the FSN's derivation assumes that studies targeted in a meta-analysis use 2-tailed (nondirectional) tests. When directional "Adjusted Fail-Safe Ns" (AFSNs) are calculated (derived from equations in Iyengar & Greenhouse, 1988a), very different conclusions may be drawn about the meanings of meta-analyses: In 12 of 16 meta-analyses targeting studies with directional predictions, AFSNs contradicted the FSN-based conclusion that the publication bias was negligible. A similarly conservative finding (11 of 16) was obtained with another directional fail-safe N method proposed by Darlington and Hayes (2000). A table of AFSNs is provided for 25 categories of meta-analyses for which Rosenthal and Rosnow (1984) reported FSNs: The FSN was up to 58 times larger than the AFSN.