Fail-safe N

Rosenthal suggested assessing the potential for publication bias to have influenced the results of a meta-analysis by calculating the ‘fail-safe N’, the number of additional ‘negative’ studies (studies in which the intervention effect was zero) that would be needed to increase the P value for the meta-analysis to above 0.05 (Rosenthal 1979). However the estimate of fail-safe N is highly dependent on the mean intervention effect that is assumed for the unpublished studies (Iyengar 1988), and available methods lead to widely varying estimates of the number of additional studies (Becker 2005). The method also runs against the principle that in medical research in general, and systematic reviews in particular, one should concentrate on the size of the estimated intervention effect and the associated confidence intervals, rather than on whether the P value reaches a particular, arbitrary threshold, although related methods for effect sizes have also been proposed (Orwin 1983). Therefore this and related methods are not recommended for use in Cochrane reviews.