“Fail safe N” is a useful mathematical measure of the stability of results

EDITOR,—Matthias Egger and George Davey Smith1 have responded to our letter2 about their editorial on meta-analysis3 by pointing out that measurement of the heterogeneity of, or variability in, study results (as a guide to the likely robustness of the averaged finding) is plagued with statistical problems, notably low statistical power. The power of tests for heterogeneity improves considerably, however, when larger numbers of studies are used in the computation.4

Egger and Davey Smith are also right to call for wider use of funnel plots in meta-analyses …