Abstract

Researchers commonly use variance-stabilizing transformations while performing meta-analyses of proportions, to obtain better approximations to the normal distribution and to avoid pooled proportions exceeding 0 or 1. Two variance-stabilizing transformations are commonly used: the logit and the Freeman–Tukey double arcsine (FTT) transformations. This presentation reports the use of simulation to address specific concerns related to the utility of double arcsine transformation for meta-analysis of binomial data and clarifies the underlying “erratic” behavior of the double-arcsine transform highlighted in recent criticisms. It also compares this transform with the logit transform and provides data on choice of transform for researchers