The effect of some estimators of between-study variance on random-effects meta-analysis

Abstract

There are different methods for estimating the between-study variance, ?² in meta-analysis, however each of the methods differs in terms of precision and bias in estimation. Consequently, each of the estimators have a different effect on the estimate of the population treatment effect parameter ?. This paper compares the effect of four estimators by DerSimonian and Laird (1986), Paule and Mandel (1982), Sidik and Jonkman (2005) and the restricted maximum likelihood method (REML). Simulations show that of all estimators, random-effects meta-analysis based on REML yielded the most accurate coverage probability for treatment effect when treatment effects are highly heterogeneous.

Keywords: Meta-analysis; random-effects model; between-study variance; and coverage probability