La estimación del tamaño del efecto medio en un meta-análisisuna comparación entre los modelos de efectos fijos y aleatorios

Abstract

The statistical techniques usually applied in meta-analysis are based on a fixed-effects model. However, recent méthodological research supports the best performance of the random-effects model to reach the true aims of a meta-analysis (Erez, Bloom & Wells, 1996; Field, 2001; Hedges & Vevea, 1998; National Research Council, 1992; Overton, 1998). This study compares the differences between the results of applying techniques based on fixed and random-effects models, when the estímate of the average effect size of a meta-analysis is the main purpose. Both sorts of techniques are applied in a Monte Cario simulation study. The simulation is based on studies with experimental and control groups, and a continuous outcome, the standardized mean difference being the effect size index. The shape of the distribution of the parametric effect size, its magnitude, the sample size of the groups, and ·the nurnber of studies of each meta-analysis were manipulated. The results showed confidence intervals narrower but less adjusted in the techniques that assurned a fixed-effects model.