Optimizando métodos estadísticos en meta-análisis

Abstract

Meta-analytic methodology allows quantitative integration of the results from a set of primary studies focused on a common topic, through the implementation of statistical methods. The advantages of meta-analysis are numerous. Nowadays, most conclusions about cumulative knowledge in psychology and in other research areas are based on meta-analytic reviews. Due to the broad scope of meta-analysis, it is really important to achieve valid results for the scientific community, by implementing the optimal inferential methods in each scenario. Many Monte Carlo simulation studies have been conducted in order to investigate which techniques and procedures are most adequate given the characteristics of a meta-analytic database. The current Doctoral Thesis gathers a total of three Monte Carlo simulation studies in which different methods for estimating an average effect size in random-effects models, and for testing the significance of moderators in mixed-effects models, were compared. The results from each simulation study show that choice of method influences the accuracy of the estimations and inferential tests. As a previous step, a methodological review of meta-analyses about the effectiveness of psychology treatments was carried out with the aim to help us, as well as other researchers, design the scenarios for Monte Carlo studies. This Doctoral Thesis concludes with several recommendations about the most appropriate conditions and methods to maximize the accuracy of the results in a meta-analysis.