Two Strategies for Meta-Analysis of Psychological　Studies

Abstract

Standard statistical texts generally focus on the analysis of a single experiment. In psychological research, a question is often investigated by several studies using similar designs. Meta-analysis is a quantitative method of combining the results of related but independent studies. Research psychologists use meta-analysis to estimate the overall treatment effect from different studies, to assess the variations between studies, and to identify the study characteristics associated with effective treatments. This talk presents the applications of both linear mixed models and Bayesian models for three meta-analyses in psychological research: (1) 14 studies on gender differences in field articulation ability, (2) 19 studies on the effects of teacher expectancy on pupil's IQ, and (3) 24 studies on the effectiveness of psychosocial treatments for addiction. In all three analyses both strategies yield comparable results, though the Bayesian models produce slightly wider estimates of between-studies variations.