COMPONENTS OF ARTIFACTUAL VARIANCE IN META‐ANALYTIC RESEARCH

Except for a few rough estimates, the literature contains no precise data on the relative size of each correctable artifact in a meta‐analysis. The purpose of the present study was to determine these proportions and see if artifacts were related to various sample and subpopulation parameters. The study confirmed the notion that sampling error accounts for the bulk of the total artifactual variance (i.e., more than 90% for small or medium samples and more than 70% for large samples, e.g., N = 500). Nevertheless, variation in the percentages of each artifact was found to be primarily a function of sample size and mean correlation. Total artifactual variance was found to be a function of subpopulation correlation differences and sample size. Practical implications of the results are presented.