Lipsey M W, Wilson D B
Eval Health Prof. 2001 Sep;24(3):236-54. doi: 10.1177/016327870102400302.
Intervention studies represent webs of interrelated substantive and methodological characteristics that take on different patterns in different studies and different intervention areas. All too often, meta-analysts do not give close attention to the possibility that these interrelated differences among studies are related in complex ways to study effect sizes and, consequently, run considerable risk of reporting results that are misleading or flatly wrong. To remedy this situation, improvements are needed in both the method and practice of meta-analysis so that greater attention can be given to effect size variation, the generalizability of study results, and the systematic multivariate relationships between study characteristics and the effect sizes reported in those studies.
