One needs to keep in mind that even a meta-analysis of findings only from RCTs is an ‘observational’
(non-experimental) study. One should approach the conduct of a meta-analysis or read a
meta-analytic report the same way one would conduct or read a report of a primary observational
study. What important variables were not assessed that might explain variation in effect sizes?
What variables were assessed but not examined? Is there overlap or confounding among moderator
variables? The fact that a moderator variable accounts for some variation in effect sizes does not
rule out the possibility that another moderator variable (assessed or not assessed in studies) might
also account for much of the same variation. For example, differences in methodological features
might account for some of the observed variation in effect sizes across studies that may have been
attributed to other substantive moderator variables (e.g. different types of interventions; variation
in sample composition – see Wilson & Lipsey, 2001).