Meta-Analysis

Types of Meta-Analyses

Meta-analytic estimates are calculated in one of two ways, either through fixed or through [Page 1052]random effect models. Fixed-effect models assume that all variabilities observed between study effect sizes are due to the differences in sampling error. Underlying this sampling error is a single, common effect (e.g., the true effect). Fixed-effect models assume a null hypothesis of no treatment effect in every study.

Fixed-effect models are appropriate if it can be assumed that all study characteristics are the same (e.g., identical conditions surrounding the characteristics of the participants, the researchers, and the experimental dosage). This assumption is rarely appropriate; multiple studies are seldom conducted in an equivalent manner. Indeed, even if a researcher were to find what the researcher believes to be appropriate conditions for a fixed-effect meta-analysis, heterogeneity of effect sizes should be tested to ensure the assumptions inherent to a fixed-effect model are tenable.

Conversely, random-effects models assume that there is no single estimate of effect. Instead, the actual influence of a given factor will vary between studies as a function of study characteristics. Effect sizes observed in studies from which a random-effects meta-analysis is calculated are assumed to be drawn from a normally distributed, random sampling of effect estimates with a mean of zero.

The assumption that a random-effects model varies across the study context is important because it provides a role for moderating influence, a factor unavailable in fixed-effect models. For instance, a study of parenting interventions for disruptive child behavior using a fixed-effect model would assume that the intervention would have the same effect across all studies; a random-effects model estimates the effect of the intervention as a function of other characteristics, such as sampling variability or a moderator such as parent-child attachment.

Preparation for Meta-Analysis

Essential to any meta-analysis is thorough preparation, beginning by specifying a precise research question. Once a clear intended thesis has been decided, determinations of how strictly to impose standards of quality can be made through the defining of criteria for study inclusion. There are many arguments for implementing strict guidelines for inclusion; however, there remains a need to present a thorough and comprehensive review, and some arguments for strictness are counterbalanced by the nature of meta-analysis. Strict guidelines on study quality can impact meta-analytic results and may be more comprehensively explored using moderator analyses.

...