In primary studies, psychotherapy researchers frequently search for mediator and moderator variables that can help them understand the relationship between treatment and outcome. Yet a review of past psychotherapy meta-analyses revealed that none examined the possible role of mediator variables; and although all of them searched for moderators of study outcome, that search was generally not as complete as it could have been. This article illustrates methods for studying such mediator and moderator variables in meta-analysis, discusses their advantages and disadvantages, and shows how the inclusion of these variables can change interpretation of meta-analytic results. In particular, the perennial interpretation of past psychotherapy meta-analyses that therapeutic orientation makes no difference to outcome--or as the dodo bird put it: "Everyone has won and all must have prizes'--may be wrong. Orientation may make significant difference, but only by virtue of its moderating and mediating effects.To the best of our knowledge, all meta-analyses ever done have concluded that (on the average) clients receiving psychotherapy do better than clients not receiving psychotherapy. In fact, the computation of average therapy effects over studies is the defining strength of meta-analysis. But this strength leads to a criticism of meta-analysis: Knowledge of average effects says nothing about when, where, why, and how therapy works. The latter questions concern mediators and moderators of therapy outcome. The present article describes methods for addressing such questions in meta-analysis.Moderators and mediators are third variables that help researchers to understand the relationship between independent and dependent variables (Baron & Kenny, 1986). "A moderator is a qualitative (e.g., sex, race, class) or quantitative (e.g., level of reward) variable that affects the direction and/or strength of the relations between an independent or predictor variable and a dependent or criterion variable" (Baron & Kenny, 1986, p. 1174. Moderators cause statistical interactions. Some moderator variables are categorical. Suppose, for example, that behavioral therapies yielded high effect sizes on behavioral presenting problems but low effect sizes on nonbehavioral presenting problems, with the opposite pattern emerging for nonbehav-

This research explored the relationships between students' multidimensional selfconcepts as a function of students' family structure. Participants were 815 adolescent and preadolescent children, ages 9 to 19, selected from 17 sites across the four major regions of the United States. Students were classified into five family types (i.e., intact, reconstituted mother-headed, reconstituted father-headed, single parent mother-headed, single parent father-headed). Results indicated that the total self-concepts of students from single-parent families were significantly lower than the global scores of students from intact families. Family self-concepts of students from reconstituted families were significantly lower than students from intact families. This finding highlights the sensitivity of domain-specific selfconcepts.

No abstract