Main Effects And Interaction in Meta-Analysis

A deeper look at Meta-Analysis and how to do it for your dissertation

Before examining several of the quantitative techniques available to synthesists, it is important to take a closer look at some of the unique features of accumulated research results. Most research syntheses first focus on tests of main effects that were carried out in the primary studies. This is large because conceptually related replications of main effects occur more frequently than tests of three or more interacting variables. So, for example, you are likely to find many more tests of interactions of whether his relationship is influenced by the number of choices given. Keep in mind that I am referring here to interaction tests within a single study, not your ability to test for the influence of number of-choice at the synthesis level because different studies have varied in the number of choices they provide in their test of the main effect.

Analyzing and integrating the outcomes of studies

It is not that interactions tested in primary studies cannot be combined. However, such replications are fever and, we shall see in the next chapter, their interpretations can be a bit more complex. There are two different ways that interactions tested in primary research could be statistically combined across studies. First, the relationship strengths associated with each study’ s interactions test could be aggregated. An alternative strategy would be to aggregate separately the relation of two of the interacting variables at each level of the third variable. For instance, assume there exists a set of studies in which the primary researchers tested whether the effect of choice in intrinsic motivation depending on the number of choices given by aggregating all motivations measures taken under conditions with one choice. They could do the same for measurements taken after, say, two or three choices. This would probably be more useful and easily interpretable than a direct estimate of the magnitude of the interaction effect. However, in order to do this, the primary research reports must contain the information needed to isolate the different simple main effects. The synthesist might also have to group a number of choices (say, three to five choices and six or more choices) in order to have enough tests to generate a good estimate.

Because main effects are most often the focus of meta-analysts and in many instances, meta-analysts interested in interactions reduce them to simple effects, my discussion of the quantitative combining techniques will refer to main effects only. The generalization to meta-analyzing interactions is mathematically straightforward.