For conducting a meta-analysis on the effectiveness of an intervention on two outcomes (attitudes, and behavioral avoidance).
The effect size measure is Cohen's d. After computing the Cohen's d, I want to proceed to subgroup analysis and meta-regression. I am looking for resources or answers to the 2 questions I came across below.
For instance, one moderator is "interactive approach VS didactic approach", and another one is gender (male VS female).
1) Referring to the Cochrane's Handbook (http://handbook.cochrane.org/chapter_9/9_6_5_1_ensure_that_there_ar... (http://handbook.cochrane.org/chapter_9/9_6_5_1_ensure_that_there_ar...), it stated that at least ten studies in a meta-analysis are required. Does this "10 studies" mean that there should be at least 10 studies/effect sizes EACH for interactive approach and didactic approach respectively? Or just at least 10 studies/effect sizes included in the meta-analysis as a whole? What should we do if there are less than 10 studies? Any remedial methods?
3) Some reviewers said individual-level moderators (e.g. percentage of male, mean age) cannot not be used to conduct subgroup analysis; only the study-level moderators (e.g. publication year, location of study) can be used. Is this always true? What is the rationale?
Thanks!
Hi Gloria,
The 10 studies per moderator is a guideline to follow for running metaregressions. They say this because often, researchers want to look at several predictors in the metaregression, just like in multiple regression. For instance, I might want to look at whether type of intervention moderated the individual-study effect size controlling for percent of males in each study.
About your second question: I disagree with the reviewers who told you you can't use mean age or percent male as moderators. Those are study-level predictors as well. For example, you can explore whether mean age of the participants in each individual study significantly predicted individual study effect size (I've done several of these myself).
So to answer question 1, in my opinion, you can run the moderator analysis of intervention approach predicting your effect size (d).
Best of luck!
Thanks a lot, Jason, for the info! May I ask few more follow-up questions?
1) How should we count the “10 studies”? Does it mean at least 10 individual studies included in the meta-analysis? Or 10 individual effect sizes in EACH category of the same moderator?
2) Referring to your explanation, //For example, you can explore whether mean age of the participants in each individual study significantly predicted individual study effect size//
Do you mean it is okay to treat mean age and percentage of male as predictors in meta-regression?
what about sub-group analysis? Its application is different from meta-regression, right?
3) Any good recommended resources or examples on the latest methodology for meta-analysis on correlations between two psychosocial constructs?
Thanks again!
- Badges
- Science topic
- Similar topics
- Geoscience
- Cartography
- Mapping