I am working on a meta-analysis on framing with cultural-level moderators (e.g. Hofstede dimensions including Individualism-Collectivism and Power Distance) and moderators based on findings in the literature (e.g. health behavior function, behavioral frequency). While the potential mechanisms and theoretical basis of moderations would be different with different moderators, they are highly correlated (r > 0.70, or V > 0.50 for two categorical moderators). What are the possible ways to address the potential confounds between these moderators *in the context* of a meta-analysis? Let's say I find support for moderations with both Moderator A (p < .001) and Moderator B, I am not sure how to know if "significant moderation" of B is due to strong correlations/confounding relationship with A, or if both moderators are really meaningful moderators. Thank you. Would appreciate if there is (ideally R) open data/code/example.
Siu Kit Yeung Subgroup analysis and meta regression in meta analysis can help in addressing your query, further detail is necessary to precisely resolve your issue.
Thank you Sasidharan Sivakumar Sorry wasn't clear. I already conducted *single-moderator* meta-regressions for each potential moderator. However, one issue in *multiple-moderator* meta-regression is multicolinearity, which sometimes result in really weird results. So one possible method is mean-centering the variables and combining some variables (that are closely related to same theoretical construct + highly correlated, for example Individualism-Collectivism and Survival-Expression values are both related to regulatory focus; Power Distance and Indulgence-Restraint are both related to psychological reactance) then ran a multiple-moderator meta-regression (https://bookdown.org/MathiasHarrer/Doing_Meta_Analysis_in_R/metareg.html#multiple-metareg-R), but I am not sure if there is a better way that I am unaware of.
Siu Kit Yeung The methods you've mentioned—mean-centering the variables and combining variables of the same theoretical construct—appear suitable for addressing multicollinearity in your multiple-moderator meta-regression analysis. While other techniques like PCA and PLS are available, the approaches you've highlighted seem adequate for your needs. Of course, you can always experiment with different methods to see what works best for your data, but starting with these techniques is a solid choice.
I just realized the correlations remain very strong (actually, exactly the same) after mean-centering. Then I read this post: https://stats.stackexchange.com/questions/95404/is-centering-a-valid-solution-for-multicollinearity What are the possible solutions to actually address multicolinearity in the context of *multiple-moderator* meta-regression? (Would really appreciate there is R code for me to follow/adapt)
- Badges
- Science topic
- Similar topics
- Psychology