[R-meta] Addressing multicollinearity among moderators in a meta-regression (metafor package)

Michael Dewey ||@t@ @end|ng |rom dewey@myzen@co@uk
Wed Feb 6 13:09:22 CET 2019

Dear Hamed

I think there are a number of possible ways forward.

1 - put all the moderators in at once. I do not think collinearity is an 
important issues for the fitting software, these days methods are 
tolerant of what you throw at them. This approach would be appropriate 
if you wanted to take out the effect of the moderators and see what 
heterogeneity was left afterwards. I sense this is in fact not your goal.

2 - since theory suggests that these moderators are in fact manifest 
variables measuring some underlying latent construct then derive a score 
for tat latent variable either formally using PCA and friends or 
informally by adding them up.

3 - examine the possible models using subsets following the approach 
outlined in
I am not convinced that this answers your scientific question but worth 
a thought perhaps.


On 04/02/2019 19:51, Hamed Qahri-Saremi wrote:
> Greetings everyone;
> I hope my message finds you well.
> I am conducting a series of moderation tests using meta-regression in the
> metafor package. I have six numeric (continuous) moderators, which are
> different cultural dimensions. In my data, several of these six moderators
> are moderately to very highly correlated (0.40 - 0.80). As a result, to
> avoid any biases in meta-regression analyses due to multicollinearity, I am
> testing these moderators separately (i.e., six univariate meta-regressions)
> rather than including all of them in one multivariate meta-regression with
> six moderators. While this solution prevents the multicollinearity from
> biasing the results, it introduces a major limitation in my tests because
> essentially I am not controlling for the relations among moderators by
> testing them separately. Therefore, I was wondering if there is a better
> solution to rectify or reduce the effects of multicollinearity among
> moderators while running a multivariate meta-regression in metafor.
> I appreciate any advice you might have,
> Warm regards,
> Hamed
> Assistant Professor of Information Systems | College of Computing and
> Digital Media, DePaul University | SAP Certified Associate &
> Instructor | Chicago,
> IL 60604 | Phone: 312-362-5841 | Email: hamed.saremi using depaul.edu | Webpage:
> http://facsrv.cdm.depaul.edu/~hqahrisa/
> 	[[alternative HTML version deleted]]
> _______________________________________________
> R-sig-meta-analysis mailing list
> R-sig-meta-analysis using r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-meta-analysis


More information about the R-sig-meta-analysis mailing list