[R-meta] Plotting rma.mv as forest plots for individual moderator levels

[Julian Packheiser]

Dear members of the mailings list,

Currently, we are conducting a multivariate multilevel meta-analysis with multiple outcomes coming from the same study. I use both study_id and effectsize_id for the calculation of V as well as as random effects in the model. We also added the laboratory in which the study was conducted as a random effect. The code thus looks like this:

V <- vcalc(vi, cluster=study_id, obs=esid, data=mydata, rho=0.6)

# random-effects model
res <- rma.mv(yi, V, mods = ~ outcome-1, random = ~ 1 | lab/study_id/esid, data=mydata)

My first question is, should we use a sensitivity approach for rho values as they are unknown?

My second question is about forest plots. Ideally, I would like to plot the results for each moderator separately. If I use forest.rma, I get all results in one plot. An overall result for this meta-analysis is however not meaningful since some outcomes are positively correlated and some are negatively correlated with the variable of interest.
I assume that I cannot simply do a forest plot of a rma.mv result for each outcome individually as the results obviously differ from a multivariate approach. Could I simply re-arrange my outcomes and only plot them ordered by outcome? I assume the plotted values are yi.f and vi.f for the means and variances, but I could not find values for the fitted polygons.

Apologies if this question has already been asked, but I could not find it in the archives.

Best
Julian

---
Julian Packheiser
Postdoc in the Social Brain Lab
Netherlands Institute for Neuroscience

	[[alternative HTML version deleted]]