I'm using metafor and the rma.mv function to conduct a multi-level random effects meta-analysis. I've used the vcalc function to impute the covariance matrix for cluster-robust estimation.
rho <- 0.6
V <- vcalc(vi = df$var_g,
cluster = df$study,
obs = df$es_id,
rho = rho)
res <- rma.mv(yi = g,
V = V,
random = ~ 1 | study / measure,
data = df,
method = "REML",
test = "t")
summary(res)
Model Results:
estimate se tval df pval ci.lb ci.ub
0.2105 0.3563 0.5908 7 0.5732 -0.6320 1.0530
However, given my dataset, it is desirable to apply a small sample correction, which I understand can be achieved by using vcov = "CR2" within the clubSandwich package. This correction slightly alters the estimate and the 95%CI's.
robust_CIs <- conf_int(res, vcov = "CR2")
print(robust_CIs)
Coef. Estimate SE d.f. Lower 95% CI Upper 95% CI
intrcpt 0.211 0.355 4.98 -0.704 1.13
My question is: since the forest function in metafor uses the res object, how can I instead use the estimate and 95%CI's with the small sample correction from clubSandwich's conf.int in the forest plot?
forest(res,
slab = df$descriptor,
addpred = TRUE)
This thread asks a similar problem but there isn't enough detail in the answer for me to understand. I can pull the estimate and CI's from conf.int but not sure how I integrate these into the forest plot.
R - Meta-Analysis - How to create a forest plot with robust estimates from clubSandwich functions
If you use the robust() function from metafor, then you can just pass the resulting object to the forest() function. So you should use:
res_robust <- robust(res, cluster=study, clubSandwich=TRUE)
and then forest(res_robust).