metafor package and bootstrapping

Question

I have issue combining metafor and boot package. My original meta-analysis model is following and this runs well:

Lu_i<- rma(yi=Hedges_g,vi=Vg, dat=intensive,  mods=~Landscape:Contrast-1, method="REML", random=list(~1|Study,~1|id))

I am using “boot” package for obtain confidence intervals. So I run following function for non-parametric bootstrapping:

boot.func <- function(dat, indices) {
    res <- try(rma(yi=Hedges_g,vi=Vg, data=intensive, mods=~Landscape:Contrast-1,                 method="REML",random=list(~1|Study,~1|id),subset=indices), silent=TRUE)
    if (is.element("try-error", class(res))) {
    NA
  } else {
    c(coef(res), vcov(res), res$tau2, res$se.tau2^2)
  }
  }

However, following command gives error message and do not perform full bootstrapping:

res.boot2 <- boot(intensive, boot.func, R=5000)

Error in boot(intensive, boot.func, R = 5000) : number of items to replace is not a multiple of replacement length In addition: There were 50 or more warnings (use warnings() to see the first 50)

Any tips for help? I think, my code should be correct, because it runs with bigger dataset and also for smaller dataset, but unfortunately not for this subgroup analysis, which I need. Actually, the code runs with very small bootstrapping number (maximum R=32), but with bigger number, I get soon error message, although, the code runs a while. Any suggestions?

Answer 1

First of all, the rma() function does not have a random argument. I think you meant to use the rma.mv() function.

If you switch to rma.mv(), then the elements you want to extract are:

c(coef(res), vcov(res), res$sigma2)

There is no res$se.sigma2 (at least not at the moment).

Also note that for models with moderators vcov(res) will be a matrix. This will get coerced to a vector which is fine, but just keep that in mind.