Random efffect BLUP COV

[chfleming]

Classification:

Feature Request

Summary

I would find it useful in some applications to have the full covariance matrix from ranef(), rather than just the standard errors.

Am I correct in assuming that vpred here:

metafor/R/ranef.rma.mv.r

Line 178 in 6a53601

vpred <- G - (GW %*% (I - Hmat) %*% G)

is what I am looking for?

Thanks.

[wviechtb]

Yes, this is correct (vpred is the entire var-cov matrix and then a few lines further se <- sqrt(diag(vpred)) just takes the square-root of its diagonal elements to get the SEs). So, you could just hack your own function to extract that.

Adding this feature more generally is not unreasonable and in fact there is a bit of a precedent to this in predict.rma(), which has a vcov argument which, when set to TRUE, returns the var-cov matrix of the predicted values. ranef.rma.mv() is a bit more tricky though, because it can return multiple sets of BLUPs simultaneously.