[R-meta] rma.mv in metafor - model assumptions
Viechtbauer, Wolfgang (SP)
wo||g@ng@v|echtb@uer @end|ng |rom m@@@tr|chtun|ver@|ty@n|
Wed Mar 3 09:37:41 CET 2021
rma.mv() does not assume homogeneity of variance (more commonly referred to as homoscedasticity). One supplies the function with the (approximately) known sampling variances via argument 'V' (or possibly even an entire var-cov matrix) and those sampling variances typically differ across studies.
The types of models that one can fit with rma.mv() do assume normality of the sampling errors and of the random effects in the model. That does imply that the observed residuals should also be approximately normally distributed, but they are not expected to have the same standard errors. One can compute standardized residuals that should behave (approximately) like z-scores.
To some degree, one can check normality assumptions by examining the distributions of the estimates of the random effects (from ranef()) and of the residuals (from rstandard()).
What to do if normality assumptions are violated is difficult to say in general. That's a very broad question.
>From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces using r-project.org] On
>Behalf Of Jessie Cait
>Sent: Tuesday, 02 March, 2021 20:57
>To: r-sig-meta-analysis using r-project.org
>Subject: [R-meta] rma.mv in metafor - model assumptions
>I am currently using the *rma.mv* function in metafor and
>I was wondering if you could explain some of the basics of the function:
>Is *rma.mv* assuming homogeneity of variance and normal
>residuals? Or making other assumptions?
>And how does one check those assumptions are met?
>And if they are not met, how does one manipulate things so that they are
>met - at least to some extent? (e.g. via transformations, and/or via by
>using robust SEs)?
>Thank you for your help.
>All the best,
More information about the R-sig-meta-analysis