Skip to content
Prev 250 / 5632 Next

[R-meta] Comparing estimates of independent subgroups

Roger,

I think the issue is in the rma.mv syntax. The model that you fit is not
equivalent to the two separate models---it has only three variance
components, rather than four. To allow for separate variances by allocation
at both the publication level and the trial level, the publication-level
random effects and trial-level random effects need to be specified as
separate terms:

(rma.mv(yi, vi, mods = ~ alloc,
        random = list(~ alloc | publication, ~ alloc | trial),
        struct= c("DIAG","DIAG"),
        data=dat, digits=3))

I hadn't realized before that the struct argument can take a vector with
one character string for each element in the random list. It looks like the
arguments will be recycled, so specifying struct = "DIAG" will give an
equivalent result.

Cheers,
James


On Mon, Oct 2, 2017 at 9:42 AM, Martineau, Roger <Roger.Martineau at agr.gc.ca>
wrote: