[R-meta] Random-effect specification in rma.mv() for multiple sources?

[Timothy MacKenzie]

Dear Michael,

Very true, thanks very much. But my question is that:  Using "rma.mv()",
would the below random-effect structure theoretically account for all the
potential sources of variation (i.e., study > sample > outcome > time >
control > obs) or I'm missing something in the syntax?

random = list(~ sample | study, ~ time | interaction(study,sample,outcome),
~ 1 | control, ~ 1 | obs)

Thanks, Tim

study  sample  outcome time  control obs
1      1       1       1     1       1
1      1       2       1     1       2
1      1       1       2     1       3
1      1       2       2     1       4
1      2       1       1     1       5
1      2       2       1     1       6
1      2       1       2     1       7
1      2       2       2     1       8
2      1       1       2     1       9
2      1       2       2     1       10
2      1       1       2     2       11
2      1       2       2     2       12
3      1       1       3     1       13
3      1       1       3     2       14
3      2       1       3     1       15
3      2       1       3     2       16

random = list(~ sample | study, ~ time | interaction(study,sample,outcome),
~ 1 | control, ~ 1 | obs)

On Tue, Jun 29, 2021 at 4:56 AM Michael Dewey <lists using dewey.myzen.co.uk>
wrote:

> Dear Tim
>
> I will leave it to the experts to check your structure but one thing
> which immediately strikes me is that you are going to need a very large
> dataset to be able to estimate all those random effects with any
> precision especially the ones with limited replicates. If you do get the
> model to converge it would be mandatory to look at diagnostics like the
> profile likelihoods.
>
> Michael
>
> On 29/06/2021 03:39, Timothy MacKenzie wrote:
> > Dear all,
> >
> > I noticed some errors in the copy-pasted data structure in my previous
> post
> > (
> https://stat.ethz.ch/pipermail/r-sig-meta-analysis/2021-June/002953.html).
> > Below is my correct data structure. From left to right, one can see the
> > hierachical structure in my dataset:   study > sample > outcome > time >
> > control > obs
> >
> > Q:  Would the following random-effect structure account for all the above
> > sources (as a first step to then drop the ones that are insignificant)?
> >
> > random = list(~ sample | study, ~ time |
> interaction(study,sample,outcome),
> > ~ 1 | control, ~ 1 | obs)
> >
> > Thanks, Tim
> >
> > study  sample  outcome time  control obs
> > 1      1       1       1     1       1
> > 1      1       2       1     1       2
> > 1      1       1       2     1       3
> > 1      1       2       2     1       4
> > 1      2       1       1     1       5
> > 1      2       2       1     1       6
> > 1      2       1       2     1       7
> > 1      2       2       2     1       8
> > 2      1       1       2     1       9
> > 2      1       2       2     1       10
> > 2      1       1       2     2       11
> > 2      1       2       2     2       12
> > 3      1       1       3     1       13
> > 3      1       1       3     2       14
> > 3      2       1       3     1       15
> > 3      2       1       3     2       16
> >
> >       [[alternative HTML version deleted]]
> >
> > _______________________________________________
> > R-sig-meta-analysis mailing list
> > R-sig-meta-analysis using r-project.org
> > https://stat.ethz.ch/mailman/listinfo/r-sig-meta-analysis
> >
>
> --
> Michael
> http://www.dewey.myzen.co.uk/home.html
>

	[[alternative HTML version deleted]]