[R-meta] Quick question about multiple independent samples within one study

[Viechtbauer, Wolfgang (SP)]

Ok, so if I understand (and remember correctly), then you will have two rows for each level of 'id', one for the correlation cor(x1, y) and one for the correlation cor(x2, y). And now you also have studies with multiple (independent) samples. So, for example:

study  id   var1var2   yi
-------------------------
1      1    var1       .
1      1    var2       .
2      2    var1       .
2      2    var2       .
2      3    var1       .
2      3    var2       .
3      4    var1       .
3      4    var2       .

So, study 2 has 2 samples and hence 4 rows. That would actually be a model with an additional level beyond what Konstantopoulos describes. So you have studies, samples within studies, and then two estimates within samples. Then I would go with:

res <- rma.mv(yi, dat$V, mods = ~ var1var2 - 1, random = list(~ var1var2 | id, ~ 1 | study/id), data=dat$dat)

Best,
Wolfgang

>-----Original Message-----
>From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces using r-
>project.org] On Behalf Of Anna-Lena Schubert
>Sent: Thursday, 13 December, 2018 16:36
>To: Guido Schwarzer; r-sig-meta-analysis using r-project.org
>Subject: Re: [R-meta] Quick question about multiple independent samples
>within one study
>
>Thanks for your helpful responses! I will quickly add a follow-up
>question regarding implementation: Previously (before accounting for
>dependencies within studies), my code looked like this:
>
>        res <- rma.mv(yi, dat$V, mods = ~ var1var2 - 1, random = ~
>var1var2 | id, struct="UN", data=dat$dat)
>
>Here, I investigate whether the correlation with a criterion variable
>differs between two predictors. Now, I have some studies with independent
>samples (different age groups), while I have other studies with different
>criterion variables within the same sample (different measures). If I
>wanted to introduce both dependencies, I believe I could change the code
>as follows:
>
>        res <- rma.mv(yi, dat$V, mods = ~ var1var2 - 1, random = list(~
>var1var2 | id, ~sample | id, ~measure | id), data=dat$dat)
>
>Three concerns: First, is "sample | id" or "id | sample" appropriate?
>Second, metafor tells me I can only have to random factors ("Only up to
>two '~ inner | outer' formulas allowed in the 'random' argument"). Which
>should I drop or is there a workaround? Third, I'm somewhat worrying I
>might be overfitting. Is this a legit worry?
>
>Best, Anna-Lena
>
>Am 13.12.2018 um 16:01 schrieb Guido Schwarzer:
>Hi,
>
>I got this wrong (and gave a nice answer for a different question).
>
>See Wolfgang's email for the correct answer.
>
>Best wishes, Guido
>--
>Dr. Anna-Lena Schubert
>Postdoc at Section of Personality
>Heidelberg University - Institute of Psychology
>Hauptstraße 47-51
>D-69117 Heidelberg
>Germany
>Phone: +49 6221 54 7746
>Mail: anna-lena.schubert using psychologie.uni-heidelberg.de
>Web: http://www.psychologie.uni-heidelberg.de/ae/diff/diff/people-
>schubert.html