[R-meta] Fwd: SV: Re:: Question: Metafor R package - rma and rma.mv function

[Viechtbauer, Wolfgang (SP)]

Hi Simona,

sigma^2, tau^2, gamma^2, or chicken^2 are just symbols that denote certain parameters (in this case, variance components). In some cases, they may denote the same thing. For example:

dat <- get(data(dat.bcg))
dat <- escalc(measure="RR", ai=tpos, bi=tneg, ci=cpos, di=cneg, data=dat.bcg)

res1 <- rma(yi, vi, data=dat)
res1

res2 <- rma.mv(yi, vi, random = ~ 1 | trial, data=dat)
res2

Here, tau^2 in res1 is the same as sigma^2 in res2. 

But one cannot say this in general. Let's say I fit this model:

res3 <- rma.mv(yi, vi, random = ~ 1 | alloc/trial, data=dat)
res3

Now there are two sigma^2 values, neither of which corresponds to tau^2 from res1.

Best,
Wolfgang

-----Original Message-----
From: Simona Frederiksen [mailto:simona.frederiksen using hotmail.com] 
Sent: Monday, 06 August, 2018 16:58
To: Viechtbauer, Wolfgang (SP); r-sig-meta-analysis using r-project.org
Subject: SV: [R-meta] Fwd: SV: Re:: Question: Metafor R package - rma and rma.mv function

Thank you for your help. I just have one final question. On some forums they write that sigma estimated from more complex models are the same as tau. Is it so?
 
Best wishes,
Simona
 
________________________________________
Fra: Viechtbauer, Wolfgang (SP) <wolfgang.viechtbauer using maastrichtuniversity.nl>
Sendt: Friday, August 3, 2018 4:01:43 PM
Til: Simona Frederiksen; r-sig-meta-analysis using r-project.org
Emne: RE: [R-meta] Fwd: SV: Re:: Question: Metafor R package - rma and rma.mv function 
 
Hi Simona,

No, I would not drop est_id if it is non-significant.

The weights depend on the model fitted. By default, the weights, or rather, the weight matrix, is the inverse of the model-implied marginal var-cov matrix of the estimates. For a bit of a discussion of this, see:

https://stat.ethz.ch/pipermail/r-sig-meta-analysis/2018-July/000909.html

For the three-level model described here:

http://www.metafor-project.org/doku.php/analyses:konstantopoulos2011

the marginal var-cov matrix of the estimates consists of blocks, where the size of each block depends on the number of estimates in the cluster. The off-diagonal elements in each block are equal to the cluster-level variance component. The diagonal elements are equal to the sampling variances plus the estimate-level variance component plus the cluster-level variance component. So, for example, a cluster with 3 estimates would look like this:

[v_1 + s^2_id + s^2_study  s^2_study                 s^2_study               ]
[                          v_2 + s^2_id + s^2_study  s^2_study               ]
[                                                    v_3 + s^2_id + s^2_study]

The inverse of this block-diagonal is the weight matrix. By default, weights() only shows the diagonal of this weight matrix. Use weights(model, type="matrix") to see the full matrix.

Best,
Wolfgang

-----Original Message-----
From: Simona Frederiksen [mailto:simona.frederiksen using hotmail.com] 
Sent: Wednesday, 01 August, 2018 13:16
To: Viechtbauer, Wolfgang (SP); r-sig-meta-analysis using r-project.org
Subject: SV: [R-meta] Fwd: SV: Re:: Question: Metafor R package - rma and rma.mv function

Dear Wolfgang,
 
Thank you very much for your reply and the link. I just came back from Scotland. 
 
I read through the link and I am wondering if I should skip est_id if I find it non-significant? 
 
Regarding the weights, I'am not quite aware of how this is computed since 1/vi does not result in the weights that I can extract from weights(res). The model that I am fitting is it as follows: yi = mean + uj[i] + ei for the i'th comparison and the j'th study when using the rma.mv function for the full model (=inclusion of est_id in addition to study id)? And how would it look if not including est_id? Finally, I am wondering how the weights are applied to the model and how it is calculated?
 
Best wishes,
Simona
 
________________________________________
Fra: Viechtbauer, Wolfgang (SP) <wolfgang.viechtbauer using maastrichtuniversity.nl>
Sendt: Monday, July 30, 2018 12:47:30 PM
Til: Simona Frederiksen; r-sig-meta-analysis using r-project.org
Emne: RE: [R-meta] Fwd: SV: Re:: Question: Metafor R package - rma and rma.mv function 
 
Dear Simona,

Did you see my reply?

https://stat.ethz.ch/pipermail/r-sig-meta-analysis/2018-July/000949.html

Best,
Wolfgang

-----Original Message-----
From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces using r-project.org] On Behalf Of Simona Frederiksen
Sent: Wednesday, 25 July, 2018 22:45
To: r-sig-meta-analysis using r-project.org
Subject: [R-meta] Fwd: SV: Re:: Question: Metafor R package - rma and rma.mv function

Dear Michael,

Thank you for your fast reply. It is very appreciated. I should have changed it to plain text now.

Maybe I'am not quite aware of how this is computed. The model that I am fitting is it as follows: yi = mean + uj[i] + ei for the i'th comparison and the j'th study when using the rma.mv function? And how is the weights applied to it or calculated?

I tried to use the profile.rma.mv() function and I got one maxima. Can I then assume that it is converged?

Best wishes,

Simona
________________________________
Fra: Michael Dewey <lists using dewey.myzen.co.uk<mailto:lists using dewey.myzen.co.uk>>
Sendt: Wednesday, July 25, 2018 3:06:06 PM
Til: Simona Frederiksen; r-sig-meta-analysis using r-project.org<mailto:r-sig-meta-analysis using r-project.org>
Emne: Re:: [R-meta] Question: Metafor R package - rma and rma.mv function

Simona Frederiksen <simona.frederiksen using hotmail.com<mailto:simona.frederiksen using hotmail.com>> wrote :

Dear Simona

Given the calculated variances (vi) the weights seem quite OK to me.

Would the question underlying the convergence issue be better answered by using profile.rma.mv()?

Please set your mailer to post in plain text not HTML as your post got mangled which made running your example rather tricky.

Michael.

> Hi,
>
> I have some question sthat I would like to get posted regarding the metafor R
> package and the rma and rma.mv function. It goes like this:
>
> Hi,
>
> I am at the moment working in the R package 'metafor' in order to perform a
> meta-analysis which I carry out in collaboration with the Danish Headache
> Center. I just figured that I have to use the rma.mv function since I have
> several effect sizes per study (for two studies).
>
> I have following questions that I hope you can help me answer:
>
>   1.  When I calculate the weights for each study, the studies that have more
> effect sizes receive a really high weight compared to the other studies with
> just one effect size. So it seems that these studies primarily are used to
> calculate the overall effect size even though N seems to be quite small?
>
>   1.  How can I see convergence when I use the rma.mv function? When using the
> rma function, it turns up when adding verbose = T. And if it does not converge,
> what would be ideal to do?
>
> Here is an example:
> library("metafor")
> study_id