package metafor: specify weights?

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

package metafor: specify weights?

Anke Stein
Dear R-experts,
Dear Wolfgang,

Weighted model fitting in metafor uses the inverse of the study specific variances as weights.
I am wondering if it is possible to specify different weights.

In my meta-analysis, there are two types of studies with (intrinsic) differences in their range of sample sizes (which are used to calculate the variances of Fisher's z).
I would like to try normalizing the sample sizes within each set of the two study types and use these normalized sample sizes as weights.
Would that be possible with rma()? So far, I only found the option "weighted = TRUE/FALSE", but no possibility to specify which weights should be used.

Many thanks in advance,
Anke






--
__________________________________________________________
Anke Stein (Dipl.-Biol.)

Biodiversity, Macroecology & Conservation Biogeography
Head Prof. Dr. Holger Kreft
Georg-August University of Göttingen
Büsgenweg 2 | 37077 Göttingen | Germany

phone +49(0)551-39-13761
fax +49(0)551-39-3618
[hidden email]
http://www.uni-goettingen.de/biodiversity

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
Reply | Threaded
Open this post in threaded view
|

Re: package metafor: specify weights?

Wolfgang Viechtbauer-2
At the moment, there is no possibility of specifying the weights with the rma() function. While the main model fitting part could be easily adapted to incorporate user-specified weights, the problem comes in with all the additional statistics that can be computed based on a fitted model. How should the predict() function now work? What would be the definition of I^2 now? How would one generalize the influence and outlier statistics to that case? Just to give some examples.

Of course, I could leave out such things when the user has specified the weights, but then things also get confusing for the user. For example, it is already less than ideal that you can only use the trim and fill method with models that do not incorporate moderators. Nobody has (as of yet) generalized the trim and fill method to that case. But when there are too many "special cases", the package becomes unusable.

I did consider user-specified weights at one point, but it opened up so many cans of worms that I preferred to quickly put the lid pack on those cans. That item is written down in my to-do list, but to be honest, it is somewhere at the very end of that list.

If you are hesitant to combine the results from those two types of studies, what about simply using a moderator to distinguish the two groups?

Best,

Wolfgang

--  
Wolfgang Viechtbauer, Ph.D., Statistician  
Department of Psychiatry and Psychology  
School for Mental Health and Neuroscience  
Faculty of Health, Medicine, and Life Sciences  
Maastricht University, P.O. Box 616 (VIJV1)  
6200 MD Maastricht, The Netherlands  
+31 (43) 388-4170 | http://www.wvbauer.com   


> -----Original Message-----
> From: [hidden email] [mailto:[hidden email]]
> On Behalf Of Anke Stein
> Sent: Thursday, May 24, 2012 15:59
> To: [hidden email]
> Subject: [R] package metafor: specify weights?
>
> Dear R-experts,
> Dear Wolfgang,
>
> Weighted model fitting in metafor uses the inverse of the study specific
> variances as weights.
> I am wondering if it is possible to specify different weights.
>
> In my meta-analysis, there are two types of studies with (intrinsic)
> differences in their range of sample sizes (which are used to calculate
> the variances of Fisher's z).
> I would like to try normalizing the sample sizes within each set of the
> two study types and use these normalized sample sizes as weights.
> Would that be possible with rma()? So far, I only found the option
> "weighted = TRUE/FALSE", but no possibility to specify which weights
> should be used.
>
> Many thanks in advance,
> Anke
>
>
>
>
>
>
> --
> __________________________________________________________
> Anke Stein (Dipl.-Biol.)
>
> Biodiversity, Macroecology & Conservation Biogeography Head Prof. Dr.
> Holger Kreft Georg-August University of Göttingen Büsgenweg 2 | 37077
> Göttingen | Germany
>
> phone +49(0)551-39-13761
> fax +49(0)551-39-3618
> [hidden email]
> http://www.uni-goettingen.de/biodiversity
>
> ______________________________________________
> [hidden email] mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-
> guide.html
> and provide commented, minimal, self-contained, reproducible code.

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
Reply | Threaded
Open this post in threaded view
|

Re: package metafor: specify weights?

Anke Stein
Dear Wolfgang,

Thank you very much for the quick reply. I already assumed that it might
get too complicated,
so I will just stick to using a moderator (as you also suggested).

Best,
Anke



On 24.05.2012 17:05, Viechtbauer Wolfgang (STAT) wrote:

> At the moment, there is no possibility of specifying the weights with the rma() function. While the main model fitting part could be easily adapted to incorporate user-specified weights, the problem comes in with all the additional statistics that can be computed based on a fitted model. How should the predict() function now work? What would be the definition of I^2 now? How would one generalize the influence and outlier statistics to that case? Just to give some examples.
>
> Of course, I could leave out such things when the user has specified the weights, but then things also get confusing for the user. For example, it is already less than ideal that you can only use the trim and fill method with models that do not incorporate moderators. Nobody has (as of yet) generalized the trim and fill method to that case. But when there are too many "special cases", the package becomes unusable.
>
> I did consider user-specified weights at one point, but it opened up so many cans of worms that I preferred to quickly put the lid pack on those cans. That item is written down in my to-do list, but to be honest, it is somewhere at the very end of that list.
>
> If you are hesitant to combine the results from those two types of studies, what about simply using a moderator to distinguish the two groups?
>
> Best,
>
> Wolfgang
>
> --
> Wolfgang Viechtbauer, Ph.D., Statistician
> Department of Psychiatry and Psychology
> School for Mental Health and Neuroscience
> Faculty of Health, Medicine, and Life Sciences
> Maastricht University, P.O. Box 616 (VIJV1)
> 6200 MD Maastricht, The Netherlands
> +31 (43) 388-4170 | http://www.wvbauer.com
>
>
>> -----Original Message-----
>> From: [hidden email] [mailto:[hidden email]]
>> On Behalf Of Anke Stein
>> Sent: Thursday, May 24, 2012 15:59
>> To: [hidden email]
>> Subject: [R] package metafor: specify weights?
>>
>> Dear R-experts,
>> Dear Wolfgang,
>>
>> Weighted model fitting in metafor uses the inverse of the study specific
>> variances as weights.
>> I am wondering if it is possible to specify different weights.
>>
>> In my meta-analysis, there are two types of studies with (intrinsic)
>> differences in their range of sample sizes (which are used to calculate
>> the variances of Fisher's z).
>> I would like to try normalizing the sample sizes within each set of the
>> two study types and use these normalized sample sizes as weights.
>> Would that be possible with rma()? So far, I only found the option
>> "weighted = TRUE/FALSE", but no possibility to specify which weights
>> should be used.
>>
>> Many thanks in advance,
>> Anke
>>
>>
>>
>>
>>
>>
>> --
>> __________________________________________________________
>> Anke Stein (Dipl.-Biol.)
>>
>> Biodiversity, Macroecology&  Conservation Biogeography Head Prof. Dr.
>> Holger Kreft Georg-August University of Göttingen Büsgenweg 2 | 37077
>> Göttingen | Germany
>>
>> phone +49(0)551-39-13761
>> fax +49(0)551-39-3618
>> [hidden email]
>> http://www.uni-goettingen.de/biodiversity
>>
>> ______________________________________________
>> [hidden email] mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-help
>> PLEASE do read the posting guide http://www.R-project.org/posting-
>> guide.html
>> and provide commented, minimal, self-contained, reproducible code.

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.