R help - Aug 2012 - "metafor" package, proportions: single groups wrt to a categorical dependent variable‏

Dear Dushanthi,

Please keep your e-mails on the R-Help list, where Michael has already given you
some excellent advice. As Michael already explained, metafor can handle
proportions, but does not have any specific functionality for categorical
variables with more than 2 levels (at the moment). So, if it is logical and
possible to do so, you could collapse the levels of the categorical outcome to 2
levels and then proceed with the proportions. To be specific, what you need is
the total sample size (let's call this variable ni) and the number of
"cases" (e.g., the number of people that improved) (let's call
this variable xi) for each study. For example:

xi <- c(12, 18, 6)
ni <- c(30, 70, 40)
res <- rma(measure="PR", xi=xi, ni=ni)
res
forest(res)

just to give you an idea. There are also various transformations that can be
applied to proportions and depending on your data, it may be more sensible to
work with one of these transformations. See "Proportions and
Transformations Thereof" under ?escalc.

Of course, by doing so, you lose some information. More advanced would be to use
an ordinal regression or a multinomial model.

Also, regarding your analysis plan -- while it is fine in principle, I hope you
are not planning to use the information obtained in this way to draw any
conclusions about the relative effectiveness of the two surgical procedures.
Based on what you wrote, it seems as if these are non-randomized studies
"with one or the other procedure applied." Therefore, any comparisons
you make would be purely observational and may be related to other variable that
confound any difference you may find.

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: Dushanthi Pinnaduwage [mailto:dushanthi at bell.net]
> Sent: Tuesday, July 31, 2012 16:40
> To: wvb at metafor-project.org; Dushanthi Pinnaduwage
> Subject: "metafor" package, proportions: single groups wrt to a
> categorical dependent variable?
> 
> Dear Dr. Wolfgang Viechtbauer,
> 
> I am using R version 2.15.0 and 'metafor' package version 1.6-0.
Can this
> version of the package handle proportions from
> a categorical dependent variable for single studies? I mean can the
> package functions handle more than 2 proportions ?
> I know it can handle binary variables.
> 
> If?it can handle?how do I set up my dataframe for the raw data from
> different studies? Also
> how do I give inputs, specially xi, mi (or ni) to the function escalc()?
> 
> Thanks,
> Dushanthi

Thanks very much for your response. Sorry for writing to you directly.
Hereafter, I'll email to the R-Help list. I agree with all your comments and
suggestions. My analysis will be descriptive type; purely observational as the
studies are non-randomized studies. I have sample sizes and counts for each
study. I'll first do the analysis taking 2 proportions. Then I'll try to
use ordinal regression or a multinomial model. This is my first time doing meta
analysis. I may have couple of questions later. If so , I will come back to this
friendly R-Help group again. Thanks very much to Michael and you again.
Dushanthi  > From:
wolfgang.viechtbauer@maastrichtuniversity.nl
> To: r-help@r-project.org; dushanthi@bell.net
> Date: Wed, 1 Aug 2012 10:58:49 +0200
> Subject: RE: "metafor" package, proportions: single groups wrt to
a categorical dependent variableþ
> 
> Dear Dushanthi,
> 
> Please keep your e-mails on the R-Help list, where Michael has already
given you some excellent advice. As Michael already explained, metafor can
handle proportions, but does not have any specific functionality for categorical
variables with more than 2 levels (at the moment). So, if it is logical and
possible to do so, you could collapse the levels of the categorical outcome to 2
levels and then proceed with the proportions. To be specific, what you need is
the total sample size (let's call this variable ni) and the number of
"cases" (e.g., the number of people that improved) (let's call
this variable xi) for each study. For example:
> 
> xi <- c(12, 18, 6)
> ni <- c(30, 70, 40)
> res <- rma(measure="PR", xi=xi, ni=ni)
> res
> forest(res)
> 
> just to give you an idea. There are also various transformations that can
be applied to proportions and depending on your data, it may be more sensible to
work with one of these transformations. See "Proportions and
Transformations Thereof" under ?escalc.
> 
> Of course, by doing so, you lose some information. More advanced would be
to use an ordinal regression or a multinomial model.
> 
> Also, regarding your analysis plan -- while it is fine in principle, I hope
you are not planning to use the information obtained in this way to draw any
conclusions about the relative effectiveness of the two surgical procedures.
Based on what you wrote, it seems as if these are non-randomized studies
"with one or the other procedure applied." Therefore, any comparisons
you make would be purely observational and may be related to other variable that
confound any difference you may find.
> 
> 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: Dushanthi Pinnaduwage [mailto:dushanthi@bell.net]
> > Sent: Tuesday, July 31, 2012 16:40
> > To: wvb@metafor-project.org; Dushanthi Pinnaduwage
> > Subject: "metafor" package, proportions: single groups wrt
to a
> > categorical dependent variableþ
> > 
> > Dear Dr. Wolfgang Viechtbauer,
> > 
> > I am using R version 2.15.0 and 'metafor' package version
1.6-0. Can this
> > version of the package handle proportions from
> > a categorical dependent variable for single studies? I mean can the
> > package functions handle more than 2 proportions ?
> > I know it can handle binary variables.
> > 
> > If it can handle how do I set up my dataframe for the raw data from
> > different studies? Also
> > how do I give inputs, specially xi, mi (or ni) to the function
escalc()?
> > 
> > Thanks,
> > Dushanthi
 		 	   		  
	[[alternative HTML version deleted]]