[R-meta] Meta-analysis with proportion and mean data

[Wolfgang Viechtbauer]

Please keep the mailing list in cc.

What you describe is a 'phylogenetic meta-analysis'. See, for example:

Adams, D. C. (2008). Phylogenetic meta-analysis. Evolution, 62(3), 567-572.

Lajeunesse, M. J. (2009). Meta-analysis and the comparative phylogenetic 
method. The American Naturalist, 174(3), 369-381.

Nakagawa, S., & Santos, E. S. A. (2012). Methodological issues and 
advances in biological meta-analysis. Evolutionary Ecology, 26(5), 
1253-1274.

And yes, that can be done with the rma.mv() function. And yes, you would 
need to have a phylogeny for the species that you are dealing with, 
which in turn can be turned into a covariance/correlation matrix under 
certain assumptions (e.g., a Brownian motion model for trait evolution).

Reading help(rma.mv) is a start. I do not currently have a nice dataset 
to illustrate this, but you can find an example (a rather complicated 
one with two phylogenies) described here:

http://www.metafor-project.org/doku.php/tips:speeding_up_model_fitting#complex_models

You can simplify this to a single phylogeny.

The 'ape' package allows you to work with phylogenies (and can convert 
them to correlation matrices).

Best,
Wolfgang

-- 
Wolfgang Viechtbauer, Ph.D., Statistician | Department of Psychiatry and
Neuropsychology | Maastricht University | P.O. Box 616 (VIJV1) | 6200 MD
Maastricht, The Netherlands | +31 (43) 388-4170 | http://www.wvbauer.com

On 08/29/2017 07:51 PM, M West wrote:
Hi Wolfgang,

Sorry, I think I wrote a very dumb question earlier - information 
overload and I tried to overly complicate the data. So I figured that 
question out. However, I do have another, hopefully, more sensible 
question: I read recently (citation below) that a unique problem for 
biological meta-analysis is that they often compare across many 
different species, and therefore, should correct for or at least account 
for the potential that phylogenetic relatedness could drive the observed 
pattern of interest in their particular analyses. I understand that 
these corrections are available in metafor - but I'm not having luck on 
finding any more detailed information on this topic. I.e., does this 
mean that in actuality, biological meta-analyses need to also know 
detailed (or just rough) phylogenetic relationships between the 
different species in their study (i.e., if the meta-analysis contains 
data with three different species of Drosophila, and monkeys, and sheep, 
how does one go about 'correcting for this'?). This particular 
recommendation seems valid, but a bit overwhelming considering that I do 
not work with phylogenies! Do I need to collaborate with someone who 
does so they can provide relatedness values for the data in the 
meta-analysis?

Many thanks for your help!

Meta-evaluation of meta-analysis: ten appraisal questions for biologists 
2017 Nakawaga et al.

On Tue, Aug 29, 2017 at 9:14 AM, Viechtbauer Wolfgang (SP) 
<wolfgang.viechtbauer at maastrichtuniversity.nl> wrote:
Dear M,

I am not sure I fully understand. My interpretation of what you describe 
is this. The first group reports:

100 * (mean_post_treatment - mean_pre_treatment) / mean_pre_treatment
100 * (mean_post_control - mean_pre_control) / mean_pre_control

or

100 * (mean_post_treatment - mean_pre_treatment) / mean_pre_treatment - 
100 * (mean_post_control - mean_pre_control) / mean_pre_control

while the second group reports:

mean_post_treatment
mean_post_control

or

mean_post_treatment - mean_post_control

Is that correct? If not, please write our in some form what exactly is 
reported.

Best,
Wolfgang

--
Wolfgang Viechtbauer, Ph.D., Statistician | Department of Psychiatry and
Neuropsychology | Maastricht University | P.O. Box 616 (VIJV1) | 6200 MD
Maastricht, The Netherlands | +31 (43) 388-4170 | http://www.wvbauer.com

-----Original Message-----
From: R-sig-meta-analysis 
[mailto:r-sig-meta-analysis-bounces at r-project.org] On Behalf Of M West
Sent: Monday, August 28, 2017 23:09
To: r-sig-meta-analysis at r-project.org
Subject: [R-meta] Meta-analysis with proportion and mean data

Hello,

I am gathering data for a meta-analysis. So far, most of the studies report
the results as the percentage reduction in the focal trait (i.e., percent
weight loss) in the control group vs. the percentage reduction in the focal
trait (percent weight loss) in the treatment group (and the SE). However, a
small subset of studies report the actual means of the control (i.e., mean
weight +/- SE) and the treatment (mean weight +/- SE).

My question is, should I convert the smaller subset into proportions so
that all of the data are the same (or v.v.,)? i.e., is that ok to do (i.e.,
since I am using other people's data)? Or would it be better to analyze the
data separately? (my apologies, I know this seems like a rather silly
question, but I feel a bit stumped).

Many thanks in advance for your helpful (and patient) answers!

Best,
~M