[R-meta] measure="CVR" in escalc

Sean Bloszies @@blo@zi @ending from nc@u@edu
Tue May 22 19:52:38 CEST 2018

Terrific, thanks a lot Wolfgang! I see the option for CVRC now.

The ri's are simply covariance(group1, group2) / (s_1 *s _2), right?

And do you have any background or reference for this (or will that be in
the help(escalc)?

Thanks again

On Tue, May 22, 2018 at 11:29 AM, Viechtbauer, Wolfgang (SP) <
wolfgang.viechtbauer at maastrichtuniversity.nl> wrote:

> Hi Sean,
> If you install the 'devel' version:
> https://wviechtb.github.io/metafor/#installation
> then escalc() also has measure="CVRC" for the coefficient of variation
> ratio for pre-post or matched designs (that is, y = logCV_1 - logCV_2 for
> matched groups 1 and 2 or time 1 and time 2). Note that computation of the
> sampling variance then requires knowing the correlation beween the
> measurements for matched pairs in groups 1 and 2 or at time 1 and time 2.
> I just realized that I forgot to document measure "CVRC" in help(escalc).
> Will do so asap.
> Best,
> Wolfgang
> -----Original Message-----
> From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-
> bounces at r-project.org] On Behalf Of Sean Bloszies
> Sent: Tuesday, 22 May, 2018 17:16
> To: r-sig-meta-analysis at r-project.org
> Subject: [R-meta] measure="CVR" in escalc
> Hi all,
> Escalc can give you an effect size that is the difference in CV's between
> two groups if you tell it measure="CVR". The calculations for difference
> and sampling variance in lnCV's this is based on assumes independence
> between groups (Nakagawa et al 2015).  One example they use is male vs.
> female animals. What if instead the source data had been weights of mated
> pairs of male and female sampled over time? Or instead, in my case, if I
> have crop yield info for paired treatment and control field plots with CV's
> over time, measured repeatedly on the same adjacent two plots year after
> year.  Do I need to take into account lack of independence between
> treatment and control by modeling this covariance somehow? If so, how do I
> do that in R?
> If this lack of independence is a problem, would an even simpler t-test of
> the ratio of CV's with a null mean of 0 be more appropriate?
> Thanks for having me on the list.
> Sean
> *Reference: Nakagawa et al. , 2015. Meta-analysis of variation:
> **ecological and evolutionary applications and beyond. Methods Ecol Evol
> **6, 143–152. doi:10.1111/2041-210X.12309

	[[alternative HTML version deleted]]

More information about the R-sig-meta-analysis mailing list