[R-meta] zero-order bivariate correlations and partial correlations
Viechtbauer, Wolfgang (SP)
wo||g@ng@v|echtb@uer @end|ng |rom m@@@tr|chtun|ver@|ty@n|
Thu Feb 20 10:38:03 CET 2020
Opinions are likely to differ on this one. I would personally be okay with that. They are both correlations, so they both measure the degree of association on the same scale. Of course, partial correlations reflect the association when 'adjusting for' one or multiple other variables, so in that sense, they are different. I would therefore code a variable that reflects the correlation type and then examine if this is an important moderator.
Note that the sampling variances are calculated in slightly different ways for zero-order and partial correlations (see the work by Ariel Aloe).
One can (and in my opinion should) apply the r-to-z variance-stabilizing transformation for both types of correlations (luckily, the transformation is the same for both types). But again, note that the sampling variance for an r-to-z transformed partial correlation is computed in a slightly different way than for zero-order correlations. The escalc() function in metafor has measures "PCOR" and "ZPCOR" for these purposes.
From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces using r-project.org] On Behalf Of Yingkai
Sent: Thursday, 20 February, 2020 10:00
To: r-sig-meta-analysis using r-project.org
Subject: [R-meta] zero-order bivariate correlations and partial correlations
Greetings! I have one question about correlations. When doing
meta-analysis using correlations, can we include both zero-order bivariate
correlations and partial correlations? Or one can just use one of them but
not mix them together?
Yingkai, Ph.D Candidate
Southwest University, ChongQing, CHINA.
More information about the R-sig-meta-analysis