[R-meta] Aggregating results from multiple reg. models in metafor
Carla Gomez Creutzberg
cgomezcre @ending from gm@il@com
Fri May 11 01:06:36 CEST 2018
I am trying to synthesize coefficients from regressions where the predictor
is a factor instead of a continuous variable (i.e. ANOVAs and ANCOVAs). The
coefficients discussed in this thread are form the *r* family but mine are
from the *d* family since they measure group differences. I was wondering
if anybody had any experience with d-type partial correlation coefficients?
So far, the only guidance I have found comes from this paper by Keef &
Keef S & Roberts LA, 2004. The meta-analysis of partial effect sizes. *British
Journal of Mathematical and Statistical Psychology*, 57: 97 – 129.
However I have not been able to adapt their methods to my case since I am
not clear on: how I can deal with studies that do not use the same
variables, the calculation of the sampling variance of the partial d-type
correlation coefficient and using a random rather than a fixed effects
model to summarize coefficients across studies .
*Carla Gómez Creutzberg*
PhD. Candidate - Tylianakis Lab
University of Canterbury - *Te Whare Wānanga o Waitaha*
Christchurch, New Zealand <http://www.tylianakislab.org/the-group.html>
cgomezcre at gmail.com
On Mon, May 7, 2018 at 11:02 PM, Viechtbauer, Wolfgang (SP) <
wolfgang.viechtbauer at maastrichtuniversity.nl> wrote:
> Indeed, the partial correlation coefficient is esp. advantageous, since we
> only need to know the test statistic (i.e., t-test) of the regression
> coefficient of interest, the sample size of the study, and the number of
> predictors in the regression model in order to compute the partial
> correlation coefficient and its sampling variance. See help(escalc) and see
> the section on "Partial and Semi-Partial Correlations".
> However, if the same variables are used across datasets, then there is no
> need to go to a standardized measure. One can just meta-analyze the
> regression coefficients directly. You will also need the standard error of
> the coefficients, but hopefully those are also available (or they can be
> back-calculated if you have the CI for the coefficients, or the p-values,
> or the test statistics). The square of the standard errors are the sampling
> variances. You can then meta-analyze the coefficients with standard
> meta-analytic methods, that is:
> rma(coef, stderror^2, data=dat)
> where 'coef' is the variable in 'dat' giving the regression coefficients
> and 'stderror' is the variable with the corresponding standard errors.
> -----Original Message-----
> From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bo
> unces at r-project.org] On Behalf Of Jens Schüler
> Sent: Monday, 07 May, 2018 11:53
> To: j.briggs; r-sig-meta-analysis at r-project.org
> Subject: Re: [R-meta] Aggregating results from multiple reg. models in
> Hi Jeffrey,
> I am not sure if this exactly what you are looking for but I came across
> practice of using partial correlations in a number of published
> meta-analysis in management journals e.g.
> Marano, V., Arregle, J. L., Hitt, M. A., Spadafora, E., & van Essen, M.
> (2016). Home country institutions and the internationalization-performance
> relationship: A meta-analytic review. Journal of Management, 42(5),
> If I have understood the practice correctly, you have to calculate partial
> correlation coefficients from the regression estimates. The "challenge"
> though is that the regression models, from which the correlations are to be
> calculated from, have to be similar or the results are prone to bias.
> Aloe published a couple of papers on that:
> Aloe, A. M., & Thompson, C. G. (2013). The synthesis of partial effect
> sizes. Journal of the Society for Social Work and Research, 4(4), 390-405.
> Aloe, A. M. (2014). An empirical investigation of partial effect sizes in
> meta-analysis of correlational data. The Journal of general psychology,
> 141(1), 47-64.
> -----Ursprüngliche Nachricht-----
> Von: R-sig-meta-analysis <r-sig-meta-analysis-bounces at r-project.org> Im
> Auftrag von j.briggs
> Gesendet: Montag, 7. Mai 2018 10:04
> An: r-sig-meta-analysis at r-project.org
> Betreff: [R-meta] Aggregating results from multiple reg. models in metafor
> Dear all,
> I would like to aggregate results from different multiple regression models
> (same variables different datasets/studies). After reading Lipsey and
> (2001, p. 67) this seems somewhat more complex. I was wondering about the
> most suitable way to do this using the metafor package?
> R-sig-meta-analysis mailing list
> R-sig-meta-analysis at r-project.org
[[alternative HTML version deleted]]
More information about the R-sig-meta-analysis