[R-meta] Estimating partial correlation coefficients from multi-level regression models

[Vainius Bartasevičius]

Dear Wolfgang,

Many thanks for your response. Yes, this issue does seem to have little influence on our summary effect sizes and meta-regression models anyway. We therefore largely follow the 2) option, although we might also try running identical analyses excluding PCCs that come from multi-level models as a kind of sensitivity analysis.

Best regards
Vainius
________________________________
From: Viechtbauer, Wolfgang (NP) <wolfgang.viechtbauer using maastrichtuniversity.nl>
Sent: 19 February 2024 11:35
To: R Special Interest Group for Meta-Analysis <r-sig-meta-analysis using r-project.org>
Cc: Vainius Bartasevi�ius <vainius.bartasevicius using tspmi.vu.lt>
Subject: RE: Estimating partial correlation coefficients from multi-level regression models

[You don't often get email from wolfgang.viechtbauer using maastrichtuniversity.nl. Learn why this is important at https://aka.ms/LearnAboutSenderIdentification ]

Dear Vainius,

I suspect you are not getting any responses to your post (and neither on CV: https://stats.stackexchange.com/questions/637996/meta-analysis-estimating-partial-correlation-coefficients-using-estimates-from) because as far as I know, nobody has ever looked into this issue.

So I would say you have two options:

1) Exclude the PCCs from multilevel analyses.
2) Include them and transparently explain how you set the dfs.

Note that ni is probably relatively large compared to the number of predictors however you count them, so this probably has little influence on the results in the first place.

Best,
Wolfgang

> -----Original Message-----
> From: R-sig-meta-analysis <r-sig-meta-analysis-bounces using r-project.org> On Behalf
> Of Vainius Bartasevicius via R-sig-meta-analysis
> Sent: Thursday, January 25, 2024 16:31
> To: r-sig-meta-analysis using r-project.org
> Cc: Vainius Bartasevi�ius <vainius.bartasevicius using tspmi.vu.lt>
> Subject: [R-meta] Estimating partial correlation coefficients from multi-level
> regression models
>
> Dear All,
>
> I am writing to you with an inquiry on the estimation of partial correlation
> coefficients (PCCs) using escalc function of the metafor package.
>
> Currently we are conducting a meta-analysis which draws on data from multiple
> regression models and uses partial correlation coefficient as an effect size.
> Some of the models included in our meta-analysis come from multi-level analyses.
> Our predictors of interest are at level 1, so is the outcome measure.
>
> We are a bit unsure about the correct way to estimate PCCs using estimates from
> multi-level models, given that the calculation of degrees of freedom in multi-
> level models is different from that applied in single-level regression. In
> particular, we are wondering what figure we should provide for mi argument
> (total number of predictors) when estimating PCCs from multi-level models.
> Should we take the sum of level 1 and level 2 predictors? Should we take into
> account the number of level 2 units when specifying this argument?
>
> Any help would be greatly appreciated - thank you very much for your time.
>
> Kind regards
> Vainius Bartasevi�ius

	[[alternative HTML version deleted]]