[R-sig-ME] nlme package, fixing variance.covariance matrix of residuals

[Viechtbauer Wolfgang (SP)]

To address the original question directly: No, this isn't possible with nlme (you can only specify known variances, not an entire var-cox matrix).

Best,
Wolfgang

-----Original Message-----
From: R-sig-mixed-models [mailto:r-sig-mixed-models-bounces at r-project.org] On Behalf Of Phillip Alday
Sent: Thursday, August 17, 2017 22:51
To: Wölwer, Anna-Lena; r-sig-mixed-models at r-project.org
Subject: Re: [R-sig-ME] nlme package, fixing variance.covariance matrix of residuals

This is an indirect (non)-answer, but Wolfgang Viechtbauer has a nice
comparison of lme and rma in the metafor online documentation:

http://www.metafor-project.org/doku.php/tips:rma_vs_lm_lme_lmer

It might be possible to figure out your problem from that -- and if you
do, please 'self-answer' on the list to document it for other users!

Best,
Phillip

On 08/17/2017 07:09 PM, Wölwer, Anna-Lena wrote:
> Dear R team,
>
> I would like to do a multivariate meta-analysis in R using the nlme package. In meta-analysis I fix the residuals to known sampling errors. As I do a multivariate analysis, I have a variance-covariance matrix of sampling errors. Unfortunately, via varFixed I can only fix a vector of sampling errors and no matrix.
>
> In the R package metafor using the rma.mv function I can insert the variance-covariance matrix in the function. I would like to do the same with lme from the nlme package as nlme provides more flexibility for my model.
>
> Do you now any possibility of doing so in nlme? Are there any changes planned to incorporate this feature?
>
> Best wishes, I am looking forward to your feedback
> Anna-Lena W�lwer