[R-meta] Bivariate generalized linear mixed model with {metafor}
I see. Thank you very much, Wolfgang. Btw, sorry for my?persistence! This subject *was* a little bit confusing until you elucidated everything. Best, Arthur M. Albuquerque Medical student Universidade Federal do Rio de Janeiro, Brazil
On Mar 9, 2022, 7:59 PM -0300, Viechtbauer, Wolfgang (SP) <wolfgang.viechtbauer at maastrichtuniversity.nl>, wrote:
Which correlation are you interested in? And are you even interested in the correlation? If not, it doesn't really matter then. Best, Wolfgang
-----Original Message-----
From: Arthur Albuquerque [mailto:arthurcsirio at gmail.com]
Sent: Wednesday, 09 March, 2022 23:51
To: r-sig-meta-analysis at r-project.org; Viechtbauer, Wolfgang (SP)
Subject: RE: [R-meta] Bivariate generalized linear mixed model with {metafor}
Wow, crazy numerical coincidence then.
I?ve been wondering about applying Reference [1] model (= your Model 6) in future
projects. Can you see any practical reason to apply the?(group | study) syntax
instead of (control+treat-1|study)?
Best,
Arthur M. Albuquerque
Medical student
Universidade Federal do Rio de Janeiro, Brazil
On Mar 9, 2022, 7:47 PM -0300, Viechtbauer, Wolfgang (SP)
<wolfgang.viechtbauer at maastrichtuniversity.nl>, wrote:
Different parameterizations of the same model.
Also, the correlations seems like they just flipped signs, but they are really
different things and I suspect it's just coincidence that they happen to be so
close in absolute value.
With (group | study), you have a random intercept (for the control group logit
risk) and a random slope for the group/treatment effect (for the log odds ratio).
With (control+treat-1|study), you have random effects for the control and
treatment group logit risks. This is the same as (0 + group | study).
So really different things are being correlated here. But in the end, it's the
same model, parameterized in different ways.
Best,
Wolfgang