[R-meta] Simple Bayesian meta-analysis in R
Thanks for the reply Wolfgang! That's a good start to know that the model is correct. But yeah, the CI for equivalence is pretty much the same as with frequentist approach. So what would be more convincing (for reviewers..) is to show Bayes factors. On Sun, 6 Oct 2019 at 00:04, Viechtbauer, Wolfgang (SP) <
wolfgang.viechtbauer at maastrichtuniversity.nl> wrote:
Dear Gaspar,
Yes, your model code is correct.
I don't know how to extract a Bayes factor here, but you can in essence
examine the equivalence by looking at the coefficient corresponding to the
'version' dummy variable (this should be called something like
'beta_version' in the output). This coefficient represents the *difference*
between the two versions, so if it is 0, then there is no difference.
Obviously, that difference will never be exactly 0, so you can look at the
2.5% and 97.5% percentiles of the posterior distribution of that
coefficient. You can be 95% certain that the difference falls within these
bounds.
Best,
Wolfgang
-----Original Message-----
From: R-sig-meta-analysis [mailto:
r-sig-meta-analysis-bounces at r-project.org] On Behalf Of Gaspar Lukacs
Sent: Friday, 27 September, 2019 17:45
To: r-sig-meta-analysis at r-project.org
Subject: [R-meta] Simple Bayesian meta-analysis in R
Hi,
I'd like to conduct a Bayesian meta-analysis in R in order to support
equivalence between the effects of two design types (within each study).
E.g., Bayes factors would be perfect.
The metaBMA package seems to imply that I can get simple Bayes factors for
a moderator in a meta-analysis. (Most other Bayesian packages don't seem to
allow moderators.) But I don't find any info about how exactly to use it
and/or how to interpret the output.
Here is a simple example:
bayes_model = metaBMA::meta_random(
y = cohens_d~version,
SE = sed,
labels = study,
data = met_bf
)
"cohens_d" is the effect in each study, separately for each of the two
levels of "version", and I would simply want to know whether there is
substantial support for the equivalence between the "version" factors. Is
this model correct? And in any case, how can I get Bayes factors out of it
(or any direct proof of equivalence)?
A more detailed description of the question is here (where I was suggested
this mailing list):
https://stats.stackexchange.com/questions/428654/simple-bayesian-meta-analysis-in-r
I'd much appreciate any help.
Best,
Gaspar