1
$\begingroup$

I am currently conducting a three-level meta-analysis and I would like to run the meta-analysis also by using a Bayesian approach. To run a meta-analysis I collected effect sizes that are correlation coefficients of the 2 variable of my interest. Since I don't have a clear assumption of the prior I should rely on, I would like to start by using a uniform distribution as my prior. To do that I need to restrict the uniform distribution's boundaries from -1 to 1 (again, due to the correlation coefficients).

Here is what I did so far: I ran brm model in R and specified the prior as the following:

brm <- brm(yi | se(sei) ~ 1 + (1|Study_ID), prior = set_prior("uniform(-1,1)", lb = -1, ub = 1), data = df, iter = 5000, warmup = 2000, cores = 4)

I received the following error message: Error: The following priors do not correspond to any model parameter: b ~ uniform. Function 'get_prior' might be helpful to you.

A further issue I am struggling with is that it is not really clear to me how the random part of the model is to be defined: In my model, I have to allow variation of both a unique study identifier and an effect size identifier to allow random intercept for the sampling variance, the variance between multiple effect sizes extracted from the same study and the variance between studies.

My questions are:

  1. What is it that I am doing wrong with the specification of the prior.
  2. Where and how do I add the random intercept for the effect size unique ID?

I hope I have made my problems clear so I can get some informative insights on them. I would appreciate any kind of help.

Thank you!!

$\endgroup$
  • $\begingroup$ You might get a better answer to your first question on an R site as debugging your code is off-topic here. You only seem to be specifying one random effect but I am not familiar with Bayesian models for regression. $\endgroup$ – mdewey Jul 13 '18 at 12:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.