Set half-normal distribution as μ prior in Bayesian meta-analysis

Modeling
, , , ,
1

Hi! I’m currently working on a meta analysis about the relationship between sleep and memory. Specifically, one aspect of my study involves analyzing the circular-linear correlation between a circular variable (EEG phase) and a linear variable (memory index). The data I will be using is about circular-linear partial r (distributed from 0 to 1) instead of the usual Pearson’s r. So after converting these values to Fisher’s z, the resulting distribution approximates a half-normal distribution rather than a normal distribution.

What I would like to ask is when I use a half-normally distributed data to build a Bayesian hierarchical model, would it be appropriate to set the μ prior to a half-normal distribution using the code below, or are additional steps necessary?

priors <- c(prior(“normal(0,1)”, class = "b", coef = "Intercept", lb = 0),
prior(“cauchy(0,0.5)”, class = “sd”))

I would greatly appreciate any insights or guidance on this question!

2

Howdy!

I’m not sure what your model is from the description, but the code you give will put a half-normal prior on the Intercept.

I’m not quite sure if you are simply asking if the code will work (it will give the half-normal prior) or if you are asking if a half-normal prior is appropriate for your use case. If the latter, just because you have an outcome that looks like half-normal, doesn’t necessarily mean that you need a half-normal prior. It depends on your model and the parameter that you are setting the prior on. What response family are you using? It would help to see your code, or even better, a reproducible example.