Bayesian Logistic Regression Priors - Bayes Factors
I am running a Bayesian mixed-model logistic regression.
As per advice from https://github.com/stan-dev/stan/wiki/Prior-Choice-Recommendations I have set my coefficient priors for main effects and interaction effects to student_t(4,0,2.5).
I wanted to check if this was an appropriate prior for the calculation of Bayes Factors for model comparison as the BFs I am getting seem somewhat implausible from looking at the data and doing equivalent frequentist tests.
Happy to share more info if needed.
Comments
Hi Whirly123,
My guess is that these priors are possibly too wide (usually, testing priors need to be more constrained than estimation priors). One way to judge whether or not the predictions are too spread out is to generate them from the prior (Andrew mentions this as Good's method of imaginary results).
Cheers,
E.J.
Resurrecting an old thread but if someone happens to find this through a search and wanted help with the same issue, I found this appendix page: https://doi.org/10.5334/joc.86.s1 from this paper https://www.journalofcognition.org/article/10.5334/joc.86/ that has some very useful info.