Dirichlet prior on ordinal regression cutpoints in brms

Interfaces brms
21

Just some quick guy guessing the night before Christmas Eve (the big day here in Sweden):

<guy guessing>
I think it is a effect of the prior having a strong diffusing effect when you have many thresholds and little data. As cyl is coded as a numeric we will have thresholds between every positive integer up to the maximum. The default t-distribution prior is easily dominated by the likelihood in comparison.
</guy guessing>

1 Like
22

Thank you for your quick thoughts. At least it doesn’t appear that I did anything grossly wrong, which was my initial concern. I’ll keep chipping away.

1 Like