Bayesian interaction analysis (subgroups)

General
#1

Hi all,

any suggestions how to do interaction analysis the Bayesian way (e.g. for survival models).

Sketch of potential frequentist approach: For a (large) number of potentially interacting variables, fit separate survival models that have an interaction between treatment variable and the corresponding interacting variable. After this is done, using some correction for multiple hypothesis tests, look at the most significant interacting variables (i.e. those that that “survive” the threshold).

Would be happy if you could share any suggestions, experiences or literature about

  • how to do a single interaction analysis (e.g. prior choice for the interaction term and variables that interact)

  • how to check for mulltiple interactions in Bayesian way?

Many thanks!

#2

The parallel to your frequentist analysis could be to fit all the plausible interactions as a single random effect. You could do better using the horseshoe but if you’re looking to identify interactions rather than quantify them precisely it’s a great first pass.

3 Likes