I'm working with Bayesian hierarchichal regressions fitted with R-INLA. I would like to simplify my model by reducing the number of covariates.

According to my understanding, Bayesian variable selection (spike & slab priors) cannot be done with R-INLA . I don't like the idea of forward/backward selection based on some information criteria (WAIC, DIC).

What approaches would you recommend for variable selection in this context? I'd appreciate it if you could cite your sources.

  • $\begingroup$ I solved this problem (not for R-INLA, but it was Gaussian Process as well) by chosing different tool for variable selection. I used R package Boruta for this purpose, along with simple GLM with backward selection. $\endgroup$
    – Tomas
    Jul 11 at 16:57
  • $\begingroup$ Thanks! I'll take a look at the Boruta method. But I'm trying to avoid stepwise selection... $\endgroup$
    – antarctica
    Jul 11 at 20:35
  • $\begingroup$ Projpred is an r library designed to help with this. Have you looked at that library? $\endgroup$ Jul 11 at 21:21
  • $\begingroup$ Thanks for the tip, projpred looks interesting. However, in this case I'm looking for methods that could be compatible with INLA. $\endgroup$
    – antarctica
    Jul 12 at 1:59

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.