Bayesian varying coefficient models using PC priors

Abstract

Varying coefficient models arise naturally as a flexible extension of a simpler model where the effect of the covariate is constant. In this work, we present varying coefficient models in a unified way using the recently proposed framework of penalized complexity (PC) priors to build priors that allow proper shrinkage to the simpler model, avoiding overfitting. We illustrate their application in two spatial examples where varying coefficient models are relevant.