Implement remaining recommendations from Freni-Sterrantino et al.'s "A note on intrinsic conditional autoregressive models for
disconnected graphs", Spatial and Spatio-Temporal Epidemiology:
give each connected component its own intercept---but avoiding assigning an intercept to islands;
for singletons/islands, replace standard normal prior with normal(0, tau) where tau is the ICAR scale parameter;
add option to scale the ICAR and BYM models (not only the BYM2 specification).
Also, could give each connected component its own scale by default since they're practically separate models, and it seems to improve sampling.
Implement remaining recommendations from Freni-Sterrantino et al.'s "A note on intrinsic conditional autoregressive models for disconnected graphs", Spatial and Spatio-Temporal Epidemiology:
tauis the ICAR scale parameter;Also, could give each connected component its own scale by default since they're practically separate models, and it seems to improve sampling.