Post date: Apr 4, 2015 7:58:11 AM
This post is to announce two new features in R-INLA
Model comparison using the WAIC criterion is recommended (by Gelman, Hwang and Vehtari) over the DIC criterion. This criterion is available in the testing version. To enable it, use
r = inla(...., control.compute = list(waic=TRUE))
and then the waic-value and p.eff is there as r$waic$waic and r$waic$p.eff
In a recent report, we suggests a way to improve Laplace approximations without extra computational costs. This is important for hard and difficult cases (see the report for details). This improvement can be enabled using
r = inla(..., control.inla = list(
correct = TRUE,
correct.factor = <VALUE>))
where 'correct.factor' is a robustness-factor (see the report for details). In the report we used a correct.factor=10, while the default value in R-INLA is currently 1. The default value in R-INLA might change at a later stage.
Please feel free to report your experience with this improved approximation to help@r-inla.org.