4.3 Priors specification

(inla.models()$prior$normal)
## $doc
## [1] "Normal prior"
## 
## $nparameters
## [1] 2
## 
## $pdf
## [1] "gaussian"
inla.doc("Normal")

You should always check the doc (or do yourself some cheat sheet) because like every software they could use other convention : use of precision instead of variance in \(N()\).

INLA use some default setting but you can overwrite at:

  • the “default level”: inla.set.control.fixed.default()

  • Inside inla(): control.fixed = prior.fixed prior.fixed is a list

The hyperparameters are specified in a list (“key=value”) with the hyper arguments inside the f():

  • inital: where to start

  • prior: prior distribution

  • param: vector of the parameters for distribution

  • fixed: fixed value or not

You can use Penalized Complexity or PC priors.