Hi all,
For robust regression, Kruschke (2015) suggests using a family of a student t distribution, with a prior for the normality parameter (nu) of a shifted exponential distribution. Specifically, he suggests an exponential distribution with a lambda of 1/29 and a shifting of 1 so its minimal value will be equal to 1.
Do you know how can I specify prior like that to the nu parameter in brms?
Thanks!
This sounds kind of like the pareto distribution. I reckon that should do the trick.
//edit
I now checked again in brms and am wondering whether it is not yet included. I can’t seem to find it.
There might be a brms way to define this prior, but in a pinch you can always pass your own prior, defined in raw Stan, to brms via the stanvars argument. See User-defined variables passed to Stan — stanvar • brms