[R-sig-ME] MCMCglmm non-informative prior
W Robert Long
longrob604 at gmail.com
Thu Aug 2 21:09:04 CEST 2012
I want to set a non-informative prior on a random effect. My simplest
model has just one random effect.
The tutorial pdf shows how to use an improper flat prior, and a proper
Cauchy prior. I don't think the former is appropriate for me because I
know the parameter will not take values above 1 (or rather values above
1 are vanishingly small). I guess a U(0,1) prior distribution would be
ideal but I don't see any way to parameterise that. Am I thinking about
this properly ?
As for the proper Cauchy prior which seems to accept a
additional parameter that controls the scale , I don't really understand
this - can anyone explain it ? Can this be used in my situation ?
University of Leeds / UK
More information about the R-sig-mixed-models