Skip to content

MCMCglmm ordinal model

2 messages · Frank Lawrence, Jarrod Hadfield

#
When I specify an ordinal single level model using MCMCglmm, part of the output includes estimates for the cut points for the response variable scale. Two questions:
1. Is the "post.mean" on the original ordinal scale or do the estimates need to be exponentiated like the coefficients? 
2. Why do I occasionally wind up with fewer cut points than the number that existed on the observed variable?
Frank Lawrence
#
Hi,

The coefficients are on the latent scale, which is the probit scale  
not log scale.

You should end up with one fewer cutpoints than categories because all  
cutpoints cannot be uniquely estimated with an intercept term in the  
model. Essentially the first cutpoint is set to 0.  See  
http://permalink.gmane.org/gmane.comp.lang.r.lme4.devel/3644

Cheers,

Jarrod

Quoting Frank Lawrence <cougar at psu.edu> on Wed, 24 Aug 2011 20:36:09 +0000: