[R] Hierarchical modeling MCMC with sample size 1

ping chen chen1984612 at yahoo.com.cn
Fri Mar 26 22:01:21 CET 2010

Dear R users: 

I am using hierchical modeling for a response varible (ordinal) given by patients for each doctor. My MCMC run with MCMMCglmm ( thin=20,nitt=208000,burnin=24000, family='ordinal') still can't have all the parameters pass the GEWEKE diagnosis test.

I am trying to find the reason for the slow convergence. 

Can the reason be 20 out of 50 doctors having 5 or less response (6 docs with 1 , 4 docs with 2,  3 docs with 3, 4 docs with 4 ,2 docs with 5 responses)? 
However, small sample sizes are the reason for bayesian modeling in the first place.

Can the reason be there are no strong variation among the doctors to start with? 

Any thoughts or suggestions will be appreciated.


More information about the R-help mailing list