05-18-2015 02:18 AM
I have been trying to implement bayesian geostatistical spatial analysis in PROC MCMC (using latitude and longitude info). After a couple of days and getting many errors, now I suspected that spatial analysis is not possible in PROC MCMC. I also found the following sentence in the SAS user guide for PROC MCMC:
" The random-effects parameters associated with each subject in the same RANDOM statement are assumed to be conditionally independent of each other, given other parameters and data set variables in the model."
So my RANDOM statement which is
random gamma ~ mvn(mean, cov) subject=cluster_id ;
is wrong, where, mean is zero vector, cov is covariance matrix with a spatial correlation and cluster_id is a variable in the dataset which identifies clusters. I will be thankful if anyone can comment on this. Aren't we really able to do spatial statistics analysis in PROC MCMC? If not, what is your suggestion?
05-18-2015 02:15 PM
Perhaps I am missing a key element here, but if the matrix cov accurately reflects a spatial correlation, and is correctly defined in the programming statements, then this should work. Can you share the entirety of your PROC MCMC code?