Proc MCMC: dirichlet random variable

[csfcgua1]

Hello,

It seems dirichlet distributions are not allowed for the random statement in Proc MCMC, so I used gamma distributions instead.

Proc MCMC Data=mydata;

   array delta[3] delta1 delta2 delta3;
   array p[3] p1 p2 p3;

   model  .........

   random delta1 ~ gamma(0.1, iscale=0.1) subject = ID ;
   random delta2 ~ gamma(0.1, iscale=0.1) subject = ID ;
   random delta3 ~ gamma(0.1, iscale=0.1) subject = ID ;

   p1 = delta1/sum(delta1, delta2, delta3); 
   p2 = delta2/sum(delta1, delta2, delta3);
   p3 = 1 - p1 - p2;

Run;

I expected that every ID should have its own post-distribution/summary of p, but SAS just generated a summary of p for the whole group.

I could compute the p from the results of delta for each ID afterward. But is it possible to obtain p directly from Proc MCMC?

Thanks.

[ballardw]

Did you try BY group processing?

BY ID;

would be my first approach if I needed output for each level of ID.

[csfcgua1]

Thanks.

Yes. I thought about using BY at first, but I immediately realized that ALL of the  parameters would be estimated by ID, which is not necessary.