topic Re: PROC MCMC for Bayesian Hierarchical Meta-Analysis in Statistical Procedures

PROC MCMC for Bayesian Hierarchical Meta-Analysis

sassos — Fri, 18 Jul 2014 10:25:28 GMT

Hi All,

I am trying to run the Bayesian hierarchical analysis example shown here in the SAS help: SAS/STAT Software Examples: Bayesian Hierarchical Modeling for Meta-Analysis

PROC MCMC Code:

proc mcmc data=multistudies outpost=nlout seed=276 nmc=50000 thin=5

monitor=(OR Pooled);

array OR[22];

parms mu tau2;

prior mu ~ normal(0, sd=3);

prior tau2~ igamma(0.01,s=0.01);

random theta ~n(mu, var=tau2) subject=studyid;

OR[studyid]=exp(theta);

Pooled=exp(mu);

model logy ~ n(theta, var=sigma2);

run;

... but I get the following error message:

ERROR: The initial value 0 for the parameter tau2 is outside of the prior distribution support set.

NOTE: The prior of tau2 is Inv-GAMMA.

NOTE: The distribution has a support set of (0, Infinity).

NOTE: The SAS System stopped processing this step because of errors.

Can anybody enlighten me as to why this is happening? I also get an error when I change the tau2 prior to another distribution (uniform, lognormal, etc) and it only seems to work when I remove the tau2 prior altogether (which obviously isn't desirable).

I'm using SAS 9.4.

Any advice would be greatly appreciated.

Thanks,

Charles.

Re: PROC MCMC for Bayesian Hierarchical Meta-Analysis

SteveDenham — Fri, 18 Jul 2014 12:53:23 GMT

Probably the easies way to get around this would be in the PARMS statement

parms mu tau2 {0.1};

The curly braces enclose an initial value, so this will move it away from the boundary of the set.

Steve Denham

Re: PROC MCMC for Bayesian Hierarchical Meta-Analysis

sassos — Fri, 18 Jul 2014 14:15:11 GMT

Hi Steve,

Thanks for getting back to me. I actually tried this already but unfortunately get the following error messages:

ERROR: Invalid Operation.

ERROR: Termination due to Floating Point Exception

Does that provide any more clues as to what might be going on?

Thanks,

Charles.

Re: PROC MCMC for Bayesian Hierarchical Meta-Analysis

SteveDenham — Fri, 18 Jul 2014 14:22:11 GMT

I missed that the sd for the igamma prior was 0.01, so my intial value was 10 sd away, if that is what you used. Are you certain about the parameters for the igamma? If so, then we just reached the end of my guesses as to things to try. I would say try opening a ticket with Tech Support if fiddling with the parameters of the prior doesn't help.

Steve Denham

Re: PROC MCMC for Bayesian Hierarchical Meta-Analysis

sassos — Fri, 18 Jul 2014 14:43:37 GMT

Hi Steve,

I tried with 0.01 instead but alas still no luck. Will try tech support as you suggested, though I'm particularly curious because this was the code provided by the official SAS software examples on their website... so I assume it must have worked for somebody!

Thanks,

Charles.

Re: PROC MCMC for Bayesian Hierarchical Meta-Analysis

SteveDenham — Fri, 18 Jul 2014 14:50:42 GMT

Any possibility of a versioning incompatibility? Tech support will hit that first, I think.

Steve Denham

Re: PROC MCMC for Bayesian Hierarchical Meta-Analysis

sassos — Fri, 18 Jul 2014 15:13:14 GMT

I'm guessing that could be the reason, let's see.

Thanks again,

Charles.

Re: PROC MCMC for Bayesian Hierarchical Meta-Analysis

Melissaglenda — Fri, 25 Jul 2014 07:11:23 GMT

data BMA;

input id ctrl ctrlN trt trtN;

datalines;

1 3 39 3 38

2 14 116 7 114

3 11 93 5 69

4 127 1520 102 1533

5 27 365 28 355

6 6 52 4 59

7 152 939 98 945

8 48 471 60 632

9 37 282 25 278

10 188 1921 138 1916

11 52 583 64 873

12 47 266 45 263

13 16 293 9 291

14 45 883 57 858

15 31 147 25 154

16 38 213 33 207

17 12 122 28 251

18 6 154 8 151

19 3 134 6 174

20 40 218 32 209

21 43 364 27 391

22 39 674 22 680

;

run;

data BMA;

set BMA;

logy=log(trt/(trtN-trt)/(ctrl/(ctrlN-ctrl)));

sigma2=1/trt+1/(trtN-trt)+1/ctrl+1/(ctrlN-ctrl);

run;

proc mcmc data=BMA outpost=nlout seed=276 nmc=50000 thin=5

monitor=(OR Pooled);

array OR[22];

parms mu tau2;

prior mu ~ normal(0, sd=3);

prior tau2~ igamma(0.01,s=0.01);

random theta ~n(mu, var=tau2) subject=id;

OR[id]=exp(theta);

Pooled=exp(mu);

model logy ~ n(theta, var=sigma2);

run;

I am unable to run this code.

Error :

random theta ~n(mu, var=tau2) subject=id;

------

180

ERROR 180-322: Statement is not valid or it is used out of proper order.

Please help me

Re: PROC MCMC for Bayesian Hierarchical Meta-Analysis

Funda_SAS — Fri, 25 Jul 2014 15:07:51 GMT

I've just copied and run your code, and got no error, clean log! One thing I suspect is that the version you use might not support the RANDOM statement. What version do you use? RANDOM statement is made available by SAS/STAT 9.3.

Funda

Re: PROC MCMC for Bayesian Hierarchical Meta-Analysis

Melissaglenda — Fri, 25 Jul 2014 17:28:37 GMT

I am using SAS 9.2. Well you are right may be,,!! I will install 9.3 soon !! I am dying to see the output..Thank you very much for ur help..

Re: PROC MCMC for Bayesian Hierarchical Meta-Analysis

ets_kps — Fri, 25 Jul 2014 17:44:03 GMT

While installing the newer version on your machine will guarantee access to all your products, if you download this, you can try running your code on the newest versions right now.

Free Statistical Software, SAS University Edition | SAS

Re: PROC MCMC for Bayesian Hierarchical Meta-Analysis

Melissaglenda — Fri, 25 Jul 2014 17:46:08 GMT

Wonderful..!! Thanks a lot..!! :smileyblush: Thank you so much...