topic Proc MIXED differences between SAS 9.3 and 9.4 in Statistical Procedures

Proc MIXED differences between SAS 9.3 and 9.4

Cate — Wed, 30 Aug 2017 09:39:00 GMT

Good morning

I need your help to understand the difference of running Proc MIXED in SAS 9.3 & SAS 9.4.

The same program (same dataset) running smoothly on SAS 9.3 when I run it on SAS 9.4 returns the 'WARNING: Stopped because of infinite likelihood.'.

In the data there are no repeated values.

ods output lsmeans=STAT00;
  proc mixed data = rawda MAXFUNC=5000;
       by paramcd param;
       class usubjid studyid avisitn basecn ; 
       model chg = studyid avisitn basecn avisitn*basecn /solution;  
       repeated avisitn / subject = usubjid(studyid) ;
       lsmeans basecn*avisitn ; 
       PARMS / NOITER ; 
  run;              
ods output close;

Please help me to understand what did change.

Many thanks in advance

Re: Proc MIXED differences between SAS 9.3 and 9.4

alexal — Wed, 30 Aug 2017 11:42:47 GMT

@Cate,

What happens if you add the option singular=1e-8 in the MODEL statement in PROC MIXED?

Re: Proc MIXED differences between SAS 9.3 and 9.4

Cate — Wed, 30 Aug 2017 13:28:31 GMT

@alexal Many thanks for your promptly reply!

The program runs smoothly with your option (it works with 1e-11 too).

When I run the program in SAS 9.3 & 9.4 with the same singluar=1e-8 I do expect the Least Squares Means being identical for both versions. This is not the case. I get the same Estimate, DF & Pr > |t|, but Standard Error & T values are different. Which is the reason?

Many thanks for helping me!

Re: Proc MIXED differences between SAS 9.3 and 9.4

alexal — Wed, 30 Aug 2017 13:35:50 GMT

@Cate,

You are welcome!

You can find the information on the SINGULAR= option in the PROC MIXED documentation here.

SINGULAR=number
tunes the sensitivity in sweeping. If a diagonal pivot element is less than D*number as PROC MIXED sweeps a matrix, the associated column is declared to be linearly dependent upon previous columns, and the associated parameter is set to 0. The value D is the original diagonal element of the matrix. The default is 1E4 times the machine epsilon; this product is approximately 1E–12 on most computers.

In your case, the default criterion might be too stringent. That is why I suggested the singular=1e-8 option.