Mixed Procedure

[k2k]

Dear Sir/Madam

I am having doubt in Estimate Statement in proc mixed procedure.

ESTIMATE 'TEST VS REFERENCE' TRT 1 -1;

When want to use 1 -1 or -1 1,

TRT (T-  Test and R- Reference) code as TRT -1 1 to compare test with reference. my results are matching,  but for TRT (A- Test and B- Reference) I used same code but my results are not matching if we use 1 -1 then my answer is matching.

What logic is behind this?

[PaigeMiller]

What results don't match? Can you show us? (Also, the complete PROC MIXED code would be helpful)

[k2k]

Dear paigemiller,

SAS Proc mixed Code,

PROC MIXED DATA = PK;

     CLASSES SEQUENCE SUBJECT PERIOD TRT;

     MODEL LN_CMAX = SEQUENCE PERIOD TRT / ddfm = satterth;

     RANDOM TRT / type = fa0(2) sub = SUBJECT G;

     REPEATED/grp = TRT sub = SUBJECT;

     LSMEANS TRT / cl alpha = 0.05 diff;

     ESTIMATE 'T vs. R' TRT  -1 1/cl alpha = 0.1;

     ODS OUTPUT ESTIMATES = CMAXUNSCL ( keep = lower upper );

     ODS OUTPUT LSMEANS   = CMAXLSM

                TESTS3    = CMAXTYPE3

                COVPARMS  = CMAXCPARMS

                DIFFS     = CMAXDIFF;

RUN;

90 % CI is not matching  CI=exp(lower);

[PaigeMiller]

That really doesn't help. We can't see what the results are, we don't see what doesn't match, you need to SHOW US the non-matching output. And you also need to show us the two versions of the code that gives the non-matching output.

[SteveDenham]

So the 90% CI in the ESTIMATE block does not match 'lower' and 'upper' in the output dataset?  That seems really unlikely, since it is an ODS dataset..

I have to echo here---show what does not match explicitly.

That particular piece of code has some interesting tricks going on, so it really makes a difference when you say things don't match that we be able to duplicate what you see.

Steve Denham