topic Re: PROC GLIMMIX - lsmeans in Statistical Procedures

PROC GLIMMIX - lsmeans

rmn0008 — Thu, 14 Nov 2019 23:20:11 GMT

Hi, all.

If I don't specify the adjust in the lsmeans, what will be the standard?

PROC GLIMMIX data = SOY plots = residualpanel;
class TRT;
model KGHA = TRT / ddfm = satterth;
lsmeans TRT / lines adjust = Tukey;
run;

PROC GLIMMIX data = SOY plots = residualpanel;
class TRT;
model KGHA = TRT / ddfm = satterth;
lsmeans TRT / lines;
run;

Re: PROC GLIMMIX - lsmeans

PaigeMiller — Fri, 15 Nov 2019 11:46:07 GMT

I believe that with ADJUST= you get the multiple comparison test that is specified. If you leave ADJUST= out, the comparisons are not based upon multiple comparisons, so it would use the estimate of error produced by the model for each comparison (which is not adjusted because of multiple comparisons).

Re: PROC GLIMMIX - lsmeans

Rick_SAS — Mon, 18 Nov 2019 13:56:15 GMT

Paige is correct. I will add that unadjusted CIs are "wrong" in the sense that the true confidence level can be much less than the nominal level. For example, an unadjusted "95% CI" might only cover the true population mean in 87% of random samples.

Re: PROC GLIMMIX - lsmeans

rmn0008 — Mon, 18 Nov 2019 13:59:35 GMT

Okay, thank you very much!