ANOVA and lsmeans

[Clg]

Hello everybody,

I have a question about the interpretation of an one-way ANOVA. The variable TRT have 3 groups and I use a proc mixed to obtain estimate and lsmeans.

My code is :

proc mixed data=base method=reml ;
class TRT;
model param= TRT / s alpha=0.10  ;
estimate "b versus A " TRTA -1 1 0 / cl alpha=0.10 ;
lsmeans TRT/ cl alpha=0.1;
run;

With the estimate, when I compare Bvs A, I obtain a pvalue=0.0478. But when I look the estimate means with the lsmeans statement I obtain estimate means and IC :

- group A : 1.0163402115 [0.9107315766 ; 1.1341952469]
- group B : 0.844887592 [0.7603604662; 0.9388113597]

We can see that the IC overlap so, I think that there is no difference between group, however, the pvalue=0.0478. How could you explain this situation according you ? Maybe because each group has a small size ? (n=12).

Thank you in advance for your answer.

[PaigeMiller]

Overlapping confidence intervals is never a valid way to perform an hypothesis test. So I would believe that pvalue=0.0478 is correct.