topic Post-Estimation Commands (Proc Mixed) - Hypothesis Testing in Statistical Procedures

Post-Estimation Commands (Proc Mixed) - Hypothesis Testing

itsd — Fri, 25 Jun 2021 16:03:58 GMT

I have run a proc mixed regression analysis and I need to run some post-estimation commands to test whether the treatment had a non-zero impact for a certain racial group.

Specifically, say we have the (simplified) model Y = B0 + B1(Treatment) + B2(Black) + B3(Treatment*Black). I want to test if (B1-B3)=0. Which command would I use in SAS?

I have read the few guides on ESTIMATE, CONTRAST, and LSESTIMATE, but I don’t know which one to use nor what the values represent.

Re: Post-Estimation Commands (Proc Mixed) - Hypothesis Testing

PaigeMiller — Fri, 25 Jun 2021 16:44:34 GMT

Let's suppose there are two treatment levels, and so the effect of treatment (your B1) also has two levels. If Black has two levels, then the effect of TREATMENT*BLACK (your B3) has 2*2 = 4 levels. So what does it mean to compare B1 with 2 levels with B3 which has four levels (via the subtraction B1 – B3) ? I suggest that the question as posed is not answerable.

... to test whether the treatment had a non-zero impact for a certain racial group.

Perhaps you are really interested in testing the effect of B3 to see if it is statistically different than zero? If that's it, then B3 can be tested to see if it is zero via the default output from PROC MIXED. You can also use the SLICE command that might provide additional information. ESTIMATE, CONTRAST and LSMESTIMATE seem not needed, in my opinion.

Re: Post-Estimation Commands (Proc Mixed) - Hypothesis Testing

itsd — Fri, 25 Jun 2021 16:50:10 GMT

I already know that the interaction B3 is significant from the mixed analysis. I suppose I may have been wrong to say that I want test B1-B3=0, I think I might have meant B1+B3=0 (since I want to know if the interaction “cancels out” the effect of the treatment for that group). Does this make sense?

Re: Post-Estimation Commands (Proc Mixed) - Hypothesis Testing

PaigeMiller — Fri, 25 Jun 2021 17:10:42 GMT

I have the same objection to B1 + B3 that I had to B1 – B3.

I still think the SLICE command may be the tool you are looking for.

Re: Post-Estimation Commands (Proc Mixed) - Hypothesis Testing

jiltao — Fri, 25 Jun 2021 17:11:20 GMT

Are Treatment and Black continuous (or dummy) variables or are they in the CLASS statement? Can you send us your complete PROC MIXED code and the Class Level Information table from PROC MIXED output?