topic Question on interpreting the result of PROC MIXED in Statistical Procedures

Question on interpreting the result of PROC MIXED

GiaLee — Fri, 01 Mar 2024 13:47:22 GMT

I'm confused about interpreting the "Solution for fixed effects" and "Type 3 Tests of Fixed Effects." I have searched through lots of SAS papers but still feel confused.

The F-test shows that the interaction term is significant, which means age has a significant effect on score changes. However, I would like to report the coefficients with confidence intervals, so I may need the results from the t-test. However, for each of the interaction terms in the test, they are not significant. How could I explain this?

Here is my code and output. Any help is appreciated

PROC MIXED DATA = test METHOD = REML COVTEST ;
CLASS id age_less55 time_dichotomous(ref="0-6") ;
MODEL Score = time_dichotomous age_less55 time_dichotomous*age_less55/ SOLUTION;
RANDOM INTERCEPT / SUBJECT = id;
repeated time_dichotomous / type=ar(1) sub=id;
store out=MixedModel;
RUN;
proc plm restore=MixedModel;
effectplot interaction (x=time_dichotomous sliceby=age_less55) / clm connect;
run;

Re: Question on interpreting the result of PROC MIXED

PaigeMiller — Fri, 01 Mar 2024 14:03:17 GMT

The F-test for an interaction and the t-test for different parts of the interaction do NOT test the same thing. They do not have to agree.

The F-test for an interaction tests to see if the different levels of the interaction are different from one another. The t-tests tests to see if the coefficient for a specific level of an interaction is different than zero.

Re: Question on interpreting the result of PROC MIXED

GiaLee — Fri, 01 Mar 2024 14:38:12 GMT

Thanks for your quick response!
So, that means for the T-test, it estimates whether the estimate is different from zero. In my case, it's whether the estimate is different from the estimate from time_dichotomous (0-6) * age_less55 (No) since time_dichotomous (0-6) is the reference group.

Regarding the F-test, it tests the estimates shown in Table 1 to see whether the estimates for time_dichotomous (0-6) * age_less55 (No), time_dichotomous (6-12) * age_less55 (No), time_dichotomous (12-18) * age_less55 (No), etc., are different from each other.
Am I understanding this correctly?

Thanks!