planned contrasts for repeated measures in glm

[montello]

I have a one-factor repeated measures design; this works fine for the main effect:

proc glm;
class;
model Meana Meanb Meanc Meand Meane Meanf Meang Meanh= /nouni;
repeated Segment 8;
run;

I have tried every syntax I could think of to carry out contrasts comparing level 1 to 5, and level 3 to 7, but none produces any contrast output.  Here is one example of my statements (yes, I have tried putting the contrast statement before the repeated statement):

proc glm;
class;
model Meana Meanb Meanc Meand Meane Meanf Meang Meanh= /nouni;
repeated Segment 8;
contrast '1 v. 5' 1 0 0 0 -1 0 0 0, '3 v. 7' 0 0 1 0 0 0 -1 0;
run;

[STAT_Kathleen]

If you are going to use  PROC GLM, the CONTRAST transformation on the REPEATED statement generates contrasts between levels of the factor and a reference level. By default, the procedure uses the last level as the reference level; you can optionally specify a reference level in parentheses after the keyword CONTRAST.  You will need to add the SUMMARY option to the REPEATED statement that will provide the respective results.  For example,

       repeated segment 8 / summary;

You would have results for 7 contrasts comparing each segment to segment 8. The results are shown under  Contrast_Variable: Segment 1,  Contrast_Variable: Segment 2,...Contrast_Variable: Segment 7.

Note the PROC GLM documentation provides a Repeated Measures Example using Polynomial Transformation located that might be helpful resource. I would also recommend reviewing the REPEATED statement SYNTAX that provides additional information.

[montello]

Thank you, I can get this to work, but it surprises me that I cannot
specify any contrast I want (such as level 3 vs. level 5). I did read all
about the REPEATED statement and its SYNTAX. Ciao.
Dan

[STAT_Kathleen]

You can specify

  repeated segment 5 / summary;

and that will provide segment 3 vs segment 5 comparison.

Another approach rather than using PROC GLM you might consider using PROC MIXED. You would need to restructure the data to univariate style, specify TYPE= covariance structure and then you can write specific CONTRASTS. Note most researchers use PROC MIXED rather than PROC GLM. The following paper

https://support.sas.com/rnd/app/stat/papers/mixedglm.pdf

provides an example of doing a repeated measures using PROC GLM and PROC MIXED.

[montello]

Actually, no that won't. These two versions produce exactly the same
output:

repeated segment 8/ summary;
repeated segment 5/ summary;

"restructure the data to univariate style"? This is but one of several
analyses I am doing. That seems like excessive trouble just to get simple
contrasts.