repeated measures ANOVA or PROC GLM...HELP!

[ksmielitz]

Dear SAS Friends, 

The good news is, the article was accepted. The bad news is one of the reviewers wants something (repeated measures ANOVA) and I don't know how to provide it.

Hopefully with the information I provide here you can tell me if I can even do it. I'd like to hope the reviewer has lost his/her mind, but I'm pretty sure I'm just not doing something correctly. 😉

Okay. I have a financial education course (treatment). I collected pretest scores (preknow) and posttest (postknow) scores. I have race (black/notblack), age (youngadult, adult, slightlyolderadult, midage, older), and education level (notgradhs hsgedordip somecollegeormore) as demographic variables. I have some other predictor variables, but we'll use these as an example and if I can run it I can figure it out (I hope, pray, think) from the information you all provide me here.

The reviewer wants to know who learned the most from the treatment/who learned more. When I googled repeated measures ANOVA, it led me to PROC GLM with some other nifty pieces. So, here's what I've got:

proc glm;
class black youngadult adult slightlyolderadult midage older notgradhs HSGEDorDip somecollegeormore;
model preknow postknow = black youngadult adult slightlyolderadult midage older notgradhs HSGEDorDip somecollegeormore black*notgradhs black*hsgedordip 
black*somecollegeormore black*youngadult black*adult black*slightlyolderadult black*midage black*older /nouni;
repeated postknow;
lsmeans black youngadult adult slightlyolderadult midage older notgradhs HSGEDorDip somecollegeormore black*notgradhs black*hsgedordip
black*somecollegeormore black*youngadult black*adult black*slightlyolderadult black*midage black*older;
run;

Now, the good news is, the procedure runs. The bad news is I get crap. 

For example:

somecollegeormore preknow LSMEAN postknow LSMEAN
0                                              Non-est                  Non-est
1                                              Non-est                  Non-est

black NotGradHS preknow LSMEAN postknow LSMEAN
0             0                           Non-est                   Non-est
0            1                            Non-est                   Non-est
1            0                            Non-est                  Non-est
1            1                           Non-est                   Non-est

and in the log

NOTE: H Matrix for postknow*youngadult has zero d.f.

NOTE: H Matrix for postknow*adult has zero d.f.

NOTE: H Matrix for postknow*slightlyolderadult has zero d.f.

NOTE: H Matrix for postknow*midage has zero d.f.

NOTE: H Matrix for postknow*older has zero d.f.

NOTE: H Matrix for postknow*black*NotGradHS has zero d.f.

NOTE: H Matrix for postknow*black*HSGEDorDip has zero d.f.

NOTE: H Matrix for postknow*black*somecollegeorm has zero d.f.

NOTE: H Matrix for postknow*black*youngadult has zero d.f.

NOTE: H Matrix for postknow*black*adult has zero d.f.

NOTE: H Matrix for postknow*black*slightlyoldera has zero d.f.

NOTE: H Matrix for postknow*black*midage has zero d.f.

NOTE: H Matrix for postknow*black*older has zero d.f.

NOTE: The Huynh-Feldt epsilon and the corresponding adjusted p-value have been enhanced to
include a correction based on Lecoutre (1991). Use the UEPSDEF=HF option on the REPEATED
statement to revert to the previous definition.

Thanks in advance for any help you can provide.

Kate

[Ksharp · Accepted solution]

If I was right "(repeated measures ANOVA) " is called MIXED model.

You have two ways to do (repeated measures ANOVA) :

1) using proc glm + manova statement -- do some multiple variable analysis

   Check proc glm's documentation ,there is an example about it .

   Example 46.7: Repeated Measures Analysis of Variance

2) using MIXED Model. Check proc mixed, proc glmmix ..... @Steve @lvm  can give you good advice.

P.S. I suggested you use Mixed model which have good ability to handle missing value.

[ksmielitz]

@Ksharp Thank you, I'll check out your suggestions!

[ksmielitz]

I ended up using PROC MIXED...still no significant results, but I'm happy because I was right...ROFL.

Thanks for your help, @Ksharp