Longitudinal Scored Data

[billfriend]

I have scored data from 3 time points, the scores range from 0-5. 

If the data is normal, can I run this?

proc glimmix data=test;

class sex(ref='1') education(ref='1') race_ethn(ref='1') age;

model score=time sex education race_ethn age hhsize/s dist=GAUSSIAN LINK=CUMLOGIT or;

random time;

run;

Do I include time in the class statement?  Also, it is correct to include time in the random statement? To account for nature of the data, dependent, do I need to include: Store=ID.

repeated subject=store / corr=ind corrw;

If the data is not normal, can I run this?

proc glimmix data=test;

class sex(ref='1') education(ref='1') race_ethn(ref='1') age;

model score=time sex education race_ethn age hhsize/s dist=multinomial LINK=CUMLOGIT or;

random time;

run;

Should the distribution actually be poisson since it's count data (i.e. scored data)?

[Ksharp]

I do not think you can do that.

LINK=CUMLOGIT

is for the general logistic regression , and for the category dependent varible which has three or more levels .

[billfriend]

I've read that you can use LINK=CUMLOGIT for multinomial data. Any thoughts on how I should treat the data? Normal vs multinomial vs poisson?