I have a truncated dependent variable, and I need to run a cross-classified growth model. If it weren’t truncated, my model would look like this:
proc mixed data=std noclprint covtest;
class sch time raceses;
model learn= raceses time raceses*time
alagcollfact alagcollfact*time alagcollfact*raceses alagcollfact*raceses*time
alagcommfact alagcommfact*time alagcommfact*raceses alagcommfact*raceses*time/ ddfm=bw s ;
repeated time/type=cs subject=childid3;
random intercept / subject=sch;
*random intercept/ subject=sch*childid3;
weight weightvar;
by _imputation_ ;
lsmeans raceses*time/at alagcommfact=-.78 diff cl;
lsmeans raceses*time/at alagcommfact=.792 diff cl;
format raceses raceses.;
run;
So, I see that nlmixed allows truncated dependent variables, but as far as I can tell, it doesn’t allow for cross-classified models because it only permits one subject. Can you recommend a way for me to turn the model above into a tobit model that is right censored (The dependent variable ranges from 1 to 4 in .1 increments and it is top censored (or truncated) at 4.
FYI: I found the following syntax online for truncated data (but again, I don’t think I can add two random statements): http://www.ats.ucla.edu/stat/sas/code/random_effect_tobit.htm
Thank you!
