topic Re: proc gee - Nesting continuous variable in Statistical Procedures

proc gee - Nesting continuous variable

SASAna — Wed, 09 Jun 2021 19:13:55 GMT

Hi SAS Users,

I am working on a statistical procedure and i am getting the below error.

ERROR: Nesting of continuous variable not allowed.

Code -

proc gee data=input_data descend;
class type plans ;
model Outcome_ind = type plans counter /
dist=bin link=logit;
repeated subject=time_period(id) / corr=ar corrw;
run;

type values are - 1,2,3 ( numeric)

plans values are - planA, planB, planC (charcter)

counter values are - 1,2,3....100+ (numeric)

Outcome_ind values are - 1,55,0,. (dot)

Thanks,

Ana

Re: proc gee - Nesting continuous variable

jiltao — Wed, 09 Jun 2021 19:36:38 GMT

You need to put ID in the CLASS statement. Also, I do not know your data, but generally speaking your subject effect might be ID, rather than time_period(id).

Re: proc gee - Nesting continuous variable

SASAna — Wed, 09 Jun 2021 20:05:42 GMT

Hello, Thanks for the reply. I tried replicating the below SAS example ,

proc gee data=input_data descend;
class type plans ;
model Outcome_ind = type plans counter /
dist=bin link=logit;
repeated subject=time_period(id) / corr=ar corrw;
run;

My scenario -

proc gee data=input_dataset descend;
class time_period type plans;
model outcome_ind = type PLANs counter /
dist=bin link=logit;
repeated subject=time_period(id) / corr=ar corrw;
run;

I have added time_period in the class and still getting the same error. i have 2 million's ID's. I do not think it should be the part of class.

Thanks,

Ana

Re: proc gee - Nesting continuous variable

StatDave — Wed, 09 Jun 2021 21:05:32 GMT

As noted in the GEE documentation, the variable(s) in the SUBJECT= effect must be in the CLASS statement. It is not a problem since you are not estimating parameters for the levels of the SUBJECT= variable(s). Also, since you say your response has three nonmissing levels, the response is not binary and DIST=BIN is incorrect. If you really just want to use the observations with levels 0 and 1, then you should remove the other observations first or use a WHERE statement to filter them out. Generally, a nested effect is not needed in the SUBJECT= option. See the discussion in this note for clarification.