topic Re: multinomial regression in panel data in SAS Health and Life Sciences

multinomial regression in panel data

fany — Mon, 06 Aug 2018 08:31:56 GMT

Hello all, i'm a master's student. I have some variable from SHARE data that i would like to use in my thesis. the data is panel data and my dependent variable is polytomic qualitative variable ( formal care, informal care, mixte care and no care) but i unable to see which procedure can I use in SAS to solve this problem. my question is:

which procedure can i use when I have panel data to make a multinomial regession with qualitative variable.

Your assistance would be greatly appreciated.

Thanks,

Fany

Re: multinomial regression in panel data

cau83 — Wed, 08 Aug 2018 00:44:35 GMT

I think your question may be more of a statistics question than a SAS question. Do you know what type of statistical model(s) you want to try? Tell us, then we can help with the Proc.

That being said, a possible answer to both is PROC LOGISTIC. I don't know what you mean by polytomic, but if the values you mentioned are ordinal (and i'm guessing they are, with no care being the lowest, up to formal care at the highest) then proc logistic is an option.

Re: multinomial regression in panel data

fany — Wed, 08 Aug 2018 07:41:01 GMT

I try to do an unordored multinomial logistic regression model with panel data. my dependents variable have the following modalities: 0=non care; 1= informal care only; 2=formal care only; 3= informal and formal care. I know that in cross-sectional data, we can use proc logistic with option link = glogit.

my question is : which procedure and option can i use to do this regression when i have panel data?

Re: multinomial regression in panel data

cau83 — Wed, 08 Aug 2018 12:33:23 GMT

Let me preface this by saying I do not have particular expertise in either logistic regression or working with panel data. That being said... I am not aware of any technical limitation on using logit + panel data. PROC LOGISTIC has a CLASS statement, so you could use your date variable in that statement or just as a continuous independent variable.

You should be okay with nominal/unordered data. Per SAS documentation

For nominal response logistic models, where the possible responses have no natural ordering, the logit model can also be extended to a multinomial model known as a generalized or baseline-category logit model

The part I'm least familiar with, this being panel data and potentially needing to track each person, I googled 'sas logit fixed effects' and found this post which mentions using PROC SURVEYLOGISTIC with the STRATA statement. That may be your best next stepl