Programming the statistical procedures from SAS

Multinomial logistic regression with robust standard errors/clustered data

Reply
New Contributor
Posts: 2

Multinomial logistic regression with robust standard errors/clustered data

Hi,

 

I have to do a multinomial logistic regression on clustedered data. My supervisor ask me to run a logistic regression with robust standard errors in order to take into account dependency between observations in the data set. I have tried to find an appropriate procedure in SAS 9.4 to do so, and my best guess is to use the PROC GLIMMIX in which I put in 'random intercept / subject= id'. However, whwn I do so I get an error in the log saying: 'Nominal model require that the repsonse variable is a group effect on the on RANDOM statements. You need to add 'GROUP=outcome '. When I do so I just get another error saying: 'Model is too large to be fit by PROC GLIMMIX in a resonable amount of time on this system. Consider changing your model.'

 

Then my question is what to do and which model to use?

 

I have used the following SAS syntax:

 

1st error reported above:

 

PROC GLIMMIX data=c;

CLASS age (ref='3') ID;

MODEL outcome = age / DIST=multinomial LINK=glogit CL ODDSRATIO;

RANDOM intercept / SUBJECT=id;

RUN;

 

and 2nd error reported above:

 

PROC GLIMMIX data=c;

CLASS age (ref='3') id;

MODEL outcome = age / DIST=multinomial LINK=glogit CL ODDSRATIO;

RANDOM intercept / SUBJECT=id GROUP=outcome;

RUN;

 

Thanks in advance

Best,

Pernille

 

Grand Advisor
Posts: 9,442

Re: Multinomial logistic regression with robust standard errors/clustered data


Could you try 
 RANDOM _residual_ / SUBJECT=id;













New Contributor
Posts: 2

Re: Multinomial logistic regression with robust standard errors/clustered data

Thanks a lot for your quick repsonse. I have tried what you suggested, but then I get another error: 'R-side random effects are not supported by the multinomial distribution.'. Any other sugesstions?

Thanks in advance.

Ask a Question
Discussion stats
  • 2 replies
  • 60 views
  • 0 likes
  • 2 in conversation