PROC GEE

[Mehdi_R]

Hello everyone,

I would like to use generalized logit GEE method. PROC GENMOD does not have this method and PROC GLIMMIX gives this error:

proc glimmix data = dataset;

class knowledge id;

model knowledge = prepost / s dist = multinomial link = glogit ;

random _residual_ / subject = id group = knowledge;

run;

ERROR: R-side random effects are not supported for the multinomial distribution.

I have heard that PROC GEE is a new experimental procedure in SAS 9.4 which applies GEE. I do have SAS 9.4 on my desktop, but when I run the code, I get this code;

proc gee data = dataset;

   class id prepost;

   model knowledge = prepost / dist=multinomial link=glogit;

   repeated subject = id / within = prepost;

run;

ERROR: Procedure GEE not found.

I will be grateful if anyone can help me how to apply generalized logit GEE to my dataset.

[jakarman]

See  SAS/STAT(R) 13.2 User's Guide  you need SAS/Stat 13.2
The experimental note is gone SAS/STAT(R) 14.1 User's Guide

Versions is a challenge: http://blogs.sas.com/content/sasdummy/2013/08/02/sas94-eg-versions/

[Mehdi_R]

Thanks jaap.

So I think I cannot use PROC GEE. So what procedure should I use to apply generalized logit GEE? Does anyone know?

[jakarman]

SAS/STAT(R) 9.3 User's Guide Example 39.5 GEE for Binary Data with Logit Link Function   The GENMOD Procedure   ?

[SteveDenham]

The following assumes that 'prepost' captures two 'trials' on each subject.  The repeated nature can be captured using a G side random statement that models the variance-covariance matrix over time.

proc glimmix data = dataset;

class knowledge id prepost ;

model knowledge (ref = '1') = prepost/ dist = multinomial link = glogit solution ;

random intercept /subject = id type = un group = knowledge;

random prepost/ type=un subject=id;

run;

As to why PROC GEE is not running, you need to be on SAS/STAT 14.1 to be able to model a nominal multinomial distribution.  See

Example 43.6 GEE for Nominal Multinomial Data (http://support.sas.com/documentation/cdl/en/statug/68162/HTML/default/viewer.htm#statug_gee_examples...)

for a worked example.

Steve Denham