Proc GEE (Multinominal Outcome) - Estimate Statement

[varatt90]

Hi,

I'm running proc gee for a multinominal outcome (0 1 2 3 4, with ref = 0). I would like to find the odds ratio for a continuous predictor using the Estimate statement (please see code below). 

However, in my estimates table, I notice that I do get an estimate but the outcome column shows "0" instead of 1, 2,3, or 4. 

Could someone verify if my code is written correctly?

Thanks in advance!

Proc GEE Data = Test;
Class =schoolid outcome year (ref = "0");
Model outcome (ref = "0") = year age;
Repeated subject = schoolid  / type = ind;
Estimate "1 vs 0"  Age 1 / exp cl;
Estimate "2 vs 0"  Age 2 / exp cl;
Estimate "3 vs 0"  Age 3 / exp cl;
Estimate "4 vs 0"  Age 4 / exp cl;
LSMEANS Year /diff oddsratio cl ; 
Run;

[SAS_Rob]

Could you post the output you are unclear about?

[varatt90]

Here's the output:

 

Also, I realized in the log, it says "Note: Results for ESTIMATE statements without a CATEGORY= option apply to the response level '0' ".

How does the category option work?

[StatDave · Accepted solution]

See this note that discusses and illustrates the use of the CATEGORY= option with multinomial models. The example shows the model fit by PROC LOGISTIC, but should apply to your GEE model.

[varatt90]

Great! Thank you! Just to be clear, the only difference between Category = Separate and Category = Joint is how the output is presented. The separate statement presents each comparison in its own table vs the joint table puts it all together. Is that correct? 

[StatDave]

Yes

[ballardw]

It may not hurt to show the LOG with the code actually submitted and any notes.

I would expect an ERROR message and no output generated because of the = in the CLASS statement.

Modifying one of the GEE example to show what happens when an equal sign is present on the CLASS statement:

540  proc gee data=Resp descend;
541     class= ID Treatment Center Sex Baseline;
        -----
        180
ERROR 180-322: Statement is not valid or it is used out of proper order.

542     model Outcome=Treatment Center Sex Age Baseline /
543           dist=bin link=logit;
544     repeated subject=ID(Center) / corr=exch corrw;
545  run;

WARNING: No MODEL statement has been specified. No analysis will be done.

So make sure the code you share is actually what generates the output.

---

@varatt90 wrote:

Hi,

I'm running proc gee for a multinominal outcome (0 1 2 3 4, with ref = 0). I would like to find the odds ratio for a continuous predictor using the Estimate statement (please see code below). 

However, in my estimates table, I notice that I do get an estimate but the outcome column shows "0" instead of 1, 2,3, or 4. 

Could someone verify if my code is written correctly?

Thanks in advance!

Proc GEE Data = Test;
Class =schoolid outcome year (ref = "0");
Model outcome (ref = "0") = year age;
Repeated subject = schoolid  / type = ind;
Estimate "1 vs 0"  Age 1 / exp cl;
Estimate "2 vs 0"  Age 2 / exp cl;
Estimate "3 vs 0"  Age 3 / exp cl;
Estimate "4 vs 0"  Age 4 / exp cl;
LSMEANS Year /diff oddsratio cl ; 
Run;

---

[varatt90]

Thanks for catching that error! I did write it as CLASS schoolid (without the equal sign) in SAS.