topic Re: Ordinal Logistic Regression model with Random factor (Glimmix) in Statistical Procedures

Ordinal Logistic Regression model with Random factor (Glimmix)

AndreyMyslivets — Tue, 24 Mar 2015 12:39:14 GMT

Dear members of community!

Could you help me with the problem of providing an Ordinal Logistic Regression model with Random factor, please?

I have the following variables:

1. Resoult - outcome ordinal var with some kind of 7-levels scale (1-very bad, 2-bad, etc.);

2. Group - predictor var with 2 levels (fixed effect in the model);

3. Sequence - predictor var with 2 levels (fixed effect in the model);

4. Baseline - baseline var 100-levels scale (as I understand we can use this var like a continious var) as a covariate;

5. Site - random effect in the model.

6. Subject ID- uniq number of patient

The main problem is:

"The Resoult will be analysed using an Ordinal Logistic Regression model with Baseline as a covariate, Group and Sequence as fixed effects as well as Site as a random effect."

I think that we need to use Proc Glimmix because this procedure involve posibility to use it for Ordinal Logistic Regression with random factor as I have read, but I'm not a very native with the procedure.

So, how I can use Proc Glimmix insead of Proc Logistic for solving the problem?

Many thanks in advance!

Best Regards,

Andrey.

Re: Ordinal Logistic Regression model with Random factor (Glimmix)

AndreyMyslivets — Wed, 25 Mar 2015 10:21:10 GMT

I think that I need to use the following code for solving my problem:

proc glimmix data=data method=laplace;

class Group Sequence Site;

model Result = Group Sequence Baseline /dist=??? link=??? ddfm=???;

random Site/ solution;

run;

I have read many similar questions here and find that "method=laplace" is often recommended method, is it applicable in my case?

Could you help me please with choosing distribution type and link function for 7-levels scale outcome and link-function? Is it multinomial distribution?

Is it correct using of random statement?

Best Regards,

Andrey.

Re: Ordinal Logistic Regression model with Random factor (Glimmix)

SteveDenham — Mon, 30 Mar 2015 17:00:19 GMT

proc glimmix data=data method=laplace;

class Group Sequence Site;

model Result = Group Sequence Baseline / dist=multinomial link=cumlogit;

random Site/ solution;

run;

I would not specify anything having to do with denominator degrees of freedom (ddfm=) at this point. The default containment method should prove adequate.

Steve Denham