Programming the statistical procedures from SAS

Interpreting GLIMMIX logit interaction term (continuous*categorical)

Reply
Occasional Contributor
Posts: 8

Interpreting GLIMMIX logit interaction term (continuous*categorical)

Hello,

I am using PROC GLIMMIX to perform a multilevel logistic regression.  I am interested in a number of interaction effects.  Breaking down interactions for categorical*categorical terms is relatively straightforward with LSMEANS; however, I have not been able to find a resource which details how to break down custom hypotheses for continuous*categorical interaction terms.   Does anyone have suggestions for resources that may detail how to explore these types of interactions for a PROC GLIMMIX logit model?

Thanks!

Respected Advisor
Posts: 2,655

Re: Interpreting GLIMMIX logit interaction term (continuous*categorical)

Do whatever it takes to get a copy of Walt Stroup's Generalized Linear Mixed Models. Then turn to page 220 where he starts a section on Analysis of Covariance.  This will walk you through how to handle covariate(continuous) by treatment(categorical) interactions and specific tests at various levels of the continuous variable.

Steve Denham

Occasional Contributor
Posts: 18

Re: Interpreting GLIMMIX logit interaction term (continuous*categorical)

Hello,

I'm sorry to use this section to my own interest. But I have the same question and I can not find the right command ...

This is my model on avian reproduction. Where age1 is a continuous variable(1, 2, 3)  and sex is categorical (1= male, 2= female).

proc glimmix data=Reproduction INFOCRIT=PQ  noreml;

class Sex nest Age2 ring;

model Reproduction (Event="1") = Sex Age1 Sex*Age1/dist= binomial link=logit solution cl;

random Age2/ subject=ring (nest) type=cs residual;

ESTIMATE 'sex 1 vs 2 at age  1 in' sex -1 1 0 Sex*Age1 -1 1 0 ;

ESTIMATE 'sex 1 vs 2 at age  2 in' sex -1 1 0 Sex*Age1 -2 2 0 ;

ESTIMATE 'sex 1 vs 2 at age  3 in' sex -1 1 0 Sex*Age1 -3 3 0 ;

run;

Could you tell me the correct command to see estimates of the interaction "Sex*Age1"? The estimates that I have shown in my model, showed me only sex differences for each age, but I need male differences along age. I'm sorry for my limited knowledge of SAS.


Thank you very much.


All the best,

Jaime

Valued Guide
Valued Guide
Posts: 684

Re: Interpreting GLIMMIX logit interaction term (continuous*categorical)

Search some old posts. This has come up several times in the past. Here is one place to start:

Occasional Contributor
Posts: 18

Re: Interpreting GLIMMIX logit interaction term (continuous*categorical)

I'm sorry but I can not understand how this works. In my case it is an interaction between a continuous and categorical variable, and I also used proc GLIMMIX with dist= binomial link=logit. Although the method could be similar (continuous*continuous), it is difficult to understand because my little knowledge in SAS. In my particular model, what should I use?

Thank you very much,

Jaime

Valued Guide
Valued Guide
Posts: 684

Re: Interpreting GLIMMIX logit interaction term (continuous*categorical)

THis was just an example. Search this website for interaction of continuous and categorical. There are several posts.

Valued Guide
Valued Guide
Posts: 684

Re: Interpreting GLIMMIX logit interaction term (continuous*categorical)

I don't quite understand your question, but I think you want to know the mean difference between age1 of 1 and 3 (for example) for males. The third contrast will do this, but I show the first two so you can see the steps.

estimate 'm age 1' int 1 sex 1 0 age1  1  sex*age1 1;

estimate 'm age 3' int 1 sex 1 0 age1  3  sex*age1 3;

estimate 'm, age 3-1               ' age1  2  sex*age1 2;

Note the the last estimate is the difference between the first two (which give you means for age 1 and age 3, both for males only [the first level of sex]). WIth the difference, terms like the intercept and sex main effect cancel.

By the way, there are only two levels to your sex*age1 interaction, so your estimate statements will not work.

Ask a Question
Discussion stats
  • 6 replies
  • 425 views
  • 0 likes
  • 4 in conversation