Hello,
I have a binomial dependent variable (0,1), a covariate "age", and a fixed factor (status: Inf, UN). The data are not fully independent because thay come from the same "strain" (with the two statuses above). The two "statuses" have been bred separatelly for several generations, so they have diverged genetically to some extent. I read that glimmix allows for "some degree of non-independence among observations" and I do not necessarily need a random factor. Therefore, believe that I can use Glimmix for my analysis.
However, I am not sure how to check the assumptions for glimmix (http://support.sas.com/documentation/cdl/en/statug/63347/HTML/default/viewer.htm#statug_glimmix_a000... Can anyone help me with this issue?
I will appreciate your help very much.
Thank you
My code would be the one below.
proc glimmix data=XXX order=data ODDSRATIO ;
class status
model Mature/Trials = age status age*status /dist=binomial link=logit solution ;
output out=results pred(ilink)=fit;
contrast 'UN - inf' inf 1 -1;
estimate 'UN - inf' inf 1 -1 /cl;
run;
It all depends on what assumptions you wish to test. GLIMMIX has a lot of options, plus there are a variety of plots that you can use to examine assumptions. I think the key I would think of for this is whether or not the results are overdispersed. Check out the examples in the GLIMMIX documentation for pointers.
Steve Denham
First, replace inf with status in your estimate and contrast statements. As it stands now, you'll get an error.
The part about "some degree of non-independence" refers to modeling correlation between observations. In this case, though, I don't think you have matched pairs or anything along those lines to consider this. If you had some measure of genetic similarity for the subjects who were in the 'inf' and 'UN' groups, you could consider that as a blocking factor, for example.
I think you will be OK once you get a model factor into the contrast and estimate statements, rather than a named level of the factor.
Steve Denham
Thank you for your answer. I did run it with your suggestions and I works well. However, I still would like to know how to test the assumptions of glimmix.
Regards.
The code is:
proc glimmix data=SASUSER.AGEMATURITYMALES_0_1 ODDSRATIO ;
class InfSta;
model Mature = InfSta CentAgeDays CentAgeDaysQ /dist=binomial link=logit solution ;
output out=results pred(ilink)=fit;
contrast 'UN - in' InfSta 1 -1;
estimate 'UN - in' InfSta 1 -1 /cl;
run;
It all depends on what assumptions you wish to test. GLIMMIX has a lot of options, plus there are a variety of plots that you can use to examine assumptions. I think the key I would think of for this is whether or not the results are overdispersed. Check out the examples in the GLIMMIX documentation for pointers.
Steve Denham
Thank you very much. I did find in the SAS webpage the diagnostic plots you mention for glimmix. I will try to work them out with my data. I find this site very helpful.
Best regards,
Build your skills. Make connections. Enjoy creative freedom. Maybe change the world. Registration is now open through August 30th. Visit the SAS Hackathon homepage.
Register today!ANOVA, or Analysis Of Variance, is used to compare the averages or means of two or more populations to better understand how they differ. Watch this tutorial for more.
Find more tutorials on the SAS Users YouTube channel.