- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Posted 05-16-2011 05:41 PM
(1437 views)
I am running a Mixed Ordinal logistic regression model, with some fixed effects and one hierarchical random effect.
I understand that Proc Glimmix could run this model, however, we are requested to report the marginal probabilities and standard errors for the marginal probabilities. Glimmix only gives marginal probability but not the error for marginal probability, does anyone know how to get the standard error?
Thanks.
I understand that Proc Glimmix could run this model, however, we are requested to report the marginal probabilities and standard errors for the marginal probabilities. Glimmix only gives marginal probability but not the error for marginal probability, does anyone know how to get the standard error?
Thanks.
2 REPLIES 2
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I think you could get the standard error you want by fitting the model using PROC NLMIXED instead of PROC GLIMMIX. Use the PREDICT statement to estimate the desired statistic and obtain its standard error.
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I assume that your proc glimmix looks something like:
proc glimmix data=datasetname;
class indepvar randomvar;
model depvar=indepvar/dist=binomial link=logistic;
random randomvar;
run;
(probably some other lines in there, but these are the critical ones)
You could add:
lsmeans indepvar/cl ilink;
ods output lsmeans=lsmeans;
This should give the "back-transformed" marginal means (which are the probabilities) and their standard errors. Note that this will NOT work for a multinomial model (which may be what you have with ordinal data), as the lsmeans statement is not supported for dist=multinomial.
Good luck,
Steve Denham
proc glimmix data=datasetname;
class indepvar randomvar;
model depvar=indepvar/dist=binomial link=logistic;
random randomvar;
run;
(probably some other lines in there, but these are the critical ones)
You could add:
lsmeans indepvar/cl ilink;
ods output lsmeans=lsmeans;
This should give the "back-transformed" marginal means (which are the probabilities) and their standard errors. Note that this will NOT work for a multinomial model (which may be what you have with ordinal data), as the lsmeans statement is not supported for dist=multinomial.
Good luck,
Steve Denham