topic Re: proc glimmix Residuals in Statistical Procedures

proc glimmix Residuals

SpanishPhD — Fri, 21 Aug 2015 11:43:05 GMT

Hello everyone,

This is my final survival model for a bird breeding colony, and I would take the residuals. I know that in proc mixed I must write "/outp=" just behind the model, but I do not know what should I use in proc glimmix.

proc glimmix data=Survival INFOCRIT=PQ noreml;

class treartment attempt year nest ring;

model Survival = treatment Year attempt Treatment*attempt size/ dist= binomial link=logit solution cl;

random Year/ subject=ring(nest) type=Cs residual;

lsmeans treatment;

run;

Thanks very much for your help.

Best wishes,

Jaime

Re: proc glimmix Residuals

PaigeMiller — Fri, 21 Aug 2015 12:31:08 GMT

The documentation for PROC GLIMMIX is very helpful. It says you should add an OUTPUT statement

SAS/STAT(R) 13.2 User's Guide

Re: proc glimmix Residuals

SteveDenham — Fri, 21 Aug 2015 14:38:58 GMT

What are you going to do with the residuals from a binomial distribution? Recall that there is no "error term" outside of the model formulation for a generalized linear mixed model that employs a binomial distribution, as there would be for models employing a normal or lognormal distribution. Original scale "residuals" will consist of a set of negative ones, zeroes, and positive ones. These are good for calculating concordances or ROC curves, if that is where you are going.

Steve Denham

Re: proc glimmix Residuals

SpanishPhD — Sun, 23 Aug 2015 19:34:09 GMT

Thank you very much!

In case of not getting the model residuals, I would not integrating the effect of "size". Then... Would it be correct to use only raw data to examine the results graphically without taking into account other significant covariates?

Best whishes,

Jaime