Programming the statistical procedures from SAS

proc glimmix fixed effects solution for logistic regression

Accepted Solution Solved
Reply
N/A
Posts: 1
Accepted Solution

proc glimmix fixed effects solution for logistic regression

Hi community,

I have question about solution of fixed effects in proc glimmix. I have data with following class variables:

ClassLevelsValues
group31 2 3
method3A B C

and one binary variable "y". I have following code:

PROC glimmix data=tmp1.data1;

class group method;

model y=method / solution;

random int / subject=group;

run;

As the result of "solution" i have following table(here is part of it):

effectgroupestimate
intercept

0.25

methodA0.3333
methodB

0.083333

methodC0

My question is: Why the result of effect method: C is equal 0? Why does it estimate the intercept as the Beta_0+Beta_C, method:B as a Beta_B-Beta_C and so on..?


regards


Accepted Solutions
Solution
‎06-19-2015 03:24 PM
Valued Guide
Valued Guide
Posts: 684

Re: proc glimmix fixed effects solution for logistic regression

THis is what is supposed to happen. In GLIMMIX (also MIXED, GLM, etc.) an overparameterized model is used.This is needed to properly test for factor effects and get expected values. With three factor levels, you need a model with three parameters, but the overparameterized model has four parameters (intercept, beta_A, beta_B, and beta_C). The last one ends up as 0 in the estimation. For one factor, this means that the intercept is the expected value for the last level (C), and the other parameters are differences from C. For instance, the expected value for A is intercept + beta_A. This is described in the User's Guide.

View solution in original post


All Replies
Solution
‎06-19-2015 03:24 PM
Valued Guide
Valued Guide
Posts: 684

Re: proc glimmix fixed effects solution for logistic regression

THis is what is supposed to happen. In GLIMMIX (also MIXED, GLM, etc.) an overparameterized model is used.This is needed to properly test for factor effects and get expected values. With three factor levels, you need a model with three parameters, but the overparameterized model has four parameters (intercept, beta_A, beta_B, and beta_C). The last one ends up as 0 in the estimation. For one factor, this means that the intercept is the expected value for the last level (C), and the other parameters are differences from C. For instance, the expected value for A is intercept + beta_A. This is described in the User's Guide.

🔒 This topic is solved and locked.

Need further help from the community? Please ask a new question.

Discussion stats
  • 1 reply
  • 264 views
  • 0 likes
  • 2 in conversation