Logistic Regression: A Self-learning Text, Third Edition (Statistics in the Health Sciences)

GEE vs. Standard Logistic
Regression

SLR equivalent to GEE model
with:

1. Independent correlation
   structure


2. fforced to equal 1


3. Model-based standard errors

Chapter 15: GEE Examples

With an SLR, there is an assumption that

each observation is independent. By using an

independent correlation structure, forcing

the scale factor to equal 1, and using model-

based rather than empirical standard errors

for the regression parameter estimates, we

can perform a GEE analysis and obtain results

identical to those obtained from a standard

logistic regression.

We suggest that you review the material cov-

ered here by reading the detailed outline that

follows. Then, do the practice exercises and

test.

In the next chapter (Chap. 15), examples are

presented to illustrate the effects of selecting

different correlation structures for a model

applied to a given dataset. The examples are

also used to compare the GEE approach with

a standard logistic regression approach in

which the correlation between responses is

ignored.

XIII. SUMMARY

üChapter 14: Logistic Regression
for Correlated
Data: GEE

Thepresentationisnowcomplete.Wehave

described one analytic approach, the GEE

model, for the situation where the outcome

variable has dichotomous correlated res-

ponses. We examined the form and interpre-

tation of the GEE model and discussed a

variety of correlation structures that may

be used in the formulation of the model. In

addition, an overview of the mathematical

theory underlying the GEE model has been

presented.

528 14. Logistic Regression for Correlated Data: GEE