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

Chapters

1. Introduction


2. Special Cases
   
   

3 7. Interaction and
Confounding Assessment

8. Additional Modeling
   Strategy Issues

This presentation is now complete. The reader

may wish to review the detailed summary and

to try the practice exercises and test that

follow.

The next chapter considers additional issues

about modeling strategy, including how to

address more than one exposure variable,

screening variables, collinearity, multiple test-

ing, and influential observations.

SUMMARY (continued)

Interaction present: compare tables

of odds ratios and confidence

intervals (subjective)

Interaction: Safe (for validity) to

keep allVs in model

However, if there is interaction, the assess-

ment of confounding is much more subjective

because it typically requires the comparison

of tables of odds ratio values. Similarly, asses-

sing precision requires comparison of tables

of confidence intervals.

Consequently, if there is interaction, it is typi-

cally safe for ensuring validity to keep all

potential confounders in the model, even

those that are candidates to be deleted as

possible nonconfounders.

232 7. Modeling Strategy for Assessing Interaction and Confounding