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

problems, e.g., collinearity; simplest choice is to

letVsbeCs themselves.

C. ChooseWs fromCs to be eitherVs or product of

twoVs; usually recommendWstobeCs

themselves or some subset ofCs.

V. Causal diagrams(pages 175–179)

A. The approach for variable selection should

consider causal structure.

B. Example of causal diagram: Smoking!Lung

Cancer!Abnormal Chest X-ray.

C. Controlling for X-ray status in above diagram

leads to bias

D. Depending on the underlying causal structure,

adjustment may either remove bias, lead to

bias, or be appropriate.

E. Causal diagram for confounding:Cis a

common cause ofEandD

C

D

E Cisacommon causeof E and D;

The path E–C–D is a (noncausal)

backdoor pathfrom E to D

F. Other types of causal diagrams:
i. Fis a common effect ofEandD;
conditioning onFcreates bias; Berkson’s
bias example.
ii. Example involving unmeasured factors.
G. Conditioning on a common cause can remove
bias, whereas conditioning on a common effect
can cause bias.
VI. Other considerations for variable specification
(pages 180–181)
A. Quality of the data: measurement error or
misclassification?
B. Qualitative collinearity, e.g., redundant
covariates.
C. Sample size
D. Complexity vs. simplicity
E. Know your data! Perform descriptive analyses.
VII. Hierarchically well-formulated models(pages
181–184)
A. Definition: given any variable in the model, all
lower order components must also be in the
model.
B. Examples of models that are and are not
hierarchically well-formulated.
C. Rationale: If the model is not hierarchically
well-formulated, then tests for significance of
the highest-order variables in the model may
change with the coding of the variables tested;
such tests should be independent of coding.

Detailed Outline 195