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

Question 2 (continued)
What if onlyEs(noCs consid-
ered)?
Answer:Use Method 0.

Why? Confounding not an issue

Initial model withEs andCs?
Screening procedure:it depends!
Option 1:ScreenCs only without
using Method 0
Option 2:ScreenCs without using
Method 0, and screenEs
using Method 0

Following up on a previous question (2), sup-

pose your starting model involves severalEs

but noVsorWs, i.e., noCs are considered.

How do you carry out (one-at-a-time) screen-

ing for this situation?

We recommend using Method 0 here, because

when there are noCs to consider, confounding

is not an issue. Consequently, using statistical

testing for one-at-a-time screening of Esis

appropriate.

Suppose your initial model involves severalEs

andCs. How do you carry out screening for this

situation?The answer is, as is often the case, it

depends!

Option 1: You may decide to screen onlyCs,

and then consider theEs during your modeling

strategy process.

Option 2: If you have large numbers ofEandC

variables, you might screen both types of vari-

ables, making sure not to use Method 0 for the

Cs and using Method 0 for theEs.

268 8. Additional Modeling Strategy Issues