EXAMPLE
Examples of Screening: SingleE,C 1 ,
C 2 ,...,C 10.
Four Scenarios: Which of these is
“legitimate”?
i. Crude analyses relatingD
to eachCiidentify only
C 1 andC 4 to be significant
predictors ofD. Starting
model then containsE,C 1 ,
C 4 ,EC 1 ,andEC 4. Best
model determined using
hierarchical backward
elimination approach
(HBWE) outlined in
Chap. 6
ii. Stratified analyses relating
DtoEand eachCiidentify
C 1 andC 4 to be individual
confounders, andC 5 to be
an effect modifier of the
E,Deffect. Starting model
then containsE,C 1 ,C 4 ,C 5 ,
EC 1 ,EC 4 ,andEC 5 .Best
model determined using
HBWE.
iii. Crude analyses relatingD
to eachCiidentify onlyC 1
andC 4 to be significant
predictors ofD. Starting
model then containsE,C 1 ,
andC 4. Backward
elimination onCs
eliminatesC 4 , but retains
C 1 (andE). Add interaction
termEC 1. Best model
determined using HBWE.
iv. Logistic regression models
relatingDtoEand each
CiidentifyC 1 andC 4 to be
individual confounders,
andC 5 to be an effect
modifier of theE,Deffect.
Starting model then
containsE,C 1 ,C 4 ,C 5 ,EC 1 ,
EC 4 , andEC 5. Best model
determined using HBWE.
8
>>
>>>
>>
>>
>>
>>
>>>
>>
>>
>>
>>
>>>
>>
>>
>>
>>>
>>
>>
>>
>>
>>>
>>
>>
>>
>>
>>>
>>
>>
>>
>>
>>>
>>
>>
>>
>>>
>>
>>
<
>>>
>>
>>
>>>
>>
>>
>>
>>
>>>
>>
>>
>>
>>
>>>
>>
>>
>>
>>
>>>
>>
>>
>>
>>
>>>
>>
>>
>>
>>>
>>
>>
>>
>>
>>>
>>
>>
>>
>>
:
We now provide a few simple examples to illus-
trate screening. We will begin by assuming that
we have a singleEvariable and 10Cvariables,
C 1 ,C 2 ,...,C 10. At the left, we describe four
different screening scenarios for this situation.
Can you determine which of these scenarios
corresponds to carrying out Method 0, and
which represents what we have described
above as a “legitimate” method of screening?
Presentation: III. Screening Variables 269
