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

State or describe at least three issues/problems that would

not have been addressed by the above screening approach.

4. Consider the following logistic regression model in
   which all predictor variables are (0, 1) variables:
   logit PðXÞ¼aþbEþg 1 C 1 þg 2 C 2 þg 3 C 3 þg 12 C 1 C 2
   þd 1 EC 1 þd 2 EC 2 þd 3 EC 3 þd 12 EC 1 C 2

Determine which of the following statements areTrue or

False (Circle T or F).

T F a. Suppose that using the collinearity macro

described in the text results in the three highest

condition indices (CNIs) being 77, 56, and 47.

Then, using a sequential approach for diagnos-

ing collinearity, there is at least 1 and possibly 3

collinearity problems associated with fitting this

model.

T F b. Suppose that using the collinearity macro

described in the text, the highest condition

index (CNI) is found to be 77 and the only

VDPs that are determined to be high for this

CNI are associated with the variablesEandC 3.

Then it is reasonable to dropC 3 from the model,

and recompute collinearity diagnostics to fur-

ther reassess collinearity for the reduced model.

T F c. Suppose that using the collinearity macro

described in the text, the three highest condition

indices (CNIs) are 77, 56, and 47. For the highest

CNI of 77, suppose there are exactly four VDPs

other than the intercept that are larger than .5.

Suppose also that the variables associated with

these four VDPs areC 1 ,C 3 ,EC 3 , andEC 1 C 2. Then

this collinearity problem can be addressed by

dropping eitherEC 3 orEC 1 C 2 from the model,

but notC 1 ,C 2 orC 3 , after which a reduced model

is fit to see if there are additional collinearity

problems.

T F d. Suppose that for the same situation as described

in part c above, the collinearity problem is

addressed by dropping EC 3 from the model,

after which a reduced model withoutEC 3 is fit.

Then the collinearity diagnostics obtained for

the reduced model will indicate another collin-

earity problem that involves the variableEC 1 C 2.

5. Suppose for the data analyzed using the model stated in
   question 4, it was desired to evaluate whether or not
   any subjects were influential observations.
   a. Assuming that the study objective was to assess the
   effect ofEcontrolling forC 1 ,C 2 , andC 3 , how can

292 8. Additional Modeling Strategy Issues