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

assess whether or not any subjects in the dataset

are influential observations. Make sure to indicate

whether you would prefer to use DeltaBeta mea-

sures or Cook’s distance-type measures, both types

of measures, or other measures.

b. Why would it be questionable to automatically

drop from your dataset any subjects that you find

to be influential observations?

12. Suppose in your analysis strategy for determining a
    “best” model, you want to reduce the number of sta-
    tistical tests that you perform. What approach(es) can
    you use? Why are you not able to adequately carry out
    a Bonferroni-type of adjustment procedure?

Answers to Practice Exercises

Practice Exercises

Practice Exercises

1. a. True
   b. False: E 2 E 3 not needed.
   c. False: E 1 C 2 also a candidate
   d. True
   e. False: Incorrect use of backward elimination.


2. a. ORd¼exp½b^ 1 þ^b 2 þ^b 3 þ^d 11 C 1 þ^d 22 C 2 Š
   b.
   C 2 ¼ 1 C 2 ¼ 0
   C 1 ¼ 1 dOR 11 ORd 10

C 1 ¼ 0 dOR 01 ORd 00

dORab¼exp½^b 1 þ^b 2 þb^ 3 þ^d 11 ðC 1 Þþ^d 22 ðC 2 ފ

c. i. 4

ii. {C 3 ,C 1 C 2 }{C 3 }{C 1 C 2 } {NeitherC 3 norC 1 C 2 }

d. Models 16, 7, 4a

e. Obtain tables of confidence intervals for the odds

ratios for each of the three models stated in part d.

Choose as the best model either:

i. the model with the narrowest confidence

interval

ii. the gold standard model (16) if all four models

have approximately the same width.

iii. the more parsimonious model

3. i. Confounding for individualCs not addressed
   by statistical testing.
   ii. Interaction of individualCs with eachEinot
   addressed
   iii. Interaction ofEiwithEjnot addressed
   iv. Screening ofCs were not distinguished from
   screening ofEs.

298 8. Additional Modeling Strategy Issues