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

the end of the interaction stage? Which of theVvari-

ables in the model cannot be deleted from any further

models considered? Explain briefly your answer to the

latter question.

4. Based on the scenario described in Question 3 (i.e., the
   only significant interaction term is SMKNS), what is
   the expression for the odds ratio that describes the
   effect of SMK on cervical cancer status at the end of
   the interaction assessment stage?


5. Based again on the scenario described in Question 3,
   what is the expression for the odds ratio that describes
   the effect of SMK on cervical cancer status if the vari-
   able NSAS is dropped from the model that remains
   at the end of the interaction assessment stage?


6. Based again on the scenario described in Question 3,
   how would you assess whether the variable NSAS
   should be retained in the model? (In answering this
   question, consider both confounding and precision
   issues.)


7. Suppose the variable NSAS is dropped from the
   model based on the scenario described in Question 3.
   Describe how you would assess confounding and preci-
   sion for any other V terms still eligible to be deleted
   from the model after interaction assessment.


8. Suppose the final model obtained from the cervical
   cancer study data is given by the following printout
   results:

Variable b S.E. Chi sq P

SMK 1.9381 0.4312 20.20 0.0000

NS 1.4963 0.4372 11.71 0.0006

AS 0.6811 0.3473 3.85 0.0499

SMKNS 1.1128 0.5997 3.44 0.0635

Describe briefly how you would use the above informa-

tion to summarize the results of your study. (In your

answer, you need only describe the information to be

used rather than actually calculate numerical results.)

Answers to Practice Exercises

Practice Exercises

1. A “chunk” test for overall significance of interaction
   terms can be carried out using a likelihood ratio test
   that compares the initial (full) model with a reduced
   model under the null hypothesis of no interaction
   terms. The likelihood ratio test will be a chi-square
   test with two degrees of freedom (because two inter-
   action terms are being tested simultaneously).

Answers to Practice Exercises 237