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

11. Model II is more appropriate than Model I if the test
    for the effect of interaction is viewed as significant.
    Otherwise, Model I is more appropriate than Model II.
    The decision here is debatable because the test result
    is of borderline significance.

Chapter 6 True–False Questions:

1. F: one stage is variable specification


2. T


3. T


4. F: no statistical test for confounding


5. F: validity is preferred to precision


6. F: for initial model,Vs chosen a priori


7. T


8. T


9. F: model needsEBalso


10. F: list needs to includeAB


11. The given model is hierarchically well formulated
    because for each variable in the model, every lower
    order component of that variable is contained in the
    model. For example, if we consider the variable
    SMKNSAS, then the lower order components are
    SMK, NS, AS, SMKNS, SMKAS, and NSAS; all
    these lower order components are contained in the
    model.


12. A test for the term SMKNSAS is not dependent
    on the coding of SMK because the model is
    hierarchically well formulated and SMKNSAS is
    the highest-order term in the model.


13. A test for the terms SMKNS is dependent on the
    coding because this variable is a lower order term in
    the model, even though the model is hierarchically
    well formulated.


14. In using a hierarchical backward elimination
    procedure, first test for significance of the highest-
    order term SMKNSAS, then test for significance
    of lower order interactions SMKNS and SMKAS,
    and finally assess confounding forVvariables in the
    model. Based on the hierarchy principle, any two-
    factor product terms andVterms which are lower
    order components of higher order product terms
    found significant are not eligible for deletion from the
    model.


15. If SMKNSAS is significant, then SMKNS and
    SMKAS are interaction terms that must remain in

674 Test Answers