any further model considered. TheVvariables that
must remain in further models are NS, AS, NSAS,
and, of course, the exposure variable SMK. Also theV*
variables must remain in all further models because
these variables reflect the matching that has been
done.- The model after interaction assessment is the same as
the initial model. No potential confounders are
eligible to be dropped from the model because NS,
AS, and NSAS are lower components of
SMKNSAS and because theV*variables are
matching variables.
Chapter 7 1. The interaction terms are SMKNS, SMKAS, and
SMKNSAS. The product term NSAS is aVterm,
not an interaction term, because SMK is not one of its
components.
- Using a hierarchically backward elimination strategy,
one would first test for significance of the highest-order
interaction term, namely, SMKNSAS. Following
this test, the next step is to evaluate the significance of
two-factor product terms, although these terms might
not be eligible for deletion if the test for SMKNSAS
is significant. Finally, without doing statistical testing,
theVvariables need to be assessed for confounding and
precision. - If SMKNS is the only interaction found significant,
then the model remaining after interaction assessment
contains theV terms, SMK, NS, AS, NSAS, and
SMKNS. The variable NS cannot be deleted from any
further model considered because it is a lower order
component of the significant interaction term SMK
NS. Also, theV terms cannot be deleted because these
terms reflect the matching that has been done. - The odds ratio expression is given by exp(bþd 1 NS).
- The odds ratio expression for the model that does not
contain NSAS has exactly the same form as the
expression in Question 4. However, the coefficientsb
andd 1 may be different from the Question 4 expression
because the two models involved are different. - Drop NSAS from the model and see if the estimated
odds ratio changes from the gold standard model
remaining after interaction assessment. If the odds
ratio changes, then NSAS cannot be dropped and is
considered a confounder. If the odds ratio does not
change, then NSAS is not a confounder. However, it
may still need to be controlled for precision reasons. To
assess precision, one should compare confidence
Chapter 7 675