from the model as possible nonconfounders? Briefly
explain your answer.- Based on the interaction assessment results described
in question 6, is it appropriate to test the significance
for the main effect ofPOST and/or OCC? Explain
briefly. - a. State the logit formula for the reduced model
obtained from the interaction results described in
question 6.
b. Based on your answer to 8 a, give a formula for the
odds ratio that compares the odds for improvement
of pain both in bed and upon rising to no improve-
ment for a subject getting a firm mattress (F¼3)
and a firm base (BASE¼1) to the corresponding
odds for a subject getting a medium firm (F¼2)
mattress and an infirm base (BASE¼0),
controlling forPOST, PF, OCC, AGE, andGEN. - Assume that the odds ratio formula obtained in ques-
tion 8 represents the gold standard odds ratio for
describing the relationship of mattress type (F) and
mattress base (BASE) to pain improvement controlling
forPOST, OCC, PF, AGE, andGEN, i.e.,your only
interestis theORcomparing (F¼3,BASE¼1) with
(F¼2, BASE¼0). One way to assess confounding
among the variables eligible to be dropped as noncon-
founders is to compare tables of odds ratios for each
subset of possible confounders to the gold standard
odds ratio.
a. How many subsets of possible confounders (other
than the set of possible confounders in the gold
standard odds ratio) need to be considered?
b. Describe what a table of odds ratios would look like
for any of the subsets of possible confounders, i.e.,
draw such a table and specify what quantities go
into the cells of the table.
c. How would you use the tables of odds ratios
described above to decide about confounding?In
your answer, describe any difficulties involved.
d. Suppose you decided thatPFandGENcould be
dropped from the model as nonconfounders, i.e.,
your reduced model now containsF, BASE, POST,
OCC, AGE, F 3 BASE, F 3 POST, and BASE 3
OCC:Describe how you would determine whether
precision was gained when droppingPFandGEN
from the model. - Suppose that after all of the above analyses described
in the previous questions, you realized that you
296 8. Additional Modeling Strategy Issues