neglected to check for the possibility ofcollinearityin
any of the models you considered so far.
a. Assuming that all the models you previously con-
sidered ran using the SAS’s LOGISTIC procedure
(i.e., they all produced output without any error or
warning messages), why should you still be
concerned about the possibility of collinearity?Suppose, further, that you use SAS’s LOGISTIC procedure
to fit a model containing the independent variables F,
BASE, POST, OCC, PF, AGE, GEN, all two-way product
terms involvingFwith the control variables (POST, OCC,
PF, AGE, GEN), and all two-way product terms involving
BASEwith these same control variables.
b. You now run the collinearity macro with the above
logistic model, and you find that there are three
condition indices with values 97, 75, and 62, with
all other condition indices less than 25. How would
you proceed “sequentially” to use this information
to assess collinearity?
c. Suppose the condition index of 97 has “high VDP
values” on the variablesF, BASE, PF, and F 3
BASE. Does this result cause you difficulty in
accepting your previous assessment of interaction
in which you found that the variablesF 3 BASE,
F 3 POST, andBASE 3 OCCneeded to remain in
the model as significant interaction effects?
Explain.
d. Based on the collinearity results in part 10 b and c,
which of the following choices is an appropriate
next step?(Circle the “best” choice)
i. Drop the product termF 3 BASEfrom the
polytomous logistic regression model and redo
the hierarchical backward elimination strategy
without further consideration of collinearity.
ii. Determine whether the next highest condition
index of 75 corresponds to high VDP loadings
two or more predictors.
iii. Drop the product termF 3 BASEfrom the
logistic regression model, and apply collinearity
diagnostics to the reduced model to determine
if there is an additional collinearity problem.
iv. Ignore the collinearity diagnostics results and
use the model obtained from the hierarchical
backward elimination strategy previously used.- a. Assuming that mattress type (F) and type of base
(BASE) are the only two exposures of interest, with
PF, GEN, POST, OCC,andAGE considered as
control variables, briefly outline how you would
Test 297