a. False: Can’t tell until you check VDPs. Possible that
all VDPs are not high (i.e., much less than 0.5)
b. False: Model won’t be HWF if C 3 is dropped.
c. True
d. False: May not be any remaining collinearity prob-
lem onceEC 3 is dropped.
a. A Cook’s distance-type measure combines the infor-
mation from all estimated regression coefficients in
one’s model, whereas it would be preferable to con-
sider either the Db orDexp[b] for theEvariable
alone, since theEvariable is the primary variable
of interest.
b. You should not automatically drop a subject from
the dataset just because you have identified it as
influential. A conservative approach is to drop only
those subjects whose data are clearly in error and
cannot be corrected.
c. The model of question 4 may not be the best model,
so that different conclusions might result about
which subjects are influential if a different (“best”)
model were used instead.
The number of tests to be performed cannot be deter-
mined in advance of the modeling process, i.e., it is not
clear whatTwill be for the individual significance level
of 0.05/T.
