15.33 I was surprised to see that frequency of the behavior was not related to the likelihood of
reporting. Can you suggest reasons why this might be so?
15.34 Malcarne, Compas, Epping, and Howell (1995) examined 126 cancer patients soon after they
were diagnosed with cancer and at a four-month follow-up. At the initial interviews (Time 1)
they collected data on the patients‘ current levels of distress (Distress1), the degree to which
they attributed the blame for the cancer to the type of person they are (BlamPer), and the degree
to which they attributed the cancer to the kind of behaviors in which they had engaged, such as
smoking or high fat diets (BlamBeh). At the four-month follow-up (Time 2) the authors again
collected data on the levels of psychological distress that the patients reported (Distress2).
(They also collected data on a number of other variables, which do not concern us here.) The
data are available on the Web site for this course and named Malcarne.dat.
a. What would you conclude if you attempt to predict Distress2 from Distress1 and
BlamPer?
b. Why would I want to include Distress1 in the analysis for part a?
15.35 In Exercise 15.4 we had a data set where BlamBeh was related to later distress at time 2.
When it is included as a predictor along with Stress1 and BlamPer it is no longer a signifi-
cant predictor. Why would this be likely to happen?
15.36What are some of the reasons why stepwise regression (broadly defined) would not find
favor with most statisticians?
15.37 Make up a very simple example with very simple variables to illustrate how one could see
the effect of an interaction between two predictors.
15.38Using the data you created in Exercise 15.37, demonstrate the effect of “centering” your
predictor variables.
15.39 As you know, the regression coefficient gives the effect of one variable holding all other
variables constant. How would you view this interpretation when you have an interaction
term in your model?
15.40 Paul Jose has a Web site referred to in the section on mediation. He discusses a problem in
which he believes that stress leads to depression through a mediating path via rumination.
(In other words, some stressed people ruminate, and as a consequence they become de-
pressed.) The path diagram derived from his analysis of actual data are shown below. The
beta given for the path from stress to depression is from the multiple regression of depres-
sion on stress and rumination. Predicting depression only from stress had a beta of .471.
Test the decline in the coefficient for the direct path from stress to depression using Sobel’s
test. (You can check your work at Jose’s Web site at http://www.victoria.ac.nz/psyc/staff/
paul-jose-files/helpcentre/help7_mediation_example.php, though the answers will not be
exactly equal.
15.41 In this chapter we spent a lot of time with Guber’s study of educational expenditures and
found that when we controlled for the percentage of students taking the SAT exam, Expend
was not a significant predictor. However, the SAT is not a good dependent variable in dis-
cussing the quality of education in a state. Perhaps the ACT, which tests something some-
what different, is a better predictor. Use the data set, which is available on the book’s Web
site and named Tab15-1.dat, to answer that question. Be complete in your answer, examin-
ing the individual variables and the residuals.
Exercises 577RuminationStress Depression5 .478
se 5 .017b5.321
se 5 .022bb 5 .3208