CHAPTER 11

RELATIONSHIPS CAN BE DECEIVING

SECTION 11.1

ILLEGITIMATE CORRELATIONS

FREE RESPONSE QUESTIONS

1. A number of anomalies can cause misleading correlations. Name two problems that can cause
distortion with correlations.
ANSWER: 1) OUTLIERS CAN SUBSTANTIALLY INFLATE OR DEFLATE THEM; 2)
GROUPS COMBINED INAPPROPRIATELY MAY MASK RELATIONSHIPS.

2. Suppose the correlation between height and weight for all the students in your class is found to be
.20. You think that you must be overlooking something, because the relationship should be
stronger than that. What additional variable may be masking an underlying relationship here?
ANSWER: ANY REASONABLE ANSWER OK. EXAMPLE: GENDER.

3. Give an example of a situation where a third variable, if not taken into account, could mask the
relationship between two measurement variables.
ANSWER: ANY REASONABLE ANSWER OK. EXAMPLES: LOOKING AT HEIGHT
VS. WEIGHT FOR ALL ADULTS, GENDER MAY BE MASKING A POTENTIAL
RELATIONSHIP.

4. Explain how (if) an outlier can affect the correlation of a data set.
ANSWER: AN OUTLIER CAN SEVERELY INFLATE OR DEFLATE THE
CORRELATION.

MULTIPLE CHOICE QUESTIONS

5. What impact can an outlier have on a correlation?
a. An outlier that is consistent with the trend of the rest of the data will inflate the
correlation.
b. An outlier that is not consistent with the rest of the data can deflate the correlation.
c. An outlier in a smaller sample has an especially large impact on a correlation, compared
to an outlier in a larger sample.
d. All of the above.
ANSWER: D

6. Which of the following could describe an outlier in a scatterplot?

a. An error in the recording of the data.
b. A gap in the explanatory variable where no data is available, followed by a point where
data is available.
c. A point in the data set whose removal changes the correlation a great deal.
 d. All of the above
ANSWER: D

7. In which case(s) should you be suspicious of a correlation that is presented?

a. When the data is likely to contain outliers.
b. When the sample size is small.
c. When removing one point in the data set actually reverses the direction of the trend.
d. All of the above
ANSWER: D

8. Suppose a correlation is found to be very weak. What does this mean about the relationship
between the two variables?
a. There is no linear relationship between the two variables being measured.
b. There may be separate linear relationships that are being masked by a third variable that
was not accounted for.
c. There may be a different type of relationship between the variables; just not a linear one.
d. All of the above.
ANSWER: D

FILL-IN-THE-BLANK QUESTIONS
9. A third variable that was not accounted for can __________ the relationship between two
measurement variables being studied.
ANSWER: MASK OR DEFLATE

10. A data point that is far removed from the rest of the data is called a(n)__________.
ANSWER: OUTLIER

SECTION 11.2

LEGITIMATE CORRELATION DOES NOT IMPLY CAUSATION

FREE RESPONSE QUESTIONS

11. Suppose two measurement variables are legitimately related or correlated; does this imply that
there is a causal connection between them? Explain why or why not.
ANSWER: NO. IT DEPENDS ON HOW THE DATA WERE COLLECTED.

12. Why, in the absence of any other evidence, can’t you use data from an observational study to
establish a causal link between two measurement variables?
ANSWER: THE SUBJECTS ARE NOT RANDOMLY ASSIGNED TO TREATMENTS;
THERE COULD BE CONFOUNDING VARIABLES.

13. Researchers noticed that happier heart patients are much more likely to still be alive 10 years
down the road than unhappy heart patients. Does this mean that if an unhappy heart patient
suddenly decides to start being happy, this will cause them to live longer?
 ANSWER: NO. THIS IS AN OBSERVATIONAL STUDY, NOT A CONTROLLED
EXPERIMENT. MANY CONFOUNDING VARIABLES EXIST, FOR EXAMPLE,
EXERCISE.

MULTIPLE CHOICE QUESTIONS

14. Which of the following can get in the way of concluding a causal connection between two
measurement variables?
a. A weak correlation.
b. An observational study.
c. Confounding variables.
d. All of the above.
ANSWER: D

15. When is it appropriate to draw a causal connection between two measurement variables?
a. When there is a strong correlation between them.
b. When the data were collected through an observational study.
c. When the observed association between the variables makes sense.
d. None of the above.
ANSWER: D

16. Which of the following statements is true?

a. Legitimate correlation never implies causation.
b. Legitimate correlation does not necessarily imply causation.
c. Legitimate correlation is equivalent to causation.
d. Legitimate correlation implies causation in the case of a single observational study, as
long as the researchers tried to control for confounding variables.
ANSWER: B

SECTION 11.3

SOME REASONS FOR RELATIONSHIPS BETWEEN VARIABLES

FREE RESPONSE QUESTIONS

17. Name three (of the seven) reasons two variables could be related.
ANSWER: ANY THREE OF THE FOLLOWING ARE OK: 1) THE EXPLANATORY
VARIABLES IS THE DIRECT CAUSE OF THE RESPONSE VARIABLE; 2) THE
RESPONSE VARIABLE IS CAUSING A CHANGE IN THE EXPLANATORY
VARIABLE; 3) THE EXPLANATORY VARIABLE IS A CONTRIBUTING CAUSE OF
THE RESPONSE VARIABLE; 4) CONFOUNDING VARIABLES MAY EXIST; 5) BOTH
VARIABLES MAY RESULT FROM A COMMON CAUSE; 6) BOTH VARIABLES ARE
CHANGING OVER TIME; OR 7) COINCIDENCE.

18. Give an example where the explanatory variable is the direct cause of the response variable.
ANSWER: ANY SIMILAR EXAMPLE TO THE FOLLOWING IS OK: AMOUNT OF
FOOD CONSUMED IN THE PAST HOUR AND LEVEL OF HUNGER.

19. Give an example where the response variable is causing a change in the explanatory variable.
 ANSWER: ANY SIMILAR EXAMPLE TO THE FOLLOWING IS OK: LOW
OCCUPANCY RATES IN HOTELS CAUSE HIGHER ADVERTISING REVENUES
(RATHER THAN THE OTHER WAY AROUND).

20. Give an example where the explanatory variable is a contributing (but not the only) cause of the
response variable.
ANSWER: ANY SIMILAR EXAMPLE TO THE FOLLOWING IS OK: DIET AND
DEVELOPING A CERTAIN TYPE OF CANCER.

MULTIPLE CHOICE QUESTIONS

21. Which of the following is a confounding variable for the relationship between happiness and
length of life?
a. Happiness level
b. Length of life
c. Emotional support
d. Age at death
ANSWER: C

22. There is a strong correlation between verbal SAT scores and college GPAs. This does not mean
however, that higher SAT scores cause higher grades in college. But what could explain this
relationship?
a. The confounding variable gender.
b. The high (low) SAT scores and high (low) GPAs both result from a common cause.
c. Both SAT scores and GPAs change over time.
d. The relationship is purely coincidental.
ANSWER: B

23. If two measurement variables are both found to be changing over time, what does this mean?
a. It means there is a causal link between the two variables.
b. It means there is a common cause of the changes in both variables (possibly other than
time).
c. It means that even though the two variables may be highly correlated, they could be
completely unrelated in terms of cause and effect.
d. None of the above.
ANSWER: C

24. Suppose within a short span of time, people in a certain town realize that there is an unusually
high rate of leukemia among children in their town. Suppose someone calculated the odds of
having that many cases in such a short time to be 1 in 15,000. Which of the following immediate
conclusions would not be correct?
a. Since the chance is only 1 out of 15,000, something in this town has to be causing our
children to get sick. We need to take immediate action.
b. We should expect this phenomenon to happen in about 1 out of every 15,000 towns
similar to this one, just by chance. There is no cause for alarm just yet.
c. It would be unusual if we did not occasionally see clusters of disease such as this one;
this town may just be unlucky. Let’s wait until more data is collected.
d. We should keep an eye on the situation but there is no reason to panic. This could very
well be just a coincidence.
ANSWER: A

FILL-IN-THE-BLANK QUESTIONS
 25. One of the ways two variables could be related without a causal connection is the existence of
__________ variables.
ANSWER: CONFOUNDING

26. Sometimes an association between two variables is due to __________, even though the odds of it
happening appear to be very small.
ANSWER: COINCIDENCE

SECTION 11.4

CONFIRMING CAUSATION

FREE RESPONSE QUESTIONS

27. What is the only legitimate way to try to establish a causal connection statistically between two
measurement variables?
ANSWER: THROUGH THE USE OF RANDOMIZED EXPERIMENTS.

28. Even in a well designed experiment, there will always be confounding variables that we have
neglected to measure. How can we reduce the chances that an observed association is due to those
confounding variables? Name two ways.
ANSWER: 1) USE A LARGE SAMPLE OF SUBJECTS; 2) RANDOMLY ASSIGN
SUBJECTS TO TREATMENTS.

29. Give an example where a randomized experiment cannot be done, even though we know that is
the best way to try to establish a causal connection between two measurement variables.
ANSWER: ANY REASONABLE ANSWER OK. EXAMPLES: DOES SMOKING CAUSE
LUNG CANCER?

30. If a randomized experiment cannot be done, then four non-statistical considerations must be used
(all together) to establish evidence of a possible causal connection between two measurement
variables. Name them.
ANSWER: 1) THERE IS A REASONABLE EXPLANATION OF CAUSE AND EFFECT;
2) THE CONNECTION HAPPENS UNDER VARYING CONDITIONS;3) POTENTIAL
CONFOUNDING VARIABLES ARE RULED OUT; 4) THERE IS A “DOSE-RESPONSE”
RELATIONSHIP.

MULTIPLE CHOICE QUESTIONS

31. Which of the following will strengthen the evidence for a causal connection?
a. Many observational studies conducted under different conditions all find the same link
between two variables.
b. Many observational studies with different confounding variables all find the same link
between the two variables.
c. The same type of relationship holds when the explanatory variables fall into different
ranges for different studies.
d. All of the above.
ANSWER: D
32. Which of the following, if left to stand alone, is the weakest evidence of a possible causal
connection?
a. There is a reasonable explanation for a cause and effect relationship.
b. The data appear to have a pattern on the scatterplot.
c. The connection was shown to hold under varying conditions.
d. Potential confounding variables have been ruled out.
ANSWER: B

33. Suppose you heard on the radio that women who overuse antibiotics have a higher chance of
developing breast cancer. You look into the literature on this, and find ten observational studies
done by different researchers under different conditions, all of which confirm the results that you
heard on the radio. What do you conclude?
a. The evidence for a causal connection between overuse of antibiotics and increased risk of
breast cancer is strengthened by these varying studies.
b. These studies have too many different confounding variables that together weaken the
evidence for a causal connection.
c. Since the studies were all done under different conditions, there is not enough
information to make a conclusion.
d. None of the above.
ANSWER: A

34. Which of the following cannot be obtained from observational studies?

a. Definitive evidence of a causal connection.
b. A reasonable explanation for a causal connection.
c. Any evidence of a causal connection.
d. Definitive evidence of a correlation.
ANSWER: A

FILL-IN-THE-BLANK QUESTIONS
35. The __________ the number of confounding factors that can be ruled out, the __________
convincing the evidence for a causal connection.
ANSWERS (RESPECTIVELY): GREATER AND MORE; OR FEWER AND LESS.

36. It is very difficult to establish a causal connection between two variables without the use of
anything except a __________.
ANSWER: RANDOMIZED EXPERIMENT