Activity 1-1 5

experience suggests that many students misunderstand these terms. Teaching tip:
Wikipedia ([Link]) has a good summary and some links for more

1
information on Kristen Gilbert.

••• Homework Activities

In this topic, the homework activities provide students with more practice identifying and
classifying variables and also with encountering research questions that can be addressed
with data. Many students find the most challenging questions are the ones that ask them
to suggest variables or research questions, so you might want to assign at least one of
these. Examples of this type of question are Activities 1-9 (part b), 1-10, and 1-17. Also
note that many of the studies mentioned in these homework activities recur later in the
book, so you might want to consider assigning these as a series. Finally, note that Activity
1-13 provides a very early preview of the concept of a sampling distribution.
Possible assignment: Activities 1-7 (identifying observational units and variables),
1-11 (these ideas within the context of a study), and 1-12 (legitimate and not so
legitimate variables). Activity 1-17 points ahead to thinking about research questions
by asking students to decide the variable to measure in a given context.

Solutions
••• In-Class Activities
Activity 1-1: Student Data
a. Answers will vary.
b. Answers will vary.
c. No, every word did not contain the same number of letters.
d. The variable is the number of letters in each word.
e. • How many hours you slept in the past 24 hours: quantitative
• Whether you slept for at least 7 hours in the past 24 hours: binary categorical
• How many states you have visited: quantitative
• Handedness: binary categorical, unless you classify “ambidextrous,” in which
case it is not binary.
• Day of the week on which you were born: categorical
• Gender: binary categorical
• Average study time per week: quantitative
• Score on the first exam in this course: quantitative
f. No, neither average height of students in the class nor percentage of students in the
class who have used a cell phone today can legitimately be considered variables when
the observational units are the students in your class. Both of these are numbers
that provide summary information about the class as a whole. They do not vary
from student to student.
g. If you record the average student height or percentage of student cell phone usage
by class taught at your school, these would become legitimate variables. Now these
numbers would (potentially) take on different values from class to class. The
observational units are no longer the students in your class, but rather all classes
taught at your school.
6 Topic 1: Data and Variables

Activity 1-2: Variables of State

a. Binary categorical variable
b. Not a variable
c. Quantitative variable
d. Quantitative variable
e. Binary categorical variable
f. Quantitative variable
g. Not a variable

Activity 1-3: Cell Phone Fraud

a. The observational units are the cell phone calls.
b. The binary categorical variables are direction, location, and whether the call took
place on a weekday or weekend. The non-binary categorical variable is day of week.
c. The quantitative variables are duration of the call and time of day.

Activity 1-4: Studies from Blink

a. Observational units: 100 CEOs
Variable: height Type: quantitative

b. Observational units: 50 marriage counselors

Variable: whether the counselor Type: (binary) categorical
makes the correct prediction about
whether a couple will still be married
in five years

c. Observational units: 200 African-American college students

Variable 1: whether their version of the Type: (binary) categorical
exam asks them to indicate race
Variable 2: score on SAT-like exam Type: quantitative

d. Observational units: 10 car dealerships

Variable 1: gender of customer Type: (binary) categorical
Variable 2: race of customer Type: (binary) categorical
Variable 3: price negotiated for the car Type: quantitative

Activity 1-5: Student Data

a. Many answers are possible, but some examples include these: Do male and female
students differ with regard to the number of states they have visited? Are sleeping
times associated with the day of the week on which a student was born?
b. Answers will vary.
 Activity 1-7 7

Activity 1-6: A Nurse Accused

a. The observational units are the eight-hour shifts.

1
b. One variable is whether Gilbert worked on the shift. This variable is categorical and
binary. The other variable is whether a patient died on the shift. This variable is
also categorical and binary.

••• Homework Activities

Activity 1-7: Miscellany
a. Variable: binary categorical; n. Variable: categorical; observational
observational units: pennies being units: automobiles
spun
o. Variable: binary categorical;
b. Variable: binary categorical; observational units: applicants for
observational units: people leaving graduate school
the washroom
p. Variable: quantitative; observational
c. Variable: quantitative; observational units: college students
units: fast-food sandwich
q. Variable: categorical; observational
d. Variable: quantitative; observational unit: person
units: residents of that country
r. Variable: binary categorical;
e. Variable: binary categorical; observational units: college students
observational units: American
s. Variable: binary categorical;
households
observational units: participants in
f. Variable: quantitative (though age sport
might make more sense to interpret);
t. Variable: quantitative; observational
observational units: colleges
units: sport participants
g. Variable: quantitative; observational
u. Variable: quantitative; observational
units: colleges
units: states
h. Variable: categorical; observational
v. Variable: quantitative; observational
units: American voters in 2004
units: bartenders (or glasses if just
i. Variable: binary categorical; one bartender)
observational unit: newborn babies
w. Variable: quantitative; observational
j. Variable: quantitative; observational unit: person
units: Alfred Hitchcock movies
x. Variable: quantitative; observational
k. Variable: quantitative; observational units: brides
units: American pennies
y. Variable: categorical; observational
l. Variable: quantitative; observational units: brides
units: automobiles
z. Variable: quantitative; observational
m. Variable: binary categorical; units: couples getting married
observational units: automobiles
 Activity 2-2 17

Solutions
••• In-Class Activities
Activity 2-1: Penny Thoughts

2
a. This is a binary categorical variable.
b. Answers will vary by class. One example is 15/18 or .833 voted to retain the
penny.
c. Answers will vary, but for the class in part b, 3/18 or .167 voted to abolish the
penny.
d. Answers will vary, but for the class in part b, the bar graph is shown here:

Opinion About Penny

1
.9
.8
.7
.6
Proportion

.5
.4
.3
.2
.1
0
Retain Abolish
Response

e. For the class in part c, more than 80% favors retaining the penny, but the vote is
not unanimous. A significant proportion of the class (.167) favors abolishing the
penny.

Activity 2-2: Hand Washing

a. The proportion of men who washed their hands is 2393/3206 or .746. The
proportion of women who washed their hands is 2802/3130 or .895.
b. Yes, these proportions are consistent with the bar graphs. The heights of the
“washed hands” bars are about .75 and .90.
c. It does appear that women are a little more likely to wash their hands after using
a public restroom than men are. About 90% of the women in this sample did,
whereas only 75% of the men did, indicating that women are about 1.2 times
more likely to wash their hands after using a public restroom.
d. Atlanta: .73 Chicago: .88 New York: .79 San Francisco: .88
e. There does not appear to be much difference between these cities in terms of
hand washing. Chicago and San Francisco appear to have identical proportions of
people who washed their hands after using a public restroom, whereas New York
seems to have about 10% fewer hand washers. Atlanta seems to have the smallest
percentage of hand washers among these four cities, with just under 75% of the
sample from Atlanta washing their hands.
18 Topic 2: Data and Distributions

Activity 2-3: Student Travels

a. This is a quantitative variable.
b. Answers will vary. Here is one example:

0 5 10 15 20 25 30 35 40 45 50
Number of States Visited

c. Answers will vary.

d. Answers will vary.
e. The number of states visited by students in this class varies from a minimum
of 6 to about 21. There are two high outliers of 30 and 31 states. The typical
number of states visited by a student in this class seems to be about 14.

Activity 2-4: Buckle Up!

a. The states are the observational units for these data.
b. The primary or secondary law variable is categorical and binary. The percentage
usage variable is quantitative.
c. Answers will vary. The typical usage percentage for a primary-type seatbelt
law state is about 86%; for states with a secondary-type law, the typical usage
percentage appears to be about 77%.
d. No, a state with a primary law does not always have a higher usage percentage
than a state with a secondary law; for example, Tennessee (p, 74.4%) and
Virginia (s, 8.4%).
e. Yes, states with a primary law tend to have higher usage percentages than states
with a secondary law. You see this in the dotplot because most of the dots for the
primary law states are clustered at the high percentage values (from 80–95%),
whereas most of the secondary law states have percentages that fall between
65%–85%.
f. Yes, the data seem to support the contention that tougher laws lead to more
seatbelt usage, but you cannot draw a definite cause-and-effect conclusion. There
might be hidden or “confounding” variables that explain the association between
primary seatbelt laws and increased seatbelt usage.

Activity 2-5: February Temperatures

a. San Luis Obispo tended to have the highest temperatures that month (its
temperatures are all clustered from 50–90°F, with a significant chunk above
75°F), and Lincoln tended to have the lowest temperatures, with many
temperatures below 45°F, whereas Sedona’s temperatures all ranged from about
48–68°F.
b. Sedona had the most day-to-day consistency in its high temperatures that month
(the temperatures stayed between 48–68°F all month), whereas Lincoln had the
least consistency because its temperatures ranged from about 10°F all the way to
about 75°F.
 Activity 2-6 19

Activity 2-6: Sporting Examples

a. The observational units are the students enrolled in one or the other of these sections.
b. The variables are section (categorical, binary), grade (categorical, not binary), and
total points earned (quantitative).

2
c. Students in the regular section tended to score more points than those in the
sports section. Scores in the regular section appear to be centered around 340
(85% of the possible points), whereas those in the sports section are centered
around 310–320 points (a bit less than 80% of the possible points). Scores in the
sports section are more spread out than those in the regular section. Students
in the sports section had the six lowest scores, all less than 260 points, but that
section also had the highest overall score, greater than 390 points.
d. Students in the regular section tended to score more points than those in
the sports section. Most students in the regular section scored between
300–380 points, with a center of approximately 340 points. In contrast, many
students in the sports section scored less than 300 points, and the center was
approximately 310–320 points.
e. No, some sports students scored more points than some regular students. The
statistical tendency means that a typical student in the regular section scored
more points than a typical student in the sports section.
f. The proportions are found by dividing the counts by 29 for the regular section
and by 28 for the sports section. These proportions are
Regular section: .552 good .379 fair .069 poor
Sports section: .250 good .536 fair .214 poor
g. The bar graphs follow:

Good 1
Fair .9
Proportion of Students

.8
Poor
.7
.6
.5
.4
.3
.2
.1
0
Regular Sports
Section

h. The bar graphs reveal similar results to the dotplots: Students in the regular
section tended to score higher than those in the sports section. More than half of
the regular students were in the good category, compared to only one-fourth of
the students in the sports section. At the other extreme, only 6.9% of the regular
students did poor work, compared to 21.4% of sports students.
i. You cannot draw a cause-and-effect conclusion between the type of section and
student performance. You will study these issues again in the next topic, but one
key is that students self-selected which section to take. Perhaps those who chose to
take the sports section had lower academic aptitude than those who selected the
20 Topic 2: Data and Distributions

regular section, or perhaps students were sleepier in the sports section because it
met earlier in the day.

••• Homework Activities

Activity 2-7: Student Data
• How many hours you slept in the past 24 hours: dotplot
• Whether you have slept for at least 7 hours in the past 24 hours: bar graph
• How many states you have visited: dotplot
• Handedness: bar graph
• Day of the week on which you were born: bar graph
• Gender: bar graph
• Average study time per week: dotplot
• Score on the first exam in this course: dotplot

Activity 2-8: Student Data

a. In general, do most female students study more than most male students? This
does not mean that you would expect to find that all female students study more
than all male students. You can also think in terms of the “typical female student”
scoring higher than the “typical male student.”
b. In general, do most students who study more score higher on exams than students
who don’t study as much? Again, this does not mean that you would expect to
find that all those who study earn higher grades than all those who do not study.

Activity 2-9: Value of Statistics

a. Answers will vary from student to student.
b. Answers will vary from class to class. Here are some sample answers:

Rating 1 2 3 4 5 6 7 8 9

Tally (Count) 0 0 1 0 5 6 11 6 6

c. Yes, 7 was chosen more often than any other value.

d. 29/35 or .829 gave a response greater than 5; 1/35 or .029 gave a response less than 5.
e. The vast majority of this class (more than 80%) feel that statistics is important to
society. In fact, more than 65% of the class feel that statistics is very important to
society. About 14% of the class are neutral about the importance of statistics and only
1 of the 35 students in this group believe that statistics is not very important to study.

Activity 2-10: Value of Statistics

i. Class C
ii. Class D
iii. Class E
iv. Class A
v. Class B
30 Topic 3: Drawing Conclusions from Studies

You might point out that asking for the explanatory and response variables puts this
question one additional step beyond what students have already begun to practice
extensively. Parts b and c then deal with both of the big issues in this topic: whether a
cause-and-effect conclusion can be drawn and to what population the results can be
generalized. You might summarize that these are the two questions students should
learn to ask about every statistical study. These are very different issues, due to two
problems (confounding and bias, respectively), and therefore require different remedies.
Those remedies are the subjects of the next two topics.

••• Homework Activities

For the homework activities, be selective about the number of activities you assign at one
time. Most require students to write quite a bit, and you don’t want to overwhelm them.
We recommend including a mix of questions about biased sampling and confounding
variables. Those that are primarily about sampling bias are Activities 3-6, 3-7, 3-8, 3-9,
3-10, 3-12, 3-13, 3-14, 3-15, and 3-16. Those that are principally about confounding
variables are Activities 3-11 and 3-17 through 3-28. One thing to watch out for is that
because almost all of these activities illustrate some type of bias or confounding, students
might become too suspicious and believe that no valid conclusions can ever be drawn.
Topics 4 and 5 should help to disavow them of this notion.
Possible assignment: Activity 3-7 requires some critical thought about representative
samples; Activity 3-8 gives practice identifying parameters and statistics; Activity 3-10
requires thinking about the correct use of statistical language; Activity 3-16 reviews the
concepts of observational unit, variable, parameter, statistic, bias, and sample size; and
Activity 3-24 asks about confounding variables.

Solutions
••• In-Class Activities
Activity 3-1: Elvis Presley and Alf Landon
a. Population: all adult Americans (record company interest)
Sample: those listening to the radio who called in
b. No, 56% is probably not an accurate reflection of the opinions of all adult
Americans on this issue. People who chose to call in (who took the time and
were willing to spend the money) probably felt differently and more strongly
about the issue than other adult Americans. The timing (on the anniversary of
Elvis’ death) could also have influenced the opinions of those who called. You
also have no indication of how widely distributed across the country the radio
stations were (perhaps there could be bias if the stations tended to be mostly
from the south).
c. Population: all Americans eligible to vote in 1936
Sample: the 2.4 million who returned the questionnaires
d. The Literary Digest’s prediction was in error because its sampling method was
biased. By sampling people who owned vehicles and telephones in 1936, Literary
Digest was sampling from a subset of the population who during that period
tended to be wealthy. Historically, the wealthy have tended to support the
Republican candidate (conservative), whereas those without money have tended
to vote Democrat (for social change). Thus, the pollsters contacted primarily
 Activity 3-2 31

Republican voters, but on election day, there was a heavy Democratic turn out.
Furthermore, those who chose to respond were probably more dissatisfied with
the incumbent (Roosevelt) than those who chose not to respond.
e. • The 56% of callers who believed that Elvis was alive: statistic
• The 57% of voters who indicated they would vote for Alf Landon: statistic
• The 63% of votes who actually voted for Franklin Roosevelt: parameter
f. • The proportion of students in your class who use instant-messaging or text-messaging
on a daily basis: statistic
• The proportion of students at your school who use instant-messaging or text-

3
messaging on a daily basis: parameter
• The average number of hours students at your school spent watching television last
week: parameter
• The average number of hours students in your class slept last night: statistic
g. • The proportion of voters who voted for President Bush in the 2004 election:
parameter
• The proportion of voters surveyed by CNN who voted for John Kerry in the 2004
election: statistic
• The proportion of voters among your school’s faculty members who voted for Ralph
Nader in the 2004 election: parameter (assuming the population is all of your
school’s voting faculty members)
• The average number of points scored in a Super Bowl game: parameter (assuming
the population is all Super Bowl games)
h. A categorical variable leads to a parameter or statistic that is a proportion; a
quantitative variable leads to a parameter or statistic that is an average.

Activity 3-2: Self-Injuries

a. Observational units: students
Variable: whether they had injured themselves Type: binary categorical
b. Population: all American college students
Sample: the 2875 students from Cornell and Princeton who responded to the
survey
c. The sample size is 2875.
d. The number 17% is a statistic; it is a proportion derived from the sample of
students.
e. This percentage is unlikely to be representative of all college students in the world
because the sample was taken from two U.S. colleges, and the college experience
in the United States is very different from the rest of the world. It is not even clear
that it would be representative of all U.S. colleges: Both schools in the survey
were Ivy League schools, so their students would hardly be “typical” U.S. college
students and might also have distinct types of stress and social reactions to stress.
32 Topic 3: Drawing Conclusions from Studies

Activity 3-3: Candy and Longevity

a. No, this is unlikely to be a representative sample of the health habits of all adult
Americans. Everyone in the sample was male and college-educated (to some
extent) at an Ivy League school at least 30 years before the study—not the picture
of the “average” American. In addition to the gender differences, the health
knowledge and access to medical care by the men in this sample could differ from
the rest of the population.
b. The proportion who consumed candy is 4529/7841 or .5776. This number is a
statistic.
c. The proportion of nonconsumers who had died is 247/3312 or .075. The
proportion of consumers who had died is 267/4529 or .059.
d. Observational units: 7841 men who entered Harvard between 1916 and 1950
Explanatory variable: whether they consumed candy Type: binary categorical
Response variable: whether they had died by the end of 1993 Type: binary categorical
e. Perhaps men who like candy also like to exercise regularly, and perhaps those who
do not tend to eat candy do not like to exercise regularly. In this case, it might
be the exercise that increases lifespan, rather than the candy. (Other possibilities
include differences in diet, differences in family size, and happiness levels.)
f. The proportion of nonconsumers who had never smoked is 1201/3312 or .363.
The proportion of consumers who had never smoked is 1852/4529 or .409.
g. A greater percentage of the candy consumers had never smoked. The higher death
rate among those who did not tend to consume candy might have been due to
smoking rather than not eating candy.

Activity 3-4: Sporting Examples

a. Observational units: statistics students
Explanatory variable: section (exclusively sports Type: binary categorical
examples or not)
Response variable: performance (points earned) Type: quantitative
b. You know this is an observational study because the students self-selected into the
two sections. The researcher (professor) merely passively observed the students’
selections and subsequent performances.
c. No, it is not legitimate to conclude that the sports examples caused the lower
academic performance. One obvious confounding variable would be the time of
the class. The section with exclusively sports examples was offered at a different
hour of the morning than the other section. Perhaps there was an honors class
being given at the same time as the earlier class, so a larger percentage of honors
students (regardless of their interests) had to sign up for the late class. Or maybe
the students in the sports-examples section meeting earlier in the day were not
as awake during classtime and that was responsible for their lower academic
performance.
 Activity 3-8 33

Activity 3-5: Childhood Obesity and Sleep

a. The explanatory variable is the amount of sleep that a child gets per night. This
is a quantitative variable, although it would be categorical if the sleep data were
reported only in intervals (more sleep vs. less sleep). The response variable is
whether the child is obese, which is a binary categorical variable.
b. This is an observational study because the researchers passively recorded
information about the children’s sleeping habits. They did not impose a certain
amount of sleep on children. Therefore, it is not appropriate to draw a cause-and-
effect conclusion that less sleep causes a higher rate of obesity. Children who get

3
less sleep might differ in some other way that could account for the increased
rate of obesity. For example, amount of exercise could be a confounding variable.
Perhaps children who exercise less have more trouble sleeping, in which case
exercise would be confounded with sleep. You have no way of knowing whether
the higher rate of obesity is due to less sleep or less exercise, or both, or due to
some other variable that is also related to both sleep and obesity.
c. The population from which these children were selected is apparently all children
aged 5–10 in primary schools in the city of Trois-Rivières. These Quebec children
might not be representative of all children in this age group worldwide, so you
should be cautious about generalizing that a relationship between sleep and
obesity exists for children around the world.

••• Homework Activities

Activity 3-6: Elvis Presley and Alf Landon
a. This is a very biased sampling method. You would expect this method to
overestimate the proportion of adults who believe that Elvis faked his death
because people who feel strongly about this are likely to be the ones responding to
such an Internet poll.
b. This number is a statistic.
c. Although this number may feel large, you really have no way of knowing, based
on the statistic alone, whether a sampling method is biased. It is better to consider
the sampling method when assessing whether you believe bias is present.
d. The sample size is 2032. Taking a larger sample would not reduce bias; if the
sampling method is flawed, increasing the sample size will not correct the problem.

Activity 3-7: Student Data

a. Answers will vary by school and class.
b. Answers will vary by school and class.
c. Answers will vary by school and class.

Activity 3-8: Generation M

a. Parameter e. Parameter
b. Statistic f. Statistic
c. Statistic g. Parameter
d. Statistic h. Statistic
46 Topic 4: Random Sampling

Solutions
••• In-Class Activities
Activity 4-1: Sampling Words
a. Answers will vary. The answers given here are one example.
b. Here are completed tables:

Word score forth whether have might

Number of Letters 5 5 7 4 5

Word did here full resolve perish

Number of Letters 3 4 4 7 6

c. Dotplot:

3 4 5 6 7
Number of Letters per Word

Observational units: words

Variable: number of letters per word
Type: quantitative
d. The average is five letters per word. This number is a statistic.
e. An example set of responses from one class follows.

4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5

4.29
Average Word Length in Sample

f. Observational units: samples of 10 words

Variable: average number of letters per word (quantitative)
g. In this example, 8/10 or .8 of the students produced a sample average greater than
4.29 letters per word.
h. Yes, this sampling method appears to be biased. It appears to overestimate the
population mean. This is evident from the dotplot because it is centered at about
5.7 (rather than 4.29), and it indicates that a large proportion of the class selected
samples with averages greater than 4.29.
i. Your eyes are most likely drawn to the longer words, and you tend to overlook the
short, common words such as a, and, is, and or. Thus, when you try to choose
representative samples, you do not select enough short words in your sample.
 Activity 4-3 47

j. If you use this method, you would also be likely to select too many long words in
your sample because the long words take up more space on the page and therefore
have a greater chance of being selected when you blindly point to a location.
k. No, increasing the sample size will not make up for the biased sampling method.
You would still tend to overrepresent the long words.
l. You need to employ a truly random method to select the words. You could
write each word on the same size slip of paper, put each slip in a hat, mix them
thoroughly, and then draw ten slips from the hat.

Activity 4-2: Sampling Words

a. Many answers are possible. The following was obtained from the Random Digits
Table starting at the beginning of line 60:

4
1 2 3 4 5

Random Digits 031 025 052 076 059

Word now That can A a

Word Length 3 4 3 1 1

b. The average word length is 2.4 letters per word.

c. Answers will vary. The following is an example from one class.

0.0 1.2 2.4 3.6 4.8 6.0 7.2 8.4

4.29
Sample Mean for Average Word Length

d. The distribution is much closer to being centered at 4.29 and has a smaller horizontal
spread than the previous one did (though the latter is not always the case).
e. The sample averages are roughly split evenly on both sides of 4.29.
f. Yes, random sampling appears to have produced unbiased estimates of the average
word length in the population.

Activity 4-3: Sampling Words

a. Answers will vary. The following are from one particular running of the applet.

1 2 3 4 5

Word The These here for Should

Number of Letters 3 5 4 3 6

Average number of letters: 4.2

b. You will probably not obtain the same sample of words or the same average length
the second time.
c. The average of the 500 sample averages is 4.31 letters per word.
48 Topic 4: Random Sampling

d. Yes, this appears to be “around” 4.29.

e. Answers will vary according to student expectation.
f. The center of this distribution should also be near 4.29 but the horizontal spread
is much smaller.
g. The distribution of the samples of size 20 has less variability (more consistency) in
the values of the sample average word length.
h. The result of a single sample is more likely to be close to 4.29 with a sample of
size 20 than with a sample of size 5.
i. No, increasing the sample size when using a biased sampling method will not
reduce the bias. The results from different samples will tend to be closer together
but will still be centered in the wrong location (not around the parameter value of
interest). If you want to reduce the bias, you must change the sampling method.

Activity 4-4: Sampling Words

a. One example set of results follows.

b. Both distributions are roughly bell-shaped, centered at about 4.29 words, with
a horizontal spread from about 2 to 7 words.
c. Yes, these distributions seem to have similar variability.
d. No, not much changed when you sampled from the larger population.

Activity 4-5: Back to Sleep

a. The population of interest is all infants younger than eight months in the United
States in those years. The sampling frame is the list of households with such
infants, generated from birth records, infant photography companies, and infant
formula companies. The sample consists of the infants in the 1002 households
whose mothers (or other caregivers) participated in the interview.
b. The sample size is 1002. (Actually, a total of 1015 infants were in the sample
because some households had twins.)
c. The researchers did not technically obtain a simple random sample of infants. One
reason is that the sampling frame did not include the entire population. Another
reason is that more than half of the numbers called did not lead to an interview.
Infants who were not included in the sampling frame or whose mothers declined to
 Activity 4-8 49

participate might differ systematically in some ways from those who were included.
Nevertheless, the researchers did use randomness to select their sample, and they
probably obtained as representative a sample as reasonably possible.
d. Perhaps mothers in those groups were in a lower economic class and therefore
less likely to have phones in the first place, or perhaps they had to work so their
children were in daycare.
e. These comparisons address the issue of bias, not precision. The sampling method
was slightly biased with regard to the mother’s race and age and the infant’s birth
weight.
f. These percentages are statistics because they are based on the sample.
g. The large sample size produces high precision. This means that the sample
statistics are likely to be close to their population counterparts. For example, the

4
population proportion of infants who sleep on their backs should be close to the
sample proportion who sleep on their backs.
h. The sample size for subgroups is smaller than for the whole group, so the sample
results would be less precise.

••• Homework Activities

Activity 4-6: Rating Chain Restaurants
a. It seems unlikely that this sample was randomly chosen as it would be extremely
difficult to give each Consumer Reports reader an equally likely chance of being
selected for the sample and to ensure that everyone selected responded. It is much
more likely that the responders self-selected by returning a survey.
b. The authors probably make the disclaimer because the sample was not randomly
selected from the entire population, but only from their readers who may have
different habits and attitudes from nonreaders and therefore cannot reasonably be
extended to the general population.
c. Answers will vary, but you probably should generalize these results only to
Consumer Reports readers who tend to visit full-service restaurant chains and like
to complete surveys.

Activity 4-7: Sampling Words

a. Categorical (binary)
b. 99/268  .369
c. The answer to part b is a parameter; .369 is the proportion of all 268 words (the
population) in the Gettysburg address that are over 5 letters long.
d. No; because of sampling variability you would not expect the sample proportion
to equal .369, but you would expect it to be reasonably close most of the time. (In
fact, with a sample of size 5, the sample proportion could not equal .369; it could
only be 0, .2, .4, .6, .8, or 1.)

Activity 4-8: Sampling Words

Answers will vary. These are based on one particular running of the applet.
a. Yes, this distribution should be centered at about .369 (it is .38 in this case).
 Activity 5-1 61

To randomize the group assignments with Fathom:

1. Create a collection with an attribute containing the names of all students in the
class (as described in the Fathom technology alternative for Topic 4).
2. Add a second attribute named Group to this collection that contains JFK for half
the students and JFKC for the other half. Don’t worry yet about doing this at
random.
3. Select the collection box and choose Collection | Scramble Attribute Values
from the Fathom menus.
4. Select the new collection box that appears and drag a table from the shelf to
view its contents. You should find that the first attribute of student names has
been scrambled, so the sequences can now be assigned at random. Fathom note:
If you want to scramble a different column, double-click on the new scrambled
collection box and choose an attribute from the Scramble panel.
Note that the new scrambled collection also provides a convenient place to enter the
data from the Preliminaries question for use in Activity 5-5.

Solutions

5
••• In-Class Activities
Activity 5-1: Testing Strength Shoes
a. No, this is anecdotal evidence based on just two of your friends. The friend who
wears the strength shoe may be much more athletic in general than your other
friend, or may be taller, etc., which would explain why he or she is able to jump
farther.
b. Explanatory: whether the individual Type: binary categorical
wears strength shoes
Response: length of jump Type: quantitative
c. No, you cannot legitimately conclude that strength shoes cause longer jumps
because the subjects self-selected which type of shoe they would wear, so your
results are from an observational study. There could be something else different
about people who choose to wear the strength shoes. For example, males might be
more likely to choose the strength shoe than females.
d. Randomly assign six of the subjects to each group.
e. You could flip a coin for each subject. If it lands heads up, that subject will wear
ordinary shoes; otherwise, the subject will wear strength shoes. Continue to flip
the coin until you have six subjects in the ordinary (or strength) shoe group. If
you have not filled both groups, the remaining subjects should all be placed in
the unfilled group. Technically, this is not random assignment, even though
this approach is used quite frequently. A more correct approach would be to
number each subject and put the numbers in a bag, and then the first six numbers
drawn out of the hat are assigned to the “strength shoe” group and the others are
assigned to the “ordinary shoe” group.
62 Topic 5: Designing Experiments

Activity 5-2: Testing Strength Shoes

a. Answers will vary. Below is one example.

Strength Shoe Group Ordinary Shoe Group

Name Gender Height Name Gender Height

Brad Male 70 Audrey Female 67

Mary Female 66 Michael Male 71

Peter Male 69 Russ Male 68

Kyle Male 71 Patrick Male 70

Barbie Female 63 Anna Female 61

Matt Male 73 Shawn Male 67

b. Strength shoe group: .667 Ordinary shoe group: .667

Difference (strength  ordinary): 0
c. Strength shoe group: 68.67 inches Ordinary shoe group: 67.33 inches
Difference (strength  ordinary): 1.33
d. No, the two groups are not identical with regard to both of these variables, but
they are similar.
e. Answers will vary from class to class. The dotplot of differences in proportions
should be roughly symmetric and centered around zero. The larger the class,
the more symmetric the plot should be. The horizontal axis should be labeled
“difference in sample proportions” with a scale from 1 to 1, and the vertical axis
should display the count/tally of each difference.
The dotplot of the differences in average heights should also be roughly
symmetric and centered at zero. The horizontal axis should be labeled “difference
in sample heights” with a scale from approximately 6 to 6, and the vertical axis
should display the count/tally of each difference. For both plots, the observational
units are the random assignments.
f. Both plots should appear to be roughly centered around zero. This indicates that
random assignment is effective because it is “balancing out” the proportion of
men/women and the heights in both groups. In the long run, both groups are
roughly the same with regard to these variables because, on average, the difference
between them is zero. In particular, you have no prior suspicion that one group
will have certain characteristics that differ from the other group.

Activity 5-3: Testing Strength Shoes

a. Answers will vary. The answers here are from one particular running of the
applet.
Group A: .5 Group B: .8333
Difference: .3333
 Activity 5-5 63

b. No, you probably will not get the exact same assignment of subjects to the groups
nor the same difference in proportions.
c. Answers will vary as this is a prediction.
d. The distribution should be centered at zero. It may or may not be what the
students predicted.
e. Random assignment does not always balance out the gender variable exactly, but it
does tend to balance out gender between the two groups. You can see this because
the dotplot is centered at zero, and zero is the difference that occurred most
frequently. Occasionally, there were differences as large or small as .8333.
f. This distribution is roughly symmetric, centered near zero, ranging from roughly
5.5 to 5.5. This indicates that the randomization also tends to balance out the
heights between the two groups.
g. Answers will vary; the question asks for student expectation.
h. Both dotplots (for gene and x-variable) are roughly symmetric and centered at
zero. This indicates that random assignment tends to balance out variables even
when they are unseen or unrecorded.

5
i. By randomly assigning the subjects to the two groups, you have (hopefully) balanced
out all other potentially confounding variables, making the only difference between
the two groups the type of shoe. Therefore, if you then find that the strength shoe
group jumps substantially farther, on average, than the ordinary shoe group, you
would be able to conclude that the increase was due to the strength shoe because
there should not be any other explanation for a difference between the two groups.

Activity 5-4: Botox for Back Pain

a. No, you should not conclude that Botox is an effective treatment for back pain.
There was no control group of patients who did not receive the treatment with which
to compare. Perhaps the patients’ pain decreased simply because time had passed.
b. This is a better design because it allows you to compare the response of a treatment
group with that of a control group and because you are randomly assigning the
subjects to the two groups so you can control for confounding variables. The only
difference between the patients in the two groups should be the Botox treatment.
c. Perhaps the patients who are given the Botox treatment will think they feel better
simply because they are given any treatment. Back pain is very subjective and hard
to measure, so any sort of treatment may cause some of the subjects to believe
their pain has decreased.
d. The researchers could inject the control group with saline or some other harmless
substance so that patients in both groups believe they are receiving a treatment,
equalizing any psychological effects between the two groups.

Activity 5-5: Memorizing Letters

a. This is an experiment because the teacher actively imposed the treatment
(grouping of letters) on each subject/student.
b. Explanatory: which sequence of letters you were given Type: binary categorical
Response: number of letters correctly memorized Type: quantitative
64 Topic 5: Designing Experiments

c. The instructor randomly decided which grouping of letters each student would
receive. This was important because it prevented self-selection and controlled for
confounding variables. You should not expect any differences between the groups
prior to the treatment.
d. The students were blind to the fact that there were two different groupings of
letters given out initially. They were unaware that you were trying to compare
the effect of these familiar and unfamiliar groupings, so they could not
unintentionally influence the results.
e. Answers will vary. Here is one representative set of answers.

JFK
4 8 12 16 20 24 28
Group

JFKC
4 8 12 16 20 24 28
Number of Letters Correct

f. Yes, these data appear to support the conjecture that those who receive the letters
in convenient three-letter chunks tend to memorize more letters. The center of
this plot is about six letters higher than for the JFCK plot.
g. Yes, because this was a well-designed, randomized controlled experiment you
could legitimately conclude that the grouping of letters into familiar chunks
caused the higher scores. Because you randomly assigned the students to each type
of grouping, there should have been roughly an equal number of good memorizers
in both the JFK and JFKC groups, so the randomization controlled for this
potentially confounding variable.

Activity 5-6: Nicotine Lozenge

a. This is an experiment because the researchers imposed the nicotine lozenge (or
placebo) on subjects.
b. The experimental units are the smokers interested in quitting who volunteered to
participate in the study.
c. The explanatory variable is whether the smoker was given a nicotine lozenge or a
placebo. This variable is categorical and binary. The response variable is whether
the smoker successfully quit smoking by the end of the study. This variable is also
categorical and binary.
d. This information validates that the random assignment achieved its goal of
balancing out all of these variables, which could potentially be related to a
smoker’s ability to quit, between the nicotine lozenge and placebo groups. Thus,
if the nicotine lozenge group has a higher proportion who quit smoking, then
the researchers can attribute that to the lozenge, not to any of these background
variables, because they were similar between the groups.
e. Yes. Because this was a randomized experiment, it is legitimate to conclude
that the nicotine lozenge caused the increase in the proportion of subjects who
successfully quit smoking.
 Activity 6-1 87

Solutions
••• In-Class Activities
Activity 6-1: Government Spending
a. Too little: 398/646  .616 About right: 198/646  .307
Too much: 50/646  .077
b. Here is the bar graph for the marginal distribution of the variable opinion about
federal spending on the environment:

Environment Spending
1
.9
.8
.7
Proportion

.6
.5
.4
.3
.2
.1
0
Too Little About Right Too Much

6
Respondents’ Opinions

c. Assuming the sample is representative, you can say that Americans tend to believe
that the government is spending too little on the environment. More than 60%
of the sample felt this way, whereas less than 8% felt that the government is
spending too much. About 30% of this sample felt that government spending on
the environment is right where it should be.
d. The proportion of liberal respondents who say the federal government spends too
little on the environment is 127/155 or .819.
e. The proportion of liberal respondents who say the federal government spends the
right amount on the environment is 27/155 or .174.
f. The proportion of liberal respondents who say the federal government spends too
much on the environment is 1/155 or .006.
g. Yes, these three proportions add up to 1.00 if you don’t use the rounded
proportions.
h. Here is the conditional distribution for the “liberal” column of the table:

Liberal Moderate Conservative

Too Little .819 .619 .479

About Right .174 .314 .385

Too Much .006 .067 1.36

Total 1.000 1.000 1.000

88 Topic 6: Two-Way Tables

i. The completed segmented bar graph follows:

Environment Spending
100
Too much
90
About right
80
Too little
70

Percentage
60
50
40
30
20
10
0
Liberal Moderate Conservative
Political Viewpoint

j. Yes, the distribution of spending opinion seems to differ among the three political
groups. The more liberal the group, the more likely they are to believe that the
government is not spending enough on the environment. Whereas 82% of the
liberals in the sample feel the government is spending too little money, 62% of the
moderates and only 48% of the conservatives believe this; on the other hand, 14%
of conservatives believe the government is spending too much on the environment
compared to less than 1% (.6%) of the liberals and 7% of the moderates.
l. Approximately 35% of the liberals in the sample believe the government spends
too much on the space program.
m. There is a difference—but not a big difference—in how the three political groups
feel about government spending on the space program. Roughly 40% of each
group stated the government is spending too much, roughly 10% stated the
government is spending too little, and roughly 50% stated the government is
spending about the right amount.
n. Government spending is closest to being independent of political viewpoint with
the issue of the space program. You can tell because the breakdown of the three
segments in the segmented bar graph is nearly identical across the three political
parties, whereas they are quite distinct across the three political parties in the
environmental spending graph.
o. Answers will vary. Here is an example graph for an issue where opinion about
government spending is perfectly independent of political viewpoint:

Independent Government Spending Issue

100
Too much
90
About right
80
Too little
70
Percentage

60
50
40
30
20
10
0
Liberal Moderate Conservative
Political Viewpoint
 Activity 6-3 89

The goal is for the three segments to have the same breakdown across the
three groups (though not necessarily being an equal size within each political
party).
p. The proportion who identify themselves as liberal is 127/398 or .319.
q. No, the proportion in part p is less than half the value found in part d (.819).

Activity 6-2: AZT and HIV

a. This is an experiment because the researchers randomly assigned the subjects to
the AZT and control (placebo) groups.
b. Explanatory variable: AZT or placebo
Response variable: whether the baby was born HIV-infected
c. Here is the 2  2 table:

Placebo AZT Total

HIV-infected 40 13 53

Not HIV-infected 143 167 310

Total 183 180 363

6
d. For the placebo group, 40/183 or .219. For the AZT group, 13/180 or .072. Yes,
these calculations are consistent with the segmented bar graph.
e. The difference is .219 – .072 or .147 (or .146 if using more than three decimal
places for the proportions). This difference does not appear to be terribly
large.
f. The ratio is .219/.072 or 3.04 (or 3.03 if using more than three decimal places for
the proportions). The risk of a baby being born HIV-infected is more than three
times greater for those whose mothers were in the placebo group.
g. Yes, you can legitimately conclude that AZT is the cause of the three-fold
reduction in HIV-infection rate compared to the placebo group because this
was a well-designed, randomized experiment. These results can be cautiously
generalized to HIV-positive pregnant women (you don’t have much information
about how the women in the sample were selected).

Activity 6-3: Lifetime Achievements

a. Gender: binary categorical
Preferred lifetime achievement: categorical (nonbinary)
b. Explanatory variable: gender
Response variable: preferred lifetime achievement
90 Topic 6: Two-Way Tables

c. Answers will vary by class. Here is one representative set of answers from a class of
78 students:

Olympic Medal Nobel Prize Academy Award Total

.346 .462 .192 1

Classmate Preferences for Lifetime Achievement

1
.9
.8
.7
Proportion

.6
.5
.4
.3
.2
.1
0
Olympic Medal Nobel Prize Academy Award
Preferred Achievement

More than 45% of this class would prefer to win a Nobel prize, whereas less than
20% aspire to win an Academy Award. Just over one-third of the class would like
their greatest lifetime achievement to be winning an Olympic medal.
d. Using the example data from part c gives these counts:

Male Female

Olympic Medal 11 16

Nobel Prize 12 24

Academy Award 1 14

e. With this information you would only be able to fill in the totals; you need
to know the information on both gender and preferred lifetime achievement
simultaneously in order to fill in each cell of the table.

Activity 6-4: Hypothetical Hospital Recovery Rates

a. A: 800/1000  .8 B: 900/1000  .9
Hospital B saved the higher percentage of its patients.
b. A: (590  210) /(600  400)  .8 B: (870  30) / (900  100)  .9
c. A: 590/600  .980 B: 879/900  .967
Hospital A saved the greater percentage of its patients who had been in fair
condition.
d. A: 210/400  .525 B: 30/100  .300
Hospital A saved the greater percentage of its patients who had been in poor
condition.
 Activity 6-5 91

e. Hospital A tends to treat a higher proportion of patients who are in poor

condition to begin with and whose chances of survival are not very high,
regardless of where they seek treatment. Hospital A does a better job with
both types of patients, but their overall survival percentage is lower because
they treat such a higher rate of patients in poor condition than Hospital B
does.
f. If you were ill, you should go to Hospital A, which does a better job of treating
both types of patients. No matter how sick you are, you stand a better chance of
surviving at Hospital A.

Activity 6-5: Back to Sleep

a. These are statistics because they pertain to the sample of parents interviewed
in the study. They are not calculated based on the population of all American
parents with infants less than eight months old.
b. These are conditional distributions. Each year has its own percentages sleeping in
each of the positions, which sum to 100%.
c. The explanatory variable is year; the response variable is infant’s sleep position.
d. The segmented bar graph is shown here:

Infant Sleeping Positions

6
No regular position 100
Back 90
Side 80
70
Stomach
Percentage

60
50
40
30
20
10
0
1992 1993 1994 1995 1996
Year

e. The segmented bar graph reveals that the distribution of infant sleep positions
changed considerably over this five-year period. The percentage of parents who
placed infants on their stomachs declined dramatically, from more than two-
thirds (70%) to less than one-fourth (24%) in these five years. The proportions
of infants placed on their backs and sides both increased during this period.
Because this is an observational study and not an experiment, you cannot say
that the recommendation or promotional campaign caused these changes, but
(assuming the sample is representative) it is nonetheless heartening to find
that parents were generally changing their habits and placing infants in safer
positions to sleep.
118 Topic 7: Displaying and Describing Distributions

4. Set the viewing window for the histogram. To do this, press the p key and
enter numbers so that the plot fits the screen properly, making sure you use the Ã
key to express the negative sign, not the subtraction key.
5. Press the s key, followed by the r key, and then the right and left
arrows to obtain information about your histogram. To reduce or increase the
number of subintervals, change the Xscl setting in the Window screen.

Solutions
••• In-Class Activities
Activity 7-1: Matching Game
Consider the following seven variables:
a. These are all quantitative variables (except perhaps “jersey number;” even though
it is numerical, it does not make sense to determine the “average jersey number”).
b. 1. Variable: point values of letters in the board game Scrabble
Explanation: The values are regularly spaced (point values are whole numbers)
with most letters being worth few points.
2. Variable: prices of properties on the Monopoly game board
Explanation: The values are very regularly spaced with two properties having
exactly the same prices.
3. Variable: annual snowfall amounts for a sample of cities around the United States
Explanation: Most cities have little or no snow, but some have quite a bit.
4. Variable: jersey numbers of Cal Poly football players in 2006
Explanation: There are virtually no repeats and the distribution covers nearly
all the values from 1–99.
5. Variable: blood pressure measurements for a sample of healthy adults
Explanation: The values are fairly symmetric, with both a high and low outlier.
6. Variable: weights of rowers on the 2004 U.S. men’s Olympic team
Explanation: There is one low outlier (coxswain) and two weight classes.
7. Variable: quiz percentages for a class of statistics students (quizzes were quite
straightforward for most students)
Explanation: Most of the values are high (near 80–90) with a few very low
outliers.
c. Dotplots 1 and 3 have a similar shape (they are skewed to the right), whereas
dotplots 6 and 7 are both skewed to the left with outliers.

Activity 7-2: Rowers’ Weights

a. The weights of the rowers on the men’s 2004 Olympic team are skewed to the
left, ranging from about 120 to 230 lbs. There are two clusters of weights, one
tight cluster around 160 lbs and one that ranges from about 180 to 230 lbs. There
is one low outlier who weighs only 120 lbs.
 Activity 7-4 119

b. Cipollone is the apparent outlier at 120 lbs. His job (the coxswain) is to call out
the cadence to the rowers—he does not row himself. So it is important that he not
add excess weight to the boat.
c. There is a cluster of rowers whose weights seem to be no more than 160 lbs. These
rowers are all involved in “lightweight” events, which require them to weigh
below a certain amount on race day. The upper cluster are not involved in the
lightweight events and have no upper limit on their weights.

Activity 7-3: British Monarchs’ Reigns

a. The current monarch is Queen Elizabeth II. She is not represented because her
reign had not ended (as of December 2006).
b. The longest reign was that of Queen Victoria. Her reign was 63 years.
c. The shortest reign was less than 12 months (0). Edward V spent the least time on
the throne. He ascended the throne at the age of 12 in 1483 upon the death of his
father Edward IV, but was deposed two months later by his uncle Richard, Duke
of Gloucester.
d. 6|3 represents 63 years. Victoria reigned for 63 years.
e. Four monarchs reigned for 13 years. You can tell this from the stemplot because
there are four 3s on the 10s-stem.
f. This distribution is skewed to the right.
g. Half of the monarchs reigned for more than 19.5 years and half of them reigned
for less than 19.5 years.

7
h. One-quarter of the monarchs reigned for fewer than 9.5 years. One-quarter of the
monarchs reigned for more than 34 years.

Activity 7-4: Population Growth

a. The completed stemplot is shown here.

Western States Eastern States

9998888550 0 033344556889999
43322 1 00014567
982100 2 136
0 3
0 4
5
6 6
leaf unit  one percentage point

b. Western states tend to have higher percentage growths than eastern states. The
distributions of percentage growth for both the eastern and western states are
both skewed to the right. The percentages in the 26 eastern states display less
variability than the western states, ranging from near 0% to about 26%, with
a center of about 10%. The percentages in the 24 western states range from a
minimum near 0% to a maximum 30%, with a high outlier at 40% and a center
of about 13%. There is also one western state with an extremely high population
growth rate (Nevada)—more than 66%.
120 Topic 7: Displaying and Describing Distributions

Activity 7-5: Diabetes Diagnoses

a. Observational units: 548 subjects with diabetes
Variable: age when diagnosed
b. No, it is not practical to construct a dotplot or stemplot to display this
distribution because there are so many data values.
c. The proportion of people who were diagnosed with diabetes before the age of 18
is (4  23  14  7)/548  48/548  .088.
d. The proportion of people who were diagnosed with diabetes after the age of 62 is
(44  40  33  16  5  1)/548  139/548  .254.
e. These ages range from about 1 year to 88 years and are centered at about 50 years.
Most of the ages are clustered between 30 and 80 years, but there is another small
cluster indicating juvenile onset diabetes—between 1 to 15 years of age.
f. Of these three choices, using 10 intervals seems to provide the most informative
histogram for these data. This provides enough information for you to see both
clusters without adding too much clutter to the graph.

300
Number of People

250
200
150
100
50
0
0 20 40 60 80
Age of Diabetes Diagnosis

140
120
Number of People

100
80
60
40
20
0
0 20 40 60 80
Age of Diabetes Diagnosis

50

40
Number of People

30

20

10

0
0 12 24 36 48 60 72 84
Age of Diabetes Diagnosis
 Activity 7-7 121

g. The first graph appears to have a uniform shape; the second is slightly skewed to
the right. But neither is a legitimate histogram of the Olympic rowers’ weights
because the variable (weight) is not on the horizontal axis—it is on the vertical
axis! For example, it would be very misleading to say the “center” of the weights
was around Dan Berry’s weight—he actually has one of the greatest weight values.
The rowers’ names are of less interest than the pattern in the weight values to the
behavior of the weights.

Activity 7-6: Go Take a Hike!

a. The stemplot is shown here.

0 68
1 00000555555588
2 00000000122555556668 leaf unit  .1 mile
3 000002224458
4 00000566
5 00055668
6 0000
7 004
8
9 5

b. The distribution of hike distances is sharply skewed to the right, indicating there
are many hikes on the short side and only a few longer hikes. A typical hike is
between 2 and 3 miles. Most hikes are between 1 and 6 miles, but two hikes are
less than a mile and a few are more than 6 miles. The longest hike is 9.5 miles,
which is a bit unusual and could be considered an outlier because this hike is

7
more than two miles longer than the next longest hike (7.4 miles). Many hikes
have a reported distance that is a multiple of a whole number or a half number
of miles.

••• Homework Activities

Activity 7-7: Newspaper Data
a. You would expect the shape of this distribution to be skewed to the left.
Most of the deceased people would be rather old, but some of them would be
unexpectedly young.

18
16
14
12
Frequency

10
8
6
4
2
0
0 20 40 60 80 100
Age at Death (in years)
 Activity 8-1 139

SetUpEditor. Press Ö and then press Õ and ¿. You can now edit any of
the entries using the arrow keys to highlight a value. To delete an entry, press the
{ key.
4. Now use 1-VarStats to recompute the mean and median.
The SetUpEditor feature can set up more than one list within the editor (up to 20 in
the order you specify). On your Home screen, after SetUpEditor, enter one or more
lists separated by commas. One nice feature is that if a list is archived, SetUpEditor will
automatically unarchive the list and place it in the editor.
Other useful features in the STAT menu include EDIT, SORTA(, SORTD(, and
CLRLIST. The SORTA( and SORTD( features are definitely worth exploring with your
students. These two features will sort a given list either in ascending or descending
order, respectively. The most useful capability of these two features is when you have
two or more dependent lists, for example, a data list and its frequency list. If you sort
the data list in either ascending or descending order, you can couple the frequency
list with the data list so you do not break the dependency. So, for example, you would
enter SORTA(datalist,frequencylist) on your Home screen if you wanted to sort your
data list in ascending order. This dependency coupling can be done with any number
of lists.

Solutions
••• In-Class Activities
Activity 8-1: Sleeping Times
a. Observational units: students in a statistics course
Explanatory variable: section of the course Type: categorical
Response variable: sleeping time in hours Type: quantitative
b. The centers are not all similar: The center for Section 1 is noticeably less than the

8
center for Section 2, which is less than the center for Section 3.
c. The center for Section 1 is about 6 hours. The peak of the distribution is about
6 hours. Approximately half of the times are less than this, and approximately
half are more.
d. The mean is 106.25/17 or 6.25 hours.
e. The median is 6.25 hours. You calculate (171)/2  9. The ninth observation,
counting from either end, is 6.25 hours.
f. Section 2 mean: 7.000 Section 3 mean: 7.523
Explanation: The mean for Section 2 should be less because there are many
observations at 8 hours in Section 3, but not in Section 2, and the classes behave
similarly at 6 hours and less.
g. Section 2 median: 7.0 Section 3 median: 7.5
h. No, based on this example, the mean and median of a dataset do not always equal
each other.
i. The mode of the section variable is Section 3. More students were enrolled in this
section than in any other section. Perhaps that is because it was offered much later
in the morning!
140 Topic 8: Measures of Center

Activity 8-2: Game Show Prizes

a. The median prize amount is (750  1000)/2 or $875.
b. Answers will vary (student prediction).
c. Mean: $131,478 Median: $875
d. Answers will vary regarding the predictions. The mean and median are not close
to each other. This makes sense because the data is strongly skewed to the right;
there are several extremely large values that “pull” the mean to the right and make
it much greater than the median.
e. Six prize amounts are greater than the mean. This proportion is 6/26 or .231.
f. Thirteen prize amounts are greater than the median. This proportion is 13/26
or .5.
g. If the producers want to give the impression that contestants win huge amounts,
they should advertise the mean amount won because it is so much greater than
the median.

Activity 8-3: Matching Game

a. Symmetric: prices of properties on the Monopoly game board
Skewed to the left: quiz percentages for a class of statistics students
Skewed to the right: annual snowfall amounts for a sampling of U.S. cities
b. Here is the completed table:

Monopoly Prices Snowfall Amounts Quiz Percentages

Mean 208.6 21.65 88.9

Median 210 19.9 95

c. When the distribution is skewed to the left, the mean is less than the median.
When the distribution is skewed to the right, the mean is greater than the
median. In symmetric distributions, the mean and median are similar.

Activity 8-4: Rowers’ Weights

a. Answers will vary (estimates).
b. The distribution is skewed to the left with an extreme low outlier (the coxswain),
so it makes sense that the mean will be less than the median.

Without With Max Weight With Max Weight

Whole Team Coxswain at 329 at 2229

Mean 188.93 191.58 195.42 268.5

Median 195 195 195 195

 Activity 8-6 141

c. Predictions. Answers will vary.

d. See table following part b.
e. Predictions. Answers will vary.
f. See table following part b. (Cipollone’s weight was not added back in.)
g. Predictions. Answers will vary.
h. See table following part b.
i. The median is resistant. This conclusion makes sense because to find its value
you use only the middle of the data, not the ends where outliers would occur. The
mean is not resistant, which also makes sense: To find the mean you use all the
data. Outliers will pull the mean toward them—potentially affecting the value of
the mean drastically.
j. The mean only has to stay between the minimum and maximum values in the
dataset. (And so, if one of those values changes without bound, so does the
mean.)

Activity 8-5: Buckle Up!

a. The states with primary seatbelt laws tend to have higher compliance percentages.
b. The primary mean is 86.4%; the primary median is 86%.
The secondary mean is 77.14%; the secondary median is 77.4%.
The compliance percentage in states with primary seatbelt laws is about 9 points
higher, on average, than in states with secondary seatbelt laws. This is a fairly
striking difference.
c. No, you cannot draw a causal conclusion between stricter laws and the higher
percentage of people who wear seatbelts because this is an observational study and
not a well-designed, comparative experiment. There could be many confounding

8
variables that explain the strong association here.

Activity 8-6: Wrongful Conclusions

a. The conclusion drawn is not valid because the mean does not divide the dataset
in half. It could be the case that most of the house prices are less than the mean,
but one house price is extremely high and is the only one greater than the mean.
b. The conclusion drawn is not valid because the median indicates the middle data
value in an ordered set. Even though the median business trip cost $600, the most
expensive trip may have cost anything, say $5000, which would definitely bring
the total cost over $3000.
c. The conclusion drawn is not valid because the mean does not indicate what
percentage of the data values fall above or below it. It is possible that the CEO,
presidents, and vice-presidents make up 10% of the company and earn enough
between them to bring the mean above the salaries of the other 90% of the
employees.
d. The conclusion drawn is not valid because the mode indicates the most frequent
choice; this choice does not have to be the preference of more than half of the
142 Topic 8: Measures of Center

customers when there are more than two choices. It could be the case that
100 customers chose chocolate, 90 chose vanilla, and 80 chose strawberry.
The mode is chocolate, but 170/270  .63, so 63% prefer a flavor other than
chocolate.

Activity 8-7: Readability of Cancer Pamphlets

a. To calculate the mean, you need to know all of the actual data values, and you do
not know the values for those patients with a reading level below grade 3 or above
grade 12.
b. There are 63 patients, so the median is the ordered value in position (63  1)/2,
or the 32nd value. If you start counting from the low end, you find 6 patients
read below grade 3, 10 patients at grade 3 or below, 14 patients at grade 4 or
below, 17 patients at grade 5 or below, 20 patients at grade 6 or below, 22 patients
at grade 7 or below, 28 patients at grade 8 or below, and 33 patients at grade 9 or
below. The 32nd value is therefore at grade level 9, which is the median patient
reading level.
c. There are 30 pamphlets, so the median readability level is the average of
the 15th and 16th pamphlets. Counting in a similar way, the 15th and 16th
readability values are both located at grade level 9, so a pamphlet’s median
readability level is grade 9.
d. These medians are identical.
e. No. The centers of the distributions (as measured by the medians) are well-
matched, but you need to look at both distributions in their entirety and consider
all values. The problem is that many patients read at a level below that of the
simplest pamphlet’s readability level. Seventeen patients read at a level below
grade 6, which is the lowest readability level of a pamphlet.
f. 17/63  .27, so 27% of the patients have a reading level below that of the simplest
pamphlet.

••• Homework Activities

Activity 8-8: Human Ages
a. Answers will vary. The mean is most likely greater than the median, because
there are probably more humans alive in the world who are young. Thus, the
distribution of ages is skewed to the right and the mean will be greater than the
median.
b. Answers will vary. Because of the relatively good healthcare in the United States
compared to many other countries, the mean age of Americans is likely to be
greater than the mean age of all human beings.
c. Answers will vary. The 2000 U.S. census gives the mean U.S. age as 36.6 years
and the median as 35.9 years. The median age for all human beings (2007) is
28 years.
Note: See [Link]
WPA2007/[Link] or [Link]
publications/the-world-factbook/fields/[Link], which
includes a country-by-country list.
 Activity 9-1 155

students have successfully rid themselves of common misconceptions; you might also
consider using this activity as a group quiz. Activity 9-22 is similar to 8-22 in that it
asks students to create examples of hypothetical data that satisfy various properties.
Note that Activity 9-23 introduces two additional terms: midrange and midhinge.
Possible Assignments:
Day 1: Activity 9-7 on standard deviations; Activity 9-18 on IQRs; Activity 9-19
if you did the memory experiment here or in Topic 5; and Activity 9-21, which asks
students to reflect on information from a study. If you did part a of Activity 9-25 in
class, then assign Activity 9-25 part b for homework.
Day 2: Activity 9-8 on the IQR, SD, and the empirical rule; Activity 9-11 on
z-scores; and Activity 9-14 on the empirical rule.

Technology Ideas
Activity 9-22: Hypothetical Exam Scores (Fathom)
In this activity, students need to create hypothetical datasets of ten exam scores to
obtain certain properties. A combination of Fathom dynamic graphs and summary
tables can help ease the computational burdens as students test ideas.
Part a of this activity asks for a dataset where less than half the exam scores fall
within one standard deviation of the mean. To explore this property, drag an empty
case table from the shelf and create a new attribute called exam with ten hypothetical
scores as an initial guess. Create a dotplot of the exam scores and choose Graph | Plot
Value to show the values of mean( ) – s( ) and mean( ) + s( ), thus
clearly showing where “within one standard deviation of the mean” lies.
Drag the data points around to try to get at least six outside the boundaries. The
mean and standard deviation, and thus the boundaries, are automatically recomputed
as you move the points. To keep track of those values, create a summary table, drag the
exam attribute to show the mean, and then choose Summary | Add Formula to also
show s( ). Once a reasonable solution is found, the student may need to edit the data
to make all the exam scores integers.
For other parts of this activity, show Q1( ) and Q3( ) on the plot, or display
IQR( ) in the summary table. For a sneaky challenge, ask students to try to get all ten
exam scores more than one standard deviation away from the mean—it’s impossible!

Solutions
•••
9
In-Class Activities
Activity 9-1: Baseball Lineups
a. Observational units: baseball players
Explanatory variable: team Type: binary categorical
Response variable: age Type: quantitative
b. Here are the comparative dotplots:

Yankees
Team

22 23 24 25 26 27 28 29 30 31 32 33 34 35

Tigers
22 23 24 25 26 27 28 29 30 31 32 33 34 35
Age (in years)
156 Topic 9: Measures of Spread

The average age of both teams seems to be about the same (about 30 years), but
the spreads are quite different. The 2006 Tigers are much closer together in age
than the 2006 Yankees.
c. Yankees Mean: 29.7 years Median: 31.5 years
Tigers Mean: 30 years Median: 29.5 years
The centers of these distributions are relatively similar.
d. No; although the centers are the same, the spreads are very different, with the
Yankees having the youngest and oldest players in the two distributions and not
much consistency in the ages of their players.
e. The Yankees’ lineup appears to have greater variability in its ages.
f. Oldest: 35 years Youngest: 22 years Difference: 13 years
g. Oldest: 34 years Youngest: 25 years Difference: 9 years
h. Lower quartile: 28 years Upper quartile: 32 years IQR: 4 years
i. The Yankees have the greater age range and greater IQR. These values are
consistent with the answer to part e.
j. The average age of the starting lineups on both teams is about 30 years, but the
Tigers’ ages are fairly tightly clustered from 28–34 years, with the exception of
one player (Granderson) who is only 25 years old. In comparison, the Yankees’
ages range from a low of 22 years to a high of 35 years and also have a larger
interquartile range. It is difficult to judge the shape with these small sample sizes,
but the distribution of the ages for the Tigers appears more symmetric, whereas
the distribution of ages for the Yankees is more skewed to the left.

Activity 9-2: Baseball Lineups

a. Here is the completed table:

Player Age Deviation from Mean Absolute Deviation Squared Deviation

I. Rodriguez 34 34  30  4 4 16

Casey 32 2 2 4

Perez 33 3 3 9

Inge 29 1 1 1

Guillen 30 0 0 0

Monroe 29 1 1 1

Granderson 25 5 5 25

Gomez 28 2 2 4

Young 32 2 2 4

Robertson 28 2 2 4

Total 300 0 22 68
 Activity 9-3 157

b. This sum in the “deviation from mean” column is zero, which makes sense
because the positive deviations from the mean “cancel out” the negative deviations
from the mean.
c. See the table in part a. The sum of the absolute deviations is 22 years.
d. The mean of the absolute deviation is 22/10 or 2.2 years.
e. See the table in part a. The sum of the squared deviations is 68 years2.
f. 68/9  7.56 years2
g. 2.749 years
h. The standard deviation for the Tiger lineup’s ages is 2.749 years. The standard
deviation for the Yankees lineup’s ages is 4.62 years.
As expected, the Yankees’ standard deviation is larger because the ages tend to be
located farther from the average age.
i. Answers will vary by student expectation. This change will definitely affect the
range and the standard deviation, but it should have little or no effect on the IQR
because you are changing only an extreme value (endpoint).
j. See table in part k.
k. Here is the completed table:

Range Interquartile Range Standard Deviation

Original Data 9 4 2.749

With Large Outlier (43) 18 4 4.86

With Huge Outlier (134) 109 4 33.1

l. See table in part k. These results demonstrate that the IQR is resistant, but the
range or standard deviation is not. You know this because the value of the IQR
does not change when the value of the outlier changes, whereas the range and
standard deviation are affected dramatically.

9
Activity 9-3: Value of Statistics
a. Answers will vary by student prediction. Many students will pick class F, focusing
incorrectly on the irregularity in the heights of the bars.
b. Answers will vary by student prediction. Many students will incorrectly predict
that class J has the most variability because more of the possible data values appear
in the histogram. Students again may incorrectly see class H as having more
variability because they are looking at the differences in the heights of the bars.
c. Here is the completed table:

Class F Class G Class H Class I Class J

Range 6 8 8 8 8

Interquartile Range 2.75 3 0 8 4.5

Standard Deviation 1.769 2.041 1.18 4 2.657

158 Topic 9: Measures of Spread

d. According to these measures of spread, class G has more variability than class F,
because class G has more data values farther from the mean.
e. According to these measures of spread, class I has the most variability and class H
has the least variability. Class I has more of its data values at the extremes (and
far from the center), whereas most of class H’s observations are close to the mean.
Class J is in between the two.
f. Class F has more bumpiness in its histogram, but has less variability than class G.
g. Class J has the greatest number of distinct values but does not have the most
variability among classes H, I, and J.
h. No, based on the two previous questions, variability does not measure either
bumpiness or variety; variability measures spread from the center (mean). A
distribution can be very “bumpy” without having a great deal of variability, and
vice versa. It is more important to consider the overall tendency for data values to
be far from the center.
i. Many answers are possible, but all ten values need to be the same so the standard
deviation is zero.
j. Only one answer is possible: {1, 1, 1, 1, 1, 9, 9, 9, 9, 9}. This dataset maximizes
the distances of observations from the mean and has a standard deviation of 4.22.
Any other combination will have a smaller standard deviation. (Note: If you did
not balance the 1s and 9s, the mean would shift away from 5 and would put the
more frequent values closer to the mean.)

Activity 9-4: Placement Exam Scores

a. Yes, this distribution appears to be roughly symmetric and mound-shaped.
_
b. x  s  14.08
_
x  s  6.362
c. The scores in this interval are 7, 8, 9, 10, 11, 12, 13, and 14. There are 16  15 
17  32  17  21  12 or 146 of them. The proportion is 146/213 or .685.
d. The scores in this interval are 3, 4, . . . 15, 16, 17. There are 202 scores in this
interval and this proportion is 202/213 or .948.
e. This would include scores from 0 to 21, that is, all the scores. Thus 100% of the
scores fall within three standard deviations of the mean.

Activity 9-5: SATs and ACTs

a. If applicant Bobby scored 1740 on the SAT, he scored 1740  1500 or 240 points
greater then the SAT mean.
b. If applicant Kathy scored 30 on the ACT, she scored 30  21 or 9 points greater
than the ACT mean.
c. No; the scales on these two tests are different, so you cannot conclude that Bobby
outperformed Kathy simply because he scored more points above the mean than
Kathy did. The 240 and the 9 points cannot be directly compared.
d. Bobby scored 240/240 or 1 standard deviation above the mean.
e. Kathy scored 9/6 or 1.5 standard deviations above the mean.
 Activity 9-6 159

f. Kathy has the higher z-score.

g. Kathy performed better on her admissions test relative to her peers because her
z-score is greater.
h. Peter: z  (1380  1500)/240  0.5 Kelly: z  (15  21)/6  1.0
i. Peter has the higher z-score (less negative as the observation is not as far below
the mean).
j. When the observation is less than the mean, the z-score will turn out to be negative.

Activity 9-6: Marriage Ages

a. With 24 people in each group, the median ages are the average of the 12th and
13th ordered values. For husbands, the median age is (30  31)/2 or 30.5 years.
For wives, the median is (28  30)/2 or 29 years. For husbands, the mean age is
35.7 years and for wives, the mean age is 33.8 years. Husbands tend to be a little
less than two years older than their wives.
b. The lower quartile is the median of the bottom 12 ordered values, so the average
of the 6th and 7th values. For husbands, the lower quartile is (25  25)/2 or
25 years and the upper quartile is (51  38)/2 or 44.5 years. The IQR is, therefore,
44.5  25 or 19.5 years. You can see this by examining the sorted ages for
husbands:
19 23 23 25 25 25 ↓ 25 26 26 29 29 30 ↓ 31 31 31 34 35 38 ↓ 51 54 54 60 62 71
For wives, the lower quartile is (24  24)/2 or 24 years and the upper quartile is
(39  44)/2 or 41.5 years, so the IQR is 41.5  24 or 17.5 years. The standard
deviations are 14.6 and 13.6 years for husbands and wives, respectively. These
calculations indicate that the middle 50% of husbands’ ages cover a slightly
greater distance than the wives’ ages by 2 years and that the husbands’ ages
typically lie slightly farther from the mean, by approximately 1 year on average.
c. The age distributions are quite similar for husbands and wives. Both are skewed to
the right, centered around the low 30s or so, with considerable variability from the
upper teens through low 70s. The husbands are a bit older on average, and their
ages are a bit more spread out than the wives’ ages.

9
d. The ordered difference in couple’s ages are
7, 5, 5, 2, 1, 1, 0, 0, 1, 1, 1, 1, 1, 2, 2, 3, 3, 3, 3, 5, 7, 8, 10, 15
The median is the average of the 12th and 13th ordered values: (1  1)/2 or 1 year.
The mean is the sum of these differences divided by 24, which turns out to be
45/24 or 1.9 years.
Notice that the mean of the age differences is equal to the difference in mean ages
between husbands and wives: 1.9  35.7  33.8. But this property does not quite
hold for the median.
e. The quartiles are 0.5 and 3, so the IQR is 3.5 years. The standard deviation of these
age differences is 4.8 years. The IQR of the differences and the standard deviations of
the differences calculated here are less than the individual IQRs (19.5 and 17.5) and
the individual standard deviations (14.56 and 13.56) calculated in part b.
160 Topic 9: Measures of Spread

f. To be within one standard deviation of the mean is to be within 1.9  4.8 years,
which means between –2.9 and 6.7 years. Seventeen of the age differences fall
within this interval, which is a proportion of 17/24 or .708, or 70.8%. This
percentage is quite close to 68%, which is what the empirical rule predicts.
Because the distribution of the age differences does look fairly symmetric and
mound-shaped, this outcome is not surprising.
g. The mean and median indicate that, on average, people marry someone within
a couple years of their own age. More importantly, the measures of spread are
fairly small for the differences, much smaller than for individual ages. This result
suggests that there is not much variability in the differences, which suggests that
people do tend to marry people of similar ages.
h. The differences have less variability because even though people get married from
their teens to seventies (and beyond), they tend to marry people within a few years
of their own age.

••• Homework Activities

Activity 9-7: February Temperatures
a. Answers will vary by student predictions.
b. For Lincoln, the standard deviation is 15.9°F. For San Luis Obispo, the standard
deviation is 9.8°F. For Sedona, the standard deviation is 6.7°F.

Activity 9-8: Social Acquaintances

a. Answers will vary by class. Here are some example answers.

1 1134579

2 1345568

3 56778

4 00248

5 012468

6 59

7 6

10 5

11

12 4

leaf unit  one person

b. The median is 37 people; the upper quartile is 52; the lower quartile is 23; and
the interquartile range is 29 people.
c. The mean is 40.83 people and the standard deviation is 25.25 people.
d. You calculate 40.83  25.25  [15.58, 66.08], so 26/35 or 0.743 of the students’
results fall within one standard deviation of the mean.
172 Topic 10: More Summary Measures and Graphs

Activity 10-5: Digital Cameras (Fathom)

For this activity, students need to compare graphical and numeric summaries between
different types of digital cameras using the data in [Link].
Fortunately, Fathom allows students to drag a categorical attribute (such as Type
of camera) to the other axis of either a summary table or a graph to show separate
summaries or graphs for each category.
For part a, drag the Price attribute to the space labeling the rows of a blank
summary table to display (by default) the mean. Choose Summary | Add Five Number
Summary to add the five-number summary (stretch the summary table if necessary to
see all the results), and then drag the Type attribute from the case table to the top of
the column to create new columns for each of the camera types. Note that you could
also place the original quantitative Price attribute in the top column and the categorical
Type attribute along the rows, but because Fathom stacks the five-number summary
vertically, this would make it more difficult to compare the groups.
You can use a similar process for comparative plots. Drag the Price attribute to the
horizontal axis of a blank graph to create a dotplot, and change the type of graph to a
Boxplot. Drag the categorical Type attribute to the vertical axis of the graph to show
comparative boxplots for the different types of digital cameras.
For parts b and c that ask you to order the categories based on median price or
the IQR, you can produce a visual ordering by dragging the category names in the
comparative boxplots to rearrange the groups. Unfortunately, you can’t do a similar
rearrangement of columns in Fathom’s summary table. Finally, dragging the Ratings
attribute to either the summary table or the graph to replace Price makes the conversion
to answer part d easy, although you might prefer to do it on a copy of the original
graph (choose Object | Duplicate Graph).

Activity 10-5: Digital Cameras (Graphing Calculator)

In this activity students are asked to use their calculator to compute five-number summaries
and produce comparative boxplots of the camera prices, based on the camera type.
Unfortunately, the TI-83/TI-84 Plus can only do three plots at a time. You will need to tell
students to make at least four different 3-plot setups to do a complete comparison of the five
camera types or have students share calculators.

Solutions
••• In-Class Activities
Activity 10-1: Natural Selection
a. Observational units: adult male sparrows
Explanatory variable: total length Type: quantitative
Response variable: whether the sparrow survived Type: binary categorical
b. This is an observational study because Bumpus simply recorded this information
about the sparrows; he did not impose any treatment on them.
c. Median: 159 mm
Lower quartile: 158 mm Upper quartile: 160 mm
Minimum: 153mm Maximum: 166 mm
 Activity 10-1 173

d. Median: 159 mm
Lower quartile: 158 mm Upper quartile: 160 mm
Minimum: 153 mm Maximum: 166 mm
e. The sparrows that died tended to be longer than the sparrows that survived.
Seventy-five percent of those that died were at least 161 mm long, but 75% of those
that survived were shorter than 160 mm. The typical length for the sparrows that
died was 162 mm, and the typical length was only 159 mm for those that survived.
f. The following boxplots display the distribution of lengths for the sparrows that
survived and died:

Died
Status

Survived

152 154 156 158 160 162 164 166

Total Length (in mm)

g. Here is the completed table:

QL QU IQR  QU − QL 1.5  IQR QL − 1.5  IQR QU  1.5  IQR

158 160 2 3 155 163

Outliers would be outside the interval [155, 163], so there are three low outliers
(153, 154, and 154) and two high outliers (165 and 166).
h. The following modified boxplots display the lengths for the surviving sparrows
and for the sparrows that died:

Died *
Status

Survived * ** * *

152 154 156 158 160

Total Length (in mm)

i. There does appear to be a substantial difference in the lengths between the

162 164 166
10
sparrows that survived and those that did not survive the storm. Those that
survived the storm tended to be about 3 inches shorter at each quartile than those
that died, although there was more variability and outliers (on both ends) in the
group that survived.
j. No, you cannot draw a cause-and-effect conclusion between sparrow length and
survival because this is an observational study, not an experiment.
174 Topic 10: More Summary Measures and Graphs

k. Yes, it is clear from this study that shorter sparrows were more likely to survive
the storm. This makes no claim about why they were more likely to survive the
storm—only that the shorter sparrows tended to survive more frequently than did
the longer sparrows.

Activity 10-2: Roller Coasters

a. Fifty percent of the steel coasters have a top speed of 60 mph or greater. The
boxplot shows that the median for the steel coasters is 60 mph.
b. Twenty-five percent of the wooden coasters have a top speed of 60 mph or greater.
The boxplot shows that the upper quartile of the wooden coasters is 60 mph.
c. The lower quartile for both types of coasters is 50 mph, so 75% of both types of
coasters have top speeds of 50 mph or greater.
d. You cannot tell which type of coaster has a higher proportion of coasters with a
top speed greater than 45 mph because 45 mph is not a quartile for either type of
coaster.
e. The steel coasters have more variability. This is obvious from the boxplots because
the boxplot for the steel coasters extends from about 25 mph to about 100 mph,
whereas the boxplot for the wooden coasters extends from only about 40 mph
to about 65 mph. Similarly, the box itself, representing the middle 50% of the
distribution, is much longer for the steel coasters than the wooden coasters.
f. The boxplots do not show how many coasters there are of either type; they
indicate only where the quartiles (percentages) of the distribution lie.

Activity 10-3: Ice Cream Calories

a. Here is the completed table:

Minimum Lower Quartile Median Upper Quartile Maximum

Ben & Jerry’s 110 220 265 290 360

Cold Stone 130 360 390 400 440

Dreyer’s 90 110 120 150 190

b. The following boxplots display the distribution of calorie amounts for the three
brands:

Ben & Jerry’s **

Cold Stone Creamery * * * * ** ** *

*

Dreyer’s

100 150 200 250 300 350 400 450

Calories per Serving
 Activity 10-4 175

Dreyer’s ice cream appears to have significantly fewer calories than the other two
brands, as well as less variability among its calorie amounts. All of the Dreyer’s
ice cream flavors have fewer than 200 calories, whereas only 25% of the Ben &
Jerry’s flavors have 220 or fewer calories. There is a great deal of variability in
the number of calories for the Ben & Jerry flavors as they range from 110 to
360 calories, whereas (excluding outliers) the Cold Stone Creamery flavors range
from 360 to 440 calories.
c. The serving sizes may not be the same for all three brands. This would make it
difficult to compare the calories as given.
d. You could convert the Cold Stone Creamery serving from 170 grams to the
comparable measure of volume in 1/2 cups.
e. Divide each of the Cold Stone Creamery listings by 170 grams, then multiply by
73 grams per 1/2 cup.
f. You calculate new value  (old value)/170 * 73.
g. Here is the five-number summary for Cold Stone’s calorie amounts using the “per
half cup” scale: min  55.82, Q L  154.59, median  167.47, Q U  171.76,
max  188.94.
h. The following boxplots display the three distributions of calorie amounts:

Ben & Jerry’s **

Cold Stone Creamery * * ** ** ** *

*

Dreyer’s

50 100 150 200 250 300 350 400

Calories per Serving

Once you adjust the ice cream servings so that they all have the same serving size,
Cold Stone Creamery still has several flavors that are low outliers (meaning that
these flavors have an unusually small amount of calories per serving). Excluding
the outliers, the Dreyer’s flavors are generally the lowest in calorie content,
followed by the Cold Stone Creamery flavors, which have a very narrow spread
(from only about 155–189 calories per 1/2 cup); at least 75% of the Ben & Jerry
flavors have more calories than either of the other two brands.

Activity 10-4: Fan Cost Index

10
a. Highest FCI team: Boston Value: $287.84
Lowest FCI team: Kansas City Value: $120.34
b. The following dotplot displays the distribution of the FCI values:

125 150 175 200 225 250 275

Fan Cost Index (in dollars)
176 Topic 10: More Summary Measures and Graphs

The fan cost index ranges from a low of about $120 to a high of about $220,
with an outlier of Boston at $287.84. The average FCI is $171, and the median
is slightly lower at $166. The standard deviation is $35.05, whereas the IQR
is $47.76.
c. Answers will vary.
d. Answers will vary.
e. Answers will vary.
f. Highest team: NY Mets Value: $4.75
Lowest team: Baltimore, Milwaukee, and Kansas Value: $2.00
g. The number of ounces in a “small” soda or beer is not the same in every park.
h. Highest team: LA Dodgers Value: $0.354/oz
Lowest team: Pittsburgh Value: $0.1125/oz

Activity 10-5: Digital Cameras

a. The table below reports the five-number summaries, as reported by the software
package Minitab. The boxplots of camera prices follow the table.

Minimum Lower Quartile Median Upper Quartile Maximum

Advanced Compact $280 $400 $560 $745 $840

Compact $140 $210 $290 $327.5 $480

Subcompact $185 $273.75 $300 $330 $450

Super-Zoom $250 $311.25 $357.5 $487.5 $720

Advanced Compact
Camera Type

Compact

Subcompact * **

Super-Zoom

100 200 300 400 500 600 700 800 900

Price of Camera (in dollars)

There is considerable overlap in prices among these four groups, but some camera
types (e.g. advanced compact) do tend to cost more than other types.
b. Advanced compact cameras tend to cost the most, followed by super-zoom
cameras. Compact and subcompact camera prices are similar, but the subcompact
cameras cost a bit more on average than the compact cameras.
c. Advanced compact cameras have the most spread in terms of prices, again
followed by super-zoom cameras. But compact cameras have more spread in prices
than do subcompact cameras, which have the least variability in prices.
 Activity 10-6 177

d. The boxplots of camera ratings are shown in the following the table. The five-
number summaries, as reported by the software package Minitab, are as shown here:

Minimum Lower Quartile Median Upper Quartile Maximum

Advanced Compact 63 69 70 73 78

Compact 62 65 71 73.5 76

Subcompact 53 61.75 65.5 69.25 75

Super-Zoom 66 72.25 75.5 79 81

Advanced Compact
Camera Type

Compact

Subcompact

Super-Zoom

50 55 60 65 70 75 80
Rating Score

The super-zoom cameras tend to have the highest ratings, and the subcompact
ones tend to rate the lowest. The subcompact cameras also have the most
variability in ratings.
e. Even though the advanced compact cameras have the highest median price by far,
their median rating is surpassed by both super-zoom and compact cameras. In
fact, compact cameras have the second-highest median rating despite having the
second-lowest median price.

••• Homework Activities

Activity 10-6: Natural Selection
a. Yes, the data suggest that the sparrows that died weighed a little more than the
sparrows that survived. The medians are very close, but every quartile for the
sparrows that died is a little larger than for the sparrows that survived. This
difference does not appear to be very substantial.
b. For the sparrows that died, the IQR is 26.95  25.2 or 1.75 grams. You calculate
1.5  1.75 or 2.625, which gives [25.2  2.625, 26.95  2.625]  [22.575,
10
29.575]. Any outliers had weights outside this range.
For the sparrows that survived, the IQR is 26.5  24.3 or 2.2 grams. You
calculate 1.5  2.2 or 3.3, which gives [24.3  3.3, 26.5  3.3]  [21, 29.8].
Any outliers had weights outside this range.
c. There were no outliers among the sparrows that survived. There was at least
one high outlier among the sparrows that died (because the maximum weight is
greater than 29.575 grams).
 Activity 11-1 203

11
Technology Ideas
Activity 11-18: Dice-Generated Ice Cream Prices (Fathom)
This activity asks students to use technology to simulate (1000 times) the prices for an ice
cream special that arise from rolling two dice and following the larger result with the smaller
result to create a two-digit price. Here’s one way to accomplish this simulation using Fathom:
1. Drag a new blank collection from the shelf, choose Collection | New Cases from
the menu, and enter 1000 to create that many cases (rows) in the collection.
2. Drag an empty table from the shelf and create two new attributes, Die1 and
Die2, to contain the random dice rolls for each price. Select the Die1 column and
choose Edit | Edit Formula to enter the formula randomInteger(1,6) to
simulate the roll of a die. Repeat for the Die2 column.
3. Create a third attribute called Price to contain the simulated prices. Use the formula
editor again and enter a formula that begins with if(Die1>Die2). When the
if( ) structure is entered in a Fathom formula, the program creates two question
marks. Replace the first (top) question mark with the computation when the
condition is true (10*Die1+Die2), and replace the second (bottom) question
mark with the computation when the condition is false (10*Die2+Die1).
4. Use the values that are computed for the Price attribute to create a dotplot of the
1000 simulated prices and compute the mean.

Activity 11-18: Dice-Generated Ice Cream Prices

(Graphing Calculator)
For a graphing calculator alternative for this activity, the randInt feature can be used to
simulate the rolling of a die. For example, to simulate 50 repetitions you would enter
randInt(1,6,50) at your prompt and then press Õ. These 50 numbers can then be
used to calculate the price of 25 ice-cream cones by pairing numbers, i.e, the first and
second entries, the third and fourth entries, etc. Remember that the highest number of
the pair is used in the tens column. This activity might be a good time to have students
explore all of the random number generators located in MATH PRB, focusing specifically
on rand and randInt. Later on in the course, you can discuss randBin and randNorm.
Note: In this topic, you can use calculator programs for Homework Activities
11-18, 11-19, and 11-21. These programs can take quite a bit of time to complete. For
this reason, avoid assigning all three activities for homework on the same night. This
topic is also a good time to show students how to archive their lists and programs.

Solutions
••• In-Class Activities
Activity 11-1: Random Babies
Answers will vary. Here is one representative set of answers:
a. One mother received her own baby.
b. Here is a record of the random “dealing”:

Number of Repetitions 1 2 3 4 5

Number of Matches 1 1 2 1 2
204 Topic 11: Probability

c. Here is a completed table for the example repetitions:

Column 3: Column 4: Cumulative

Column 1: Number Column 2:
Cumulative Number Proportion of Trials
Student # of Trials with at Cumulative Number
of Trials with at with at Least One
Least One Match of Trials
Least One Match Match

1 5 5 5 1

2 2 10 7 .7

3 4 15 11 .7333

4 5 20 16 .8

5 4 25 20 .8

6 4 30 24 .8

7 4 35 28 .8

8 0 40 28 .7

9 3 45 31 .6889

10 3 50 34 .6800

11 2 55 36 .6545

12 1 60 37 .6167

13 3 65 40 .6154

14 3 70 43 .6143

15 4 75 47 .6267

16 4 80 51 .6375

17 4 85 55 .6471

18 3 90 58 .6444

19 5 95 63 .6632

20 2 100 65 .6500

d. The following graph displays the cumulative proportion of trials with at least one
match vs. the cumulative number of trials.

1.2

1.0
Trials with at Least One Match
Cumulative Proportion of

0.8

0.6

0.4

0.2

0
5 15 25 35 45 55 65 75 85 95
Cumulative Number of Trials
 Activity 11-1 205

11
e. The proportion of trials that result in at least one mother getting the correct baby
fluctuates more at the beginning of this process.
f. Yes, the relative frequency appears to be settling down and approaching one
particular value. Answers will vary as to what that value is, but it should be in the
ballpark of .627.
g. Here is a completed table with counts and proportions recorded:

Number of Matches 0 1 2 3 4 Total

Count 34 33 28 0 5 100

Proportion .34 .33 .28 0 .05 1.00

h. The proportion is .33  .28  .05 or .66.

i. Pr(no matches)  .34
j. Pr(at least one match)  .66
k. Answers will vary. The following are results of one simulation:

Number of Matches 0 1 2 3 4 Total

Count 359 335 267 0 39 1000

Proportion .359 .335 .267 0 .039 1.00

l. Yes, these simulation results are reasonably consistent with the class results.
m. Pr(at least one match)  1  .359  .641.
n. Yes, this graph appears to be fluctuating less as more trials are performed,
approaching a limiting value:

o. Getting exactly three matches is impossible because if three mothers and

babies matched, then the fourth mother would have to be matched correctly
with her baby. There would be no “incorrect” baby left for her to be paired
with.
206 Topic 11: Probability

p. It is not impossible to get four matches, but it seems very rare/unlikely because it
appears to have happened only about 39 out of 1000 times.
q. No, results of zero, one, or two matches do not seem to be unlikely. Each would
appear to occur more than 25% of the time; a zero match and one match even
occur more than one-third of the time each.

Activity 11-2: Random Babies

a. There are 24 different arrangements for returning the four babies to the four
mothers.
b. Here are the number of mothers who get the correct baby for each arrangement:
1234 4 1243 2 1324 2 1342 1 1423 1 1432 2
2134 2 2143 0 2314 1 2341 0 2413 0 2431 1
3124 1 3142 0 3214 2 3241 1 3412 0 3421 0
4123 0 4132 1 4213 1 4231 2 4312 0 4321 0
c. 4: 1 3: 0 2: 6 1: 8 0: 9
d. 4:1/24  .04167 3: 0 2:6/24  .250 1: 8/24  .333 0: 9/24  .375
e. The empirical probabilities from our class are reasonably close to these theoretical
probabilities. The applet simulation probabilities are even closer.
f. The mean number of matches per repetition is
0.34  1.33  2.28  3.0  4.5  ____
___________________________ 109  1.09
100 100
(Results will differ.)
g. The expected value for the number of matches from the probability distribution is
0  2 ___
1  3 ___ 6  1 ___ 9  ___
8  0 ___ 24  1
4  ___
24   24   24   24   24  24
This is very close to the value found in part f.
h. Pr(number of matches  1)  .333. No, you do not expect this value to occur most
of the time; you expect it to occur about 1/3 of the time, which is significantly
less than half of the time.

Activity 11-3: Family Births

a. Probability does not have a memory. With each birth, there is a 50% chance of a
girl, but that probability does not change after the first child is born. If the first
child is a girl, the probability that the second child is a girl is still 1/2—not zero.
This probability applies “in the long run”—about half of all children born will be
girls—but not for every family.
b. There are four equally likely outcomes (not three) for a family with two children.
The outcomes are: Boy first and then Boy; Boy first and then Girl; Girl first and
then Boy; and Girl first and then Girl. So, the probability that the couple has one
boy and one girl (GB or BG) is 2/4 or 50%, and the probability that the couple
has two children of the same gender (GG or BB) is also 2/4 or 50%.
 Activity 11-4 207

11
c. Answers will vary. The following were obtained using row 31 of the table.

Family 1 Family 2 Family 3 Family 4

Child 1 Child 2 Child 1 Child 2 Child 1 Child 2 Child 1 Child 2

Random Digit 3 4 2 8 2 5 0 7

Gender B G G G G B G B

Number of Girls 1 2 1 1

d. Here are the proportions of each type:

Two Girls Two Boys One of Each

Tally (Count) 4 5 11

Proportion .20 .25 .55

e. No, it does not appear that the probability of each of these outcomes is 1/3. It
appears that the probability of having one child of each gender is about twice that
of having two girls or two boys.
f. You can obtain better empirical estimates of these probabilities if you simulate
more families.
g. Here is a completed table of simulation results:

Two Girls Two Boys One of Each

Tally (Count) 96 205 99

Proportion 96/400  .24 205/400  .5125 99/400  .2475

h. No, it does not appear that the probability of each of these outcomes is 1/3. It
appears that the probability of having one girl and one boy each is about .5,
whereas the probability of having two girls or of two boys is about .25.
i. Two girls: Pr(GG)  1/4  .25 Two boys: Pr(BB)  1/4  .25
One of each: Pr(BG or GB)  1/2  .5
Yes, these probabilities are reasonably close to the empirical estimates from the
class simulations.

Activity 11-4: Jury Selection

a. The probability that he/she will be age 65 or older is .20.
b. Answers will vary by student guess.
c. At least one-third: (135  60  20)/1000; the approximate probability is .25.
d. Answers will vary by student guess.
e. The approximate probability is 2/1000 or .002.
208 Topic 11: Probability

f. A sample of size 12 is more likely to contain at least one-third senior citizens. This
result makes sense because only 20% of the population is senior citizens, so it is
easier to select 4 of 12 jurors who are senior citizens than it is to select 25 of 75.
There should be more sampling variability (easier to get an unlucky result) with
the smaller sample size.
g. Based on the simulations, you see that a sample size of 75 is more likely to contain
between 15% and 25% (11.25 and 21) senior citizens than a sample size of 25.
This result makes sense because with the larger sample size, the sample proportion
should be close to the population proportion more often.
h. The previous questions show that a larger sample size is more likely to produce a
sample proportion that is close to the population proportion, and smaller sample sizes
are more likely to produce samples with greater variability from sample to sample.

Activity 11-5: Treatment Groups

a. You could assign each subject a number from 1–6. Then, as the die is rolled, you
could assign the subjects corresponding to the first three (distinct) numbers rolled
to the new treatment group. The process would be similar using a random digit
table, where you could simply skip over the digits 7–9 and 0.
b. Answers will vary. (You should find roughly 2 randomizations with 0 women
in the new group, 18 randomizations with 1 woman, 18 randomizations with
2 women, and 2 randomizations with 3 women.)
c. Answers will vary. Determine the total number of repetitions with 1 woman
or 2 women in the new group, and divide that number by 40. (Your empirical
estimate should be fairly close to .9.)
d. Performing many more than 40 repetitions of the random assignment should
produce a more accurate estimate.
e. Answers will vary. Calculate this empirical estimate of the expected value by
multiplying the number of women by the count of repetitions that produced that
number, and then divide that value by 40. (Your empirical estimate should be
fairly close to 1.5.)
f. Random assignment ensures that all possible ways to assign these subjects to the
treatment groups are equally likely to occur.
g. The exact probabilities are 1/20 or .05 for 0 women in the new group, 9/20 or .45
for 1 woman in the new group, 9/20 or .45 for 2 women in the new group, and
1/20 or .05 for 3 women in the new group.
h. The probability of a 1/2 or 2/1 gender breakdown is 18/20 or .9.
i. The probability that the two genders are completely separated into two groups (all
men in one group and all women in the other) is 2/20 or .1. This result is small
enough to be perhaps mildly surprising, but it is not small enough to be very
surprising. Although randomization should balance out the gender breakdown
between the two groups in the long run, with such a small study group, you could,
simply by chance, still end up with all males in one group and all females in the
other group.
j. The expected value is 0(1/20)  1(9/20)  2(9/20)  3(1/20) or 30/20 or
1.5 women.
 Activity 11-7 209

11
k. This expected value makes sense because with three women randomly assigned
among two groups, you would expect half of them to be assigned to each group
in the long run, and half of 3 is 1.5.

••• Homework Activities

Activity 11-6: Random Cell Phones
a. Write the names of the five executives on five index cards of the same size. Shuffle
the cards well and deal them onto a sheet of paper on which you have drawn five
boxes that are labeled with the names of the executives. Check how many matches
you have between names on the index cards and names in the boxes. Record this
value. Then reshuffle the cards and deal again. Repeat many times.
b. Let 123 mean that the first cell phone went to the first executive, the second cell
phone to the second executive, and the third cell phone to the third executive. All
of the possibilities are listed here:
123 213 321
132 231 312
c. Pr(nobody gets the correct phone)  2/6  1/3
Pr(exactly one person gets the correct phone)  3/6
Pr(exactly two people get the correct phone)  0
Pr(all three people get the correct phone)  1/6
Pr(at least one person gets the correct phone)  1  1/3  2/3
d. If you repeated this exact “experiment” over and over again, in the long run you
would expect to observe nobody getting the correct phone in about 84.6% of the
trials.

Activity 11-7: Equally Likely Events

a. These outcomes are equally likely. Because the die is fair, each side should be just
as likely to come up as any other side.
b. No, these outcomes are not equally likely. A 2 can only be summed from a roll
of snake eyes (1  1), but a 7 can be summed from rolls of 1  6, 2  5, 3  4,
4  3, 5  2, and 6  1. So Pr(sum of 2)  1/36, whereas Pr(sum of 7)  6/36.
c. You would assume that a coin flip would have equally likely outcomes.
d. You would assume that spinning a coin would result in equally likely outcomes.
e. You would assume this spinning would result in equally likely outcomes.
f. No, these outcomes are not equally likely. Among other reasons, an A is typically
90–100, whereas an F is generally any average less than 60. These do not cover
equivalent ranges. You might not expect to see a similar proportion of Fs as As.
g. No, these outcomes are not equally likely. (We hope the probability that California
experiences a catastrophic earthquake within the next year is less than 1/2.)
h. No, these outcomes are not equally likely, unless you frequent bad restaurants.
The probability that your waiter or waitress brings you the meal you ordered is
greater than 1/2.
226 Topic 12: Normal Distributions

Solutions
••• In-Class Activities
Activity 12-1: Body Temperatures and Jury Selection
a. Both distributions are roughly symmetric and mound-shaped.
b. The following sketch approximates the general shape in the two histograms:

c. The dashed curve (C) has a mean of 50 and a standard deviation of 10. The dashed
curve (B) has a mean of 70 and a standard deviation of 10. The solid curve (A) has a
mean of 70 and a standard deviation of 5. Here is the labeled drawing:

10 20 30 40 50 60 70 80 90 100 110
Exam Score

d. Yes, these predictions are quite close to each other and to what the empirical rule
would predict.
e. The z-score for the value 97.5 in the distribution of body temperatures is z 
(97.5  98.249)/.733 or 1.02.
f. The z-score for the value 11.5 in the distribution of number of senior citizens in
the jury pool is z  (11.5  14.921)/3.336 or 1.025.
g. The two z-scores are almost identical.
h. Both z-scores indicate that the observations are just over one standard deviation
below their respective means.
 Activity 12-2 227

i. The value is .1515.

j. Yes, this value is reasonably close to .146 and .153.

Activity 12-2: Birth Weights

12
a. The following graph shows the shaded region whose area corresponds to the
probability that a baby will have a low birth weight:

1000 1500 2000 2500 3000 3500 4000 4500 5000 5500
Birth Weight (in grams)

b. Answers will vary by student guess.

c. The z-score for a birth weight of 2500 grams is z  (2500  3300)/570 or 1.40.
d. Pr(Z  1.40)  .0808
e. The applet gives the probability of .0802.
f. Using the applet, Pr(birth weight  4536)  Pr(Z  2.17)  .0151.
g. One way to use Table II to answer part f is to look up the z-score and subtract the
given area from 1.000. A second way is to look up the opposite z-score (2.17)
because the curve is symmetric.
h. Using the applet, Pr(3000  birth weight  4000)  Pr(.53  Z  1.23) 
.5910. Using Table II, .8907  .2981  .5926.
i. For babies with low birth weight, you calculate (331772/4112052) or .081, which
is very close to the predicted .08 from your calculations in part d. For babies
between 3000 and 4000 kg, you calculate (2697819/4112052) or .656, which is
not as close to (but not unreasonably far from) the prediction in part h.
j. The lightest 2.5% corresponds to a z-score of 1.96 (using Table II or technology).
You calculate z  1.96  (x  3300)/570; weight  x  2182.8 kg.
k. The heaviest 10% (or bottom 90%) corresponds to a z-score of 1.28 (using Table II
or technology). You calculate z  1.28  (x  3300)/570; weight  x  4029.6 kg.
228 Topic 12: Normal Distributions

Activity 12-3: Blood Pressures and Pulse Rate Measurements

a. The diastolic blood pressure is the most symmetric and mound-shaped of the
three dotplots, so it is most likely to have come from a normal population.
b. The systolic blood pressure is least likely to have come from a normal distribution
because its distribution is not symmetric or mound-shaped.
c. Here are normal probability plots for systolic blood pressure, diastolic blood
pressure, and pulse rate:

99.9 99.9
99 99
90 90
Percentage

Percentage
50 50
10 10
1 1
0.1 0.1
50 100 150 200 30 60 90 120
Systolic Blood Pressure Diastolic Blood Pressure

99.9
99
90
Percentage

50
10
1
0.1
30 60 90 120

Pulse Rate

These probability plots confirm that the systolic pressures were unlikely to have
come from a normal distribution, and the diastolic pressures could quite possibly
have come from a normal distribution.

Activity 12-4: Criminal Footprints

a.

Females

Males

10 15 20 25 30 35 40
Footprint Length (in cm)
 Activity 12-4 229

b. Because you are dealing with a male footprint, the z-score is (22 – 25)/4 or –0.75.
Using Table II (or technology), the probability of the footprint being less than
22 centimeters is .2266. So, roughly 22.66% of men have a footprint smaller than
22 centimeters and would be misclassified as female.

12
Females

Males

10 15 20 25 30 35 40
z= -0.75
Footprint Length ( in cm)

c. Now you are dealing with a female footprint, so the z-score is (22 – 19)/3 or 1.00.
Using Table II (or technology), the probability of the footprint being longer than
22 centimeters is 1 – .8413 or .1587. This indicates that about 16% of females
would be mistakenly identified as male.

Females

Males

10 15 20 25 30 35 40
z = 1.00
Footprint Length (in cm)

d. Using Table II or technology, the z-score to produce a probability of .08 is –1.41.

To find the corresponding male foot length, you need to solve –1.41  (x – 25)/4,
which gives you x  25 – 1.41(4)  25 – 5.64  19.36 centimeters. You can also
think of this as subtracting 1.41 standard deviations of 4 from the mean of 25.
Notice that this new cutoff value (19.36 centimeters) is much smaller than before
230 Topic 12: Normal Distributions

(22 centimeters) in order to reduce the probability of classifying a male footprint

as having come from a woman.

Females

Males

.08

10 15 20 25 30 35 40
z = -1.41
Footprint Length (in cm)

e. Using this new cutoff value of 19.36 centimeters, the z-score for a female footprint
is (19.36 – 19)/3 or 0.12. The probability of a female footprint being longer than
19.36 centimeters is 1 – .5478 or .4522 (using Table II or technology).

Females

Males

.4522

10 15 20 25 30 35 40
z = 0.12
Footprint Length (in cm)

f. The probability of misclassifying a female footprint as a male footprint is

much greater than before (.4522 as opposed to .1587). In order to reduce the
probability of one type of error (misclassifying a man’s footprint as having
come from a woman) from .2266 to .08, the probability of making the other
kind of error increases substantially. This exercise reveals that there is a trade-
off between the probabilities of making the two kinds of errors that can occur:
you can reduce one error probability but only by increasing the other error
probability.
 Activity 13-1 249

••• Homework Activities

As we’ve repeated several times, the concept of sampling distribution is very challenging,
so it’s more important than ever to give students opportunities to assess and develop
their understanding with homework activities. Do not be surprised if students struggle,
and we recommend giving as much detailed feedback as possible. Try not to overwhelm
students, however. In fact, you may want to assign fewer activities than usual but then
require very clear and detailed solutions. We especially recommend helping students

13
to practice distinguishing between the population, sample, and sampling distributions,
perhaps asking them to sketch all three distributions. Activities 13-5 and 13-6 provide
practice with the terms parameter and statistic. Note that Activities 13-9, 13-10, 13-11,
and 13-13 ask students to perform simulations. The remaining homework activities ask
students to work with the CLT more mathematically.
Possible assignment: Activities 13-5 (parameter vs. statistic; it is not necessary to assign
all 15 parts), 13-9 (requires a simulation), and 13-12 or 13-16 (thinking about plausibility)

Technology Ideas
Activity 13-9: Pet Ownership (Fathom)
Part d of this activity asks students to use technology to simulate the random selection
of 500 samples, each of size 200 from a population of households where 25% have cats.
Although this can be done within the Reece’s Pieces applet, students need to ignore the
candy context. To accomplish this simulation using Fathom, follow these steps:
1. Create a new collection and choose Collection | New Cases to specify 500 cases
for the samples.
2. Label the first attribute in this collection as Cats and choose Edit | Edit Formula
to enter the formula randomBinomial(200,0.25) to create the simulated
counts of cats in each sample.
3. Create a second attribute Phat to contain the proportion for each sample, and in
the formula editor, enter Cats/200 as the sample proportion.
4. Counting how many sample proportions are within .05 of the population propor-
tion can be a bit tricky because attempting to sort the column will cause Fathom
to recalculate the random values. Instead, double-click on the collection to show its
inspector, and on the Measures panel, define a measure named Close and enter the for-
mula 500-Count(Phat.20)-Count(Phat.30) to compute the count.
These simulation steps can also be applied to Activities 13-10, 13-11, and 13-13.

Solutions
••• In-Class Activities
Activity 13-1: Candy Colors
Answers will vary. Here is one representative set of answers.
a. The count and proportion of each color in this sample are recorded in the table:

Orange Yellow Brown

Count 13 7 5

Proportion (Count/25) .52 .28 .2

250 Topic 13: Sampling Distributions: Proportions

b. This is a statistic. The symbol used to denote the proportion is p̂.

c. This is a parameter. The symbol used to denote the proportion is ␲.
d. No, you do not know the proportion of orange candies manufactured by
Hershey.
e. Yes, you know the proportion of orange candies among the 25 candies that you
individually selected.
f. It is very unlikely that every student in the class obtained the same proportion of
orange candies in his or her sample.
Answers will vary. Here is one representative set of answers:
g. Here is a dotplot of the sample proportions of orange candies:

.42 .48 .54 .60 .66 .72 .78 .84

Sample Proportion of Orange Candies

h. The observational units in this graph are the samples of 25 candies. The variable
being measured from unit to unit is the proportion of the sample that is colored
orange.
i. This dotplot of the sample proportions is symmetric and mound-shaped
(roughly normal), with a center of .5, and with most of the sample
proportions falling between .4 and .6 (min .36, max .68). The standard
deviation is .08.
j. Based on the sample results from this class, a reasonable guess for ␲ would be .5.
k. Most estimates would be reasonably close to ␲, but a very few estimates
would be way off. You can see this from the dotplot. Most of the class results
are the same (near .5), but a few of the class results are quite extreme (far
from .5).
l. If each student had taken samples of size 10 instead of size 25, you would expect
more variability (greater horizontal spread) in the dotplot.
m. If each student had taken samples of size 75 instead of size 25, you would expect
less variability (less horizontal spread) in the dotplot.

Activity 13-2: Candy Colors

Answers will vary. The following results are from one particular running of the applet.
a. p̂  .44
b. p̂  .52, .48, .52, .56, .40. No, you did not get the same sample proportion
each time.
 Activity 13-2 251

c. Here is a sketch of the results displayed in the dotplot:

13
.1 .2 .3 .4 .5 .6 .7 .8
∏
Mean = .448 SD = .099 ^p

d. Yes, the distribution appears roughly normal, centered at about .45, with a
standard deviation of about .1.
e. A normal curve seems to model the simulated sample proportions very well.
f. Mean of p̂ values: .449 Standard deviation of p̂ values: .100
g. Roughly speaking, more sample proportions are close to .45 than are far away
from it.
h. Here is the completed table:

Number of 500 Percentage of 500

Sample Proportions Sample Proportions

Within  .10 of .45 354 71.5%

Within  .20 of .45 473 95.6%

Within  .30 of .45 491 99.2%

i. About 95% would capture the actual population proportion.

j. No, you would not have any definite way of knowing whether your sample
proportion was within .20 of the population proportion. However, you could be
reasonably confident that your sample proportion was within .20 of the population
proportion because about 95% of the sample proportions would be within .2 of ␲.
k. Mean of p̂ values: .446 Standard deviation of p̂ values: .057
l. The shape is still roughly normal and the center is still about .45. The spread,
however, has decreased significantly (from .1 to about .057).
m. The applet reports that 460/500  92% of the sample proportions are within
.1 of .45.
n. This percentage is much greater (92% versus 71.5%) than it was when the sample
size was n  25.
o. The sample proportion is more likely to be closer to the population proportion
with a larger sample size.
p. You calculate .057  2  .114, and .45  .114  [.336, .564].
252 Topic 13: Sampling Distributions: Proportions

q. The applet reports that 336/500 or 95% of the sample proportions are within
.114 of .45.
r. About 95% of the students’ intervals would contain the actual population
proportion of .45.
s. Theoretical mean of p̂ values: .45
________


(.45)(.55)
Theoretical standard deviation of p̂ values: ________  .0995  .1
25
t. Theoretical mean of p̂ values: .45
________


(.45)(.55)
Theoretical standard deviation of p̂ values: ________  .057
75
u. No, the normal model does not summarize this distribution well. This is not a
contradiction to the Central Limit Theorem because n␲  25(.1)  2.5 10.

Activity 13-3: Kissing Couples

a. This is a parameter; ␲  .5.
b. You calculate n␲  124(.5)  62  10 and n(1  ␲)  124(.5)  62  10, so
the CLT does apply.
Shape: approximately normal
Center: ␲  .5
_______


(.5)(.5)
Spread: ______  .0449
124
c. Yes, the histogram does appear to be consistent with what the CLT predicts. It is
bell-shaped, centered at about .5, and extends from about .5  3(.0449) or .3653
to about .5  3(.0449) or .6347.
d. p̂  80/124  .645
e. Yes, it would be very surprising to observe such a sample proportion (.645) if 1/2
of all kissing couples lean their heads to the right; this sample proportion never
occurred in 1000 simulations.
f. The z-score for the observed sample proportion is z  (.645  .5)/.0449  3.23.
g. Yes, this is a very surprising z-score; Pr(Z  2.33)  .0099. If 1/2 of all kissing
couples lean their heads to the right, you would see a sample result as or more
extreme than .645 in less than 1% of random samples.

Activity 13-4: Kissing Couples

a. Recall that the observed sample proportion of kissing couples who lean their
heads to the right is p̂  80/124  .645. This value is not at all uncommon in
the first histogram.
b. The CLT says that the sample proportion in this case would vary approximately
normally with mean equal to .667 and standard deviation equal to
__________


(.667)(.333)
__________  .042
124
 Activity 13-5 253

The z-score for the observed sample proportion of .645 is therefore

.645  .667  0.52
z  __________
.042
so the observed sample proportion .645 lies only about half of a standard deviation
from the population proportion when ␲  .667.
c. The observed sample proportion is barely one-half of a standard deviation away
from what you would expect if the population proportion were equal to 2/3, not

13
a surprising result at all. Therefore, the sample data provide no reason to doubt
that the population proportion of kissing couples who lean their heads to the
right equals 2/3.
d. The value .645 is pretty far along the lower tail of the second histogram.
This indicates that the observed sample proportion would rarely occur if the
population proportion were equal to 3/4. Further evidence of this result is
provided by the rather large negative z-score:
.645
z  ____________
__________ .645  .750  2.69
 .750  __________
.39

(.750)(.250)
__________
124
Therefore, the sample data provide fairly strong evidence that the population
proportion of kissing couples who lean their heads to the right is not 3/4 (because
it would be rather surprising to find a sample proportion so far from this
population proportion by chance alone).
e. A reasonable estimate of the population proportion ␲ is the sample proportion
.645. An estimate of the standard deviation of p̂ would then be
__________


(.645)(.355)
__________  .043
124
Doubling this standard deviation gives .086. The interval is, therefore, .645 
.086, which runs from .559 to .731. Notice that 1/2 and 3/4 are not in this
interval, but 2/3 is. The interval is consistent with the earlier analysis of the
plausibility of the values 1/2, 2/3, and 3/4 for the population proportion of kissing
couples who lean their heads to the right.

••• Homework Activities

Activity 13-5: Parameters vs. Statistics
a. Statistic p̂ i. Statistic p̂
b. Parameter ␲ j. Parameter μ
_
c. Statistic x k. Statistic p̂
_
d. Parameter ␮ l. Statistic x (population
is all American
e. Parameter ␴
households)
f. Statistic s
m. Parameter ␮
g. Parameter ␲ (population
n. Statistic p̂
is all voters)
_
o. Statistic x
h. Statistic p̂
 Activity 14-1 269

4. To calculate the mean of this sample, choose the Measures panel in the inspector
for the sample collection and define a new measure called XBar with the formula
mean(Change _ Amount).
5. To save the sample means for repeated samples, select the sample collection box
and choose Collection | Collect Measures from the top menu. This creates
a third collection with, by default, the sample means (XBar) for five different
samples. Drag a new table from the shelf to view the contents of this collection.
6. Double-click on the new collection to show its inspector, and on the Collect
Measures panel, turn off animation by unchecking the Animation on box, check
Replace existing cases, and change the number of measures to collect from 5 to
500. Press the Collect More Measures button to perform the 500 simulations.

14
7. Use the Xbar attribute in the Measure from Sample of Change collection to
create a dotplot, normal quantile plot, and summary statistics to describe the
distribution of the sample means.

Activity 14-2: Coin Ages (Graphing Calculator)

As a graphing calculator alternative, use the CHANGE program, which simulates
the taking of 500 random samples of size n  30 from a population representing
hypothetical amounts of change (in cents) given out by a store. Consider pairing
students so that one student can use her calculator for the simulation and the other
student can use his calculator for the activity.

Solutions
••• In-Class Activities
Activity 14-1: Coin Ages
a. Observational units: pennies Variable: age
Quantitative or categorical? quantitative
b. These values are parameters, represented by symbols ␮ (mean) and ␴ (standard
deviation).
c. No, the distribution of ages does not follow a normal distribution; it is strongly
skewed to the right.
d. Answers will vary. One example (using row 48 of the Random Digits Table, three
digits at a time) is coins numbered 788, 929, 977, 718, 049, which have ages 20,
28, 35, 17, and 2, respectively. Note that we would delete any three-digit numbers
that correspond to a coin that has already been selected (sampling without
replacement). You are also free to read the table vertically, backward, selecting a
new row, etc. In order for someone to evaluate your procedure though, include
enough details so your method is clear.

0 5 10 15 20 25 30 35 40 45 50 55 60
Penny Age

e. The sample mean of your five penny ages is 20.4 years.

270 Topic 14: Sampling Distributions: Means

f. The following table gives the sample mean age for each sample:

1 2 3 4 5
_
Sample Mean (x ) 20.4 18.8 1.2 18 10.6

g. No, you do not get the same value for the sample mean all five times. This reveals
the phenomenon of sampling variability. This variable is quantitative, whereas the
variable in the Candy Colors activity (Activities 13-1 and 13-2) was binary categorical.
__ __
h. Mean of x values: 13.8 years SD of x values: 7.99 years
__
i. The mean of x values is reasonably close to the population mean (␮  12.264 years).
The standard deviation is less than the population standard deviation of 9.613 years.
j. The following dotplot displays the sample means:

10.5 11.0 11.5 12.0 12.5 13.0 13.5

Mean Age of Pennies (Sample Size of 25)

The observational units are the samples of 25 coins.

k. Yes, this distribution should be roughly centered at 12.264 years.
l. These values appear to be less spread out than either the population or the
samples of five pennies.
m. Yes, this distribution does appear to be closer to a normal shape than the
distribution of ages in the original population, which was so strongly skewed to
the right.

Activity 14-2: Coin Ages

Answers will vary. The following are examples from one running of the applet.
a. A rough sketch of the results displayed in the dotplot follow:

Total Samples 500

Average Age Estimate

0 4 8 12 16 20 24 28
Mean = 13.05 SD = 9.95
 Activity 14-2 271

Yes, this dotplot does resemble the distribution of ages in the population; it is
definitely skewed to the right. Note: Any observations to the right of the applet
window are piled in the last bin.
b. The mean is 13.05 years and the standard deviation is 9.95 years. The shape
is somewhat skewed to the right.
c. A sketch of the results displayed in the dotplot follow (500 sample means
with n  5):

Total Samples 500

Average Age Estimate

14
0 4 8 12 16 20 24 28
SD = 4.21 Mean = 12.32 x–

Now the shape of the distribution is much more symmetric and less skewed. The
center is 12.32 years and the standard deviation is 4.2 years.
d. The mean of the sampling distribution is very close to the population mean, but
the standard deviation of the sampling distribution is much smaller.
e. A sketch of the results displayed in the dotplot follow (500 sample means
with n  25):

Average Age Estimate

0 4 8 12 16 20 24 28
Mean=12.27 SD=1.94

The shape is even more symmetric, and the spread is much smaller. (The center
has remained the same.)
f. The mean of the sampling distribution is very close to the population mean, but
the standard deviation of the sampling distribution is much smaller.
272 Topic 14: Sampling Distributions: Means

g. A sketch of the results displayed in the dotplot follow (500 sample means
with n  50):

Average Age Estimate

0 4 8 12 16 20 24 28
Mean=12.21 SD=1.29

The shape is even more symmetric, and the spread is even smaller, but the center
has remained at about 12.2 years.
h. Once again the mean of the sampling distribution is very close to the population
mean, but the standard deviation of the sampling distribution is much smaller.

Population Mean Population SD Population Shape:

(␮  12.264 years) (␴  9.613 years) Skewed to the Right

Sample Size Mean of Sample Means SD of Sample Means Shape of Sample Means

1 13.05 9.95 Skewed to right

5 12.32 4.21 Symmetric

25 12.27 1.94 Normal

50 12.21 1.29 Normal

i. No, this population follows a uniform distribution (from $0.01 to $0.99).

j. The following histogram displays the results of 500 sample means with n  30:

80
70
60
Frequency

50
40
30
20
10
0
0.00 0.14 0.28 0.42 0.56 0.70 0.84 0.98
Sample Mean

The histogram of sample means is fairly bell-shaped (approximately normal). The

mean is $0.50 and the standard deviation is $0.0517.
 Activity 14-3 273

k. Yes, this probability plot indicates the distribution is definitely well modeled by a
normal curve:

99.9

99

95
90
80
70
Percentage 60
50
40
30

14
20
10
5

0.1
0.3 0.4 0.5 0.6 0.7
Sample Mean

__ ___
l. You calculate ␴/n  9.613/50  1.359 years. This value is reasonably close
to the standard deviation of the 500 simulated sample means (1.29).
__ ___
m. You calculate ␴/n  28.866/30  4.08¢. This is reasonably close to the
simulated standard deviation of 5¢.

Activity 14-3: Christmas Shopping

_
a. The value $857 is a statistic, represented with the symbol x. This value is a
statistic because it comes from a sample, not from the entire population of
American adults.
b. i. Population of interest: all American adults
ii. Sample selected: 922 American adults who expected to buy Christmas gifts
in 1999
iii. Parameter of interest: average amount all American adults expected to spent on
Christmas gifts in 1999
iv. Statistic calculated: $857, or the average amount this sample of 922 adults
expected to spend on Christmas gifts
_
c. No, ␮ does not necessarily equal $857. It is possible that x  $857, even if ␮ 
$850, or $800, or $1000. But such a sample mean is not very likely for ␮  $800,
and it is extremely unlikely for ␮  $1000.
__
d. The CLT says that that sampling distribution of x would be approximately
____
normal, with mean $850 and standard deviation $250/ 922  $8.23. Because
the sample size is large, your answer does not depend on the shape of the
distribution of expected expenditures in the population.
274 Topic 14: Sampling Distributions: Means

e. Here is a sketch for this distribution:

820 830 840 850 860 870 880

Sample Mean Expected Christmas Expenditures (in dollars)

f. No, a sample mean of $857 would not be at all surprising as it is located close to
the center of the distribution and there is a great deal of area to the right of this
value in the plot shown in part e.
_
g. The CLT says that that sampling distribution of x would be approximately
____
normal with mean $800 and standard deviation $250/ 922  $8.23. Because
the sample size is large, your answer does not depend on the shape of the
distribution of expected expenditures in the population. Here is a sketch for this
distribution:

770 780 790 800 810 820 830

Sample Mean Expected Christmas Expenditures (in dollars)

This time a sample mean of $857 would be very surprising because this value does
not appear on the normal curve modeling the behavior of the sample means.
____
h. The standard deviation should be $250/ 922  $8.23.
 Activity 14-4 275

i. You calculate 2  $8.23  $16.46, so $857  $16.46  $840.54 and

$857  $16.46  $873.46. The interval of values is [$840.54, $873.46].
Note: These could be considered the plausible values of ␮ based on observing
_
a sample mean of x  857.

Activity 14-4: Looking Up to CEOs

a. The mean height among all adult American males, 69 inches, is a parameter
because it describes the entire population. It is denoted by the symbol ␮.
b. The sample size is 100 and is denoted by the symbol n.
c. You need to know the population standard deviation of the heights of adult
American males, denoted by ␴. Because the sample size is fairly large (100 is

14
larger than 30), you do not need to know whether the population distribution
of heights is normal because the Central Limit Theorem tells you the shape of
the sampling distribution of the sample mean will be approximately normal
with this sample size, regardless of the shape of the original population
distribution.
d. The CLT establishes that the sampling distribution of the sample mean height
____ is
approximately normal, with mean 69 inches and standard deviation 3/ 100 
0.3 inches.

0.3

68.0 68.5 69.0 69.5 70.0

Average Height in Sample (in inches)

e. Doubling the standard deviation of the sample mean gives you 0.6 inches, so
the sample mean height among the CEOs would have to be at least 69.6 inches
to persuade the psychologist that, on average, CEOs are indeed taller than the
average adult male.
f. If the sample size were 30, the normal approximation should still be valid, and
___
the standard deviation of the sampling distribution would increase to 3/ 30 
0.548 inches. Doubling this standard deviation gives you 1.096 inches, so the
sample mean height would have to be at least 70.096 inches to be persuasive. The
smaller sample size produces more sampling variability, hence a larger standard
deviation. As a result, the cut-off value needed for a persuasive sample mean
height is larger.
276 Topic 14: Sampling Distributions: Means

n = 100

0.548

n = 30

67 68 69 70 71
Average Height in Sample (in inches)

••• Homework Activities

Activity 14-5: Heart Rates
a. Yes, this distribution appears approximately normal, although it is a little skewed
to the right.
b. The probability plot indicates that this data is not entirely normally distributed;
there is too much data in the right tail and not enough in the left, but it is fairly
close to normal.
c. The following graph displays the distribution of the 1000 sample means
with n  3:

100

80
Frequency

60

40

20

0
80 90 100 110 120 130 140 150
Sample Mean (in bpm)

This distribution appears normally distributed with mean 108.19 bpm and
standard deviation 11.58 bpm.
d. Yes, the distribution appears roughly normal in spite of the small sample size.
This is because the population itself was so close to being normally distributed.
288 Topic 15: Central Limit Theorem

good example of thinking through the effect of increasing sample size and the danger
of automatically thinking that a larger sample size produces a smaller probability.
Instead, you can point out that if an interval includes the population parameter, then
larger samples produce larger probabilities. If an interval is in one of the tails of the
distribution, larger samples produce smaller probabilities. For other kinds of intervals,
the issue is less clear.
Possible assignment: Activity 15-6 reminds students to identify the type of variable;
Activity 15-9 reviews proportions and statistical significance; and Activity 15-11 works
with means and statistical confidence.

Solutions
••• In-Class Activities
Activity 15-1: Smoking Rates
a. The symbol ␲ represents the proportion .209.
b. No; in general, the sample result will not equal .209 exactly because of sampling
variability.
c. The CLT predicts the sampling distribution of p̂ will be approximately normal,
centered at .209 with a standard deviation equal to
__________
(.209)(.791)
√
__________
100
 .04066.

d. The following graph shows the shaded area that corresponds to a sample
proportion exceeding .25:

.05 .10 .15 .20 .25 .30 .35 .40

Sample Proportion Who Smoke

e. Using the CLT result, z  (.25  .209)/.04066  1.01.

f. Pr(Z  1.01)  .1562 (Table II) or .1566 (applet)
g. When the sample size increases to 400, the standard deviation of the sampling
distribution will decrease to .0203. This means there will be fewer sample
proportions as far from the center of .209, and it will be less likely that you will
 Activity 15-2 289

have a sample proportion greater than .25. The following sketch displays these
results:

.150 .175 .200 .225 .250 .275

15
Sample Proportion of Smokers (n = 400)

h. You calculate z  (.25 .209)/.0203  2.02; Pr(Z  2.02)  .0217. Yes, this
probability has decreased as predicted.
i. No, the population size of the United States did not enter into the calculations.
j. The previous calculations would not change in any way.

Activity 15-2: Smoking Rates

a. You have N(.105, .0307). The following sketch displays these results:

.00 .05 .10 .15 .20

Sample Proportion of Smokers in Utah (n = 100)

b. You calculate z  (.25  .105)/.0307  4.72; Pr(Z  4.72)  0.000.

c. Yes, you would have strong reason to doubt that the state was Utah, because the
probability of finding a random sample of 100 people from Utah containing
25 smokers is essentially zero—this never happens by chance alone. So if you find
a random sample of 25/100 smokers, you have very strong evidence the sample is
from some other state where the proportion of smokers is greater than 10.5%.
290 Topic 15: Central Limit Theorem

Activity 15-3: Body Temperatures

a. These numbers are parameters: ␮  98.6°F, ␴  0.7°F.
b. Yes, the sample size is greater than 30 and you have a simple random sample, so
the CLT applies.
c. The CLT says the sampling distribution of the sample means will
____be approximately
normal with a mean of 98.6°F and a standard deviation of .7/√130  .061.

98.4 98.5 98.6 98.7 98.8

Sample Average Body Temperature (in °F)

d. Pr(98.5  sample mean  98.7)  Pr(1.64  Z  1.64)  .9495  .0505 

.8990 (Table II) or .8989 (applet)
e. Pr(98.2  sample mean  98.4)  Pr(1.64  Z  1.64)  .9495  .0505 
.8990 (Table II) or .8989 (applet)
f. These answers are the same; you have simply shifted the center of the plot, but the
area within 0.1 degrees of the center has not changed.
g. Pr(0.1/.061  Z  0.1/.061)  Pr(1.64  Z  1.64)  .9495  .0505 
.8990 (Table II) or .8989 (applet)
So, there is about a 90% chance that a random sample of 130 adults will result
in a sample mean body temperature that is within .1 degrees of the actual
population mean ␮ if you assume the population standard deviation is ␴  0.7°F.

Activity 15-4: Solitaire

a. The CLT says the sampling distribution will be approximately N(.1111, .0994).
You calculate z  (.1  .1111)/.0994  .11. So Pr(Z  .11)  .4562 (Table II)
or .4562 (applet).
b. Pr( p̂  .10)  .308  .385  .693
c. No, the probabilities in parts a and b are not close.
d. The technical conditions for the CLT are not satisfied. You calculate n␲  10 
(1/9)  1.111 10 and n(1  ␲)  10  (8/9)  8.888 10.
 Activity 15-7 291

Activity 15-5: Capsized Tour Boat

First, weight is a quantitative variable, so the relevant statistic is the sample mean weight of
the 47 passengers. Because the question is phrased in terms of the total weight in a sample
of 47 adults, you must rephrase it in terms of the sample mean weight. If total weight
exceeds 7500 pounds, then the sample mean weight must exceed 7500/47 or 159.574
_
pounds. So, you want to find the probability that x  159.574 (with n  47 and ␴  35).
The CLT applies because the sample size (n  47) is fairly large, greater than 30.
_
The sampling distribution of x is, therefore, approximately ___normal with mean 167
__
pounds and standard deviation equal to ␴/ n  35/ 47  5.105 pounds. A sketch
of this sampling distribution is shown here:

15
5.105

150 155 160 165 170 175 180 185

167
Average Weight in Sample (in pounds)

Now you can use the Normal Probability Calculator applet or the
Standard Normal Probabilities Table to find the probability of interest. The z-score
corresponding to a sample mean weight of 159.574 pounds is (159.574  167)/5.105 
1.45. The probability of the weight being less than 159.574 pounds is found from the
table to be .0736, so the probability of exceeding this weight is 1  .0736  .9264. It’s
not surprising the boat capsized with 47 passengers!

••• Homework Activities

Activity 15-6: Means or Proportions
a. Sample mean
b. Sample proportion
c. Sample mean
d. Sample mean
e. Sample proportion

Activity 15-7: Candy Colors

a. The CLT says the sample proportions will be approximately
_______ normally distributed,
______

with a mean of .45 and a standard deviation of (.45)(.55)  .568.
75
 Activity 16-1 317

Technology Ideas
Activity 16-3: Generation M (Fathom)
Part n of this activity asks students to compute 95% confidence intervals for
proportions based on sample proportions of boys and girls who have a television in
their bedrooms. This calculation may be done with an estimate box in Fathom to
allow students to quickly generate the intervals needed for this comparison, using just
the summary counts given for each sample. To get the actual counts for each gender,
students need to multiply the sample proportion given by the sample size and round to
the nearest integer, so that 0.72 * 996 or 718 boys and 0.64 * 1036 or 663 girls in the
samples have a television in their bedrooms.
1. Starting with a new Fathom document (no need to enter any data), drag an empty
estimate box from the shelf.
2. Click on Empty Estimate to show a menu of types of confidence intervals and
choose Estimate Proportion to set up the confidence interval.
3. Any of the blue values in the estimate box may be edited to reflect the actual
sample values, so change “10 out of 20” to 718 out of 996 for the sample of boys
to produce the desired 95% confidence interval.

16
4. Choose Estimate | Verbose from the main Fathom menu to turn on and off
concise vs. verbose versions of the results.
5. Edit the blue values in the estimate box to find an interval based on the 663 out
of 1036 girls in the sample.

Activity 16-5: Elvis Presley and Alf Landon

(Graphing Calculator)
For part a of this activity students can use their calculator’s 1-PropZInt command to
form a 99.9% confidence interval for the actual proportion of all adult Americans who
preferred Landon over Roosevelt.
1. Press the Ö key.
2. From the TESTS menu, arrow down to select 1-PropZInt.
3. Enter 57% of 2.4 million for x (multiply 2.4 million by .57), 2.4 million for n,
and .999 for C-Level. Arrow down to select Calculate.
4. Press Õ.

Solutions
••• In-Class Activities
Activity 16-1: Generation M
a. The observational units are American youth ages 8–18.
b. The variable is whether each youth has a television in his or her bedroom (binary
categorical).
c. The value .68 is a statistic, which can be denoted by the symbol p̂.
318 Topic 16: Confidence Intervals: Proportions

d. The relevant parameter of interest is the population proportion of all American

youth ages 8–18 who have a television in their bedrooms, which can be denoted
by the symbol ␲.
e. The Kaiser survey does not allow the researchers to determine the exact value of
the parameter because they did not survey all American youth in the population.
f. The parameter value is more likely to be close to the survey’s sample proportion
than to be far from it, because the surveyed sample was randomly selected. You
expect some sampling variability, but with a large sample size like this, you expect
the sample proportion to be reasonably close to the true parameter.
g. The value p̂ (.68) seems like a reasonable replacement to use as an estimate for ␲.
_________
(.68)(.32)
h. The standard error of p̂ is
√ ________
2032
 .01035.

i. You calculate .68  2(.01035)  .68  .0207  (.6593, .7007). Consider the
value of the population parameter, ␲, to be somewhere in this interval.
j. You do not know for sure whether the actual value of ␲ is contained in this
interval (as in part e).

Activity 16-2: Critical Values

a. Here is the sketch:

.9800

.01 .01

–2.33 0 2.33
–z* z-values z*

b. The total area to the left of z* is .9900.

c. z*  2.33
 Activity 16-3 319

d. Here is the sketch:

.9500

.025 .025

–1.96 0 1.96
–z* z-values z*

The total area to the left of z* is .975.

16
z*  1.96

Activity 16-3: Generation M

____________
a. For a 95% confidence interval, you calculate √ .68(.32)2032  .01035. So
.68  1.96(.01035)  .68  .020286  (.6597, .7003).
b. You are 95% confident the population proportion of all American 8–18-year-olds
who have a television in their bedrooms is between .66 and 70.
c. No, you cannot be certain that this interval contains the actual value of ␲.
d. Width  .7003  .6587  .0406
e. Half-width  .04062  .0203
________
p̂(1  p̂)
f. This half-width is also z *
margin-of-error.
√ _______
n or 1.96  (.01035), which is the

g. Midpoint  (.7003  .6597)2  .68

h. Yes, this value looks familiar. This value is p̂. This value makes sense because
the interval is created extending an equal distance above and below the observed
sample proportion p̂.
i. Technical conditions: (i) The sample is a simple random sample from the
population of interest (probably more complicated than a “simple random
sample,” but still random), and (ii) the sample size is large relative to ␲:
(2032)(.68)  1382  10 and 2032(.32)  650  10.
j. Answers will vary based on student intuition, but students should expect the
interval to be wider; if they specify more values, they will be more confident that
the actual value is in the specified range.
k. You calculate .68  2.576(.01035)  .68  .0267  (.653, .707).
320 Topic 16: Confidence Intervals: Proportions

l. The midpoints of both intervals are the same (.68), but the 99% confidence
interval is wider; it has a greater margin-of-error (.267) vs. (.203).
m. .50: not plausible .75: not plausible Two-thirds: plausible
Explanation: Both .50 and .75 do not seem to be plausible values for ␲ as they
are not contained in either the 95% or 99% confidence intervals. However,
.67 is contained in both confidence intervals, so it seems to be a plausible
value for ␲.
___________
n. Boys: .72  (1.96) √ .72(.28)996  .72  (1.96)(.0142)  (.692, .748)
____________
Girls: .64  (1.96) √ .64(.36)1036  .64  (1.96)(.0149)  (.611, .669)
o. These intervals do seem to indicate that there is a difference in the population
proportion of boys and girls who have a television in their bedrooms. You are 95%
confident the population proportion of boys with a television in their bedrooms
is at least .69 (and no more than .75), whereas you are 95% confident that the
proportion of girls with a television in their bedrooms is between .61 and .67.
There is no overlap in these intervals; the values of ␲ that are plausible values for
boys are not plausible values for girls.
p. The margins-of-error for these intervals are .028 (boys) and .029 (girls). These are
greater than the margin-of-error based on the entire sample, which makes sense
because the entire sample is roughly twice as large as the single gender samples,
so you would expect it to have less sampling variability and therefore a smaller
margin-of-error.
q. You set the margin-of-error formula equal to .01 and then solve for the sample
size n, as follows:
_________
(.68)(.32) 2
1.96  8359.322; n  8,360
n  (.68)(.32)  ____
√ ________ .01 
.01  1.96 n

r. Answers will vary by student expectation, but the required sample size will
increase.
s. To determine the sample size, you calculate
_________
(.68)(.32) 2
2.576  14439.448; n  14,440
n  (.68)(.32)  _____
√ ________ .01 
.01  2.576 n

Activity 16-4: Candy Colors

Answers will vary. Here is one representative set of answers.
a. Endpoints: .315, .539.
Yes, this interval succeeds in capturing ␲  .45.
b. Yes, the interval changed each time. No, the intervals did not all succeed in
capturing the true value of ␲ (4 out of 5 did).
c. In this simulation, the proportion of intervals that succeed in capturing ␲ is
190200  95% (you expect something close to 95% but it may not match
exactly with 200 samples).
 Activity 16-4 321

d. The intervals that fail to capture ␲ have midpoints that are fairly far away (more
than 2 standard deviations) from .45 (either on the low end of the scale or the
high end).
e. No; if you had taken a single sample in a real situation, you would have no way of
knowing whether the true value of ␲ was contained in your interval because you
would not know what ␲ was; you are not guaranteed that the constructed interval
will capture the value of ␲.
f. In the following simulation, the proportion of intervals that succeed in capturing
␲ is 9521000  95.2%:

16
g. Yes, this percentage is close to 95%. Yes, it should be close to 95% because these
are 95% confidence intervals; the procedure should be “successful” 95% of the
time. This simulation reveals that the phrase “95% confidence” indicates that
your method of creating confidence intervals is successful in capturing the true
population parameter (␲) 95% of the time and that it fails 5% of the time in the
long run (over many, many intervals).
h. Answers will vary by student prediction, but the intervals will become less wide
(more narrow) as the sample size increases.
322 Topic 16: Confidence Intervals: Proportions

i. In the following simulation, the proportion of intervals that succeed in capturing

␲ is 9421000  94.2%:

This is reasonably close to the percentage from part f. The noticeable difference
about these intervals is that they are not as wide as those that were generated with
samples of 75 candies. [Example interval: (.341, .499).]
j. Answers will vary by student prediction, but students should predict that the
length of the intervals will shorten and the success rate will decrease when the
confidence level is changed to 90%.
k. In the following simulation, the proportion of intervals that succeed in capturing
␲ is 9021000  90.2%.
 Activity 16-6 323

16
This is not particularly close to the percentages from parts f and i, but it shouldn’t
be because you changed the confidence level. The noticeable differences about
these intervals are that they are not as wide as the previous intervals [example
interval: (.334, .466)], nor are they as successful in capturing ␲.

Activity 16-5: Elvis Presley and Alf Landon

_______________
a. For a 99.9% confidence interval, you calculate .57  (3.291) √ .57(.43)2400000 
.57  (3.22291)(.00032)  (.569, .571).
b. This interval is so narrow because the sample size is so very, very large.
c. The confidence interval did such a poor job predicting the election results because
the necessary technical conditions were not satisfied. In particular, the sample was
not randomly selected; it was chosen in a very biased fashion and therefore vastly
overestimated the support for the Republican candidate.

Activity 16-6: Kissing Couples

a. The observational units are the kissing couples. The variable is which direction the
couples lean their heads while kissing.
b. The sample consists of the 124 kissing couples observed by the researchers in
various public places. The statistic is the sample proportion of couples who lean to
the right when kissing:
p̂  80124  .645
324 Topic 16: Confidence Intervals: Proportions

A bar graph is shown here:

90
80

Number of Couples
70
60
50
40
30
20
10
0
Right Left
Direction that Kissing
Couple Leans

c. A 95% confidence interval for the population proportion of all couples who lean
to the right is
_______________
(.645)(1  .645)
.645  1.96
√ ______________
124
which is .645  .084, or (.561, .729). You are 95% confident that the population
proportion of all kissing couples who lean to the right is somewhere between
.561 and .729. This “95% confidence” means that if you were to take many
random samples and generate a 95% confidence interval (CI) from each, then
in the long run, 95% of the resulting intervals would succeed in capturing the
actual value of the population proportion, in this case the proportion of all kissing
couples who lean their heads to the right.
d. A 90% CI is
_______________
(.645)(1  .645)
.645  1.645
√ ______________
124
which is .645  .071, or (.574, .716).
A 99% CI is
_______________
(.645)(1  .645)
.645  2.576
√ ______________
124
which is .645  .111, or (.534, .756).
The higher confidence level produces a wider confidence interval. All of these
intervals have the same midpoint: the sample proportion .645.
e. Because none of these intervals includes the value .5, it does not appear to be
plausible that 50% of all kissing couples lean to the right. In fact, all of the
intervals lie entirely above .5, so the data suggest that more than half of all kissing
couples lean to the right. The value 23 is quite plausible for this population
proportion because .667 falls within all three confidence intervals. The value
34 is not very plausible because only the 99% CI includes the value .75; the 90%
CI and 95% CI do not include .75 as a plausible value.
f. The sample size condition is clearly met, as np̂  80 is greater than 10, and
n(1  p̂)  44 is also greater than 10. But the other condition is that the sample
be randomly drawn from the population of all kissing couples. In this study, the
couples selected for the sample were those who happened to be observed in public
 Activity 16-9 325

places while the researchers were watching. Technically, this is not a random
sample, and so you should be cautious about generalizing the results of the
confidence intervals to a larger population.

••• Homework Activities

Activity 16-7: Generation M
_________
(.68)(.32)
a. TV margin-of-error: (1.96)
√ ________
2032
 .0203
_________
(.86)(.14)
CD/tape player margin-of-error: (1.96)
√ ________
2032
_________
 .0151

(.49)(.51)
Video game player margin-of-error: (1.96)
_________
√ ________
2032
 .0217

(.31)(.69)
Computer margin-of-error: (1.96)
√ ________
2032
 .0201

b. Yes, the margin-of-error also depends on the confidence level and on the sample
proportion. You can tell this from the above results because all four devices have
the same sample size and confidence level, but the sample proportions differ, as do

16
the margins-of-error. The more similar the sample proportions (similar distance
from .5), the more similar the margins-of-error, all other factors being equal.
c. The video game player produces the greatest margin-of-error and the CDtape
player produces the smallest.
d. Answers will vary by student conjecture. The sample proportion that produces the
greatest margin-of-error is p̂  .5.

Activity 16-8: Cursive Writing

a. The number of essays needed for a 99% CI is
_________
(.15)(.85) 2
2.576  8460.614; n  8461
n  (.15)(.85)  _____
√ ________ .01 
.01  2.576 n
b. You could use a lower confidence level (95% or 90% confidence, for example),
or you could use a wider margin-of-error, say .02. Either of these choices would
allow you to select a smaller (random) sample.

Activity 16-9: Penny Activities

Results will vary. The following are meant to be representative results:
a. A 95% confidence interval is
__________
2650  (1.96) √ .52(.48)50  .52  (1.96)(.07065)
 .52  .1385  (.382, .658)
You are 95% confident that the probability a flipped penny lands heads is
between .382 and .658.
b. You calculate 2750  (1.96)(.07048)  .54  .1381  (.402, .678). You are 95%
confident that the probability a spun penny lands heads is between .407 and .678.
c. You calculate 3250  (1.96)(.0679)  .64  .1330  (.507, .773). You are 95%
confident that the probability a tilted penny lands heads is between .507 and .773.
340 Topic 17: Tests of Significance: Proportions

Technology Ideas
Activity 17-2: Flat Tires (Fathom)
As an alternative to using the Test of Significance Calculator applet
in part b of this activity, students can calculate the details of the test of a proportion,
using a test box in Fathom. The procedure is very similar to constructing a confidence
interval with an estimate box based on the summary counts, as described in the
previous topic.
1. Drag down an empty test box from the shelf, and select the Empty Test label and
change it to Test Proportion.
2. As with the estimate box, edit the blue values to change “10 out of 20” to 24 out
of 74 to reflect those students who chose the right-front tire in the sample.
3. In the sentence or expression that describes the alternative hypothesis, change the
default “is not equal to” to is greater than and change the hypothesized value
from “.5” to .25.
4. The verbose version of the test output shows the p-value at the end of a
long sentence that carefully describes what the p-value measures. Choosing
Test | Verbose to remove the check mark gives more concise output.
5. To show a graphical display of the p-value as an area in the tail of a normal
distribution approximating the distribution of sample proportions when the
null hypothesis is true, choose Test | Show p_hat Distribution.

Activity 17-2: Flat Tires (Graphing Calculator)

Suppose this sample result (30% answering “right front”) had come from a sample of
n  50 people. In part b of this activity, students can use their calculators to conduct
the appropriate test of significance by using the 1-PropZTest command.
1. The 1-PropZTest can be found by pressing Ö and then selecting the TESTS
menu. Your TI-83 Plus or TI-84 Plus calculator uses p0 instead of ␲0.
2. Enter 0.3*50  15 for x and 50 for n.
3. Select Calculate and press Õ.
4. As an alternative, you may want to explore the Draw feature.

Solutions
••• In-Class Activities
Activity 17-1: Flat Tires
a. 14 or .25
b. This value is a parameter because it represents the overall process, not simply
results that you observed, and is represented by ␲.
c. ␲  .25
d. The sampling distribution of the sample proportion will be______________
approximately normal,
with mean equal to .25, and standard deviation equal to √ .25(1  .25)73 
.0507. Here is a sketch of the sampling distribution:
 Activity 17-1 341

.10 .15 .20 .25 .30 .35 .40

Sample Proportion of Right-Front Tires

e. The conditions necessary for the CLT to be valid are that n␲ 10 and
n(1  ␲) 10. These conditions are met (73(.25)  10 and 73(.75)  10).
However, you also need to believe that the sample is representative of the
larger population process. This is less clear, but there may not be any reason to
believe that this professor’s class would behave substantially differently on this
issue than college students in general, or you may want to think more carefully
about how you define the population in this activity (e.g., students of similar
age and major).

17
34  .466. Yes, this
f. To determine the sample proportion, you calculate p̂  __
73
sample proportion is greater than 14.
g. The following graph displays the shaded area:

Mean  .25, SD  .05068

.0000101
.25 .466
Sample Proportion of Right-Front Tires

For the z-score, you calculate

.466  .25  4.26
z  _________
.0507
p-value  Pr(Z  4.26)  0.0002
342 Topic 17: Tests of Significance: Proportions

h. This sample result would be surprising if there were nothing special about the
right-front tire. With a p-value of approximately zero, you would consider this
sample result so surprising that you would conclude the right-front tire would be
chosen by more than one-fourth of the population.
i. Let ␲ represent the proportion of the population who will select the right-front
tire when asked this “flat tire” question.
j. H0: ␲  .25
k. Ha: ␲  .25
.466  .25
l. The test statistic is z  _______________
______________ .466  .5  4.26.
 ________
√ .25(1  .25)73 .0507

m. The p-value  .0002.

n. Yes, this probability suggests that it is very unlikely for 34 or more of 73 randomly
selected students to choose the right-front tire if one-fourth of the population
would choose the right-front tire. Such a sample result would occur in less than
1% of random samples if ␲  .25.
o. Calculate 73(.25)  18.25  10 and 73(.75)  54.75  10, so this condition
is met. You are not certain this is a simple random sample from the population
of interest, but it is likely to be a representative sample of introductory statistics
students.
p. You have found very strong statistical evidence that introductory statistics students
tend to choose the right-front tire more than one-fourth of the time.

Activity 17-2: Flat Tires

a. Define parameter of interest: Let ␲ represent the proportion of all introductory
statistics students who will select the right-front tire when asked this “flat tire”
question.
H0: ␲  .25
Ha: ␲  .25
Check technical conditions: 74(.25)  18.5  10 and 74(.75)  55.5  10, so
this condition is met. You are not certain this is a simple random sample from the
population of interest, but it is likely to be a representative sample of introductory
statistics students.
.324  .25  1.47
Test statistic: z  __________
_________
(.25)(.75)
√
________
73
p-value  Pr(Z  1.47)  .0708
Test decision: At ␣  .05 level, do not reject H0 ( p-value  .0708  .05).
 Activity 17-3 343

Conclusion in context: You do not have sufficient statistical evidence at the 5%

level (though you do at the 10% level) to conclude that the proportion of statistics
students who will choose the right-front tire is greater than one-fourth. You will
continue to believe that the right-front tire is no more likely to be chosen than any
other tire.
b. The following graph confirms calculation of the test statistic and p-value:

Activity 17-3: Racquet Spinning

17
a. Population parameter of interest: Let ␲ represent the proportion of times a spun
tennis racquet will land “up.”
b. H0: ␲  .5 (A spun tennis racquet will land “up” half the time.)
H a: ␲ .5 (A spun tennis racquet will not land “up” half the time.)
c. Technical conditions: As long as there was nothing unusual about the spinning
process, you will consider these data a “random” sample, and 100(.5) 
100(1  .5)  50  10, so the technical conditions for the validity of this test
procedure are satisfied.
.46  .5  0.80.
d. The test statistic is z  ________
_______
(.5)(.5)
√ ______
100
e. p-value  2  Pr(Z  0.80)  2  .2119  .4238
344 Topic 17: Tests of Significance: Proportions

Here is a sketch of the standard normal curve with the p-value shaded:

Mean = .5, SD = .05

.212 .212

.46 .5 .54
Sample Proportion of Spins Landing “Up”

f. This p-value is very large (.4238  .05). Therefore, do not reject the null
hypothesis at the .05 significance level.
g. Conclusion in context: You do not have statistical evidence that would allow you
to conclude that a spun tennis racquet will fail to land “up” half the time.

Activity 17-4: Flat Tires

a. You need more information in order to decide whether this constitutes strong
evidence that the right-front tire would be chosen more than one-quarter of the
time in the long run. You need to know the number of people surveyed (the
sample size).
b. See values in table following part c.
c. Here is the completed table:

Sample # “right-
Size front”
p̂ z statistic p-value ␣  .10? ␣  .05? ␣  .01? ␣  .001?

50 15 .30 0.82 .207 No No No No

100 30 .30 1.15 .124 No No No No

150 45 .30 1.41 .079 Yes No No No

250 75 .30 1.83 .034 Yes Yes No No

500 150 .30 2.58 .005 Yes Yes Yes No

1000 300 .30 3.65 .000 Yes Yes Yes Yes

d. When the sample size is small, a sample result of .30 is not statistically
significant at any level. But, as the sample size increases, this result becomes
more significant—meaning that it becomes more unlikely that you would obtain
a sample result of .30 (or more extreme) if the population parameter is actually
.25 as you use larger and larger samples.
 Activity 17-5 345

Activity 17-5: Baseball “Big Bang”

a. The null hypothesis is that the proportion of all major-league baseball
games that contain a big bang is three-fourths. In symbols, the null hypothesis
is H0: ␲  .75.
b. The alternative hypothesis is that less than three-fourths of all major-league baseball
games contain a big bang. In symbols, the alternative hypothesis is Ha: ␲  .75.
c. A week of games was randomly selected. Although this does not constitute a
simple random sample of all games, you do hope it is representative of the scores
in such games. The CLT applies here because 95(.75)  71.25 is greater than
10, and 95(.25)  23.75 is also greater than 10. According to the CLT, the sample
proportion would vary approximately normally, with mean .75 and standard
deviation equal to
_________
(.75)(.25)
√ ________
95
 .0444

17
.55 .60 .65 .70 .75 .80 .85 .90 .95
Sample Proportion of Games with Big Bang

d. The sample proportion of games in which a big bang occurred is

47  .495
p̂  ___
95
e. Yes, this sample proportion is less than .75, as Marilyn conjectured.
f. The test statistic is
.495
z  __________
_________ .495  .75  5.74.
 .75  _________
(.75)(.25) .0444
√________
95
This test statistic says that the observed sample result is almost six standard
deviations below what the grandfather conjectured. This z-score is way off the
chart in Table II, indicating that the p-value is virtually zero ( .0002).
g. Yes, this very small p-value indicates that the sample data provide extremely
strong evidence against the grandfather’s claim. There is extremely strong
evidence that less than 75% of all major-league baseball games contain a big bang.
The null (grandfather’s) hypothesis would be rejected at the ␣  .01 level.
h. The hypotheses for testing Marilyn’s claim are H0: ␲  .5 vs. Ha: ␲ .5.
346 Topic 17: Tests of Significance: Proportions

i. The test statistic is

.495  .5  ________
z  ________
_______ .495  .5  0.10
(.5)(.5) .0513
√ ______
95
The p-value is 2(.4602)  .9204.

.3 .4 .5 .6 .7
Sample Proportion of Games with Big Bang

j. This p-value is not small at all, suggesting that the sample data are quite
consistent with Marilyn’s hypothesis that half of all games contain a big bang.
The sample data provide no reason to doubt Marilyn’s hypothesized value for ␲.
k. A 95% confidence interval for ␲ (the population proportion of games that
contain a big bang) is given by
________
p̂(1  pˆ )
√
p̂  z* ________
n
with z*  1.96, which is
___________
(.495)(.505)
√
.495  1.96 __________
95
which is .495  .101, which is the interval from .394 to .596. Therefore, you are
95% confident that between 39.4% and 59.6% of all major-league baseball games
contain a big bang. The grandfather’s claim (75%) is not within this interval
or even close to it, which explains why it was so soundly rejected. Marilyn’s
conjecture (50%) is well within this interval of plausible values, which is consistent
with it not being rejected.

••• Homework Activities

Activity 17-6: Interpreting p-values
a. You would reject the null hypothesis at the ␣  .10 level because the p-value is
less than .05 and therefore less than .10.
b. You do not have enough information to know whether you would reject the null
hypothesis at the ␣  .01 level. Although you know the p-value is less than .05,
you do not know whether it is less than .01 (e.g., it could equal .03).
 Activity 18-1 365

Activity 18-5: Hypothetical Baseball Improvements

Activity 18-5 introduces the important but challenging concept of power. It
also defines the two types of errors that can arise with a test of significance. You
might remind students that they have seen the idea of trade-offs between two
error probabilities before, for example, in Activity 12-4 when they applied normal
calculations to predicting whether a footprint came from a man or woman. Part f is a
key one: The considerable overlap between the two distributions (one where the success
probability ␲  .250 and one where ␲  .333) indicates that it’s hard to distinguish
a .250 hitter from a .333 hitter in a sample of 30 trials, which means that the power
of this test is low. When students increase the sample size in part j, they should find
much less overlap, indicating much higher power. We suggest that you focus students’
attention on this sample size issue and also on parts k and l, which reveal that the
alternative value of the parameter and the significance level are two other factors that
affect the power of a test.

Activity 18-6: West Wing Debate

Activity 18-6 is the self-check activity, giving students practice applying what they have
learned about interval-test duality and also about power and types of errors.

••• Homework Activities

Because this topic addresses very specific and fairly subtle issues, you may want to
choose homework activities even more carefully than usual. We suggest that you aim
for a mixture of activities that deal with interval-test duality, statistical vs. practical
significance, and statistical power. Notice that Activities 18-16, 18-17, and 18-18 make
use of the Power Simulation applet.
Possible assignment: Activity 18-10 deals with interval-significance test duality
and sampling bias, and Activity 18-15 explores statistical significance vs. practical

18
significance.

Solutions
••• In-Class Activities
Activity 18-1: Generation M
a. This is a statistic because it is a number that represents a sample.
b. For a 99% confidence interval, you calculate .68  (2.576)(.010348) 
.68  (.02666)  (.6533, .7066).
c. The values .70 and .6667 are in this interval; .65 and .707 are not.
d. A significance test should reject the hypothesis that ␲  .65, but .7 is contained
in the interval, so this could be a plausible value for ␲. A significance test would
not reject the hypothesis that ␲  .7.
e. Using Minitab’s Test and CI for One Proportion:
Test of p  0.65 vs p not  0.65
Sample X N Sample p 99% CI Z-Value P-Value
1 1382 2032 0.680118 (0.653465, 0.706771) 2.85 0.004
Using the normal approximation.

f. See table following part g.

366 Topic 18: More Inference Considerations

g. Here is the completed table:

Hypothesized Contained in 99% Test Significant at

p-value
Value Confidence Interval? Statistic .01 Level?

.65 No 2.85 .004 Yes

.6667 Yes 1.28 .199 No

.7 Yes 1.96 .05 No

.707 No 2.66 .008 Yes (barely)

h. If a hypothesized value is contained in the 99% confidence interval, then this

value is not significant at the .01 level, and vice versa.

Activity 18-2: Pet Ownership

a. Because this number (.316) describes a sample, it is a statistic, represented by p̂.
b. Let ␲ represent the proportion of all American households who own a pet cat.
The null hypothesis is that one-third of all American households own a pet cat. In
symbols, the null hypothesis is H0: ␲  .333.
The alternative hypothesis is that the proportion of American households who
own a pet cat differs from one-third. In symbols, the alternative hypothesis is
Ha: ␲ .333.
.316  .333  10.20.
The test statistic is z  ____________
__________
(.333)(.667)
√__________
80,000
Using Table II, p-value  2  Pr(Z  10.20)  .0002.
Reject H0 with this very small p-value.
You have overwhelming statistical evidence that the proportion of all American
households who own a cat differs from one-third.
___________________
c. For a 99.9% CI, you calculate .316  (3.291) √.316(1  .316)80000 
.316  (3.291)(.001644)  (.310591, .321409). You are 99.9% confident the
proportion of all American households who own a pet cat is between .311
and .321.
d. Yes, this confidence interval is consistent with the test results because 13  .333
is not contained in the interval.
e. Yes, the sample data provide very strong evidence that the population proportion
(␲) is not one-third. The p-value is what helps you decide this; the p-value is so
small (essentially zero) that it easily convinces you that ␲ is not one-third.
f. No, the sample data do not provide strong evidence that the population proportion
of households who own a pet cat is very different from one-third. The evidence
suggests that this proportion is between .311 and .321, which are awfully close to
.33. The confidence interval helps you decide how much ␲ differs from one-third.
 Activity 18-5 367

Activity 18-3: Racquet Spinning

a. Sample proportion: .565 Test statistic: 1.84 p-value: .066 Significant at .05? no
b. Sample proportion: .575 Test statistic: 2.12 p-value: .034 Significant at .05? yes
c. Sample proportion: .65 Test statistic: 4.24 p-value: .000 Significant at .05? yes
d. The sample results are most similar in parts a and b, where you had almost the
same number of “ups.”
e. The decisions are the same in parts b and c (where the sample results are quite
dissimilar).

Activity 18-4: Female Senators

a. For a 95% CI, you calculate .16  1.96(.0367)  (.088, .231).
b. No, this confidence interval is not a reasonable estimate of the actual proportion
of all humans who are female.
c. The confidence interval procedure fails in this case because the alien did
not select a simple random sample of all humans. The U.S. Senate is not
representative of the population of all humans with respect to gender, so the
sampling method is extremely biased and you cannot legitimately use the
confidence interval procedure.
d. You do not need to estimate the proportion of women in the 2007 U.S. Senate.
You know this proportion is .16.

Activity 18-5: Hypothetical Baseball Improvements

a. The null hypothesis is that this player is still a .250 hitter. In symbols,
H0: ␲  .250.

18
The alternative hypothesis is that this player has improved and is now better than
a .250 hitter. In symbols, Ha: ␲  .250.
b. A Type I error would be deciding that the player has improved his batting
performance when, in fact, he is still batting no better than .250.
c. A Type II error would be failing to realize that the player has improved.
d. Answers will vary. The following is a representative set:

This distribution is roughly normal, centered at about 7.5 hits, and extends from
about 1 hit to about 17 hits.
368 Topic 18: More Inference Considerations

e. A player would need to get at least 13 hits.

f. There is a great deal of overlap between the two distributions.

g. 28200  14%
h. No, it does not appear very likely that a .333 hitter will be able to establish that
he is better than a .250 hitter in 30 at-bats. Based on this simulation, he had only
about a 14% chance of establishing his improvement (performing well enough to
convince the manager that his success rate was now greater than .250).
i. Power  .14
j. The following is based on one representative running of the applet:

Based on this simulation, a player would need at least 33 hits (out of 100) in order
for the probability of a .250 hitter to do that well by chance alone to be less than
.05. The approximate power of this test is 123200 or .615.
k. Answers will vary by student expectation, but the test will be more powerful if the
player improves to a .400 hitter. It should be easier to detect the improvement to
.400 because it is farther away from .250 than .333 is.
 Activity 18-6 369

The simulation confirms this result; the approximate power is 108200 or .54
(using 30 at-bats).

l. Answers will vary by student expectation, but the test will be more powerful if
you use a higher significance level.
The simulation confirms this result; the approximate power is 52200 or .26
(using 30 at-bats, alternative value of ␲  .333 and the significance level .10).

m. i. The magnitude of the difference between the hypothesized value ␲0 and the
18
particular alternative value of ␲
ii. The significance level

Activity 18-6: West Wing Debate

a. The population of interest is all adult Americans who are familiar with these
fictional candidates. The parameter (call it ␲) is the proportion of this population
who would have supported Santos if they had been asked.
b. The 90% CI for ␲ is .54  .024, which is (.516, .564).
370 Topic 18: More Inference Considerations

The 95% CI for ␲ is .54  .028, which is (.512, .568).

The 99% CI for ␲ is .54  .037, which is (.503, .577).
c. The midpoints are all the same, namely .54, the sample proportion of Santos
supporters. The 99% CI is wider than the 95% CI, and the 90% CI is the
narrowest.
d. Yes. All three intervals contain only values greater than .5, so they do suggest,
even with 99% confidence, that more than half of the population would have
favored Santos.
e. H0: ␲  .5 (half of the population favored Santos)
Ha: ␲  .5 (more than half of the population favored Santos)
f. Because all three intervals fail to include the value .5, you know that the p-value
for a two-sided alternative would be less than .10, .05, and .01. Because you
have a one-sided alternative in this case, you know that the p-value will be less
than .01 divided by 2, or .005 (because the observed sample proportion is in the
conjectured direction).
g. A Type I error occurs when the null hypothesis is really true but is rejected. In
this case, a Type I error would mean that you conclude that Santos was favored
by more than half of the population when in truth he was not favored by more
than half. In other words, committing a Type I error means concluding that
Santos was ahead (favored by more than half) when he wasn’t really. A Type II
error occurs when the null hypothesis is not really true but is not rejected (you
continue to believe a false null hypothesis). In this case, a Type II error means that
you conclude Santos was not favored by more than half of the population when
in truth he was favored by more than half of the population. In other words,
committing a Type II error means concluding that Santos was not ahead when he
really was.
h. The test would be more powerful if Santos really were favored by 55% rather
than 52%. The higher population proportion would make it more likely to
reject the null hypothesis that only half of the population favored Santos
because the distribution of sample proportions would center around .55 rather
than .52 (further from .5).
i. The larger sample (10,000) would produce stronger evidence that more than
half of the population favored Santos. With less variability in the sampling
distribution, the p-value would be much smaller.

••• Homework Activities

Activity 18-7: Charitable Contributions
__________________
a. For a 90% CI, you calculate .788  (1.645) √ .788(1  .788)1334  (.769,
.806).
b. For a 99% CI, you calculate .788  (2.576)(.0112)  (.759, .817).
c. Yes, the sample proportion does differ from .75 at the  .01 significance level
because .75 is not in the 99% confidence interval. Thus, you would reject the null
hypothesis H0: ␲  .75 vs. the two-sided alternative Ha: ␲ .75, and would
decide that .75 is not a plausible value for ␲.
 Activity 19-1 387

2. Rather than editing the blue values in the estimate box directly to reflect the
summary statistics from the sample, drag the Body_Temp attribute to the
“Attribute (Numeric)” label at the top of the estimate box to find the relevant
statistics automatically (mean, standard deviation, and sample size) and compute
the 95% confidence interval for the mean. If desired, choose Estimate | Verbose
to display a less wordy version of the output.
3. Select the blue “95%” and change it to 90% and then 99% to compute the other
intervals needed for the comparison in part g.

Activity 19-3: Body Temperatures (Graphing Calculator)

In part f, students can use their calculator’s T-Interval command to confirm their
calculation of the 95% confidence interval (CI) in part a.
1. The T-Interval can be found by pressing Ö and selecting the TESTS menu.
2. The T-Interval test gives you a choice of inputs, either Data or Stats. For this
activity, select Data and then input the list named BDTMP.
3. Set the frequency to 1.
4. Set the confidence level (C-Level) to the level of interest.
5. If you select the Stats option for the T-Interval, then you will need to enter the
sample mean, standard deviation, and the sample size.
6. Use your calculator to produce a 90% CI and a 99% CI for the population mean
body temperature.

Solutions
••• In-Class Activities
Activity 19-1: Christmas Shopping
a. The amount expected to be spent on Christmas presents in 1999 is a quantitative
variable.

19
b. The value $857 is a statistic because it is a number that describes a sample. This
_
statistic is represented by x.
c. The parameter is the average (mean) amount expected to be spent by all American
adults on Christmas presents in 1999. This parameter is represented by ␮.
d. You do not know the value of the ␮, but it is more likely to be close to $857 than
to be far from it.
_ __ ____
e. The standard deviation of the sample mean x is √n  250√922  $8.23.
f. This interval estimate works out to be $857  2($8.23)  $857  $16.47 
($840.53, $873.47).
g. The sample standard deviation, s, is a reasonable substitute for ␴.
388 Topic 19: Confidence Intervals: Means

h. Answers will vary. The following is from one representative running of the applet:
You find 96% of the intervals succeed in capturing the value of ␮. Your total
should be roughly 95%.

i. The running total percentage of intervals that succeed in capturing the

population mean is 95.3%, which is very close to the expected 95%.

j. The running total percentage of intervals that succeed in capturing the

population mean is 92.4%, which is noticeably less than 95%.
 Activity 19-2 389

Activity 19-2: Exploring the t-Distribution

a. Here is a sketch of the t(9) distribution:

.9500

19
.025 –t* 0 t* .025
T, df = 9

b. See shading in plot above.

c. The area to the right of t*  1  .975  .025.
d. Using the t-table (Table III), t*  2.262.
e. This t* critical value is greater than the z* critical value for a 95% confidence
interval because of the greater uncertainty introduced by estimating with s rather
than ␴. This will make the margins-of-error of your confidence intervals wider in
order to achieve the stated confidence level in the long run.
390 Topic 19: Confidence Intervals: Means

f. For a 95% confidence interval, t*  2.045. This value is less than the previous t*
value, which makes sense because you have increased the sample size, decreasing
the uncertainty in estimating ␴ by s.
g. For a 90% confidence interval t*  1.699; for a 99% confidence interval t* 
2.759. The t* for a 99% confidence interval is greater, which is appropriate
because this interval claims more confidence (more certainty). In order to be more
confident, the interval will need to be wider.
h. With 100 degrees of freedom, t*  1.984.

Activity 19-3: Body Temperatures

a. For a 95% CI with 129 degrees of freedom, you calculate 98.249  (1.984)
____
(.733)√130  (98.1215, 98.3765).
b. You are 95% confident the average body temperature of a sample of 130 healthy
adults is between 98.12°F and 98.38°F.
When you say 95% confident, you mean that if you repeated this procedure
of creating confidence intervals in this same manner (using random samples
of 130 healthy adults), in the long run 95% of the intervals would contain the
population mean body temperature of all healthy adults and 5% of the intervals
would not contain this parameter.
c. Here are graphs of the sample data:

25

20
Frequency

15

10

0
96.75 97.50 98.25 99.00 99.75 100.50
Body Temperature (in °F)

96.6 97.2 97.8 98.4 99.0 99.6 100.2 100.8

Body Temperature (in °F)
 Activity 19-3 391

99.9

99

95
90
80
70

Percentage
60
50
40
30
20
10
5

0.1
95 96 97 98 99 100 101
Body Temperature (in ⴗF)

Yes, the body temperatures appear to be roughly normally distributed.

d. No, the normality of the population of body temperatures is not required for this
t-procedure to be valid with these data because the sample size is large (n  130).
e. You do not know whether this was a simple random sample of healthy adults.
If you assume that it was, then the other technical condition required for the
validity of this t-interval is satisfied.
f. Using Minitab’s One-Sample T procedure,
Variable N Mean StDev SE Mean 95% CI
body temp 130 98.2492 0.7332 0.0643 (98.1220, 98.3765)

90% CI: (98.1427, 98.3558) 99% CI: (98.0811, 98.4174)

g. The midpoints of all three intervals are the same: 98.249 (the sample mean). The

19
90% confidence interval is the most narrow (width  .2131), followed by the
95% confidence interval (width  .2545), whereas the 99% confidence interval
is the widest (width  .3363).
h. It does not appear that 98.6 is a plausible value for the mean body temperature
for the population of all healthy adults because this value is not contained in any
of the confidence intervals.
i. If the sample size had been only 13, but the sample mean and standard deviation
had been the same, the 95% CI would be much wider (though it would have the
same center) because (i) the t* value used to create the interval would be much
greater and (ii) the standard error would be greater (the square root of 13 is much
smaller than the square root of 130).
j. For a 95% confidence interval
___ with 12 degrees of freedom, you calculate
98.249  (2.179) (.733)√13  (97.807, 98.692)°F.
As predicted, the midpoint is still 98.249, but the width is much greater (.88597).
In fact, with this interval, 98.6 would be a plausible value for the mean body
temperature for the population of all healthy adults.
392 Topic 19: Confidence Intervals: Means

Activity 19-4: Sleeping Times

a. Here is the completed table:

Sample Number Sample Size Sample Mean Sample SD

3 30 6.6 0.825

1 10 6.6 0.825

2 10 6.6 1.597

4 30 6.6 1.597

b. They all have a sample mean of 6.6 hours.

c. The most important difference between samples 1 and 2 are the spreads (sample
SDs). Sample 1 has a much smaller SD than does sample 2.
d. The most important difference between samples 1 and 3 is the sample size.
Sample 3 uses a sample of size 30, whereas sample 1 uses a sample of size 10.
e. Sample 1 produced a more precise estimate of ␮. This result makes sense because
sample 1 has a smaller SD than sample 2, and so sample 1 will have a smaller
margin-of-error.
f. Sample 3 produced a more precise estimate of ␮. This result makes sense
because sample 3 has a larger sample size, and so sample 3 will have a smaller
margin-of-error.

Activity 19-5: Sleeping Times

Results will vary by class, but here are the results from one college class.
a. The following graph displays the results:

0.0 1.6 3.2 4.8 6.4 8.0 9.6 11.2

Sleep Time (in hours)

These 40 sleep times are roughly normally distributed, centered around

7 hours, with a minimum of 3 hours and a maximum of 10.5 hours. There are
no noticeable outliers.
_
b. The sample size is n  40; the sample mean is x  6.981 hours; and the sample
standard deviation is s  1.981 hours.
c. The sample size is large (40  30), but the sample was not randomly selected,
because it consisted of the students in this one class. It might not be representative
of students at the entire school with regard to sleep hours, as students in a statistics
class may tend to be mostly from one type of major who may tend to study and
sleep more or less than the typical student.
d. For a 90% CI, with 39 degrees of freedom, you calculate 6.981  (1.685)(.31322) 
(6.45322, 7.50878) hours.
 Activity 19-6 393

You are 90% confident the average amount of sleep per night obtained by all
students at the school is between 6.45 and 7.51 hours.
e. Seven of the 40 sleep times fall within this interval. This is 17.5%.
f. No, this percentage is not close to 90%, but there is no reason that it should be.
Your interval is designed to estimate the average sleep time. It is not telling you
anything about the individual sleep times. They may or may not fall within this
interval.

Activity 19-6: Backpack Weights

a. The observational units are the students. The variable is the ratio of backpack
weight to body weight, which is quantitative. The sample is the 100 Cal Poly
students whose weights were recorded by the student researchers. The population
is all Cal Poly students at the time the study was conducted.
b. The following histogram reveals that the distribution of these weight ratios is a bit
_
skewed to the right. The center is around .07 or .08 (mean x  .077, median 
.071). The five-number summary is (.016, .050, .071, .096, .181), so students in
the sample carried as little as 1.6% of their weight in their backpacks and as much
as 18.1% of their weight in their backpacks. The standard deviation of these ratios
is s  .037.

20

16
Frequency

12

0
.00 .02 .04 .06 .08 .10 .12 .14 .16 .18 .20

19
Ratio of Backpack Weight to Body Weight

c. Calculating a 99% CI for the population mean by hand using the formula
_ s__
x  t* ___
√n

with 100  1  99 rounding down to 80 degrees of freedom gives .0771 

____
2.639  .0366√100 , which is .0771  .0097, which corresponds to the interval
from .0674 through .0868. (This t* value should be based on 99 degrees of
freedom, but we used 80 degrees of freedom here, the closest value less than 99
that appears in Table III.) Using technology gives a slightly more accurate 99%
CI for ␮ of .0675 through .0867.
d. You are 99% confident that the mean weight ratio of backpack-to-body weights
among all Cal Poly students at the time of this study is between .0674 and .0868.
In other words, you are 99% confident that the average Cal Poly student carries
between 6.74% and 8.68% of his/her body weight in his/her backpack. By “99%
confidence,” you mean that 99% of all intervals constructed with this method
would succeed in capturing the actual value of the population mean weight ratio.
394 Topic 19: Confidence Intervals: Means

e. The first condition is that the sample be randomly selected from the population.
This is not literally true in this case because the student researchers did not obtain
a list of all students at the university and select randomly from that list, but they
did try to obtain a representative sample. The second condition is either that the
population of weight ratios is normal or that the sample size is large. In this case,
the sample size is large (n  100, which is greater than 30), so this condition is
satisfied even though the distribution of ratios in the sample is somewhat skewed
(and so presumably is the population).
f. You do not expect 99% of the sample, nor 99% of the population, to have
a weight ratio between .0673 and .0869. You are 99% confident that the
population mean weight ratio is between these two endpoints. In fact, only 18 of
the 100 students in the sample have a weight ratio in this interval.

••• Homework Activities

Activity 19-7: Body Temperatures
___
a. For males, 98.394  (2.00) .743√65   (98.2097, 98.5783)°F using Table III
with 60 degrees of freedom.
Or (98.2096, 98.5781)°F using technology.
___
For females, 98.105  (2.00) .699√65   (97.932, 98.278)°F using Table III
with 60 degrees of freedom.
Or (97.9315, 98.2778)°F using technology.
b. You are 95% confident the average body temperature of healthy adult males is
between 98.2°F and 98.6°F. You are 95% confident the average body temperature
of healthy adult females is between 97.9°F and 98.3°F.
c. These intervals suggest that, on average, women may have a slightly lower body
temperature than men.
d. Half-width males: .184 half-width females: .173.
The confidence interval for the males is wider. This is because the sample
standard deviation is slightly larger for the males than it is for the females
(.743 vs. .699).
e. The half-width of the interval based on the entire sample is .25452  .12725.
This is significantly smaller than the half-width of either the male or female
interval. You expect this, however, because the sample size is twice as __large for this
interval, so the margin-of-error will be smaller by at least a factor of √2 .
f. For both genders, the sample sizes are large, so you do not need to check that
the populations are normally distributed. You do need to assume that you have
a simple random sample of healthy adult males and a simple random sample of
healthy adult females in order for the t-intervals to be valid.

Activity 19-8: Credit Card Usage

a. You need to know the sample standard deviation.
_____
b. For a 95% CI, you calculate $2169  (1.965)($1000√1074 )  ($2109.04,
$2228.96) using Table III with 500 degrees of freedom.
With technology: ($2109.10, $2228.90)
 Activity 20-1 407

••• Homework Activities

The first two homework activities simply ask students to practice reading the t-table.
You might want to assign a mixture of some activities that use summary statistics
(such as Activities 20-8, 20-9, 20-10, 20-14, and 20-15) and some that use raw
data (such as Activities 20-11, 20-12, 20-13, 20-16, and 20-17). Notice that the last
several homework activities concern some of the ideas presented in Topic 18, such as
types of errors and the duality of tests and intervals.
Suggested assignment: It might be good to choose an activity that involves testing
by hand and one that involves technology. Consider Activities 20-7 and 20-13 (if you
collected age-guess data) or Activity 20-10.

Technology Ideas
Activity 20-2: Sleeping Times (Fathom)
Part e of this activity asks students to use technology to calculate test statistics and
p-values for four different samples of hypothetical sleep times.
1. Open the [Link] file, drag an empty test box from the shelf, and
change the type of test to Test Mean.
2. Drag one of the samples to the top row of the test box, and edit the blue value
in either hypothesis to change the hypothesized mean from “0” to 7. Select
the test box or press Õ to accept the change. You might want to choose
Test | Verbose from the main Fathom menu to see more concise results.
3. Drag each of the other samples to the same test box in turn to find the results
needed to summarize each test in the table in part b.

Activity 20-2: Sleeping Times (Graphing Calculator)

For each of the four sleeping time samples, students can use their calculator’s T-Test
command in part e of this activity to compute the test statistic and p-value for testing
that the population mean differs from seven hours.
1. The T-Test can be found by pressing Ö and selecting the TESTS menu.
2. For the T-Interval, select Data (lists SLP1, SLP2 , SLP3, and then SLP4) as input.
3. Set the frequency to 1.
4. Select Calculate.

Solutions
As an alternative, you can explore the Draw feature.
20
••• In-Class Activities
Activity 20-1: Basketball Scoring
a. If the rule changes had no effect on scoring, ␮ would have the value 183.2. This
is the null hypothesis.
b. If the rule changes had the desired effect on scoring, then ␮  183.2, which
would be the alternative hypothesis.
408 Topic 20: Tests of Significance: Means

c. The following graph displays the NBA game data:

140 154 168 182 196 210 224 238

Points Scored

Scoring definitely seems to have increased over the previous season’s mean of
183.2 points per game. In only 4 of these 25 games were fewer than 183 points
scored, and the center of this dotplot is now about 196 points per game.
_
d. Sample mean x : 195.88 points Sample standard deviation s: 20.27 points
e. Yes, the sample mean is in the direction specified in the alternative hypothesis
(greater than 183.20 points).
f. Yes, it is possible to have gotten such a large sample mean even if the new rules had
no effect on scoring.
g. H0: ␮  183.2
Ha: ␮  183.2
h. Technical conditions: The sample size is not large (n  25  30), so the
population must follow a normal distribution. A probability plot of the data
provides evidence that these sample data are not arising from a normal population.
Still, the data are reasonably symmetric and the sample size is moderately large, so
this condition could be considered met with caution.

99

95
90

80
70
Percentage

60
50
40
30
20

10
5

1
150 175 200 225 250
Points Scored

However, the data are not a simple random sample gathered from the population
as the data are all the NBA games played from December 10–12, 1999. These
games occurred relatively early in the season and are probably not representative of
scoring over the course of the entire season, especially with respect to a new rule
change. So the technical conditions for the validity of this t-test have not been met.
195.88  183.2
i. The test statistic is t  _____________
___  3.13.
20.27√25
j. Here is a sketch of the t-distribution:
 Activity 20-2 409

df = 24

.00227

0 3.13
t-values

k. You find 2.797  3.13  3.467, so .005  p-value  .001.

l. Using the applet, the p-value is .0023. The picture shows the information that
should be entered into the applet:

m. If the average number of points per game for all NBA games in this season were still
183.2 points, there is only a .0023 chance that you would find a random sample
of 25 games with a mean of at least 195.88 points. Because finding a sample as
extreme as this one is so unlikely by chance alone, you conclude that the mean
number of points per game this season has increased; it is no longer 183.2 points.
n. Yes, you would reject the null hypothesis at the .10, .05, .01, and .005 levels
because the p-value is less than each of these significance levels.
20
o. If these data had been a random sample from a normal population, you would have
very strong statistical evidence that the mean points per game in the 1999–2000
season were greater than in the previous season. However, you would not be able to
conclude that the rule change caused the average point increase because this is not
a randomized experiment.

Activity 20-2: Sleeping Times

Answers will vary by class. The following is one representative set of answers.
a. Let ␮ represent the mean sleep time of all students at your school.
410 Topic 20: Tests of Significance: Means

The null hypothesis is that the mean sleep time of the population is 7 hours. In
symbols, the null hypothesis is H0: ␮  7.0 hours.
The alternative hypothesis is that the mean sleep time of the population is not
7 hours. In symbols, the alternative hypothesis is Ha: ␮ 7.0 hours.
6.981 ___7  0.06.
The test statistic is t  _________
1.981√40
Using Table III with 39 degrees of freedom, p-value  2  .20  .40.
Using Minitab, p-value  2  .476231  .952462.
Because the p-value is not small, do not reject H0.
You do not have any statistical evidence to suggest that the mean sleep time of all
students at your school differs from 7.0 hours.
b. Technical conditions: The sample size is large (40  30), but the sample was
not randomly selected because it consisted of only the students in your class. It
might not be representative of students at your school with regard to sleep hours,
as students in a statistics class may tend to have similar majors and may tend to
study and sleep more or less than the typical student. Here is the completed table:

Sample Number Sample Size Sample Mean Sample SD Test Statistic p-value

1 10 6.6 0.825 1.53 .1596

2 10 6.6 1.597 0.79 .4487

3 30 6.6 0.825 2.66 .0127

4 30 6.6 1.597 1.37 .1806

c. Sample 1 would produce a smaller p-value because it has the smaller standard deviation.
d. Sample 3 would produce a smaller p-value because it has the larger sample size.
e. See the table following part b.
f. Only sample 3 gives enough evidence to reject the null hypothesis at the .05 level.
g. Answers will vary by student conjecture.

Activity 20-3: Golden Ratio

a. The following histogram displays data for width-to-length ratios for a sample of
20 beaded rectangles:

8
7
6
Frequency

5
4
3
2
1
0
.6 .7 .8 .9
Width-to-Length Ratio
 Activity 20-4 411

These twenty width-to-length ratios are skewed right, with a minimum of

.553 and a maximum of .933. The median is .6410, the mean is .6605, and the
standard deviation is .0925.
b. Let ␮ represent the average width-to-length ratio for all beaded rectangles made
by the Shoshoni Indians.
The null hypothesis is that this average ratio is the golden ratio (.618). In symbols,
the null hypothesis is H0: ␮  .618.
The alternative hypothesis is that this average ratio is not the golden ratio. In
symbols, the alternative hypothesis is Ha: ␮ .618.
Technical conditions: You don’t know whether the sample was randomly selected,
but it is small (n  20), and the sample (and therefore the population) does not
appear to be normally distributed. So the technical conditions for this procedure
to be valid are not satisfied.

99

95
90

80
70
Percentage

60
50
40
30
20

10
5

1
.3 .4 .5 .6 .7 .8 .9 1.0
Width-to-Length Ratio

.6695  ___
.618  2.05.
The test statistic is t  ___________
.0925√20
The p-value is 2  Pr(T  2.05).
Using Table III with 19 degrees of freedom, 1.729  2.05  2.093, so
2  .025  p-value  2  .05 or .05  p-value  .10. You would reject
H0 at the .10 significance level.
Using the applet, the p-value  .0539  .10. You would reject H0 at the
.10 significance level.
20
You can conclude that the average width-to-length ratio for all beaded rectangles
made by the Shoshoni Indians is not the golden ratio.

Activity 20-4: Children’s Television Viewing

The observational units are third- and fourth-grade students. The sample consists of the
198 students at two schools in San Jose. The population could be considered all American
third- and fourth-graders, but it might be more reasonable to restrict the population to be
all third- and fourth-graders in the San Jose area at the time the study was conducted.
412 Topic 20: Tests of Significance: Means

The variable measured here is the amount of television the student watches in a
typical week, which is quantitative. The parameter is the mean number of hours of
television watched per week among the population of all third- and fourth-graders.
This population mean is denoted by ␮. The question asked about watching an average
of two hours of television per day, so convert that to be 14 hours per week.
The null hypothesis is that third-and fourth-graders in the population watch an
average of 14 hours of television per week (H0: ␮  14). The alternative hypothesis is that
these children watch more than 14 hours of television per week on average (Ha: ␮  14).
Check technical conditions:
• The sample of children was not chosen randomly; they all came from two schools
in San Jose. You might still consider these children to be representative of third-
and fourth-graders in San Jose, but you might not be willing to generalize to a
broader population.

• The sample size is large enough (198 is far greater than 30) that the second
condition holds regardless of whether the data on television watching follow a
normal distribution. You do not have access to the child-by-child data in this case,
so you cannot examine graphical displays; however, the large sample size assures
you that this condition may be considered satisfied.
_
Test statistic: The sample size is n  198; the sample mean is x  15.41 hours;
and the sample standard deviation is s  14.16 hours. The test statistic is
15.41 ____
14  1.401
t  __________
14.16√198
indicating the observed sample mean lies 1.401 standard errors above the conjectured
value for the population mean. Using Table III and the 100 degrees of freedom line
(rounded down from the actual number of degrees of freedom of 198  1  197)
reveals the p-value (probability to the right of t  1.401) to be between .05 and .10.
Technology calculates the p-value more exactly to be .081.

Test decision: This p-value is not less than the .05 significance level. The sample
data, therefore, do not provide sufficient evidence to conclude that the population
mean is greater than 14 hours of television watching per week.
Conclusion in context: This conclusion stems from realizing that obtaining a
sample mean of 15.41 hours or greater would not be terribly uncommon when the
population mean is really 14 hours per week. If you had used a greater significance level
(such as .10), which requires less compelling evidence in order to reject a hypothesis,
then you would have concluded that the population mean exceeds 14 hours per week.
 Activity 21-1 441

21
2. Edit the blue values to reflect the counts and sample sizes for each of the two
groups (740 out of 1000 in one sample and 240 out of 1000 in the other).
3. Select the blue text, “is not equal to” in the alternative hypothesis statement ( Ha )
and change it to is greater than for the lower tail test.
4. (Optional) For a more informative display, change the generic names
“FirstAttribute” and “SecondAttribute” to 1992 and 1996, respectively, and
change “Category” to Stomach as a reminder of what the proportions measure.
To confirm the interval calculation, follow these steps:
1. Drag a new estimate box from the shelf and change the type of estimation to
Difference of Proportions.
2. As in the steps above, edit the counts to show those found in the two samples.

Activity 21-2: Back to Sleep (Graphing Calculator)

In part k of this activity, the students can use their calculator’s 2-PropZTest command:
1. Press Ö and select the TESTS menu.
2. Select 2-PropZTest and press Õ.
3. Confirm the test and interval calculations using x1  700, n1  1000, x2 
240, and n2  1000.
4. Select Calculate and press Õ.
5. As an alternative, you can explore the Draw feature.
In part m of this activity, students can use the 2-PropZInt command:
1. Press Ö and arrow over to the TESTS menu.
2. Select 2-PropZTest and press Õ.
3. Compute a 95% confidence interval for ␲1996 and ␲1992.
4. For this exercise, you will need to select Stats rather than Data for Inpt.
5. As an alternative, you can explore the Draw feature.

Solutions
••• In-Class Activities
Activity 21-1: Friendly Observers
a. Explanatory: whether the observer shares Type: binary categorical
the prize
Response: whether the participants beat Type: binary categorical
the threshold time
b. This is an experiment because the researchers randomly assigned the subjects to
the control and treatment groups.
c. The sample proportions of success for each group are p̂A  .25; p̂B  .6667.
The following segmented bar graph displays the results:
442 Topic 21: Comparing Two Proportions

100
Did not beat threshhold
90
Beat threshhold
80
70

Percentage
60
50
40
30
20
10
0
A: Observer B: No Sharing
Shares Prize of Prize
Sharing of Prize

d. Yes, these sample proportions differ in the direction conjectured by the

researchers. The proportion in group A is substantially smaller than in group B.
e. Yes; even if there were no effect due to the observer’s incentive, it is possible to
have obtained such a big difference between the two groups simply from the
random assignment process.
For parts f–l, answers will vary by student. The following are one representative set
of answers.
f. Five of the 12 cards are successes. No, this result is not at least as extreme as the
result in the actual study.
g. Here is a completed table:

Repetition # 1 2 3 4 5

Number of “Successes” Assigned to Group A 5 6 4 8 2

Is Result as Extreme as in Actual Study? No No No No Yes

h. Here is a dotplot of the number of successes randomly assigned to group A:

0 1 2 3 4 5 6 7 8 9 10 11
Number of Successes Assigned to Group A

i. The most common values for the number of successes are 5 and 6. This result
makes sense because if the cards are randomly assigned to groups A and B, then
half of the 11 successes should end up in group A. So, typically you expect 5 or 6
of the successes in group A.
j. Forty repetitions were performed by this class. Five of these repetitions gave a
result of 3 or fewer successes in group A. This represents a proportion of .125.
k. Based on this class’s simulated results, it does not appear unlikely for random
assignment to produce a result as extreme as (or more extreme than) the actual
sample when the observer has no effect on subjects’ performance.
 Activity 21-2 443

21
l. No, in light of your answer to the previous question, the data do not provide
reasonably strong evidence in support of the researchers’ conjecture; this type of
sample result appears to happen in about 12–13% of random assignments when
the observer has no effect on subjects’ performance.
m. Yes, the number of successes randomly assigned to group A varies. The five values
that occur in this running of the applet are 6, 5, 5, 8, and 7.
n. In this simulation, the resulting distribution is approximately normal, centered at
about 5.4 successes, with a standard deviation of 1.327 successes.
o. Here are the simulation results:

The approximate p-value is 32500, or .064.

p. Based on this approximate p-value from the larger simulation, it does appear
somewhat unlikely for random assignment to produce a result at least as extreme
as the actual sample when the observer has no effect on subjects’ performance.
The data provide moderate evidence in support of the researchers’ conjecture
because this type of sample result appears to happen in no more than about 6% of
random assignments when the observer has no effect on subjects’ performance.

Activity 21-2: Back to Sleep

a. This study involves random sampling from populations.
b. The null hypothesis is the proportion of all infants who sleep on their stomachs is
the same in 1996 as it was in 1992. In symbols, the null hypothesis is H0: ␲1996 
␲1992 or H0: ␲1996  ␲1992  0.
The alternative hypothesis is the proportion of all infants who sleep on their
stomachs in 1996 is less than it was in 1992. In symbols, the alternative
hypothesis is Ha: ␲1996  ␲1992 or Ha: ␲1996  ␲1992  0.
444 Topic 21: Comparing Two Proportions

c. The combined sample proportion of infants who slept on their stomachs is

700  240  .47
p̂c  _________
2000

.24  .70
d. The test statistic is z  ______________________
_____________________  20.61.
1
1  _____
√ (
(.47)(.53) _____
1000 1000 )
Yes, this is a very large z-score (in absolute value).
e. p-value  Pr(Z  20.61)  .000
f. If the proportion of infants who sleep on their stomachs is the same in 1996 as
it was in 1992, the probability that you would find a difference in sample results
as or more extreme than this study by random sampling alone is essentially 0,
which means it would just about never happen. Because you did find this sample
difference, you have very strong evidence that the null hypothesis was an incorrect
conjecture, and the proportion of infants who sleep on their stomachs in 1996 was
less than it was in 1992.
g. Reject the null hypothesis at the  .01 significance level because the p-value is
smaller than .01.
h. This sample difference provides extremely strong evidence that the proportion
of all households that place infants to sleep on their stomachs decreased between
1992 and 1996.
___________________
(.70)(.30) (.24)(.76)
1000 √
i. For a 95% CI, you calculate (.70  .24)  (1.96) ________  ________ ,
1000
which is .46  (1.96) (.0198), which is .46  .0388  (.4212, .4988).
j. You are 95% confident the difference in population proportions is between
.420 and .499. Because this interval does not contain 0, you are 95% confident
the population proportions are not the same. Because the values in the interval are
strictly positive, you are 95% confident that the percentage of infants who sleep
on their stomachs has decreased (between 42 and 50 percentage points) in the
years from 1992 to 1996.
k. Here are the applet results:
 Activity 21-4 445

21
l. Answers will vary by student expectation. ___________________
(.70)(.30) (.24)(.76)
√
m. For a 95% CI, you calculate (.24  .70)  (1.96) ________  ________ 
1000 1000
(.4988, .4212). The endpoints are reversed and multiplied by 1 in this
interval. The interval has the same width as the previous interval, however, and its
midpoint is the negative of that of the previous one.

Activity 21-3: Preventing Breast Cancer

a. This is an experiment because the women were randomly assigned to receive one
of the treatments.
b. Explanatory: which drug (tamoxifen or raloxifene) the woman received
Response: whether the woman developed invasive breast cancer
c. H0: ␲T  ␲R (drugs are equally effective)
Ha: ␲T ␲R (there is a difference in the breast cancer rate between the two drugs)
d. Here is the 2  2 table:

Taxmoxifen Raloxifene Total

Developed Invasive Breast Cancer 163 167 330

Did Not Develop Breast Cancer 9,563 9,578 19,141

Total 9,726 9.745 19,471

e. The sample proportions are p̂T  .01676 and p̂R  .01714.

.01676  .01714
f. The test statistic is z  _______________________________
______________________________  0.20.
1
1  _____
√(.01695(1  .01695) _____(
9726 9745 )
g. No, this is not a large z-score. This is not surprising in light of the values
calculated in part e, because the two proportions calculated for the tamoxifen and
raloxifene groups are quite similar.
h. Using Table II, p-value  2  Pr(Z  0.20)  2  .4207  .8414. Using
Minitab, the p-value is .838.
i. Because the p-value is not small, do not reject H0. You do not have statistical
evidence that there is any difference between these two drugs in terms of their
effectiveness in preventing breast cancer for five years. If you had found a
difference, you could safely conclude it was caused by a difference in the drugs
because this was a well-designed experiment. Because the study did not select a
random sample, you should be cautious in generalizing these results to a larger
population. In particular, you should not extend any of your conclusions beyond
postmenopausal women who are at increased risk for breast cancer.

Activity 21-4: Perceptions of Self-Attractiveness

a. You need to know how many men and women were surveyed.
b. Answers will vary, but one example is if only 100 men and 100 women were
surveyed because this would mean that 71 women and 81 men were satisfied
446 Topic 21: Comparing Two Proportions

with their appearance, and this difference of 10 people would not seem to be
significant (it could have arisen simply from random sampling variability).
c. Answers will vary, but one example is if 1000 men and 1000 women were
surveyed. In this case, you would not expect much sampling variability by chance,
and the observed difference would seem much more convincing of a difference in
the populations.
d. Here is the completed table:

Significant at:

Sample Size Satisfied Satisfied Test

p-value ␣  .10? ␣  .05? ␣  .01?
(each group) Women Men Statistic

100 71 81 1.67 .095 Yes No No

200 142 162 2.36 .018 Yes Yes No

500 355 405 3.73 .000 Yes Yes Yes

e. The larger the sample size, the more statistically significant a difference in
proportions will be. A large difference may not be significant with a small sample
size, but even a small difference may be statistically significant when the sample
size is large.

Activity 21-5: Graduate Admissions Discrimination

a. The null hypothesis is that the probability of admission to the graduate programs
at the University of California at Berkeley among male applicants is the same as
the probability of admission among female applicants. In symbols, H0: ␲m  ␲f .
The alternative hypothesis is that the probability of admission to the graduate
programs differs for male and female applicants. In symbols, Ha: ␲m ␲f .
The sample proportions are p̂m  .4457 and p̂f  .3046.
.4457  .3046
The test statistic is z  ____________________________
___________________________  9.56.
1 1
(.388)(1  .388)
√ (
_____  _____
2681 1835 )
This test statistic is very large (the difference between the acceptance rates is more
than 9 standard deviations), so the p-value is 2  Pr(Z  9.56), which is very
close to 0.
Because the p-value is small, reject H0 at any commonly used level of significance.
You have extremely strong statistical evidence that the probability of admission
for males differs from the probability of admission for females, much more than
can be accounted for by random chance alone.
b. No, this is not evidence of discrimination. This sample was not a randomized
experiment, so there could be other explanations for the observed difference.
Recall that this is an example of Simpson’s paradox: The women tended to apply
to the graduate programs with low acceptance rates, whereas the men tended
to apply to the graduate programs with high acceptance rates. Within each
individual program, women were accepted for admission at roughly the same or
higher rates than the men were.
 Activity 21-6 447

21
Activity 21-6: Nicotine Lozenge
a. The explanatory variable is the type of lozenge (nicotine or placebo). The response
variable is whether the smoker successfully abstains from smoking for the year. Both
variables are categorical and binary.
b. This is an experiment because the researchers assigned the subjects to take a
particular kind of lozenge (nicotine or placebo).
c. The null hypothesis is that the nicotine lozenge is no more (or less) effective than
the placebo, where effectiveness is measured by the proportion of smokers who
successfully abstain from smoking for a year. The alternative hypothesis is that the
nicotine lozenge is more effective than the placebo, meaning a higher proportion
of smokers (who are interested in and might potentially use such a product)
would successfully quit with a nicotine lozenge than with a placebo lozenge. In
symbols, the hypotheses are H0: ␲nicotine  ␲placebo vs. Ha: ␲nicotine  ␲placebo, where
␲ represents the population proportion of smokers who successfully abstain for a
year if given the nicotine lozenge or the placebo.
d. The 2  2 table is shown here:

Nicotine Lozenge Placebo Lozenge Total

Successfully Abstained 82 44 126

Resumed Smoking 377 414 791

Total 459 458 917

The following segmented bar graph shows those smokers taking the nicotine
lozenge had a higher success rate (proportion) in this study than those smokers
taking the placebo lozenge, almost twice as high (.179 vs. .096). But the graph
also reveals that in both groups, many more smokers resumed smoking than were
able to abstain successfully.

100
Resumed smoking
90
Successfully abstained
80
70
Percentage

60
50
40
30
20
10
0
Nicotine Placebo
Type of Lozenge

e. The sample proportions of smokers who successfully abstained from smoking in

each group are
82  .179,
p̂nicotine  ____ 44  .096
p̂placebo  ____
459 458
448 Topic 21: Comparing Two Proportions

The combined sample proportion of smokers who abstained is

82  44  ____
p̂c  _________ 126  .137
459  458 917
The test statistic is
.179  .096
z  __________________________
_________________________  3.65
1
1  ____
√(.137)(1  .137) ____
459 458 ( )
The p-value is the area to the right of 3.65 under the standard normal
curve, which Table II reveals to be .0001. This test is valid because random
assignment was used to put subjects in groups and because the sample size
condition is also met: 459(.137)  62.88 is greater than 5, as are 458(.137) 
62.75, 459(1 .137)  396.12, and 458(1  .137)  395.25.
This very small p-value indicates the experimental data provide very strong
evidence against the null hypothesis, which means very strong evidence that the
population proportion of smokers who would successfully abstain from smoking
is higher with the nicotine lozenge than with the placebo lozenge. Because this
was a randomized experiment, you can conclude the nicotine lozenge causes
an increase in the proportion of smokers who would successfully abstain as
compared to the placebo group. You are not told how the subjects were selected
for the study, but you do know that they were hoping to quit smoking, so this
conclusion should be limited to smokers hoping to quit. They were probably
chosen from a particular geographic area, so you might not want to generalize
beyond that area, but it is hard to imagine why smokers in one area would
respond differently to these lozenges than smokers in another area.
f. A 95% confidence interval for the difference ␲nicotine  ␲placebo is
_______________________________
(.096)(1  .096)
(.179)(1  .179) ______________
(.179  .096)  1.96
√ ______________
459

458
which is .083  1.96(.023), which is .083  .044, which is the interval from .039
through .127. The researchers are 95% confident that the proportion of smokers
using a nicotine lozenge who successfully quit would be higher than the successful
proportion of smokers using a placebo lozenge by somewhere between .039 and
.127. This interval procedure is valid because of random assignment and because
the number of successes and failures in both groups exceeds 5 (the smallest of
these numbers is 44 successes in the placebo group) with the sampling issues
cautioned about in part e.
g. Notice this question calls for a confidence interval for a single proportion: ␲nicotine ,
so you need to use the procedure from Topic 16:
________________
p̂nicotine( 1  p̂nicotine )
p̂nicotine  z* √ _______________
nnicotine
which is .179  1.96(.018), which is .179  .035, which is the interval from .144
through .214. The researchers can be 95% confident that if the population of all
smokers who wanted to quit were to use the nicotine lozenge, the proportion who
would successfully abstain from smoking for one year would be between .144 and
.214. So, even though the experiment provides strong evidence that the nicotine
lozenge works better than a placebo, most smokers (more than 34) would be
unable to quit even with the nicotine lozenge.
472 Topic 22: Comparing Two Means

Solutions
••• In-Class Activities
Activity 22-1: Close Friends
a. This is an observational study because the researcher simply observed the gender
of each subject—he/she did not randomly assign gender to subjects.
b. Explanatory: gender Type: binary categorical
Response: number of “close friends” Type: quantitative
c. The null hypothesis is men and women tend to mention the same average number
of close friends in response to this question. In symbols, H0: ␮m  ␮f .
The alternative hypothesis is men and women differ in the average number of close
friends they tend to mention in response to this question. In symbols, Ha: ␮m ␮f .
d. The two-sample z-test procedure from Topic 21 does not apply in this situation
because you are testing hypotheses about population means rather than population
proportions. The one-sample t-test procedure from Topic 20 does not apply
because you are comparing two means in this situation, not testing a claim about
a single mean.
e. These values are statistics because they describe samples, not populations.
f. The following boxplots compare the distributions of the number of close friends:

Male

Female

0 1 2 3 4 5 6
Number of Close Friends

g. There are very few differences in the distributions of the number of close friends
mentioned between males and females. The males have a slightly lower mean,
median, and lower quartile, but the upper quartiles and maximums are identical
to those of females.
h. Yes, it would be possible to obtain sample means this far apart even if the
population means were equal.
1.861  2.089  2.45.
i. The test statistic is t  _______________
______________
1.777 2  _____
______ 1.76 2
√ 654 813
j. Using 500 degrees of freedom, 2.586  2.45  2.334, so 2  .005 
p-value  2  .01, which means .01  p-value  .02.
k. The correct interpretation of the p-value is “The p-value is the probability of
getting sample data so extreme if, in fact, males and females have the same mean
number of close friends in the populations.”
l. Yes, this p-value is small enough to reject the null hypothesis at the ␣  .05
significance level (p-value  .02  .05).
 Activity 22-2 473

m. No, the observed difference in sample means is not statistically significant at the
␣  .01 significance level (p-value  .01).
n. Technical conditions:

22
i. The data are a random sample broken into two distinct groups (see page 418).
ii. Both sample size are large (greater than 30).
Because both sample sizes are large, you do not need to worry about the strong
skewness in the sample data. It does not provide any reason to doubt the validity
of this test.
o. For a 95% CI for ␮f  ␮m with 500 degrees of freedom, you calculate (2.089 
1.861)  (1.965)(0.0929)  (0.045, 0.411). This interval is entirely positive (and
does not include zero), which means you can be 95% confident that the mean
number of close friends that women have is between .045 and .411 greater than
the mean number of close friends that men have.
p. Here are the applet results:

Activity 22-2: Hypothetical Commuting Times

a. No, one route does not always get Alex to school more quickly than the other.
b. Yes, the data suggest that Route 1 tends to get Alex to school more quickly than
Route 2.
c. Here is the completed table:

Sample Size Sample Mean Sample SD p-value

Alex: Route 1 10 28 6
.155
Alex: Route 2 10 32 6

d. H0: ␮1  ␮2 Ha: ␮1 ␮2
28  32  1.49.
The test statistic is t  _________
________
2
6 62
___  ___
√ 10 10
474 Topic 22: Comparing Two Means

Using Table III with 9 degrees of freedom, 1.833  1.49  1.383, so

2  .05  p-value  2  .1, which means .1  p-value  .2. Using Minitab
with 18 degrees of freedom, the p-value is .154. Using the applet (which uses
9 degrees of freedom), the p-value is .1702.
e. No, Alex’s data are not statistically significant at any of the commonly used
significance levels. Alex cannot reasonably conclude that one route is faster than the
other route for getting to school because his p-value is not small and will not allow you
to reject the null hypothesis that the average time required for both routes is the same.
f. Using Minitab, (28  32)  t18*(2.683)  (8.65, 0.65). Using the applet or
Table III, t* with 9 degrees of freedom is 1.833, and the resulting confidence
interval is (8.92, 0.92).
g. Yes, this interval includes the value zero. This means that zero is a plausible value
for the difference in the population means or that you cannot conclude there is a
difference in the population mean travel times (with 90% confidence).
h. The most important difference between Alex’s and Barb’s results appears to be
their mean travel times. The difference between the mean travel times on Barb’s
two routes appears to be significantly greater than that of Alex.
i. Carl’s travel times appear to have a much smaller standard deviation than Alex’s.
j. Donna appears to have used a much larger sample size than Alex did.
k. Here is the completed table:

Sample Size Sample Mean Sample SD p-value

Barb: Route 1 10 25 6
.002 (Minitab)
Barb: Route 2 10 35 6 .0047 (applet)

Carl: Route 1 10 28 3
.008 (Minitab)
Carl: Route 2 10 32 3 .0154 (applet)

Donna: Route 1 40 28 6
.004 (Minitab)
Donna: Route 2 40 32 6 .0049 (applet)

l. Barb: The difference in the sample means (35  25) is large.

Carl: The sample standard deviations (3) are small.
Donna: The sample sizes are large (40).

Activity 22-3: Memorizing Letters

a. The observational units are the students in your class.
b. Explanatory: grouping of letters Type: binary categorical
Response: number of letters correctly memorized Type: quantitative
c. This study involves random assignment to two experimental groups. The students
were randomly assigned to the convenient three-letter chunk group or to the
inconvenient grouping group.
d. The null hypothesis is that the mean number of letters correctly memorized under
both conditions is the same. In symbols, H0: ␮JFK  ␮JFKC.
 Activity 22-3 475

The alternative hypothesis is that mean number of letters correctly memorized

under the three-letter grouping condition is greater than the mean number
of letters correctly memorized under the inconvenient grouping condition. In

22
symbols, Ha: ␮JFK  ␮JFKC.
Answers will vary by class. The following is one representative set of answers.
e. The following visual displays display the data:

JFK
4 8 12 16 20 24 28
Group

JFKC
4 8 12 16 20 24 28
Number of Correctly Memorized Letters

JFK

JFKC

0 5 10 15 20 25 30
Number of Correctly Memorized Letters

Yes, these data appear to support the conjecture that those who receive the letters
in convenient three-letter chunks tend to correctly memorize more letters. The
center of this plot is about six letters higher than for the JFKC plot, whereas the
spreads are not very different.
f. The data arise from random assignment of subjects to two treatment groups, but the
sample sizes are not quite large enough (n  27 and 26, respectively). Probability
plots indicate that the JFKC (inconvenient chunks) distribution is likely to be
normal, but the JFK (three-letter chunk) distribution is skewed left and not normal.

99

95
90

80
70
Percentage

60
50
40
30
20

10
5

1
0 10 20 30 40
JFK
476 Topic 22: Comparing Two Means

99

95
90

80
70
Percentage 60
50
40
30
20

10
5

1
10 0 10 20 30
JFKC

Therefore, for these data, the results of the test of significance should be
interpreted with caution.
16  10.88  3.03.
g. The test statistic is t  ______________
_____________
6.45 2  _____
_____ 5.86 2
√27 26
Using Table III with 25 degrees of freedom, 2.787  3.03  3.450, so
.001  p-value  .005. Using Minitab, the p-value is .002. Using the applet,
the p-value is .0028.
With a p-value less than .05, reject H0. You have strong statistical evidence that
the average number of letters correctly memorized is higher under the convenient
three-letter condition than under the inconvenient condition. If there were no
difference in the average number of letters correctly memorized under these two
conditions, you would see a result as extreme or more extreme than in this study
by chance alone in only about .2% of random assignments.
h. The confidence interval formula is: (16  10.88)  t*(1.69). Using Table III
with 25 degrees of freedom, t*  1.708, so the confidence interval is (2.23,
8.01). Using Minitab, the confidence interval is (2.28, 7.95). Using the applet,
the interval is (2.23, 8.01). You are 90% confident that the convenient three-
letter grouping helps individuals memorize between 2.3 and 8.0 more letters, on
average, than the inconvenient grouping.
i. Because this is a well-designed experiment in which the subjects were randomly
assigned to the two groups, you can conclude that the convenient three-letter
grouping did cause higher memory scores on average.

Activity 22-4: Got a Tip?

a. This is a randomized experiment because the waitress randomly assigned the
two-person parties either to be told her name or not as part of her greeting. You
do have to consider that the waitress was not blind to the treatment condition and
may have subconsciously provided better service for the customers she expected to
give her a larger tip.
 Activity 22-4 477

b. The explanatory variable is whether the waitress included her name as part of
her greeting; this variable is categorical and binary. The response variable is the
amount of the tip, which is quantitative.

22
c. The null hypothesis is that there is no effect of the waitress using her name in
her greeting. In other words, the null hypothesis says that the population mean
tip amount would be the same when she uses her name as when she does not.
The alternative hypothesis is that using her name has a positive effect, that the
population mean tip amount would be greater when she uses her name than when
she does not. In symbols: H0: ␮name  ␮no name vs. Ha: ␮name  ␮no name.
d. A Type I error would mean that the waitress decides that using her name helps
when it really doesn’t, so she would waste the minimal effort of giving her name
and reap no benefit from it. A Type II error means the waitress decides using her
name is not helpful even though it actually is helpful, so she would not bother
to give customers her name and would lose out on that benefit. Because the cost
of giving her name is minimal, losing out on potential tips is probably more of a
concern.
5.44  3.49
e. The test statistic is t  ________________
_______________ 1.95  4.19.
 _____
(1.75) 2
(1.13) 2 0.466
√ ______
20
 ______
20
Using Table III with 19 degrees of freedom reveals that this test statistic is off the
chart, so the p-value is less than .0005.
f. Because the p-value is less than .05, reject the null hypothesis at the ␣  .05 level.
Indeed, you would also reject the null hypothesis at the ␣  .01 and even at the
␣  .001 levels. The data provide very strong evidence that giving her name as
part of her greeting does lead to higher tips on average.
g. You do not have enough information to check the technical conditions
thoroughly. You do know that the parties were randomly assigned to one group or
the other. But the sample sizes are not large, so you should check whether the tip
data could reasonably have come from normal distributions; however, you only
have the summary statistics, not the actual tip amounts from each party, so you
cannot check this condition. You should ask the waitress to provide the actual
party-by-party tip amounts to help you assess the shape of the distribution of tip
amounts.
h. A 95% confidence interval for ␮name  ␮no name is
_______________
(1.75) 2 (1.13) 2
√
(5.44  3.49)  2.093 ______  ______
20 20
which is 1.95  2.093(0.466), which is 1.95  0.97, which is the interval (0.98,
2.92). You can be 95% confident that the waitress would earn between $0.98 and
$2.92 more per party, on average, with a $23.21 check, by giving her name as part
of her greeting (assuming that the tip amounts are roughly normally distributed).
i. Conclude a causal link between the waitress giving her name and receiving
higher tips on average. Random assignment would have assured that the only
difference between the groups was whether the party was given the waitress’s
name. Because the group who was told her name gave significantly higher tips
on average (p-value  .0005), you can attribute that to being told her name in
greeting unless the waitress gave better service to customers to whom she gave
478 Topic 22: Comparing Two Means

her name. The confidence interval enables you to say more: that giving her
name to customers increases the waitress’ tips by an average of about
1–3 dollars per dining party at Sunday brunch in this restaurant. But you
must be cautious about generalizing this result to other waitresses because only
one waitress participated in this study. Even for this particular waitress, you
should be cautious about generalizing the results to customers beyond those
who partake of Sunday brunch at that particular Charlie Brown’s restaurant
in southern California. You should also remember that these p-value and
confidences interval calculations are only valid if the tip amounts roughly
follow a normal distribution.

••• Homework Activities

Activity 22-5: Close Friends
a. If all of the women sampled had one more friend, the mean for the women would
increase by one, and all other statistics would remain the same. This would
increase the absolute value of the test statistic (the difference between the sample
means would increase, but the denominator would not change) and thus would
decrease the p-value.
b. If all of the men sampled had one more friend, the mean for the men would
increase by one, and all other statistics would remain the same. This would
increase the absolute value of the test statistic (the difference between the sample
means would increase, but the denominator would not change) and thus would
decrease the p-value.
c. If every man and every woman sampled had one more close friend than they
originally reported, then the sample means for the men and for the women
would increase by one. The sample standard deviations and sample sizes
would not change. The difference between the sample means would not
change, so the test statistic value would not change, and thus the p-value
would not change.
d. If both sample standard deviations were larger, the denominator of the test
statistic would be larger, and thus the test statistic would be smaller. This would
make the p-value larger.
e. If both sample sizes were larger, the denominator of the test statistic would be
smaller, and thus the test statistic would be larger (in absolute value). This would
make the p-value smaller.

Activity 22-6: Hypothetical Commuting Times

Earl: not statistically significant because the difference in his means is less than the
difference in Alex’s means, and his sample sizes and sample standard deviations are the
sample as Alex’s.
Fred: statistically significant because his sample sizes are much larger than Donna’s,
and his sample means and standard deviations are the same
Grace: statistically significant because the difference in her sample means is greater
than that of Barb, and her sample sizes and sample standard deviations are the same
Harry: not statistically significant because his sample standard deviations are larger
than those of Alex whereas his sample sizes and sample means are the same
Ida: statistically significant because her sample standard deviations are less than
those of Carl whereas her sample sizes and sample means are the same
 Activity 23-1 503

3. Rather than finding the mean and standard deviation of the two lists, select Data
for Inpt and insert HUSB as List1 and WIFE as List2 .
4. Leave the frequencies as 1 and Pooled as No.
5. Select Calculate and press Õ.
6. As an alternative, use the Draw feature of the 2-SampTTest.

23
Solutions
••• In-Class Activities
Activity 23-1: Marriage Ages
a. The null hypothesis is that the population mean marriage age for husbands is the
same as the mean marriage age for their wives. In symbols, H0: ␮husbands  ␮wives.
The alternative hypothesis is that the population mean marriage age for husbands is
greater than the mean marriage age for their wives. In symbols, Ha: ␮husbands  ␮wives.
35.71  33.83  0.46.
The test statistic is t  ________________
_______________
14.56 2  ______
______ 13.56 2
√ 24 24
Using Table III with 23 degrees of freedom, 0.46  0.858, so the p-value is off
the chart. This means the p-value  .20. Using Minitab, the p-value is .323.
Using the applet with 23 degrees of freedom, the p-value is .3226.
Do not reject H0 at the ␣  .05 significance level (.32  .05). You do not have
statistical evidence that the population mean marriage age for husbands is greater
than for wives.
b. For a 90% confidence interval: (35.71  33.83)  t*(4.06). Using Table III with
23 degrees of freedom, t*  1.714, so the confidence interval is (5.08, 8.84).
Using Minitab, the interval is (4.95, 8.70). Using the applet, the interval is
(5.08, 8.84).
c. Most of the values are above the y  x line, which indicates a clear tendency for
husbands to be older than their wives.
d. The husband’s and wife’s ages appear to be related. People tend to marry people
in the same age group: younger people tend to marry younger people, and older
people tend to marry older people.
e. For a 90% CI with ___23 degrees of freedom, you calculate 1.875 
(1.714) ( 4.812√ 24 )  1.875  .98225  (.1914, 3.5586). You are 90%
confident the population mean difference in ages between husbands and
their wives is between .19 and 3.6 years.
f. This interval is entirely positive, indicating that there is a difference in the
population mean ages of husbands and their wives. The incorrect interval in part
b contained both positive and negative values (and zero). The midpoint of this
interval is 1.875 years and its width is 3.367 years. The midpoint of the previous
(incorrect) interval was also 1.875 years, but the width was 13.65 years.
g. The null hypothesis is that the population mean difference in ages between
husbands and their wives is zero. In symbols, H0: ␮d  0. The alternative
504 Topic 23: Analyzing Paired Data

hypothesis is that the population mean difference in ages between husbands and
their wives is greater than 0. In symbols, Ha: ␮d  0.
1.875___  1.91.
The test statistic is t  __________
4.812√ 24
Using Table III with 23 degrees of freedom, 1.714  1.91  2.069, so .025 
p-value  .05. Using the applet with 23 degrees of freedom, the p-value is .0344.
Reject H0 at the ␣  .05 significance level (.0344  .05). You have moderate
statistical evidence the population mean difference in ages between husbands and
their wives is greater than zero.
h. This is the opposite of the (incorrect) conclusion you drew in part a.
i. Technical conditions: This is not a simple random sample, but you have no
reason to suspect that this sample is not representative of at least marriages in
Cumberland County, Pennsylvania. The sample size of pairs is not large (n 
24  30), but a probability plot of the sample differences indicates that it is
plausible the sample comes from a normally distributed population.

99

95
90
80
70
Percentage

60
50
40
30
20
10
5

1
10 5 0 5 10 15 20
Difference (in years)

j. The paired analysis produces such a different conclusion from the independent-
samples analysis because it reduces the variability so much. The standard
deviation of the differences in ages is much smaller than the standard deviations
of the ages. This makes the observed difference in the mean ages seem larger (less
likely to happen by chance). You can also see that this reduces the denominator of
the test statistic creating a larger value and lowering the p-value.
k. Yes, the researcher was wise to gather paired data. Because the husband’s age and
the wife’s age were related, pairing was very helpful for estimating the population
mean age difference among married couples.

Activity 23-2: Chip Melting

a. Explanatory: type of chip (chocolate or peanut butter) Type: binary categorical
Response: time (in seconds) it takes the chip to melt Type: quantitative
b. This is an experiment. The researchers are imposing the chips and randomly
deciding the order of the chips.
 Activity 23-2 505

c. Answers will vary by class. The following is one representative set:

Chocolate Chip
56 70 84 98 112 126 140

Peanut Butter Chip

56 70 84 98 112 126 140

23
Melting Time (in seconds)

Chocolate Chip

Peanut Butter Chip

50 75 100 125 150

Melting Time (in seconds)

_
n x s Min QL Median Qu Max
Chocolate Chip 16 79.9 23.3 55.0 56.3 77.5 95.0 140
Peanut 16 93.8 27.5 45.0 71.3 100.0 113.8 148
Butter Chip

It appears that for these students, peanut butter chips tended to take a little longer
than chocolate chips to melt (mean 93.8 vs. 79.9 seconds; almost all five numbers
in five number summary are larger).
d. The null hypothesis is that the population mean melting time for both chips
is the same. In symbols, H0: ␮chocolate  ␮peanut butter. The alternative hypothesis is
that the population mean melting time for both chips is not the same. In symbols,
Ha: ␮chocolate ␮peanut butter.
Technical conditions: The data are from a completely randomized experiment.
However, the sample sizes are not large (16 and 16), but probability plots do not
provide strong evidence that the data come from nonnormal distributions (though
there are some outliers with the chocolate chip melting times).

99

95
90

80
70
Percentage

60
50
40
30
20

10
5

1
20 40 60 80 100 120 140
Chocolate Chip Melting Time (in seconds)
506 Topic 23: Analyzing Paired Data

99

95
90

80
70
Percentage 60
50
40
30
20

10
5

1
20 40 60 80 100 120 140 160
Peanut Butter Chip Melting Time (in seconds)

79.9  93.8  1.54.

The test statistic is t  ______________
_____________
27.5 2
23.3 2  _____
√
_____
16 16

Using Table III with 15 degrees of freedom, 1.341  1.54  1.753, so

2  .05  p-value  2  .1. This means .1  p-value  .2. Using Minitab, the
p-value is .135. Using the applet with 15 degrees of freedom, the p-value is .1438.
Because the p-value is not small, do not reject H0 at any common significance
level. You do not have statistical evidence that the population mean melting time
differs between the two kinds of chips.
e. These data call for a paired t-test because the chips are paired-up—one chip of
each type was given to each student, and each student tested the melting time of
both chips. The observations in each row are clearly related—they are from the
same individual!
f. Here is the numerical summary and graphical display:
_
n x s Min. QL Median QU Max.
Differences (Chocolate 32 8.8 22.0 55.0 28.0 9.0 8.8 30.0
– Peanut Butter)

48 36 24 12 0 12 24

Difference in Melting Times (in seconds)

The null hypothesis is that the population mean difference in melting times
between chocolate and peanut butter chips is 0. In symbols, H0: ␮d  0. The
alternative hypothesis is that the population mean difference in melting times
between chocolate and peanut butter chips is not 0. In symbols, Ha: ␮d 0.
8.8___  2.27.
The test statistic is t  _________
22.0√ 32
 Activity 23-4 507

Using Table III with 31 degrees of freedom, 2.040  2.27  2.453, so 2 

.01  p-value  2  0.025. This means .02  p-value  .05. Using Minitab, the
p-value is .030. Using the applet with 31 degrees of freedom, the p-value is .0311.
Because the p-value .030 is less than .05, reject H0 at the ␣  .05 significance
level.
A 90% confidence interval for the mean difference in the melting time is
___

23
8.8  (1.696)( 22.0√ 32 )  (15.4, 2.2)
Because the p-value of .03 is less than .05, you have some statistical evidence that
there is a difference in the mean melting time between the two kinds of chips.
You are 90% confident the melting time of the peanut butter chips is between 2.2
and 15.4 seconds longer than that of the chocolate chips, on average.
g. The conclusions for these two studies were quite different. In the first study, you had
no statistical evidence of a difference in the chip melting time (p-value  .135), and
in the paired data study, you had some evidence of a difference (p-value  .03).
Pairing the data reduced the variability in the data, as the standard deviation of
the differences was 22.0 seconds, whereas the standard deviations of the individual
melting times were both more than 23 seconds.

Activity 23-3: Body Temperatures,

Natural Selection, and Memorizing Letters
a. No, it would not be appropriate to analyze these data with a paired t-test because
there is no link between the men and women in this study. You could change the
order of the men or the women (or both) without creating a problem in your data.
b. It would be impossible to analyze these data with a paired t-test because the
sample sizes for the two groups are not the same. You could not pair up each
surviving sparrow with a dead sparrow because more sparrows survived than died.
c. No, it would not be appropriate to analyze these data on memorizing letters with
a paired t-test because there is no link between the students who memorized the
JFK chunks and the students who memorized the JFKC chunks. Randomization
was used to assign subjects to the two separate groups, so you should use the two-
sample t-procedures instead.
d. You could alter this experiment by having each student memorize the letters
twice—once in the convenient three-letter (JFK) chunks and once in the
inconvenient (JFKC) chunks. Randomization would still be important in this
experiment because you would need to determine the order in which the student
receivedmemorized the type of chunk randomly.

Activity 23-4: Alarming Wake-Up

a. The explanatory variable is the sound used to wake the child.
b. Use randomization to divide the children into two groups—one group to be
awakened by recordings of their mothers’ voices (personalized alarm) and the
other group to be awakened by a conventional smoke alarm. Record the time it
takes to wake each child. Use a two-sample t-test to determine whether there is a
statistically significant difference in the average times of the two groups.
508 Topic 23: Analyzing Paired Data

c. The null hypothesis is that the population mean time required to wake up to the
conventional smoke alarm is the same as the population mean time required to
wake up to the personalized alarm. In symbols, H0: ␮conventional  ␮personalized.
The alternative hypothesis is the population mean time required to wake up to
the conventional smoke alarm is greater than the population mean time required
to wake up to the personalized alarm. In symbols, Ha: ␮conventional  ␮personalized.
d. Wake each child on two different nights—once with the personalized alarm and
once with the conventional smoke alarm. Use randomization to decide the order
of the alarms. Record the time required for the child to wake each time, and then
calculate the difference in the times (for instance, conventional  personalized) for
each child. Use the one-sample t-test to determine whether the average of these
differences is greater than zero.
e. Let ␮d represent the average of the population differences in the times required to
wake a child (conventional alarm  personalized alarm).
H0: ␮d  0 vs. Ha: ␮d  0
f. You should recommend the matched-pairs design because it would have less
variability. The times required to wake the children might vary quite a bit, but
the difference in the times required to wake the children using the two methods
should vary much less.

Activity 23-5: Muscle Fatigue

a. The researchers expected that strength would be related to time until fatigue;
therefore, matching men and women based on strength would result in less
variability in the time differences than in the individual times. This reduction
in variability allows for a more powerful test, one that is better able to detect a
difference in fatigue times between the genders if a difference really exists.
b. The explanatory variable is gender, which is categorical and binary. The response
variable is time until fatigue, which is quantitative.
c. This is an observational study. Subjects were not assigned to a gender.
d. The differences in times until fatigue, subtracting the man’s time from the
woman’s time in each pair, are 890, 706, 3584, 16, 58, 335, 322, 1817,
1484, 110. A dotplot follows.

0 600 1200 1800 2400 3000 3600

Difference (Woman's Fatigue Time – Man's Fatigue Time, in seconds)

Three of these differences are negative, and seven are positive. Therefore, in most
pairs, the woman lasted longer until fatigue set in than the man did. The mean
of these differences is 895 seconds, so the women in the sample outlasted the men
by almost 15 minutes on average. The standard deviation is 1148 seconds. The
distribution of differences appears to be a bit skewed to the right.
e. Because of the small sample size (n  10 pairs), the distribution of differences
must be approximately normal in order for the technical conditions to be satisfied.
But you have already seen that the distribution is a bit skewed to the right. This
skewness is also seen in a normal probability plot, which does not look very linear:
 Activity 23-5 509

99

95
90

80
70

Percentage
60

23
50
40
30
20

10
5

1
4000 3000 2000 1000 0 1000 2000 3000 4000 5000
Difference (in seconds)

You also lack either a randomly selected sample or random assignment to treatment
groups, and therefore the paired t-test conditions were not met. We should stop
the analysis at this point, but we will continue with the calculations for the sake of
practice, remembering that we cannot take the results very seriously.
f. You are testing H0: ␮d  0 vs. the alternative Ha: ␮d 0, where ␮d represents the
population mean of the differences in times until fatigue between healthy young
adult men and women. The test statistic is
895 ___ 2.465
t  _________ 
1148√ 10
The p-value, from Table III (t-Distribution Critical Values) with 9 degrees of
freedom and remembering the alternative hypothesis is two-sided, is between
2(.01) and 2(.025), which means the p-value is between .02 and .05. Technology
gives a p-value of .036. This p-value is fairly small, so you would reject the null
hypothesis at the ␣  .05 level. The sample data provide fairly strong evidence
that the mean time until fatigue differs between men and women.
g. A 90% confidence interval for ␮d is
1148
895  1.833 _____
___
√ 10

which is 895  665.4, which gives the interval (229.6, 1560.4). This interval means
you can be 90% confident that women take between 230 seconds and 1560 seconds
longer, on average, to reach fatigue than men do (between 426 minutes).
h. This is not a randomized experiment, so even if the test were valid, you would
not be able to attribute the longer time to fatigue for women to any particular
cause. Moreover, because the subjects volunteered, you should be cautious about
generalizing the results even to all healthy young adults in the area of the study.
Finally, the technical conditions of the paired t-test were not met because of the
small sample size and skewed distribution of differences, so you cannot take the
inference results very seriously. Researchers might want to repeat the study with
large sample sizes.
 Activity 24-1 529

6. For the built-in calculator  2 GOF-Test, you can explore the Draw feature as an
alternative. (The CHIGOF program does not have the Draw feature.)

Activity 24-12: Wayward Golf Balls (Fathom)

In this activity, students use technology to simulate a sample of 486 golf balls where
the numbers on the balls are randomly chosen from among the digits 1, 2, 3, and 4.
A goodness-of-fit test for such a simulated sample shows what should be expected when
the null hypothesis of equally likely proportions is true. Collecting the results for many
simulated samples shows the behavior of the test statistic and p-value.
To create the simulated sample and goodness-of-fit test with Fathom, follow
these steps:

24
1. Create a new (empty) collection, and choose Collection | New Cases to add
486 cases to the collection. Drag a case table from the shelf to view the (empty)
collection.
2. Name the first attribute in the collection Ball and use the formula editor to enter
the formula RandomPick(1,2,3,4) to create a sample in this column of ball
numbers where each of the four digits is equally likely.
3. Drag a test box from the shelf and change the type of test to Goodness of Fit.
Drag the Ball attribute to the top slot, labeled “Attribute (categorical),” to produce
the goodness-of-fit test results needed to answer parts a–c of Activity 24-12.
To collect the results for tests on 1000 such simulated samples in Fathom, continue
with these steps:
4. Select the test box and choose Test | Collect Results as Measures from the main
Fathom menu to create a new collection with the results (including the chi-square
statistic and p-value) from five simulations. Drag a new table from the shelf to
view these results.
5. Double-click the new Measures from Test of Collection 1 box to show its
inspector. On the Collect Measures tab, uncheck the “Animation on” box
to turn off animation, check Replace existing cases, and change the number
of measures to 1000. Click Collect More Measures to perform the 1000
simulations and tests.
6. To obtain the count for part f of the simulated test statistics that exceed the test
statistic from the original data in Activity 24-11, select the test statistic attribute
in the new Measures collection and choose Table | Sort Descending from the
main Fathom menu to reorder the 1000 simulated p-values. Compare this count
out of 1000 to the p-value for the original test.

Solutions
••• In-Class Activities
Activity 24-1: Birthdays of the Week
a. Observational units: writers
Variable: days of the weeks on which the birthdays fall Type: categorical
530 Topic 24: Goodness-of-Fit Tests

b. The following bar graph displays the data:

Birthdays of the Week

.20
.18
.16
.14

Percentage
.12
.10
.08
.06
.04
.02
0
Monday Tuesday Wednesday Thursday Friday Saturday Sunday
Day of Week

There does not appear to be a great difference in the proportion of people born on
any given day of the week.
c. These values ( ␲Tu, ␲We, . . . , ␲Su ) describe the proportion of all people who were
born on Tuesday, Wednesday,… Sunday. The values are parameters because they
describe an entire population.
d. The values are ␲M  ␲Tu  …  ␲Su  17  .1429.
e. You would expect the count of Monday birthdays to be (17)  147  21.
f. See the middle row of the following table.

Mon Tues Wed Thu Fri Sat Sun Total

Observed Count (O) 17 26 22 23 19 15 25 147

Expected Count (E) 21 21 21 21 21 21 21 147

(O  E)
________
2
.7619 1.1905 .0476 .1905 .1905 1.7143 .7619 4.857
E

g. See bottom row of the previous table.

h. You calculate X 2  .7619  1.1905  . . .  .7619  4.857.
i. Large values of the test statistic constitute evidence against the null hypothesis.
Answers will vary about whether this test statistic provides convincing evidence.
The additional evidence you need is the p-value. You need to know how likely it is
that you would obtain a test statistic this large (or larger) by random chance alone
if all seven days of the week are equally likely to be a person’s birthday.
j. Using Table IV with 6 degrees of freedom, 4.587  8.56, so the p-value is greater
than .2.
 Activity 24-2 531

.598

24
0 4.587

k. If all seven days of the week are equally likely to be a person’s birthday, the
probability that you would obtain a test statistic this large (or larger) by random
chance alone is at least .2.
l. With the larger p-value, do not reject H0 at the ␣  .10, ␣  .05, or ␣  .01
levels.
m. Yes; all the expected counts are 21, which is greater than 5. However, the random
sampling condition is not met. These are the birthdays of “noted writers of the
present,” which is not a random sample of the population of U.S. citizens.
n. You have no statistical evidence against the null hypothesis that the seven days of
the week are all equally likely to be a person’s birthday (at least for the population
of famous writers).

Activity 24-2: Birthdays of the Week

a. See middle row of the following table.

Mon Tues Wed Thu Fri Sat Sun Total

Observed 17 26 22 23 19 15 25 147
Count

Expected 24.5 24.5 24.5 24.5 24.5 12.25 12.25 147

Count

(O  E)
________
2
2.2959 .0918 .2551 .0918 1.235 .6173 13.270 17.857
E

b. The test statistic is X 2  17.857.

c. Using Table IV with 6 degrees of freedom, 16.81  17.857  18.55, so .005 
p-value  .01.
d. With the smaller p-value, reject H0 at the .01 level.
e. Sunday contributes the most to the test statistic (13.720). Its observed count is
greater than its expected count, which means you have many more births on
Sunday than you would expect if only 112 of the births occur on Sunday.
532 Topic 24: Goodness-of-Fit Tests

f. You have strong statistical evidence that the null hypothesis is not true; that is,
that one or more of the population proportions is not equal to the values that you
proposed in the null hypothesis. (So, one or more of ␲M  16, ␲Tu  16, ␲W 
16, ␲Th  16, ␲F  16, ␲Sa  112, ␲Su  112 is not true.) If all of these
were true, you would see a sample result (chi-square test statistic) as large or larger
by random chance alone somewhere between .5% and 1% of the time (of random
samples from such a population). Because this would occur so rarely by random
chance, you conclude the null hypothesis is false.

Activity 24-3: Birthdays of the Week

a. Answers will vary, but students should expect the p-value to be smaller because
the sample size is so much larger.
b. The sample proportions are the same as those in the actual data, so the bar graph
would be identical.
c. The test statistic is X 2  48.572 (ten times larger than before).
d. Using Table IV with 6 degrees of freedom, 24.1  48.572, so p-value  .0005.
With the small p-value, reject H0. You have extremely strong evidence that at least
one of the seven days of the week is more or less likely to be a person’s birthday (at
least for the population of writers).
e. This is a different conclusion and much smaller p-value than in Activity 24-1.
This makes sense because you have a much larger (ten times as large) sample size.

Activity 24-4: Kissing Couples

a. Observational units: 124 kissing couples
Variable: whether the couple leaned their heads to the right when they kissed
Type: binary categorical
b. Yes, the one-sample z-test from Topic 17 applies to these data.
c. Let ␲right refer to the proportion of all kissing couples who turn their heads to the
right.
H0: ␲right  .75, ␲left  .25
Ha: At least one of ␲right  .75, ␲left  .25 is not true. Equivalently: Ha: ␲right .75
(Note that when there are only two proportions, if one proposed value is false, the
other must also be false.)
d. Expected count for “leans right”  .75  124  93
Expected count for “leans left”  .25  124  31
The test statistic is X 2  (80  93)293  (44  31)231  1.8172  5.4516 
7.2688.
Using Table IV with 1 degree of freedom, 6.63  7.27  7.88, so .005  p-value
 .01. Using Minitab, the p-value is .007.
 Activity 24-5 533

e. Technical conditions: The expected counts are at least five in both categories
(smallest  31). The couples selected for the sample were those who happened
to be observed in public places while the researchers were watching. This is not a
random sample, so you should be cautious about generalizing the results of this
test to a larger population.
f. Considering these data to be a representative sample, you have very strong
statistical evidence that H0 is false. That is, you are confident ␲right .75.
g. You calculate (2.7)2  7.29  7.27. The X 2-test statistic appears to be
(roughly) the square of the z-test statistic. Without rounding the z-test statistic
value as much, z  2.696 and z 2  7.2688, and the two test statistics are
identical.

24
h. The p-values are roughly the same (.0069  .007). Without rounding, the values
are identical.

Activity 24-5: Leading Digits

a. A bar graph follows:

Leading Digits of Populations for Countries in the World

Proportion of Countries

.3
.25
.2
.15
.1
.05
0
1 2 3 4 5 6 7 8 9
Leading Digit

This graph does appear to be consistent with the Benford probabilities, insofar
as the most common leading digit in this sample is 1 by a wide margin, followed
by 2, and then gradually tapering off from there. Note: Even though the leading
digits are numbers, you can consider this variable to be categorical, because the
number simply separates the countries into groups. That’s why a bar graph is
appropriate rather than a histogram and also why a chi-square goodness-of-fit test
is appropriate.
b. The null hypothesis is H0: ␲1  .301, ␲2  .176, ␲3  .125, ␲4  .097, ␲5 
.079, ␲6  0.067, ␲7  .058, ␲8  .051, and ␲9  .046. In other words, the
null hypothesis says the Benford probability model is correct for the data. The
alternative hypothesis simply says this model is not correct, meaning at least one
of the hypothesized Benford probabilities is wrong. In symbols, you can write
Ha: at least one ␲i differs from its hypothesized value.
c. The expected count in the “1” category is 194(.301)  58.394 countries. This
is fairly close to the observed count of 55 countries with a leading digit of 1.
Calculating the expected counts for the other categories in the same way produces
these results:
534 Topic 24: Goodness-of-Fit Tests

Leading Digit Observed Count Expected Count

1 55 58.394

2 36 34.144

3 22 24.250

4 23 18.818

5 15 15.326

6 14 12.998

7 12 11.252

8 12 9.894

9 5 8.924

Total 194 194.000

All of these expected counts are greater than five, so that technical condition
is satisfied. If you consider the population to be all numbers appearing in
the almanac, you do not literally have a random sample of numbers from the
almanac, but these are likely to be representative of population values.
d. The contribution to the test statistic from the “1” category is
(55  58.394) 2
_____________  0.197
58.394
Calculating the other contributions in the same way produces these results:

(O  E)
________
2
Leading Digit Observed Count Expected Count
E
1 55 58.394 0.197

2 36 34.144 0.100

3 22 24.250 0.209

4 23 18.818 0.929

5 15 15.326 0.007

6 14 12.998 0.077

7 12 11.252 0.050

8 12 9.894 0.448

9 5 8.924 1.725

Total 194 194.000 3.742

The test statistic equals 3.742, the sum of the nine values in the last column. (You
might get a slightly different, more accurate answer if you carry more than
three decimal places in your intermediate calculations.) Looking at Table IV
with 8 degrees of freedom (one less than the nine categories), this test statistic
is way off the chart to the left. Therefore, the p-value is much greater than .2.
 Activity 24-6 535

Technology calculates the p-value to be about .88, as seen in the following

graph:

24
0 5 10 15 20 25 30
3.742
Chi-Square Values (df = 8)

e. This is a very large p-value, much greater than all common significance levels, so fail
to reject the null hypothesis that Benford’s probabilities adequately model these data.
f. The large p-value (based on the small value of the chi-square test statistic) from
this chi-square analysis reveals the sample data are extremely consistent with the
Benford probabilities. If Benford’s model were correct, it would not be the least
bit surprising to observe sample data as found with these nations’ populations.
In other words, the sample data provide no reason to doubt Benford’s model
describes the distribution of leading digits in the almanac.
g. An accountant or IRS agent could apply a chi-square test to see how well the
numbers in a tax return follow Benford’s probabilities. If the p-value turns out to
be very small, that suggests the leading digits of numbers on that tax return differ
substantially from what Benford’s probabilities predict. This certainly does not
prove the tax return is fraudulent, but it might suggest the numbers were made
up and not legitimate. A tax return whose leading digits do not follow Benford’s
probabilities might be worth a closer look.

••• Homework Activities

Activity 24-6: Mendel’s Peas
Define parameters of interest: Let ␲RY refer to the proportion of all peas that are round
and yellow, and so on.
H0: ␲RY  916, ␲WY  316, ␲RG  316, ␲WG  116
Ha: At least one of these hypothesized values is incorrect.
Check technical conditions: The expected counts are all at least five. You do
not know if these data are a random sample of peas, but they are most likely a
representative sample.

Round, Yellow Wrinkled, Yellow Round, Green Wrinkled, Green Total

Observed Count 315 101 108 32 556

Expected Count 312.75 104.25 104.25 34.75 556

(O  E)2
________ .016187 .101319 .13489 .217626 .4700
E
 Activity 25-1 553

3. Fill in the counts as they appear in the two-way table. The expected counts and
results for the chi-square test are updated as you enter the values.
4. (Optional) For more informative labeling of the categories or attributes, edit any
of the blue text or table headings (e.g., “FirstAttribute” or “RowCategory1”) to
better match the categories shown in the original table.

Activity 25-2: Pursuit of Happiness (Graphing Calculator)

In part b of this activity, students are asked to use their calculators to compute the
expected values, test statistic, and p-value. To do this, they can use the  2-Test. Before
using the  2-Test, however, the students need to set up a matrix containing the
observed counts:
1. Press 2ND [MATRIX].

2. Arrow over to the EDIT menu and press Õ to select matrix [A].

25
Note: The table at the beginning of this activity has three categories and three years, so
your matrix needs three rows and three columns (i.e., the dimensions should be 3  3).
3. Now enter the numbers from the table into the matrix, using the arrow keys to
move the cursor.
4. Once matrix [A] is completed, press Õ and arrow over to the TESTS menu and
select  2-Test.
5. Select CALCULATE and then press Õ. The expected values are automatically
entered into the matrix [B].

Solutions
••• In-Class Activities
Activity 25-1: Pursuit of Happiness
a. The observational units are the 4409 American adults who were surveyed.
b. Explanatory: year Type: categorical
Response: happiness level Type: categorical
c. This is an observational study because the researchers did not randomly assign the
subjects to the years.
d. A segmented bar graph is the appropriate graphical display for these data.
e. The proportions of Americans who considered themselves very happy in each of
these three years were very similar (about 30%).
f. The two-sample z-test from Topic 21 is not appropriate for testing whether
these population proportions differ significantly because there are three years to
consider—not just two.
g. H0: ␲1972  ␲1988  ␲2004, where ␲1972 represents the proportion of all American
adults who considered themselves happy in 1972, and so on.
h. For the three years combined, the proportion of respondents who were very happy
was 1403/4409 or .318.
554 Topic 25: Inference for Two-Way Tables

i. For 1972, you calculate .318  1606 or 511.05.

j. 1988: .318  1466  466.50; 2004: .318  1337  425.45
k. Here is the completed table:

1972 1988 2004 Total

486 498 419 1403

Very Happy
(511.05) (466.50) (425.45)

1120 968 918 3006

Less Than Very Happy
(1094.95) (999.50) (911.55)

Total 1606 1466 1337 4409

l. For the “less than very happy” people in 1988, you calculate (968  999.5)2/
999.5  .993.
m. The test statistic is X 2  5.065.
n. Large values of the test statistic would constitute evidence against the null
hypothesis that the three populations have the same proportions of very happy
Americans because if the proportions are the same, then the observed and
expected values in each year will be very similar, and each year’s contribution
to the test statistic will be small, so the overall test statistic value will be small.
However, if one or more of the proportions is not the same, then there will be a
large difference between the observed and expected values in that year, and this
difference will make a large contribution to the test statistic.
o. Using Table IV with (2  1)  (3  1)  2 degrees of freedom, 4.61  5.065 
5.99, so .05  p-value  .10.
p. These sample data are a bit unlikely to have occurred by chance alone if the
population proportions of very happy people had been identical for these three
years, but you would fail to reject the null hypothesis at the ␣  .05 level.
q. You have weak statistical evidence that the population proportions of very happy
people were not identical for these three years. (Because these were random
samples, you are safe in generalizing this conclusion to the populations of all
American adults in each year.)

Activity 25-2: Pursuit of Happiness

a. H0: The population distributions of happiness levels were the same in all
three years.
Ha: At least one year’s distribution of happiness levels differs from the others.
b. Using Minitab, the test statistic is X 2  35.655 with 4 degrees of freedom,
p-value  .000.
c. Yes, the sample data provide evidence that the distributions of self-reported
happiness levels differed among all adult Americans these three years, at the
.05 significance level.
 Activity 25-4 555

d. “1988/not too happy” has the largest contribution (16.927) to the test statistic.
The observed count (136) is less than the expected count (193.18).
“1972/not too happy” has the next largest contribution (13.458) to the test
statistic. The observed count (265) is greater than the expected count (211.63).
e. You have strong statistical evidence that the population distributions of happiness
levels were not the same for all three years. In 1972, the sample proportion of
those who were not too happy was substantially greater than you would expect if
the population proportions were equal in all three years, and in 1988, the sample
proportion of those who were not too happy was substantially less than you would
expect.

Activity 25-3: Got a Tip?

a. The observational units are the customers in the coffee bar.
b. Explanatory: type of card received Type: categorical

25
Response: whether the customer left a tip Type: binary categorical
c. This is an experiment because the waiter randomly assigned the customers to
treatment groups.
d. This study involves random assignment to groups.
e. H0: The population proportions of potential customers who would leave a tip is
the same regardless of the type of card they receive.
Ha: The population proportion of potential coffee bar customers who would leave
a tip is affected by the type of card they receive.
Using Minitab, the test statistic is X 2  9.953 with 2 degrees of freedom and the
p-value is .007.
As the p-value is .007, which is less than .10, reject H0 at the .10 significance level.
The “joke/left a tip” cell makes the largest contribution to the test statistic, and
the observed count is much greater than the expected count. So, you conclude
that potential coffee bar customers are not equally likely to leave a tip regardless
of the type of card they receive—leaving a joke card is more likely to generate a
tip than leaving an advertisement card or no card.
f. Because this was an experiment where the only difference between the groups
should be the type of card the customers received, you can conclude that the
joke card caused customers to be more likely to leave a tip. (Note also that the
technical conditions for this test to be valid were satisfied—the subjects were
randomly assigned to treatment groups, and the expected counts in each cell
of the table were at least five.)
g. You should not generalize your conclusion to all waiters and waitress in all food
and drink establishments. This experiment was carried out in only one coffee bar
with one waiter.

Activity 25-4: Pursuit of Happiness

a. Explanatory: political inclination Type: categorical
Response: self-reported happiness level Type: categorical
556 Topic 25: Inference for Two-Way Tables

b. In Activities 25-1 and 25-2, the researchers took independent random samples
from three populations (1972, 1988, and 2004). In this activity, the researchers
took one random sample and then classified the subjects by both explanatory and
response variables.
c. H0: Political inclination and self-reported happiness level are independent in the
population of adult Americans.
Ha: Political inclination is related to self-reported happiness level.
Here is a bar graph:

Political Inclination and Happiness Level

Not too happy 100
Pretty happy 90
80
Very happy
70
Percentage

60
50
40
30
20
10
0
Liberal Moderate Conservative
Political Inclination

Using Minitab, the test statistic is X 2  22.454 with 4 degrees of freedom and
p-value  .000.
With such a small p-value, reject H0 and conclude political inclination and
self-reported happiness level are related in the population of adult Americans.
The largest contributions to the test statistic are made by the “conservative/very
happy” cell, in which the observed count (187) is greater than the expected
count (152.88), and the “conservative/not too happy” cell, in which the observed
count (48) is less than the expected count (66.06).

Activity 25-5: Pets as Family

a. Here is the two-by-two table:

Dog Cat Total

Feels Close to Pet 1110 577 1687

Does Not Feel Close to Pet 71 110 181

Total 1181 687 1868

H0: Type of pet and feelings of closeness to pet are independent in the population
of adult Americans.
Ha: Type of pet is related to feelings of closeness to pet.
Technical conditions: The observations arose from a random sample of the
population, and the expected counts are at least five for each cell in the table.
 Activity 25-6 557

Using Minitab, the test statistic is X 2  49.63 with 1 degree of freedom and the
p-value  .000.
With such a small p-value, reject H0 and conclude type of pet and feeling close to
the pet are related in the population of adult Americans. The largest contribution
to the test statistic is made by the “cat/does not feel close to pet” cell, in which the
observed count (110) is greater than the expected count (66.57).
b. The null hypothesis is that the proportion of the population of all dog owners
who feel close to their pets is the same as the proportion of all cat owners who feel
close to their pets. In symbols, the null hypothesis is H0: ␲dog  ␲cat.
The alternative hypothesis is the proportion of the population of all dog owners
who feel close to their pets is not the same as the proportion of all cat owners who
feel close to their pets. In symbols, the alternative hypothesis is Ha: ␲dog ␲cat.
Technical conditions: The number of successes and failures in each group is at
least five, and the data were randomly selected.

25
.93988  .83988
The test statistic is z  ___________________________
__________________________  7.045.
1  ____
.9031(1  .9031) _____ 1
√ 1181 687 
p-value  2  Pr(Z  7.045)  .0000
With such a small p-value, reject H0 at any commonly used significance level.
You have very strong statistical evidence that the population proportion of dog
owners who feel close to their pets is not the same as the population proportion
of cat owners who feel close to their pets.
c. The p-values are the same in both cases.
d. (7.045) 2  49.63; 49.63  X 2

Activity 25-6: Newspaper Reading

a. The General Social Survey took one random sample, and this table summarizes
responses on two categorical variables (gender and newspaper reading frequency), so
the third scenario applies.
b. The following segmented bar graph displays the distributions (conditional
proportions) of newspaper reading frequency for each gender:

How Often American Adults

Read the Newspaper
100
Never
Less than once a week 90
80
Once a week
70
Few times a week
Percentage

60
Every day
50
40
30
20
10
0
Male Female
Gender
558 Topic 25: Inference for Two-Way Tables

This graph reveals that a higher proportion of men than women read the
newspaper every day (191/422  .453 for men vs. 167/484  .345 for women). A
higher proportion of women than men read the paper a few times a week (.211 for
men vs. .275 for women) and once a week (.126 for men vs. .167 for women).
c. The null hypothesis asserts newspaper reading frequency is independent of
gender. In other words, the null hypothesis says the proportional distribution
of the various reading frequency categories is identical for men and women
in the population of all American adults. The alternative hypothesis says
newspaper reading frequency is related to gender, which means these population
distributions are not the same for men and women.
The expected counts appear in parentheses in the table:

Male Female Total

Every Day 191 (166.75) 167 (191.25) 358

Few Times a Week 89 (103.40) 133 (118.60) 222

Once a Week 53 (62.42) 81 (71.58) 134

Less Than Once a Week 56 (56.36) 65 (64.64) 121

Never 33 (33.07) 38 (37.93) 71

Total 422 484 906

For an example of one of these calculations, the expected count for the “every
day/male” cell is found by taking (358)(422)/906  166.75.
Technical conditions: All of these expected counts are greater than five (smallest
33.07). The subjects were randomly selected from the population of adult
Americans, so the technical conditions are satisfied.
The chi-square test statistic, computed as
(O  E )2
 ________
E
turns out to be
X 2  3.526  3.075 
2.006  1.749 
1.420  1.238 
0.002  0.002 
0.000  0.000  13.020
Comparing this value to a chi-square distribution with (5  1)(2  1) 
4 degrees of freedom reveals that the p-value is slightly greater than .01 (13.02 is
just below 13.28 in Table IV). This is less than .05, so the test decision is to reject
the null hypothesis at the .05 significance level. The data provide fairly strong
evidence that newspaper reading frequency is related to gender.
In particular, the table cells in the top row contribute the most to the test statistic
calculation. This finding indicates more men than expected read the newspaper
 Activity 25-7 559

every day and fewer women than expected read the newspaper every day. The
sample proportions of men and women who read the newspaper every day are
191/422  .453 for men and 167/484  .345 for women.

••• Homework Activities

Activity 25-7: Government Spending
a. The following segmented bar graph displays the data:

Political Inclination and Spending

on the Environment
100
Too much
90
About right
80
Too little
Percentage 70
60
50

25
40
30
20
10
0
Liberal Moderate Conservative
Political Inclination

This graph reveals that a very high proportion (.819) of liberals in this sample feel
the govenernment is spending too little on the envirnoment.
b. No; although the observed count in this category is only 1, it is the expected
count that needs to be at least five in order for the chi-square test to be valid.
c. A chi-square test in this setting would be one of independence because the data
were collected by one random sample from a population and then classified by
both the independent and response variables.
d. H0: Political affiliation and feelings about government spending on the
environment are independent variables in the population of adult Americans.
Ha: Political affiliation and feelings about government spending on the
environment are related.
Technical conditions: The observations arose from a random sample of the
population, and the expected counts are at least five for each cell in the table.
Using Minitab, the test statistic is X 2  52.117 with 4 degrees of freedom and
p-value  .000. Using Table IV, 52.117  20.00, so p-value  .0005.
With a p-value less than .025, reject H0 at the .025 significance level. You can
conclude that political affiliation and feelings about government spending on the
environment are related in the population of adult Americans.
e. The largest contributions to the test statistic are made by the “liberal/too little
cell,” in which the observed count (127) is greater than the expected count (95.5);
the “conservative/too much” cell, in which the observed count (32) is also greater
than the expected count (18.27); and the “liberal/too much” cell, in which the
observed count (1) is less than the expected count (12).
 Activity 26-1 591

3. Press q and then Æ to display the graph (do not forget to turn off the other plots).
To create a labeled scatterplot, which is necessary for part b, you have two choices.
Choice 1:
a. Divide these data between men and women (four lists total: two for men and
two for women).
b. Use Plot1 and Plot2 simultaneously using different symbols.
Choice 2:
c. Create a new list (call it GNDR for gender) and enter a 0 if the corresponding
data is from a male student and a 1 if the data is from a female student (your
calculator only accepts numeric values in lists).
d. Download the program named LBLSCAT and use SPAN as the X-LIST, HGHT as
the Y-LIST, and GNDR as the CATEGORICAL VARIABLE LIST. You will need to type
in names for the different categories (i.e., you could use M for male and F for
female).

Solutions

26
••• In-Class Activities
Activity 26-1: House Prices
a. The observational units are the houses.
b. Two variables reported: price and size. Both are quantitative variables.
c. Yes, the data suggest that bigger houses tend to cost more than smaller ones, as
the house sizes tend to increase down the list as well.
d. Bottom left: 2130 Beach St. Top right: 833 Creekside Dr.
e. The point corresponding to the house at 845 Pearl Drive is circled on the
scatterplot below:

650
Price (in thousands of dollars)

600
550
500
450
400
350
300

500 750 1000 1250 1500 1750 2000

Size (in sq ft)
592 Topic 26: Graphical Displays of Association

f. Yes, the scatterplot reveals a positive relationship between a house’s size and price.
The upward trend of the graph indicates that, in general, the larger the house, the
greater the price.
g. Direction: positive
Strength: moderate
Form: linear
h. Answers will vary. One example is 2545 Lancaster Dr. (1030 sq ft, $344,720) and
415 Golden West Pl. (883 sq ft, $359,500). (The smaller house has the greater
price.)

Activity 26-2: Birth and Death Rates

a. This association is weak, negative, and linear.
b. A: Florida; reasoning: This state has the highest death rate and lowest birth rate
(of the three) because this is a state to which people like to retire.
B: Alaska; reasoning: This state has a lower than average death rate because
people typically do not retire to Alaska.
C: Utah; reasoning: This state has an unusually high birth rate that may be due
to the prevalence of the highly procreative Mormon (Latter Day Saints) faith
in Utah.

Activity 26-3: Car Data

Negative None Positive

Strongest Weakest Weakest Strongest

Letter of Scatterplot D G A H C E I F B

Activity 26-4: Marriage Ages

a. This scatterplot shows a strong, positive, linear association between the husband’s
age and the wife’s age.
b. Two of the ages fall exactly on the y  x line.
c. For 16 of the couples, the husband is older than the wife.
d. For six of the couples, the husband is younger than the wife.
e. The fact that the majority of couples produce points that lie above the y  x line
tells you that, in general, husbands tend to be older than their wives.
 Activity 26-5 593

Activity 26-5: Heights, Handspans, and Foot Lengths

a. The scatterplot of height vs. handspan follows:

75

Height (in inches)

70

65

60

55

17 18 19 20 21 22 23
Handspan (in inches)

There is a moderate, positive, linear association between height and handspan.

b. The labeled scatterplot of height vs. handspan follows:

26
Gender 75
Female
Male
Height (in inches)

70

65

60

55

17 18 19 20 21 22 23
Handspan (in inches)

Yes, men and women tend to differ with regard to handspan. Men tend to have
longer handspans than women and less variability in their handspans. This result
is indicated by the red points clustered fairly close together on the upper end of
the horizontal scale, whereas the black points are spread widely on the lower to
middle end of the horizontal scale.
Yes, men and women tend to differ with regard to height. Most of the men in this
class are taller than most of the women, although there are some fairly tall women
in this class. You can tell because the black points (women) range from the low to
the highest end of the vertical scale, whereas the red points are primarily clustered
toward the upper end of the vertical scale.
Yes, the association between handspan and height appears to be different between
men and women. There is a much stronger linear relationship between these
variables for the women than there is for the men. There is actually a very weak,
positive association been height and handspan for the men.
c. The scatterplot of height vs. foot length follows:
594 Topic 26: Graphical Displays of Association

75

Height (in inches)

70

65

60

55

19 20 21 22 23 24 25 26 27 28
Foot Length (in inches)

There is a moderately strong, positive, linear relationship between height and foot
length. The scatterplot shows one observation (21 in, 55 cm) that does not seem to
fit the overall trend. In general, the taller the person, the longer her or his feet.
The labeled scatterplot of height vs. foot length follows:

Gender 75
Female
Male
Height (in inches)

70

65

60

55

19 20 21 22 23 24 25 26 27 28
Foot Length (in inches)

As with handspans, most of the men’s feet are longer than most of the women’s
feet. However, there is one man of medium height with (relatively) short feet.
Both genders display a moderate, positive, linear association between height and
foot length, although the association is somewhat stronger among the women.

Activity 26-6: Televisions and Life Expectancy

a. Fewest: Haiti (5) Most: United States (844)
b. The scatterplot of life expectancy vs. televisions per thousand people follows:

80
Life Expectancy (in years)

70

60

50

40

0 100 200 300 400 500 600 700 800 900

Televisions per Thousand People

Yes, there is a moderate, positive, curved association.

 Activity 26-8 595

c. This is an absurd argument. Simply sending televisions to countries such as Haiti

will not lead to an increase in the life expectancies of inhabitants.
d. No; this is a very good example of a situation where two variables are strongly
associated, but there is no cause-and-effect relationship between them. Having
lots of televisions does not cause life expectancy to increase. There are several
other variables that affect both life expectancy and number of televisions per
1000 people.
e. Many answers are possible. One possibility is the wealth (GNP) of the nation.
Nations with high GNPs will also tend to have a large number of televisions per
capita and will have good medical care and thus long life expectancies. Similarly,
countries with low GNPs will tend to have few televisions per capita, poor
national medical care, and shorter life expectancies.

Activity 26-7: Airline Maintenance

a. The observational units are the airlines.
b. The scatterplot follows:
Delays Caused by Airline (percentage)

26
70

60

50

40

30

20

10

10 20 30 40 50 60 70 80 90 100
Maintenance Outsourced (percentage)

c. The scatterplot reveals a fairly strong positive association between these variables.
Airlines that outsource a higher percentage of their maintenance tend to have a
greater percentage of delays caused by the airline, as compared to airlines that
outsource a smaller percentage of their maintenance. The form of the association
is clearly nonlinear, as the relationship appears to follow a curved pattern.
d. No, you cannot conclude from these data that outsourcing maintenance causes
airlines to experience more delays. This is an observational study, not an
experiment. Many other variables could explain the observed association between
outsourcing and delays. For example, perhaps less financially successful airlines
tend to outsource more of their maintenance and also to have more delays.

••• Homework Activities

Activity 26-8: Miscellany
Answers will vary from student to student, but here are some examples:
a. SAT score and college GPA: positive, strong (You can expect students who do well
on the SAT to tend to also obtain above average college grades.)
 Activity 27-2 611

The correlations may be calculated in one of two ways:

Option 1 (if you only want the students to see the correlation coefficient):
a. Download the grouped file CARS99, which contains the lists PNUM (page
number), CMPG (city mpg), HMPG (highway mpg), FCAP (fuel capacity), WGT
(weight), FWGT (front weight), ACC30 (acceleration 0–30), ACC60 (acceleration
0–60), MILE (time to travel ¼ mile), and TYPE (0 for family, 1 for small, 2 for
luxury, 3 for large, 4 for sports, and 5 for upscale).
b. Next, download the program CORR.
c. Use the CORR program on your calculator to compute the value of the
correlation coefficient between time to travel 1/4 mile and weight.
Option 2 (if you do not care whether the students see the equation for the least
squares line along with the correlation coefficient):
a. Press the Ö button and then select LinReg(aⴙbx) from the CALC menu. Be
careful! The calculator contains two very similar commands: LinReg(aⴙbx)
and LinReg(axⴙb).
b. Enter the list names after LinReg(aⴙbx), separated by a comma.

Solutions
••• In-Class Activities

27
Activity 27-1: Car Data

Negative None Positive

Strongest Weakest Weakest Strongest

Letter of D G A H C E I F B
Scatterplot

Correlation
.907 .721 .450 .244 .081 .235 .510 .889 .994
Coefficient

a. The correlation coefficient is .450.

b. See table preceding part a.
c. Largest: 1.0 Smallest: 1.0
d. The correlation coefficient will assume its largest or smallest value when the
scatterplot displays a perfect linear relationship (positive or negative, respectively).
e. Positive correlation coefficients correspond to positive relationships.
f. The stronger the linear relationship, the closer the absolute value of the correlation
coefficient to one.

Activity 27-2: Governors’ Salaries

a. The 50 U.S. states are the observational units.
b. The scatterplot of governor’s salary vs. median housing price follows:
612 Topic 27: Correlation Coefficient

180,000

Governor’s Salary (in dollars)

160,000

140,000

120,000

100,000

80,000 Hawaii

60,000

100,000 150,000 200,000 250,000 300,000

Median Housing Value (in dollars)

The scatterplot shows a weak positive linear association with one outlier (Hawaii).
c. Answers will vary by student guess.
d. The correlation coefficient is .334.
e. Yes, one of the states appears to be unusual. Hawaii has an unusually large
median housing value ($272,700) and the salary of Hawaii’s governor is in the
lower half of the distribution.
f. The new correlation coefficient is .576. The revised scatterplot follows:

200,000
Governor’s Salary (in dollars)

180,000 Hawaii

160,000
140,000
120,000
100,000
80,000
60,000

100,000 150,000 200,000 250,000 300,000

Median Housing Value (in dollars)

The correlation coefficient is now a bit stronger (.334 vs. .576).

g. The correlation is now .673. The revised scatterplot follows:

300,000
Governor’s Salary (in dollars)

Hawaii
250,000

200,000

150,000

100,000

100,000 150,000 200,000 250,000 300,000

Median Housing Value (in dollars)

The correlation coefficient has again increased, but not as substantially.

 Activity 27-4 613

h. The correlation coefficient plummets to .061. The revised scatterplot follows:

200,000

Governor’s Salary (in dollars)

150,000

100,000

50,000
Hawaii
0

100,000 150,000 200,000 250,000 300,000

Median Housing Value (in dollars)

i. No, the correlation coefficient is not a resistant measure of association.

j. The government agencies tend to report the median housing price because
housing prices are usually quite skewed to the right with several high outliers,
and so the mean (which is not resistant) would not be an appropriate measure of
center to report with these data.

Activity 27-3: Televisions and Life Expectancy

a. This scatterplot shows a moderately strong positive association. The association
does not follow a linear form; it appears to be curved.

27
b. Answers will vary by student guess.
c. The correlation coefficient is .743.
d. Yes, the value of the correlation coefficient is fairly high even though the
association between the variables is not linear.
e. No, the fairly high value of the correlation coefficient is not evidence of a cause-
and-effect relationship between the variables. There are many confounding
variables that could explain the association, as discussed in Activity 26-6.

Activity 27-4: Guess the Correlation

a. Answers will vary. The following is one representative set of answers.

Repetition Number 1 2 3 4 5 6 7 8 9 10

Your Guess .6 .7 .7 .8 .85 .7 .15 .9 .85 .1

Actual Value .69 .87 .75 .87 .76 .74 .11 .84 .95 .37

b. See table in part a.

c. Guess: r  .8.
d. The scatterplot of your guess vs. actual values follows:
614 Topic 27: Correlation Coefficient

.8

.6

.4

.2

Guess
0

.2

.4

.6

.8

1
1 .8 .6 .4 .2 0 .2 .4 .6 .8 1
Actual (Correlation : .99)

The correlation coefficient r is .99. Most students will be surprised that their
guesses were so consistent.
e. The scatterplot of your errors vs. the actual values follows:

.8

.6

.4

.2
Error

.2

.4

.6

.8

1
1 .8 .6 .4 .2 0 .2 .4 .6 .8 1
Actual

For most students, errors are for correlations near 1 or 1.

f. The scatterplot of your errors vs. the repetition (trial) number follows:
 Activity 27-5 615

.8

.6

.4

.2

Error
0

.2

.4

.6

.8

1
0 4 8 12
Trial

In this example, there is no real evidence that the accuracy of the guesses changed
over time.
g. The correlation would be 1.00.
h. The correlation would be 1.00 in this case also.
i. No; if the correlation is 1.00, it does not mean that you guessed perfectly

27
every time. It means your guesses are consistent: either consistently correct or
consistently incorrect by the same amount and in the same direction. You cannot
use the correlation coefficient to tell whether you guess correctly, but you can use
it to tell whether your guesses change appropriately with the size and direction of
the actual correlation.

Activity 27-5: House Prices

a. For the z-score for the price of 2130 Beach St., you calculate (311,000 
482,386)/79601.5  2.148. For the product of z-scores, you calculate 2.148 
2.243  4.817.
For the z-score for the size of 2461 Ocean St., you calculate (1755 
1288.1)/369.191  1.265. For the product of z-scores, you calculate 1.223 
1.265  1.547.
b. Correlation coefficient r  14.819/19 or .77997  .78.
c. Most of the houses with negative price z-scores are paired with negative size
z-scores, making the product of their z-scores positive. This is a result of the
strong positive association between house price and size.
d. Using Minitab, the correlation coefficient is .780.
616 Topic 27: Correlation Coefficient

Activity 27-6: Exam Score Improvements

a. The scatterplot of exam 2 score vs. exam 1 score (ignoring the student who did not
take exam 2) follows:

100

90
Exam 2 Score
80

70

60

50

70 75 80 85 90 95 100
Exam 1 Score

The scatterplot shows a moderate positive linear association.

b. The correlation coefficient is .602.
c. Answers will vary by student expectation.
d. The scatterplot of exam 2 score vs. new exam 1 score follows:

100

90
Exam 2 Score

80

70

60

50

60 65 70 75 80 85 90
New Exam 1 Score

The correlation coefficient did not change (it remains .602).

e. Answers will vary by student expectation.
 Activity 27-6 617

f. The scatterplot of new exam 2 score vs. new exam 1 score follows:

200
190
180

New Exam 2 Score

170
160
150
140
130
120
110
100

70 75 80 85 90 95 100
New Exam 1 Score

The correlation coefficient did not change (it is still .602).

g. Answers will vary by student expectation.
h. The hypothetical data and scatterplot will vary by student. The following is a
representative example.

100
(hypothetical exam 1 score ⴙ 10)
Hypothetical Exam 2 Score

90

27
80

70

60

50 60 70 80 90
Hypothetical Exam 1 Score

The correlation coefficient is 1.0.

i. Answers will vary by student expectation.
j. The hypothetical data and scatterplot will vary by student. The following is a
representative example.

90
(twice hypothetical exam 1 score)
Hypothetical Exam 2 Score

80

70

60

50

25 30 35 40 45
Hypothetical Exam 1 Score

The correlation coefficient is 1.0.

618 Topic 27: Correlation Coefficient

Activity 27-7: Draft Lottery

a. Answers will vary.
b. With a perfectly fair, random lottery, there should be no association between draft
number and sequential date for the birthday. In other words, these variables should
be independent, so the correlation coefficient would equal zero. With an actual
lottery, you would not expect the correlation coefficient to equal exactly zero, but
it should be close to zero.
c. The scatterplot is shown here:

350

300

250
Draft Number

200

150

100

50

0
0 50 100 150 200 250 300 350
Sequential Date

It’s hard to see a relationship between the variables in this scatterplot, so a

reasonable guess for the value of the correlation coefficient would be close to zero.
d. Technology reveals the correlation coefficient to equal 0.226. This indicates a
weak negative association between draft number and sequential date. Although
not large, this correlation value is farther from zero than most people expect.
Looking at the scatterplot more closely, you can see there are few points in the top
right and bottom left of the graph. This result suggests few birthdays late in the
year were assigned high draft numbers, and few birthdays early in the year were
assigned low draft numbers, which means young men born late in the year were at
a disadvantage and had a better chance of getting a low draft number. Birthdays
late in the year were not mixed as thoroughly as those earlier in the year, so
they tended to be selected early in the process and thereby assigned a low draft
number.
e. The scatterplot for the 1971 draft lottery data is shown here:

350

300

250
Draft Number

200

150

100

50

0
0 50 100 150 200 250 300 350
Sequential Date
 Activity 27-9 619

The correlation coefficient is .014, which is very close to 0. This value indicates
there is no evidence of association between draft number and sequential date,
suggesting the lottery process was fair and random in 1971. The mixing
mechanism was greatly improved after the anomaly with the 1970 results was
spotted.

••• Homework Activities

Activity 27-8: Hypothetical Exam Scores
a. Yes, most of the exam scores follow a linear pattern in class A, with one exception.
b. Yes, most of the exam scores in class B are scattered haphazardly with no
particular pattern except for the exam in the lower left of the plot, which gives a
linear appearance to the overall plot.
c. For class A, the correlation coefficient is .037. For class B, the correlation
coefficient is .705. Both values are surprising. In class A, the value of r is
surprisingly small, and in class B, the value of r is surprisingly large. Clearly a
single outlier can have a huge effect on the value of r.
d. It only takes one exception to a linear pattern to throw off what could be a strong
correlation. Similarly, a seemingly haphazard pattern could have a fairly strong
correlation if one or more extreme points line up linearly. Because correlation
coefficients are so easily affected by these outliers, you say that they are not
resistant.

27
e. In class C, students who scored less than 50 on exam 1 also scored less than
50 on exam 2, although some scores increased and some decreased. Students who
scored 70 and greater on exam 1 also performed this well on exam 2, but some
scores increased and some decreased. There were no students who scored
between 50 and 70 on either exam.
f. The correlation coefficient is .954. This value is deceptively high. The two
clusters of scores, which have no particular pattern within each cluster, form a
line (any two points will form a line) that causes an artificially inflated overall
correlation coefficient.

Activity 27-9: Proximity to the Teacher

a. The correlation coefficient cannot be less than 1.0.
b. The order in which you state the variables will not change the correlation
coefficient. The correlation coefficient of x and y is always the same as the
correlation coefficient of y and x.
c. A correlation coefficient of .8 indicates a strong negative association.
d. You cannot find a correlation between one quantitative and one categorical
variable.
e. If the correlation coefficient is .8, then the association is negative, so students
who sit father away tend to score lower.
f. You can never conclude cause-and-effect based solely on the value of the
correlation coefficient.
634 Topic 28: Least Squares Regression

To see the least squares line:

7. Choose Graph  Least Squares Line from the menus.
Note: In practice, you can go directly to Graph  Least Squares Line to show the
regression line on a scatterplot.
To see the effect of an influential point (part n of Activity 28-1):
1. Construct the scatterplot and add the regression line. Use the Graph menu to
remove the movable line and residual squares if they are still visible on the graph.
2. Select any data point and drag it to the right to make the foot length much
greater, and then drag it up and down to see the effect on the least squares line.
For a more dramatic effect, drag the horizontal scale to allow even greater foot
lengths.
3. Choose Edit  Undo (or CtrlZ), several times if needed, to restore the point to
its original value (or reopen the data file).
To investigate the coefficient of determination, r 2 (parts p and q of Activity 28-1):
1. Construct the scatterplot and add the movable line to the original data as
described in the previous steps.
2. Drag the line to adjust the slope to be as close to horizontal (slope  0) as possible.
3. Choose Graph  Show Squares again and move the horizontal line up and down
by dragging its center until you minimize the sum of the squares. This should
occur when the line is at a level equal to the mean of the heights. The sum of
squares at this value is the total variability in the heights, which serves as the
denominator for calculating the coefficient of determination, r 2.
4. Choose Graph  Least Squares Line again to show the least squares line and find
its SSE.
The difference between the two sums of squares is the amount of variability in the
heights that is explained by the least squares regression on the foot lengths. Use it as
the numerator in the calculation of r 2 and multiply by 100% to express the result as
the percentage of total variability in the heights that is explained by the foot lengths.
Note the value of r 2 is also displayed next to the equation of the least squares line in the
bottom corner of the plot in Fathom.

Solutions
••• In-Class Activities
Activity 28-1: Heights, Handspans, and Foot Lengths
a. There is a moderate positive linear association between height and foot length in
this scatterplot. Answers will vary by student.
The following is one representative set of answers.
ˆht  38.293  1.048 foot size.
b. The equation reported is (predicted) heig
c. No, everyone in the class did not obtain this same line.
d. Answers will vary.
e. The SAE value is 56.46. Yes, someone had a smaller value (55.11).
 Activity 28-1 635

f. The SSE value is 238.25. A graph follows:

ˆht  37.503  1.074 foot size; the SSE is 238.19. Yes, someone
g. The equation is heig
had a smaller value (235.28). A graph follows:

28
636 Topic 28: Least Squares Regression

ˆht  38.302  1.033 foot size

h. Regression line: heig
SSE (least squares): 235
A graph follows.

i. Using the least squares regression line, predicted height is 38.302  1.033  28
or 67.226 inches. Yes, this prediction seems reasonable based on the scatterplot.
j. Using the least squares regression line, predicted height is 38.302  1.033  29
or 68.259 inches.
k. These predictions differ by 1.033 inches, which is the slope of the least squares line.
l. The least squares line would predict a height of 38.302 inches for a person with a
0 cm foot length. No, this does not make any sense because a foot length of 0 cm
is outside the range of plausible values.
m. Using the least squares regression line, predicted height is 38.302  1.033 
45 or 84.79 inches, which is more than 7 feet tall. You should not consider this
prediction as reliable as the one for a person with a 28 cm foot length, because for
this prediction (84.79 in) you do not have much data for someone more than
7 feet tall. You should not feel comfortable assuming the same relationship will
hold for such extreme observations.
n. No, the least squares regression line does not move much at all when you change
this student’s height.
o. Changing the height of the student with the shortest or longest foot has a much
more drastic effect on the regression line. Points on the end of the regression line
are clearly more influential than points near the middle.
_
p. SSE (y )  475.75
q. Here is the percentage change in the SSE:
475.75  235  50.6%
100% ____________
475.75
 Activity 28-2 637

Activity 28-2: House Prices

a. Yes, the least squares line appears to provide a reasonable model for predicting the
price of a house based on its size (sq ft, or ft2).
b. Slope b 
79,802  $168.6/ft2
.780  ______
369.2
Intercept a 
168.6  (1288.1)  $265,212.34
482,386  _____
ft2
ˆe  265,212.34  168.6/ft2  size
Least squares line equation: pric
c. Using Minitab, predicted price  $265,222  $169  size.
d. Predicted price  $265,222  $169  (1242)  $475,120. Yes, this answer
seems reasonable.
e. The prediction was too high by $475,120  $459,000 or $16,120.
f. Fitted value: $475,120 Residual: $16,120
g. Points that lie above the line will have positive residuals.
h. The scatterplot, with the point circled, follows:

^
Price ⴝ 265,222 ⴙ 168.6 Size
650,000
600,000

28
Price (in dollars)

550,000
500,000
450,000
400,000
350,000 Greatest negative residual
300,000

500 750 1000 1250 1500 1750 2000

Size (in sq ft)

The house with the greatest negative residual is located at 2545 Lancaster Drive.
i. Each increase of one square foot in the house size increases the predicted house
price by $168.60.
j. You calculate 100  $168.60 or $16,860.
k. The predicted price for a house with an area of 0 square feet is $265,222. This
value makes no sense in this context because there is no such thing as a house
with zero area.
l. The percentage of variability in house prices explained by the least squares
regression line with size is 60.8% as r2  correlation coefficient2  (.78)2.
m. No, it would not be reasonable to use this regression line to predict the price of a
3500 ft2 house because the given house sizes are from about 500–2000 ft2. This
638 Topic 28: Least Squares Regression

would be extrapolation; you have no idea whether the given association continues
beyond this range of house sizes.
n. No, it would not be reasonable to use this regression line to predict the price of a
2000 ft2 house in Canton, New York. This regression line was created with data
collected solely from one region in California. The housing market in New York
is very different, and you cannot expect this regression line to make accurate
predictions for any market other than Arroyo Grande, California.

Activity 28-3: Animal Trotting Speeds

a. Yes, it appears that larger animals tend to break into trot at higher speeds than
smaller animals. The scatterplot reveals a moderate positive nonlinear association.
b. No, the least squares line does not appear to provide a reasonable model for
summarizing the relationship between trot speed and body mass because the
relationship between these variables is nonlinear.
c. Small animals: tend to have negative residuals
Medium-sized animals: tend to have positive residuals
d. The residual plot of residual vs. body mass follows:

0.50

0.25

0.00
Residual

–0.25

–0.50

–0.75

–1.00

0 50 100 150 200

Body Mass

Yes, the residual plot reveals a pattern that suggests the linear model is not
appropriate.
e. The scatterplot of trot speed vs. log(body mass) follows:

^
Trot Speed ⴝ 0.8209 ⴙ 0.4637 Log(Body Mass)
Trot Speed (in meters per second)

2.0

1.5

1.0

0.5

0.0

–2 –1 0 1 2
Log(Body Mass)

Yes, the association between trot speed and log10(body mass) appears to be fairly
linear.
 Activity 28-4 639

f. Here is the least squares line:

ˆeed  0.8209  0.4637  log(body mass); r 2  87.5%
trotsp
g. The residual plot for the least squares line follows:

0.3
0.2
0.1
0.0

Residual
–0.1
–0.2
–0.3
–0.4
–0.5
–0.6

0 50 100 150 200

Body Mass

Yes, the points in this plot are fairly randomly scattered, indicating the linear
model is reasonable for the transformed data.
h. 10 kilograms: predicted trot speed  0.8209  0.4637  log(10)  1.2846 m/sec
100 kilograms: predicted trot speed  0.8209  0.4637  log(100)  1.7483 m/sec
i. The difference between these predictions is 0.4637 m/sec  slope of regression
line.

Activity 28-4: Textbook Prices

a. It seems reasonable to regard price as the response variable because it is natural to
take an interest in predicting a textbook’s price from other variables.

28
b. These scatterplots follow:

180
160
140
Price (in dollars)

120
100
80
60
40
20
0

0 200 400 600 800 1000 1200

Pages
640 Topic 28: Least Squares Regression

180
160
140

Price (in dollars)

120
100
80
60
40
20
0

1970 1980 1990 2000 2010

Year

The scatterplot of price vs. pages reveals a fairly strong positive linear association.
The scatterplot of price vs. year also indicates a positive association, but the
association is much weaker and not very linear. Two unusual textbooks from
the early 1970s, with very low prices, appear to be outliers, because they differ
substantially from the pattern of the other textbooks, and potentially influential
observations.
c. Number of pages appears to be a much better predictor of price than year. The
relationship is much stronger and also more linear.
d. The equation of this least squares line is predicted price  3.42  0.147 pages. It
is shown on the scatterplot here:

180
160
140
Price (in dollars)

120
100
80
60
40
20
0
0 200 400 600 800 1000 1200
Pages

e. The predicted price for a 500-page textbook is predicted price  3.42 

0.147(500)  $70.08.
f. The slope coefficient is $0.147/page, which indicates the predicted price of a
textbook increases by $0.147 (about 15 cents) for each additional page.
g. The value of the correlation coefficient between price and pages is r  0.823, so
r 2  (0.823)2  0.677. This coefficient says that 67.7% of the variability in
textbook prices is explained by the least squares line with number of pages. The
other 32.3% is explained by other factors, which could include random variation.
 Activity 28-4 641

h. A plot of residual vs. pages follows:

75

50

25

Residual
0

25

50

0 200 400 600 800 1000 1200

Pages

This residual plot reveals no obvious pattern, which suggests the least squares line
is a reasonable model for the relationship between price and pages.
i. The equation of this least squares line is predicted price  4969  2.516 year,
with r 2  .186. The least squares line is shown on the scatterplot here:

180
160
140
Price (in dollars)

120
100
80
60
40

28
20
0

1970 1980 1990 2000 2010

Year

j. The residual plot is shown here:

100

50
Residual

50

1970 1980 1990 2000 2010

Year

This plot reveals that the middle years (19902000) tend to have negative
residuals. This pattern suggests a linear model is not very appropriate for
predicting price from year of publication. A transformation might lead to a more
appropriate linear model.
670 Topic 29: Inference for Correlation and Regression

Solutions
••• In-Class Activities
Activity 29-1: Studying and Grades
a. Observational units: students at UOP
Explanatory variable: study hours per week Type: quantitative
Response variable: GPA Type: quantitative
b. The scatterplot of GPA vs. study hours follows:

4.0

3.5
GPA

3.0

2.5

2.0

0 1 2 3 4 5 6 7 8 9
Study Hours per Week

There is a weak positive association between these variables.

ˆ
c. GP A  2.89  .0894  (study hours)
d. The value of r 2 is 11.8%, which means that only 11.8% of the variation in the
GPAs is explained by the least squares line with study hours per week. The
remaining 89.2% is due to other factors, including random variation.
e. This equation predicts the GPA will rise by 0.0894 points on average for each
additional hour of study. This number is the slope in the regression equation.
f. This least squares line is based on a sample. Its coefficients are statistics because
they are based on a sample.
g. No; if the student researchers took another sample of 80 students, they probably
would not get the exact same regression equation from this new sample. They
should get a similar equation, but sampling variability would most likely cause the
equation to be somewhat different.
h. Answers will vary with each running of the applet. The following is one
representative set:
ŷ  0.023x  3.38
No, this is not the same equation as in part c.
i. yˆ  0.031x  3.348 ŷ  0.032x  3.339
Yes, you get a different sample line each time.
j. The generated regression lines appear to rotate about the middle, sometimes with
a positive slope and sometimes with a negative slope.
 Activity 29-1 671

k. The following graph displays the results:

–1.12 –0.1 –0.08 –0.06 –0.04 –0.02 0.0 0.02 0.04 0.06 0.08 0.1 0.12
0.0

The distribution of slope coefficients is roughly normal, centered at zero, with a

standard deviation of 0.028.
l. None of the 100 simulated sample slopes are 0.0894. This suggests that the
p-value for testing whether the population slope is zero against the alternative
hypothesis that the population slope is positive is roughly zero. You would
therefore conclude that the population slope is positive.
m. The null hypothesis is that there is no relationship between the population hours
studied and GPA, or the slope of the regression line between these two variables is
zero. In symbols, H0: ␤  0.
The alternative hypothesis is that students who spend more hours studying
tend to have higher GPAs, or the slope of the regression line between these two
variables is positive. In symbols, Ha: ␤  0.
n. Using Minitab, SE(b) is 0.02771. Yes, this is very close to the standard deviation
of the 100 simulated sample slope coefficients.
0.0894  3.23.
o. The test statistic is t  _______
0.02771
p. You calculate df  80  2  78. Using Table III with 60 degrees of freedom,
.001  p-value  .005.
Yes, this p-value is consistent with the simulation results, which found a p-value of
roughly zero.
q. Yes, this p-value suggests the association found in the sample of 80 students

29
would happen by random chance alone less than .1% of the time if there were no
association between GPA and hours studied in the population.
r. Using t*  2.000 with 60 degrees of freedom, a 95% CI for ␤ is
0.0894  2  (0.02771), which is (0.034, 0.145).
s. You are 95% confident the slope of the population regression line between GPA
and hours studied is somewhere between 0.034 and 0.145 points/hour.
t. You have strong statistical evidence of a positive association between GPA and
hours studied at UOP. You are 95% confident that an additional hour of study
corresponds to an average increase of 0.034 to 0.145 points in overall student
GPA. You are not necessarily drawing a cause-and-effect relationship, however, as
this is an observational study and not an experiment.
672 Topic 29: Inference for Correlation and Regression

Activity 29-2: Studying and Grades

a. The following histogram and normal probability plot display the distribution of
the residuals.

99.9

99

95
90

14 80
70

Percentage
12 60
10 50
40
Frequency

8 30
6 20
4 10
2 5
0
1
–0.8 –0.4 0.0 0.4 0.8
Residual
0.1
–2 –1 0 1 2
Residual

The distribution of residuals appears roughly normal. These plots do not reveal
any marked features suggesting nonnormality.
b. The scatterplot of residual vs. study hours follows:

1.0

0.5
Residual

0.0

–0.5

–1.0

0 1 2 3 4 5 6 7 8 9
Study Hours per Week

This residual plot does not reveal a strong pattern (curvature). The variability of
the residuals does not differ substantially at various x-values (although it is a little
different at the lowest and highest x-values than it is at the middle x-values).

Activity 29-3: House Prices

ˆe  $265,222  168.6  (size).
a. The equation is pric
SE(b)  31.88 ($/ft2)
b. The null hypothesis is that there is no relationship in the population between
house size and price, or the slope of the population regression line between these
two variables is zero. In symbols, H0: ␤  0.
 Activity 29-3 673

The alternative hypothesis is that larger houses in this population tend to have
greater purchase prices, or the slope of the population regression line between
these two variables is positive. In symbols, Ha: ␤  0.
168.6  5.29.
The test statistic is t  _____
31.88
Using Table III with 18 degrees of freedom, p-value  .0005. Using Minitab, the
p-value is .000.
With the small p-value, reject H0 at any commonly used significance level, and
conclude there is very strong statistical evidence of a positive linear association
between house size and price in this population.
The residual plot follows:

100,000

50,000
Residual

–50,000

–100,000

500 750 1,000 1,250 1,500 1,750 2,000

Size (in sq ft)

A residual plot does not reveal any strong curvature and shows that the standard
deviations of the y-values can be reasonably considered the same at each x-value.
The data are from a simple random sample, so technical conditions 1, 2, and 4 are
met. A histogram and probability plot indicate the residuals are not beautifully
normal, but the nonnormality is probably not strong enough to convince you that
technical condition 3 has been violated.

6 99
5

29
95
Frequency

4
90
3
80
2
70
Percentage

1 60
0 50
40
0 30
0

00

00
00
0

0
00
00

00

00

,0

,0
,0
,0

20
5,

0,

5,

25

75
50
00
–7

–5

–2
–1

10
Residual
5

1
–200,000 –100,000 0 100,000 200,000
Residual

c. A 90% CI for ␤ is 168.6  (1.734)  (31.88)  (113.32, 223.88).

You are 90% confident the slope of the population regression line between
house price and size is between 113.22 and 223.88 $/ft 2. Therefore, you are 90%
confident the average increase in the price of a house is between $113.22 and
$223.88 for each additional square foot of area in a house in Arroyo Grande,
California.
674 Topic 29: Inference for Correlation and Regression

The value zero is not in this interval, which is consistent with the test result in
part b where you found that zero is not a plausible value for the population slope
coefficient.

Activity 29-4: Plasma and Romance

a. The scatterplot and correlation coefficient indicate there is a weak positive linear
association between plasma levels and passionate feelings.
b. The correlation coefficient is a statistic because it is computed from a sample.
c. Yes, it is possible to obtain such a large correlation coefficient by random chance,
even if there is no correlation between these variables in the population.
d. H0: There is no correlation between plasma levels and strength of romantic feelings
in the population, ␳  0.
Ha: There is a positive correlation between plasma levels and strength of romantic
feelings, ␳  0.
___
.347  √56  ___________
e. The test statistic is t  ___________
__________ 2.597
__________  2.77.
√1  (.347)2 √1  (.1204)
f. You calculate df  58  2  56. Using Table III with 50 degrees of freedom,
.001 p-value  .005.
g. With the small p-value, you have strong statistical evidence there is a positive
correlation between plasmas levels and strength of romantic feelings in this
population. If there were no correlation in the population, you would obtain a
sample result (a correlation coefficient as great or greater than .347) less than .1%
of the time by random sampling alone. Because this is so unlikely to have
occurred by random chance, but did happen, you conclude there must be a
positive correlation between these variables in the population.
Technical conditions: Neither has been assumed met. The sample was not a
simple random sample, and you have no way of testing whether the variables
are normally distributed because you only have the summary statistic r and the
sample size n (which is large).

Activity 29-5: Textbook Prices

a. The scatterplot with least squares line follows:

180
160
140
Price (in dollars)

120
100
80
60
40
20
0

0 200 400 600 800 1000 1200

Pages
 Activity 29-5 675

The equation of this line is predicted price  3.42  0.147  (pages). The value
of r 2 is .677, indicating that 67.7% of the variability in textbook prices can be
explained by the number of pages in the texts.
b. The slope coefficient is b  0.147, and technology reports the standard error to be
0.019. The slope indicates that the predicted price increases by $0.147 on average
for each additional page in the book. The standard error measures the variability
in the sample slopes from repeated random samples of 30 textbooks from this
population.
c. The population of interest in this study is all textbooks in the Cal Poly Bookstore
in November 2006.
d. Let ␤ represent the slope of the least squares line for predicting textbook price
from number of pages in the population. Then the null hypothesis is H0: ␤  0,
meaning there is no association between textbook price and number of pages in the
population. The alternative is Ha: ␤  0, meaning there is a positive association
between these variables in the population. The test statistic is
0.147  7.74
b  _____
t  _____
SE(b) 0.019
(Without rounding, technology reports the test statistic to be 7.65.) Comparing
this to the t-distribution using Table III with 30  2 or 28 degrees of freedom
reveals that the p-value is much less than .0005. Such a small p-value provides
extremely strong evidence of a positive relationship between a textbook’s price
and its number of pages in the population of all textbooks in that bookstore at
that time.
e. Using Table III, the t* critical value for 90% confidence with 28 degrees of
freedom is 1.701. A 90% confidence interval for the population slope ␤ is
0.147  1.701(0.019), which is 0.147  0.032, which is the interval from 0.115
through 0.179. You can be 95% confident that in the population of all textbooks
in that bookstore, the predicted price of a textbook increases between 11.5 and
17.9 cents for each additional page.
f. A 99% confidence interval for the population slope ␤ has the same midpoint,

29
namely the sample slope 0.147. But the 99% interval is wider than the 90%
interval in order to achieve higher confidence of capturing the population slope
coefficient.
g. Technical conditions: First, Shaffer and Kaplan did take a random sample of
textbooks, so that condition is satisfied as long as you restrict your population to
the Cal Poly Bookstore in November 2006. To check the normality condition,
consider the following histogram and normal probability plot of the residuals.

7
6
5
Frequency

4
3
2
1
0
80 40 0 40 80
Residual
676 Topic 29: Inference for Correlation and Regression

99

95
90
80

Percentage
70
60
50
40
30
20
10
5

1
150 100 50 0 50 100 150
Residual

Both plots indicate the distribution of the residuals is approximately normal,

so that condition is satisfied. For the conditions regarding linearity and equal
variability, consider a plot of residual vs. pages:

75

50

25
Residual

25

50

0 200 400 600 800 1000 1200

Pages

There is no obvious pattern to the residuals in this graph, so the linearity

condition is met. The variability in residuals appears to be similar across all values
of number of pages, although there might be a bit more variability in the residuals
for larger numbers of pages. All technical conditions are met, so the significance
test and confidence interval are valid.

••• Homework Activities

Activity 29-6: Proximity to Teacher
a. H0: There is no correlation between distance from teacher and quiz average. In
symbols, ␳  0.
Ha: There is a negative correlation between distance from teacher and quiz average.
In symbols, ␳  0.
b. The value r  .3 is a statistic because it is calculated from a sample. It is not a
population value.
c. You need to know the sample size.
___
.3
d. The test statistic is t  __________
________ 1.2728
 √ 18  _________
________  1.33.
√1  (.3)2 √1  (.09)
With 18 degrees of freedom, the p-value is .10.