10 Confidence Intervals and

Hypothesis Tests Based on

Two Samples or Treatments
Looking Back
■ Recall the formal, four-part hypothesis test process.
■ Remember the specific inference procedures concerning a single population parameter: m,
p, or s2 .
Looking Forward
■ Adapt and extend the single-sample hypothesis test procedures.
■ Construct confidence intervals to estimate the difference between two population parameters.
■ Conduct hypothesis tests to compare two population parameters.

Are people who live at higher

altitudes slimmer?
A recent study suggests that people who live at higher altitudes tend to be thinner
than those who live in low-lying areas. This does not mean we should all pack up and
move to the mountains. There are certainly other factors affecting obesity. However,
the data suggest that Americans who live near sea level are more likely to be obese,
compared with people who live at higher altitudes, for example, in Colorado.
A possible explanation for a difference in obesity rates is that altitude can
affect appetite hormones. In addition, altitude can also affect how many calories
the body burns in ordinary daily activities. People who live at higher altitudes
also tend to drink more water, which may help weight loss.
To examine this trend more closely, random samples were obtained for 125
adults living in Denver, Colorado (the Mile High City), and 150 adults living in
New Orleans, Louisiana (where the mean elevation is zero feet above sea level).
Using body mass index as a measure of obesity, 38 of those living in Denver and
61 of those living in New Orleans were classified as obese.
The hypothesis test procedures presented in this chapter will be used to compare
parameters (means or variances, for example) from two different populations. In
this case, we will compare the population proportion of obese adults in Denver
with the population proportion of obese adults in New Orleans. These tests are
constructed using methods of standardization similar to those in Chapter 9.

CONTENTS
10.1 Comparing Two Population Means Using Independent Samples When Population
Variances Are Known
10.2 Comparing Two Population Means Using Independent Samples from Normal
Populations
10.3 Paired Data
10.4 Comparing Two Population Proportions Using Large Samples
10.5 Comparing Two Population Variances or Standard Deviations
Katja Kreder/AWL Images/Getty Images

461
462 CH A PT ER 1 0 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

Notation

To conduct a hypothesis test to compare two (similar) population parameters, we will sim-
ply modify the single-sample procedures presented in the previous chapter. Perhaps the
most tricky aspect of these procedures is the notation. The following table summarizes the
notation used to represent similar parameters associated with two different populations.

Population parameters
Standard
Mean Variance deviation Proportion
Population 1 m1 s21 s1 p1
Population 2 m2 s22 s2 p2

The following table summarizes the notation used to represent values of summary statis-
tics associated with samples from two different populations.
Note: We do not necessarily use Sample statistics
every summary statistic associated
Sample Standard
with a sample in every problem.
size Mean Variance deviation Proportion
For example, we may only need
the sample size and proportion in Sample from =
n1 x1 s21 s1 p1
one case, but use the sample size, population 1
mean, and standard deviation in Sample from =
another problem. population 2
n2 x2 s22 s2 p2

To compare two population parameters to see whether there is any evidence that they
are different, we often consider a difference. For example, to compare two population
means, m1 and m2, we consider the difference m1 2 m2. In searching for evidence that p1
is larger than p2, we look at the difference p1 2 p2.
There are two reasons to consider a difference.
1. A typical relationship between two population parameters can be written in terms of a
difference. For example, suppose we need to compare the means from two popula-
tions, m1 and m2.

Standard Difference
notation notation
m1 5 m2 is equivalent to m1 2 m2 5 0
m1 . m2 is equivalent to m1 2 m2 . 0
m1 , m2 is equivalent to m1 2 m2 , 0

Therefore, a statistical test with null hypothesis H0: m1 2 m2 5 0 corresponds to a

test of H0: m1 5 m2 . And Ha: m1 2 m2 . 0 is equivalent to Ha: m1 . m2 . The hypoth-
esized difference between the two means may be nonzero. The null hypothesis
H0: m1 5 m2 1 5 written using a difference is equivalent to H0: m1 2 m2 5 5.
2. In addition, a difference (for example, m1 2 m2) is itself a single population parameter.
A natural, intuitive statistic, X 1 2 X 2 , may be used to estimate the value of this param-
eter. The properties of X 1 2 X 2 will be used to develop a test statistic.
As in the statistical tests presented in the last chapter, in any two-sample hypothesis
test we usually make certain assumptions. The assumptions associated with the hypothe-
sis tests in this chapter include a statement concerning the selection of individuals or
objects from two different populations.
 10.1 Comparing Two Population Means Using Independent Samples When Population Variances Are Known 463

Definition
1. Two samples are independent if the process of selecting individuals or objects in
sample 1 has no effect on, or no relation to, the selection of individuals or objects in
sample 2. If the samples are not independent, they are dependent.
Similar means the individuals or 2. A paired data set is the result of matching each individual or object in sample 1 with
objects share some common, a similar individual or object in sample 2. A common experiment in which paired data
fundamental characteristic. They are obtained involves a before and after measurement on each individual or object.
may even be the same individual Each before observation is matched, or paired, with an after observation.
or object!

The notation, the idea of using differences, and the extra assumptions are all used in
the following sections to construct hypothesis tests for comparing various characteristics
of two populations.

10.1 Comparing Two Population Means Using Independent

Samples When Population Variances Are Known
As in Chapters 8 and 9, the first hypothesis test presented here, for comparing two popula-
tion means, is instructive but not very realistic. Because X1 is a good estimator for m1 and
X2 is a good estimator for m2, it is reasonable to use the estimator X 1 2 X 2 to estimate the
parameter m1 2 m2. To develop a hypothesis test, we need to know the properties of the
estimator, or the distribution of the random variable, X 1 2 X 2.

Properties of X 1 ! X 2
Suppose
1. X 1 is the mean of a random sample of size n1 from a population with mean m1 and
variance s21.
2. X2 is the mean of a random sample of size n2 from a population with mean m2 and
variance s22.
3. The samples are independent.

If the distributions of both populations are normal, then the random variable X 1 2 X 2 has
the following properties.
1. E (X1 2 X2 ) 5 mX12X 2 5 m1 2 m2 .
X 1 2 X 2 is an unbiased estimator of the parameter m1 2 m2. The distribution is cen-
tered at m1 2 m2.
s21 s22
2. Var ( X1 2 X2 ) 5 sX212X2 5 1 and the standard deviation is
n1 n2
s21 s22
sX12X 2 5 1 .
Å n1 n2
3. The distribution of X 1 2 X 2 is normal.

Can you see the standardization If the underlying distributions are not known, but both n1 and n2 are large, then X 1 2 X 2
coming? is approximately normal (by the central limit theorem).

Because the distribution of X 1 2 X 2 is (approximately) normal, the usual standardiza-

tion can be used to obtain a Z random variable. The resulting hypothesis test has a very
typical form.
464 CHA PT E R 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

Hypothesis Test Concerning Two Population Means

When Population Variances Are Known
Given two independent random samples, the first of size n1 from a population with mean
m1 and the second of size n2 from a population with mean m2, assume that:
For reference, we’ll call these the 1. The underlying populations are normal and/or both sample sizes are large, and
two-sample Z-test assumptions. 2. The population variances, s21 and s22 , are known.

A hypothesis test concerning two population means, in terms of the difference in means
m1 2 m2, with significance level a, has the form
This is the template for a H0: m1 2 m2 5 D0
hypothesis test concerning two Ha: m1 2 m2 . D0, m1 2 m2 , D0, or m1 2 m2 2 D0
population means when variances
( X1 2 X2 ) 2 D0
are known, sometimes called a TS: Z 5
two-sample Z test. s21 s22
1
Å n1 n2
RR: Z $ za, Z # 2za, or 0 Z 0 $ za / 2

A CLOSER L OK
D is the uppercase Greek letter 1. The value D0 is the fixed, hypothesized difference in means. Usually D0 5 0, that is,
delta. the means are assumed equal. The null hypothesis is then H0: m1 2 m2 5 0, which is
equivalent to H0: m1 5 m2. However, D0 may be some nonzero value. For example, two
population means may historically differ by 12 so that H0: m1 2 m2 5 12 ( 5 D0 ) . We
may want to conduct a test to see whether there is any change in this difference, with
Ha: m1 2 m2 2 12.
2. Just a reminder: Use only one (appropriate) alternative hypothesis and the correspond-
ing rejection region. The z critical values are from the standard normal distribution.
3. This hypothesis test procedure can be used only if both population variances are
known. If they are unknown but both sample sizes are large, some statisticians substi-
tute s21 for s21 and s22 for s22. This produces an approximate test statistic. Section 10.2
presents an exact test procedure for comparing population means (under certain
assumptions) when the population variances are unknown.

The following example illustrates this hypothesis test procedure.

Example 10.1 Turning off the TV

Some social science researchers believe that with increased usage of social media, mobile
devices, and other communication devices, younger people are watching less TV. Inde-
pendent random samples of people in two age groups were obtained, and the weekly time
spent watching TV (in hours) was recorded for each. The summary statistics and known
variances are given in the following table.1

Sample Sample Population

Age group size mean variance
18–24 (1) n1 5 18 x1 5 23.4 s21 5 44.89
25–34 (2) n2 5 24 x2 5 28.9 s22 5 65.61

Is there any evidence to suggest that the mean weekly time spent watching TV for 18–24
year olds is less that the mean weekly time spent watching TV for 25–34 year olds? Use
 10.1 Comparing Two Population Means Using Independent Samples When Population Variances Are Known 465

Solution Trail 10.1 a 5 0.01 and assume that each underlying distribution of weekly time spent watching TV
is normal.
K EY WO R D S
■ Is there any evidence? SOLUTION
■ Less than STEP 1 Arbitrarily, let the 18–24-year-old group be population 1, and the 25–34-year-
■ Known variances old group be population 2.
■ Independent random samples The current state, or assumption, is that the two population mean weekly times
■ Each underlying distribution is spent watching TV are equal:
normal
m1 5 m2 1 m1 2 m2 5 0 ( 5 D0 ) .
TR A NSL ATI O N
The sample sizes, sample means, and population variances are given.
■ Conduct a one-sided,
left-tailed test to compare m1 We are trying to find evidence that the 18–24-year-old group has smaller mean
and m2 . weekly time spent watching TV: m1 , m2, which is the same as m1 2 m2 , 0.
Therefore, the alternative hypothesis is one-sided, left-tailed.
CON C EPT S
STEP 2 The four parts of the hypothesis test are
■ Hypothesis test concerning
two population means when H0: m1 2 m2 5 0
variances are known Ha: m1 2 m2 , 0
V I SI O N ( X1 2 X2 ) 2 0
TS: Z 5
Use the template for this s21 s22
hypothesis test. The samples are 1
Å n1 n2
random and independent, the
underlying populations are RR: Z # 2za 5 2z0.01 5 22.3263
normal, and the population STEP 3 The value of the test statistic is
variances are known. Use a
one-sided alternative hypothesis ( x1 2 x2 ) 2 0 23.4 2 28.9
and the corresponding rejection z5 5 5 22.4055 ( #22.3263 )
s21 s22 44.89 65.61
region, find the value of the test 1 1
statistic, and draw a conclusion. Å n1 n2 Å 18 24
STEP 4 Because 22.4055 lies in the rejection region, we reject the null hypothesis at the
a 5 0.01 significance level. There is evidence to suggest that the mean weekly
time spent watching TV for 18–24-year-olds is less than the mean weekly time
spent watching TV for 25–34-year-olds.
The p value for this hypothesis test is

p 5 P ( Z # 22.4055 ) 5 0.0081 ( # 0.05 ) Use Table III in the Appendix.

Because p # a, we reject the null hypothesis. (See Figure 10.1.)

−2.4055 0
Figure 10.1 p-Value illustration:
p 5 P(Z # 22.4055)
5 0.0081 # 0.05 5 a

Figures 10.2 through 10.4 show a technology solution.

466 CH AP TE R 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

Figure 10.2 Figure 10.3 Figure 10.4

2-SampZTest input screen. 2-SampZTest Hypothesis 2-SampZTest Draw
test results. results.

TRY IT NOW GO TO EXERCISE 10.14

The following example involves a hypothesis test with a nonzero value for the hypoth-
esized difference in means, D0.

Example 10.2 Low-Carb Ice Cream

DATA SET Low-carbohydrate foods are very popular as many Americans try to avoid this sugar and
LOWCARB starch combination that many believe causes weight gain. An advertisement for a low-
carb ice cream claims that the product has 16 fewer grams of carbohydrates per serving
than the leading store brand. To check this claim, independent random samples of each
Solution Trail 10.2 type of ice cream were obtained, and the amount of carbohydrates in each serving was
measured. The data are given (in grams) in the following table.
KEY W OR DS
■ Is there any evidence? Store brand (l)
■ Difference in population
means 15.4 20.4 21.0 24.3 23.3 18.7 19.8 22.5 18.9 22.8
■ 16 grams 25.4 25.1 20.3 24.1 16.6 22.6 22.1 19.4 16.6 24.4
■ The variance is known 17.8 18.6 14.9 24.6 19.1 17.9 18.7 20.1 26.3 18.4
■ Independent random samples 21.8 17.1 21.5 19.6 22.9 22.2 21.5 18.3

T RA NSL AT I O N Low-carb brand (2)

■ Conduct a two-sided test to
3.7 3.9 4.5 4.3 3.2 3.6 3.7 3.6 3.7 4.0
compare m1 and m2
■ Large samples 4.1 3.1 4.3 3.4 3.4 3.5 4.4 4.9 3.7 3.8
■ D0 5 16 4.1 4.7 3.7 4.2 3.1 4.4 4.2 3.4 4.8 3.6
3.2 3.4 4.2 3.0 3.9
CO NCE P TS
■ Hypothesis test concerning The variance in carbohydrates per serving is known to be 8.5 for the store brand and 0.253
two population means when for the low-carb brand. Is there any evidence to suggest that the difference in population
variances are known means of carbohydrates per serving is not 16 grams? Use a 5 0.01.
VIS I ON
SOLUTION
Use the template for this
hypothesis test. The samples are STEP 1 The hypothesized difference is m1 2 m2 5 16 ( 5 D0 ) .
random, independent, and large, The sample sizes are n1 5 38 and n2 5 35.
and the population variances are
The known population variances are s21 5 8.5 and s22 5 0.253.
known. Use the two-sided
alternative hypothesis and the The significance level is a 5 0.01.
corresponding rejection region, We are testing for any difference in population means other than 16 grams of
find the value of the test statistic, carbohydrates. This is a two-sided test.
and draw a conclusion.
The samples are random and independent, and the population variances are
known. The underlying population distributions are unknown, but both sample
sizes are large ($ 30). A hypothesis test concerning two population means when
variances are known is relevant.
 10.1 Comparing Two Population Means Using Independent Samples When Population Variances Are Known 467

STEP 2 The four parts of the hypothesis test are

H0: m1 2 m2 5 16
Ha: m1 2 m2 2 16
( X1 2 X2 ) 2 16
TS: Z 5
s21 s22
1
Å n1 n2
0 0
RR: Z $ za/2 5 z0.005 5 2.5758
STEP 3 The sample means are

1
x1 5 ( 15.4 1 20.4 1 c1 18.3 ) 5 20.6579
38
1
x2 5 ( 3.7 1 3.9 1 c1 3.9 ) 5 3.8486
35
The value of the test statistic is
( x1 2 x2 ) 2 16 ( 20.6579 2 3.8486 ) 2 16
z5 5 5 1.6842
s21 s22 8.5 0.253
1 1
Å n1 n2 Å 38 35
STEP 4 The value of the test statistic, z 5 1.6842, does not lie in the rejection region.
We do not reject the null hypothesis. There is no evidence to suggest that the
difference in population mean carbohydrates is different from 16 grams at the
a 5 0.01 significance level.
This is a two-sided test and the value of the test statistic is positive, so p/2 is a
Z
right-tail probability.
p/2 5 P ( Z $ 1.6842 ) Definition of p value for a two-sided test.

5 1 2 P ( Z # 1.6842 ) The complement rule.

5 1 2 0.9539 5 0.0461 Use Table III in the Appendix.

−1.6842 0 1.6842
p 5 2 ( 0.0461 ) 5 0.0922 Solve for p.

Figure 10.5 p-Value illustration: Because p 5 0.0922 . 0.01 ( 5 a ) , we do not reject the null hypothesis. See
p 5 2P(Z $ 1.6842) Figure 10.5.
5 0.0922 . 0.01 5 a Figure 10.6 shows a technology solution.

Figure 10.6 Crunchlt! z

2-Sample Hypothesis test
results.

TRY IT NOW GO TO EXERCISE 10.16

Given the two-sample Z test assumptions and the properties of the random variable
X1 2 X2 , we can construct a confidence interval (CI) for the (difference) parameter
m1 2 m2.
468 CH AP T ER 1 0 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

As usual, to find a general CI, start with an appropriate symmetric interval about 0

▲
such that the probability Z lies in this interval is 1 2 a.
( X1 2 X2 ) 2 ( m1 2 m2 )
P £ 2za/2 , , za / 2 § 5 1 2 a (10.1)
s21 s22
1
Å n1 n2
Z
Manipulate the inequality in Equation 10.1 to sandwich the parameter m1 2 m2. We
obtain the following probability statement:

s21 s22 s21 s22

P £ ( X1 2 X2 ) 2 za/2 1 , m1 2 m2 , ( X1 2 X2 ) 1 za/2 1 § 512a
Å n1 n2 Å n1 n2

This leads to the following general result.

▲
How to Find a 100(1 ! !)% Confidence Interval
for "1 ! "2 When Variances Are Known
Given the two-sample Z test assumptions, a 100 ( 1 2 a ) % confidence interval for
m1 2 m2 has as endpoints the values

s21 s22
( x1 2 x2 ) 6 za/2 1 (10.2)
Å n1 n2

Example 10.3 No Anchovies

Pizza stones designed for home use help cooks produce baked goods with brick-oven
qualities, for example, a crusty loaf of bread or crispy-crust pizza. However, pizza stones
can be very heavy and can also take up a lot of space in a traditional residential oven.
Independent random samples of two similar types of round pizza stones were obtained,
and the weight (in pounds) of each was recorded. The summary statistics and known vari-
ances are given in the following table.
Tracy Hornbrook/[Link]
Sample Sample Population
Pizza stone size mean variance
Kitchen Depot (1) n1 5 35 x1 5 6.21 s21 5 2.1
Head Chef (2) n2 5 31 x2 5 7.08 s22 5 3.5

Find a 95% confidence interval for the difference in population mean pizza-stone weights.

SOLUTION
STEP 1 Sample sizes, sample means, and known variances are given.
The underlying weight distributions are unknown, but the sample sizes are both
large ( $ 30 ) .
1 2 a 5 0.95 1 a 5 0.05 1 a /2 5 0.025 Find a/2.

za/2 5 z0.025 5 1.960 Find the z critical value.

STEP 2 Use Equation 10.2.

s21 s22
( x1 2 x2 ) 6 za/2 1 Equation 10.2.
Å n1 n2
 10.1 Comparing Two Population Means Using Independent Samples When Population Variances Are Known 469

2.1 3.5
5 ( 6.21 2 7.08 ) 6 ( 1.96 ) 1 Use summary statistics and critical value.
Å 35 31
5 20.87 6 0.8150 Simplify.

5 ( 21.6850, 20.0550 ) Compute endpoints.

( 21.6850, 20.0550 ) is a 95% confidence interval for the difference in popu-

lation mean weights (in pounds) of the pizza stones, m1 2 m2 . This interval
represents a set of very plausible values for the difference in population mean
weights.
Figures 10.7 and 10.8 together show a technology solution.

Figure 10.7 Figure 10.8 Resulting

2-SampZInt input screen. confidence interval.

TRY IT NOW GO TO EXERCISE 10.17

Technology Corner
Procedure: Hypothesis tests and confidence intervals concerning two population means when the population variances
are known.
Reconsider: Example 10.2, solution, and interpretations.

Crunchlt!
Use the function z 2-Sample to conduct a hypothesis test concerning two population means.
1. Enter the store-brand data into column Var1 and the low-carb brand data into column Var2.
2. Select Statistics; z; 2-Sample. Under the Columns tab, select Var1 for Sample 1 and Var2 for Sample 2. Enter the
standard deviation for each group.
3. Under the Hypothesis Test tab, enter the Difference of means under null hypothesis, 16 ( 5 D0 ) . Choose the appropriate
Alternative (hypothesis).
4. Click Calculate. The results are shown in Figure 10.6.

TI-84 Plus C
Use the calculator functions 2-SampZTest and 2-SampZInt. Input is either summary statistics or data in lists.
1. Enter the store brand data into list L1 and the low-carb brand data into list L2.
2. Subtract D0 5 16 from each observation in list L1 and store the results in list L1.
3. Select STAT ; TESTS; 2-SampZTest. Highlight Data. Enter s1 , s2 , List1, and List2. Set each frequency to 1.
Highlight the alternative hypothesis. See Figure 10.9.
4. Highlight Calculate and press ENTER . The results are displayed on the Home screen. See Figure 10.10. The Draw
results are shown in Figure 10.11.
470 CH AP TE R 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

Figure 10.9 Figure 10.10 Figure 10.11

2-SampZTest input screen. 2-SampZTest Hypothesis 2-SampZTest Draw
test results. results.
5. To construct a confidence interval for the difference in population means, use the original data, in list L1.
6. Select STAT ; TESTS; 2-SampZInt. Highlight Data. Enter s1 , s2 , List1, and List2. Set each frequency to 1
and enter the C-Level. See Figure 10.12.
7. Highlight Calculate and press ENTER . The resulting confidence interval is displayed on the Home screen. See Figure 10.13.

Figure 10.12 Figure 10.13 Resulting

2-SampZInt input screen. 95% confidence interval.

Minitab
There is no built-in function to conduct hypothesis tests and construct confidence intervals concerning two population
means when the population variances are known. Remember, this is an instructive situation, not very realistic. It is pretty
unlikely we would know the population variances, but not the population means.

Excel
Use the built-in function z-test: Two Sample for Means.
1. Enter the store brand data into column A and the low-carb brand data into column B.
2. Under the Data tab, select Data Analysis; z-test: Two Sample for Means.
3. Enter the Variable 1 Range and Variable 2 Range, the Hypothesized Mean Difference, and the known population
variances. Enter the value for Alpha and choose an Output option. See Figure 10.14. Click OK.
4. Summary statistics along with the value of the test statistic, critical values, and p values are displayed. See Figure 10.15.

Figure 10.14 z-Test: Two Sample for Means input Figure 10.15 Hypothesis test
screen. results.
 10.1 Comparing Two Population Means Using Independent Samples When Population Variances Are Known 471

SECTION 10.1 EXERCISES

Concept Check Sample Sample Population
Group size mean variance
10.1 True/False A paired data set often involves a before-
and-after measurement on each individual or object. One 25 186 14.7
Two 24 190 23.8
10.2 True/False In a two-sample Z test, the hypothesized
difference in means must be 0.
a. Write the four parts of a hypothesis test of
10.3 True/False The two-sample Z test can be used only H0: m1 2 m2 5 2 versus Ha: m1 2 m2 , 2. Use a 5 0.01.
if both population variances are known. b. Compute the value of the test statistic and draw a
conclusion.
10.4 True/False In a two-sample Z test, both sample sizes c. Carefully sketch a graph to illustrate the p value
must be large. associated with this hypothesis test. Compute the p value.
10.5 True/False In a two-sample Z test, the observations 10.11 Given the two-sample Z test assumptions, consider the
may be dependent. following table of sample sizes, sample means, and known
10.6 Short Answer Given the two-sample Z test assump- variances.
tions, a 100 ( 1 2 a ) % confidence interval for m1 2 m2 has as Sample Sample Population
endpoints the values _____________. Group size mean variance
Practice One 37 1025.6 225.3
10.7 In each of the following problems, rewrite the standard-
Two 42 1031.3 107.6
notation hypothesis concerning two population means in terms a. Write the four parts of a hypothesis test of
of a difference, m1 2 m2. H0: m1 2 m2 5 0 versus Ha: m1 2 m2 2 0. Use
a. m1 5 m2 b. m1 , m2 a 5 0.001.
c. m1 2 m2 1 7 d. m1 . m2 2 4 b. Compute the value of the test statistic and draw a
e. m1 2 m2 f. m1 2 10 5 m2 conclusion.
10.8 In each of the following problems, m1, m2, s1, s2, n1, and c. Is the normality assumption necessary in order to conduct
n2 are given. Assume the underlying distributions are normal. this hypothesis test? Justify your answer.
Find the mean, variance, and standard deviation of the random 10.12 Two random samples were obtained independently and
variable X1 2 X2 , and carefully sketch the probability density the resulting data are given on the text website. Assume both
function. populations are normal with s1 5 8 and s2 5 12. EX10.12
a. m1 5 12, m2 5 9, s1 5 3, s2 5 7, n1 5 15, n2 5 11 a. Find a 95% confidence interval for the true difference in
b. m1 5 25.6, m2 5 37.8, s1 5 7.5, s2 5 10.5, means, m1 2 m2.
n1 5 10, n2 5 25 b. Using the confidence interval in part (a), is there any
c. m1 5 125.3, m2 5 250.6, s1 5 15.6, s2 5 25.6, evidence to suggest that the two population means are
n1 5 8, n2 5 12 different? Justify your answer.
d. m1 5 3.1, m2 5 2.2, s1 5 0.50, s2 5 0.75,
n1 5 21, n2 5 21 10.13 Suppose a random sample of size 15 is taken from a
normal population with mean 25 and standard deviation 5; and
10.9 Given the two-sample Z test assumptions, consider the a second, independent random sample of size 21 is taken from
following table of sample sizes, sample means, and known a normal population with mean 10 and standard deviation 4.
standard deviations. a. Describe the distribution of the difference in sample
Sample Sample Population means, X1 2 X2 (in terms of type of distribution, mean,
Group size mean standard deviation variance, and standard deviation).
b. Carefully sketch the probability distribution for X1 2 X2 .
One 18 17.5 1.5 c. Find P ( X1 2 X2 $ 17 ) .
Two 26 16.2 2.6 d. Find P ( 13.5 , X1 2 X2 , 14.5 ) .
e. Find P ( X1 , X2 1 14 ) .
a. Write the four parts of a hypothesis test of
H0: m1 2 m2 5 0 versus Ha: m1 2 m2 . 0. Use a 5 0.05.
b. Compute the value of the test statistic and draw a Applications
conclusion. 10.14 Manufacturing and Product Development The
c. Find the p value associated with this hypothesis test. efficiency of an electric toothbrush is often judged by the rota-
10.10 Given the two-sample Z test assumptions, consider the tion speed, in revolutions per minute (rpm). Two brands were
following table of sample sizes, sample means, and known selected for comparison, and independent random samples of
variances. each electric toothbrush were obtained. The rotation speed for
472 CHAP T ER 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

each toothbrush was measured, and the summary statistics are Sample Sample Population
given in the following table. State size mean variance

Electric Sample Sample Population Ohio (1) 16 427.40 8500

toothbrush size mean variance Texas (2) 24 419.50 7400
Sonicare Elite 23 7992.2 1260.25 a. Assume the underlying distributions are normal and find a
Oral-B 25 7988.2 1697.44 95% confidence interval for the true difference in
population mean amounts stolen.
a. Is there any evidence to suggest that the Sonicare Elite b. Use the confidence interval to determine whether there is
has a greater population mean rotation speed than the any evidence that the mean dollar amount of copper stolen
Oral-B? Assume normality and use a 5 0.05. differs for the two states.
b. Find the p value associated with this hypothesis test.
10.18 Marketing and Consumer Behavior A new
10.15 Business and Management Gift cards have advertising program involves placing small screens on
become a popular present. Retailers like these cards the back of taxi front seats in order to run several advertise-
because they are easier to process than paper gift certificates ments continuously. The theory is that riders give their undi-
and more difficult to forge. Customers appreciate the conve- vided attention to these ads during the entire trip. However,
nience; the cards make great stocking stuffers and are easy to advertisers worry that their ad may not be viewed during a
mail. Independent random samples of credit card–type gift cer- typical ride. Independent random samples of taxi ride times
tificates from two merchants were obtained, and the purchased (in minutes) in two cities were obtained. Is there any evidence
value (in dollars) of each was recorded. The summary statistics to suggest that the mean taxi ride time is different in San
and known variances are given in the following table. Diego and Phoenix? Assume normality, with s1 5 6.2 and
s2 5 4.9, and use a 5 0.01. Write a Solution Trail for this
Sample Sample Population
problem. TAXIRIDE
Store size mean variance
10.19 Manufacturing and Product Development The
Nordstrom 41 24.07 16.81
total weight (with the case) of a portable sewing machine is an
Macy’s 38 26.61 10.24 important consideration. Suppose Singer claims to have the
lightest machine by five pounds. Independent random samples
Is there any evidence to suggest that the true mean Nordstrom of a Singer machine and a comparable Simplicity machine were
gift-certificate purchased value is different from the true mean obtained, and the weight (in pounds) of each was recorded. The
Macy’s gift-certificate purchase value? Use a 5 0.01. Write a summary statistics and known variances are given in the fol-
Solution Trail for this problem. lowing table.
10.16 Manufacturing and Product Development The
Sewing Sample Sample Population
energy rating, water consumption, and noise level of an elec-
machine size mean variance
tric dishwasher are all-important selling features. Suppose the
makers of the Hotpoint DF55 claim that this model has a Simplicity 42 17.99 2.89
lower noise-level rating than any other comparable dish- Singer 38 13.26 2.25
washer. Independent random samples of the Hotpoint DF55
and of a similar Maytag dishwasher were obtained, and the a. Is there any evidence to refute the claim made by Singer?
noise level (in decibels) was measured for each. Is there any Use a 5 0.01.
evidence to suggest that the population mean noise level for b. Find the p value associated with this hypothesis test.
the Hotpoint dishwasher is less than the population mean c. Is the normality assumption necessary in this problem?
noise level for the Maytag? Assume the underlying distribu- Why or why not?
tions are normal, with s1 5 3.75 and s2 5 4.14. Use
a 5 0.05. DISHWASH
10.20 Medicine and Clinical Studies Many people con-
sume protein shakes to help build muscle mass and eliminate
10.17 Copper Thieves As a result of increasing commodity body fat. In a recent study, the amount of protein in two com-
prices, the theft of certain metals has increased dramatically. peting drinks was compared. Independent random samples
Because copper is used in so many items and is difficult to were obtained, and the protein content (in grams) in each drink
trace, thieves across the United States have become more bra- was measured. The summary statistics and known variances are
zen. A major target for thieves has been copper wiring in elec- given in the following table.
trical power substations and utility poles. The leading states for
thefts are Ohio, Texas, and Georgia.2 Independent random Protein Sample Sample Population
samples of copper theft reports were obtained from two states, drink size mean variance
and the estimated dollar amount was recorded for each. The Met-Rx 12 39.38 5.06
summary statistics and known variances are given in the
Pure Gro 24 39.01 6.01
following table.
 10.1 Comparing Two Population Means Using Independent Samples When Population Variances Are Known 473

Is there any evidence to suggest that the mean amount of pro- a. Assume the underlying distributions are normal. Is there
tein is different in these two products? Use a 5 0.01 and any evidence to suggest that the population mean
assume normality. airspeeds are different? Use a 5 0.05.
b. Find the p value associated with this hypothesis test.
10.21 Manufacturing and Product Development Several
factors determine how well a ceiling fan cools a room, includ- 10.24 Medicine and Clinical Studies The time it takes for
ing blade pitch, height from the ceiling, and revolutions per general anesthesia to work (time to induction) is an important
minute. Independent random samples of two types of ceiling consideration during an emergency and for scheduled surgeries.
fans were obtained, and the revolutions per minute (on high) for Recently, a study was conducted to compare the mean induction
each was measured. The summary statistics and known vari- time of similar drugs administered via inhalation and intrave-
ances are given in the following table. nously. Independent random samples of patients requiring
general anesthesia were obtained, and the induction times (in
Sample Sample Population minutes) were measured. Assume the variance in induction time
Ceiling fan size mean variance for inhalation administration is 0.0625 and for intravenous
Hampton 34 295.05 11.55 administration is 0.1225. Is there any evidence to suggest that
Altura 35 300.38 6.25 the mean time to induction for intravenous administration is
less than the mean time to induction for inhalation administra-
a. Find a 99% confidence interval for the true mean tion? Use a 5 0.05. ANESTH

difference in revolutions per minute, m1 2 m2.

b. Using the interval in part (a), is there any evidence to Extended Applications
suggest that the mean revolutions per minute for the
10.25 Public Health and Nutrition Magnesium is used by
Altura ceiling fan is greater than the mean revolutions per
every cell in your body, is required for over 300 biochemical
minute for the Hampton ceiling fan? Justify your answer.
reactions, and helps muscles and nerves function properly.
10.22 Travel and Transportation The recommended tire According to the U.S. Department of Agriculture National
pressure for an off-road bicycle depends on the weight of Nutritional Database, 12 cup of vegetarian baked beans and one
the rider. As you would expect, the greater the rider’s weight, medium baked potato without the skin contain the same amount
the greater the recommended tire pressure. At a well-used bicy- of magnesium (40 milligrams). To check this claim, indepen-
cle trail in the Black River State Forest in Wisconsin, indepen- dent random samples of baked beans and potatoes were
dent random samples were obtained from two different weight obtained, and the amount of magnesium in each serving was
groups. The front tire pressure (in psi) was measured for each recorded (in milligrams). The summary statistics and known
person, and the summary statistics and known variances are variances are given in the following table.
given in the following table.
Sample Sample Population
Weight Sample Sample Population Food size mean variance
group size mean variance Vegetarian 18 39.58 2.47
< 150 pounds 18 38.91 2.25 baked beans (1)
< 180 pounds 23 41.99 6.25 Medium 18 40.12 0.87
potato (2)
Is there any evidence to suggest that the difference between the
180-pound riders’ mean tire pressure and the 150-pound riders’ a. Assume the underlying distributions are normal. Is there
mean tire pressure is greater than 3 psi? Assume normality and any evidence to refute the claim? Use a 5 0.01.
use a 5 0.05. b. Suppose that, instead, the sample sizes are n1 5 n2 5 38.
Now, is there any evidence to refute the claim? Find the p
10.23 Manufacturing and Product Development value for this hypothesis test.
Although they are called leaf blowers, these hand-held c. How large would the sample sizes (n1 5 n2) have to be for
machines are used to sweep patios, clean driveways, and even the hypothesis test to be significant at the a 5 0.01 level?
move light snow. Leaf blowers are often compared using
weight, noise, and airspeed. Independent random samples of 10.26 Physical Sciences The manufacturer of a Kenmore
two types of leaf blowers were obtained, and the airspeed residential stove can order parts from two different suppliers:
(in mph) was measured for each. The summary statistics and The Repair Clinic and The Parts Pros. The small burners, or
the known variances are given in the following table. elements, are designed to produce 7.4 kilowatts at 240 volts.
To decide which supplier to use, independent random samples
Leaf Sample Sample Population from each supplier were obtained, and each element’s output
blower size mean variance (in kilowatts) was carefully measured. For The Repair Clinic,
n1 5 12, x1 5 7.361, and s21 5 0.81; for The Parts Pros,
Craftsman 18 200.28 24.5
n2 5 15, x2 5 7.307, and s22 5 0.64. Assume the underlying
Echo 19 196.74 35.7 distributions are normal and use a 5 0.05 for the following.
474 CHAP T E R 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

a. Is there any evidence to suggest that the population mean Florida still has the most flood insurance policies, followed by
output of elements from The Repair Clinic is different Texas, then Louisiana. A random sample of flood insurance
from 7.4? policies in these three states was obtained, and the premium in
b. Is there any evidence to suggest that the population dollars for each was recorded. The summary statistics are given
mean output of elements from The Parts Pros is different in the following table.4
from 7.4?
c. Is there any evidence to suggest that m1 is different from Sample Sample Population
m2? State size mean standard deviation
d. Using the results from parts (a), (b), and (c), which Louisiana 18 716.15 250
supplier should the manufacturer use? Florida 22 498.71 275
10.27 Marketing and Consumer Behavior The Press Texas 26 560.50 300
Association Mediapoint recently released a housing market
report for England that included how long it would take a Assume the underlying populations are normal. Is there any
typical first-time buyer to save for a deposit in their local area.3 evidence to suggest that any pairs of population mean flood
Independent random samples of first-time buyers were obtained insurance premiums are different? That is, conduct three sepa-
in two areas, and the time (in years) needed for a couple to save rate hypothesis tests to consider m1 2 m2, m1 2 m3, and
for the deposit was recorded for each. The summary statistics m2 2 m3. Use a 5 0.05 in each case.
are given in the following table.
Challenge
Sample Sample Population
Area size mean variance 10.29 Sample Size Calculation Suppose a 100 ( 1 2 a ) %
confidence interval is needed for the difference in two popula-
Yorkshire & The Humer 60 4.5 1.56
tion means, m1 2 m2. In addition, suppose the underlying popu-
East Midlands 75 4.8 3.24
lations are normal, the population variances, s21 and s22, are
known, and the samples sizes are equal, n1 5 n2 5 n.
a. Is there any evidence to suggest that the mean time
a. Find an expression for the sample size necessary (from
needed for a couple to save for a deposit is different in
each population) in order for the resulting confidence
these two areas? Use a 5 0.05.
interval to have a bound on the error of estimation B
b. Find the p value associated with this hypothesis test.
(half the width of the confidence interval).
c. The sample means, 4.5 and 4.8, seem close together. Can
b. How large a sample size is necessary if s1 5 12.7,
you find a value n 5 n1 5 n2 such that the hypothesis test
s2 5 9.5, B 5 5, and the confidence level is 95%?
is significant at the a 5 0.05 level?
c. Use the sample size in part (b) with x1 5 57.3 and
10.28 Flood Insurance According to the U.S. National x2 5 48.6 to construct a 95% confidence interval for
Flood Insurance Program, the cost of policies in New Orleans m1 2 m2. Compute the exact bound on the error of
continued to rise for several years after Hurricane Katrina. estimation. How does this compare with B 5 5?

10.2 Comparing Two Population Means Using Independent

Samples from Normal Populations
In Section 10.1, the hypothesis tests concerning the difference between two population
means (or for comparing two population means) were based on the standard normal, or Z,
distribution. These tests are valid only if both population variances are known (and with
normality and/or large samples, and independent random samples). It is unrealistic to
assume that the population variances are known. As in Chapter 9, we will assume that the
underlying populations are normal. But one additional assumption is necessary to con-
struct a similar two-sample t test.
Suppose that

For reference, these are the 1. X1 is the mean of a random sample of size n1 from a normal population with mean m1.
two-sample t test assumptions. 2. X2 is the mean of a random sample of size n2 from a normal population with mean m2.

A test of equality of population 3. The samples are independent.

variances will be discussed in 4. The two population variances are unknown but equal. The common variance is denoted
Section 10.5. s2 ( 5 s21 5 s22 ) .
 10.2 Comparing Two Population Means Using Independent Samples from Normal Populations 475

The last assumption is new and implies we are comparing populations with the same
variability. If we do not assume equal variances, there is no nice test procedure. More on
this later.

Properties of X1 ! X2
If the two-sample t test assumptions are true, then the estimator X1 2 X2 has the following
properties.

1. E ( X1 2 X2 ) 5 m X1 2X2 5 m1 2 m2
X1 2 X2 is still an unbiased estimator of the parameter m1 2 m2.
s21 s22 s2 s2 1 1
2. Var ( X1 2 X2 ) 5 s2X1 2X 2 5 1 5 1 5 s2 a 1 b and the standard
n1 n2 n1 n2 n1 n2
1 1
deviation is sX 2X 5 s2 a 1 b.
1 2
Å n1 n2
3. Both underlying populations are normal, so the distribution of X1 2 X2 is also normal.

In the previous section, we used the known population variances, standardized, and
constructed a test based on the Z distribution. Here, an estimate of the common variance
s2 is necessary. The appropriate standardization results in a t distribution.
S 21 and S 22 are separate estimators for the common variance, but using only one of these
means ignoring additional, useful information. Because s2 is the variance for both under-
lying populations, an estimator for this common variance should depend on both samples.
However, it also seems reasonable for the estimator to rely more on the larger sample.
Therefore an estimate of the common variance uses both S12 and S22 in a weighted average.

Definition
The pooled estimator for the common variance s2, denoted Sp2, is
( n1 2 1 ) S 21 1 ( n2 2 1 ) S 22
S 2p 5 (10.3)
n1 1 n2 2 2
n1 2 1 n2 2 1
5a bS 21 1 a bS 22
n1 1 n2 2 2 n1 1 n2 2 2

The pooled estimator for the common standard deviation s is Sp 5 "S 2p .

A CLOSER L OK
l is the lowercase Greek letter 1. S 2p is indeed a weighted average. This estimator can be written in the form
lambda and represents a constant.
Sp2 5 lS 21 1 ( 1 2 l ) S 22 where 0#l#1
1 1 2 1 2
If n1 5 n2, then l 5 and 2 5 Sp2
1 If n1 2 n2, then more weight is given
2 S1 2S 2.
to the larger sample.
2. The constants in Equation 10.3 are related to the number of degrees of freedom. S 21
contributes n1 2 1 degrees of freedom and S 22 contributes n2 2 1 degrees of freedom.
Consequently, there are a total of ( n1 2 1 ) 1 ( n2 2 1 ) 5 n1 1 n2 2 2 degrees of
freedom associated with the estimator S 2p .

The hypothesis test procedure is based on the following theorem.

476 CHAP T E R 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

Theorem
If the two-sample t test assumptions are true, then the random variable
( X1 2 X2 ) 2 ( m1 2 m2 )
T5
1 1
S 2p a 1 b
Å n1 n2
has a t distribution with n1 1 n2 2 2 degrees of freedom.

As in a two-sample Z test, the null and alternative hypotheses are stated in terms of the
difference m1 2 m2. The critical values are from the appropriate t distribution.

Hypothesis Tests Concerning Two Population Means

When Variances Are Unknown but Equal
Given the two-sample t test assumptions, a hypothesis test concerning two population
means in terms of the difference in means m1 2 m2, with significance level a, has the
form
H0: m1 2 m2 5 D0
This is the template for a Ha: m1 2 m2 . D0, m1 2 m2 , D0, or m1 2 m2 2 D0
hypothesis test concerning two ( X1 2 X2 ) 2 D0
population means when variances TS: T 5
are unknown but equal: a 1 1
S 2p a 1 b
two-sample t test, with pooled Å n1 n2
variance. RR: T $ ta, n11 n222, T # 2 ta,n11 n222, or 0 T 0 $ ta/2,n11 n222

Example 10.4 Surgical Wait Times

Frequently, patients must wait a long time for elective surgery. For those with chronic
pain, the wait can be unbearable. Suppose the wait time for patients needing a knee
replacement at two hospitals in British Columbia was investigated. Independent random
samples of patients were obtained, and the wait time for each (in weeks) was recorded.
The resulting summary statistics are given in the following table.5

Sample Sample Sample

Hospital size mean variance
Abbotsford Regional (1) 15 17.4 34.81
Earl Ridge (2) 17 12.1 46.24

a. Is there any evidence to suggest that there is a difference in the population mean wait-
ing time for a knee replacement between Abbotsford Regional Hospital and Earl Ridge
Hospital? Use a 5 0.05 and assume the underlying distributions are normal, with
equal variances.
b. Find bounds on the p value associated with this hypothesis test.

SOLUTION
STEP 1 Let Abbotsford Regional Hospital be population 1 and Earl Ridge Hospital be
population 2.
The null hypothesis is that the two population means are equal—with the same
waiting time for a knee replacement: m1 5 m2 1 m1 2 m2 5 0 ( 5 D0 ) .
 10.2 Comparing Two Population Means Using Independent Samples from Normal Populations 477

Solution Trail 10.4 The summary statistics are given, and the samples were obtained independently.
The population variances are unknown but assumed equal. A two-sample t test is
K EY WO R D S appropriate.
■ Is there any evidence? We are looking for any difference in population means, so this is a two-sided test.
■ Difference in population mean
STEP 2 The four parts of the hypothesis test are
■ Underlying distributions are
normal, with equal variances H0: m1 2 m2 5 0
■ Independent random samples Ha: m1 2 m2 2 0
( X1 2 X2 ) 2 0
TR A NSL ATI O N TS: T 5
Conduct a two-sided test to 1 1
■
S 2p a 1 b
compare m2 and m2 Å n1 n 2
■ Variances are unknown but RR: 0 T 0 $ ta/2,n11 n222 5 t0.025,30 5 2.0423
assumed equal
STEP 3 The pooled estimate of the common population variance is
CON C EPT S ( n1 2 1 ) s21 1 ( n2 2 1 ) s22 ( 14 ) ( 34.81 ) 1 ( 16 ) ( 46.24 )
■ Hypothesis test concerning s2p 5 5 5 40.906
n1 1 n2 2 2 30
two population means when
variances are unknown but The value of the test statistic is
equal
( x1 2 x2 ) 2 0 17.4 2 12.1
V I SI O N
t5 5 5 2.3393 ( $ 2.0423 )
1 1 1 1
Use the template for this s2p a 1 b ( 40.906 ) a 1 b
hypothesis test. The samples are
Å n1 n2 Å 15 17
random and independent, the The value of the test statistic, t 5 2.3393, lies in the rejection region, hence we
underlying distributions are reject the null hypothesis at the a 5 0.05 significance level. There is evidence to
normal, and the population suggest the mean waiting time for a knee replacement is different at Abbotsford
variances are unknown but
Regional Hospital and Earl Ridge Hospital.
assumed equal. Use the
two-sided alternative hypothesis STEP 4 Recall, because of the nature of the table of critical values for t distributions, we
and the corresponding rejection can only bound the p value.
region.
0 t 0 5 0 2.3393 0 5 2.3393
In Table V in the Appendix, row n1 1 n2 2 2 5 15 1 17 2 2 5 30, place
t30 2.3393 in the ordered list of critical values.
2.0423 # 2.3393 # 2.4573
t0.025,30 # 2.3444 # t0.01,30
Therefore, 0.01 # p/2 # 0.025
And 0.02 # p # 0.05
−2.3393 0 2.3393
See Figure 10.16.
Figure 10.16 p Value illustration:
p 5 2P(T $ 2.3393) Figures 10.17 through 10.19 together show a technology solution.
5 0.0262 # 0.05 5 a

VIDEO TECH MANUALS

EXEL SAMPLE
TWO DISCRIPTIVE
MEAN
INFERENCE - t -
SUMMARIZED DATA

Figure 10.17 Figure 10.18 Figure 10.19

2-SampTTest input 2-SampTTest Hypothesis 2-SampTTest Draw
screen. test results. results.
TRY IT NOW GO TO EXERCISE 10.45
478 CH AP T ER 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

Example 10.5 Weight of Aluminum Cans

DATA SET Aluminum cans are made from huge solid ingots pressed under high-pressure rollers and
ALUMCAN are cut like cookies from thin sheets. Aluminum is ideal for cans because it is lightweight,
strong, and recyclable. A company claims that a new manufacturing process decreases the
Why do you suppose a small amount of aluminum needed to make a can, and therefore, decreases the weight. Indepen-
significance level is important dent random samples of aluminum cans made by the old and new processes were obtained,
here? and the weight (in ounces) of each is given in the following table.

Old process (1)

Solution Trail 10.5
0.52 0.49 0.47 0.47 0.48 0.52 0.55 0.49 0.52 0.50 0.50
KEY W OR DS 0.50 0.51 0.51 0.50 0.53 0.49 0.51 0.52 0.51 0.51
■ Is there any evidence?
■ Smaller population mean New process (2)
■ Populations are normal, with 0.51 0.51 0.50 0.48 0.47 0.49 0.46 0.46 0.52 0.50 0.48
equal variances
0.51 0.50 0.48 0.51 0.44 0.48 0.47 0.50 0.51 0.48
■ Independent random samples

T RA NSL AT I O N Is there any evidence that the new-process aluminum cans have a smaller population
■ Conduct a one-sided test to mean weight? Assume the populations are normal, with equal variances, and use a 5 0.01.
compare m1 and m2
■ Variances are unknown SOLUTION
but equal STEP 1 The null hypothesis is that the two mean weights are the same: m1 2 m2 5 0. We
are looking for evidence that the new-process cans have a smaller mean weight.
CO NCE P TS
The alternative hypothesis is m1 2 m2 . 0.
■ Hypothesis test concerning
two population means when The underlying populations are assumed normal with equal variances, and the
variances are unknown but samples were obtained independently. A two-sample t test is relevant.
equal STEP 2 The four parts of the hypothesis test are

VIS I ON H0: m1 2 m2 5 0
Use the template for this Ha: m1 2 m2 . 0
hypothesis test. The samples are ( X1 2 X2 ) 2 0
random and independent, the TS: T 5
underlying distributions are 1 1
normal, and the population S 2p a 1 b
Å n1 n2
variances are unknown but RR: T $ ta,n11 n222 5 t0.01,40 5 2.4233
assumed equal. Use a one-sided
alternative hypothesis and the STEP 3 The summary statistics are
corresponding rejection region.
1
x1 5 ( 0.52 1 0.49 1 c1 0.51 ) 5 0.5048
21
1
x2 5 ( 0.51 1 0.51 1 c1 0.48 ) 5 0.4886
21
1 1
s21 5 c 5.3580 2 ( 10.6 ) 2 d 5 0.0003762
20 21
1 1
s22 5 c 5.0216 2 ( 10.26 ) 2 d 5 0.0004429
20 21
The pooled estimate of the common population variances is
( 20 ) ( 0.0003762 ) 1 ( 20 ) ( 0.0004429 )
s2p 5 5 0.0004095
40
The value of the test statistic is
( x1 2 x2 ) 2 0 0.5048 2 0.4886
t5 5 5 2.5941
1 1 1 1
s2p a 1 b ( 0.0004095 ) a 1 b
Å n1 n2 Å 21 21
 10.2 Comparing Two Population Means Using Independent Samples from Normal Populations 479

STEP 4 The value of the test statistic lies in the rejection region (t 5 2.5941 $ 2.4233;
p 5 0.0066 # .01, see Figure 10.20). We reject the null hypothesis at the
a 5 0.01 significance level. There is evidence to suggest that new-process alu-
minum cans have a smaller mean weight.

t40

Figure 10.20 p-Value illustration:

p 5 P(T $ 2.5941)
5 0.0066 # 0.01 5 a 0 2.5941

Figures 10.21 and 10.22 show technology solutions.

Figure 10.21 Minitab hypothesis test (and confidence Figure 10.22 JMP two-sample t test.
interval) results.

TRY IT NOW GO TO EXERCISE 10.47

This methodology has been used Using the assumptions presented in this section and the technique presented in Section
several times, beginning in 10.1, a confidence interval for m1 2 m2 can be derived. Start with a symmetric interval
Chapter 8. about 0 such that the probability T lies in this interval is 1 2 a. Manipulate the inequal-
ity to sandwich the parameter m1 2 m2.

How to Find a 100(1 ! !)% Confidence Interval

for "1 ! "2 When Variances Are Unknown but Equal
Given the two-sample t test assumptions, a 100 ( 1 2 a ) % confidence interval for m1 2 m2
has as endpoints the values

1 1
( x1 2 x2 ) 6 ta/2,n1 1 n222 s2p a 1 b (10.4)
Å n1 n2

Example 10.6 Iron Man

VIDEO TECH MANUALS Iron is an essential mineral. It is used by the body to carry oxygen, and even a slight defi-
EXEL SAMPLE
TWO DISCRIPTIVE
MEAN ciency can cause fatigue and weakness. Certain kinds of mollusks are very high in iron
INFERENCE - t - WITH content, for example, clams, mussels, and oysters. Independent random samples of
DATA.
3-ounce servings of clams and oysters were obtained, and the iron content (in mg) was
measured in each.
480 CH AP TER 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

The summary statistics are given in the following table.

Sample Sample Sample

Mollusk size mean standard deviation
Clams (1) 12 23.17 4.38
Oysters (2) 15 24.19 3.35

Assume the populations are normal and the variances are equal. Find a 99% confidence
interval for the difference in population mean iron content.6

SOLUTION
STEP 1 The summary statistics are given, the underlying distributions are assumed normal,
and the population variances are assumed equal. Equation 10.4 can be used to
construct a confidence interval for the difference m1 2 m2.
1 2 a 5 0.99 1 a 5 0.01 1 a /2 5 0.005 Find a/2.

ta/2,n11 n222 5 t0.005,25 5 2.7874 Find the t critical value.

STEP 2 Find the pooled estimate of the common variance.

( n1 2 1 ) s21 1 ( n2 2 1 ) s22 ( 11 ) ( 4.38 ) 2 1 ( 14 ) ( 3.35 ) 2
s2p 5 5 5 14.7257
n1 1 n2 2 2 25
STEP 3 Use Equation 10.4.

1 1
( x1 2 x2 ) 6 ta/2 # s2p a 1 b Equation 10.4.
Å n1 n2
1 1
5 ( 23.17 2 24.19 ) 6 ( 2.7874 ) ( 14.7257 ) a 1 b
Å 12 15
Use summary statistics and critical values.

5 21.02 6 4.1427 Simplify.

5 ( 25.1627, 3.1227 ) Compute endpoints.

( 25.1627, 3.1227 ) is a 99% confidence interval for the difference (in mg) in
population mean iron content, m1 2 m2. Note that because 0 is included in, or
captured by, this interval, there is no evidence to suggest the mean iron content
is different.
Figures 10.23 and 10.24 together show a technology solution.

Figure 10.23 Figure 10.24 Resulting

2-SampTInt input screen. 99% confidence interval.

TRY IT NOW GO TO EXERCISE 10.52

The underlying distributions might
only be approximately normal, or The hypothesis test procedure and the confidence interval formula presented in this
the population variances might section are robust. That is, if the assumptions aren’t entirely true, the hypothesis test and
not be exactly the same. the confidence interval are still very reliable. Even if the population variances are very
 10.2 Comparing Two Population Means Using Independent Samples from Normal Populations 481

different, as long as the underlying populations are normal and n1 5 n2, the results are
still very reliable.
If the underlying populations are normal, the population variances are unequal, and
Nice means a reasonable the sample sizes are different, there is no nice test procedure concerning m1 2 m2 (or
standardization to produce a confidence interval for m1 2 m2 ). It is reasonable to use each sample variance as an
common random variable. approximation for the corresponding population variance. However, the resulting log-
ical standardization produces only an approximate test statistic. If the sample sizes are
small and the underlying populations are not normal, then a nonparametric test must
be used.

Hypothesis Tests and Confidence Interval Concerning

This is the template for an Two Population Means When Variances Are Unknown
approximate two-sample t test. and Unequal
Given the modified two-sample t test assumptions (population variances unknown and
assumed unequal), an approximate hypothesis test concerning two population means in
terms of the difference, m1 2 m2, with significance level a, has the form
H0: m1 2 m2 5 D0
Ha: m1 2 m2 . D0, m1 2 m2 , D0, or m1 2 m2 2 D0
( X1 2 X2 ) 2 D0
TS: Tr 5
S 21 S 22
1
Å n1 n2
The formula for n is the RR: Tr $ ta,v, Tr # 2ta,v, or 0 Tr 0 $ ta/2,v
Satterthwaite approximation for s21 s22 2
the number of degrees of a 1 b
n1 n2
freedom. where v <
( s21 /n1 ) 2 ( s22 /n2 ) 2
1
n1 2 1 n2 2 1
An approximate 100 ( 1 2 a ) % confidence interval for m1 2 m2 has as endpoints the
values

s21 s22
( x1 2 x2 ) 6 ta/2,v 1 (10.5)
Å n1 n2

A CLOSER L OK
1. The random variable Tr has an approximate t distribution with v degrees of freedom.
2. It is likely that the value of v will not be an integer. To be conservative, always round
down (to the nearest integer).
3. A test for equality of population variances is presented in Section 10.5. This hypothe-
sis test is often used to determine whether equal population variances is a reasonable
assumption.

Example 10.7 Poker Chip Weights

DATA SET Clay-composite poker chips used in Las Vegas and Atlantic City weigh between 8.5 and
POKER 10 grams each, and last between 3 and 6 years. In a recent study of poker-chip weights, a
casino obtained independent random samples of $100 and $500 chips. The weight of each
chip (in grams) is given in the following table.
482 CHA PT ER 1 0 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

Solution Trail 10.7 $100 chips (1)

9.17 9.21 9.25 9.29 9.16 9.08 9.39 9.23 9.15 9.14
KEY W OR DS
■ Is there any evidence? 9.34 9.26 9.08 9.11
■ Population mean is different $500 chips (2)
■ Both populations are normal
■ Independent random samples
9.37 9.98 9.04 8.74 9.58 9.45 9.08 9.96 9.69

T RA NSL AT I O N Is there any evidence to suggest that the population mean weight of $100 chips is different
■ Conduct a two-sided test to from that of $500 chips? Assume both populations are normal, and use a 5 0.05.
compare m1 and m2
■ Variances are unknown and SOLUTION
assumed unequal STEP 1 The null hypothesis is that the two mean weights are the same, and the alterna-
tive is two-sided. The underlying populations are assumed normal and the sam-
CO NCE P TS
ples were obtained independently. However, there is no assumption of equal
■ Hypothesis test concerning variances. The approximate two-sample t test is appropriate.
two population means when
variances are unknown and STEP 2 The summary statistics are
unequal 1
x1 5 ( 9.17 1 9.21 1 c1 9.11 ) 5 9.2043
VIS I ON 14
1
Use the template for this x2 5 ( 9.37 1 9.98 1 c1 9.69 ) 5 9.4322
hypothesis test. The samples are 9
random and independent, the 1 1
underlying distributions are s21 5 c 1186.1804 2 ( 128.86 ) 2 d 5 0.008934
13 14
normal, and the population
1 1
variances are unknown and s22 5 c 802.1295 2 ( 84.89 ) 2 d 5 0.1785
unequal. Use the two-sided 8 9
alternative hypothesis and the The approximate number of degrees of freedom are
corresponding rejection region,
the test statistic, and draw a 0.008934 0.1785 2
conclusion. a 1 b
14 9
v< 5 8.5177
( 0.008934/14 ) 2 ( 0.1785/9 ) 2
1
13 8
We round v down to 8.
STEP 3 The four parts of the hypothesis test are

H0: m1 2 m2 5 0
Ha: m1 2 m2 2 0
( X1 2 X2 ) 2 0
TS: Tr 5
S 21 S 22
1
Å n1 n2
RR: 0 Tr 0 $ ta/2,v 5 t0.025,8 5 2.3060
STEP 4 The value of the test statistic is
( x1 2 x2 ) 2 0 9.2043 2 9.4322
tr 5 5 5 21.5928
s21 s22 0.008934 0.1785
1 1
Å n1 n2 Å 14 9
The value of the test statistic does not lie in the rejection region. Equivalently,
p 5 0.1475 . 0.05, illustrated in Figure 10.25. We do not reject the null hypoth-
esis at the a 5 0.05 significance level. There is no evidence to suggest that the
mean weight of $100 chips is different from the mean weight of $500 chips.
Figure 10.26 shows a technology solution.
 10.2 Comparing Two Population Means Using Independent Samples from Normal Populations 483

t8

−1.5928 0 1.5928
Figure 10.25 p Value illustration: Figure 10.26 Excel hypothesis test results.
p 5 2P(T $ 1.5928)
5 0.1475 . 0.05 5 a

TRY IT NOW GO TO EXERCISE 10.55

Technology Corner
Procedure: Hypothesis tests and confidence intervals concerning two population means when the population variances
are unknown.
Reconsider: Example 10.5, solution, and interpretations.

Crunchlt!
Use the built-in function t 2-Sample. Input is either summary statistics or data in columns.
1. Enter the old process data into column Var1 and the new process data into column Var2.
2. Select Statistics; t; 2-sample. Using the pull-down menus, select Var1 for Sample 1 and Var2 for Sample 2. Check the
Pooled Variance box.
3. Under the Hypothesis tab, enter the Difference of means under null hypothesis (0) and select the appropriate Alternative
(Greater than). See Figure 10.27.
4. Click Calculate. The results are displayed in a new window (Figure 10.28).
5. Use the Confidence Interval tab to construct a confidence interval for the difference of two population means.

Figure 10.27 t 2-Sample input Figure 10.28 t 2-Sample results.

screen.
484 CH AP T ER 1 0 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

TI-84 Plus C
Use the built-in functions 2-SampTTest and 2-SampTInt. Input is either summary statistics or data in lists.
1. Enter the old process data into list L1 and the new process data into list L2.
2. Select STAT ; TESTS; 2-SampTTest. Highlight Data. Enter List1, List2, and set each frequency to 1.
Highlight the alternative hypothesis and Yes for Pooled. See Figure 10.29.
3. Highlight Calculate and press ENTER . The results are displayed on the Home screen. See Figure 10.30.
4. The Draw results are shown in Figure 10.31.
5. Use the function STAT ; TESTS; 2-SampTInt to construct a confidence interval for the difference of two
population means.

Figure 10.29 Figure 10.30 Figure 10.31

2-SampTTest input 2-SampTTest hypothesis 2-SampTTest Draw
screen. test results. results.

Minitab
Use the built-in function 2-Sample t to conduct a hypothesis test and to construct a confidence interval. Input is
either data in one or two columns (data in one column requires a subscript, or group-identifying, column) or
summarized data.
1. Enter the old process data into column C1 and the new process data into column C2.
2. Select Stat; Basic Statistics; 2-Sample t.
3. Choose Each sample is in its own column and enter C1 in the Sample 1 input window and C2 in the Sample 2 input
window.
4. Choose the Options option button. Enter a Confidence level, the hypothesized Test difference, and choose the appropri-
ate Alternative, check the Assume equal variances box.
5. The hypothesis test results and confidence interval are displayed in a session window. Refer to Figure 10.21.

Excel
The Data Analysis toolkit contains two functions for comparing population means, assuming equal variances and
assuming unequal variances. Use the appropriate formula and ordinary spreadsheet calculations to find the endpoints
of a confidence interval.
1. Enter the old process data into column A and the new process data into column B.
2. Under the Data tab, select Data Analysis; t-Test: Two-Sample Assuming Equal Variances.
3. Enter the Variable 1 Range, Variable 2 Range, the Hypothesized Mean Difference, and the value for a. Choose
an Output option and click OK.
4. Summary statistics along with the value of the test statistic, critical values, and p values are displayed. See
Figure 10.32.
 10.2 Comparing Two Population Means Using Independent Samples from Normal Populations 485

Figure 10.32 Excel

hypothesis test results.

SECTION 10.2 EXERCISES

Concept Check a. Conduct a hypothesis test of H0: m1 2 m2 5 0 versus
Ha: m1 2 m2 , 0. Use a 5 0.05.
10.30 Short Answer State the two-sample t test b. Find bounds on the p value associated with this test.
assumptions.
10.38 Given the two-sample t test assumptions, consider the
10.31 Short Answer Under the two-sample t test assump- following table of summary statistics.
tions, Var ( X1 2 X2 ) 5 _____________.
10.32 True/False In a two-sample t test, the pooled estima- Sample Sample Sample
tor for the common variance is the sample variance associated Group size mean standard deviation
with the larger sample. One 10 156.5 26.5
10.33 Short Answer The two-sample t test is robust. Two 11 132.6 21.5
Explain what this means in practice.
a. Conduct a hypothesis test of H0: m1 2 m2 5 0 versus
10.34 Short Answer If the population variances are
Ha: m1 2 m2 . 0. Use a 5 0.01.
unequal, there is no nice test procedure to compare the popula-
b. Find bounds on the p value associated with this test.
tion means. Explain what nice means in terms of statistics.
10.39 Given the two-sample t test assumptions, consider
10.35 True/False Consider a hypothesis test or confidence
the independent random samples from two different
interval concerning two population means when the variances
populations. EX10.39
are unknown and unequal. If the approximate number of
a. Conduct a hypothesis test of H0: m1 2 m2 5 0 versus
degrees of freedom is a decimal, round down to the nearest
Ha: m1 2 m2 2 0. Use a 5 0.05.
integer.
b. Find bounds on the p value associated with this test.
10.36 True/False In a hypothesis test or confidence interval
10.40 Given the two-sample t test assumptions, consider the
concerning two population means when the variances are
following table of summary statistics.
unknown and unequal, the hypothesized difference in means
must always be 0.
Sample Sample Sample
Group size mean standard deviation
Practice
One 23 49.03 9.24
10.37 Given the two-sample t test assumptions, consider the Two 23 49.57 8.15
following table of summary statistics.
Sample Sample Sample a. Find a 95% confidence interval for the difference in
Group size mean variance population means, m1 2 m2.
b. Using the confidence interval in part (a), is there any
One 14 49.6 134.56
evidence to suggest that the two population means are
Two 16 50.2 243.36 different? Justify your answer.
486 CH AP T ER 1 0 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

10.41 In each of the following problems, n1, n2, s1, and s2 are Suppose independent random samples of rotors from two
given. Assume normal underlying distributions, independent different manufacturers were obtained. The largest deviation
random samples, and unknown, unequal variances. Find the from perfect flatness of each rotor was measured, and the
approximate number of degrees of freedom, v, in the critical resulting summary statistics are given in the following table.
value of an approximate two-sample t test.
a. n1 5 12, n2 5 15, s1 5 11.7, s2 5 16.7 Sample Sample Sample
b. n1 5 8, n2 5 23, s1 5 5.46, s2 5 6.78 Manufacturer size mean standard deviation
c. n1 5 18, n2 5 26, s1 5 57.8, s2 5 49.9 Tire Rack 11 26.74 8.31
d. n1 5 32, n2 5 34, s1 5 5.51, s2 5 5.03 JC Whitney 14 29.53 6.85
10.42 Consider the following table of summary statistics.
Assume the underlying distributions are normal and the popula-
Sample Sample Sample tion variances are equal.
Group size mean variance a. Is there any evidence to suggest that there is a difference
One 8 173.9 320.41 in population mean deviations from perfect flatness for
these two rotor brands? Use a 5 0.05.
Two 9 150.3 655.36
b. Find bounds on the p value associated with this hypothesis
test.
Assume normal underlying distributions, independent random
samples, and unknown, unequal variances. 10.46 Manufacturing and Product Development The mean
a. Conduct a hypothesis test of H0: m1 2 m2 5 0 versus weight of an ordinary key is an important consideration, as most
Ha: m1 2 m2 . 0. Use a 5 0.05. Americans carry a pocketful of keys. A manufacturer claims that
b. Find bounds on the p value associated with this test. a new process produces a lighter and more durable key. Indepen-
dent random samples of both types of keys were obtained, and
10.43 Consider the following table of summary statistics.
each key was carefully weighed and its weight (in ounces) was
Sample Sample Sample recorded. The resulting summary statistics are given in the follow-
Group size mean standard deviation ing table, with the sample means and sample variances.

One 7 76.83 3.30 Key Sample Sample Sample

Two 16 66.80 14.00 type size mean variance
Old process 10 0.321 0.0137
Assume the underlying distributions are normal and the random New process 10 0.199 0.0202
samples were obtained independently.
a. Suppose the population variances are assumed equal.
Assume the underlying distributions are normal and the popula-
Conduct a hypothesis test of H0: m1 2 m2 5 0 versus tion variances are equal.
Ha: m1 2 m2 2 0. Use a 5 0.05. a. Is there any evidence that the population mean weight of a
b. Suppose the population variances are assumed unequal.
new-process key is less than the population mean weight
Conduct a hypothesis test of H0: m1 2 m2 5 0 versus of an old-process key? Use a 5 0.05.
Ha: m1 2 m2 2 0. Use a 5 0.05. b. Find bounds on the p value associated with this hypothesis
c. Which of the two tests do you think is more appropriate
test.
here? Justify your answer.
10.47 Biology and Environmental Science The tiny
10.44 Independent random samples from two normal popula-
zebra and quagga mussels have invaded at least 600 bodies
tions were obtained. The summary statistics were n1 5 17, of water in the United States and are causing many problems.
x1 5 32.3, s1 5 12.9, n2 5 19, x2 5 43.8, s2 5 14.9. These mollusks disrupt the natural food chain, clog pipes, cling
a. Assume the population variances are unequal. Find a 99%
to machinery, and foul water-delivery systems.7 A random sam-
confidence interval for the difference in population ple of quagga mussels was obtained from Lake Texoma and Lake
means, m1 2 m2. Mead, and each was carefully measured. The size of each (in cm)
b. Using the confidence interval in part (a), is there any
is given on the text website. Is there any evidence to suggest that
evidence to suggest the two population means are the population mean size of quagga mussels is larger in Lake
different? Justify your answer. Texoma than in Lake Mead? Use a 5 0.05, and assume the
underlying distributions are normal with equal population vari-
Applications ances. Write a Solution Trail for this problem. MUSSELS

10.45 Fuel Consumption and Cars The durability and flat- 10.48 Manufacturing and Product Development Shelf
ness of the front rotors on an automobile are important for Safe Milk does not need to be refrigerated until it is opened.
braking and for a smooth ride. The flatness of a rotor can be Although it is convenient, there is some concern that this Grade
determined by a special optical measuring device that measures A milk contains less protein than regular milk. Independent
the largest deviation from perfect flatness in microinches. random samples of 8-ounce servings of Shelf Safe Milk and
 10.2 Comparing Two Population Means Using Independent Samples from Normal Populations 487

regular milk were obtained, and the amount of protein (in the boarding time (in minutes) for each was recorded. The sum-
grams) in each was measured. The summary statistics are given mary statistics are given in the following table.
in the following table.
Sample Sample Sample
Sample Sample Sample Airline size mean standard deviation
Milk size mean standard deviation
American 21 44.5 12.3
Shelf Safe 25 13.95 3.93 US Airways 26 50.7 15.5
Regular 23 19.09 5.91
Is there any evidence to suggest that the new boarding proce-
Is there any evidence to suggest that the population mean dure has decreased the population mean boarding time? Use
amount of protein in Shelf Safe Milk is less than the population a 5 0.05, and assume the populations are normal with equal
mean amount of protein in regular milk? Use a 5 0.01, and variances.
assume the underlying distributions are normal, with equal
10.52 Manufacturing and Product Development A
variances.
recent study was conducted to determine the curing efficiency
10.49 Medicine and Clinical Studies A study was con- (time to harden) of dental composites (resins for the restoration
ducted to determine standard reference values for musculoskel- of damaged teeth) using two different types of lights. Indepen-
etal ultrasonography in healthy adults. Independent random dent random samples of lights were obtained and a certain
samples of men and women were obtained, and the sagittal composite was cured for 40 seconds. The depth of each cure
diameter (in mm) of the biceps tendon was measured in each (in mm) was measured using a penetrometer. The summary
subject. The resulting summary statistics are given in the fol- statistics for the Halogen light were n1 5 10, x1 5 5.35, and
lowing table. s1 5 0.7. The summary statistics for the LuxOMax light were
n2 5 10, x2 5 3.90, and s2 5 0.8. Assume the underlying
Sample Sample Sample populations are normal, with equal variances.
Group size mean standard deviation a. The maker of the Halogen light claims that they produce
Women 54 2.5 0.49 a larger cure depth after 40 seconds than LuxOMax
Men 48 2.8 0.49 lights. Is there any evidence to support this claim? Use
a 5 0.01.
Assume the underlying populations are normal, with equal b. Construct a 99% confidence interval for the difference in
variances. population mean cure depths.
a. Is there any evidence to suggest that the population mean 10.53 Manufacturing and Product Development
sagittal diameter of women’s biceps tendons is different Certain masonry ties used in residential construction
from that of men’s biceps tendons? Use a 5 0.01. receive a hot-dipped galvanized finish for strength and protec-
b. Construct a 95% confidence interval for the difference in tion against moisture. Independent random samples of masonry
population mean sagittal diameters, m1 2 m2. ties from two competing companies were obtained. The amount
10.50 Manufacturing and Product Development A com- of coating on one side of each tie was measured (in g/m2). The
pany that produces hospital furniture has two assembly lines resulting summary statistics are given in the following table.
dedicated to cutting and drilling wood for medical cabinets. Sample Sample Sample
Each computer-controlled process is designed to drill holes in Company size mean variance
a certain cabinet part with depth 12.7 mm. Independent ran-
dom samples of drilled holes were obtained from the two Fero 10 331.4 201.64
assembly lines, the resulting hole depths (in mm) were Cintex 18 298.7 1190.25
recorded. Assume the underlying populations are normal, with
equal variances. CABINET Assume the underlying populations are normal.
a. Is there any evidence to suggest that Line 2 is producing a. Managers at Fero claim that their product has a larger
holes with a greater population mean depth than Line 1? mean coating than Cintex. Is there any evidence to
Use a 5 0.05. support this claim? Use a 5 0.01. Write a Solution Trail
b. Find bounds on the p value associated with this hypothesis for this problem.
test. b. Find a 95% confidence interval for the difference in
population mean coatings.
10.51 Travel and Transportation During Summer 2013,
American Airlines introduced a new method for passengers to 10.54 Sports and Leisure The curve in a hockey stick is
board a narrowbody aircraft. The new procedure affected pas- measured by first placing the face of the blade against a flat sur-
sengers traveling light, those carrying one item that fits under face. The curvature of the stick is restricted so that the perpen-
the seat. The system was designed to decrease the total boarding dicular distance from any point at the heel to the end of the
time, and improve on-time performance.8 Independent random blade is at most 34 inch.9 Independent random samples of hockey
samples of American and US Airways flights were obtained and sticks used by players on the Toronto Maple Leafs and
488 CHAP TE R 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

Montreal Canadiens teams were obtained. The curve in each 10.58 Medicine and Clinical Studies An abdominal aortic
stick was measured (in inches), and the resulting data are sum- aneurysm (AAA) is often signaled by inflammation. A cardio-
marized in the following table. vascular magnetic resonance study was conducted to identify
wall edema as a marker for inflammation. Independent ran-
Sample Sample Sample dom samples of AAA and normal patients were obtained, and
Team size mean standard deviation the MR-STIR intensity values were recorded for each.12
Toronto 10 0.361 0.122 Assume the underlying populations are normal with unequal
Montreal 20 0.425 0.051 variances. EDEMA
a. Is there any evidence to suggest that the population mean
Assume the underlying distributions are normal, with unequal intensity values are different for the two patient groups?
variances. Is there any evidence to suggest that the mean curve Use a 5 0.05.
in Toronto sticks is different from the mean curve in Montreal b. Find bounds on the p value for this hypothesis test.
sticks? Use a 5 0.001.
10.55 Manufacturing and Product Development The Extended Applications
tear strength, tensile strength, backing, and thickness all con- 10.59 Physical Sciences Tinted residential windows have
tribute to the durability of vinyl wallpaper. A new company become popular because they help a home absorb solar energy,
(Aries Wallcoverings) claims to sell the thickest vinyl wallpaper keep out harmful ultraviolet rays, and add privacy. Two inde-
of any currently on the market. Independent random samples of pendent random samples of tinted windows were obtained, each
Aries wallpaper and all others were obtained. The thickness of produced by applying a thin film of a specified color and den-
each wallpaper (in inches) was measured. Assume the underly- sity. The shading coefficient of each tinted window (a unitless
ing distributions are normal, with unequal variances. Is there quantity) was measured, and the summary statistics are given in
any evidence to support the claim made by Aries Wallcover- the following table.
ings? Use a 5 0.01. WALLCVR
Tinted Sample Sample Sample
10.56 Manufacturing and Product Development Avid
window size mean standard deviation
video-game players are always searching for the best graphics
card. Overall performance is improved and everything just runs Silver 8 0.601 0.113
more smoothly with a better graphics card, creating a more Neutral 11 0.741 0.077
enjoyable game experience. Independent random samples of
two video cards were obtained and the frame rate (in fps) was
Assume the underlying populations are normal and the popula-
measured for each. The summary statistics are given in the
tion variances are equal.
following table.10
a. Is there any evidence to suggest the population mean
Sample shading coefficients are different? Use a 5 0.01.
Sample Sample standard b. Construct a 99% confidence interval for the difference in
Card size mean deviation population mean shading coefficients, m1 2 m2.
c. Use the confidence interval in part (b) to determine
GeForce GTX 680 SLI 15 54.7 10.75 whether there is any evidence to suggest the shading
Radeon HD 7970 CrossFire 18 58.8 12.00 coefficients are different. Does your answer agree with
part (a)? If so, why? If not, why not?
Assume the underlying populations are normal and the popula-
10.60 Economics and Finance The U.S. Bureau of
tion variances are equal. Is there any evidence to suggest that the
mean frame rate for the two cards is different? Use a 5 0.05. Engraving and Printing produces $1, $5, $10, $20, $50, and
$100 bills. The $2 banknote is still legal tender but is currently
10.57 Biology and Environmental Science Over the last not in production. Each bill is designed to have the same
few years, the bee population has declined by approximately width, but many people perceive larger-denomination bills to
one-third as a result of mites, fungus, and colony collapse dis- be larger in size. Independent random samples of newly
order.11 Some states have been more affected than others. To minted $1 and $20 bills were obtained, and the width of each
compare the effect on honey production, independent random (in mm) was recorded. The summary statistics are given in the
samples of hives in Kentucky and Missouri were selected and following table.
the amount of honey harvested (in pounds) from each was
recorded. Assume the underlying populations are normal and Sample Sample Sample
the population variances are equal. BEES Bill size mean variance
a. Is there any evidence to suggest that the mean honey
$1 23 66.5990 0.0132
harvest per hive is different in the two states? Use
a 5 0.05. $20 24 66.6924 0.0057
b. Find bounds on the p value for the hypothesis test in
part (a). Assume the underlying distributions are normal.
 10.2 Comparing Two Population Means Using Independent Samples from Normal Populations 489

a. If the population variances are assumed equal, is there any designed to last longer. Independent random samples of the two
evidence to suggest the mean width of a $20 bill is greater types of fuel rods were obtained, and the lifetime of each
than the mean width of a $1 bill? Use a 5 0.01. (in months) was recorded. The summary statistics are given in
b. If the population variances are assumed unequal, is there the following table.
any evidence to suggest the mean width of a $20 bill is
greater than the mean width of a $1 bill? Use a 5 0.01. Fuel-rod Sample Sample Sample
c. Why do both tests lead to the same conclusion (with very design size mean standard deviation
similar p values)? Old 11 34.91 3.20
10.61 Marketing and Consumer Behavior Many home- New 11 39.55 3.55
owners use TIKI torches for outside decoration and to burn
special oil to repel insects. Independent random samples of two Assume the underlying populations are normal and the
types of oil were obtained, and the burn time for 3 ounces of variances are equal.
each was recorded (in hours). The summary statistics are given a. Conduct the relevant hypothesis test to determine
in the following table. whether the new fuel rod does last longer. Use
a 5 0.01.
Sample Sample Sample b. Construct a 99% confidence interval for the difference in
Oil size mean variance population mean lifetimes. Does this confidence interval
Citronella Torch Fuel 21 6.25 1.04 support the hypothesis test conclusion in part (a)?
Explain.
Black Flag Mosquito Control 28 5.98 0.77
10.64 Manufacturing and Product Development Root
Assume the underlying populations are normal. beer was originally made using the sarsaparilla root. However,
a. Do you think the assumption of equal variances is the oil from this root was shown to be carcinogenic (cancer-
reasonable? Why or why not? causing). Since then, many varieties are now made with cane
b. Based on your answer to part (a), conduct the appropriate sugar, herbs, spices, and vanilla. Independent random samples
hypothesis test to determine whether there is any evidence of 12-ounce cans of A&W root beer and Barq’s root beer were
that the mean burn time is different for these two brands. obtained, and the amount of sugar (in grams) was measured in
Use a 5 0.01. each. Assume the underlying populations are normal, with
c. Find bounds on the p value associated with the hypothesis unequal variances. ROOTBEER
test in part (b). a. Is there any evidence to suggest that the population mean
amount of sugar in A&W root beer is greater than in
10.62 Physical Sciences A pressure-relief valve (PRV) is
Barq’s root beer? Use a 5 0.05.
installed on a residential hot-water heater to protect against
b. Construct a 95% confidence interval for the difference in
overheating and, of course, high pressure. Independent random
the population mean sugar amounts. Does this confidence
samples of PRVs from different companies were obtained. Each
interval support the hypothesis test conclusion in part (a)?
value was tested by recording the pressure (in psi) required to
Explain.
cause the valve to open. The summary statistics are given in the
following table.
Challenge
Sample Sample Sample
Company size mean variance 10.65 Robust Statistics The two-sample t test for compar-
ing population means when the variances are equal is a
Delta 30 147.6 7.09 robust statistical procedure. If the population variances are
Gamma 35 147.8 13.70 different, as long as the underlying populations are normal
and the sample sizes are equal, then the hypothesis test is still
Assume the underlying populations are normal, with unequal very reliable.
variances. Generate a random sample of size 25 from a normal distribu-
a. Is there any evidence to suggest that the mean pressure tion with mean m1 5 100 and standard deviation s1 5 5.
required to open each valve is different? Use a 5 0.05. Generate a second random sample of size 25 from a normal
b. Find a 95% confidence interval for the difference in distribution with mean m2 5 100 and standard deviation
population mean pressure required to open each valve. Is s2 5 5. Conduct a two-sided, two-sample t test for comparing
this confidence interval consistent with the results in part population means assuming the population variances are equal
(a)? Explain. and with a 5 0.05. Do this 100 times and record the number of
10.63 Physical Sciences The lifetime of a fuel rod in a com- times you reject the null hypothesis.
mercial light-water nuclear reactor is related to the internal Repeat the same procedure but use s2 5 7. Record the number
pressure. Typically, a fuel rod lasts for 36 months, and one-third of times you reject the null hypothesis. Repeat the same proce-
of all fuel rods are replaced each year during a plant shutdown. dure for s2 5 10, 15, 20, 25, 30, 50. Use your results to explain
A new type of fuel rod includes a gas-relief capsule and is the robust nature of this hypothesis test.
490 CHAP T E R 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

10.3 Paired Data

When comparing population means in the previous two sections, one necessary assumption
was that the samples were obtained independently. The n1 observations from the first popu-
lation and the n2 observations from the second population were unrelated. Many experi-
ments, however, involve only n individuals, or objects, where two observations are made of
each individual. A classic example involves a diet-and-exercise program designed to help
people lose weight. A random sample of n individuals is selected and each is weighed. Each
person follows the regimen for a specified time period and is weighed again at the end of the
experiment. There are two observations of each individual, a before weight and an after
weight. The data are used to determine whether the diet-and-exercise program is effective.
Trial-and-error learning or memory experiments in animals present another good
example. In a typical psychology experiment, a random sample of rats is obtained and
each is timed as it maneuvers through a maze. After several weeks of training, each rat is
timed again. This produces two observations of each animal, a before time and an after
time. This experiment might be designed to determine whether animals can learn the cor-
rect path through a maze and hence decrease the mean time needed to traverse the course.
The difference between the experiments described above and those in the previous two
sections is that here, the paired observations are dependent. We are still interested in
comparing population means (the before mean m1 and the after mean m2), and, therefore,
still interested in the difference m1 2 m2. However, the sample means, X1 and X2 , are not
independent. The standardizations used previously are not applicable, because the vari-
The variance of X1 2 X2 must ance of X1 2 X2 is more complicated. Therefore, another method is necessary, one that
account for the dependence. addresses the dependence and yet still considers the difference m1 2 m2.
Suppose that
For reference, these are the 1. There are n individuals or objects, or n pairs of individuals or objects, that are related
two-sample paired t test in an important way or share a common characteristic; and
assumptions. 2. There are two observations of each individual. The population of first observations is
normal, and the population of second observations is also normal.

Even through the word individual Let X1 represent a randomly selected first observation and let X2 represent the cor-
▲

is used, this means individual or responding second observation on the same individual. Consider the random variable
object. D 5 X1 2 X2 , the difference in the observations, and the n observed differences
di 5 ( x1 ) i 2 ( x2 ) i, i 5 1, 2, . . . , n. X1 and X2 are both normal, so D is also normal.
More important, the differences are independent. A hypothesis test concerning
m1 2 m2 is based on the sample mean of the differences, D. This random variable has
the following properties.

Properties of D
1. E ( D ) 5 m1 2 m2 . D is an unbiased estimator for the difference in means m1 2 m2.
2. The variance of D is unknown, but it can be estimated using the sample variance of
the differences.
3. Both underlying populations are normal, so D is normal, and hence, D is also
normal.
▲

Here’s what all of these results mean for us. To compare population means, m1 and m2,
when the data are paired, we focus on the difference m1 2 m2. As in earlier two-sample
tests, the null hypothesis H0: m1 5 m2 is equivalent to H0: m1 2 m2 5 0. A test to determine
whether the underlying population means of two paired samples are equal is equivalent to
a test to determine whether the population mean of the paired differences is zero. We com-
pute the differences, d1, d2, . . . , dn, and conduct a one-sample t test (with n 2 1 degrees
of freedom) using the differences.
 10.3 Paired Data 491

Hypothesis Tests Concerning Two Population Means

When Data Are Paired
Given the two-sample paired t test assumptions, a hypothesis test concerning the two
population means in terms of the difference mD 5 m1 2 m2, with significance level a, has
the form
This is the template for a H0: mD 5 m1 2 m2 5 D0
hypothesis test concerning two Ha: mD . D0, mD , D0, or mD 2 D0
population means when data are
D 2 D0
paired: a paired t test. TS: T 5
SD/ !n
where SD is the sample standard deviation of the differences.
RR: T $ ta,n21, T # 2ta,n21, or 0 T 0 $ ta/2,n21

A CLOSER L OK
1. D0 is the hypothesized difference in the population means. Usually D0 5 0: The null
hypothesis is that the two population means are equal. However, D0 may be nonzero.
For example, the null hypothesis H0: mD 5 m1 2 m2 5 5 5 D0 specifies that the dif-
ference in population means is 5.
2. A paired t test is valid even if the underlying population variances are unequal, that is,
even if s21 2 s22. The sample variance of the differences, S2D, is a good estimator of
Var ( X1 2 X2 ) when the observations are paired.
3. If a paired t test is appropriate, the test statistic is based on n 2 1 degrees of freedom.
A two-sample t test (incorrect here) would be based on a test statistic with
n 1 n 2 2 5 2n 2 2 degrees of freedom. Therefore, the correct analysis is based on
a distribution with greater variability and is more conservative.

Example 10.8 Relaxing Music

DATA SET There is no direct scientific measure of stress, but some physical properties of the body
MUSIC that are believed to be related to stress include pulse rate, blood pressure, breathing rate,
brain waves, muscle tension, skin resistance, and body temperature. Some researchers
claim that music can be relaxing and, therefore, reduce stress. Twelve patients who claim
to be suffering from job-related stress were selected at random. An initial resting pulse
rate (in beats per minute, bpm) was obtained, and each person participated in a month-
The common characteristic is long music-listening, relaxation-therapy program. A final resting pulse rate was taken at
patient identity. the end of the experiment. The data are given in the following table.

Subject 1 2 3 4 5 6
Initial pulse rate 67 71 67 83 70 75
Final pulse rate 61 72 70 76 58 61
Difference 6 21 23 7 12 14
Subject 7 8 9 10 11 12
Initial pulse rate 71 68 72 88 78 70
Final pulse rate 74 59 61 64 71 77
Difference 23 9 11 24 7 27
Is there any evidence to suggest that the music-listening, relaxation-therapy program
reduced the mean pulse rate and, therefore, the stress level? Assume the underlying distri-
butions of initial and final pulse rate are normal, and use a 5 0.05.
492 CHAP TE R 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

Solution Trail 10.8 SOLUTION

STEP 1 Traditionally, the before measurements are population 1 and the after measure-
KEY W OR DS
ments are population 2.
■ Initial pulse rate
The null hypothesis is that the two population means are equal (i.e., the music-
■ Final pulse rate
listening, relaxation-therapy program has no effect): m1 5 m2 1 m1 2 m2 5
■ Is there any evidence?
mD 5 0 ( 5 D0 ) .
■ Reduce the mean
Each population is assumed normal, and there are two observations on each
■ Random sample
individual. A paired t test is appropriate.
T RA NSL AT I O N The therapy program is designed to reduce stress, so the alternative hypothesis
■ Before and after measure- is m1 . m2 1 m1 2 m2 5 mD . 0. This is a one-sided, right-tailed test.
ments on the same individual
STEP 2 The four parts of the hypothesis test are
■ Conduct a one-sided, right-
tailed test to compare the H0: mD 5 0
before and after mean pulse Ha: mD . 0
rates.
D 2 D0
TS: T 5
CO NCE P TS SD/ !n
■ Hypothesis test concerning RR: T $ ta,n21 5 t0.05,11 5 1.7959
two population means when
data are paired STEP 3 In anticipation of a paired t test, the differences are given in the table above.
The sample mean of the differences is
VIS I ON
1
The data are certainly paired—
d5 3 6 1 ( 21 ) 1 c1 ( 27 ) 4 5 6.3333
there is a before and an after 12
measurement on each The sample variance of the differences is
individual—and each population
is assumed normal. Compute the 1 1
s2D 5 c 1320 2 ( 76 ) 2 d 5 76.2424
differences, use a one-sided 11 12
alternative hypothesis and the
corresponding rejection region, The sample standard deviation of the differences is
find the value of the test statistic,
sD 5 !76.2424 5 8.7317
and draw a conclusion.
The value of the test statistic is
d20 6.3333
t5 5 5 2.5126 ( $ 1.7959 )
sd / !n 8.7317/ !12
Try to find bounds for the p value STEP 4 The value of the test statistic, t 5 2.5126, lies in the rejection region, so we
associated with this hypothesis reject the null hypothesis at the a 5 0.05 significance level. There is evidence to
test. suggest that the music-listening, relaxation-therapy program does reduce a per-
son’s resting pulse rate (and therefore the stress level).
Figure 10.33 shows a technology solution.

Figure 10.33 JMP Matched

Pairs output.

TRY IT NOW GO TO EXERCISE 10.79

 10.3 Paired Data 493

Example 10.9 Radiation Leakage

DATA SET In August 2013, radioactive water from the damaged Fukushima Daiichi nuclear plant
LEAKAGE leaked into the Pacific Ocean. The contaminated groundwater seeped through an under-
ground barrier created through chemical injections, and the Japanese Nuclear Regula-
tory Authority declared an emergency. In response, Tepco (the plant’s operator) began
pumping out more groundwater to ease the spill into the ocean. Concentrations of
cesium-137 (in Bq/1) in the seawater 10–30 kilometers off the Japanese coast were taken
at eight different locations one week apart, before and after the additional pumping. The
data are given in the following table.
Location 1 2 3 4 5 6 7 8
August 23 16.0 12.0 14.0 18.0 12.0 12.5 10.5 15.5
August 30 16.0 9.0 11.0 16.0 14.0 8.5 6.5 14.5

Is there any evidence to suggest that the population mean cesium-137 level decreased
from August 23 to August 30? Assume the underlying populations are normal, use
a 5 0.05, and find bounds on the p value associated with this hypothesis test.

Solution Trail 10.9 SOLUTION

K EY WO R D S
STEP 1 Let population 1 be the cesium-137 concentrations on August 23, and let popula-
tion 2 be the cesium-137 concentrations on August 30.
■ August 23, August 30
■ Is there any evidence? The null hypothesis is that the two population means are equal and the cesium-137
■ Mean level decreased concentrations are the same on both days:
m1 5 m2 1 m1 2 m2 5 mD 5 0
TR A NSL ATI O N
■ Two measurements from each Each population is assumed normal and there are two observations at each loca-
location tion. A paired t test is appropriate.
■ Conduct a one-sided test We would like to determine if the additional pumping helped to reduce the
to compare the cesium-137 cesium-137 concentrations. Therefore, the alternative hypothesis is
levels at different times.
m1 . m2 1 m1 2 m2 . mD 5 0
CON C EPT S STEP 2 The four parts of the hypothesis test are
■ Hypothesis test concerning
two population means when H0: mD 5 0
data are paired Ha: mD . 0
V I SI O N D 2 D0
TS: T 5
The data are paired, and each SD/ !n
population is assumed normal. RR: T $ ta,n21 5 t0.05,7 5 1.8946
Compute the differences, use the
one-sided, right-tailed alternative STEP 3 The summary statistics for the differences are
hypothesis and the
d 5 1.875, sD 5 2.1002
corresponding rejection region,
find the value of the test statistic, The value of the test statistic is
and draw a conclusion.
d20 1.875
t5 5 5 2.5251 ( $ 1.8946 )
sd / !n 2.1002/ !8
STEP 4 The value of the test statistic, t 5 2.5251, lies in the rejection region. Therefore,
we reject the null hypothesis. There is evidence to suggest that the population
mean cesium-137 concentration was less on August 30.
STEP 5 Using Table V in the Appendix, we can only bound the p value. In Table V, row
n 2 1 5 8 2 1 5 7, place 2.5251 in the ordered list of critical values.
2.4469 # 2.5251 # 3.1427
t0.025,7 # 2.5251 # t0.01,7
Therefore, 0.01 # p # 0.025
494 CH APT ER 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

See Figure 10.34.

Figure 10.35 shows a technology solution.

t1

0 2.5251
Figure 10.34 p-Value illustration: Figure 10.35 Minitab Paired t (Test and Confidence
p 5 P(T $ 2.5251) Interval) results.
5 0.0198 # 0.05 5 a

TRY IT NOW GO TO EXERCISE 10.81

The random variable T here is The usual technique can be used to construct a confidence interval for the difference in
D 2 mD means, mD 5 m1 2 m2, when the observations are paired. Start with a symmetric interval
T5
SD / !n about 0 such that the probability T lies in this interval is 1 2 a. Manipulate the inequality
to sandwich mD.

How to Find a 100(1 ! !)% Confidence Interval for "D

Given the paired t test assumptions, a 100 ( 1 2 a ) % confidence interval for mD has as
endpoints the values
sD
d 6 ta/2,n21 (10.6)
!n

Example 10.10 Improved Mileage

A local automotive repair shop advertises a special maintenance package, including tire
balancing, new spark plugs, engine oil additive, and a front-end alignment, that will cer-
tainly improve gas mileage. To check this claim, 18 cars (and drivers) were randomly
selected. Each car was driven on a specially designed route and the miles per gallon for each
car was recorded. Following the maintenance package, each driver took the same route, and
the miles per gallon were measured again. The summary statistics for the differences (before
maintenance mpg ! after maintenance mpg) were d 5 21.28, sD 5 5.62. Assuming nor-
mality, find a 99% confidence interval for the true difference in mean miles per gallon.

SOLUTION
STEP 1 The sample size and summary statistics are given, and the underlying distribu-
tions (before maintenance mpg and after maintenance mpg) are assumed normal.
The observations are paired, so we can use Equation 10.6.
1 2 a 5 0.99 1 a 5 0.01 1 a /2 5 0.005 Find a/2.

Andresr/Shutterstock ta/2,n21 5 t0.005,17 5 2.8982 Find the t critical value with n 5 17.

STEP 2 Use Equation 10.6.

sD
d 6 ta/2,n21 Equation 10.6.
!n
5.62
5 21.28 6 ( 2.8982 ) Use summary statistics and critical value.
!18
5 21.28 6 3.8391 Simplify.

5 ( 25.1191, 2.5591 ) Compute endpoints.

 10.3 Paired Data 495

( 25.1191, 2.5591 ) is a 99% confidence interval for the true mean difference in
miles per gallon, mD.
Note that because 0 is included in this confidence interval, there is no evidence
VIDEO TECH MANUALS to suggest that mD is different from 0, and there is no evidence to suggest that the
EXEL DISCRIPTIVE
PAIRED SAMPLES maintenance program improves mileage.
INFERENCE
Figures 10.36 and 10.37 together show a technology solution.

Figure 10.36 TInterval Figure 10.37 Resulting

input screen. confidence interval.

TRY IT NOW GO TO EXERCISE 10.93

Technology Corner
Procedure: Hypothesis tests and confidence intervals concerning paired data.
Reconsider: Example 10.9, solution, and interpretations.

Crunchlt!
Use the built-in function Statistics; t; Paired. There are tabs to conduct a hypothesis test and construct a confidence interval.
1. Enter the data from the first date into column Var1 and the data from the second date into column Var2.
2. Select Statistics; t; Paired. Choose the First and Second Variables from the drop-down menus. Under the Hypothesis
Test tab, enter the Mean difference under the null hypothesis and select the appropriate alternative. See Figure 10.38.
3. Click Calculate. The results are displayed in a separate window. See Figure 10.39.

Figure 10.38 Crunchlt! t Paired input Figure 10.39 Crunchlt! t Paired

screen. hypothesis test results.

TI-84 Plus C
Compute the differences if necessary. Use the built-in functions T-Test and TInterval. Input is either summary statistics
or data in lists.
1. Enter the August 23 data into list L1 and the August 30 data into list L2.
2. Find the paired differences on the Home screen, and store them in the list L3.
496 CHA PT E R 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

3. Select STAT ; TESTS; T-Test. Highlight Data. Enter m0, the hypothesized difference in means and the list
containing the paired differences. Set the frequency to 1 and highlight the alternative hypothesis. See Figure 10.40.
4. Highlight Calculate and press ENTER . The results are displayed on the Home screen. See Figure 10.41. The Draw
results are shown in Figure 10.42.
5. Use STAT ; TESTS; TInterval to construct a confidence interval for the paired differences. This procedure is described
in the Technology Corner in Section 8.3. A 95% confidence interval for the paired differences in shown in Figure 10.43.

Figure 10.40 T-Test Figure 10.41 T-Test Figure 10.42 T-Test Figure 10.43 A 95%
input screen. hypothesis test results. Draw results. confidence interval for the
paired differences.

Minitab
Use the built-in function Paired t to conduct a hypothesis test and to construct a confidence interval. Input is either paired
data in two columns or summarized data for the paired differences.
1. Enter the August 23 data into column C1 and the August 30 data into column C2.
2. Select Stat; Basic Statistics; Paired t.
3. Choose Each sample is in a column, and enter C1 in the Sample 1 input window and C2 in the Sample 2 input window.
4. Choose the Options option button. Enter a Confidence level, the Hypothesized difference ( D0 ) , and the Alternative
hypothesis.
5. The hypothesis test results and the confidence interval are displayed in a session window. Refer to Figure 10.35.

Excel
Use the Data Analysis toolkit function t-Test: Paired Two Sample for Means. If only summarized data for the paired
differences are given, use the function CONFIDENCE.T to find the bound on the confidence interval and compute the
endpoints of the CI as described in the Technology Corner in Section 8.3.
1. Enter the August 23 data into column A and the August 30 data into column B.
2. Under the Data tab, select Data Analysis; t-Test: Paired Two Sample for Means.
3. Enter the Variable 1 Range (A1:A8), Variable 2 Range (B1:B8), the Hypothesized Mean Difference ( D0 ) , and a value
for Alpha.
4. Choose an Output option and click OK.
5. Summary statistics along with the value of the test statistic, critical values, and p values are displayed. See Figure 10.44.

Figure 10.44 Excel paired

t test results.
 10.3 Paired Data 497

SECTION 10.3 EXERCISES

Concept Check e. Random samples of 45 new home sites in Kansas and
52 new home sites in upstate New York were selected. The
10.66 Short Answer State the two-sample paired t test flatness coefficient (a unitless quantity between 0 and 1) of
assumptions. each lot was measured. The data will be used to determine
10.67 True/False In a two-sample paired t test, the before whether the mean flatness coefficient for new home sites
and after sample sizes must be the same. in Kansas is less than the mean flatness coefficient for
new home sites in upstate New York.
10.68 True/False In a two-sample paired t test, the sample
mean of the differences, D, is an unbiased estimator for 10.73 In each experiment, determine whether the data are
m1 2 m2. obtained independently or are paired. If the data are indepen-
dent, indicate the two distinct populations. If the data are
10.69 True/False A paired t test is valid only if the underly- paired, indicate the common characteristic.
ing population variances are equal. a. A surgeon is investigating the effect of physical therapy on
10.70 Short Answer Explain how a confidence interval for patients who have had rotator-cuff injuries. Twenty-eight
mD can be used to make an inference concerning m1 and m2. patients were selected at random. The range of motion in
the affected arm was measured in each patient prior to
10.71 Short Answer Suppose a paired t test is appropriate. starting physical therapy. After three weeks of consistent
How does this reference distribution compare to the distribution therapy, the range of motion in each patient was measured
based on a two-sample t test? again. The data will be used to determine whether physical
therapy has increased the mean range of motion.
Practice b. A woodwork manufacturer has several lathes used for
10.72 In each experiment, determine whether the data are shaving parts. The accuracy of each is determined by
obtained independently or are paired. If the data are indepen- measuring the production width of a designed 1-mm
dent, indicate the two distinct populations. If the data are wood strip. Nine lathes were selected at random. The
paired, indicate the common characteristic. accuracy of each was measured. Then the gib screws on
a. School board members believe adding a teacher’s aide to each lathe were adjusted, and the width of another 1-mm
each K–4 class will improve classroom management and wood strip was measured. The data will be used to
increase instruction time. Twenty-six elementary determine if the mean width before adjustment is different
classrooms are selected at random, and the daily from the mean width after adjustment.
instruction time for each is recorded. A teacher’s aide is c. A random sample of sixty 20-year-old males and a
then added to each classroom, and the daily instruction random sample of forty-five 70-year-old males were
time is recorded again. The data will be used to determine obtained. The length of each person’s right ear was
whether there is any evidence that adding a teacher’s aide measured (in inches). The data will be used to determine
increases the mean daily instruction time. whether the mean ear length of a 70-year-old male is
b. A researcher investigating home-insurance costs obtained greater than the mean ear length of a 20-year-old male.
a random sample of homes in the Northeast and another d. Modular homes are built in a closed, factory setting,
random sample of homes in the South. The yearly indoors, and are not subject to adverse weather conditions.
insurance cost for each home was recorded. The data will The quality control is often better than for on-site
be used to determine whether there is any difference in construction, but the cost may also differ. Ten different
the mean yearly home-insurance costs between the styles of homes were selected, and the cost of each built
Northeast and the South. as a modular home and on-site was estimated. The data
c. Officials at the transit authority of a large city would like will be used to determine whether the mean cost of
to compare the route times during the morning and modular homes is greater than for on-site construction.
evening rush hours. Eleven routes are selected at random. e. Random samples of 25 frequent flyers on United Airlines
A morning and an evening route completion time are and on Delta Airlines were obtained. The total number of
recorded for each. The data will be used to determine accumulated frequent-flyer miles for each person was
whether the mean evening route time is less than the mean recorded. The data will be used to determine whether the
morning route time. mean number of frequent-flyer miles is greater for United
d. A supplementary health-insurance provider is Airlines passengers than for Delta Airlines passengers.
investigating changes in claim patterns. A random sample 10.74 In each experiment, determine whether the data are
of 32 policyholders was selected. The total amounts obtained independently or are paired. If the data are indepen-
claimed in 2013 and in 2014 were recorded for each dent, indicate the two distinct populations. If the data are
person. The data will be used to determine whether there paired, indicate the common characteristic.
is any evidence that the mean amount claimed has a. A random sample of soccer players on the Brazil national
increased from 2013 to 2014. team and a random sample of soccer players on the
498 CHA PT ER 1 0 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

Argentina national team was obtained. The amount of Applications

time (in minutes) each played during a World Cup match
was obtained. The data will be used to determine whether 10.78 Technology and the Internet Twenty-one computer
the mean time is different for the two teams. programmers from IT firms around the country were selected at
b. A random sample of bananas from Columbia and Costa random. Each was asked to write code in C++ and in Java for a
Rica was obtained. The amount of fiber in the peel of each specific application. The runtime (in seconds) for each pro-
banana was carefully measured and recorded. The data gram, by computer language, was recorded. RUNTIME

will be used to determine whether there is a difference in a. What is the common characteristic that makes these data
the mean amount of fiber in the banana peel by country. paired?
c. A new stent has been designed for use in patients with b. Assume normality. Conduct the appropriate hypothesis
diseased arteries. A random sample of patients in need of test to determine whether there is any evidence that the
a stent was selected. The intrasaccular pressure of each mean runtime for Java programs is greater than the mean
was carefully measured, and then a stent was surgically runtime for C++ programs. Use a 5 0.001.
placed in the diseased artery. The intrasaccular pressure c. Find bounds on the p value associated with this hypothesis
was measured again following surgery. The data will be test.
used to determine whether the stent placement reduced 10.79 Physical Sciences A consultant working for a State
intrasaccular pressure. Police barracks contends that service weapons will fire with a
d. An automobile manufacturer claims that using a lighter higher muzzle velocity if the barrel is properly cleaned. A
weight engine oil can actually increase a car’s miles per random sample of Glock 9-mm handguns was obtained, and
gallon (of gasoline). A random sample of cars (and the muzzle velocity (in feet per second) of a single shot from
drivers) was selected, all who currently use a heavy weight each gun was measured. Each gun was professionally
oil in their engines. The miles per gallon for each car was cleaned, and the muzzle velocity of a second shot (with the
recorded. The engine oil was drained, a lighter weight oil same bullet type) was measured. The data are given in the
was used, and the miles per gallon was recorded again. following table. GLOCK
The data will be used to determine if the mean miles per
gallon has increased. Gun 1 2 3 4 5 6
e. A new process has been developed that theoretically Before 1505 1419 1504 1494 1510 1506
improves the nutritional value of barley for use in fish After 1625 1511 1459 1441 1472 1521
feed.* A random sample of traditional barley and another
random sample of the new barley was obtained. The a. What is the common characteristic that makes these data
percentage of protein in each grain was carefully paired?
measured. The data will be used to determine whether the b. Assume normality. Conduct the appropriate hypothesis
new barley has a greater mean percentage of protein. test to determine whether there is any evidence that a
10.75 The following summary statistics were obtained in a clean gun fires with a higher muzzle velocity. Use
paired-data study: d 5 15.68, sD 5 33.55, and n 5 17. Assume a 5 0.01.
normality and conduct a test of H0: mD 5 0 versus Ha: mD . 0. c. Find bounds on the p value associated with this hypothesis
Use a 5 0.05. test.

10.76 Consider the following paired data. EX10.76

10.80 Sports and Leisure In 1969, the height of Major
League Baseball pitching mounds was lowered from 15 to 10
Subject 1 2 3 4 5 inches. This decreased a pitcher’s leverage and presumably the
Before treatment 332.5 289.3 288.2 268.0 278.0 speed of a typical fastball. Fifteen Major League pitchers were
selected at random, and each threw his best fastball from a
After treatment 317.3 302.5 312.9 325.4 267.3
15-inch mound and from a 10-inch mound. The speed of each
pitch (in mph) was recorded. FASTBALL
a. Assume normality and conduct a test of H0: mD 5 0
a. Assume normality. Construct a 99% confidence interval
versus Ha: mD , 0. Use a 5 0.01.
for the true mean difference in fastball speeds from a
b. Find bounds on the p value associated with this hypothesis
15-inch mound and a 10-inch mound.
test.
b. Using the confidence interval in part (a), is there any
10.77 Consider the paired data provided. EX10.77 evidence to suggest that pitching speed is, on average,
a. Assume normality and conduct a test of H0: mD 5 0 faster from a higher mound? Justify your answer.
versus Ha: mD 2 0. Use a 5 0.01.
10.81 Public Health and Nutrition A filtration
b. Find bounds on the p value associated with this hypothesis
system made for small businesses is designed to remove
test.
particulate matter from the air. To test the new device, a random
sample of businesses was obtained. The concentration of par-
*
US Department of Agriculture, accessed on July 15, 2014, [Link] ticulate matter was measured (in mg/ m3). The filtration system
[Link]/is/pr/2014/[Link]. was then allowed to run for 24 hours, and the concentration of
 10.3 Paired Data 499

particulate matter was measured again. The differences (before ballads. The steps per minute for each person for each genre
filtration – after filtration) were recorded. Assume normality. Is were recorded. WALKING
there any evidence to suggest that the new filtration system a. What is the common characteristic that makes these data
improves air quality by removing particulate matter? Use paired?
a 5 0.05. Write a Solution Trail for this problem. FILTER b. Assume normality. Conduct the appropriate hypothesis to
determine whether there is any evidence to suggest that
10.82 Biology and Environmental Science The best hur-
the population mean steps per minute listening to hard
ricane forecasting models use global data and take hours to run
rock is greater than the population mean steps per minute
on the world’s fastest supercomputers. A random sample of 26
listening to ballads. Use a 5 0.01.
tropical storms or hurricanes that passed within 25 miles of
Miami were selected. The European Center for Medium-Range 10.86 Public Health and Nutrition The cost of
Weather Forecasting (ECMWF) and the Global Forecast Sys- long-term health care continues to rise each year, but it
tem (GFS) models were used to predict storm surge (in feet). varies considerable from state to state. Many older adults who
The summary statistics for the differences (ECMWF – GFS) must enter a nursing home may select either a semi-private or
were d 5 1.4923 and sD 5 2.8097. private room. Several states were selected at random and the
a. Is there any evidence to suggest that the ECMWF model cost of each type of room was recorded.14 ROOMS
predicts higher storm surges? Assume normality and use a. Assume normality and conduct the appropriate hypothesis
a 5 0.05. test to determine whether the population mean cost of a
b. Find bounds on the p value associated with this hypothesis private room is $15 greater than the population mean cost
test. of a semi-private room. Use a 5 0.001. Write a Solution
Trail for this problem.
10.83 Physical Sciences It is important to maintain a low
b. What characteristic of the differences suggests that the
ammonia-ion concentration in freshwater aquariums to ensure hypothesis test in part (a) will be significant?
healthy fish (and plants). An ammonia neutralizer is advertised c. Find bounds on the p value associated with this hypothesis
to almost instantly detoxify ammonia (i.e., reduce the concen- test.
tration of ammonia ions) in order to protect fish. Fourteen
untreated 20-gallon aquariums were selected at random, and the 10.87 Manufacturing and Product Development The
ammonia-ion concentration (in ppm) in each was measured. porosity of a concrete block is a measure (as a percentage) of
One hour after the directed amount of the neutralizer was used, the amount of empty space in the block. In residential homes
the ammonia ion concentration was measured again. Assuming with concrete-block foundations, a larger porosity leads to
normality, is there any evidence to suggest that the neutralizer damper, colder basements. A contractor recommends pretreat-
decreases the mean ammonia-ion concentration? Use ment of concrete blocks with a product designed to decrease the
a 5 0.025. AQUARIUM porosity. A random sample of concrete blocks was obtained,
and the porosity of each was measured. The clear, paintlike
10.84 Public Health and Nutrition Beef boullion generally product was applied to each block, and the porosity was mea-
has a high salt content, which can cause health problems. A sured again. The differences (before treatment – after treatment)
food columnist for a local newspaper suggested simmering in porosities were recorded. Assume normality. Is there any evi-
boullion with slices of raw potato to remove salt. To check this dence to suggest the new product decreases the mean porosity
claim, 10 different boullion brands were selected at random of concrete blocks? Use a 5 0.001. CONCRETE
and the salt content in each was measured (in mg/cup of
10.88 Psychology and Human Behavior Each employee
water). Five potato slices were then added to each broth and
hired at an electronics parts assembly line in Edmonton, Alberta,
the mixtures were left to simmer for 15 minutes. Following
is given a general intelligence test. To determine which method
this procedure, the salt content was measured again. The dif-
of training is more effective, eight pairs of new hires were
ference between the initial and the final salt content was com-
matched according to their exam scores. One set of employees
puted for each boullion brand, and the data are given in the
was asked to read appropriate training manuals, while the other
following table. BOULLION
group watched interactive training videos. Each employee was
then asked to assemble a part used in a locater-beacon transmit-
2169 2222 431 110 2168 ter, and the time (in minutes) to completion was recorded. The
353 2207 68 25 203 data are given in the following table. TRAINING

Assume normality. Is there evidence to suggest that simmering Employee pair 1 2 3 4

with raw potatoes decreases the mean salt content in beef boul- Written manual 4.9 4.6 5.3 4.9
lion? Use a 5 0.05. Interactive video 3.1 4.1 4.4 4.9
10.85 Psychology and Human Behavior A new study sug-
gests that the speed of a person’s step is related to the genre of Employee pair 5 6 7 8
music being listened to while walking.13 A random sample of Written manual 4.9 5.4 5.5 5.0
18 adults was obtained. Each walked in a circular path while
Interactive video 3.6 3.9 6.5 5.3
listening to two different genres of music, hard rock and
500 CH APT ER 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

a. What is the common characteristic that makes these data to lower the cloud point of this type of fuel. A random sample
paired? of six different biodiesel fuels was obtained and the cloud point
b. Is there any evidence to suggest the true mean time was measured for each. One ounce of the chemical additive was
difference, mD, is different from 0? Assume normality and mixed in with every fuel sample and the cloud point was mea-
use a 5 0.05. sured again. The resulting data are given in the following table
(temperatures in °C). CLOUDING
10.89 Public Health and Nutrition Americans love ham-
burgers, but the high fat content in some cooked patties pres-
ents a severe health threat. Certain electric grills are designed to Fuel 1 2 3 4 5 6
drain fat away from the patty, resulting in a healthier, although Before 11.7 12.9 14.2 12.7 11.3 12.4
perhaps less tasty, meal. A random sample of ground beef pack- additive
ages was obtained (with various fat contents). Two patties were
After 10.3 10.7 14.1 10.0 11.2 12.1
made from each package. One was cooked in an electric grill,
additive
while the other was prepared in a frying pan on top of a stove.
The fat content (as a percentage) in each cooked patty was
measured. HAMBURG a. Assume normality, and conduct the appropriate
a. Conduct the appropriate hypothesis test to determine hypothesis test to determine whether the additive lowers
whether the true mean fat content in hamburgers cooked the mean cloud point in biodiesel fuel. Use a 5 0.05.
on an electric grill is less than the true mean fat content of b. Conduct an inappropriate two-sample t test to compare
hamburgers cooked in a frying pan. Assume normality the population mean cloud point before treatment with the
and use a 5 0.001. population mean cloud point after treatment. Assume the
b. Find bounds on the p value associated with this hypothesis population variances are unequal and use a 5 0.05.
test. c. Compare the conclusions in parts (a) and (b). How are the
test statistics the same, and how do they differ?
Extended Applications 10.92 Biology and Environmental Science In August
2013, Vancouver Coastal Health warned swimmers that the
10.90 Medicine and Clinical Studies A new drug designed
coliform count at East False Creek was approximately twice the
to reduce fever (and relieve aches and pains) is being tested for
safe level.15 The contamination was attributed to boats, birds,
efficacy and side effects. Ten patients entering a hospital with a
geese, and warm weather. A random sample of locations along
high fever were selected at random. The temperature (in °F) of
the creek was obtained, and the coliform count (bacteria per
each patient was measured, the drug was administered, and two
100 ml of water) was measured in early August and again fol-
hours later the temperature was measured again. The data are
lowing several rain storms. COLIFORM
given in the following table. FEVER
a. What is the common characteristic that makes these data
Patient 1 2 3 4 5 paired?
b. Use a one-sided paired t test to determine whether the rain
Before drug 102.6 99.2 102.3 101.1 102.7
caused a decrease in the coliform count. Assume
After drug 99.8 98.8 97.5 100.3 99.6 normality and use a 5 0.001.
c. Find the safe level of coliform. Do you think it was safe to
Patient 6 7 8 9 10 swim in the creek after the rain? Why or why not?
Before drug 102.6 100.5 103.5 105.7 104.3 10.93 Travel and Transportation For anyone planning to
After drug 102.8 99.0 101.8 97.1 99.2 travel, either for vacation or on business, the pricing plans of
airlines remain a guarded mystery. A study by [Link]
a. What is the common characteristic that makes these data
suggested that the cheapest fares are found 49 days before a
paired?
flight. However, according to Travelers Today, the best prices
b. Assume normality. Conduct the appropriate hypothesis
are offered 21 days before a flight.16 The Washington, D.C.-to-
test to determine whether there is any evidence that the
Los Angeles route was selected as a test case. A random sample
new drug reduces the mean patient temperature after two
of days was selected, and the best price for a one-way ticket was
hours. Use a 5 0.05.
recorded 21 and 49 days prior to the flight. AIRPRICE
c. Find bounds on the p value associated with this hypothesis
a. Assume normality. Is there any evidence to suggest that
test.
the price of a ticket on this route differs if purchased 21 or
d. What characteristic of the differences suggests that a
49 days in advance? Use a 5 0.05.
hypothesis test will be significant?
b. Find bounds on the p value associated with this hypothesis
10.91 Fuel Consumption and Cars Biodiesel fuel has a test.
cloud point, the temperature at which the fuel becomes cloudy, c. Find a 95% confidence interval for the difference in
of approximately 13°C. This clouding can lead to poor engine population mean cost per ticket. Does this confidence
performance and can even cause an engine to stop completely. interval support your conclusion in part (a)? Why or
An industrial chemical company produces an additive designed why not?
 10.4 Comparing Two Population Proportions Using Large Samples 501

10.4 Comparing Two Population Proportions

Using Large Samples
The methods presented in this section can be used to compare two population proportions.
For example, a social scientist may conduct an experiment to determine whether the true
proportion of men who favor legalized gambling is the same as the true proportion of
women. Or an advertising agency might be interested in comparing the true propor-
tions of children who saw a certain television commercial in two different regions of
the country.
Here is a quick review of the notation (presented at the beginning of this chapter) asso-
ciated with populations 1 and 2 and samples 1 and 2.
Population proportion of successes: p1, p2
Sample size: n1, n2
Number of successes: x1, x2
Corresponding random variables: X1, X2
= =
Sample proportion of successes: p1 5 x1 /n1, p2 5 x2 /n2
= =
Corresponding random variables: P1 5 X1 /n1, P2 5 X2 /n2
The general null hypothesis is stated (as usual) in terms of a difference, H0: p1 2 p2 5 D0.
However, there are two cases to consider: (1) D0 5 0 and (2) D0 2 0. In both cases, a
reasonable estimator for p1 2 p2 is the difference between the sample proportions,
= =
P1 2 P2. The following properties are used to construct a hypothesis test (and confidence
interval) concerning the difference between two population proportions.

= =
Properties of the Sampling Distribution of P1 ! P2
= =
1. The mean of P1 2 P2 is the true difference between population proportions, p1 2 p2 .
That is,
= =
E ( P1 2 P2 ) 5 mP=1 2 P= 25 p12 p2
= =
2. The variance of P1 2 P2 is
= = p1 ( 1 2 p1 ) p2 ( 1 2 p2 )
Var ( P1 2 P2 ) 5 s2P= 12P= 2 5 1
= n=1 n2
The standard deviation of P1 2 P2 is

p1 ( 1 2 p1 ) p2 ( 1 2 p2 )
sP=12P= 2 5 1
Å n1 n2
3. If
(a) both n1 and n2 are large,
Items (b) and (c) are the (b) n1p1 $ 5 and n1 ( 1 2 p1 ) $ 5, and
nonskewness criterion. (c) n2 p2 $ 5 and n2 ( 1 2 p2 ) $ 5,
= =
then the distribution of P1 2 P2 is approximately normal.
= = p1 ( 1 2 p1 ) p2 ( 1 2 p2 )
In symbols: P1 2 P2 , N c p1 2 p2, d
d
1
n1 n2

The appropriate standardization will result in an approximate Z distribution. The esti-

mate of the standard deviation, sP=12P= 2, is determined by the value of D0.
502 CH A PTE R 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

Case 1: H0: p1 2 p2 5 0, or p1 5 p2 ( D0 5 0 )
If this null hypothesis is true, there is one common value for the two population propor-
= =
tions, denoted p ( 5 p1 5 p2 ) . The variance of P1 2 P2 becomes
p(1 2 p) p(1 2 p) 1 1
s2P=12P= 2 5 1 5 p(1 2 p) a 1 b (10.7)
n1 n2 n1 n2
= =
Using the properties of P1 2 P2, the random variable
= =
( P1 2 P2 ) 2 0
Z5 (10.8)
1 1
p(1 2 p) a 1 b
Å n 1 n2

is approximately standard normal. As for the common variance in Section 10.2, an esti-
mator for the common proportion, p, is obtained by using information from both samples.
The pooled or combined estimate of the common population proportion is
= = X1 1 X2 n1 = n2 =
P c is another weighted average. Pc 5 5a bP1 1 a bP2 (10.9)
n1 1 n2 n1 1 n2 n1 1 n2
The general hypothesis test procedure is based on the standardization in Equation 10.8
=
with pc as an estimate of p.

Hypothesis Tests Concerning Two Population Proportions

When D0 " 0
Given two random samples of sizes n1 and n2, a large-sample hypothesis test concerning
two population proportions in terms of the difference p1 2 p2 (with D0 5 0) with sig-
nificance level a has the form
This is the template for a large- H0: p1 2 p2 5 0
sample hypothesis test concerning Ha: p1 2 p2 . 0, p1 2 p2 , 0, or p1 2 p2 2 0
two population proportions when = =
D0 5 0. P1 2 P2
TS: Z 5
= = 1 1
Pc ( 1 2 Pc ) a 1 b
Again, use only one (appropriate) Å n1 n2
alternative hypothesis and the RR: Z $ za, Z # 2za, or 0 Z 0 $ za / 2
corresponding rejection region.

There is no confidence interval for the difference in population proportions in this

case. If we assume p1 5 p2, then there is no reason to construct a confidence interval for
the difference p1 2 p2 5 0.

A CLOSER L OK
1. This test is valid as long as the nonskewness criterion holds for both samples. Use the
= =
estimates p1 and p2 to check the inequalities.
2. Just as a reminder, the z critical values for this test are from the standard normal distri-
bution.
3. Remember, we can also determine whether to reject or not to reject the null hypothesis
by comparing the p value associated with the value of the test statistic to the signifi-
cance level a.

The following example illustrates this hypothesis test procedure.

 10.4 Comparing Two Population Proportions Using Large Samples 503

Solution Trail 10.11 Example 10.11 Living with Parents

A high percentage of the millennial generation, adults approximately 18 to 31 years old,
K EY WO R D S
are living in their parents’ homes. This may be due to unemployment, lower marriage
Is there any evidence?
rates, and/or the recession.17 In a random sample of 275 male millennials, 110 lived with
■

■ Greater than the true their parents, and in a random sample of 300 female millennials, 96 lived with their par-
proportion
ents. Is there any evidence to suggest that the true proportion of male millennials living
■ Random sample with their parents is greater than the true proportion of female millennials living with
TR A NSL ATI O N their parents? Use a 5 0.05.
■ Conduct a one-sided
hypothesis test about p1 2 p2 SOLUTION
■ D0 5 0. STEP 1 This is a one-sided test in which we are looking for evidence that a greater pro-
CON C EPT S portion of males than females are living with their parents. Therefore, D0 5 0,
■ Large-sample hypothesis test
and case 1 is appropriate. Arbitrarily, let male millennials be population 1 and
concerning two population female millennials be population 2. The given information:
proportions when D0 5 0

V I SI O N Males Females
Check the large-sample Sample size n1 5 275 n2 5 300
assumptions. Use the template
Number of successes x1 5 110 x2 5 96
for a one-sided, right-tailed test = =
concerning p1 2 p2 when Sample proportion p1 5 110/275 5 0.40 p2 5 96/300 5 0.32
D0 5 0. Use a 5 0.05 to find
the critical value, compute the
value of the test statistic, and STEP 2 Check the nonskewness criterion using estimates for p1 and p2.
draw a conclusion. = =
n1p1 5 ( 275 ) ( 0.40 ) 5 110 $ 5 n1 ( 1 2 p1 ) 5 ( 275 ) ( 0.60 ) 5 165 $ 5
= =
n2 p2 5 ( 300 ) ( 0.32 ) 5 96 $ 5 n2 ( 1 2 p2 ) 5 ( 300 ) ( 0.68 ) 5 204 $ 5
= =
All of the inequalities are satisfied, so P1 2 P2 is approximately normal, and
the large-sample hypothesis test concerning population proportions can be
used.
STEP 3 The four parts of the hypothesis test are

H0: p1 2 p2 5 0
Ha: p1 2 p2 . 0
= =
P1 2 P2
TS: Z 5
= = 1 1
Pc ( 1 2 Pc ) a 1 b
Å n1 n2
RR: Z $ za 5 z0.05 5 1.6449
STEP 4 The estimate of the common population proportion is

= x1 1 x2 110 1 96
pc 5 5 5 0.3583
n1 1 n2 275 1 300
The value of the test statistic is
= =
p1 2 p2 0.40 2 0.32
z5 5 5 1.9985
= = 1 1 1 1
( )
pc 1 2 pc a 1 b ( ) (
0.3583 0.6417 a) 1 b
Å n1 n2 Å 275 300
STEP 5 The value of the test statistic lies in the rejection region. Equivalency, p "
0.0228 # 0.05, as illustrated in Figure 10.45. We reject the null hypothesis at the
a 5 0.05 significance level. There is evidence to suggest that the true proportion
of male millennials living with their parents is greater than the true proportion of
female millennials living with their parents.
504 CH AP TER 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

Figure 10.45 p-Value illustration:

p 5 P(Z $ 1.9985)
5 0.0228 # 0.05 5 a 0 1.9985

Figures 10.46 through 10.48 together show a technology solution.

Figure 10.46 2-Prop Figure 10.47 2-PropZ Figure 10.48 2-PropZ

ZTest input screen. Test hypothesis test Test Draw results.
results.

TRY IT NOW GO TO EXERCISE 10.106

Case 2: H0: p1 2 p2 5 D0 2 0
This case, with D0 2 0, is less common. Because p1 and p2 are assumed unequal, there is
no hypothesized, common population proportion. The hypothesis test follows routinely
= =
from the properties of P1 2 P2.

Hypothesis Tests Concerning Two Population

Proportions When D0 u 0
Given two random samples of sizes n1 and n2, a large-sample hypothesis test concerning
two population proportions in terms of the difference p1 2 p2 (with D0 2 0) with sig-
nificance level a has the form
This is the template for a large- H0: p1 2 p2 5 D0
sample hypothesis test concerning Ha: p1 2 p2 . D0, p1 2 p2 , D0, or p1 2 p2 2 D0
two population proportions when = =
( P1 2 P2 ) 2 D0
D0 2 0. The nonskewness TS: Z 5 = = = =
criterion must also be met. P1 ( 1 2 P1 ) P2 ( 1 2 P2 )
1
Å n1 n2
RR: Z $ za, Z # 2za, or 0 Z 0 $ za/2

Suppose two random samples of sizes n1 =and n2= are obtained, and the nonskewness
criterion is satisfied. Using the properties of P1 2 P2, a confidence interval for p1 2 p2
can be derived ( p1 and p2 are assumed unequal). Start with a symmetric interval about 0
such that the probability Z lies in this interval is 1 2 a. As usual, manipulate the inequal-
ity to sandwich the parameter p1 2 p2.
 10.4 Comparing Two Population Proportions Using Large Samples 505

STEPPED
STEPPED TUTORIAL
TUTORIALS
How to Find a 100(1 2 #)% Confidence Interval for p1 2 p2
CONFIDENCE
BOX PLOTS Given two (large) random samples of sizes n1 and n2, a 100 ( 1 2 a ) % confidence interval
INTERVALS FOR THE
DIFFERENCE IN POPU- for p1 2 p2 has as endpoints the values
LATION PROPORTIONS
= = = =
= = p1 ( 1 2 p1 ) p2 ( 1 2 p2 )
( p1 2 p2 ) 6 za/2 1 (10.10)
Å n1 n2

Solution Trail 10.12 Example 10.12 Make Room for Canadian Fliers
More Canadians are using smaller U.S. airports near the border when they travel to cities
K EY WO R D S
in the United States. Reasons include a strong Canadian dollar and higher taxes and fees
Is there any evidence?
■
on air travel in Canada.18 In a random sample of 500 fliers at Buffalo Niagara Interna-
■ True proportion is more than tional Airport, 245 were Canadian, and in a random sample of 400 fliers at Bellingham
0.05 greater than
(Washington) International Airport, 160 were Canadian.
■ Random sample
a. Conduct the appropriate hypothesis test to determine whether there is evidence that the
TR A NSL ATI O N true proportion of Canadian fliers at Niagara is more than 0.05 greater than the true
■ Conduct a one-sided proportion of Canadian fliers at Bellingham. Use a 5 0.01.
hypothesis test about p1 2 p2 b. Find the p value associated with this hypothesis test.
■ D0 5 0.05
SOLUTION
CON C EPT S
STEP 1 Let Niagara fliers be population 1 and Bellingham fliers be population 2. We are
■ Large-sample hypothesis test
concerning two population looking for evidence that the difference p1 2 p2 is greater than 0.05 5 D0 2 0.
proportions when D0 2 0 Therefore, case 2 is appropriate.
The given information is presented here.
V I SI O N
Check the large-sample Niagara fliers Bellingham fliers
assumptions. Use the template
for a one-sided test concerning Sample size n1 5 500 n2 5 400
p1 2 p2 when D0 2 0. Use Number of successes x1 5 245 x2 5 160
a 5 0.01 to find the critical = =
Sample proportion p1 5 245/500 5 0.49 p2 5 160/400 5 0.40
value, compute the value of the
test statistic, and draw a
conclusion. Check the nonskewness criterion using estimates for p1 and p2.
= =
n1p1 5 ( 500 ) ( 0.49 ) 5 245 $ 5 n1 ( 1 2 p1 ) 5 ( 500 ) ( 0.51 ) 5 255 $ 5
= =
n2 p2 5 ( 400 ) ( 0.40 ) 5 160 $ 5 n2 ( 1 2 p2 ) 5 ( 400 ) ( 0.60 ) 5 240 $ 5
= =
All of the inequalities are satisfied, so P1 2 P2 is approximately normal, and the
large-sample hypothesis test concerning population proportions can be used.
STEP 2 The four parts of the hypothesis test are
H0: p1 2 p2 5 0.05
Ha: p1 2 p2 . 0.05
= =
( P1 2 P2 ) 2 0.05
TS: Z 5 = = = =
P1 ( 1 2 P1 ) P2 ( 1 2 P2 )
1
Å n1 n2
RR: Z $ za 5 z0.01 5 2.3263
STEP 3 The value of the test statistic is
= =
( p1 2 p2 ) 2 0.05 ( 0.49 2 0.40 ) 2 0.05
z5 = = = = 5 5 1.2062
p1 ( 1 2 p1 ) p2 ( 1 2 p2 ) ( 0.49 ) ( 0.51 ) ( 0.40 ) ( 0.60 )
1 1
Å n1 n2 Å 500 400
506 CH APT ER 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

The value of the test statistic does not lie in the rejection region. (Equivalently,
p 5 0.1139 . 0.05, as illustrated in Figure 10.49.) We do not reject the null
hypothesis. At the a 5 0.01 significance level, there is no evidence to suggest
that the population proportion of Canadian fliers at Niagara is more than 0.05
greater than the population proportion of Canadian fliers at Bellingham.

Figure 10.49 p-Value illustration:

p 5 P(Z $ 1.2062)
5 0.1139 . 0.05 5 a 0 1.2062

STEP 4 This is a one-sided, right-tailed test, so the p value is a right-tail probability.

p 5 P ( Z $ 1.2062 ) Definition of p value for Ha: p1 2 p2 . 0.05.

5 1 2 P ( Z # 1.2062 ) The complement rule.

5 1 2 0.8861 5 0.1139 ( . 0.05 5 a ) Use Table III in the Appendix.

Figure 10.50 shows a technology solution.

Figure 10.50 The 2

Proportions hypothesis test
results and (one-sided)
confidence interval from
Minitab.

TRY IT NOW GO TO EXERCISE 10.121

VIDEO TECH MANUALS The computations for a confidence interval for p1 2 p2 are illustrated in the next
EXEL PROPORTIONS
TWO DISCRIPTIVE example.
INFERENCE CI TEST -
SUMMARIZED DATA
Example 10.13 Storm Watch
The Weather Channel (TWC) is one of the most popular cable TV networks. Who hasn’t
seen Jim Cantore in the middle of some wild weather? However, the number of viewers
varies greatly according to geographic region and the current weather conditions. Random
samples of cable TV viewers in the Northeast (population 1) and in the West (population
2) were obtained. The number of viewers who watched TWC in the past week was recorded.
The data are given in the following table.

Northeast West
Carolina K. Smith MD/Shutterstock Sample size n1 5 1000 n2 5 1500
Number of successes x1 5 446 x2 5 303
= =
Sample proportion p1 5 446/1000 5 0.4460 p2 5 303/1500 5 0.2020

Construct a 99% confidence interval for the true difference in proportions of cable TV
viewers who watched TWC in the past week.
 10.4 Comparing Two Population Proportions Using Large Samples 507

SOLUTION
STEP 1 The sample sizes, number of successes, and sample proportions are given. Check
the nonskewness criterion using estimates for p1 and p2.
= =
n1p1 5 ( 1000 ) ( 0.4460 ) 5 446 $ 5 n1 ( 1 2 p1 ) 5 ( 1000 ) ( 0.5540 ) 5 540 $ 5
= =
n2 p2 5 ( 1500 ) ( 0.2020 ) 5 303 $ 5 n2 ( 1 2 p2 ) 5 ( 1500 ) ( 0.7980 ) 5 1197 $ 5
All of the inequalities are satisfied, so the distribution of the difference in sample
proportions is approximately normal. A large-sample confidence interval is
appropriate.
STEP 2 Find the critical value.
1 2 a 5 0.99 1 a 5 0.01 1 a /2 5 0.005 Find a/2.

za/2 5 z0.005 5 2.5758 Common critical value.

STEP 3 Use Equation 10.10.

= = = =
= = p1 ( 1 2 p1 ) p2 ( 1 2 p2 )
( p1 2 p2 ) 6 za/2 1 Equation 10.10.
Å n1 n2
( 0.4460 )( 0.5540 ) ( 0.2020 ) ( 0.7980 )
5 ( 0.4460 2 0.2020 ) 6 ( 2.5758 ) 1
Å 1000 1500
Use summary statistics and critical value.

5 0.2440 6 0.0485 Simplify.

5 ( 0.1955, 0.2925 ) Compute endpoints.

(0.1955, 0.2925) is a 99% confidence interval for the difference in the proportion
of cable TV viewers who watched TWC in the past week in the Northeast and in
the West, p1 2 p2. Note that because 0 is not included in this interval, there is
evidence to suggest that the two proportions are different.
Figures 10.51 and 10.52 together show a technology solution.

Figure 10.51 2-PropZ Figure 10.52 Resulting

Int input screen. confidence interval.

TRY IT NOW GO TO EXERCISE 10.122

Technology Corner
Procedure: Hypothesis tests and confidence intervals concerning two population proportions.
Reconsider: Example 10.11, solution, and interpretations.

Crunchlt!
Use Proportion 2-sample to conduct a hypothesis test and concerning two population proportions, D0 5 0 or D0 2 0, and
to construct a confidence interval for the difference in population proportions.
508 CHAP TER 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

1. Select Statistics; Proportion; 2-sample.

2. Under the Summarized tab, enter n and the number of successes for each sample.
3. Under the Hypothesis Test tab, enter the Difference of proportions under the null hypothesis ( D0 ) , and select the
appropriate Alternative hypothesis. See Figure 10.53.
4. Click Calculate. The results are displayed in a separate window. See Figure 10.54.
5. To construct a confidence interval, under the Confidence Interval tab, enter the Confidence Interval Level, and click
Calculate. See Figure 10.55.

Figure 10.53 Proportion Figure 10.54 Proportion Figure 10.55

2-Sample input screen. 2-Sample hypothesis test Proportion 2-Sample
results. confidence interval.

TI-84 Plus C
Use 2-PropZTest to conduct a hypothesis test concerning two population proportions, D0 5 0, and 2-PropZInt to
construct a confidence interval for the difference in population proportions. There is no built-in function to conduct a
hypothesis test if D0 2 0.
1. Select STAT ; TESTS; 2-PropZTest.
2. Enter the number of successes and the number of trials for each sample: x1, n1, x2, n2. Highlight the appropriate alterna-
tive hypothesis. See Figure 10.46.
3. The Calculate and Draw results are shown in Figures 10.47 and 10.48.
4. To construct a confidence interval, select STAT ; TESTS; 2-PropZInt.
5. Enter the number of successes and the number of trials for each sample, x1, n1, x2, n2, and the confidence level
(Figure 10.56).
6. Highlight Calculate and press ENTER . The confidence interval is displayed on the Home Screen. See Figure 10.57.

Figure 10.56 2-PropZ Figure 10.57 Resulting

Int input screen. confidence interval.

Minitab
Use the function 2 Proportions to conduct a hypothesis test and to construct a confidence interval. Input is samples in one
column, samples in different columns, or summarized data.
1. Select Stat; Basic Statistics; 2 Proportions.
2. Choose Summarized data and enter the number of successes (Events) and number of trials for each sample.
 10.4 Comparing Two Population Proportions Using Large Samples 509

3. Choose the Options option button. Enter a Confidence level, the (hypothesized) Test difference ( D0 ) , and select the
appropriate Alternative hypothesis. If D0 5 0, use the Test method: Use the pooled estimate of the proportion. If
D0 2 0, use Estimate the proportions separately.
4. The hypothesis test results and confidence interval are displayed in a session window. See Figure 10.58.

Figure 10.58 The 2 Proportions

hypothesis test results and
(one-sided) confidence interval
from Minitab.

Excel
There are no built-in functions to conduct a hypothesis test concerning two population proportions or to construct a
confidence interval for the difference in population proportions. However, functions associated with the standard normal
distribution may be used to find critical values and p values. Use ordinary spreadsheet calculations where necessary.
1. Enter the number of successes and number of trials for each sample.
= = =
2. Compute p1, p2, and pc.
3. Compute the value of the test statistic, z, and use the function [Link] to find the p value.
= =
4. To construct the confidence interval, compute the difference, p1 2 p2, use the function [Link] to find the
= = = =
p1 ( 1 2 p1 ) p2 ( 1 2 p2 )
critical value, and compute 1 .
Å n1 n2
5. Find the left endpoint and the right endpoint of the confidence interval. See Figure 10.59.

Figure 10.59 Excel calculations

to conduct a hypothesis test and
construct a confidence interval
concerning two population
proportions.

SECTION 10.4 EXERCISES

Concept Check 10.97 True/False If p1 and p2 are assumed unequal, there is
no hypothesized common population proportion.
10.94 Short Answer State the nonskewness criterion.
10.98 True/False In a hypothesis test concerning two popu-
10.95 True/False The estimate of sP=12P= 2 is determined by lation proportions, D0 can be negative.
the value of D0.
10.99 True/False A confidence interval for p1 2 p2 can be
10.96 True/False The pooled estimate of the common popu- used to draw a conclusion about the equality of the two popula-
lation proportion is based on the largest sample. tion proportions.
510 CH APT ER 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

Practice e. Independent random samples of videos on U.S. and

Canadian newspaper websites were obtained. The number
10.100 In each of the following problems, the size and the of videos that were advertisements was recorded. The data
number of individuals or objects with a certain characteristic will be used to determine whether the proportion of
are given for samples from two populations. Check the two videos that are advertisements in the United States is
nonskewness inequalities for both samples, and determine higher than in Canada.
whether a large-sample test concerning two population
proportions is appropriate. 10.103 In each of the following problems, use the given data to
a. n1 5 303, x1 5 175, n2 5 463, x2 5 250 conduct the appropriate hypothesis test concerning two popula-
b. n1 5 560, x1 5 140, n2 5 530, x2 5 125 tion proportions, find the p value, and state your conclusion.
c. n1 5 160, x1 5 155, n2 5 185, x2 5 170 a. n1 5 500, x1 5 400, n2 5 525, x2 5 405,
d. n1 5 1020, x1 5 700, n2 5 1277, x2 5 950 H0: p1 2 p2 5 0, Ha: p1 2 p2 . 0, a 5 0.05
e. n1 5 842, x1 5 319, n2 5 755, x2 5 280 b. n1 5 646, x1 5 280, n2 5 680, x2 5 330,
f. n1 5 4375, x1 5 237, n2 5 5005, x2 5 245 H0: p1 2 p2 5 0, Ha: p1 2 p2 , 0, a 5 0.01
c. n1 5 255, x1 5 81, n2 5 266, x2 5 110,
10.101 In each of the following problems, the sample sizes and H0: p1 2 p2 5 0, Ha: p1 2 p2 2 0, a 5 0.025
population proportions are given. Find
= the= mean, variance, and d. n1 5 1440, x1 5 907, n2 5 1562, x2 5 970,
standard deviation of the estimator P1 2 P2, and compute each H0: p1 2 p2 5 0, Ha: p1 2 p2 2 0, a 5 0.001
probability.
a. n1 5 645, p1 5 0.24, n2 5 650, p2 5 0.26, 10.104 In each of the following problems, use the given data to
= =
P ( P1 2 P2 $ 0.045 ) conduct the appropriate hypothesis test concerning two popula-
b. n1 5 250, p1 5 0.37, n2 5 270, p2 5 0.33, tion proportions, find the p value, and state your conclusion.
= = a. n1 5 200, x1 5 100, n2 5 300, x2 5 165,
P ( P1 2 P2 # 20.04 )
c. n1 5 144, p1 5 0.87, n2 5 156, p2 5 0.86, H0: p1 2 p2 5 0.05, Ha: p1 2 p2 , 0.05, a 5 0.01
= = b. n1 5 480, x1 5 384, n2 5 490, x2 5 367,
P ( 20.05 , P1 2 P2 , 0.05 )
H0: p1 2 p2 5 0.02, Ha: p1 2 p2 . 0.02, a 5 0.05
d. n1 5 520, p1 5 0.65, n2 5 480, p2 5 0.72,
= = c. n1 5 610, x1 5 450, n2 5 675, x2 5 470,
P ( P1 2 P2 . 20.10 ) H0: p1 2 p2 5 0.10, Ha: p1 2 p2 2 0.10, a 5 0.01
e. n1 5 1200, p1 5 0.73, n2 5 1150, p2 5 0.85,
= = d. n1 5 2500, x1 5 710, n2 5 3100, x2 5 770,
P ( P1 2 P2 , 20.06 ) H0: p1 2 p2 5 0.07, Ha: p1 2 p2 2 0.07, a 5 0.001
f. n1 5 500, p1 5 0.645, n2 5 525, p2 5 0.604,
= = 10.105 In each of the following problems, use the given data
P ( 20.02 # P1 2 P2 # 0.10 )
and confidence level to construct a confidence interval for the
10.102 A hypothesis test concerning two population propor- difference of two population proportions, p1 2 p2.
tions is described in each of the following problems. Identify a. n1 5 388, x1 5 230, n2 5 402, x2 5 250, 95%
each population, and determine the appropriate null and alter- b. n1 5 528, x1 5 475, n2 5 530, x2 5 497, 95%
native hypotheses in terms of p1 and p2. c. n1 5 180, x1 5 92, n2 5 194, x2 5 100, 99%
a. A study was conducted to determine whether there is any d. n1 5 2300, x1 5 1705, n2 5 2404, x2 5 1690, 90%
difference in the proportion of people who listen to
satellite radio in California versus Tennessee.
b. Two random samples, of men and of women who tried to Applications
talk their way out of a traffic ticket, were obtained. The
10.106 Public Health and Nutrition Over the last decade,
number of each who said they missed a street sign was
many Americans have been able to stop smoking. However, a
recorded. The data will be used to determine whether the
recent survey suggests that asthmatic children are more likely
proportion of women who say they missed a street sign is
to be exposed to second-hand smoke than children without
greater than the proportion of men who say so.
asthma.19 In a random sample of 300 children without asthma,
c. Random samples of teens ages 13–19 from two different
132 were regularly exposed to second-hand smoke, and in a
school districts were obtained. The number of teens who
random sample of 325 children with asthma, 177 were regularly
own a cell phone was recorded. The data will be used to
exposed to second-hand smoke. Is there any evidence to sug-
determine whether there is any difference in the proportion
gest that the proportion of children with asthma who are
of teens who own a cell phone in the two districts.
exposed to second-hand smoke is greater than the proportion
d. Random samples of Americans who received an income
for children without asthma? Use a 5 0.01.
tax refund were obtained. All were classified by income
level (low versus high) and asked whether they intended 10.107 Public Policy and Political Science A recent
to use their refund to pay outstanding bills. The data will survey suggested that more than one-quarter of registered
be used to determine whether the proportion of high- U.S. voters believe an armed revolution might be necessary to
income Americans who pay bills with tax refunds is 0.10 protect our liberties.20 In a random sample of 250 voters in
greater than the proportion of low-income Americans who Western states, 73 indicated that an armed revolution might be
pay bills with tax refunds. necessary, and in a random sample of 275 voters in Eastern
 10.4 Comparing Two Population Proportions Using Large Samples 511

states, 85 said that an armed revolution might be necessary. Is 10.111 Travel and Transportation Many people who com-
there any evidence to suggest that the proportion of voters in mute to work by car in New York City every day use either the
the West who believe that an armed revolution might be neces- George Washington Bridge or the Lincoln Tunnel. A random
sary is different from the proportion of voters in the East? Use sample of commuters who use one of these two routes was
a 5 0.05. Write a Solution Trail for this problem. obtained, and each was asked whether they carpooled to work.
The data are given in the following table.
10.108 Technology and the Internet In its Millennium
Development Goals Report, the United Nations suggested that Commuting Sample Number who
by the end of 2013 there will be 6.8 billion cell-phone subscrip- route size carpool
tion plans, and approximately 2.7 billion people will be con-
Bridge 1055 530
nected to the Internet.21 A random sample of households in
Brazil and Russia was obtained, and the number of people Tunnel 1663 825
connected to the Internet was recorded for each. The data are a. Verify that the nonskewness criterion inequalities are
given in the following table. satisfied.
Sample Number connected b. Is there any evidence to suggest that the proportion of
Country size to the Internet carpoolers crossing the George Washington Bridge is
greater than the proportion of carpoolers using the
Brazil 326 161 Lincoln Tunnel? Use a 5 0.01.
Russia 387 210 c. Find the p value associated with the hypothesis test in
part (b).
Is there any evidence to suggest that the proportion of house-
holds connected to the Internet is greater in Russia than in 10.112 Conspiracy Theory In a recent survey, Americans
Brazil? Use a 5 0.01. were asked about 20 popular conspiracy theories. Thirty-seven
percent of voters believe global warming is a hoax, 21% are
10.109 Marketing and Consumer Behavior Many critics
certain a UFO crashed in Roswell, New Mexico, and 7% believe
have been known to say, “They sure don’t make movies like the moon landings were faked.22 (You might consider investi-
they used to.” To assess Americans’ opinions of movies, a ran- gating the theory about lizard people who control our society.)
dom sample of people was obtained and each was asked about The survey results were often very different according to politi-
the quality of movies. The data are given in the following table. cal affiliation. A random sample of Democrats and Republicans
Sample Number who said were asked if they believe pharmaceutical companies invent
Age group size movies are getting better new diseases to make money. The resulting data are given in the
following table.
18–29 347 238
30–49 387 221 Number who believe
Political Sample pharmaceutical companies
a. Conduct the appropriate hypothesis test to determine affiliation size invent diseases
whether there is any evidence to suggest that the true
Democrats 788 137
proportion of 18–29-year-olds who believe movies are
getting better is greater than the proportion of 30–49-year- Republicans 866 105
olds. Use a 5 0.001. Is there any evidence to suggest that the proportion of voters
b. Find the p value associated with the hypothesis test in
who believe pharmaceutical companies invent diseases to make
part (a). money is different for Democrats and Republicans? Use
10.110 Public Health and Nutrition Several years a 5 0.01.
ago, most doctors believed that it was not necessary to 10.113 Medicine and Clinical Studies According to the
take any dietary supplement. Now, because many Americans do National Institute of Allergy and Infectious Diseases, approxi-
not eat a healthy, balanced diet, many physicians recommend a mately 54.6% of all U.S. citizens test positive to one or more
once-a-day multivitamin. A random sample of people was allergens. Between 9% and 16% suffer from hay fever. A ran-
obtained and asked whether they regularly take a multivitamin. dom sample of people who suffer from hay fever was obtained,
The data are given in the following table. and each was treated with either a conventional antihistamine
Sample Number who take or butterbur extract. The number of subjects who experienced
Group size a multivitamin relief from hay fever was recorded for each group. The result-
ing data are given in the following table.
Men 490 181
Women 428 214 Sample Number who
Treatment size experienced relief
Is there any evidence that the proportion of women who take
Antihistamine 255 71
a multivitamin is greater than the proportion of men? Use
a 5 0.005. Write a Solution Trail for this problem. Butterbur extract 237 55
512 CH AP T ER 1 0 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

a. Compute the sample proportion of people who Respondents were asked to rate how safe or unsafe they felt in
experienced relief for each treatment. their neighborhood. The data are given in the following table.24
b. Conduct the appropriate hypothesis test to determine
Number who felt safe
whether the proportion of people who experience relief
Sample in their neighborhood
due to the antihistamine is different from the proportion
District size after dark
of people who experience relief from butterbur extract.
Use a 5 0.01. Hunter Woods 213 141
10.114 Public Health and Nutrition A survey was con- Lake Anne 218 155
ducted concerning physical activity of adults in two states.
Is there any evidence to suggest that the true proportion of resi-
Random samples of adults were obtained from Arizona and
dents who feel safe after dark is different in the two districts?
from West Virginia, and they were all asked whether they
Use a 5 0.05.
consider themselves physically inactive. The data are given
in the following table. 10.118 Technology and the Internet The percentage of
Sample Number who are Canadians who use technology is very high, but a recent survey
State size physically inactive suggests that they greatly overrate their tech savviness. Approx-
imately 60% of Canadians rated themselves as B or better for
Arizona 1122 163 tech savviness, but a large proportion of respondents could not
West Virginia 1181 205 explain roaming, data usage, or online security.25 A random
sample of Canadians from two regions was obtained and asked
Is there any evidence to suggest that the proportion of adults to rate their tech savviness. The results are given in the follow-
who consider themselves physically inactive is greater in West ing table.
Virginia than in Arizona? Use a 5 0.001 and find the p value. Number who rated
10.115 Manufacturing and Product Development Sample their tech savviness
Blenko Specs has two different processes for the manufacture Region size B or better
of optical lenses supplied to the military. A random sample of Edmonton 566 345
finished lenses was obtained from each process, and each lens Thunder Bay 617 330
was carefully inspected for defects. Of the 106 lenses from
Process A, eight were defective, and 12 of the 121 lenses from Is there any evidence to suggest that the true proportion of
Process B were defective. Canadians who rate their tech savviness as B or better is greater
a. Compute the sample proportion of defectives for each in Edmonton than in Thunder Bay? Use a 5 0.01.
process.
b. Check the nonskewness criterion and verify that the
Extended Applications
inequalities are satisfied.
c. Conduct a hypothesis test to determine whether there 10.119 Marketing and Consumer Behavior Americans
is any evidence that the sample proportion of defective have many sources for daily news, for example, local television
lenses is different for Process A and Process B. Use shows, public radio, or national newspapers. A random sample
a 5 0.05. of Americans was obtained and classified by age. Each person
was asked whether he or she obtained news every day from
10.116 Sports and Leisure A major league sports franchise
three specific sources. The data are given in the following table.
can contribute a great deal to the local economy and unite an
entire region. In a recent survey, a random sample of adults in Age group
the Portland, Oregon, area were asked if they would support a
18- to 29-year-olds 30- to 49-year-olds
National Football League team. The data are given in the fol-
lowing table.23 Number who Number who
News Sample obtained news Sample obtained news
Sample Number who would
source size every day size every day
County size support an NFL team
Nightly 570 103 462 120
Clackamas 469 117 network
Multnomah 1985 337 news
Cable 450 108 520 182
Is there any evidence to suggest that the proportion of residents
news
who would support an NFL team is different in the two coun-
networks
ties? Use a 5 0.01.
Internet 546 197 568 239
10.117 Psychology and Human Behavior In a recent sur-
vey, residents of Reston, Virginia, were asked about their quality a. Conduct the appropriate hypothesis test to determine
of life, characteristics of the community, child care, and crime. whether there is evidence that the true proportion of
 10.5 Comparing Two Population Variances or Standard Deviations 513

18- to 29-year-olds who obtain news every day from obtained asked whether they planned any landscaping within
nightly network news shows is less than the true the next year. The data are given in the following table.
proportion of 30- to 49-year-olds who obtain news every
day from nightly network news shows. Use a 5 0.05. Sample Number who are
Find the p value associated with this hypothesis test. Residence size planning to landscape
b. Conduct the appropriate hypothesis test to determine Homeowner 261 90
whether there is evidence that the true proportion of 18- to Condominium owner 303 65
29-year-olds who obtain news every day from cable news
networks is less than the true proportion of 30- to Is there any evidence to suggest that the proportion of home-
49-year-olds who obtain news every day from cable news owners planning a landscaping project is more than 0.10 greater
networks shows. Use a 5 0.01. Find the p value than the proportion of condominium owners planning a land-
associated with this hypothesis test. scaping project? Use a 5 0.01.
c. Conduct the appropriate hypothesis test to determine
whether there is evidence that the true proportion of 18- to 10.122 Medicine and Clinical Studies Young children usu-
29-year-olds who obtain news every day from the Internet ally get 5–10 colds each year. To ease cold symptoms, for exam-
is different from the true proportion of 30- to 49-year-olds ple, a runny nose or sore throat, some parents give their children
who obtain news every day from the Internet. Use over-the-counter cough and cold medicines. However, many of
a 5 0.005. Find the p value associated with this these medicines are not effective and can cause serious side
hypothesis test. effects in young children. A random sample of parents of young
children was obtained and asked whether they give their children
10.120 Manufacturing and Product Development Two cough or cold medicine. The data are given in the following table.
different machines in a manufacturing facility are designed to
fill cans with 280 grams of Tang orange-flavored drink mix. A Sample Number who give
random sample of filled cans from each machine was obtained, Parent size cold medicine
and each can was carefully weighed. Of the 134 cans from Male 376 142
machine A, 10 were underfilled, and 7 of 114 cans from
Female 428 183
machine B were underfilled.
a. Compute the sample proportion of underfilled cans for
a. Is there any evidence to suggest that the true population
each machine.
proportion of males who give their children cough
b. Verify the nonskewness criterion.
medicine is different from the true proportion of females?
c. Find a 95% confidence interval for the true difference in
Use a 5 0.05.
the proportion of underfilled cans for machines A and B.
b. Find the p value associated with this hypothesis test.
d. Using the confidence interval in part (c), is there any
c. Find a 95% confidence interval for the difference in the
evidence to suggest that the proportion of underfilled cans
proportion of males who give their children cold medicine
is different for the two machines? Justify your answer.
and the proportion of females who give their children cold
10.121 Psychology and Human Behavior Historically, the medicine. Does this confidence interval support your
three most popular home-improvement projects are interior conclusion in part (a)? Why or why not?
decorating, landscaping, and expansion (respectively). A ran- d. What must be true of the respondents in order for the
dom sample of homeowners and condominium owners was hypothesis test in part (a) to be valid?

10.5 Comparing Two Population Variances

or Standard Deviations
Many practical business decisions are based on a comparison of variability, or manufac-
turing precision. For example, a company that produces a certain drug via fermentation
would like to maintain a very small variability in yield. One fermentation process may be
more reliable and less variable than another. A hardware store wants very little variability
in paint color from gallon to gallon. One paint mixer may be more precise (less variable)
than another. Even food manufacturers strive for small differences in product taste from
one batch to the next. And we may compare population variances to decide which two-
sample t test is appropriate, the pooled test or the approximate test.
S 21 and S 22 are good (unbiased) estimators for the population variances s21 and s22,
respectively. However, a hypothesis test for comparing s21 and s22 is based on a new stan-
dardization and an F distribution, introduced below.
514 CHAP T E R 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

An F distribution has positive probability only for non-negative values. The probabil-
ity density function for an F random variable is 0 for x , 0. Once again, it is important
to focus on the properties of an F distribution and the method for finding critical values
associated with this distribution.

Properties of an F Distribution
The numerator and denominator 1. An F distribution is completely determined by two parameters, the number of degrees
designations will make more of freedom in the numerator and the number of degrees of freedom in the denominator,
sense as you read on. given in that order. Both values must be positive integers (1, 2, 3, . . .) and there is, of
course, a different F distribution for every combination.
2. If X has an F distribution with v1 and v2 degrees of freedom, ( X , Fv1,v2 ) , then

v2 2v22 ( v1 1 v2 2 2 )
Why are these restrictions on v2 mX 5 , v2 $ 3 and s2X 5 , v2 $ 5 (10.11)
necessary? What do you suppose
v2 2 2 v1 ( v2 2 2 ) 2 ( v2 2 4 )
the mean is if n2 5 2? 3. Suppose X , Fv1,v2. The density curve for X is positively skewed (not symmetric), and
gets closer and closer to the x axis but never touches it. As both degrees of freedom
increase, the density curve becomes taller and more compact. See Figure 10.60.

f(x)
X ~ F10,15
X ~ F7,7

X ~ F3,4

Figure 10.60 Density curves for

several F distributions. 0 1 2 3 x

The definition and notation for an F critical value are analogous to those for Z, t, and
x2 critical values.

Definition
Fa,v1,v2 is a critical value related to an F distribution with v1 and v2 degrees of freedom. If
X , Fv1,v2, then P ( X $ Fa,v1,v2 ) 5 a.

Yes, there are three subscripts, A CLOSER L OK

but don’t panic. The notation
looks more complicated than it is. 1. Fa,v1,v2 is a value on the measurement axis in an F world such that there is a of the area
Fa,n1,n2 is simply consistent, concise (probability) to the right of Fa,v1,v2. Remember, as for a chi-square distribution, there is
notation to represent a critical no symmetry in F critical values.
value related to an F distribution. 2. Critical values are defined in terms of right-tail probability, and the F distribution
is not symmetric, so the notation here is similar to that for chi-square critical val-
ues. It will be necessary to find critical values denoted F12a,v1,v2 , where 1 2 a is
large. By definition, P ( X $ F12a,v1,v2 ) 5 1 2 a, and by the complement rule,
P ( X # F12a,v1,v2 ) 5 a. See Figure 10.61.
 10.5 Comparing Two Population Variances or Standard Deviations 515

f(x)
X ~ F!1,!2

P(X ≤ F1− ",!1,!2) = "

P(X ≥ F ",!1,!2) = "

0 F1−",!1,!2 F ",!1,!2 x

Figure 10.61 An illustration of F distribution critical values.

Notice how the degrees of 3. F critical values are related according to the following equation:
freedom switch positions.
1
F12a,v1,v2 5 . (10.12)
Fa,v2,v1
Table VII in the Appendix presents selected critical values associated with various F
distributions and right-tail probabilities. The degrees of freedom in the numerator are
given in the top row and the degrees of freedom in the denominator are given in the left
column. In the body of the table, Fa,v1,v2 is at the intersection of the appropriate row and
column. Left-tail probabilities are found using Equation 10.12. The following example
illustrates the use of Table VII in the Appendix for finding critical values associated
with an F distribution.

Example 10.14 Critical Value Look–ups

Find each critical value: a. F0.05,8,10 and b. F0.99,9,15.

SOLUTION
a. F0.05,8,10 is a critical value related to an F distribution with 8 and 10 degrees of free-
dom. By definition, if X , F8,10 , then P ( X $ F0.05,8,10 ) 5 0.05. Using Table VII in
the Appendix, for a 5 0.05, find the intersection of the v1 5 8 column and the
v2 5 10 row.
a 5 0.05
vl
v2 6 7 8 9 10
( ( ( ( ( ( ( (
8 c 3.58 3.50 3.44 3.39 3.35 c
9 c 3.37 3.29 3.23 3.18 3.14 c
10 c 3.22 3.14 3.07 3.02 2.98 c
11 c 3.09 3.01 2.95 2.90 2.85 c
12 c 3.00 2.91 2.85 2.80 2.75 c
( ( ( ( ( ( ( (
Therefore, F0.05,8,10 5 3.07 and if X , F8,10, then P ( X $ 3.07 ) 5 0.05, as illustrated
in Figure 10.62.
b. F0.99,9,15 is a critical value related to an F distribution with 9 and 15 degrees of freedom.
By definition, if X , F9,15, then P ( X $ F0.99,9,15 ) 5 0.99. Because F0.99,9,15 is in the
left tail of the distribution, use Equation 10.12.
1
F0.99,9,15 5 F120.01,9,15 5
F0.01,15,9
516 CH APT ER 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

f(x)
f(x)

X ~ F8, 10 X ~ F9,15

0.05 0.01 0.99

0 3.07 x 0 0.202 x
Figure 10.62 Visualization of F0.05,8,10 . Figure 10.63 Visualization of F0.99,9,15 .

Using Table VII in the Appendix, for a 5 0.01, find the intersection of the v1 5 15
column and the v2 5 9 row.

a 5 0.01
v1
v2 9 10 15 20 30
( ( ( ( ( ( ( (
7 c 6.72 6.62 6.31 6.16 5.99 c
8 c 5.91 5.81 5.52 5.36 5.20 c
9 c 5.35 5.26 4.96 4.81 4.65 c
10 c 4.94 4.85 4.56 4.41 4.25 c
11 c 4.63 4.54 4.25 4.10 3.94 c
( ( ( ( ( ( ( (
Figure 10.64 Use the Excel
function [Link] to find F Therefore, F0.99,9,15 5 1/4.96 5 0.202. If the random variable X , F9,15 then
critical values. P ( X $ 0.202 ) 5 0.99 and P ( X # 0.202 ) 5 0.01, as illustrated in Figure 10.63.

TRY IT NOW GO TO EXERCISE 10.128

A CLOSER L OK
1. Table VII in the Appendix is very limited. There are only three values for a and a lim-
ited number of values for v1 and v2. The TI-84 Plus C does not have a built-in function
for finding F critical values. However, the SOLVE feature may be used to find a critical
value related to any F distribution.
2. Minitab and Excel may also be used to find a critical value related to any F distribu-
tion. Minitab uses inverse cumulative probability and Excel uses right-tail probability.
For example, to find F0.99,9,15 using Minitab, let the random variable X , F9,15.
P ( X $ F0.99,9,15 ) 5 0.99 Definition of F critical value.

P ( X # F0.99,9,15 ) 5 1 2 0.99 5 0.01 The complement rule.

F0.99,9,15 5 0.2015 Use Minitab.

Hypothesis tests concerning two population variances and a confidence interval for the
ratio of two population variances are based on the following results.
For reference, we’ll call these the Let S 21 be the sample variance of a random sample of size n1 from a normal distribution
two-sample F test assumptions. with variance s21, let S 22 be the sample variance of a random sample of size n2 from a
normal distribution with variance s22, and suppose the samples are independent.
 10.5 Comparing Two Population Variances or Standard Deviations 517

This is yet another kind of 1. The random variable

standardization, a transformation
to an F distribution. S12 /s21
F5 (10.13)
S 22 /s 22
has an F distribution with n1 2 1 (from the numerator) and n2 2 1 (from the denomi-
nator) degrees of freedom.
If s21 and s22 are equal, they 2. If the null hypothesis is H0: s21 5 s22 , then the random variable simplifies to
cancel out. S 21 /s21
F 5 2 2 5 S 21 /S 22 . This simple ratio is the test statistic for comparing two population
S 2 / s2
variances.

Hypothesis Tests Concerning Two Population

Variances
Given the two-sample F test assumptions, a hypothesis test concerning two population
variances with significance level a has the form
This is the template for a H0: s21 5 s22
hypothesis test concerning two Ha: s21 . s22, s21 , s22, or s21 2 s22
population variances, sometimes
called a two-sample F test. TS: F 5 S 21 /S 22
RR: F $ Fa,n121,n221, F # F12a,n121,n221 or
F # F12a/2,n121,n221 or F $ Fa/2,n121,n221

Using the same assumptions, a confidence interval for the ratio of two population vari-
ances can be derived. Let X , Fn121,n221 and find an interval that captures 1 2 a in the
middle of this F distribution. Manipulate the inequality to sandwich the ratio s21 /s22.

How to Find a 100(1 2 a)% Confidence Interval

for the Ratio of Two Population Variances
Given the two-sample F test assumptions, a 100 ( 1 2 a ) % confidence interval for s21 /s22
is given by
s21 1 s21 1
a , b (10.14)
s22 Fa/2,n121,n221 s22 F12a/2,n121,n221
Using Equation 10.12, the confidence interval can be written as
s21 1 s21
a , Fa/2,n221,n121 b (10.15)
s22 Fa/2,n121,n221 s22

The hypothesis test procedure described above can also be used to compare two popu-
lation standard deviations. And, you can take the square root of each endpoint of Equation
10.14 to find a 100 ( 1 2 a ) % confidence interval for the ratio of two population standard
deviations. The following example illustrates the hypothesis test procedure.

Example 10.15 Long-Term-Care Cost

DATA SET The cost of long-term care in a nursing home varies considerably by region and may be as
CARECOST much as $50,000 per year. Two independent samples of nursing homes in Connecticut and
in Colorado were obtained, and the cost-per-day for each was recorded. The data are given
in the following table.
518 CH AP TER 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

Connecticut (1)
270 294 174 180 314 274 160 210 255 187 271

Colorado (2)
161 150 164 109 168 172 133 148 120 157 138 94
166 116 98 168 153 118 138 116 120

a. Conduct the appropriate hypothesis test to determine whether there is any evidence
that the population variance in cost per day is different in Connecticut and
Colorado. Assume the costs per day underlying populations are normal and use
a 5 0.02.
b. Find bounds on the p value for the hypothesis test in part (a).

SOLUTION
STEP 1 The null hypothesis is that the two population variances are equal. We are look-
ing for any difference in the variances, so the alternative hypothesis is two-sided.
The underlying populations are assumed normal and the samples were obtained
independently. A two-sample F test is appropriate.
In this case, n1 5 11 and n2 5 21; a /2 5 0.01 and 1 2 a /2 5 0.99.
STEP 2 The four parts of the hypothesis test are

H0: s21 5 s22

Ha: s21 2 s22
TS: F 5 S 21 /S 22
RR: F # F12a/2,n121,n221 5 F0.99,10,20 5 1/4.41 5 0.2268 or
F $ Fa/2,n121,n221 5 F0.01,10,20 5 3.37
STEP 3 The summary statistics are

1 1
s21 5 c 638,819 2 ( 2589 ) 2 d 5 2946.25
10 11
1 1
s22 5 c 414,601 2 ( 2907 ) 2 d 5 609.46
20 21
The value of the test statistic is
s21 2946.25
f5 5 5 4.83 ( $ 3.37 )
s22 609.46
The value of the test statistic lies in the rejection region. Therefore, we reject the
null hypothesis at the a 5 0.02 significance level. There is evidence to suggest
that the two population variances are different.
STEP 4 Because the tables of critical values for F distributions are very limited, we can
X ~ F10,20
only bound the p value. Place the value of the test statistic, f 5 4.83, in an
ordered list of critical values with df 10 and 20.
3.37 # 4.83 # 5.08
F0.01,10,20 # 4.83 # F0.001,10,20
0 4.83
Therefore, 0.001 # p / 2 # 0.01
Figure 10.65 p-Value illustration: and, 0.002 # p # 0.02
p 5 2P(X $ 4.83)
5 0.0027 # 0.02 5 a The exact p value is illustrated in Figure 10.65.
Figures 10.66 through 10.69 show technology solutions.
 10.5 Comparing Two Population Variances or Standard Deviations 519

Figure 10.66 Figure 10.67 Figure 10.68 Figure 10.69 JMP tests for
2-SampFTest input 2-SampFTest hypothesis 2-SampFTest Draw equality of two population
screen. test results. results. variances.

TRY IT NOW GO TO EXERCISE 10.138

The following example involves constructing a confidence interval for the ratio of two
population variances.

Example 10.16 Importance of Coral Reefs

Coral reefs are diverse and valuable ecosystems. These vast ocean environments support
approximately 4000 species of fish, lead to medical advances, and contribute to local
economies through tourism. Independent random samples of brain corals were obtained
from two Caribbean reefs. The diameter of each polyp was carefully measured (in mm).
The data are summarized in the following table.

Navassa Island n1 5 16, s 21 5 0.267

U.S. Virgin Islands n2 5 21, s 22 5 0.172

Construct a 90% confidence interval for the ratio of population variances in diameter of
brain coral polyps.
SOLUTION
STEP 1 The samples are independent; the sample sizes and sample variances are given.
A confidence interval for the ratio of two population variances is appropriate.
STEP 2 Find the critical values.

1 2 a 5 0.90 1 a 5 0.10 1 a /2 5 0.05 Find a/2.

Fa/2,n121,n221 5 F0.05,15,20 5 2.20 Critical value, left endpoint. Use Table VII in the Appendix.

Fa/2,n221,n121 5 F0.05,20,15 5 2.33 Critical value, right endpoint. Use Table VII in the Appendix.

STEP 3 Use Equation 10.15.

s 21 1 s 21
a , Fa/2,n221,n121 b Equation 10.15.
s 22 Fa/2,n121,n221 s 22
0.267 1 0.267
5a , ( 2.33 ) b Use sample variances and critical values.
0.172 2.20 0.172
5 ( 0.7056, 3.6169 ) Simplify.

(0.7056, 3.6169) is a 90% confidence interval for the ratio of the population
variances.

TRY IT NOW GO TO EXERCISE 10.141

Suppose a two-sample t test will be used to compare two population means. The
hypothesis test presented in this section is often used first to compare the population
variances. The results and conclusion suggest the appropriate hypothesis test concerning
population means from Section 10.2, according to whether or not there is evidence that
the two population variances are unequal.
520 CH A PTE R 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

Technology Corner
Procedure: Hypothesis tests and confidence intervals concerning two population variances.
Reconsider: Example 10.15, solution, and interpretations.

Crunchlt!
There is no built-in function to conduct hypothesis tests nor to construct a confidence interval concerning two population
variances. However, the F Distribution Calculator can be used to find critical values.

TI-84 Plus C
Use the built-in function 2-SampFTest to conduct a hypothesis test concerning two population variances. Input is either
data in lists or summary statistics. There is no built-in function to construct a confidence interval for the ratio of population
variances.
1. Enter the Connecticut data into list L1 and the Colorado data into list L2.
2. Select STAT ; TESTS; 2-SampFTest.
3. Highlight Data and enter the two lists. Set each frequency to 1 and highlight the appropriate alternative hypothesis.
See Figure 10.66.
4. Highlight Calculate and press ENTER to display the hypothesis test results. See Figure 10.67. The Draw results
are shown in Figure 10.68.

Minitab
Use the built-in function 2 Variances to conduct a hypothesis test concerning two population variances and to construct a
confidence interval for the ratio of the population variances. Input is samples in one column (with subscripts), samples in
different columns, or summarized data. Several graph options are also available.
1. Enter the Connecticut data into column C1 and the Colorado data into column C2.
2. Select Stat; Basic Statistics; 2 Variances.
3. Choose Each sample is in its own column, and enter C1 in the Sample 1 input window and C2 in the Sample 2 input
window.
4. Choose the Options option button. Select the Ratio of sample variances, enter a Confidence level, enter a Hypothesized
ratio (and value, usually 1), and an Alternative hypothesis.
5. Click OK and the results are displayed in a session and graph window. See Figures 10.70 and 10.71.

Figure 10.70 The 2 Variances output from Figure 10.71 The 2 Variances graph results
Minitab. from Minitab.
 10.5 Comparing Two Population Variances or Standard Deviations 521

Excel
Use the built-in function F-Test Two-Sample for Variances to conduct a hypothesis test concerning two population
variances. There is no built-in function to construct a confidence interval for the ratio of population variances.
1. Enter the Connecticut data into column A and the Colorado data into column B.
2. Under the Data tab, select Data Analysis; F-Test Two-Sample for Variances.
3. Enter the Variable 1 range, Variable 2 Range, a value for Alpha, and specify an Output option. There is no alternative
hypothesis option.
4. Click OK to view the summary statistics and the hypothesis test results. See Figure 10.72. The p value displayed is for a
one-sided hypothesis test. Double this value for a two-sided test.

Figure 10.72 Excel

Hypothesis test results.

SECTION 10.5 EXERCISES

Concept Check and the value of the test statistic are given. Find bounds on the
p value associated with each hypothesis test.
10.123 True/False Every F distribution is symmetric about
a. Ha: s 21 . s 22, n1 5 16, n2 5 17, f 5 2.29
its mode.
b. Ha: s 21 , s 22, n1 5 11, n2 5 16, f 5 0.34
10.124 True/False An F distribution has positive probability c. Ha: s 21 2 s 22, n1 5 7, n2 5 10, f 5 7.36
only for non-negative values. d. Ha: s 21 . s 22, n1 5 31, n2 5 26, f 5 4.26
10.125 True/False Fa,v1,v2 5 F12a,v2,v1 10.131 Consider independent random samples of sizes
10.126 True/False A hypothesis test to compare two popula- n1 5 31 and n2 5 25 from normal populations.
tion variances can also be used to test for equality of population a. Write the four parts for a one-sided, right-tailed hypothesis
standard deviations. test concerning the population variances with a 5 0.05.
b. Suppose s21 5 44.89 and s22 5 17.64. Find the value of the
10.127 Short Answer S 12 and S 22 are unbiased estimators for test statistic, and draw a conclusion about the population
the population variances s12 and s 22 , respectively, Explain what variances.
this means. c. Find bounds on the p value associated with this hypothesis
test and carefully sketch a graph to illustrate this value.
Practice
10.132 Consider the two independent random samples from
10.128 Find each of the following critical values. normal distributions given in the following table. EX10.132
a. F0.05,7,19 b. F0.05,30,25
c. F0.01,6,19 d. F0.001,40,40 Sample 1
e. F0.95,17,15 f. F0.95,12,10 89.6 61.1 83.7 74.2 60.6 50.4 82.4 79.0
g. F0.99,21,30 h. F0.999,11,8
56.5 72.2 72.4 77.1 58.2 72.3 71.6 70.3
10.129 Find each of the following critical values. 76.1 73.2
a. F0.01,20,60 b. F0.01,15,19
c. F0.05,6,8 d. F0.001,8,24 Sample 2
e. F0.99,12,9 f. F0.99,6,8
g. F0.95,10,10 h. F0.999,23,20 73.7 37.8 76.2 64.7 74.8 75.8 67.9 61.6
76.2 82.7 88.9 60.3 34.4 74.5 68.0 100.2
10.130 In each of the following problems, the null hypothesis
73.7 41.5 76.9 55.0 76.0
is H0: s21 5 s22. The alternative hypothesis, the sample sizes,
522 CHAP T ER 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

a. Write the four parts for a one-sided, left-tailed hypothesis 10.138 Medicine and Clinical Studies A study in the British
test concerning the two population variances with a 5 0.05. Medical Journal suggested that children who received a diag-
b. Compute each sample variance, find the value of the test nostic CT scan using ionizing radiation were more likely to
statistic, and draw a conclusion. develop a cancer 10 years after radiation exposure. Newer CT
c. Find bounds on the p value associated with this hypothesis scanners use less radiation. Therefore, the increased risk for
test. children today may be decreased.26 Independent random sam-
ples of hospital CT scans in the United States and England were
10.133 Consider independent random samples of sizes
obtained, and the amount of radiation for each was recorded (in
n1 5 10 and n2 5 16 from normal populations.
mSV). For the United States, n1 5 10 and s21 5 1.075; for
a. Write the four parts for a two-sided hypothesis test
concerning the population variances with a 5 0.01. England n2 5 16 and s22 5 2.786. Is there any evidence to
b. Suppose s 21 5 426.42 and s 22 5 88.36. Find the value of suggest that the variability in CT radiation per scan is different
the test statistic, and draw a conclusion about the for these two countries? Use a 5 0.05 and assume normality.
population variances. 10.139 Travel and Transportation Most airlines now
c. Find bounds on the p value associated with this hypothesis charge passengers to check a bag and impose a surcharge if the
test. weight of the bag is over 50 pounds. Independent random sam-
10.134 In each of the following problems, the sample sizes and ples of checked luggage on Delta and American flights were
the confidence level are given. Find the appropriate F critical obtained, and the weight (in pounds) of each was recorded. For
values for use in constructing a confidence interval for the ratio Delta, n1 5 25, and s21 5 96.23; for American, n2 5 21 and
of the population variances. s22 5 194.02. Is there any evidence to suggest that the variabil-
a. n1 5 10, n2 5 10, 90% ity in checked baggage weight is different for these two air-
b. n1 5 21, n2 5 31, 98% lines? Use a 5 0.05 and assume normality.
c. n1 5 9, n2 5 7, 98% 10.140 Sports and Leisure Many basketball purists believe
d. n1 5 41, n2 5 31, 99.8% that the three-point shot (a shot from behind the three-point
10.135 In each of the following problems, the sample sizes, line, 22 feet from the basket) has dramatically changed the
the sample variances, and the confidence level are given. game, for the worse. Independent random samples of attempted
Assume the underlying populations are normal and the samples shots from National Basketball Association games played in
were obtained independently. Find the associated confidence 1975 (prior to the three-point shot) and in 2013 were obtained.
interval for the ratio of the population variances. The shot distance (in feet) was recorded for each attempt. The
data are given in the following table.
a. n1 5 10, s 21 5 17.2, n2 5 9, s 22 5 15.6, 90%
2
b. n1 5 16, s 1 5 54.1, n2 5 16, s 22 5 32.6, 98% Sample Sample
2
c. n1 5 16, s 1 5 3.35, n2 5 31, s 22 5 4.59, 98% Year size variance
d. n1 5 31, s 1 5 126.8, n2 5 41, s 22 5 155.3, 99.8%
2
1975 61 12.25
10.136 Use Table VII in the Appendix and linear interpolation 2013 61 26.01
to approximate each critical value. Verify each approximation
using technology. Is there any evidence to suggest that the variability in shot
a. F0.05,25,15 b. F0.99,20,32 distance is greater in the year 2013 than it was in 1975? Use
c. F0.01,10,56 d. F0.025,15,20 a 5 0.01 and assume normality. (Why do you suppose there is
e. F0.995,10,7 f. F0.05,35,35 greater variability in shot distance with a three-point shot?)
10.141 Sports and Leisure A Laurel Downs racetrack
Applications official believes there is less variability in winning times for a
10.137 Biology and Environmental Science In a recent race in which the purse is at least $10,000, called a stakes race.
study conducted by the NOAA, the aerosol light absorption Independent random samples of ordinary races and stakes races
coefficient was measured (in Mm21) at randomly selected loca- were obtained, and the winning time (in seconds) for each race
tions in Africa and in South America. The resulting data are was recorded. The summary statistics were as follows: ordinary
summarized in the following table. race, n1 5 26 and s21 5 110.25; stakes race, n2 5 26 and
s22 5 38.44.
Sample Sample a. Write the four parts for a hypothesis test to check for
Country size variance evidence of the official’s assertion. Use a 5 0.01, assume
Africa 10 243.36 normality, and find the critical value using technology.
Conduct the hypothesis test and draw a conclusion.
South America 21 51.84
b. Construct a 98% confidence interval for the ratio of
population variances.
Is there any evidence that the population variance in aerosol
light absorption coefficient is greater in Africa than in South 10.142 Sports and Leisure A study was conducted to
America? Use a 5 0.05 and assume normality. compare the variability in times for men and women involved
 10.5 Comparing Two Population Variances or Standard Deviations 523

in collegiate swimming events. Independent random samples 12.42-mile race up Pikes Peak in Colorado features over 156
of 800-meter freestyle competitors were obtained, and the time turns, grades of 7%, and a finish line at 14,110 feet. Indepen-
(in minutes) was recorded for each swimmer. The data are dent random samples of two classes of cars in the 2013 race
summarized in the following table. were obtained, and the speed (in mph) for each was recorded.
The summary statistics are given in the following table.28
Sample Sample
Group size variance Sample Sample
Class size variance
Men 11 0.1025
Women 12 0.1241 Exhibition Powersports 9 44.05
Heavyweight Supermoto 9 9.76
a. Find the critical values necessary to construct a 95%
confidence interval for the ratio of population variances. Is there any evidence to suggest that the variability in speed for
b. Construct the confidence interval. the Exhibition Powersports class is greater than the variability
in speed for the Heavyweight Supermoto class? Use a 5 0.01
10.143 Take the Stairs The World Summit Wing Hotel in
and assume normality.
China is Beijing’s tallest hotel and hosts the Vertical Run. There
are 81 floors, 330 meters, and 2041 steps to reach the roof. 10.147 Manufacturing and Product Development The
Independent random samples of men’s and women’s times (in Akashi-Kaikyo bridge in Japan is the longest suspension bridge
minutes) from the 2013 run were obtained.27 VERTRUN in the world, with a main span of 1991 meters. Two million
a. Is there any evidence of a difference in variability of times workers took 10 years to construct this bridge using 181,000
for men and women who finished this vertical run? Use tons of steel and 1.4 million cubic meters of concrete. Indepen-
a 5 0.05 and assume normality. dent random samples of suspension bridges in China and the
b. Find bounds on the p value associated with this hypothesis United States were obtained, and the span (in meters) of each
test. was recorded. For China, n1 5 14 and s21 5 66,096.8; and for
the United States, n2 5 10 and s 22 5 59,524.5. Is there any
10.144 Manufacturing and Product Development A sail-
evidence to suggest a difference in the variability of the span
boat manufacturer has two machines for constructing main
of suspension bridges in China and the United States? Use
mast poles with diameter designed to be 76.2 mm. Small vari-
a 5 0.05 and assume normality.
ability in production is very important to ensure boat control
and safety. Independent random samples of mast poles pro-
duced on each machine were obtained, and the diameter of each Extended Applications
was carefully measured. The summary statistics were as fol- 10.148 Physical Sciences Crude oil pumped from ocean
lows: Machine A, n1 5 7, s21 5 0.0231; Machine B, wells contains salt that must be removed before the oil is
n2 5 8, s22 5 0.0096. Conduct the appropriate hypothesis test refined. Otherwise, equipment would erode quickly. Indepen-
to determine whether there is any evidence of a difference in dent random samples of unrefined crude oil from two ocean
variability of mast-pole diameter between the two machines. wells were obtained, and the percentage of salt in each sample
Assume normality, use a 5 0.05, find the p value associated was recorded. The data are given in the following table.
with this test, and use this value to draw a conclusion.
Sample Sample
10.145 Manufacturing and Product Development Tower
Oil well size variance
cranes along a city skyline often indicate the success of eco-
nomic development efforts. In early January 2013, more than North Sea 21 56.40
50 tower crane permits were in use in Washington, D.C. Inde- Antarctica 31 82.42
pendent random samples of items lifted by cranes at two differ-
ent sites were obtained, and the weight (in tons) of each item a. Conduct the appropriate test to determine whether there is
was recorded for each. The summary statistics are given in the evidence of any difference in variability of salt content
following table. between these two wells. Use a 5 0.10 and assume
normality.
Sample Sample
b. Use Table VII in the Appendix to find bounds on the p
Site size variance
value for this hypothesis test. Use technology to find an
New York Avenue 31 109.86 exact p value.
Jefferson at Market Place 31 339.43 10.149 Public Health and Nutrition Saccharin is a low-
calorie sweetener used in sugar-free foods and beverages.
Is there any evidence to suggest that the variability in item
According to the U.S. Food and Drug Administration, the
weight at Jefferson is greater than at New York Avenue? Use
acceptable daily intake (ADI) of saccharin is 5 mg for a per-
a 5 0.01 and assume normality.
son with a body weight of 60 kg. If saccharin is used as an
10.146 Sports and Leisure The Pikes Peak International additive, it must be included on the food label and cannot
Hill Climb is also known as The Race to the Clouds. This exceed certain limits. Independent random samples of
524 CH AP T ER 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

12-ounce bottles of iced tea from two different manufacturers the null hypothesis, H0: s 21 5 s 22 , is true, then the test statistic
were obtained, and the amount of saccharin in each drink was F 5 S 21 / S 22 is approximately normal with
measured (in mg). The summary statistics were as follows:
Fishing Creek, n1 5 20, s21 5 7.84; Honest Tea, n2 5 15, n2 2 1 2 ( n2 2 1 ) 2 ( n1 1 n2 2 4 )
mF 5 and s2F 5
s 22 5 2.89. n2 2 3 ( n1 2 1 )( n2 2 3 ) 2 ( n2 2 5 )
a. Conduct the appropriate test to determine whether there is An approximate hypothesis test is based on standardizing F to a
any difference in the population variance of saccharin Z random variable.
amounts. Use a 5 0.02.
b. Find bounds on the p value associated with this hypothesis
The four parts of the hypothesis test are
test. H0: s 21 5 s 22
10.150 Fuel Consumption and Cars The length of time Ha: s 21 . s 22, s 21 , s 22, or s 21 2 s 22
brake pads last in an automobile varies depending on driving ( S 21 / S 22 ) 2 3 ( n2 2 1 ) / ( n2 2 3 ) 4
style and the type of car. Brake pads are made from organic, TS: Z 5
semimetallic, metallic, or synthetic materials, and typically 2 ( n2 2 1 ) 2 ( n1 1 n2 2 4 )
last between 30,000 and 70,000 miles. Independent random Å ( n1 2 1 )( n2 2 3 ) 2 ( n2 2 5 )
samples of cars in for an inspection at dealerships and private RR: Z $ za, Z # 2za or 0 Z 0 $ za / 2
garages were obtained and the width (in mm) of the brake pad
The National Wind Energy Assessment contains data from
on the front driver’s side was measured for each. The data are
975 stations and includes measurements of wind speed and
given in the following table.
wind power density. Suppose independent random samples of
Sample Sample wind power density (in watts/m2) during the winter were
Location size variance obtained from two stations. The data are summarized in the
following table.
Dealership 41 1.056
Private garage 41 2.771 Sample Sample Sample
Location size mean variance
a. Is there any evidence to suggest that the variance in
brake-pad widths of cars in for inspection is greater at Chanute 31 207 95.35
private garages than at dealerships? Use a 5 0.01 and Dodge City 31 283 53.68
assume normality.
b. Find bounds on the p value associated with this hypothesis a. Write the four parts of a large-sample, two-sided,
test. approximate test based on the standard normal
distribution to determine whether there is any evidence to
suggest that the two population variances are different.
Challenge
Conduct the test using a 5 0.05.
10.151 Physical Sciences When we are comparing two pop- b. Conduct an exact hypothesis test based on the F
ulation variances, if both sample sizes, n1 and n2, are large and distribution. Compare your answer to part (a).

CHAPTER 10 SUMMARY
Concept Page Notation / Formula / Description

Independent samples 463 Two samples are independent if the process of selecting individuals or objects in
sample 1 has no effect on the selection of individuals or objects in sample 2.
Paired data set 463 The result of matching each individual or object in sample 1 with a similar indi-
vidual or object in sample 2.
( n1 2 1 ) S 21 1 ( n2 2 1 ) S 22
Pooled estimator for 475 S 2p 5
n1 1 n2 2 2
the common variance
= X1 1 X2
Combined estimate of 502 Pc 5
n1 1 n2
the common population
proportion
 Chapter 10 Summary 525

Summary of confidence intervals

Parameter Assumptions 100 ( 1 2 a ) % Confidence interval
m1 2 m2 n1, n2 large, independence, s21 s22
s21, s22 known, or ( x1 2 x2 ) 6 za/2 1
Å n1 n2
normality, independence,
s21, s22 known.
m1 2 m2 Normality, independence, 1 1
s21, s22 unknown but equal. ( x1 2 x2 ) 6 ta/2,n11 s 2p a 1
n222 b
Å n1 n2
( n1 2 1 ) s21 1 ( n2 2 1 ) s22
s2p 5
n1 1 n2 2 2
m1 2 m2 Normality, independence, s21 s22
s21, s22 unknown and unequal. ( x1 2 x2 ) 6 ta/2,v 1
Å n1 n2
s21 s22 2
a 1 b
n1 n2
v5
( s21 /n1 ) 2 ( s22 /n2 ) 2
1
n1 2 1 n2 2 1
mD 5 m1 2 m2 Normality, n pairs, sD
d 6 ta/2,n21
dependence. !n
p1 2 p2 n1, n2 large, = = = =
= = p1 ( 1 2 p1 ) p2 ( 1 2 p2 )
nonskewness, ( p1 2 p2 ) 6 za/2 1
Å n1 n2
independence.
s21 Normality, independence. s 21 1 s 21 1
a , b
s22 s 22 Fa/2,n121,n221 s 22 F12a/2,n121,n221

Summary of hypothesis tests

Null Alternative
hypothesis Assumptions hypothesis Test statistic Rejection region
m1 2 m2 5 D0 n1, n2 large, m1 2 m2 . D0 ( X1 2 X2 ) 2 D0 Z $ za
independence, Z5
m1 2 m2 , D0 s21 s22 Z # 2za
s21, s22 known, or m1 2 m2 2 D0 1 0 Z 0 $ za / 2
Å n1 n2
normality,
independence,
s21, s22 known.
m1 2 m2 5 D0 Normality, m1 2 m2 . D0 ( X1 2 X2 ) 2 D0 T $ ta,n11 n222
independence, T5
m1 2 m2 , D0 1 1 T # 2ta,n11 n222
s21, s22 unknown, m1 2 m2 2 D0 Sp2 a1 b 0 T 0 $ ta/2,n11 n222
Å n1 n2
s 21 5 s 22 .
( n1 2 1 ) S21 1 ( n2 2 1 ) S22
S2p 5
n1 1 n2 2 2
 526 CH APT ER 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

Null Alternative
hypothesis Assumptions hypothesis Test statistic Rejection region
m1 2 m2 5 D0 Normality, m1 2 m2 . D0 ( X 1 2 X2 ) 2 D0 Tr $ ta,v
independence, Tr 5
m1 2 m2 , D0 S 21 S 22 Tr # 2ta,v
s21, s22 unknown, m1 2 m2 2 D0 1 0 Tr 0 $ ta/2,v
Å n1 n2
s21 2 s22. 2 2 2
s1 s2
a 1 b
n1 n2
v5
( s21 /n1 ) 2 ( s22 /n2 ) 2
1
n1 2 1 n2 2 1
mD 5 D0 Normality, mD . D0 D 2 D0 T $ ta,n21
n pairs, T5
mD , D0 SD/ !n T # 2ta,n21
dependence. mD 2 D0 0 T 0 $ ta/2,n21
= =
p1 2 p2 5 0 n1, n2 large, p1 2 p2 . 0 P1 2 P2 Z $ za
nonskewness, p1 2 p2 , 0 Z5
= = 1 1 Z # 2za
independence. p1 2 p2 2 0 Pc ( 1 2 Pc ) a 1 b 0 Z 0 $ za / 2
Å n 1 n2
= X1 1 X2
Pc 5
n1 1 n2
= =
p1 2 p2 5 D0 n1, n2 large, p1 2 p2 . D0 ( P1 2 P2 ) 2 D0 Z $ za
nonskewness, p1 2 p2 , D0 Z5 = = = = Z # 2za
P1 ( 1 2 P1 ) P2 ( 1 2 P2 )
independence. p1 2 p2 2 D0 1 0 Z 0 $ za / 2
Å n1 n2

s21 5 s22 Normality, s21 . s22 S 21 F $ Fa,n121,n221

independence. F5
s21 , s22 S 22 F # F12a,n121,n221
s21 2 s22 F # F12a/2,n121,n221 or
F $ Fa/2,n121,n221

CHAPTER 10 EXERCISES

10 APPLICATIONS different from the mean amount of corrosive material carried

by trucks in Virginia? Use a 5 0.01 and assume each underly-
10.152 Travel and Transportation The U.S. Department ing distribution of weight is normal.
of Transportation requires vehicles transporting hazardous
materials to use special placards indicating the type of cargo. 10.153 Taser Accuracy Tasers are nonlethal weapons used
There are many other regulations involving containers, by police to subdue dangerous people. This electroshock
separation of various materials, and gross weight. Independent weapon uses electrical current to disrupt control of an individu-
random samples of trucks carrying corrosive materials were al’s muscles. The police forces in England maintain careful
stopped on highways in North Carolina and in Virginia, and records concerning the use of Tasers. Records from 2010–2012
the weight (in kg) of the hazardous material was recorded. indicate the following.29
The summary statistics and known variances are given in the
following table. Police force Lancashire West Mercia

Sample Sample Population Number of Taser uses 186 138

State size mean variance Number of chest hits 120 62

North Carolina 22 835.6 3192.25 Is there any evidence to suggest that the true proportion of chest
Virginia 25 884.2 3956.41 hits is greater in Lancashire than in West Mercia? Use a 5 0.01.
Is there any evidence to suggest that the mean amount of 10.154 Biology and Environmental Science Soybeans are
corrosive material carried by trucks in North Carolina is an important source of oil and protein and are also used to
 Chapter 10 Exercises 527

produce many food additives. The leading producers of a. What is the common characteristic that makes these data
soybeans are the United States, Brazil, Argentina, and China. paired?
The first genetically modified (GM) soybeans were grown in b. Assume normality. Conduct the appropriate hypothesis
the United States in 1996, and now GM soybeans are grown in test to determine whether there is any evidence that the
at least nine countries. Independent random samples of soybean aluminum arrow flies faster. Use a 5 0.05.
farmers in the United States and Brazil were obtained, and the c. Find bounds on the p value associated with this
number growing GM soybeans was recorded. The data are hypothesis test.
given in the following table.
10.158 Manufacturing and Product Development
Sample Number of Raytheon Aircraft is now manufacturing business jets with a
Country size GM soybean farmers molded carbon fiber fuselage instead of aluminum. This
reduces the overall weight of the plane, speeds production time,
United States 238 202
and increases cabin space. The total wall thickness of a carbon
Brazil 162 104 fiber fuselage is 0.81 inch versus 3 inches for aluminum, and
a. Find the sample proportion of GM soybean farmers for the variability in thickness is theoretically much smaller also.
each country. Verify the nonskewness criterion. Independent random samples of the two fuselage types were
b. Conduct the appropriate hypothesis test to determine obtained, and the thickness was measured (in inches) on each.
whether there is any evidence that the true proportion of The data are given in the following table.
GM soybean farmers in the United States is 0.15 greater Fuselage Sample Sample
than in Brazil. Use a 5 0.05. type size variance
c. Find the p value associated with this hypothesis test.
Aluminum 9 0.0196
10.155 Biology and Environmental Science Maple syrup
Carbon fiber 11 0.0025
producers in New York and Vermont collect sweet-water sap
from sugar maples and black maples in early spring. It takes Is there any evidence to suggest that the variability in fuselage
approximately 30–50 gallons of sap to yield, through boiling thickness is less for carbon fiber fuselages? Use a 5 0.01.
and evaporation, 1 gallon of maple syrup. Independent
random samples of maple trees in both states were obtained, 10.159 Biology and Environmental Science Piers on public
and the amount of sap collected from each tree was recorded. beaches are usually supported by widely spread piles or pillars and
Assume the underlying populations are normal, with equal can extend a thousand feet into the ocean. Many piers are
variances. Is there any evidence to suggest that the population extensions of boardwalks, and visitors frequently fish or simply
mean amount of sap from trees in New York is different from sightsee along these walkways. Longer piers tend to be more
the population mean amount of sap from trees in Vermont? susceptible to wind and storm damage. Independent random
Use a 5 0.01. samples of concrete and wooden piers on public beaches along the
SAP
California and Florida coasts were obtained and the length (in feet)
10.156 Physical Sciences Recycling of aluminum, glass, of each was recorded. Assume the underlying populations are
newspapers, and magazines is good for the environment and normal, with equal variances. Is there any evidence to suggest that
the economy. In 2013, San Francisco had the highest recy- the population mean pier length in California is greater than the
cling rate in the United States30 (recycling rate 5 tons population mean pier length in Florida? Use a 5 0.01. PIERS
collected for recycling/tons of all waste generated). Despite
efforts to make the process easier, many people still do not 10.160 Public Health and Nutrition In case you missed it,
recycle. Independent random samples of residents in Ohio the United Nations declared 2008 as the International Year of the
and in Florida were obtained and asked whether they recycle Potato. Seriously, potatoes are a good source of carbohydrates,
newspapers. Of the 909 Ohio residents, 700 said they protein, fiber, and potassium. However, the amount of each
recycled newspapers, and 691 of the 923 Florida residents element varies depending on where the potato is grown. Indepen-
said they recycled newspapers. dent random samples of medium-sized potatoes from Russia and
a. Is there any evidence to suggest that the population China were obtained, and the amount of potassium (in mg) was
proportion of residents in Ohio who recycle newspapers measured in each. The data are summarized in the following table.
is greater than the population proportion of residents in Sample
Florida? Use a 5 0.01. Sample Sample standard
b. Find the p value for this hypothesis test. Location size mean deviation
10.157 Sports and Leisure Archery target shooters use a Russia 25 896.8 92.9
variety of arrows made from wood, carbon, aluminum, or even China 30 866.0 120.0
platinum. One measure of the quality of an arrow (and bow) is
the speed of the arrow when shot. A random sample of archers Assume normality and equal population variances. Is there any
was obtained, and each was asked to shoot a carbon arrow and a evidence to suggest that the population mean potassium level is
similarly made aluminum arrow. The speed (in feet per second) different for a medium-sized potato in Russia and China? Use
of each arrow was measured. ARCHERY a 5 0.05.
528 CH APT ER 10 Confidence Intervals and Hypothesis Tests Based on Two Samples or Treatments

10.161 Biology and Environmental Science The n2 5 1448 and 87 were contaminated. Is there any evidence
moisture content in bulk grain is important, because high to suggest that the true proportion of contaminated cat food
values can encourage the development of fungi. Potential is different from the true proportion of contaminated dog
buyers want to know how much water they are buying along food? Use a 5 0.05.
with their grain. Two direct methods for measuring the
moisture content are by means of a chemical reaction (with
iodine in the presence of sulfur dioxide) and by distillation. A EXTENDED APPLICATIONS
random sample of bulk grain was obtained, and the moisture 10.165 Economics and Finance The U.S. Internal
content of each grain sample was measured as a percentage of Revenue Service estimates that the average taxpayer takes
water using each method. Assuming normality, conduct the approximately six hours to complete Form 1040. A study was
appropriate hypothesis test to determine whether there is any conducted to examine the amount of time it takes to complete
difference in the population mean moisture content of bulk this dreaded form, by income level. Independent random
grain measured by chemical reaction and by distillation. Use samples of federal filers in two income ranges were obtained,
a 5 0.05. GRAIN and the length of time (in hours) to complete Form 1040 was
10.162 Psychology and Human Behavior Two recent recorded for each. The summary statistics are given in the
studies suggest that people who drive really nice cars exhibit following table.
some very bad habits. In one study, as a car approached a Income level Sample Sample Sample
crosswalk, a person stepped into the road, and the driver’s (in dollars) size mean variance
reaction was recorded. In another, similar study, independent
50,000–<100,000 17 4.56 1.5625
random samples of drivers were selected, and their behavior
was observed at a four-way intersection. For luxury-car drivers, 100,000–<200,000 14 6.58 15.0544
n1 5 217 and 130 cut ahead in the usual four-way rotation. For
Assume the underlying populations are normal.
ordinary-car drivers, n2 5 182 and 82 violated the four-way-
a. Conduct an F test to determine whether there is any
intersection rotation rule. Is there any evidence to suggest that
evidence that the two population variances are different.
the proportion of luxury-car drivers with insufferable driving
Use a 5 0.02.
habits is greater than the proportion of ordinary-car drivers with
b. Using your conclusion from part (a), conduct the
similar habits? Use a 5 0.01. Note: The largest group of
appropriate test for evidence that the mean time to
driving-rule etiquette violators were men, ages 35–50, with
complete Form 1040 for the lower-income level is less
blue BMWs.31
than the mean time for the higher-income level. Use
10.163 Public Policy and Political Science California law a 5 0.05. State your conclusion and find bounds on the
requires fuel outlets to install special catch basins designed to p value.
contain gasoline leaks in underground storage tanks. Owners
10.166 Public Policy and Political Science In many
who do not comply can face stiff fines and other penalties.
Independent random samples of gasoline stations around Los states, lawyers are encouraged to do pro bono work by both
Angeles and around San Francisco were obtained, and each their firms and judicial advisory councils. However, in recent
station was inspected for catch basins. Sixteen of 140 stations years lawyers have been devoting more time to paying clients
near Los Angeles had no catch basins, and 12 of 126 in San and less time to pro bono legal aid. Independent random
Francisco were not complying with the law. samples of lawyers from two large firms were obtained, and
a. Find the sample proportion of stations without
the number of pro bono hours for the past year was recorded
catch basins near each city. Verify the nonskewness for each lawyer. The summary statistics are given in the
criterion. following table.
b. Is there any evidence that the population proportion Sample Sample Sample
of stations in noncompliance with the law is different Law firm size mean variance
near Los Angeles and near San Francisco? Use
Dewey, Cheatum, & Howe 26 75.1 5.92
a 5 0.01.
Fine, Howard, & Fine 26 80.9 5.65
10.164 Manufacturing and Product Development Dur-
ing Summer 2013, Procter and Gamble (P&G) recalled 30 Assume the underlying populations are normal, with equal
different types of cat and dog food because they may have variances.
been contaminated with Salmonella. While pets can become a. Is there any evidence to suggest the mean number of
ill from eating contaminated foods, the Centers for Disease yearly pro bono hours is different at these two law firms?
Control and Prevention also reminded people to wash their Use a 5 0.01.
hands thoroughly after handling pet food. Independent b. Construct a 99% confidence interval for the difference in
random samples of P&G cat foods and dog foods were mean pro bono hours.
obtained, and each was tested for Salmonella. For cat food, c. Does the confidence interval in part (b) support your
n1 5 1250 and 50 were contaminated, and for dog food, conclusion in part (a)? Explain.
 Chapter 10 Exercises 529

10.167 Economics and Finance Online investing has grown Sample

with the Internet and with companies like E*TRADE and Sample Sample standard
Ameritrade. Independent random samples of investors were Dwelling size mean deviation
obtained and asked whether they traded online within the past
Single detached 15 8.68 6.32
year. The data are given in the following table, by portfolio size.
Double/duplex 15 13.16 9.20
Sample Number of Apartment 16 23.70 16.52
Portfolio size online traders
Assume all three underlying populations are normal, with equal
Less than $100,000 348 132 variances. Conduct the appropriate hypothesis tests to deter-
At least $100,000 226 65 mine which pairs of population mean toluene levels are
different. Use a 5 0.01 in each test.
a. Compute the sample proportions and verify the
nonskewness criterion. 10.170 Physical Sciences Independent random samples of
b. Construct a 95% confidence interval for the difference in ore taken from two high-grade gold mines were obtained, and
population proportions of online investors. the gold value (in grams/tonne) was measured for each. The
c. Using the interval in part (b), is there any evidence to summary statistics were
suggest that the population proportion of online investors
is different for these two portfolio classifications? Justify El Aguila mine n1 5 8 x1 5 15.6 s1 5 5.2
your answer. Dolaucothi mine n2 5 11 x2 5 26.8 s2 5 21.6
10.168 Look, in the Sky The American Meteor Society Assume normality and unequal variances.
maintains a running fireball-tracking system. All reports are a. Conduct the appropriate hypothesis test to determine
analyzed and grouped according to several variables. In 2012, whether there is any evidence that the population mean
there were 2302 fireball reports, and 590 of these were con- gold value at the El Aguila mine is less than at the
firmed by 2–5 witnesses. As of August 2013, there were 507 Dolaucothi mine. Use a 5 0.05.
reports and 142 were confirmed by 2–5 witnesses.32 Assume the b. Your conclusion in part (a) should be that there is no
samples are independent. Is there any evidence to suggest that evidence to suggest a difference. Explain why this result is
the proportion of fireball reports that are confirmed by 2–5 correct even though the sample means are very far apart.
witnesses is different for these two time periods? Use
a 5 0.05. LAST STEP
10.169 Biology and Environmental Science Benzene, 10.171 Are people who live at higher altitudes
toluene, ethylbenzene, m-, p-xylenes, and o-xylene are volatile slimmer? Random samples were obtained for 125 adults
organic compounds that are found in residential environments living in Denver, Colorado (the Mile High City), and 150 adults
and can cause severe health problems, for example, dizziness, living in New Orleans, Louisiana (where the mean elevation
tremors, eye, ear, and throat irritation. The Canadian Health is zero feet above sea level). Using body mass index as a
Measures Survey was administered to more than 5000 measure of obesity, 38 of those living in Denver and 61 of
individuals in order to predict the presence of these com- those living in New Orleans were classified as obese. Is there
pounds. Suppose a random sample of these respondents was any evidence to suggest that the people living in Denver are
selected, and the concentration of toluene was measured in thinner? That is, is there any evidence to suggest that the
each residence (in mg/m3 ). The results are given in the proportion of adults classified as obese in Denver is less than
following table.33 the proportion of adults classified as obese in New Orleans?
Use a 5 0.05.