ch15 PDF

Chapter
15
© Kitch/Age Fotostock America, Inc.
Nonparametric Methods1
W hat kind of soda is that on your desk? Regardless of the particular brand or flavor, it is more 15.1 The Sign Test
likely than last year that it is a diet soda. According to a Dow Jones report (August 19, 2002), diet 15.2 The Wilcoxon Signed-
sodas represent 30% of the soft drink market, a 6.6% sales increase over the prior year (compared Rank Test for Two
to a 3.1% increase in sales of regular soft drinks). Even so, diet soft drinks represent only 18.2% Dependent Samples
of the total U.S. carbonated soft drink market, according to John Sicher, editor and publisher of
15.3 The Wilcoxon Rank
Beverage Digest. We can conduct hypothesis tests to determine people’s preferences for one type Sum Test for Two
of soft drink over another. Independent Samples
15.4 The Kruskal-Wallis Test
The hypothesis tests discussed so far in this text are called parametric tests. In those tests, we used 15.5 The Spearman Rho
the normal, t, chi-square, and F distributions to make tests about population parameters such as means, Rank Correlation
Coefficient Test
proportions, variances, and standard deviations. In doing so, we made some assumptions, such as the
15.6 The Runs Test for
assumption that the population from which the sample was drawn was normally distributed. This chap-
Randomness
ter discusses a few nonparametric tests. These tests do not require the same kinds of assumptions,
and hence, they are also called distribution-free tests.
Nonparametric tests have several advantages over parametric tests: They are easier to use and
understand; they can be applied to situations in which parametric tests cannot be used; and they do
not require that the population being sampled is normally distributed. However, a major problem with
nonparametric tests is that they are less efficient than parametric tests. The sample size must be larger
for a nonparametric test to have the same probability of committing the two types of errors.
Although there are a large number of nonparametric tests that can be applied to conduct tests
of hypothesis, this chapter discusses only six of them: the sign test, the Wilcoxon signed-rank test, the
Wilcoxon rank sum test, the Kruskal-Wallis test, the Spearman rho rank correlation coefficient test, and
the runs test for randomness.
1
Tables VIII to XII that are needed for this chapter are given at the end of this chapter. Tables IV and VI are in
Appendix C of the book.
562
563 Chapter 15 Nonparametric Methods
15.1 The Sign Test

The sign test is one of the easiest tests to apply to test hypotheses. It uses only plus and minus
signs. The sign test can be used to perform the following types of tests:
1. To determine the preference for one product or item over another, or to determine whether
one outcome occurs more often than another outcome in categorical data. For example, we
may test whether or not people prefer one kind of soft drink over another kind.
2. To conduct a test for the median of a single population. For example, we may use this pro-
cedure to test whether the median rent paid by all tenants in a city is different from $1250.
3. To perform a test for the median of paired differences using data from two dependent sam-
ples. For example, we may use this procedure to test whether the median score on a stan-
dardized test increases after a preparatory course is taken.
Definition
Sign Test The sign test is used to make hypothesis tests about preferences, a single median, and
the median of paired differences for two dependent populations. We use only plus and minus
signs to perform these tests.
In the following subsections we discuss these tests for small and large samples.
15.1.1 Tests About Categorical Data

Data that are divided into different categories for identification purposes are called categorical
data. For example, people’s opinions about a certain issue—in favor, against, or no opinion—
produce categorical data. This subsection discusses how to perform tests about such data using
the sign test procedure. We discuss two situations in which such tests can be performed: the
small-sample case and the large-sample case.
The Small-Sample Case

When we apply the sign test for categorical data, if the sample size is 25 or less (i.e., n 25),
we consider it a small sample. Table VIII: Critical Values of X for the Sign Test (that appears
at the end of this chapter), is based on the binomial probability distribution. This table gives
the critical values of the test statistic for the sign test when n 25 using the binomial proba-
bility distribution.
The sign test can be used to test whether or not customers prefer one brand of a product
over another brand of the same type of product. For example, we can test whether customers
have a higher preference for Coke or for Pepsi. This procedure can also be used to test whether
people prefer one of two alternatives over the other. For example, given a choice, do people pre-
fer to live in New York City or in Los Angeles?
EXAMPLE 15–1
The Top Taste Water Company produces and distributes Top Taste bottled water. The company
Performing sign test
wants to determine whether customers have a higher preference for its bottled water than for its
with categorical data:
small sample.
main competitor, Spring Hill bottled water. The Top Taste Water Company hired a statistician
to conduct this study. The statistician selected a random sample of 10 people and asked each
of them to taste one sample of each of the two brands of water. The customers did not know
the brand of each water sample. Also, the order in which each person tasted the two brands
of water was determined randomly. Each person was asked to indicate which of the two sam-
ples of water he or she preferred. The following table shows the preferences of these 10
individuals.
15.1 The Sign Test 564
Person Brand Preferred

1 Spring Hill
2 Top Taste
3 Top Taste
4 Neither
5 Top Taste
6 Spring Hill
7 Spring Hill
8 Top Taste
9 Top Taste
10 Top Taste
Based on these results, can the statistician conclude that people prefer one brand of bottled
water over the other? Use the significance level of 5%.
Solution We use the same five steps to perform this test of hypothesis that we used in ear-
lier chapters.
Step 1. State the null and alternative hypotheses.

If we assume that people do not prefer either brand of water over the other, we would expect
about 50% of the people (among those who show a preference) to indicate a preference for Top
Taste water and the other 50% to indicate a preference for Spring Hill water. Let p be the pro-
portion of all people who prefer Top Taste bottled water. The two hypotheses are as follows:
H0: p .50 1People do not prefer either of the two brands of water2
H1: p .50 1People prefer one brand of water over the other2
The null hypothesis states that 50% of the people prefer Top Taste water over Spring Hill water
(and, hence, the other 50% prefer Spring Hill water). Note that we do not consider people who
have no preference, and we drop them from the sample. If we fail to reject H0, we will conclude
that people do not prefer one brand over the other. However, if we reject H0, we will conclude
that the percentage of people who prefer Top Taste water over Spring Hill water is different from
50%. Thus, the conclusion will be that people prefer one brand over the other.
Step 2. Select the distribution to use.
We use the binomial probability distribution to make the test. Note that here there is only one
sample and each member of the sample is asked to indicate a preference if he or she has one. We
drop the members who do not indicate a preference and then compare the preferences of the re-
maining members. Also note that there are three outcomes for each person: (1) prefers Top Taste
water, (2) prefers Spring Hill water, or (3) has no preference. We are to compare the two outcomes
with preferences and determine whether more people belong to one of these two outcomes than
to the other. All such tests of preferences are conducted by using the binomial probability distri-
bution. If we assume that H0 is true, then the number of people who indicate a preference for Top
Taste bottled water (the number of successes) follows the binomial distribution, with p .50.
Step 3. Determine the rejection and nonrejection regions.
Note that 10 people were selected to taste the two brands of water and indicate their pref-
erences. However, one of these individuals stated no preference. Hence, only 9 of the 10 people
have indicated a preference for one or the other of the two brands of bottled water. To con-
duct the test, the person who has shown no preference is dropped from the sample. Thus, the
true sample size is 9; that is, n 9.
The significance level for the test is .05. Let X be the number of people in the sample of 9
who prefer Top Taste bottled water. Here X is called the test statistic. To establish a decision
rule, we find the critical values of X from Table VIII for n 9. From that table, for n 9
and .05 for a two-tailed test, the critical values of X are 1 and 8. Note that in a two-tailed
test we read both the lower and the upper critical values, which are shown in Figure 15.1.
Thus, we will reject the null hypothesis if either fewer than two or more than seven people in
nine indicate a preference for Top Taste bottled water.
Figure 15.1 0 or 1 2 to 7 8 or 9
X
Rejection region Nonrejection region Rejection region
Critical Value(s) of X In a sign test for a small sample, the critical value of X is obtained from
Table VIII. If the test is two-tailed, we read both the lower and the upper critical values from that
table. However, we read only the lower critical value if the test is left-tailed, and only the upper
critical value if the test is right-tailed. Also note that which column we use to obtain this critical
value depends on the given significance level and on whether the test is two-tailed or one-tailed.
Step 4. Calculate the value of the test statistic.

To record the results of the experiment, we mark a plus sign for each person who prefers
Top Taste bottled water, a minus sign for each person who prefers Spring Hill bottled water,
and a zero for the one who indicates no preference. This listing is shown in Table 15.1.
Table 15.1
Person Brand Preferred Sign

1 Spring Hill
2 Top Taste
3 Top Taste
4 Neither 0
5 Top Taste
6 Spring Hill
7 Spring Hill
8 Top Taste
9 Top Taste
10 Top Taste
Now, we count the number of plus signs (the sign that belongs to Top Taste bottled water
because p in H0 refers to Top Taste water). There are six plus signs, indicating that six of the
nine people in the sample stated a preference for Top Taste bottled water. Note that the sam-
ple size is 9, not 10, because we drop the person with zero sign. Thus,
Observed value of X 6
Observed Value of X The observed value of X is given by the number of signs that belong to the
category whose proportion we are testing for.
Step 5. Make a decision.

Because the observed value of X 6 falls in the nonrejection region (see Figure 15.1), we
fail to reject H0. Hence, we conclude that our sample does not indicate that people prefer
either of these two brands of bottled water over the other.
Note that it does not matter which outcome p refers to. If we assume that p is the propor-
tion of people who prefer Spring Hill water, then X will denote the number of people in a sam-
ple of n who prefer Spring Hill water. The observed value of X this time would be 3, which
is the number of minus signs in Table 15.1. From Figure 15.1, X 3 also falls in the nonre-
jection region. Hence, we again fail to reject the null hypothesis.
The Large-Sample Case

If we are testing a hypothesis about preference for categorical data and n 25, we can use the
normal probability distribution as an approximation to the binomial probability distribution.
The Large-Sample Case If n 25, the normal distribution can be used as an approximation to
the binomial probability distribution to perform a test of hypothesis about the preference for cat-
egorical data. The observed value of the test statistic z, in this case, is calculated as
1X .52 m
z .
s
where X is the number of units in the sample that belong to the outcome referring to p.
We either add .5 to X or subtract .5 from X to correct for continuity (see Section 6.7 of
Chapter 6). We will add .5 to X if the value of X is less than or equal to n2, and we will
subtract .5 from X if the value of X is greater than n2. The values of the mean and standard
deviation are calculated as
m np and s 1npq
Example 15–2 illustrates the procedure for the large-sample case.
EXAMPLE 15–2
A developer is interested in building a shopping mall adjacent to a residential area. Before
Performing sign test with
granting or denying permission to build such a mall, the town council took a random sample
categorical data:
of 75 adults from adjacent areas and asked them whether they favor or oppose construction of large sample.
this mall. Of these 75 adults, 40 opposed construction of the mall, 30 favored it, and 5 had no
opinion. Can you conclude that the number of adults in this area who oppose construction of
the mall is higher than the number who favor it? Use .01.
Solution Again, each adult in the sample has to pick one of three choices: oppose, favor,
or have no opinion. And we are to compare two outcomes—oppose and favor—to find out
whether more adults belong to the outcome indicated by oppose. We can use the sign test here.
To do so, we drop the subjects who have no opinion—that is, the adults who belong to the
outcome that is not being compared. In our example, five adults have no opinion. Thus, we
drop these adults from our sample and use the sample size of n 70 for the purposes of this
test. Let p be the proportion of adults who oppose construction of the mall and q be the pro-
portion who favor it. We apply the five steps to make this test.

H0: p .50 and q .50 1The two proportions are equal2
H1: p 7 .50 or p 7 q 1The proportion of adults who oppose the mall is greater
than the proportion who favor it2
The null hypothesis here states that the proportion of adults who oppose and the proportion who
favor construction of the mall are both .50, which means that p .50 and q .50. The alter-
native hypothesis is that p q, which means that more adults oppose construction of the mall
than favor it. Note that H1 states that p .50 and q .50. In other words, of those who have
an opinion, more than 50% oppose and less than 50% favor construction of the shopping mall.
As explained earlier, we will use the sign test to perform this test. Although 75 adults were
asked their opinion, only 70 of them offered it and 5 did not. Hence, our sample size is 70;
that is, n 70. Because it is a large sample (n 25), we can use the normal approximation
to perform the test.

Because H1 states that p .50, our test is right-tailed. Also, .01. From Table IV (the
standard normal distribution table) in Appendix C, the z value for 1.0 .01 .9900 area to
the left is approximately 2.33. Thus, the decision rule is that we will not reject H0 if z 2.33
and we will reject H0 if z
2.33. Thus, the nonrejection region lies to the left of z 2.33
and the rejection region to the right of z 2.33, as shown in Figure 15.2.
Figure 15.2
α = .01
0 z
Nonrejection region Rejection region
2.33

Assuming that the null hypothesis is true, we expect (about) half of the adults in the
population to oppose construction of the mall and the other half to favor it. Thus, we expect
p .50 and q .50. Note that we do not count the people who have no opinion. The mean
and standard deviation of the binomial distribution are
m np 701.502 35
s 1npq 1701.5021.502 117.5 4.18330013
In this example, p refers to the proportion of adults who oppose construction of the mall.
Hence, X refers to the number in 70 who oppose the mall. Thus,
n 70
X 40 and 35
2 2
Because X is greater than n2, the observed value of the test statistic z is
1X .52 m 140 .52 35
z 1.08
s 4.18330013
Because the observed value of z 1.08 is less than the critical value of z 2.33, it falls
in the nonrejection region. Hence, we do not reject H0. Consequently, we conclude that the
number of adults who oppose construction of the mall is not higher than the number who favor
its construction.
15.1.2 Tests About the Median of a Single Population

The sign test can be used to test a hypothesis about a population median. Recall from Chapter 3
that the median is the value that divides a ranked data set into two equal parts. For example, if
the median age of students in a class is 24, half of the students are younger than 24 and
half are older. This section discusses how to make a test of hypothesis about the median of a
population.

If n 25, we use the binomial probability distribution to test a hypothesis about the median of
a population. The procedure used to conduct such a test is very similar to the one explained in
Example 15–1.
EXAMPLE 15–3
A real estate agent claims that the median price of homes in a small midwest city is $137,000.
Performing sign test about a
A sample of 10 houses selected by a statistician produced the following data on their prices.
population median:
Home 1 2 3 4 5 6 7 8 9 10 small sample.
Price ($) 147,500 123,600 139,000 168,200 129,450 132,400 156,400 188,210 198,425 215,300
Using the 5% significance level, can you conclude that the median price of homes in this city
is different from $137,000?
Solution Using the given data, we prepare Table 15.2, which contains a sign row. In this
row, we assign a positive sign to each price that is above the claimed median price of $137,000
and a negative sign to each price that is below the claimed median price.
In Table 15.2, there are seven plus signs, indicating that the prices of seven houses are higher
than the claimed median price of $137,000, and there are three minus signs, showing that the
prices of three homes are lower than the claimed median price. Note that if one or more val-
ues in a data set are equal to the median, then each of them is assigned a zero value and dropped
from the sample. Next, we perform the following five steps to perform the test of hypothesis.
Table 15.2
Home 1 2 3 4 5 6 7 8 9 10
Sign

H0: Median price $137,000 1Real estate agent’s claim is true2
H1: Median price $137,000 1Real estate agent’s claim is false2
For a test of the median of a population, we employ the sign test procedure by using the
binomial probability distribution if n 25. Since in our example n 10, which is less than
25, we use the binomial probability distribution to conduct the test.
In our example, n 10 and .05. The test is two-tailed. Let X be the test statistic that
represents the number of plus signs in Table 15.2. From Table VIII, the (lower and upper) crit-
ical values of X are 1 and 9. Using these critical values, Figure 15.3 shows the rejection and
nonrejection regions. Thus, we will reject the null hypothesis if the observed value of X is ei-
ther 1 or 0, or 9 or 10. Note that because X represents the number of plus signs in the sam-
ple, its lowest possible value is 0 and its highest possible value is 10.
Figure 15.3 0 or 1 2 to 8 9 or 10
X

The observed value of X is given by the number of plus signs in Table 15.2. Thus,
Observed value of X 7
Observed Value of X When using the sign test to perform a test about a median, we can use
either the number of positive signs or the number of negative signs as the observed value of X if
the test is two-tailed. However, the observed value of X is equal to the larger of these two num-
bers (the number of positive and negative signs) if the test is right-tailed, and equal to the small-
er of these two numbers if the test is left-tailed.

The observed value of X 7 falls in the nonrejection region in Figure 15.3. Hence, we do
not reject H0 and conclude that the median price of homes in this city is not different from
$137,000.

For a test of the median of a single population, we can use the normal approximation to the
binomial probability distribution when n 25. The observed value of z, in this case, is calculated
as in a test of hypothesis about the preference for categorical data (see the rule described on
page 635 and Example 15–2). Example 15–4 explains the procedure for such a test.
EXAMPLE 15–4
A long-distance phone company believes that the median phone bill (for long-distance calls)
Performing sign test about a
is at least $70 for all the families in New Haven, Connecticut. A random sample of 90 fami-
population median:
large sample.
lies selected from New Haven showed that the phone bills of 51 of them were less than $70
and those of 38 of them were more than $70, and 1 family had a phone bill of exactly $70.
Using the 1% significance level, can you conclude that the company’s claim is true?
Solution We use the usual five steps to test this hypothesis.

The company’s claim is that the median phone bill is at least $70. Hence, the two hypotheses
are as follows:
H0: Median
$70 1Company’s claim is true2
H1: Median 6 $70 1Company’s claim is false2
This is a test about the median and n 25. Hence, to conduct this test, we can use the nor-
mal distribution as an approximation to the binomial probability distribution.
The test is left-tailed and .01. From Table IV (the standard normal distribution table),
the z value for .01 area in the left tail is 2.33. Note that the z value is negative because it is
a left-tailed test. Thus, we will reject H0 if the observed value of z is 2.33 or lower, and we
will not reject H0 if the observed value of z is greater than 2.33. Figure 15.4 shows the re-
jection and nonrejection regions.
Figure 15.4
α = .01
z
Rejection 0
region Nonrejection region
–2.33

In our example, 51 phone bills out of 90 are below the hypothesized median, 38 are above
it, and 1 is exactly equal to the median. When we perform this test, we drop the value or val-
ues that are exactly equal to the median. Thus, after dropping one value that is equal to the
median, our sample size is 51 38 89; that is, n 89. Let a phone bill below the median
be represented by a minus sign and one above the median by a plus sign. Then, in these 89
bills there are 51 minus signs (for values less than the median) and 38 plus signs (for values
greater than the median). If the given claim is true, we would expect (about) half plus signs
and half minus signs. Let p be the proportion of plus signs in 89. Then, we would expect
p .50 if H0 is true. Hence, the mean and the standard deviation of the binomial distribution
are calculated as follows:
n 89 p .50 q 1 p .50
m np 891.502 44.50
s 1npq 1891.5021.502 4.71699057
In our example, 51 phone bills are below the median and 38 are above the median. Because
it is a left-tailed test, X 38, which is the smaller of the two numbers (51 and 38). Conse-
quently, the z value is calculated as follows. Note that we have added .5 to X because the value
of X is less than n2, which is 892 44.5.
1X .52 m 138 .52 44.50
z 1.27
s 4.71699057
Because z 1.27 is greater than the critical value of z 2.33, we do not reject H0.
Hence, we conclude that the company’s claim that the median phone bill is at least $70 seems
to be true.
Observation
Note that in Example 15–4 there were 51 minus signs and 38 plus signs. We assigned the smaller
of these two numbers to X so that X 38 to calculate the observed value of z. We did so to obtain
a negative value of the observed z because the test is left-tailed and the critical value of z is nega-
tive. If we assigned 51 as the value of X, we would obtain z 1.27 as the observed value of z,
which does not make sense. Let X1 be the number of plus signs and X2 the number of minus signs
in a test about the median. Then, we can establish the following rules to calculate the observed
value of X.
1. If the test is two-tailed, it does not matter which of the two values, X1 and X2, is assigned
to X to calculate the observed value of z.
2. If the test is left-tailed, X should be assigned a value equal to the smaller of the values of
X1 and X2.
3. If the test is right-tailed, X should be assigned a value equal to the larger of the values of
X1 and X2.
Note that the rule to calculate the observed value of z here is the same as explained on page
635 for the large-sample case for a test of hypothesis about the preference for categorical data.
15.1.3 Tests About the Median Difference

Between Paired Data
We can use the sign test to perform a test of hypothesis about the difference between the me-
dians of two dependent populations using the data obtained from paired samples. We learned
in Section 10.4 of Chapter 10 that two samples are paired samples when, for each data value
collected from one sample, there is a corresponding data value collected from the second sam-
ple, and both data values are collected from the same source. In this section we discuss the
small-sample and the large-sample cases to conduct such tests.

If n 25, we use the binomial probability distribution to perform a test about the difference
between the medians of paired data. In such a case, Table VIII is used to find the critical val-
ues of the test statistic. Example 15–5 illustrates this procedure.
EXAMPLE 15–5
A researcher wanted to find the effects of a special diet on systolic blood pressure in adults.
Performing sign test
She selected a sample of 12 adults and put them on this dietary plan for three months. The
about the median of paired dif-
following table gives the systolic blood pressure of each adult before and after the completion
ferences: small samples.
of the plan.
Before 210 185 215 198 187 225 234 217 212 191 226 238
After 196 192 204 193 181 233 208 211 190 186 218 236
Using the 2.5% significance level, can we conclude that the dietary plan reduces the median
systolic blood pressure of adults?
Solution We find the sign of the difference between the two blood pressure readings of each
adult by subtracting the blood pressure after completion of the dietary plan from the blood
pressure before the plan. A plus sign indicates that the plan reduced that person’s blood pres-
sure and a minus sign means that it increased blood pressure. Table 15.3 gives the signs of the
differences.
Table 15.3
Before 210 185 215 198 187 225 234 217 212 191 226 238
After 196 192 204 193 181 233 208 211 190 186 218 236
Sign of difference
(before after)
Next we perform the five steps for testing the hypothesis.

Let M denote the difference in median blood pressure readings before and after the dietary
plan. The null and alternative hypotheses are as follows:
H0: M 0 1The dietary plan does not reduce the median blood pressure2
H1: M 7 0 1The dietary plan reduces the median blood pressure2
The alternative hypothesis is that the dietary plan reduces the median blood pressure,
which means that the median systolic blood pressure of all adults after the completion
of the dietary plan is lower than the median systolic blood pressure before the completion
of the dietary plan. In this case, the median of the paired differences will be greater
than zero.
The sample size is small (that is, n 12 25), and we do not know the shape of the dis-
tribution of the population of paired differences. Hence, we use the sign test with the binomial
probability distribution.
Because the test is right-tailed, n 12, and .025, the critical value of X from Table VIII
is 10. Note that we use the upper critical value of X in Table VIII because, as indicated by the
sign in H1, the test is right-tailed. Thus, we will reject the null hypothesis if the observed value
of X is greater than or equal to 10, and we will not reject H0 otherwise. The rejection and non-
rejection regions are shown in Figure 15.5.
Figure 15.5 0 to 9 10 to 12
X

In our sample data, the blood pressure of 10 adults decreases and that of 2 adults increases
after the dietary plan. Note that there are 10 plus signs and 2 minus signs in Table 15.3. When-
ever the test is right-tailed, the observed value of X is equal to the larger of these two num-
bers. Thus, for our example, observed value of X 10.
Because the observed value of X 10 falls in the rejection region, we reject H0. Hence,
we conclude that the dietary plan reduces the median blood pressure of adults.

In Example 15–5, we used Table VIII to find the critical value of the test statistic X. However,
Table VIII goes up to n 25 only. If n 25, we can use the normal distribution as an ap-
proximation to the binomial distribution to conduct a test about the difference between the me-
dians of paired data. The following example illustrates such a case.
EXAMPLE 15–6
Many students suffer from math anxiety. A statistics professor offered a two-hour lecture Performing sign test
on math anxiety and ways to overcome it. A total of 42 students attended this lecture. The about the median of paired dif-
students were given similar statistics tests before and after the lecture. Thirty-three of ferences: large samples.
the 42 students scored higher on the test after the lecture, 7 scored lower after the lecture,
and 2 scored the same on both tests. Using the 1% significance level, can you conclude
that the median score of students increases as a result of attending this lecture? Assume
that these 42 students constitute a random sample of all students who suffer from math
anxiety.
Solution Let M be the median of the paired differences between scores of students before
and after the test, where a paired difference is obtained by subtracting the score after the lec-
ture from the score before the lecture. In other words,
Paired difference Score before Score after
Thus, a positive paired difference means that the score before the lecture is higher than the
score after the lecture for that student, and a negative paired difference indicates that the score
before the lecture is lower than the score after the lecture for that student. Thus, there are
33 minus signs, 7 plus signs, and 2 zeros.

H0: M 0 1The lecture does not increase the median score2
H1: M 6 0 1The lecture increases the median score2
The alternative hypothesis is that the lecture increases the median score, which means that the
median score after the lecture is higher than the median score before the lecture. In this case,
the median of the paired differences will be less than zero.

Here, n 40. Note that to find the sample size, we drop the students whose score did not
change. Because n 25, we can use the normal distribution to test this hypothesis about the
median of paired differences.

The test is left-tailed and .01. From Table IV, the critical value of z for .01 area in the
left tail is 2.33. Thus, the decision rule is that we will reject the null hypothesis if the observed
value of z is 2.33 or smaller, and we will not reject H0 otherwise. The rejection and nonre-
jection regions are shown in Figure 15.6.
Figure 15.6
α = .01
z
Rejection 0
region Nonrejection region
–2.33

If the null hypothesis is true (that is, the lecture does not increase the median score),
then we would expect (about) half of the students to score higher and the other half to score
lower after the lecture than before. Thus, we would expect (about) half plus signs and half
minus signs in the population. In other words, if p is the proportion of plus signs, we would
expect p .50 when H0 is true. Hence, the mean and standard deviation of the binomial
distribution are
m np 401.502 20
s 1npq 1401.5021.502 3.16227766
In our example, 33 of the students scored higher after the lecture and 7 scored lower after the
lecture. Thus, there are 33 minus signs and 7 plus signs. We assign the smaller of these two
values to X when the test is left-tailed. Here X 7, and the observed value of z is calculated
as follows. Here, because the value of X is less than n2, we add .5 to X.
1X .52 m 17 .52 20
z 3.95
s 3.16227766
Because the observed value of z 3.95 is less than the critical value of z 2.33, it
falls in the rejection region. Consequently, we reject H0 and conclude that attending the math
anxiety lecture increases the median test score.
Remember Again, remember that if the test is left-tailed, X is assigned the value equal to the smaller num-
ber of plus or minus signs. On the other hand, if the test is right-tailed, X is assigned the value
equal to the larger number of plus or minus signs. Note that the rule to calculate the observed
value of z here is the same as explained on page 635 for the large-sample case for a test of hy-
pothesis about the preference for categorical data.
EXERCISES
CONCEPTS AND PROCEDURES
15.1 Briefly explain the meaning of categorical data and give two examples.
15.2 When we use the sign test for categorical data, how large a sample size is required to permit the use
of the normal distribution for determining the rejection region?
15.3 When we use the sign test for the median of a single population, how small must the sample size be
to require the use of Table VIII ?
15.4 When we use the sign test for the difference between the medians of two dependent populations,
how large must n be for the large-sample case?
15.5 Determine the rejection region for each of the following sign tests for categorical data.
a. H0: p .50, H1: p .50, n 15, .05
b. H0: p .50, H1: p .50, n 20, .01
c. H0: p .50, H1: p .50, n 30, .05
15.6 In each case below, n is the sample size, p is the proportion of the population possessing a certain
characteristic, and X is the number of items in the sample that possess that characteristic. In each case,
perform the appropriate sign test using .05.
a. n 14, X 10, H0: p .50, H1: p .50
b. n 10, X 1, H0: p .50, H1: p .50
c. n 30, X 12, H0: p .50, H1: p .50
d. n 27, X 20, H0: p .50, H1: p .50
15.7 In each case below, n is the sample size and X is the appropriate number of plus or minus signs as
defined in Section 15.1.2. In each case, perform the appropriate sign test using .05.
a. n 10, X 8, H0: Median 28, H1: Median 28
b. n 11, X 1, H0: Median 100, H1: Median 100
c. n 26, X 3, H0: Median 180, H1: Median 180
d. n 30, X 6, H0: Median 55, H1: Median 55
15.8 In each case below, M is the difference between two population medians, n is the sample size, and
X is the appropriate number of plus or minus signs as defined at the end of Section 15.1.3. In each case,
perform the appropriate sign test using .01.
a. n 20, X 6, H0: M 0, H1: M 0
b. n 8, X 8, H0: M 0, H1: M 0
c. n 29, X 4, H0: M 0, H1: M 0
APPLICATIONS
15.9 In Pine Grove, the city water is safe to drink but some people think it has a slightly unpleasant taste
due to chemical treatment. Some residents prefer to buy bottled water (B) but others drink the city water
(C). A random sample of 12 residents is taken. Their preferences are shown here.
B C B C C B C C C C B C
At the 5% significance level, can you conclude that the residents of Pine Grove prefer either type of drink-
ing water over the other?
15.10 A consumer organization wanted to compare two rival brands of infant car seats, Brand A and Brand
B. Fifteen families, each with a child under 12 months of age, were selected at random. Each family tested
each of the two brands of car seats for one week. The order in which each family tried the two brands
was decided by a coin toss. At the end of two weeks, each family indicated which brand it preferred. Their
preferences are listed here. The 0 indicates that one family had no preference.
A A A B A A B A A A A 0 A B A
At the 5% level of significance, can you conclude that families prefer Brand A over Brand B?
15.11 Twenty randomly chosen loyal drinkers of JW’s Beer are tested to see if they can distinguish be-
tween JW’s and its chief rival. Each of the 20 drinkers is given two unmarked cups, one containing JW’s
and the other containing the rival brand. Thirteen of the drinkers correctly indicate which cup contains
JW’s, but the other seven are incorrect. At the 2.5% level of significance, can you conclude that drinkers
of JW’s Beer are more likely to correctly identify it than not?
15.12 Three weeks before an election for state senator, a poll of 200 randomly selected voters shows that
95 voters favor the Republican candidate, 85 favor the Democratic candidate, and the remaining 20 have no
opinion. Using the sign test, can you conclude that voters prefer one candidate over the other? Use .01.
15.13 One hundred randomly chosen adult residents of North Dakota are asked whether they would pre-
fer to live in another state or to stay in North Dakota. Of these 100 adults, 55 indicate that they would like
to move to another state, 41 would prefer to stay, and 4 have no preference. At the 2.5% level of signifi-
cance, can you conclude that less than half of all adult residents of North Dakota would prefer to stay?
15.14 Three hundred randomly chosen doctors were asked, Which is the most important single factor in
weight control: diet or exercise? Of these 300 doctors, 162 felt that diet was more important, 117 favored
exercise, and 21 thought that diet and exercise were equally important. At the 1% level of significance,
can you conclude that for all doctors, the number who favor diet exceeds the number who favor exercise?
15.15 In a Gallup Poll of adults taken December 6–9, 2001, 42% reported that they frequently experi-
enced stress in their daily lives (USA TODAY, January 24, 2002). Suppose that in a recent sample of 700
adults, 370 indicated that they frequently experience such stress. Using the sign test with .01, can
you conclude that currently more than half of all adults frequently experience stress in their daily lives?
15.16 A past study claims that adults in the United States spend a median of 18 hours a week on leisure
activities. A researcher took a sample of 10 adults and asked them how many hours they spend per week
on leisure activities. She obtained the following data:
14 25 22 38 16 26 19 23 41 33
Using .05, can you conclude that the median amount of time spent per week on leisure activities by
all adults is more than 18 hours?
15.17 The manager of a soft-drink bottling plant wants to see if the median amount of soda in 12-ounce
bottles differs from 12 ounces. Ten filled bottles are selected at random from the bottling machine, and
the amount of soda in each is carefully measured. The results (in ounces) follow:
12.10 11.95 12.00 12.01 12.02 12.05 12.02 12.03 12.04 12.06
Using the 5% level of significance, can you conclude that the median amount of soda in all such bottles
differs from 12 ounces?
15.18 According to the annual USA TODAY/NFL salary survey, the median salary of offensive linemen in
the National Football League (NFL) was $589,133 in 2001 (USA TODAY, July 29, 2002). Suppose that a
recent random sample of 10 NFL offensive linemen yielded the following salaries (in thousands of dollars).
700 615 710 805 630 575 900 730 710 695
Using the 5% level of significance, can you conclude that the current median salary of all offensive line-
men in the NFL exceeds $589,133?
15.19 A city police department claims that its median response time to 911 calls in the inner city is four
minutes or less. Shown below is a random sample of 28 response times (in minutes) to 911 calls in the
inner city.
6 5 7 12 2 1.5 3.5 4 10 11 4.5 6 5 8.5
7 15 9 8 3 10 8 4.5 9 4 6 3 6 7.5
Using .01, can you conclude that the median response time to all 911 calls in the inner city is longer
than four minutes?
15.20 According to the American Community Survey conducted during the 2000 census, New Jersey’s
median household income of $54,226 was the highest among the 50 states (USA TODAY, August 6, 2001).
Suppose that in a recent random sample of 400 New Jersey households, 220 had incomes higher than
$54,226 and 180 had incomes lower than $54,226. Using the sign test at the 2% level of significance,
can you conclude that the current median household income in New Jersey differs from $54,226?
15.21 The following numbers are the times served (in months) by 35 prison inmates who were released
recently.
37 6 20 5 25 30 24 10 12 20
24 8 26 15 13 22 72 80 96 33
84 86 70 40 92 36 28 90 36 32
72 45 38 18 9
Using .01, test the null hypothesis that the median time served by all such prisoners is 42 months
against the alternative hypothesis that the median time served is less than 42 months.
15.22 Twelve sixth-grade boys who are underweight are put on a special diet for one month. Each boy is
weighed before and after the one-month dietary regime. The weights (in pounds) of these boys are recorded
here.
Before 65 63 71 60 66 72 78 74 58 59 77 65
After 70 68 75 60 69 70 81 81 66 56 79 71
Can you conclude that this diet increases the median weight of all such boys? Use the 2.5% level of sig-
nificance. Assume that these 12 boys constitute a random sample of all underweight sixth-grade boys.
15.23 Refer to Exercise 10.52 of Chapter 10. The following table shows the self-confidence test scores
of seven employees before and after they attended a course on building self-confidence.
Before 8 5 4 9 6 9 5
After 10 8 5 11 6 7 9
At the 5% level of significance, can you conclude that attending this course increases the median self-
confidence test score of all employees?
15.24 The manager at a large factory suspects that the night-shift workers use more hours of sick leave
than the day-shift workers. The workers at this factory are rotated between shifts. Each worker works the
day shift for two months, then works the night shift for two months, then goes back to the day shift for
two months, and so on. The manager at the factory selected 12 workers randomly and recorded the total
number of hours of sick leave each of these workers used during the two months of day shift and then
during the two months of night shift. The results are given in the following table.
Day shift 20 32 12 24 16 0 22 8 10 38 16 12
Night shift 16 56 0 28 36 24 40 29 30 26 32 20
Using the 5% level of significance, can you conclude that the median number of hours of sick leave used
by workers is lower for the day shift than for the night shift?
15.25 At a large bicycle factory, employees are paid by the hour to assemble bicycles. The plant man-
ager decides to test a modified-piecework payment schedule, whereby each worker will be paid a lower
hourly wage plus an additional amount for each bicycle assembled. The manager randomly selects 27 workers
and places them on the new payment schedule. For each worker in the sample, the number of bicycles
assembled during the last week under the old payment system is recorded, and then the number of bicycles
assembled during the first week under the new system is recorded. Nineteen of the workers assembled
more bicycles under the new system, seven assembled fewer, and one assembled the same number. Using
the 2% level of significance, can you conclude that the median number of bicycles assembled by all such
workers is the same under both payment systems?
15.26 A researcher suspects that two medical laboratories, A and B, tend to give different results when
determining the cholesterol content of blood samples. The researcher obtains blood samples from 30 ran-
domly selected adults and divides each sample into two parts. One part of each blood sample is sent to
Lab A, the other part to Lab B. Each lab determines the cholesterol content of each of its 30 samples and
reports the results to the researcher. The following table gives the cholesterol levels (in milligrams per
hundred milliliters) reported by the two labs.
Sample Lab A Lab B Sample Lab A Lab B

1 135 137 16 214 202
2 202 195 17 255 242
3 239 250 18 233 217
4 210 202 19 246 231
5 180 185 20 292 262
6 195 195 21 229 212
7 188 177 22 170 172
8 200 204 23 261 243
9 320 300 24 310 281
10 290 269 25 302 277
11 285 271 26 283 264
12 210 216 27 221 199
13 185 176 28 208 211
14 194 184 29 344 321
15 181 182 30 170 164
Can you conclude that the median cholesterol level for all such adults as determined by Lab A is higher
than that determined by Lab B? Use a significance level of 1%.
15.27 A dairy agency wants to test a hormone that may increase cows’ milk production. Some members
of the group fear that the hormone could actually decrease production, so a “matched pairs” test is arranged.
Thirty randomly selected cows are given the hormone, and their milk production is recorded for four weeks.
Each of these 30 cows is matched with another cow of similar size, age, and prior record of milk pro-
duction. This second group of 30 cows do not receive the hormone. The milk production of these cows is
recorded for the same time period. In 19 of these 30 pairs, the cow taking the hormone produced more
milk; in 9 of the pairs, the cow taking the hormone produced less; and in 2 of the pairs, there was no dif-
ference. Using the 5% level of significance, can you conclude that the hormone changes the median milk
production of such cows?
15.2 The Wilcoxon Signed-Rank Test

for Two Dependent Samples
The Wilcoxon signed-rank test for two dependent (paired) samples is used to test whether or
not the two populations from which these samples are drawn are identical. We can also test the
alternative hypothesis that one population distribution lies to the left or to the right of the other.
Actually, the null hypothesis in this test states that the medians of the two population distribu-
tions are equal. The alternative hypothesis states that the medians of the two populations are
not equal, or that the median of the first population is less than that of the second population,
or that the median of the first population is greater than that of the second population. This test
is an alternative to the paired-samples test discussed in Section 10.4.2 of Chapter 10. In that
section, we assumed that the paired differences have a normal distribution. Here, in the Wilcoxon
signed-rank test, we do not make that assumption. In this test, we rank the absolute differences
between the pairs of data values collected from two samples and then assign them the sign based
on which of the paired data values is larger. Then we compare the sums of the ranks with plus
and minus signs and make a decision.

If the sample size is 15 or smaller, we find the critical value of the test statistic, denoted by T,
from Table IX (given at the end of this chapter) which gives the critical values of T for the
Wilcoxon signed-rank test. We also calculate the observed value of the test statistic differently
in this test. However, when n 15, we can use the normal distribution to perform the test.
Example 15–7 describes the small-sample case for the Wilcoxon signed-rank test.
EXAMPLE 15–7
A private agency claims that the crash course it offers significantly increases the writing speed
Performing the Wilcoxon
of secretaries. The following table gives the writing speeds of eight secretaries before and after
signed-rank test for two
dependent populations:
they attended this course.
small samples.
Before 84 75 88 91 65 71 90 75
After 97 72 93 110 78 69 115 75
Using the 2.5% significance level, can you conclude that attending this course increases the
writing speed of secretaries? Use the Wilcoxon signed-rank test.
Solution We use the five steps to make the hypothesis test.

H0: The crash course does not increase the writing speed of secretaries
H1: The crash course does increase the writing speed of secretaries
15.2 The Wilcoxon Signed-Rank Test for Two Dependent Samples 578
Note that the alternative hypothesis states that the population distribution of writing speeds of
secretaries moves to the right after they attend the crash course. In other words, the center of
the population distribution of writing speeds after the crash course is greater than the center
of the population distribution of writing speeds before the crash course. If we measure the cen-
ters of the two populations by their respective medians, with MA the median of the population
distribution after the course and MB the median of the population distribution before the course,
we can rewrite the two hypotheses as follows:
H0: MA MB
H1: MA 7 MB
We are making a test for paired samples, and the distribution of paired differences is unknown.
Since n 15, we use the Wilcoxon signed-rank test procedure for the small-sample case.
As mentioned earlier, we denote the test statistic in this case by T. The critical value of T
is found from Table IX, which lists the critical values of T for the Wilcoxon signed-rank test
for small samples (n 15). Our test is right-tailed because the alternative hypothesis is that
the “after” distribution lies to the right of the “before” distribution. Also, .025 and n 7.
Note that for one pair of data, both values are the same, 75. We drop such cases when deter-
mining the sample size for the test. From Table IX, the critical value of T is 2. Thus, our
decision rule will be: Reject H0 if the observed value of T is less than or equal to the critical
value of T, which is 2. Note that in the Wilcoxon signed-rank test, the null hypothesis is rejected
if the observed value of T is less than or equal to the critical value of T. This rule is true for
a two-tailed, a right-tailed, or a left-tailed test. The observed value of T is calculated differ-
ently, depending on whether the test is two-tailed or one-tailed. This is explained in the next
step. Figure 15.7 shows the rejection and nonrejection regions.
0, 1, or 2 3 or higher Figure 15.7

T
Rejection region Nonrejection region
Decision Rule For the Wilcoxon signed-rank test for small samples (n 15), the critical value of
T is obtained from Table IX. Note that in the Wilcoxon signed-rank test, the decision rule is to reject
the null hypothesis if the observed value of T is less than or equal to the critical value of T. This rule
is true for a two-tailed, a right-tailed, or a left-tailed test.

The observed value of the test statistic, T, is calculated as follows. The given data on writ-
ing speeds before and after the course are reproduced in the first two columns of Table 15.4.
Table 15.4
Differences Absolute Ranks of Signed

Before After (Before After) Differences Differences Ranks
84 97 13 13 4.5 4.5
75 72 3 3 2 2
88 93 5 5 3 3
91 110 19 19 6 6
65 78 13 13 4.5 4.5
71 69 2 2 1 1
90 115 25 25 7 7
75 75 0 0 — —
1. We obtain the differences column by subtracting each data value after the course from
the corresponding data value before the course. Thus,
Difference Writing speed before the course Writing speed after the course
These differences are listed in the third column of Table 15.4.
2. In the fourth column, we write the absolute values of differences. In other words, the
numbers in the fourth column of the table are the same as those in the third column, but
without plus and minus signs.
3. Next, we rank the absolute differences listed in the fourth column from lowest to highest.
These ranks are listed in the fifth column. Note that the difference of zero is not ranked and
is dropped from the sample. Among the remaining absolute differences, the smallest dif-
ference is 2, which is assigned a rank of 1. The next smallest absolute difference is 3, which
is assigned a rank of 2. Next, the absolute difference of 5 is given a rank of 3. Then, two
absolute differences have the same value, which is 13. We assign the average of the next
two ranks, (4 5)2 4.5, to these two values. Thus, as a rule, whenever some of the ab-
solute differences have the same value, they are all assigned the average of their ranks.
4. In the last column of Table 15.4, we write the ranks of the fifth column with the signs
of the corresponding paired differences. For example, the first difference of 13 has a
minus sign in the third column. Consequently, we assign a minus sign to its rank of 4.5
in the sixth column. The second difference of 3 has a positive sign. Hence, its rank of
2 is assigned a positive sign.
5. Next, we add all the positive ranks and we add the absolute values of the negative ranks
separately. Thus, we obtain:
Sum of the positive ranks 2 1 3
Sum of the absolute values of the negative ranks 4.5 3 6 4.5 7 25
The observed value of the test statistic is determined as shown in the next box.
Observed Value of the Test Statistic T

I. If the test is two-tailed with the alternative hypothesis that the two distributions are not the
same, then the observed value of T is given by the smaller of the two sums, the sum of the
positive ranks and the sum of the absolute values of the negative ranks. We will reject H0 if
the observed value of T is less than or equal to the critical value of T.
II. If the test is right-tailed with the alternative hypothesis that the distribution of after values
is to the right of the distribution of before values, then the observed value of T is given by
the sum of the values of the positive ranks. We will reject H0 if the observed value of T is
less than or equal to the critical value of T.
III. If the test is left-tailed with the alternative hypothesis that the distribution of after values is
to the left of the distribution of before values, then the observed value of T is given by the
sum of the absolute values of the negative ranks. We will reject H0 if the observed value of
T is less than or equal to the critical value of T.
Remember, for the above to be true, the paired difference is defined as the before value minus
the after value. In other words, the differences are obtained by subtracting the after values from
the before values.
Our example is a right-tailed test. Hence,

Observed value of T sum of the positive ranks 3
Whether the test is two-tailed, left-tailed, or right-tailed, we will reject the null hypothesis if:
Observed value of T Critical value of T
where the observed value of T is calculated as explained in Step 4. In this example, the ob-
served value of T is 3 and the critical value of T is 2. Because the observed value of T is greater
than the critical value of T, we do not reject H0. Hence, we conclude that the crash course
does not seem to increase the writing speed of secretaries.

If n 15, we can use the normal distribution to make a test of hypothesis about the paired dif-
ferences. Example 15–8 illustrates the procedure for making such a test.
EXAMPLE 15–8
The manufacturer of a gasoline additive claims that the use of its additive increases gasoline
Performing the Wilcoxon
mileage. A random sample of 25 cars was selected, and these cars were driven for one week
signed-rank test for paired pop-
without the gasoline additive and then for one week with the additive. Then, the miles per ulations: large samples.
gallon (mpg) were estimated for these cars without and with the additive. Next, the paired
differences were calculated for these 25 cars, where a paired difference is defined as
Paired difference mpg without additive mpg with additive
The differences were positive for 4 cars, negative for 19 cars, and zero for 2 cars. First, the
absolute values of the paired differences were ranked, and then these ranks were assigned the
signs of the corresponding paired differences. The sum of the ranks of the positive paired dif-
ferences was 58, and the sum of the absolute values of the ranks of the negative paired dif-
ferences was 218. Can you conclude that the use of the additive increases gasoline mileage?
Use the 1% significance level.
Solution We perform the five steps to conduct this test of hypothesis.

We are to test whether or not the gasoline additive increases gasoline mileage. This will be
true if the distribution of gasoline mileages with the additive lies to the right of the distribution
of gasoline mileage without the additive. The median mileage with the additive will be higher
than the median mileage without the additive. Let MA and MB be the median mileage after
(with) and before (without) the gasoline additive. Then, the null and the alternative hypothe-
ses can be written as follows:
H0: MA MB
H1: MA 7 MB
We are given information about the sums of positive and negative ranks. The sample size
is greater than 15. We use the Wilcoxon signed-rank test procedure with the normal distribu-
tion approximation.
We are using the normal distribution as an approximation to make this test. Hence, we will
find the critical value of z from Table IV in Appendix C. The test is right-tailed. The significance
level is .01, which gives the area to the left of critical point as 1 .01 .9900. Therefore, the
critical value of z is 2.33. The rejection and nonrejection regions are shown in Figure 15.8.
Figure 15.8
α = .01
z
0 Rejection
Nonrejection region region
2.33

Because the sample size is larger than 15, the test statistic T follows (an approximate)
normal distribution.
Observed Value of z In a Wilcoxon signed-rank test for two dependent samples, when the sam-
ple size is large (n 15), the observed value of z for the test statistic T is calculated as
T mT
z
sT
n1n 12 n1n 1212n 12
where mT and sT
4 B 24
The value of T that is used to calculate the value of z is determined based on the alternative
hypothesis, as explained next.
1. If the test is two-tailed with the alternative hypothesis that the two distributions are not the
same, then the value of T may be equal to either of the two sums, the sum of the positive ranks
or the sum of the absolute values of the negative ranks. We will reject H0 if the observed value
of z falls in either of the rejection regions.
2. If the test is right-tailed with the alternative hypothesis that the distribution of after values is
to the right of the distribution of before values, then the value of T is equal to the sum of the
absolute values of the negative ranks. We will reject H0 if the observed value of z is greater
than or equal to the critical value of z.
3. If the test is left-tailed with the alternative hypothesis that the distribution of after values is to
the left of the distribution of before values, then the value of T is equal to the sum of the
absolute values of the negative ranks. We will reject H0 if the observed value of z is less than
or equal to the critical value of z.
Remember, for the above to be true, the paired difference is defined as the before value minus
the after value. In other words, the differences are obtained by subtracting the after values from
the before values. Also, note that whether the test is right-tailed or left-tailed, the value of T in
both cases is equal to the sum of the absolute values of the negative ranks.
Using the given information, we calculate the values of T and T and the observed value
of z as follows. Note that here n 23 because two of the paired differences are zero.
n1n 12 23123 12
mT 138
4 4
n1n 1212n 12 23123 12146 12
sT 32.87856445
B 24 B 24
T mT 218 138
z 2.43
sT 32.87856445
The observed value of z 2.43 falls in the rejection region. Hence, we reject the null hy-
pothesis and conclude that the gasoline additive increases mileage.
EXERCISES
15.28 When would you use the Wilcoxon signed-rank test procedure instead of the paired-samples test of
Chapter 10?
15.29 Explain how the null hypothesis is usually stated in the Wilcoxon signed-rank test.
15.30 How are ranks assigned to two or more absolute differences that have the same value in the Wilcoxon
signed-rank test?
15.31 Determine the rejection region for the Wilcoxon signed-rank test for each of the following. Indi-
cate whether the rejection region is based on T or z.
a. n 10, H0: MA MB, H1: MA MB, .05
b. n 12, H0: MA MB, H1: MA MB, .01
c. n 20, H0: MA MB, H1: MA MB, .025
d. n 30, H0: MA MB, H1: MA MB, .01
15.32 In each case, perform the Wilcoxon signed-rank test.
a. n 8, T 5, left-tailed test using .05
b. n 15, T 20, right-tailed test using .01
c. n 25, T 51, two-tailed test using .02
d. n 36, T 238, left-tailed test using .01
APPLICATIONS
15.33 Refer to Exercise 10.96 of Chapter 10, which deals with Gamma Corporation’s installation of gov-
ernors on its salespersons’ cars to regulate their speeds. The following table gives the number of contacts
made by each of seven randomly selected sales representatives during the week before governors were
installed and the number of contacts made during the week after installation.
Salesperson A B C D E F G
Before 50 63 42 55 44 65 66
After 49 60 47 51 50 60 58
a. Using the Wilcoxon signed-rank test at the 5% level of significance, can you conclude that the
use of governors tends to reduce the number of contacts made per week by Gamma Corporation’s
sales representatives?
b. Compare your conclusions of part a with the result of the hypothesis test that was performed
(using the t distribution) in Exercise 10.96.
15.34 Refer to Exercise 10.96 of Chapter 10. The following table gives the gas mileage (in miles per gal-
lon) for each of seven randomly selected sales representatives’ cars during the week before governors were
installed and the gas mileage in the week after installation.
Before 25 21 27 23 19 18 20
After 26 24 26 25 24 22 23
use of governors tends to increase the median gas mileage for Gamma Corporation’s sales rep-
resentatives?
b. Compare your conclusion of part a with the result of the hypothesis test that was performed
(using the t distribution) in Exercise 10.96.
15.35 Refer to Exercise 15.23. The following table shows the self-confidence test scores of seven
employees before and after they attended a course designed to build self-confidence.
Before 8 5 4 9 6 9 5
After 10 8 5 11 6 7 9
a. Using the Wilcoxon signed-rank test at the 5% level of significance, can you conclude that
attending this course increases the median self-confidence test score of employees?
b. Compare your conclusion of part a with the result of Exercise 15.23.
15.36 Refer to Exercise 15.25, which compares productivity of 27 bicycle assemblers under an hourly
payment system and under a modified-piecework payment scheme. The paired difference for each assembler
was calculated by subtracting the number of bicycles assembled during the first week under the new pay-
ment system from the number of bicycles assembled during the last week under the hourly wage system.
These paired differences are positive for 7 assemblers, negative for 19, and zero for 1 assembler. The sum
of the ranks of the positive paired differences is 61, and the sum of the absolute values of the ranks of the
negative paired differences is 290.
a. Using the Wilcoxon signed-rank test at the 2% level of significance, can you conclude that the me-
dian number of bicycles assembled by all such assemblers is the same under both payment systems?
b. Compare your conclusion of part a with the result of the sign test performed in Exercise 15.25.
15.37 Twenty randomly selected adults who describe themselves as “couch potatoes” were given a six-
week course in physical fitness. Before starting the course, each adult took a two-mile hike on the
same trail. The time required to complete the hike was recorded for each adult. After finishing the
course, they all took the same hike again, and their times were recorded again. The following table
lists the times recorded (in minutes) before and after the course for each of the 20 adults.
Before After Before After Before After

41 37 64 55 100 78
91 71 37 31 48 40
35 30 54 57 50 48
58 64.5 70 59 94 102
45 44 40 33 42.5 40
48.5 44 78 70.5 75 63
84 78 66 56
Does the fitness course appear to reduce the time required to complete the two-mile hike? Use the Wilcoxon
signed-rank test at the 2.5% level of significance.
15.38 Many adults in the United States have accumulated excessive balances on their credit cards. Ninety
such adults were randomly chosen to participate in a group therapy program designed to reduce their debts.
Each adult’s total credit card balance was recorded twice: before the program began and three months af-
ter the program ended. The paired difference was then calculated for each adult by subtracting the balance
after the program ended from the balance before the program began. These paired differences were posi-
tive for 49 adults and negative for 41 adults. The sum of the ranks of the positive paired differences was
2507, and the sum of the absolute values of the ranks of the negative paired differences was 1588. Can
you conclude that this group therapy program reduces credit card debt? Use the 5% level of significance.
15.3 The Wilcoxon Rank Sum Test for

Two Independent Samples
In Chapter 10 we discussed tests of hypotheses about the difference between the means of
two independent populations using the normal and t distributions. In this section we compare
two independent populations using the results obtained from samples drawn from these pop-
ulations. In a Wilcoxon rank sum test, we assume that the two populations have identical
shapes but differ only in location, which is measured by the median. Note that identical shapes
do not mean that they have to have a normal distribution. To apply this test, we must be
able to rank the given data. Note that the Wilcoxon rank sum test is almost identical to the
Mann-Whitney test.
In the hypothesis tests discussed in this section, the null hypothesis is usually that the two
population distributions are identical. The alternative hypothesis can be that the two population
distributions are not identical or that one distribution is to the right of the other or that one dis-
tribution is to the left of the other. Assuming that the null hypothesis is true and that the two
15.3 The Wilcoxon Rank Sum Test for Two Independent Samples 584
populations are identical, we rank all the (combined) data values of the two samples as if they
were drawn from the same population. Any tied data values are assigned the ranks in the same
manner as in the preceding section. Then we sum the ranks for the data values of each sample
separately. If the two populations are identical, the ranks should be spread randomly (and evenly)
between the two samples. In this case, the sums of the ranks for the two samples should be
almost equal, given that the sizes of the two samples are almost the same. However, if one of
the two samples contains mostly lower ranks and the other contains mostly higher ranks, then
the sums of the ranks for the two samples will be quite different. The larger the difference in
the sums of the ranks of the two samples, the more convincing is the evidence that the two pop-
ulations are not identical and that the null hypothesis is not true.
In this section, we discuss the Wilcoxon rank sum test first for small samples and then for
large samples.

If the sizes of both samples are 10 or less, we use the Wilcoxon rank sum test for small sam-
ples. Example 15–9 illustrates how the test is performed. To make this test, the population that
corresponds to the smaller sample is labeled population 1 and the one that corresponds to the
larger sample is called population 2. The respective samples are sample 1 and sample 2. If the
sizes of the two samples are equal, either of the two populations can be labeled population 1.
EXAMPLE 15–9
A researcher wants to determine whether the distributions of daily crimes in two cities are
Performing the Wilcoxon rank
identical. The following data give the numbers of violent crimes on eight randomly selected
sum test for two independent
days for City A and on nine days for City B. populations: small samples.
City A 12 21 16 8 26 13 19 23
City B 18 25 14 16 23 19 28 20 31
Using the 5% significance level, can you conclude that the distributions of daily crimes in the
two cities are different?
Solution We apply the following five steps to perform the hypothesis test.

We are to test if the two populations are identical or different. Hence, the two hypotheses
are as follows:
H0: The population distributions of daily crimes in the two cities are identical
H1: The population distributions of daily crimes in the two cities are different
Let the distribution of daily crimes in City A be called population 1 (note that it corre-
sponds to the smaller sample) and that in City B be called population 2. The respective sam-
ples are called sample 1 and sample 2. Because n1 10 and n2 10, we use the Wilcoxon
rank sum test for small samples.
The test statistic in Wilcoxon’s rank sum test is denoted by T. The critical value or val-
ues of T are obtained from Table X that appears at the end of this chapter. In this table, TL
gives the lower critical value and TU gives the upper critical value. If the test is two-tailed,
we use both TL and TU. For a left-tailed test we use TL only, and for a right-tailed test we
use TU only.
In our example, the test is two-tailed. Also, .05, n1 8, and n2 9. Hence, from
Table X, the values of TL and TU are 51 and 93, respectively. We will reject the null hy-
pothesis if the observed value of T is either less than or equal to TL or greater than or equal
to TU. The rejection and nonrejection regions are shown in Figure 15.9. Thus, the decision
rule is that we will reject H0 if either the observed value of T 51 or the observed value
of T
93.
Figure 15.9 51 or lower 52 to 92 93 or higher

Table 15.5
City A City B
Crimes Rank Crimes Rank

12 2 18 7
21 11 25 14
16 5.5 14 4
8 1 16 5.5
26 15 23 12.5
13 3 19 8.5
19 8.5 28 16
23 12.5 20 10
31 17
Sum 58.5 Sum 94.5
To find the observed value of T, first we rank all the data values of both samples as if they
belonged to the same population. Then, we find the sum of the ranks for each sample sepa-
rately. The observed value of the test statistic T is given by the sum of the ranks for the smaller
sample. If the sizes of the samples are the same, we can use either of the rank sums as the ob-
served value of T.
In Table 15.5, we rank all the values of both samples and find the sum of the ranks for each
sample. Note that 8 is the smallest data value in both samples. Hence, it is assigned a rank of 1.
The next smallest value in both samples is 12, which is assigned a rank of 2. The remaining
values are assigned ranks in the same way.
Because n1 8 and n2 9, the sample size for City A is smaller. Hence, the observed
value of T is given by the sum of the ranks for city A. Thus,
Observed value of T 58.5

Comparing the observed value of T with TL and TU (obtained from Table X in Step 3), we
see that the observed value of T 58.5 is between TL 51 and TU 93. Hence, we do not
reject H0, and we conclude that the two population distributions seem to be identical.
Below we describe the Wilcoxon rank sum test procedure for small samples for two-tailed,
right-tailed, and left-tailed tests.
Wilcoxon Rank Sum Test for Small Independent Samples

1. A two-tailed test: The null hypothesis is that the two population distributions are identical,
and the alternative hypothesis is that the two population distributions are different. The
critical values of T, TL, and TU for this test are obtained from Table X for the given
significance level and sample sizes. The observed value of T is given by the sum of the
ranks for the smaller sample. The null hypothesis is rejected if T TL or T TU. Otherwise,
the null hypothesis is not rejected.
Note that if the two sample sizes are equal, the observed value of T is given by the
sum of the ranks for either sample.
2. A right-tailed test: The null hypothesis is that the two population distributions are identical,
and the alternative hypothesis is that the distribution of population 1 (the population that
corresponds to the smaller sample) lies to the right of the distribution of population 2. The
critical value of T is given by TU in Table X for the given for a one-tailed test and the
given sample sizes. The observed value of T is given by the sum of the ranks for the smaller
sample. The null hypothesis is rejected if T TU. Otherwise, the null hypothesis is not
rejected.
sum of the ranks for sample 1.
3. A left-tailed test: The null hypothesis is that the two population distributions are identical,
and the alternative hypothesis is that the distribution of population 1 (the population that
corresponds to the smaller sample) lies to the left of the distribution of population 2. The
critical value of T in this case is given by TL in Table X for the given for a one-tailed test
and the given sample sizes. The observed value of T is given by the sum of the ranks for
the smaller sample. The null hypothesis is rejected if T TL. Otherwise the null hypothesis
is not rejected.
sum of the ranks for sample 1.

If either n1 or n2 or both n1 and n2 are greater than 10, we use the normal distribution as an
approximation to the Wilcoxon rank sum test for two independent samples.
Observed Value of z In the case of a large sample, the observed value of z is calculated as
T mT
z
sT
Here, the sampling distribution of the test statistic T is approximately normal with mean T and
standard deviation T. The values of T and T are calculated as
n1 1n1 n2 12 n1n2 1n1 n2 12
mT and sT
2 B 12
Note that in these calculations sample 1 refers to the smaller sample and sample 2 to the larger
sample. However, if the two samples are of the same size, we can label either one sample 1. The
value of T used in the calculation of z is given by the sum of the ranks for sample 1.
The critical value or values of z are obtained from Table IV in Appendix C for the given
significance level. We will reject the null hypothesis if the observed value of z is in the rejec-
tion region. Otherwise, we will not reject H0. Example 15–10 illustrates the procedure for per-
forming such a test.
EXAMPLE 15–10
A researcher wanted to find out whether job-related stress is lower for college and univer-
Performing the Wilcoxon rank
sity professors than for physicians. She took random samples of 14 professors and 11 physi-
sum test for two independent
cians and tested them for job-related stress. The following data give the stress levels for
populations: large samples.
professors and physicians on a scale of 1 to 20, where 1 is the lowest level of stress and
20 is the highest.
Professors 5 9 4 12 6 15 2 8 10 4 6 11 8 3
Physicians 10 18 12 5 13 18 14 9 6 16 11
Using the 1% significance level, can you conclude that the job-related stress level for profes-
sors is lower than that for physicians?
Solution Because the smaller sample should be labeled sample 1, the sample of 11 phy-
sicians will be called sample 1 and that of 14 professors will be called sample 2. The respective
populations are populations 1 and 2. Thus, n1 11 and n2 14. We perform the five steps
of the hypothesis test.
We are to test whether or not professors have lower job-related stress than physicians. Be-
cause physicians are labeled population 1 and professors population 2, professors will have a
lower stress level if the distribution of population 1 is to the right of the distribution of pop-
ulation 2. Thus, we can state the two hypotheses as follows.
H0: The two population distributions are identical
H1: The distribution of population 1 is to the right of the distribution of population 2

Because n1 10 and n2 10, we use the normal distribution to make this test as the test
statistic T follows an approximately normal distribution.
The test is right-tailed and .01. The area to the left of the critical point under the
normal distribution curve is 1 .01 .9900. From Table IV in Appendix C, the critical
value of z for .9900 is 2.33. The rejection and nonrejection regions are shown in Figure 15.10.
Thus, we will reject H0 if the observed value of z is 2.33 or greater. Otherwise, we will not
reject H0.
Figure 15.10
α = .01
z
0 Rejection
Nonrejection region region
2.33

Table 15.6 shows the rankings of all the data values for the two samples and the sums of
these ranks for each sample separately.
Table 15.6
Physicians Professors
Stress Level Rank Stress Level Rank

10 14.5 5 5.5
18 24.5 9 12.5
12 18.5 4 3.5
5 5.5 12 18.5
13 20 6 8
18 24.5 15 22
14 21 2 1
9 12.5 8 10.5
6 8 10 14.5
16 23 4 3.5
11 16.5 6 8
11 16.5
8 10.5
3 2
Sum 188.5 Sum 136.5
Hence, we calculate the value of the test statistic as follows:

n1 1n1 n2 12 11111 14 12
mT 143
2 2
n1n2 1n1 n2 12 111142111 14 12
sT 18.26654501
B 12 B 12
T mT 188.5 143
z 2.49
sT 18.26654501
Thus, the observed value of z is 2.49. Note that in the calculation of z, we used the value of
T that belongs to sample 1, which should always be the case.
Because the observed value of z 2.49 is greater than the critical value of z 2.33, it
falls in the rejection region. Hence, we reject H0 and conclude that the distribution of popu-
lation 1 is to the right of the distribution of population 2. Thus, the job-related stress level of
physicians is higher than that of professors. This can also be stated as the job-related stress
level of professors is lower than that of physicians.
Below we describe the Wilcoxon rank sum test procedure for large samples for two-tailed,
right-tailed, and left-tailed tests.
Wilcoxon Rank Sum Test for Large Independent Samples When n1 10 or n2 10 (or both
samples are greater than 10), the distribution of T (the sum of the ranks of the smaller of the two
samples) is approximately normal with mean and standard deviation as follows:
n1 1n1 n2 12 n1n2 1n1 n2 12
mT and sT
2 B 12
For two-tailed, right-tailed, and left-tailed tests, first calculate T, T, T, and the value of the
test statistic, z (T T)T. If n1 n2, T can be calculated from either sample 1 or sample 2.
1. A two-tailed test: The null hypothesis is that the two population distributions are
identical, and the alternative hypothesis is that the two population distributions are
different. At significance level , the critical values of z are obtained from Table IV in
Appendix C. The null hypothesis is rejected if the observed value of z falls in the
rejection region.
2. A right-tailed test: The null hypothesis is that the two population distributions are
identical, and the alternative hypothesis is that the distribution of population 1 (the
population with the smaller sample size) lies to the right of the distribution of popu-
lation 2. At significance level , the critical value of z is obtained from Table IV in
Appendix C. The null hypothesis is rejected if the observed value of z falls in the rejection
region.
3. A left-tailed test: The null hypothesis is that the two population distributions are identical,
and the alternative hypothesis is that the distribution of population 1 (the population with
the smaller sample size) lies to the left of the distribution of population 2. At significance
level , the critical value of z is found from Table IV in Appendix C. The null hypothesis
is rejected if the observed value of z falls in the rejection region.
EXERCISES
15.39 Explain what determines whether to use the Wilcoxon signed-rank test or the Wilcoxon rank sum
test.
15.40 Find the rejection region for the Wilcoxon rank sum test in each of the following cases.
a. n1 7, n2 8, right-tailed test using .05
b. n1 10, n2 10, two-tailed test using .10
c. n1 18, n2 20, left-tailed test using .05
d. n1 25, n2 25, two-tailed test using .01
15.41 In each of the following cases, perform the Wilcoxon rank sum test.
a. n1 6, n2 7, T 22, two-tailed test with .05
b. n1 10, n2 12, T 137, right-tailed test with .025
c. n1 9, n2 11, T 68, left-tailed test with .05
d. n1 22, n2 23, T 638, two-tailed test with .01
APPLICATIONS
15.42 A consumer agency wants to compare the caffeine content of two brands of coffee. Eight jars
of each brand are analyzed, and the amount of caffeine found in each jar is recorded as shown in the
table.
Brand I 82 77 85 73 84 79 81 82
Brand II 75 80 76 81 72 74 73 78
Using .10, can you conclude that the two brands have different median caffeine contents per jar?
15.43 In a Winter Olympics trial for women’s speed skating, seven skaters use a new type of skate, while
eight others use the traditional type. Each skater is timed (in seconds) in the 500-meter event. The results
are given in the following table.
New skates 40.5 40.3 39.5 39.7 40.0 39.9 41.5

Traditional skates 41.0 40.8 40.9 39.8 40.6 40.7 41.1 40.5
15.4 The Kruskal-Wallis Test 590
Assuming that these 15 skaters make up a random sample of all Olympic-class 500-meter female speed
skaters, can you conclude that the new skates tend to produce faster times in this event? Use the 5% level
of significance.
15.44 During April–June 2004, the median price of homes sold in Phoenix was $252,400, and the me-
dian price in Las Vegas was $255,800. The following table gives the prices (in thousands of dollars) of 9
randomly selected homes in Phoenix and 10 homes in Las Vegas that were sold recently.
Phoenix 258 269 229 279 249 260 242 240 307
Las Vegas 280 245 319 289 259 268 295 239 262 250
Using the 5% level of significance, can you conclude that the current median price of homes in Phoenix
is different from the current median price of homes in Las Vegas?
15.45 A factory’s management is concerned about the number of defective parts produced by its machinists.
Management suspects that production may be improved by giving machinists frequent breaks to reduce
fatigue. Twenty-four randomly chosen machinists are randomly divided into two groups (A and B) of 12
each. During the next week all 24 machinists work to manufacture similar parts. The workers in Group A
get a five-minute break every hour, whereas the workers in Group B stay on the usual schedule. The num-
ber of good parts produced by each machinist during the week is recorded in the following table.
Group A 157 139 188 143 172 144 191 128 177 160 175 162
Group B 160 118 150 165 158 159 127 133 170 164 152 142
At the 1% level of significance, can you conclude that the median number of good parts produced by ma-
chinists who take a five-minute break every hour is higher than the median number of good parts produced
by the machinists who do not take a break?
15.46 Two brands of tires are tested to compare their durability. Eleven Brand X tires and 12 Brand Y
tires are tested on a machine that simulates road conditions. The mileages (in thousands of miles for each
tire) are shown in the following table.
Brand X 51 55 53 49 50.5 57 54.5 48.5 51.5 52 53.5

Brand Y 48 47 54 55.5 50 51 46 49.5 52.5 51 49 45
Using the 5% level of significance, can you conclude that the median mileage for Brand X tires is greater
than the median mileage for Brand Y tires?
15.47 Two Midwestern towns that are 120 miles apart are served by an airline that has been plagued
by delays in the past few months. Consequently, many passengers who formerly traveled by air between
these towns are taking advantage of a new express bus service. Some statistics students at a local col-
lege conducted a survey to see if the bus service between these towns was faster than the air flight. The
students took random samples of 15 (one-way) plane trips and 17 (one-way) bus trips between the towns,
recording the times for all 32 trips. The time recorded for each trip was measured from the scheduled
departure time to the actual arrival time. The sum of the ranks for the 15 plane trips was 295; the sum
of the ranks for the 17 bus trips was 233. At the 5% level of significance, can you conclude that the
median time for the plane trip is higher than the median time for the bus trip?
15.4 The Kruskal-Wallis Test

In Chapter 12 we used the one-way analysis of variance (ANOVA) procedure to test whether
or not the means of three or more populations are all equal. To apply the ANOVA procedure
using the F distribution, we assumed that the populations from which the samples were drawn
were normally distributed with equal variance, 2. However, if the populations being sampled
are not normally distributed, then we cannot apply the ANOVA procedure of Chapter 12. In
such cases, we can use the Kruskal-Wallis test, also called the Kruskal-Wallis H test. This
is a nonparametric test because to use it we do not make any assumptions about the
distributions of the populations being sampled. The only assumption we make is that all
populations under consideration have identical shapes but differ only in location, which is
measured by the median. Note that identical shapes do not mean that they have to have a normal
distribution.
In a Kruskal-Wallis test, the null hypothesis is that the population distributions under con-
sideration are all identical. The alternative hypothesis is that at least one of the population dis-
tributions differs and that, therefore, not all of the population distributions are identical. Note
that we use the Kruskal-Wallis test to compare three or more populations. Also note that to
apply the Kruskal-Wallis test, the size of each sample must be at least five.
Kruskal-Wallis Test To perform the Kruskal-Wallis test, we use the chi-square distribution that
was discussed in Chapter 11. The test statistic in this test is denoted by H, which follows
(approximately) the chi-square distribution. The critical value of H is obtained from Table VI in
Appendix C for the given level of significance and df k 1, where k is the number of popu-
lations under consideration. Note that the Kruskal-Wallis test is always right-tailed.
To find the observed value of the test statistic H, we first rank the combined data from
all samples in the same way as in a Wilcoxon rank sum test. The tied data values are handled
the same way as in a Wilcoxon test. Then the observed value of H is calculated as explained
below.
Observed Value of the Test Statistic H The observed value of the test statistic H is calculated
using the following formula:
R21 R22 R2k
a # # # b 31n 12
12
H
n1n 12 n1 n2 nk
where R1 sum of the ranks for sample 1
R2 sum of the ranks for sample 2
..
.
Rk sum of the ranks for sample k
n1 sample size for sample 1
n2 sample size for sample 2
..
.
nk sample size for sample k
n n1 n2 # # # nk
k number of samples
The test statistic H measures the extent to which the k samples differ with regard to the ranks
assigned to their data values. Basically, H is a measure of the variance of ranks (or of the variance
of the means of ranks) for different samples. If all k samples have exactly the same mean of
ranks, H will have a value of zero. The value of H becomes larger as the difference between
the means of ranks for different samples increases. Thus, a larger observed value of H indicates
that the distributions of the given populations do not seem to be identical.
Example 15–11 illustrates the procedure for applying the Kruskal-Wallis test.
EXAMPLE 15–11
A researcher wanted to find out whether the population distributions of salaries of computer
Performing the
programmers are identical in three cities, Boston, San Francisco, and Atlanta. Three different
Kruskal-Wallis test.
samples—one from each city—produced the following data on the annual salaries (in thou-
sands of dollars) of computer programmers.
Boston San Francisco Atlanta

43 54 57
39 33 68
62 58 60
73 38 44
51 43 39
46 55 28
34 49
57
Using the 2.5% significance level, can you conclude that the population distributions of salaries
for computer programmers in these three cities are all identical?
Solution We apply the five steps to perform this hypothesis test.

H0: The population distributions of salaries of computer programmers in the
three cities are all identical
H1: The population distributions of salaries of computer programmers in the
three cities are not all identical
Note that the alternative hypothesis states that the population distribution of at least one city
is different from those of the other two cities.
The shapes of the population distributions are unknown. We are comparing three popula-
tions. Hence, we apply the Kruskal-Wallis procedure to perform this test, and we use the
chi-square distribution.
In this example,
a .025 and d f k 1 3 1 2
Hence, from Table VI in Appendix C, the critical value of 2 is 7.378, as shown in
Figure 15.11.
Figure 15.11
df = 2
α = .025
χ2
7.378

To calculate the observed value of the test statistic H, we first rank the combined data for
all three samples and find the sum of ranks for each sample separately. This is done in Table 15.7.
Table 15.7
Boston San Francisco Atlanta
Salary Rank Salary Rank Salary Rank

43 7.5 54 13 57 15.5
39 5.5 33 2 68 20
62 19 58 17 60 18
73 21 38 4 44 9
51 12 43 7.5 39 5.5
46 10 55 14 28 1
34 3 49 11
57 15.5
n1 6 R1 75 n2 7 R2 60.5 n3 8 R3 95.5
We have
n n1 n2 n3 6 7 8 21
and
R21 R22 # # # R2k
a b 31n 12
12
H
n1n 12 n1 n2 nk
1752 2 160.52 2 195.52 2
a b 3121 12
12

21121 12 6 7 8
1.543
Because the observed value of H 1.543 is less than the critical value of H 7.378 and
it falls in the nonrejection region, we do not reject the null hypothesis. Consequently, we con-
clude that the population distributions of salaries of computer programmers in the three cities
seem to be all identical.
EXERCISES
15.48 Briefly explain when the Kruskal-Wallis test is used to make a test of hypothesis.
15.49 What assumption that is required for the ANOVA procedure of Chapter 12 is not necessary for the
Kruskal-Wallis test?
15.50 Describe the form of the null and alternative hypotheses for a Kruskal-Wallis test.
15.51 Here, ni is the size of the ith sample and Ri is the sum of ranks for the ith sample. For each of the
following cases, perform the Kruskal-Wallis test using the 5% level of significance.
a. n1 9, n2 8, n3 5, R1 81, R2 102, R3 70
b. n1 n2 n3 n4 5, R1 27, R2 30, R3 83, R4 70
c. n1 6, n2 10, n3 6, R1 93, R2 70, R3 90
d. n1 8, n2 9, n3 8, n4 10, n5 9,
R1 210, R2 195, R3 178, R4 212, R5 195
15.52 The following table gives the ranked data for three samples. Perform the Kruskal-Wallis test using
the 1% level of significance.
Sample I Sample II Sample III

3 14 2
1 11 4.5
10 16 13
7 15 4.5
9 12 8
6
APPLICATIONS
15.53 Refer to Examples 12–2 and 12–3 of Chapter 12. Fifteen randomly selected fourth-grade students were
randomly assigned to three groups, and each group was taught arithmetic by a different method. At the end
of the semester, all 15 students took the same arithmetic test. Their test scores are given in the following table.
Method I Method II Method III

48 55 84
73 85 68
51 70 95
65 69 74
87 90 67
a. At the 1% level of significance, can you reject the null hypothesis that the median arithmetic
test scores of all fourth-grade students taught by these three methods are all equal?
b. Compare your answer to part a with the result of the hypothesis test in Example 12–3.
15.54 A consumer agency investigated the premiums charged by four auto insurance companies. The
agency randomly selected five drivers insured by each company who had similar driving records, autos,
and insurance coverages. The following table gives the monthly premiums paid by the 20 drivers.
Company A Company B Company C Company D

$65 $48 $57 $62
73 69 61 53
54 88 89 45
43 75 77 51
70 72 69 44
Can you reject the null hypothesis that the distributions of auto insurance premiums paid per month by all
such drivers are the same for all four companies? Use .05.
15.55 Refer to Problem 10 of the Self-Review Test in Chapter 12. A small college town has four pizza
parlors that make deliveries. A student doing a research paper for her business management class
decides to compare how promptly the four parlors deliver. On six randomly chosen nights, she orders
a large pepperoni pizza from each establishment and then records the elapsed time until the pizza is
delivered to her apartment. Assume that her apartment is approximately at the same distance from the
four pizza parlors. The following table shows the delivery times (in minutes) for these orders.
Tony’s Luigi’s Angelo’s Kowalski’s

20.0 22.1 22.3 23.9
24.0 27.0 26.0 24.1
18.3 20.2 24.0 25.8
22.0 32.0 30.1 29.0
20.8 26.0 28.0 25.0
19.0 24.8 25.8 24.2
a. Test the null hypothesis that the distributions of delivery times are identical for the four pizza
parlors. Use the 5% level of significance.
b. Compare your conclusion of part a here with that of part a of Problem 10 of the Self-Review
Test in Chapter 12.
15.56 Refer to Exercise 12.27 of Chapter 12. A resort area has three seafood restaurants, which employ
students during the summer season. The local chamber of commerce took a random sample of five servers
from each restaurant and recorded the tips they received on a recent Friday night. The results of the sur-
vey are shown in the table below. Assume that the Friday night for which the data were collected is typ-
ical of all Friday nights of the summer season.
Barzini’s Hwang’s Jack’s

$ 97 $67 $ 93
114 85 102
105 92 98
85 78 80
120 90 91
a. Would a student seeking a server’s job at one of these three restaurants conclude that the pop-
ulation distributions of tips on Friday nights are identical for the three restaurants? Use the 5%
level of significance.
b. Compare your conclusion of part a with that of part a of Exercise 12.27 of Chapter 12.
c. What would your decision be if the probability of making a Type I error were zero in part a?
Explain.
15.57 A factory operates three shifts a day, five days per week, each with the same number of workers
and approximately the same level of production. The following table gives the number of defective parts
produced during each shift over a period of five days.
First Shift Second Shift Third Shift

23 25 33
36 35 44
32 41 50
40 38 52
45 50 60
At the 5% level of significance, can you conclude that the median number of defective parts is the same
for all three shifts?
15.58 A consumer group wanted to compare the service time at three fast-food restaurants, Al’s, Eduardo’s,
and Patel’s. Every Tuesday and Wednesday for four weeks, three staff members of the group were randomly
assigned to these three restaurants. Each staff member went to his or her assigned restaurant and ordered a
hamburger, fries, and a Coke and then recorded the time that elapsed from entering the restaurant until re-
ceiving the food. The service times (in minutes) for these eight days for the three restaurants are listed below.
Al’s Eduardo’s Patel’s

7.0 3.3 1.1
8.3 11.0 2.4
6.9 5.7 1.8
1.3 8.1 3.0
6.7 6.6 4.1
7.1 13.0 12.0
5.5 2.3 1.5
6.6 5.9 3.1
Assume that these service times make up random samples of all service times at the respective restaurants.
At the 10% level of significance, can you conclude that there is a difference in the median service times
at these three restaurants?
15.5 The Spearman Rho Rank Correlation Coefficient Test 596
15.5 The Spearman Rho Rank Correlation

Coefficient Test
In Chapter 13 we discussed the linear correlation coefficient between two variables x and y. We also
learned how to make a test of hypothesis about the population correlation coefficient using the in-
formation from a sample. In that chapter we used the t distribution to perform this test about .
However, using the procedure of Chapter 13 and using the t distribution to make this test of
hypothesis about require that both variables x and y are normally distributed.
The Spearman rho rank correlation coefficient (Spearman’s rho) is a nonparametric analog
of the linear correlation coefficient of Chapter 13. It helps us decide what type of relationship, if
any, exists between data from populations with unknown distributions. The Spearman rho rank cor-
relation coefficient is denoted by rs for sample data and by s for population data. This correlation
coefficient is simply the linear correlation coefficient between the ranks of the data on variables x
and y. To make a test of hypothesis about the Spearman rho rank correlation coefficient, we do not
need to make any assumptions about the populations of x and y variables.
Spearman Rho Rank Correlation Coefficient The Spearman rho rank correlation coefficient is
denoted by rs for sample data and by s for population data. This correlation coefficient is simply
the linear correlation coefficient between the ranks of the data. To calculate the value of rs, we rank
the data for each variable, x and y, separately and denote those ranks by u and v, respectively. Then
we take the difference between each pair of ranks and denote it by d. Thus,
Difference between each pair of ranks d u v
Next, we square each difference d and add these squared differences to find d 2. Finally, we
calculate the value of rs using the formula:
6d 2
rs 1
n 1 n 2 12
In a test of hypothesis about the Spearman rho rank correlation coefficient s, the test statistic
is rs and its observed value is calculated by using the above formula.
Example 15–12 shows how to calculate the Spearman rho rank correlation coefficient rs
and how to perform a test of hypothesis about s.
EXAMPLE 15–12
Suppose we want to investigate the relationship between the per capita income (in thousands
Performing the Spearman rho
of dollars) and the infant mortality rate (in percent) for different states. The following table rank correlation coefficient test.
gives data on these two variables for a random sample of eight states.
Per capita income (x) 29.85 19.0 19.18 31.78 25.22 16.68 23.98 26.33
Infant mortality (y) 8.3 10.1 10.3 7.1 9.9 11.5 8.7 9.8
Based on these data, can you conclude that there is no significant (linear) correlation between
the per capita incomes and the infant mortality rates for all states? Use .05.
Solution We perform the five steps to test the null hypothesis that there is no correlation between
the two variables against the alternative hypothesis that there is a significant correlation.

The null and alternative hypotheses are as follows:
H0: There is no correlation between per capita incomes and infant mortality rates in
all states
H1: There is a correlation between per capita incomes and infant mortality rates in
all states
If we denote the Spearman correlation coefficient by s, the null hypothesis and the alterna-
tive hypothesis can be written as
H0: rs 0
H1: rs 0
Note that this is a two-tailed test.
Because the sample is taken from a small population and the variables do not follow a normal
distribution, we use the Spearman rho rank correlation coefficient test procedure to make this test.
The test statistic that is used to make this test is rs, and its critical values are given in Table XI
that appears at the end of this chapter. Note that, for this example,
n 8 and a .05
To read the critical value of rs from Table XI, we locate 8 in the column labeled n and .05 in
the top row of the table for a two-tailed test. The critical values of rs are .738, or .738 and
.738. Thus, we will reject the null hypothesis if the observed value of rs is either .738 or
less, or .738 or greater. The rejection and nonrejection regions for this example are shown
in Figure 15.12.
Figure 15.12
rs
–.738 +.738
Critical Value of rs The critical value of rs is obtained from Table XI for the given sample size
and significance level. If the test is two-tailed, we use two critical values, one negative and one
positive. However, we use only the negative value of rs if the test is left-tailed, and only the pos-
itive value of rs if the test is right-tailed.

In the Spearman rho rank correlation coefficient test, the test statistic is denoted by rs, which
is simply the linear correlation coefficient between the ranks of the data. As explained in the be-
ginning of this section, to calculate the observed value of rs, we use the formula:
6d 2
rs 1
n1n2 12
where d u v, and u and v are the ranks of variables x and y, respectively.
Table 15.8
u 7 2 3 8 5 1 4 6
v 2 6 7 1 5 8 3 4
d 5 4 4 7 0 7 1 2
d2 25 16 16 49 0 49 1 4 d 2 160
Table 15.8 shows the ranks for x and y, which are denoted by u and v, respectively. The
table also lists the values of d, d 2, and d 2. Note that if two or more values are equal, we use
the average of their ranks for all of them. Hence, the observed value of rs is
611602 960
rs 1 1 .905
8164 12 504
15.5 The Spearman Rho Rank Correlation Coefficient Test 598
Note that Spearman’s rho rank correlation coefficient has the same properties as the linear
correlation coefficient (discussed in Chapter 13). Thus, 1 rs 1 or 1 s 1, depending
on whether sample or population data are used to calculate the Spearman rho rank correlation
coefficient. If s 0, there is no relationship between the x and y data. If 0 s 1, on average
a larger value of x is associated with a larger value of y. Similarly, if 1 s 0, on average a
larger value of x is associated with a smaller value of y.
Because rs .905 is less than .738 and it falls in the rejection region, we reject H0 and
conclude that there is a correlation between the per capita incomes and the infant mortality
rates in all states. Because the value of rs from the sample is negative, we can also state that
as per capita income increases, infant mortality tends to decrease.
Decision Rule for the Spearman Rho Rank Correlation Coefficient The null hypothesis is always
H0: s 0. The observed value of the test statistic is always the value of rs computed from the
sample data. Let denote the significance level, and c and c be the critical values for the
Spearman rho rank correlation coefficient test obtained from Table XI.
1. For a two-tailed test, the alternative hypothesis is H1: s 0. If c are the critical values
corresponding to sample size n and two-tailed , we reject H0 if either rs c or rs
c; that is, reject H0 if rs is “too small” or “too large.”

2. For a right-tailed test, the alternative hypothesis is H1: s 0. If c is the critical value
corresponding to sample size n and one-sided , we reject H0 if rs
c; that is, reject H0
if rs is “too large.”
3. For a left-tailed test, the alternative hypothesis is H1: s 0. If c is the critical value
corresponding to sample size n and one-sided , we reject H0 if rs c; that is, reject H0
if rs is “too small.”
EXERCISES
15.59 What assumptions that are required for hypothesis tests about the linear correlation coefficient in
Chapter 13 are not required for testing a hypothesis about the Spearman rho rank correlation coefficient?
15.60 Two sets of paired data on two variables, x and y, have been ranked. In each case, the ranks for x
and y are denoted by u and v, respectively, and are shown in the tables. Calculate the Spearman rho rank
correlation coefficient for each case.
a.
u 2 1 3 4 6 5 7 8
v 8 6 7 4 5 2 1 3
b.
u 1 2 3 4 5 6 7
v 4 2 1 5 3 7 6
15.61 Calculate the Spearman rho rank correlation coefficient for each of the following data sets.
a.
x 5 10 15 20 25 30
y 17 15 12 14 10 9
b.
x 27 15 32 21 16 40 8
y 95 81 102 88 75 120 62
15.62 Perform the indicated hypothesis test in each of the following cases.
a. n 9, rs .575, H0: s 0, H1: s 0, .025
b. n 15, rs .575, H0: s 0, H1: s 0, .005
c. n 20, rs .554, H0: s 0, H1: s 0, .01
d. n 20, rs .554, H0: s 0, H1: s 0, .01
APPLICATIONS
15.63 The following data are a random sample of the heights (in inches) and weights (in pounds) of 10
NBA players selected at random.
Height 84 76 79 79 84 74 83 81 83 75
Weight 240 208 205 215 265 182 225 220 250 190
a. Based on the reasonable assumption that as height increases, weight tends to increase, do you
expect the value of rs to be positive or negative? Why?
b. Compute the value of rs. Does it agree with your expectation of its value in part a?
15.64 Let s be the Spearman rho rank correlation coefficient between heights (in inches) and weights (in
pounds) for the entire population of NBA players listed in Data Set II. Using the value of rs computed
from the sample data in Exercise 15.63, test the null hypothesis H0: s 0 against the alternative hypothesis
H1: s 0 at the significance level of .01.
15.65 In Example 13–1 of Chapter 13, we estimated the regression line for the data given in Table 13.2
on food expenditures (in hundred dollars) and incomes (in hundred dollars). Those data are reproduced
here.
Income (x) 35 49 21 39 15 28 25
Food expenditure (y) 9 15 7 11 5 8 9
The estimated regression line in Example 13–1 had a slope of .2642.

a. Do you expect the Spearman rho rank correlation coefficient for these data to be positive or neg-
ative? Why?
b. Compute rs for these data. Did it come out as expected?
15.66 In Example 13–7 of Chapter 13, the null hypothesis H0: 0 was tested against the alternative
hypothesis H1: 0, where is the linear correlation coefficient for the population. At a significance
level of .01, H0: 0 was rejected there. Hence, it seemed that in fact 0.
a. If s is the Spearman rho rank correlation coefficient for the entire population for food ex-
penditure and income data, and you test H0: s 0 against H1: s 0 at the significance
level of .01, based on the results of Example 13–7, do you expect to reject or accept
H0? Why?
b. Perform the hypothesis test stated in part a.
15.67 The following table shows the combined (math and verbal) SAT scores (denoted by x) and college
grade point averages (denoted by y) on completion of a bachelor’s degree for nine randomly chosen recent
college graduates who had taken the SAT test.
x 1105 990 1040 1215 1405 975 1300 1010 1080

y 3.33 2.62 3.05 3.60 3.85 2.43 3.90 2.40 2.95
a. Find the Spearman rho rank correlation coefficient for this data set.
b. Test H0: s 0 against H1: s 0 using the 5% level of significance.
c. Does your test result indicate a positive relationship between the variables x and y?
15.68 A day-care center operator is concerned about the aggressive behavior of young boys left in her
care. She feels that prolonged watching of television tends to promote aggressive behavior. She selects
seven boys at random and ranks them according to the level of aggressiveness in their behavior, with a
rank of 7 indicating most aggressive and a rank of 1 denoting least aggressive. Then she asks each boy’s
parent(s) to estimate the average number of hours per week the boy spends watching television. The fol-
lowing table shows the data collected on the aggressiveness ranks of these boys and the number of hours
spent watching TV per week.
15.6 The Runs Test for Randomness 600
Aggressiveness rank 1 2 3 4 5 6 7
Weekly TV hours 15 21 28 8 24 32 20
a. Calculate rs for these data.

b. At the 5% level of significance, can you conclude that there is a positive relationship between
aggressiveness and hours spent watching television?
15.6 The Runs Test for Randomness

The runs test for randomness tests the null hypothesis that a sequence of events has occurred
randomly against the alternative hypothesis that this sequence of events has not occurred randomly.

As an example of a runs test, in apartment complexes families with children are often assigned
to units close to one another to lessen the impact of noise on childless tenants. We would like
to determine whether a landlord has randomly assigned units to tenants irrespective of whether
they have children, or has tried to cluster tenants with children. Here what we really mean by
random is independence in the sense that if, say, a childless tenant lives in unit 3, this provides
no additional information about whether the tenants in units 2 and 4 are childless. With non-
randomness, we would have some additional information.
Suppose that a part of the apartment complex consists of a single building with 10 adja-
cent units numbered 1 to 10. We specify the family status of the 10 tenants via a string of 10
letters, using C for “has children” and D for “no children.” One possible string is D D C D C
C D D C C, which would mean that the tenants in units 3, 5, 6, 9, and 10 have children and
those in 1, 2, 4, 7, and 8 do not. (Note that in our example the numbers of tenants with and
without children are equal, but they do not have to be equal.)
Two extreme arrangements that may not be random are C C C C C D D D D D and C D C
D C D C D C D. In the first arrangement, all the tenants with children are adjacent, as are those
without children. In the second case, tenants with and without children alternate perfectly. Note
that in these two examples exactly half of the tenants have children and half do not.
A characteristic of a string of the letters C and D that helps determine the randomness of
the string is called a run. A run is a sequence of the same symbol (in this case the letter C or
D) appearing one or more times. The arrangement C C C C C D D D D D has two runs; the
arrangement C D C D C D C D C D has 10 runs. Intuitively, we know that in the string with
2 runs the arrangement is not random because there are too few runs, whereas in the string with
10 runs the arrangement is not random due to too many runs.
If the number of tenants with children, the number without children, and their arrangement
in the 10 units are all random, then the number of runs in the arrangement, which we denote
by R, will also be random. Thus, R is a statistic with its own sampling distribution. Table XII
(that appears at the end of this chapter) gives the critical values of R for a significance level of
5%—that is, for .05. There are two parameters associated with the distribution of R, n1 and
n2. Here n1 is the number of times the first symbol (in our example the letter C) appears in the
string, and n2 is the number of times the second symbol (in our example the letter D) appears.
Table XII provides critical values of R for values of n1 and n2 up to 15. If either n1 15 or
n2 15, we can apply the normal approximation (discussed later in the section on the large-
sample case) to perform the test. For each pair of n1 and n2, there are two critical values: a
smaller value (denoted by c1) and a larger value (denoted by c2).
Definition
Run A run is a sequence of one or more consecutive occurrences of the same outcome in a
sequence of occurrences in which there are only two outcomes. The number of runs in a
sequence is denoted by R. The value of R obtained for a sequence of outcomes for a sample gives
the observed value of the test statistic for the runs test for randomness.
Suppose we formally set up the following hypotheses:

H0: Tenants with and without children are randomly mixed among the 10 units
H1: These tenants are not randomly mixed
We will reject H0 if either of the following occurs:
R c 1 1 too few runs2 or R
c 2 1 too many runs2
Let’s apply these rules to the hypothetical strings listed earlier to determine whether or not to
reject H0 at a significance level of .05.
1. Let the string of letters be C C C C C D D D D D. Here n1 5, n2 5, and R 2. From
Table XII, c1 2 and c2 10. Since R c1, we reject H0 on the basis that there are too
few runs.
2. Let the string of letters be C D C D C D C D C D. Here n1 5, n2 5, and R 10. From
Table XII, c1 2 and c2 10. Since R
c2, we reject H0 on the basis that there are too
many runs.
3. Let the string of letters be D D C D C C D D C C. Here n1 5, n2 5, and R 6. From
Table XII, c1 2 and c2 10. Because the value of R 6 is between c1 2 and c2
10, we do not reject H0.
Example 15–13 illustrates the application of the runs test for randomness.
EXAMPLE 15–13
A college admissions office is interested in knowing whether applications for admission
Performing the runs test for
arrive randomly with respect to gender. The genders of 25 consecutively arriving applications
randomness: small sample.
were found to arrive in the following order (here M denotes a male applicant and F a female
applicant).
M F M M F F F M F M M M F F F F M M M F F M F M M
Can you conclude that the applications for admission arrive randomly with respect to gender?
Use .05.
Solution We perform the following five steps in this hypothesis test.

H0: Applications arrive in a random order with respect to gender
H1: Applications do not arrive in a random order with respect to gender
Let n1 and n2 be the number of male and female applicants, respectively. Then
n1 13 and n2 12
Because both n1 and n2 are less than 15, we use the runs test to check for randomness.
For n1 13, n2 12, and .05, the critical values from Table XII are c1 8 and c2 19.
Thus, we will not reject the null hypothesis if the observed value of R is in the interval 9 to
18. We will reject the null hypothesis if the observed value of R is either 8 or lower, or 19 or
higher. The rejection and nonrejection regions are shown in Figure 15.13.
Figure 15.13 8 or lower 9 to 18 19 or higher

R

As the given data show, the 25 applications included in the sample were received in the
following order with respect to gender:
M F M M F F F M F M M M F F F F M M M F F M F M M
Because this string of the letters M and F has 13 runs,
Observed value of R 13
Because R 13 is between 9 and 18, we do not reject H0. Hence, we conclude that the
applications for admission arrive in a random order with respect to gender.

If either n1 15 or n2 15, the sample is considered large for the purpose of applying the
runs test for randomness and we use the normal distribution to perform the test.
Observed Value of z For large values of n1 and n2, the distribution of R (the number of runs in
the sample) is approximately normal with its mean and standard deviation given as
2n1n2 2n1n2 12n1n2 n1 n2 2
mR 1 and sR
n1 n2 B 1n1 n2 2 2 1n1 n2 12
The observed value of z for R is calculated using the formula
R mR
z
sR
In this case, rather than using Table XII to find the critical values of R, we use the stan-
dard normal distribution table (Table IV in Appendix C) to find the critical values of z for the
given significance level. Then, we make a decision to reject or not to reject the null hypothesis
based on whether the observed value of z falls in the rejection or the nonrejection region.
Example 15–14 describes the application of this procedure.
EXAMPLE 15–14
Refer to Example 15–13. Suppose that the admissions officer examines 50 consecutive
Performing the runs test for
applications and observes that n1 22, n2 28, and R 20, where n1 is the number of
randomness: large sample.
male applicants, n2 the number of female applicants, and R the number of runs. Can we
conclude that the applications for admission arrive randomly with respect to gender? Use
.01.
Solution We perform the following five steps to make this test.
H0: Applications arrive in a random order with respect to gender
H1: Applications do not arrive in a random order with respect to gender
Here, n1 22 and n2 28. Since n1 and n2 are both greater than 15, we use the normal
distribution to make the runs test. Note that only one of n1 and n2 has to be greater than 15 to
apply the normal distribution.

The significance level is .01 and the test is two-tailed. From Table IV in Appendix C, the
critical values of z for .005 and .9950 areas to the left are 2.58 and 2.58, respectively. The
rejection and nonrejection regions are shown in Figure 15.14.
Figure 15.14
α /2 = .005 α /2 = .005
0 z
Rejection Rejection
region Nonrejection region region
– 2.58 2.58

To find the observed value of z, we first find the mean and standard deviation of R as fol-
lows:
2n1n2 12212221282
mR 1 1 25.64
n1 n2 22 28
2n1n2 12n1n2 n1 n2 2 21222128212 22 28 22 282

sR 3.44783162
B 1n1 n2 2 1n1 n2 12
2
B 122 282 2 122 28 12
The observed value of the test statistic z is
R mR 20 25.64
z 1.64
sR 3.44783162
Note that the value of R is given in the example to be 20.
Since z 1.64 is between 2.58 and 2.58, we do not reject H0, and we conclude that
the applications for admission arrive in a random order with respect to gender.
EXERCISES
15.69 Briefly explain the term run as used in a runs test.
15.70 What is the usual form of the null hypothesis in a runs test for randomness?
15.71 Under what conditions may we use the normal distribution to perform a runs test?
15.72 Using the runs test for randomness, indicate whether the null hypothesis should be rejected in each
of the following cases.
a. n1 10, n2 12, R 17, .05
b. n1 20, n2 23, R 35, .01
c. n1 15, n2 17, R 7, .05
d. n1 14, n2 13, R 21, .05
15.73 In Example 15–13, if we use the symbol 0 for male and 1 for female instead of M and F, would
this affect the test in any way? Why or why not?
15.74 For each of the following sequences of observations, determine the values of n1, n2, and R.
a. X X Y X Y Y X Y X Y X X X Y Y
b. F M F F F F M M F F F F F F
c.
d. 1 1 0 0 0 0 1 1 0 0 1 1 1 1
APPLICATIONS
15.75 A psychic claims that she can cause a nonrandom sequence of heads (H) and tails (T) to appear
when a coin is tossed a number of times. A fair coin was tossed 20 times in her presence, and the fol-
lowing sequence of heads and tails was obtained.
H H T H T H T T H T H H T T H H H T T H
Can you conclude that the psychic’s claim is true? Use .05.
15.76 At a small soda factory, the amount of soda put into each 12-ounce bottle by the bottling ma-
chine varies slightly. The plant manager suspects that the machine has a nonrandom pattern of overfilling
and underfilling the bottles. The following are the results of filling 18 bottles, where O denotes 12 ounces
or more of soda in a bottle and U denotes less than 12 ounces of soda.
U U U O O O O U U O O O U O U U U U
Using the runs test at the 5% significance level, can you conclude that there is a nonrandom pattern of
overfilling and underfilling such bottles?
15.77 An experimental planting of a new variety of pear trees consists of a single row of 20 trees. Several
of these trees were affected by an unknown disease. The order of diseased and normal trees is shown be-
low, where D denotes a diseased and N denotes a normal tree.
N N N D D D N N N N N N D D D D N N N N
If the sequence of diseased and normal trees falls into a nonrandom pattern with clusters of diseased trees,
that would suggest that the disease may be contagious. Perform the runs test at the 5% significance level
to determine if there is evidence of a nonrandom pattern in the sequence.
15.78 A fourth-grade teacher asks her class orally one by one whether “potato” or “potatoe” is the cor-
rect spelling for the common vegetable. She suspects the children may engage in copycat behavior, in
which an incorrect spelling by one student is likely to be followed by an incorrect spelling by the next
student. If this theory is true, there should be fewer runs than expected. The responses of the 22 students
in the class are as follows, where a correct answer is denoted by C and an incorrect answer by I.
C C C I I I C C C C C I I C C C C I C C C C
At the 5% significance level, test the null hypothesis that the correct (and incorrect) responses are ran-
domly distributed in the population of all fourth-graders against the alternative hypothesis that they are
not randomly distributed in the population. Assume that this class is a random sample of all fourth-graders.
15.79 Do baseball players’ hits come in streaks? Seventy-five consecutive at-bats of a baseball player were
recorded to determine whether the hits were randomly distributed or nonrandomly distributed for him, thus
possibly indicating the presence of “hitting streaks.” The observation of these 75 at-bats produced the fol-
lowing data.
n1 number of hits 22, n2 number of nonhits 53, R number of runs 37
Can you conclude that hits occur randomly for this player? Use .01.
15.80 A researcher wanted to determine whether the stock market moves up and down randomly. He
recorded the movement of the Dow Jones Industrial Average for 40 consecutive business days. The
observed data showed that the market moved up 16 times, moved down 24 times, and had 11 runs during
these 40 days. Using a significance level of 5%, do you think that the movements in the stock market are
random?
15.81 Many state lotteries offer a “daily number” game, in which a three-digit number is randomly drawn
every day. Suppose that a certain state generates its daily number by a computer program, and the state
lottery commissioner suspects that the process is flawed. Specifically, she feels that if today’s lucky num-
ber is higher than 500, tomorrow’s number is more likely to exceed 500; and if today’s number is below
500, tomorrow’s number is more likely to be under 500. Suppose that a sequence of the state’s lucky
numbers for a period of consecutive days is analyzed for runs above 500 and runs below 500. If the com-
missioner is right, there should be fewer runs than expected by chance. Analysis of a sequence for 50 con-
secutive days produced the following information.
n1 27 numbers above 500 n2 23 numbers below 500 R number of runs 11
Using the 2.5% level of significance, can you conclude that the sequence of all this state’s daily numbers
for this game is nonrandom?
USES AND MISUSES...
1. I’M F R E E! and alternative hypotheses, and proper application requires a detailed

Imagine that you turn on the news one morning. Still a bit sleepy, specification. Similarly, if you were to drive your car down the road
you hear a report: “And this just in from the state government: all on that dangerous morning, you would probably find that everyone
road signs have been removed. That’s right, folks. No more traffic was still driving on the right side of the road.
lights, no more ‘no parking’ signs, no more speed limits! And now,
the weather… ” You would rightfully be scared to death. During your 2. M O R E POWE R TO YO U!
drive to school or work, you would be much more vigilant than the Chapters 8 through 10, and 12 through 14 discussed a variety of
day before. You would approach an intersection more slowly, you (parametric) procedures related to statistical inference and each of
would have to rely on published maps given the absence of exit signs, these procedures requires that certain conditions hold true in order
and you would keep your eye out for very fast and very dangerous to produce valid results. The nonparametric methods presented in
drivers. Quite simply, you would not be able to rely on assumptions Chapter 11 and 15 also require a set of conditions that must hold
regarding the rules of the road. true, but these conditions are much less restrictive than the ones in
Assumptions can be good. This text has emphasized that the other chapters. If the results of a parametric test are valid, then the
parametric methods (including hypothesis testing, chi-square tests, results from an equivalent nonparametric test will also be valid. How-
F-tests, and linear regression models), require that your samples, pop- ever, the inverse of this relationship is not true. Having said that, many
ulations, and errors be normally distributed. Because of the central people wonder why not just forget about the stricter conditions and
limit theorem, the assumption that population data and large sam- instead use a nonparametric test each time? That is a good question,
ples taken from them are normally distributed is typically pretty good. and it has a good answer.
Standard statistical computer packages are able to compare your sam- As you may recall from Chapter 9, every hypothesis test has the
ples to a normal distribution and should be used to do so. The as- potential to produce an incorrect result. The possible incorrect results
sumption that a statistical sample comes from a normal distribution were referred to as Type I and Type II errors in that chapter. You may
is equivalent to augmenting your data set. In the example above, the remember from Chapter 9 that the significance level (denoted by )
assumption is roughly equivalent to the rules of the road: You do not gives the probability of a Type I error. In practice, the statistician spec-
have to check on the driver next to you all the time because you can ifies the significance level that is used in a hypothesis test. The prob-
assume that he or she will behave in a certain way. ability of a Type II error (denoted by ) in a hypothesis test problem
However, for every case in which the assumption of normality is a function of the test being used, the sample size, and the signif-
applies, there is at least one in which it does not. Without the distri- icance level. The power of a hypothesis test is defined to be 1 minus
bution parameters of the mean and variance, you are going to need the probability of a Type II error, i.e., 1 . So we try to use a test
to gather more data to make up for the fact that you do not have that has the highest power of the (hypothesis) test given a specific
these parameters, or accept a larger confidence interval than you significance level.
would have preferred. You might even find that as you collect more Parametric tests, such as the single-sample t test for the mean
data, they resemble a normal distribution after all. Additionally, the and a one-way ANOVA test, have higher powers of the test than their
nonparametric tests have their own applications and assumptions, nonparametric compatriots. Therefore, if it is reasonable to conclude
and keeping them straight can be tricky: The Kruskal-Wallis test in- that the stricter conditions of a parametric test have been met, using
vestigates if the distributions under investigation are identical but tells a parametric test will result in a higher power of the test. However,
us no more. if it is not reasonable to assume that the conditions of a parametric
Fortunately, the nonparametric methods do not abandon the test have been met, then using a parametric test will not produce
rules of the road. Each nonparametric method is based on proper- valid results. So while the power of the test will be lower when us-
ties of probability distributions and the assumption of random sam- ing a nonparametric test, you can rely on the fact that the designated
pling of the populations under investigation. Each has explicit null significance level will be met.
Glossary
Categorical data Data divided into different categories for iden- Kruskal-Wallis test A distribution-free method used to test the
tification purposes only. hypothesis that three or more populations have identical
Distribution-free test A hypothesis test in which no assumptions distributions.
are made about the specific population distribution from which the Nonparametric test A hypothesis test in which the sample data
sample is selected. are not assumed to come from a specific type of population distri-
H The test statistic used in the Kruskal-Wallis test. bution, such as the normal distribution.
Supplementary Exercises 606
s The value of the Spearman rho rank correlation coefficient Spearman rho rank correlation coefficient The linear correla-
between the ranks of the values of two variables for population data. tion coefficient between the ranks of paired data for two samples or
rs The value of the Spearman rho rank correlation coefficient be- populations.
tween the ranks of the values of two variables for sample data. T The test statistic used in the Wilcoxon signed-rank test and
R The number of runs in a runs test used to test for randomness. Wilcoxon rank sum test.
Run A sequence of one or more consecutive occurrences of the TU, TL The upper and lower critical values for the Wilcoxon rank
same outcome in a sequence of occurrences in which there are only sum test obtained from the table.
two possible outcomes. Wilcoxon rank sum test A nonparametric test that is used to test
Runs test for randomness A test that is used to test the null hy- whether two independent samples come from identically distributed
pothesis that a sequence of events has occurred randomly. populations by analyzing the ranks of the pooled sample data.
Sign test A nonparametric test that is used to test a population pro- Wilcoxon signed-rank test A nonparametric test that is used to
portion (with categorical data), a population median (with numeri- test whether two paired and dependent samples come from identi-
cal data), or the difference in population medians for two dependent cally distributed populations by analyzing the ranks of the paired
and paired sets of numerical data. differences of the samples.
Supplementary Exercises
15.82 Fifteen cola drinkers are given two paper cups, one containing Brand A cola and the other con-
taining Brand B cola. Each person tries both drinks and then indicates which one he or she prefers. The
drinks are offered in random order (some people are given Brand A first, and others get Brand B first).
Ten of the people prefer Brand A, while five prefer Brand B. Using the sign test at the .05 significance
level, can you conclude that among all cola drinkers there is a preference for Brand A?
15.83 Twenty-four randomly selected people are given samples of two brands of low-fat ice cream. Sev-
enteen of them prefer Brand B, and 7 prefer Brand A. Using the sign test at the .05 significance level, can
you conclude that among all people there is a preference for Brand B?
15.84 Four hundred randomly selected football fans were asked whether they prefer watching college foot-
ball or professional football. Of these fans, 220 said they prefer the professional games, 168 prefer col-
lege games, and 12 have no preference. At the 2% level of significance, can you conclude that among all
football fans there is a preference for either professional or college football?
15.85 A random sample of 200 customers of a large bank are asked whether they prefer using an auto-
matic teller machine (ATM) or seeing a human teller for deposits and withdrawals. Of these customers,
122 said they prefer an ATM, 66 prefer a teller, and 12 have no opinion. At the 1% level of significance,
can you conclude that more than half of all customers of this bank prefer an ATM?
15.86 Suppose that a polling agency is conducting a telephone survey. When prospective participants
answer the phone, they are told that the survey will take just five minutes of their time. Ten randomly
selected calls are monitored. The lengths of time (in minutes) required for the survey in these 10 cases
are shown here.
7.1 6.3 4.9 5.0 5.7 9.0 8.2 5.9 6.5 7.7
Using the sign test at the 5% level of significance, can you conclude that the median time for the survey
exceeds 5 minutes?
15.87 In 2001, the median age of buyers of Harley-Davidson motorcycles was 45 years (USA TODAY,
June 7, 2002). Suppose that a random sample of 25 persons who bought Harley-Davidson motorcycles re-
cently showed that 16 of them were over 45 years of age, 7 were under 45, and 2 were 45 years old. At
the 5% level of significance, can you conclude that the current median age of Harley-Davidson buyers is
over 45 years?
15.88 A state motor vehicle department requires auto owners to bring their autos to state emission centers
periodically for testing. State officials claim that the median waiting time between the hours of 8 A.M. and
11 A.M. on weekdays at a particular site is 25 minutes. In a check of 30 randomly selected motorists dur-
ing this time period at this site, 9 motorists waited for less than 25 minutes, 2 waited exactly 25 minutes,
and 19 waited longer than 25 minutes.
a. Using the sign test at the 5% significance level, can you conclude that the median waiting time
at this site during these hours exceeds 25 minutes?
b. Perform the test of part a at the 2.5% level of significance.
c. Comment on the results of parts a and b.
15.89 The following data give the amounts (in dollars) spent on textbooks by 35 college students during
the 2005–2006 academic year.
475 418 680 610 655 488 710 375 250
695 420 610 380 98 530 415 757 357
409 611 455 618 395 612 468 610 780
450 880 490 490 626 850 688 588
Using .05, can you conclude that the median expenditure on textbooks by all such students in
2005–2006 was different from $650?
15.90 Two brothers, Bob and Morris, who work the same hours at the same company in a large city,
share an apartment on the outskirts of the city. When weather permits, Bob rides his bicycle to work,
but Morris always drives. Although they always leave for work at exactly the same time each morning,
Morris often arrives later than Bob because of the heavy traffic. Last year, on 21 randomly selected days
of good weather, Bob arrived at work first 16 times, and Morris was first on 5 days. At the 5% level of
significance, can you conclude that the median morning commuting time for Bob is less than that for
Morris?
15.91 Refer to Exercise 15.34 and to Exercise 10.96 of Chapter 10, which concern Gamma Corporation’s
installation of governors on its salespersons’ cars to regulate their speeds. The following table gives the
gas mileage (in miles per gallon) for each of seven sales representatives’ cars during the week before gov-
ernors were installed and the gas mileage in the week after installation.
Before 25 21 27 23 19 18 20
After 26 24 26 25 24 22 23
a. Using the sign test at the 5% level of significance, can you conclude that the use of gover-
nors tends to increase the median gas mileage for Gamma Corporation’s sales representatives’
cars?
b. Compare your conclusion of part a with the result of the Wilcoxon signed-rank test that was
performed in part a of Exercise 15.34 and with the result of the corresponding hypothesis test
(using the t distribution) of Exercise 10.96.
c. If there is a difference in the three conclusions, how can you account for it?
15.92 A reporter for a travel magazine wanted to compare the effectiveness of two large travel agencies
(X and Y) in finding the lowest airfares to given destinations. She randomly chose 32 destinations from
the many offered by both agencies. She and her assistants requested the lowest available fare for each des-
tination from each agency. For 18 of these destinations Agency X quoted a fare lower than that of Agency
Y, for 8 destinations Agency Y found a lower fare, and in 6 cases the fares were the same. At the 2% level
of significance, can you conclude that there is any difference in the median fares quoted by Agency X and
Agency Y for all destinations they both offer?
15.93 Thirty-five patients with high blood pressures are given medication to lower their blood pressures.
For all 35 patients, their blood pressures are measured before they begin the medication and again after
they have finished taking medication for 30 days. For 25 patients the blood pressures were lower after fin-
ishing the medication, in 7 cases they were higher, and for 3 patients there was no change. Assume that
these 35 patients make up a random sample of all people suffering from high blood pressures. At the 2.5%
level of significance, can you conclude that the median blood pressure in all such patients is lower after
the medication than before?
15.94 The following table shows the one-week sales of six salespersons before and after they attended a
course on “how to be a successful salesperson.”
Before 12 18 25 9 14 16
After 18 24 24 14 19 20
a. Using the Wilcoxon signed-rank test at the 5% significance level, can you conclude that the
weekly sales for all salespersons tend to increase as a result of attending this course?
b. Perform the test of part a using the sign test at the 5% significance level.
c. Compare your conclusions from parts a and b.
15.95 An official at a figure skating competition thinks that two of the judges tend to score skaters
differently. Shown next are the two judges’ scores for eight skaters.
Skater 1 2 3 4 5 6 7 8
Judge A 5.8 5.7 5.6 5.9 5.8 5.9 5.8 5.6
Judge B 5.4 5.5 5.7 5.4 5.6 5.3 5.4 5.6
Using the Wilcoxon signed-rank test at the 5% level of significance, can you conclude that either judge
tends to give higher median scores than the other?
15.96 Refer to Exercise 15.26. Consider the data given in that exercise on the cholesterol levels (in mil-
ligrams per hundred milliliters) for 30 randomly selected adults as determined by two laboratories, A and B.
median cholesterol level for all such adults as determined by Lab A is higher than that deter-
mined by Lab B?
b. Compare your conclusion in part a to that of Exercise 15.26.
15.97 A consumer agency conducts a fuel economy test on two new subcompact cars, the Mouse (M) and the
Road Runner (R). Each of 18 randomly selected drivers takes both cars on an 80-mile test run. For each driver,
the gas mileage (in miles per gallon) is recorded for both cars; then the gas mileage for the R car is subtracted
from the gas mileage for the M car. Thus, a minus difference indicates a higher gas mileage for the R car. One
of the 18 drivers obtains exactly the same gas mileage for both cars. For the other drivers, the differences are
ranked. The sum of the positive ranks is 31, and the sum of the absolute values of the negative ranks is 122.
Can you conclude that the R car gets better gas mileage than the M car? Use the 2.5% level of significance.
15.98 Each of the two supermarkets, Al’s and Bart’s, in River City claims to offer lower-cost shopping. Fifty
people who normally do the grocery shopping for their families are chosen at random. Each shopper makes up
a list for a week’s supply of groceries. Then these items are priced and the total cost is computed for each store.
The paired differences are then calculated for each of the 50 shoppers, where a paired difference is defined as
the cost of a cart of groceries at Al’s minus the cost of the same cart of groceries at Bart’s. These paired dif-
ferences were positive for 21 shoppers and negative for 29 shoppers. The sum of ranks of the positive paired
differences was 527 and the sum of the absolute values of the ranks of the negative paired differences was 748.
Using the 1% level of significance, can you conclude that either store is less expensive than the other?
15.99 A consumer advocate is comparing the prices of eggs at supermarkets in the suburbs with the prices
of eggs at supermarkets in the cities. The following data give the prices (in dollars) of a dozen large eggs
in 13 supermarkets, 6 of which are in cities and 7 are in suburbs.
City 1.49 1.29 1.35 1.58 1.33 1.47

Suburb .99 1.09 1.39 1.28 1.16 1.44 1.05
Using the .05 level of significance and the Wilcoxon rank sum test, can you conclude that egg prices tend
to be higher in the city?
15.100 Many VCR owners have difficulty learning to program the VCR to record TV programs. A con-
sumer magazine tested two new VCRs, Brands X and Y, which are claimed to be user-friendly by their
manufacturers. A random sample of 13 adults (6 for Brand X, 7 for Brand Y) are observed to see how
quickly they can learn to program the VCRs properly. The following table gives the times (in minutes).
Brand X 32 36 28 43 98 39
Brand Y 33 18 21 25 24 27 17
Using .05 and the Wilcoxon rank sum test, can you conclude that learning times tend to be longer
for Brand X?
15.101 A researcher obtains a random sample of 24 students taking elementary statistics at a large uni-
versity and divides them randomly into two groups. Group A receives instruction to use Software A to
do a statistics assignment, whereas Group B is taught to use Software B to do the same statistics
assignment. The time (in minutes) taken by each student to complete this assignment is given in the table.
Group A 123 101 112 85 87 133 129 114 150 110 180 115
Group B 65 115 95 100 94 72 60 110 99 102 88 97
a. Using the 5% level of significance and the Wilcoxon rank sum test, can you conclude that the
median time required for all students taking elementary statistics at this university to complete
this assignment is longer for Software A than for Software B?
b. Would a paired-samples sign test be appropriate here? Why or why not?
15.102 Refer to Exercise 15.101. The scores on the homework assignment for the 24 students are given
in the table.
Group A 48 38 45 31 42 25 40 43 50 30 33 46
Group B 37 21 40 27 49 44 36 41 20 39 18 40
Using the 10% significance level and the Wilcoxon rank sum test, can you conclude that there is a dif-
ference in the median scores for all students using Software A and all students using Software B?
15.103 Manufacturers of luxury cars are very much interested in knowing the age distribution of their
customers because then they can change these models to attract younger buyers without losing the older
customers who have traditionally favored such cars. According to data from CNW Marketing Research,
the median ages of drivers (primary drivers using vehicles for personal use only) of Rolls-Royce, Mer-
cedes, and Cadillac automobiles were 62.9, 58.7, and 53.4 years, respectively, at the time of the survey
(USA TODAY, February 17, 2005). The following table gives the ages of seven randomly selected primary
drivers of each of these three makes of cars.
Rolls-Royce Mercedes Cadillac

64 61 52
61 47 63
70 66 39
68 71 55
55 44 50
64 53 47
68 58 61
At the 5% level of significance, can you reject the null hypothesis that the median age of drivers for each
of these three makes of cars is the same?
15.104 An academic employment service compared the starting salaries of May 2005 graduates in three
major fields. Random samples were taken of 8 engineering majors, 10 business majors, and 7 mathemat-
ics majors. The starting salaries of all 25 graduates were determined and then ranked, yielding the following
rank sums:
Engineering: 137 Business: 126 Mathematics: 62
At the 5% level of significance, can you reject the null hypothesis that the median starting salary is the
same for May 2005 graduates in these three fields?
15.105 A sports magazine conducted a test of three brands (A, B, and C) of golf balls by having a pro-
fessional golfer drive six balls of each brand. The lengths of the drives (in yards) from this test are listed
in the table.
Brand A Brand B Brand C

275 245 267
266 256 283
301 261 259
281 270 250
288 259 263
277 262 256
At the 5% level of significance, can you reject the null hypothesis that the median distance of drives by
this golfer is the same for all three brands of golf balls?
15.106 A students’ group at a state university wanted to compare textbook costs for students majoring in
economics, history, and psychology. The group obtained data from random samples of 10 economics
majors, 9 history majors, and 11 psychology majors, all in the second semester of their junior year. The
total textbook costs of the 30 students were recorded and ranked. The rank sums for economics and his-
tory majors were 134 and 157, respectively.
a. Find the rank sum for psychology majors. [Hint: The sum of n integers from 1 through n is
given by n(n 1)2.]
b. At the 2.5% significance level, can you reject the null hypothesis that the median textbook
costs are the same for students in all three majors who are in the second semester of the
junior year?
15.107 The following table shows the average verbal SAT score and the percentage of high school grad-
uates who took the SAT in 2002 for a random sample of 10 states.
Percentage of Graduates
State Average Verbal SAT Score Taking SAT
Connecticut 509 83
Georgia 489 65
Illinois 578 11
Kentucky 550 12
Michigan 558 11
New Jersey 498 82
South Carolina 488 59
South Dakota 576 5
Vermont 512 69
Wisconsin 583 7
Source: The College Board, The World Almanac and Book of Facts, 2003.
a. For all 50 states, would you expect s to be positive, negative, or near zero? Why?
b. Calculate rs for the sample of 10 states and indicate whether its value is consistent with your
answer to part a.
c. Using the value of rs calculated in part b, test H0: s 0 against H1: s 0 using the 5% level
of significance.
15.108 The Spearman rho rank correlation coefficient may be used in cases where data for one or both
variables are given in the form of ranks. Suppose that a film critic views 10 randomly chosen new movies
and ranks them, with a rank of 10 being assigned to the film that he thinks will have the highest box of-
fice receipts, a rank of 9 to the next most profitable, and so forth. Three months after each film is released,
its total box office receipts (in millions of dollars) are tabulated. The following table shows the ranking
and receipts for each of these 10 films.
Rank 7 3 10 1 4 5 2 6 8 9
Receipts 40 5 66 2 3 10 28 15 30 17
a. Calculate rs for the sample of 10 films.

b. Using the 5% level of significance, test H0: s 0 against H1: s 0.
c. Based on your conclusion in part b, is there sufficient evidence that this critic can predict the
box office performance of a film?
15.109 Refer to Example 13–8, which contained data on monthly auto insurance premiums and years of
driving experiences. Those data are reproduced here.
Driving Experience Monthly Auto

(years) Insurance Premium
5 $64
2 87
12 50
9 71
15 44
6 56
25 42
16 60
The estimated regression line was found to be ŷ 76.6605 1.5476x in Example 13–8 and the simple
linear correlation coefficient for the sample data was .77. The true regression slope B was found to be
significantly less than zero at the significance level of 5%.
a. Based on this information, if s is the Spearman rho rank correlation coefficient for the en-
tire population from which this sample was taken, what would you expect from a test of H0:
s 0 against H1: s 0 at the significance level of 5%?
b. Perform the hypothesis test mentioned in part a.
15.110 A researcher wonders if men still tend to stand aside and let women board elevators ahead of them.
She observes 10 men and 10 women boarding the same elevator. The order in which they boarded is given
here.
W W W M M W W M W W W M M M M W W M M M
Using the 5% level of significance, can you conclude that the order of boarding is nonrandom with re-
spect to gender?
15.111 A machinist is making precision cutting tools. Because of the exacting specifications for these
tools, about 20% of them fail to pass inspection and are judged defective. The shop supervisor feels that
the machinist tends to produce defective tools in clusters, perhaps due to fatigue or distraction. If this
is true, then a sequence of tools produced by this machinist will tend to have fewer runs of defective
and good tools than expected by chance. The supervisor chooses a day at random and observes the
following sequence of 18 tools, where G denotes a tool that passed inspection and D indicates a de-
fective tool.
G G G G G D D G G G G D G G G D D G
Do you think that there is evidence of nonrandomness in this sequence? Use the 5% level of significance.
15.112 Some states require periodic testing of cars to monitor the emission of pollutants. A state official
suspects that the inspection process at a particular station is faulty, that a car’s test result may be affected
by the tests of preceding cars. Analysis of the sequence of test results for a random sample chosen on a
day yields the following information:
n1 number of cars that passed the test 157
n2 number of cars that failed the test 143
R number of runs 41
Using the 1% level of significance, can you conclude that the test results for this emissions test station are
not random?
15.113 The following data give the sequence of wins and losses in 30 consecutive games for a baseball
team during a season.
L W L W L W L L W W W L L W L L L L L W L L L L L W L W L L
Can you conclude that the wins are randomly distributed for this baseball team? Use the 2% significance
level.
Advanced Exercises
15.114 A medical researcher wants to study the effects of a low-calorie diet on the longevity of lab-
oratory mice. She randomly divides 20 mice into two groups. Group A gets a standard diet, while
Group B receives a diet that contains all the necessary nutrients but provides only 70% as many calo-
ries as Group A’s diet. The experiment is conducted for 36 months and the length of life (in days) of
each mouse is recorded. The data obtained on the lives of these mice are shown in the following table.
In these data, the asterisk (*) indicates that this mouse was still alive at the end of the 36-month
experiment.
Group A 900 907 751 833 920 787 850 877 848 901
Group B 1037 905 1023 988 1078 1011 * 1063 898 1033
a. Using the 2.5% level of significance and the Wilcoxon rank sum test, does the low-calorie
diet appear to lengthen the longevity of laboratory mice? Should the mouse that was still
alive at the end of the experiment be eliminated from your analysis or is there a way to
include it?
b. Would a Wilcoxon signed-rank test be appropriate in this example? Why or why not?
15.115 The editor of an automotive magazine has asked you to compare the median gas mileages for driv-
ing in the city for three models of compact cars. The editor has made available to you one car of each of
the three models, three drivers, and a budget sufficient to buy gas and pay the drivers for approximately
500 miles of city driving for each car.
a. Explain how you would conduct an experiment and gather the data for a magazine article com-
paring gas mileages.
b. Suppose you wish to test the null hypothesis that the median gas mileages for driving in
the city are the same for all three models of cars. Outline the procedure for using your data to
conduct this test. Do not assume that the gas mileages for all cars of each model are normally
distributed.
15.116 Refer to Exercise 10.96 in Chapter 10. Suppose Gamma Corporation decides to test the governors
on seven cars. However, management is afraid that the speed limit imposed by the governors will reduce
the number of contacts the salespersons can make each day. Thus, both the fuel consumption and the num-
ber of contacts made are recorded for each car/salesperson for each week of the testing period, both be-
fore and after the installation of governors.
Number of Contacts Fuel Consumption (mpg)
Salesperson Before After Before After

A 50 49 25 26
B 63 60 21 24
C 42 47 27 26
D 55 51 23 25
E 44 50 19 24
F 65 60 18 22
G 66 58 20 23
Suppose that you are directed to prepare a brief report that includes statistical analysis and interpretation
of the data. Management will use your report to help decide whether or not to install governors on all
salespersons’ cars. Use 5% significance levels for any hypothesis tests you perform to make suggestions.
In contrast to Exercise 10.96, do not assume that the numbers of contacts, fuel consumption, or differ-
ences are normally distributed.
15.117 Suppose that you are a newspaper reporter and your editor has asked you to compare the
hourly wages of carpenters, plumbers, electricians, and masons in your city. Because many of these
workers are not union members, the wages may vary considerably among individuals in the same
trade.
a. What data should you gather to make this statistical analysis and how would you collect them?
What sample statistics would you present in your article and how would you calculate them?
Assume that your newspaper is not intended for technical readers.
b. Suppose that you must submit your findings to a technical journal that does require statistical
analysis of your data. If you want to determine whether or not the median hourly wages are
the same for all four trades, briefly describe how you would analyze the data. Do not assume
that the hourly wages for these populations are normally distributed.
15.118 Consider the data in the following table.
x 10 20 30 40 50 60
y 12 15 19 21 25 30
a. Suppose that each value of y in the table is increased by 5 but the x values remain unchanged.
What effect will this have on the rank of each value of y? Do you expect the value of rs to in-
crease, decrease, or remain the same? Explain why.
b. Now, first calculate the value of rs for the data in the table, and then increase each value of
y by 5 and recalculate the value of rs. Does the value of rs increase, decrease, or remain the
same? Does this result agree with your expectation in part a?
15.119 The English department at a college has hired a new instructor to teach the composition course to
first-year students. The department head is concerned that the new instructor’s grading practices might not
be consistent with those of the professor (let us call him Professor A) who taught this course previously.
She randomly selects 10 essays written by students for this class and makes two copies of each essay. She
asks Professor A and this instructor (working independently) to assign a numerical grade to each of the
10 essays. The results are shown in the following table.
Essay 1 2 3 4 5 6 7 8 9 10
Professor A 75 62 90 48 67 82 94 76 78 84
Instructor 80 50 85 55 63 78 89 81 75 83
a. Suppose the department head wants to determine whether the instructor tends to grade higher
or lower than Professor A. Which of the statistical tests discussed in this chapter could she use?
Note that more than one test may be appropriate.
b. Using an appropriate test from your answer in part a, can you conclude that the instructor tends
to grade higher or lower than Professor A? Use .05.
c. Suppose the department head wants to determine whether the instructor is consistent with Pro-
fessor A in the sense that they tend to agree on which paper is best, which is second best, and
so forth. Which test from this chapter would be appropriate to use? State the relevant null and
alternative hypotheses.
d. Using the test you chose in part c, can you conclude that Professor A and the instructor are
consistent in their grading? Use the 5% level of significance.
15.120 Three doctors are employed at a large clinic. The manager at the clinic wants to know whether
these three doctors spend the same amount of time per patient. The manager randomly chooses 10 rou-
tine appointments of patients with each of the three doctors and times them. Thus, the data set consists of
10 observations on the time spent with patients by each doctor.
a. To test the null hypothesis that the mean or median times are equal for all three doctors against
the alternative hypothesis that they are not all equal, which tests from Chapters 12 and 15 are
appropriate?
b. For each test that you indicated in part a, specify whether the test is about the means or the
medians.
c. What assumptions are required for the test from Chapter 12?
15.121 An educational researcher is studying the relationship between high school grade point averages
(GPAs) and SAT scores. She obtains GPAs and SAT scores for a random sample of 25 students and wants
to test the null hypothesis that there is no correlation between GPAs and SAT scores against the alterna-
tive hypothesis that these variables are positively correlated.
a. If she wants to base her test on the linear correlation coefficient of Chapter 13, what assump-
tions are required about the two variables (GPAs and SAT scores)?
b. If the assumptions required in part a are not satisfied, what other test(s) might she use?
15.122 To test the effectiveness of a new six-week body-building course, 12 tenth-grade boys are ran-
domly selected. Each boy is tested before and after the course to see how much weight he can lift.
a. To test whether or not the mean or median weight lifted by all such boys tends to be greater
after the course than before, which tests from Chapters 10 and 15 might be used?
b. For each test that you indicated in part a, specify whether it involves the mean or median.
c. If the paired differences in weights lifted before and after the test are not normally distributed,
which of the tests indicated in part a could be used?
15.123 Suppose in a sample we have 10 A’s and 15 B’s. What is the maximum number of runs possible
in a sequence of these 25 letters?
15.124 Refer to Exercises 12.27 and 15.56. In these two problems you were asked to perform an ANOVA
and a Kruskal-Wallis test, respectively, on the data. In both cases, the results were significant at the 5%
significance level. Change the values of the tips in those data so that the p-value for the Kruskal-Wallis
test remains the same, but the ANOVA results are no longer significant at the 5% level. (Hint: When mak-
ing the changes, the ranks of the fifteen data points should not change.)
15.125 A student who typically does not do his homework was asked to toss a coin 20 times and write
down the sequence of results. Instead of tossing the coin, the student simply wrote down the following
sequence (reading from left to right) of hypothetical outcomes.
H T H T H T H H T H
T T H T H T H T H T
Use the appropriate test to show that the professor was justified in accusing the student of not actually
tossing the coin.
Self-Review Test 614
Self-Review Test
1. Nonparametric tests
a. are more efficient than the corresponding parametric tests
b. do not require that the population being sampled has a normal distribution
c. generally require more assumptions about the population than parametric tests do
2. For small samples (n 25), the critical value(s) for the sign test are based on the
a. binomial distribution b. normal distribution c. t distribution
3. Which of the following tests may be used to test hypotheses about one median?
a. The sign test b. The Kruskal-Wallis test c. The Wilcoxon rank sum test
4. The Wilcoxon signed-rank test may be used to test
a. for a difference between the medians of two independent samples
b. for a preference for one product over another
c. hypotheses involving paired samples
5. When we use the Wilcoxon signed-rank test,
a. all observations are ranked
b. the difference for each pair is calculated and then all the differences are ranked according to their
absolute values
c. only the signs of the differences are used to calculate the value of the test statistic
6. In order to perform a Wilcoxon rank sum test, one must calculate the
a. standard deviation of each sample
b. range of the data
c. rank of each observation
7. Which of the following tests may be used with paired samples? Circle all that apply.
a. Sign test
b. Wilcoxon signed-rank test
c. Wilcoxon rank sum test
d. Spearman rho rank correlation coefficient test
8. The Spearman rho rank correlation coefficient is calculated as the
a. simple linear correlation coefficient between the two sets of observations
b. simple linear correlation coefficient between the ranks of the two sets of observations
c. square of the simple linear correlation coefficient between two sets of observations
9. In order to test a hypothesis about the Spearman rho rank correlation coefficient
a. both sets of data must come from normally distributed populations
b. one set of data must come from a normally distributed population
c. either set of data can have any distribution
10. The Spearman rho rank correlation coefficient is positive when
a. there is no relationship between the two sets of observations
b. the values in one set of observations increase as the values of the corresponding observations in
the other set decrease
c. the values in one set of observations increase as the values of the corresponding observations in
the other set increase
11. For the runs test for randomness, which of the following statements are true?
a. We notice which of the two possible outcomes has occurred at each stage in a list of consecutive
outcomes.
b. A run is one or more consecutive occurrences of either one of the two possible outcomes.
c. We are testing the hypothesis that one of the two possible outcomes occurred significantly more
frequently than the other.
12. In the runs test for randomness, we reject the null hypothesis
a. only when there is a very large number of runs
b. only when there is a very small number of runs
c. if there is either a very large or a very small number of runs
13. In the runs test for randomness, the distribution of R (the total number of runs) is approximately
normal when
a. R is greater than 10
b. at least one of the two possible outcomes occurs more than 15 times
c. each of the two possible outcomes occurs more than 15 times
14. A large pool of prospective jurors is made up of an equal number of men and women. A 12-person
jury selected from this pool consists of 2 women and 10 men. At the 5% level of significance, can we
reject the null hypothesis that the selection process is unbiased in terms of gender?
15. A September 2002 USA TODAY/Gallup poll asked Americans whether they favored a proposal to put
Social Security payroll taxes into personal retirement accounts. Fifty-two percent of the respondents were
in favor of the proposal (USA TODAY, September 25, 2002). Suppose that the 2002 poll consisted of 1000
respondents, so that 520 favored the proposal. Using the 2.5% level of significance, can you conclude that
more than half of all Americans favor putting Social Security payroll taxes into personal retirement
accounts?
16. The past records of a supermarket show that its customers spent a median of $65 per visit. After a
promotional campaign designed to increase spending, the store took a sample of 12 customers and recorded
the amounts (in dollars) they spent. The amounts are listed here.
88 69 141 28 106 45 32 51 78 54 110 83
Using .05, can you conclude that the median amount spent by all customers at this store after the
campaign exceeds $65?
17. According to a U.S. Census Bureau survey of households, women living alone had a median income
of $20,264 in 2001 (USA TODAY, September 25, 2002). Suppose that in a recent random sample of 400
women living alone, 229 had incomes under $20,264 and 171 had incomes over $20,264. At the 1% level
of significance, can you conclude that the median income of women living alone currently is different
from $20,264?
18. The following table gives the cholesterol levels for seven adults before and after they completed a
special dietary plan.
Before 210 180 195 220 231 199 224

After 193 186 186 223 220 183 233
a. Using the sign test at the 5% significance level, can you conclude that the median cholesterol lev-
els are the same before and after the diet?
b. Using the Wilcoxon signed-rank test at the 5% significance level, can you conclude that the me-
dian cholesterol levels are the same before and after the diet?
c. Compare your conclusions for parts a and b.
19. An archeologist wants to compare two methods (I and II) of radioactive dating of artifacts. He sub-
mits a random sample of 33 artifacts that are suitable for radioactive dating. Each one of these artifacts is
dated by both methods. The paired differences are then calculated for each of the 33 artifacts, where a
paired difference is defined as the age of an artifact dated by Method I minus the age of the same artifact
dated by Method II. These paired differences were positive for 11 of the artifacts, negative for 20, and
zero for 2 artifacts. Using the sign test at the 2% level of significance, can you conclude that the median
estimated ages of such artifacts differ for the two methods?
20. A professor at a large university suspects that the grades of engineering majors tend to be lower in
the spring semester than in the fall semester. He randomly selects 10 sophomore electrical engineering
majors and records their grade point averages (GPAs) for the fall and the spring semesters. The data ob-
tained are shown in the table.
Student 1 2 3 4 5 6 7 8 9 10
Fall GPA 3.20 3.56 3.05 3.78 4.00 2.85 3.33 2.67 3.00 3.67
Spring GPA 3.15 3.40 2.88 3.67 4.00 3.00 3.30 3.05 2.95 3.50
Using the Wilcoxon signed-rank test at the 5% level of significance, can you conclude that the median
GPA of all sophomore electrical engineering majors at this university tends to be lower in the spring se-
mester than in the fall semester?
21. A random sample of 30 students was selected to test the effectiveness of a course designed to im-
prove memory. Each student was given a memory test before and after taking the course. Each student’s
score after taking the course was subtracted from his or her score before the course; then the 30 differ-
ences were ranked. Thus, a negative rank denotes an improved score after taking the course. The sum of
the positive ranks was 102; the sum of the absolute values of the negative ranks was 276. Three students
scored exactly the same on both tests. Using the 2.5% level of significance, can you conclude that the
course tends to improve scores on memory tests?
Self-Review Test 616
22. A commuter has two alternative routes (Route 1 and Route 2) to drive to work. Picking days at random,
she drives to work using each route for eight days and records the time (in minutes) taken to commute
from home to work on each day. These times are shown in the following table.
Route 1 45 43 38 56 41 43 46 44
Route 2 38 40 39 42 50 37 46 36
Using the Wilcoxon rank sum test at the 5% level of significance, can you reject the null hypothesis that
the median commuting time is the same for both routes?
23. An accounting firm has hired two temporary employees, A and B, to prepare individual federal income
tax returns during the tax season. Clients who have relatively simple tax situations are randomly assigned to
either A or B. The firm randomly selected 18 income tax returns prepared by each of these two employees
and recorded the times taken to prepare these tax returns. After these times taken to prepare 36 tax returns
were ranked, the sum of the ranks for A was found to be 298 and the sum of the ranks for B was equal to
368. Using the Wilcoxon rank sum test at the 2.5% level of significance, can you conclude that there is a
difference in the median times taken to prepare such income tax returns by A and B?
24. The following table lists the numbers of cases of telemarketing fraud reported to law-enforcement officials
during several randomly chosen weeks in 2002 for three large cities of approximately equal populations.
City A City B City C

53 29 75
46 35 49
59 44 62
33 31 68
60 50 52
48
a. At the 2.5% level of significance, can you reject the null hypothesis that the distributions of the
numbers of such reported cases are identical for all three cities?
b. Can you reject the null hypothesis of part a at the 1% level of significance?
c. Comment on the results of parts a and b.
25. The following is a list of home runs (denoted by x) and runs batted in (denoted by y) as of July 1,
2005, by 10 players selected at random from a minor league baseball team.
Player 1 2 3 4 5 6 7 8 9 10
x 10 7 13 2 8 4 16 11 5 4
y 49 38 54 20 41 27 62 40 22 19
a. As home runs increase, runs batted in tend to increase. From this, do you expect the value of the
Spearman rho rank correlation coefficient to be positive or negative?
b. Compute rs for the data.
c. Suppose s is the value of the Spearman rho rank correlation coefficient for all players in this
league. Using the 2.5% significance level, test the null hypothesis H0: s 0 against the alternative
hypothesis H1: s 0.
26. Ramon fishes in a lake where the minimum size for bass to be kept is 12 inches long; all smaller bass
must be returned to the water. He thinks that most of the “keepers” (bass 12 inches or longer) are caught early
in the morning. If he is right, there should be a few long runs of keepers caught in the early morning followed
by a few long runs of smaller bass caught later on during the day. Thus, if the fish are recorded in sequence,
there should be fewer runs than expected by chance. Last Saturday Ramon fished from 6 A.M. to 11 A.M. and
caught 14 bass in the following order, where K denotes a keeper and S denotes a bass shorter than 12 inches.
K K K K S K K S K S S S S S
Using the 5% significance level, does this sequence support Ramon’s theory?
27. As of June 1, 2005, a minor league baseball team had played 54 games, winning 30 and losing 24.
In these 54 consecutive games, there were 15 runs (in the statistical sense). Using the runs test for ran-
domness, can we conclude that the 30 wins and 24 losses are randomly spread out among the 54 games?
Use a significance level of 5%.
Mini-Projects
MINI-PROJECT 15–1
For a period of 30 business days, record the daily price of crude oil and the price of a stock that you think
might be affected by the oil price (for example, an oil company or an alternative energy company).
a. Compute the Spearman rho rank correlation coefficient for these data.
b. Can you conclude that there is a relationship between the two sets of prices? Use .05.
MINI-PROJECT 15–2
For a period of 30 business days, record whether the Dow Jones Industrial Average moves up or down.
Use your data to perform an appropriate test at the 1% level of significance to see if the sequence of up-
ward and downward movements of the Dow appears to be random over this period. (As an alternative,
you might use the NASDAQ index, the price of an individual stock, or the price of gold.)
MINI-PROJECT 15–3
On December 8, 2005, the U.S. national debt was approximately $8.13 trillion. Take random samples of
10 or more students from each of three different majors and ask each student to estimate the size of the
national debt.
a. At the 5% level of significance and using the Kruskal-Wallis test, can you conclude that the me-
dian perceived values of the national debt are the same for all three majors?
b. Find the current value of the national debt. On the whole, did the students in your sample tend to
overestimate its value, or did they tend to underestimate?
DECIDE FOR YOURSELF

Using Nonparametric Methods
Prior to this chapter, you used inferential methods to work with quan-
titative data that are classified as scale data or with categorical data that
are classified as nominal data.2 A third type of data, called ordinal data,
often requires nonparametric methods for analysis. Ordinal data are
data that can be ranked. For example, insurance companies will clas-
sify policy holders by their age groups, as opposed to their specific
ages, for assessing risk. Since nobody can be in two different age
groups at the same time, the data identifying age groups can be ranked.
Suppose you are interested in studying the relationship between
the successes of Division I men’s and women’s basketball teams at
colleges that have both of these sports. We can use Spearman’s Rho
to calculate the rank correlation between the power index rankings
(RPIs) of the men’s and women’s teams. We have provided three
scatterplots here. Specifically we have the scatterplot of the ranks of
the men’s teams versus the ranks of the women’s teams for various Figure 15.15
colleges (Figure 15.15), the scatterplot of the RPIs of men’s teams
Answer the following questions.
versus the RPIs of women’s teams (Figure 15.16), and the scatterplot
of the RPIs of men’s teams versus their ranks (Figure 15.17). These 1. Which of the three relationships should have its strength meas-
data are for all Division I programs that had men’s and women’s ured by Spearman’s Rho rank correlation instead of the Pearson cor-
basketball programs during the 2004–2005 season. relation coefficient?
2
Based on what are called the scales or levels of measurement, data can be of such data is an evaluation of a product as excellent, good, or poor. Data
classified into four scales or levels—nominal, ordinal, interval, and ratio that can be ranked and for which the difference between any two values
scales. Data that can be divided into different categories only for identifi- can be calculated (and is meaningful) are said to have an interval scale. An
cation purposes are said to have a nominal scale. An example of such data example of such data is temperatures in two cities. Data that can be ranked
is the names given to different makes of cars, such as Town Car, Toyota and for which all arithmetic operations (such as addition, subtraction, mul-
Camry, and so forth. Data that can be divided into different categories so tiplication and division) can be done are said to have ratio scale. An exam-
that categories can be ranked are said to have an ordinal scale. An example ple of such data is gross sales of two companies.
Technology Instruction 618
Figure 15.16
2. Which of the three relationships is inappropriate to be discussed

in terms of its correlation? Figure 15.17
3. There is one point that stands out on the scatterplot of RPI values
(Figure 15.16). Identify the location of this point on the scatterplot
of ranks (Figure 15.15).
T ECH NOLOGY
I NSTR UCTION
TI-84
Nonparametric Methods
1. The TI-84 does not contain any built-in nonparametric methods.
MINITAB
1. To perform a sign test about a population
median, select Stat>Nonparametrics>1-
Sample Sign. Enter the name of the col-
umn containing your sample data in the
box below Variables, select Test median,
enter the hypothesized value of the median,
and select your alternative hypothesis.
Click OK to see the results. (See Screens
15.1 and 15.2.) Then use the p-value ob-
tained in this output to make a decision.
2. To run a Wilcoxon rank sum test to deter-

mine if the populations from which two
independent samples are drawn are identi-
cal, select Stat>Nonparametrics>Mann-
Whitney. (Note that MINITAB does not
have procedures for the Wilcoxon rank
sum test. The Mann-Whitney test is very
Screen 15.1
Screen 15.2
similar to the Wilcoxon rank sum test and we can use it

here.) Enter the names of the two columns containing your
data in their respective boxes. Select the alternative hypoth-
esis. Click OK to see the results. (See Screens 15.3 and
15.4.) Then use the p-value obtained in this output to make
a decision.
3. To run a Kruskal-Wallis test to determine if three or

more populations have identical distributions, enter the
data on the response variable in one column and the
factor in another column. Select Stat>Nonpara-
metrics>Kruskal-Wallis and enter the columns
Screen 15.3
containing the response variable and factor in their

respective boxes. Click OK to see the results. (See
Screens 15.5 and 15.6.) Then use the p-value
obtained in this output to make a decision.
4. To perform a Wilcoxon signed-rank test, enter the
pairwise differences in a single column, making sure
that you calculate each difference in the same order.
Screen 15.4 Select Stat>Nonparametrics>1-Sample Wilcoxon.
Enter the name of the column containing the differ-
ences in the box below Variables, select Test
median, enter the hypothesized value of the median,
and select your alternative hypothesis. Click OK to
Screen 15.6
Screen 15.5
Technology Assignments 620
see the results. (See Screens 15.7 and

15.8.) Then use the p-value obtained in
this output to make a decision.
5. To perform a runs test to see if a set of
data is random, enter the given data into
a column. Note that categorical data must
be given numeric values. Select
Stat>Nonparametrics>Runs Test and
enter the column containing data in the
box below Variables. Click OK to see
the results. (See Screens 15.9 and 15.10.)
Then use the p-value obtained in this out-
put to make a decision.
Screen 15.7
Screen 15.8
Screen 15.10
Screen 15.9
Excel
Excel does not contain any built-in nonparametric methods.
TECHNOLOGY ASSIGNMENTS
TA15.1 Fifteen coffee drinkers are selected at random and asked to test and state their preferences for
Brand X coffee, Brand Y coffee, or neither (N). The results are as follows:
X X Y X N Y Y X Y X X Y Y X X
Let p be the proportion of coffee drinkers in the population who prefer Brand X. Using the sign test, per-
form the test H0: p .50 against H1: p .50. Use a significance level of 2.5%.
TA15.2 Twelve sixth-graders were selected at random and asked how many hours per week they spend
watching television. The data obtained are shown here.
23 30 22.5 28 29 24.5 25 32 31 26 27 21
Using the sign test, can you conclude that the median number of hours spent per week watching televi-
sion by all sixth-graders is less than 28? Use a significance level of 5%.
TA15.3 The manufacturer of an engine oil additive, Hyper-Slick, claims that this product reduces the en-
gine friction and, consequently, increases the miles per gallon (mpg). To test this claim, 10 cars are driven
on a fixed 300-mile course without the oil additive, and each car’s mpg is calculated and recorded. Then the
engine oil additive is added to each car and the process is repeated. The data obtained are shown in the table.
Car MPG without Additive MPG with Additive

1 20.00 19.90
2 23.60 27.85
3 29.40 28.70
4 25.70 28.20
5 35.80 37.30
6 32.20 31.30
7 26.30 26.10
8 31.80 36.80
9 29.00 32.75
10 24.70 29.20
Using the sign test, can you conclude that the manufacturer’s claim is true? Use a significance level of 5%.
TA15.4 Do Technology Assignment TA15.3 using the Wilcoxon signed-rank test and a significance level
of 5%. Compare your conclusion with that of Technology Assignment TA15.3 and comment.
TA15.5 Refer to Exercise 15.43. In a winter Olympics trial for women’s speed skating, seven skaters use
a new type of skate, while eight others use the traditional type. Each skater is timed (in seconds) in the
500-meter event. The results are given in the table.
New skates 40.5 40.3 39.5 39.7 40.0 39.9 41.5

Traditional skates 41.0 40.8 40.9 39.8 40.6 40.7 41.1 40.5
Assume that these 15 skaters make up a random sample of all Olympic-class 500-meter female speed
skaters. Using the Wilcoxon rank sum test (Mann-Whitney test), can you conclude that the new skates
tend to produce faster times in this event? Use the 5% significance level.
TA15.6 Refer to Exercise 15.46. Two brands of tires are tested to compare their durability. Eleven Brand
X tires and 12 Brand Y tires are tested on a machine that simulates road conditions. The mileages (in thou-
sands of miles for each tire) are shown in the following table.
Brand X 51 55 53 49 50.5 57 54.5 48.5 51.5 52 53.5

Brand Y 48 47 54 55.5 50 51 46 49.5 52.5 51 49 45
Using the Wilcoxon rank sum (Mann-Whitney) test, can you conclude that the median mileage for Brand X
tires is greater than the median mileage for Brand Y tires? Use the 5% level of significance.
TA15.7 Three brands of 60-watt lightbulbs—Brand A, Brand B, and a generic brand—are tested for their
lives. The following table shows the lives (in hours) of these bulbs.
Brand A Brand B Generic

975 1001 899
1050 1099 789
890 915 824
933 959 1011
962 986 907

925 957 923
1007 987 937
855 881 865
1025 1024
Using the Kruskal-Wallis test with a significance level of .05, can you conclude that the distributions of
the lives of lightbulbs are the same for all three brands?
TA15.8 Refer to Exercise 15.54. A consumer agency investigated the premiums charged by four auto
insurance companies. The agency randomly selected five drivers insured by each company who had sim-
ilar driving records, autos, and insurance coverages. The following table gives the monthly premiums paid
by these 20 drivers.
Company A Company B Company C Company D

$65 $48 $57 $62
73 69 61 53
54 88 89 45
43 75 77 51
70 72 69 44
Using the Kruskal-Wallis test at the 5% significance level, can you reject the null hypothesis that the distribu-
tions of auto insurance premiums paid per month by all such drivers are the same for all four companies?
TA15.9 Refer to Exercise 15.75. A fair coin is tossed 20 times in the presence of a psychic who claims
that she can cause a nonrandom sequence of heads (H) and tails (T) to appear. The following sequence of
heads and tails is obtained in these 20 tosses.
H H T H T H T T H T H H T T H H H T T H
Using the runs test, can you conclude that the psychic’s claim is true? Use a significance level of 5%.
TA15.10 At a small soda factory, the amount of soda put into each 12-ounce bottle by the bottling ma-
chine varies slightly for each filling. The plant manager suspects that the machine has a random pattern
of overfilling and underfilling the bottles. The following are the results of filling 18 bottles, where O
denotes 12 ounces or more of soda in a bottle and U denotes less than 12 ounces of soda.
U U U O O O O U U O O O U O U U U U
Using the runs test at the 5% significance level, can you conclude that there is a nonrandom pattern of
overfilling and underfilling such bottles?
Table VIII Critical Values of X for the Sign Test
One tail .005 One tail .01 One tail .025 One tail .05
Two tail .01 Two tail .02 Two tail .05 Two tail .10
Lower Upper Lower Upper Lower Upper Lower Upper

critical critical critical critical critical critical critical critical
n value value value value value value value value
1 — — — — — — — —
2 — — — — — — — —
3 — — — — — — — —
4 — — — — — — — —
5 — — — — — — 0 5
6 — — — — 0 6 0 6
7 — — 0 7 0 7 0 7
8 0 8 0 8 0 8 1 7
9 0 9 0 9 1 8 1 8
10 0 10 0 10 1 9 1 9
11 0 11 1 10 1 10 2 9
12 1 11 1 11 2 10 2 10
13 1 12 1 12 2 11 3 10
14 1 13 2 12 2 12 3 11
15 2 13 2 13 3 12 3 12
16 2 14 2 14 3 13 4 12
17 2 15 3 14 4 13 4 13
18 3 15 3 15 4 14 5 13
19 3 16 4 15 4 15 5 14
20 3 17 4 16 5 15 5 15
21 4 17 4 17 5 16 6 15
22 4 18 5 17 5 17 6 16
23 4 19 5 18 6 17 7 16
24 5 19 5 19 6 18 7 17
25 5 20 6 19 7 18 7 18
Source: D. B. Owen, Handbook of Statistical Tables. © 1962 by Addison-Wesley Publishing Company, Inc. Reprinted
by permission of Addison Wesley Longman.
Table IX Critical Values of T for the Wilcoxon Signed-Rank Test
One-tailed .005 One-tailed .01 One-tailed .025 One-tailed .05

n Two-tailed .01 Two-tailed .02 Two-tailed .05 Two-tailed .10
1 — — — —
2 — — — —
3 — — — —
4 — — — —
5 — — — 1
6 — — 1 2
7 — 0 2 4
8 0 2 4 6
9 2 3 6 8
10 3 5 8 11
11 5 7 11 14
12 7 10 14 17
13 10 13 17 21
14 13 16 21 26
15 16 20 25 30
Source: Some Rapid Approximate Statistical Procedures, 1964. Reprinted with permission of Lederle Pharmaceutical
Division of American Cyanamid Company, Philadelphia, PA.
Table X Critical Values of T for the Wilcoxon Rank Sum Test
a. One-tailed .025; Two-tailed .05

n1 3 4 5 6 7 8 9 10
n2
TL TU TL TU TL TU TL TU TL TU TL TU TL TU TL TU
3 5 16 6 18 6 21 7 23 7 26 8 28 8 31 9 33
4 6 18 11 25 12 28 12 32 13 35 14 38 15 41 16 44
5 6 21 12 28 18 37 19 41 20 45 21 49 22 53 24 56
6 7 23 12 32 19 41 26 52 28 56 29 61 31 65 32 70
7 7 26 13 35 20 45 28 56 37 68 39 73 41 78 43 83
8 8 28 14 38 21 49 29 61 39 73 49 87 51 93 54 98
9 8 31 15 41 22 53 31 65 41 78 51 93 63 108 66 114
10 9 33 16 44 24 56 32 70 43 83 54 98 66 114 79 131
b. One-tailed .05; Two-tailed .10

n1 3 4 5 6 7 8 9 10
n2
TL TU TL TU TL TU TL TU TL TU TL TU TL TU TL TU
3 6 15 7 17 7 20 8 22 9 24 9 27 10 29 11 31
4 7 17 12 24 13 27 14 30 15 33 16 36 17 39 18 42
5 7 20 13 27 19 36 20 40 22 43 24 46 25 50 26 54
6 8 22 14 30 20 40 28 50 30 54 32 58 33 63 35 67
7 9 24 15 33 22 43 30 54 39 66 41 71 43 76 46 80
8 9 27 16 36 24 46 32 58 41 71 52 84 54 90 57 95
9 10 29 17 39 25 50 33 63 43 76 54 90 66 105 69 111
10 11 31 18 42 26 54 35 67 46 80 57 95 69 111 83 127
Source: Some Rapid Approximate Statistical Procedures, 1964. Reprinted with the permission of Lederle Pharmaceutical Division of American
Cyanamid Company, Philadelphia, PA.
Table XI Critical Values for the Spearman Rho Rank

Correlation Coefficient Test
One-tailed
.05 .025 .01 .005
Two-tailed
n .10 .05 .02 .01
5 .900 — — —
6 .829 .886 .943 —
7 .714 .786 .893 .929
8 .643 .738 .833 .881
9 .600 .700 .783 .833
10 .564 .648 .745 .794
11 .536 .618 .709 .755
12 .503 .587 .678 .727
13 .475 .566 .672 .744
14 .456 .544 .645 .714
15 .440 .524 .622 .688
16 .425 .506 .601 .665
17 .411 .490 .582 .644
18 .399 .475 .564 .625
19 .388 .462 .548 .607
20 .377 .450 .534 .591
21 .368 .438 .520 .576
22 .359 .428 .508 .562
23 .351 .418 .496 .549
24 .343 .409 .485 .537
25 .336 .400 .475 .526
26 .329 .392 .465 .515
27 .323 .384 .456 .505
28 .317 .377 .448 .496
29 .311 .370 .440 .487
30 .305 .364 .432 .478
Table XII Critical Values for a Two-Tailed Runs Test with .05
n2
n1 5 6 7 8 9 10 11 12 13 14 15
2 — — — — — — — 2 2 2 2
6 6 6 6
3 — 2 2 2 2 2 2 2 2 2 3
8 8 8 8 8 8 8 8 8 8
4 2 2 2 3 3 3 3 3 3 3 3
9 9 10 10 10 10 10 10 10 10 10
5 2 3 3 3 3 3 4 4 4 4 4
10 10 11 11 12 12 12 12 12 12 12
6 3 3 3 3 4 4 4 4 5 5 5
10 11 12 12 13 13 13 13 14 14 14
7 3 3 3 4 4 5 5 5 5 5 6
11 12 13 13 14 14 14 14 15 15 15
8 3 3 4 4 5 5 5 6 6 6 6
11 12 13 14 14 15 15 16 16 16 16
9 3 4 4 5 5 5 6 6 6 7 7
12 13 14 14 15 16 16 16 17 17 18
10 3 4 5 5 5 6 6 7 7 7 7
12 13 14 15 16 16 17 17 18 18 18
11 4 4 5 5 6 6 7 7 7 8 8
12 13 14 15 16 17 17 18 19 19 19
12 4 4 5 6 6 7 7 7 8 8 8
12 13 14 16 16 17 18 19 19 20 20
13 4 5 5 6 6 7 7 8 8 9 9
12 14 15 16 17 18 19 19 20 20 21
14 4 5 5 6 7 7 8 8 9 9 9
12 14 15 16 17 18 19 20 20 21 22
15 4 5 6 6 7 7 8 8 9 9 10
12 14 15 16 18 18 19 20 21 22 22
Source: Frieda S. Swed and C. Eisenhart, “Tables for Testing Randomness of Grouping in a Sequence of Alternatives,”
The Annals of Statistics 14(1943). Reprinted with permission of the Institute of Mathematical Statistics.

ch15 PDF

Uploaded by

Copyright:

Available Formats

ch15 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

ch15 PDF

Uploaded by

Copyright:

Available Formats

Chapter

15.1 The Sign Test

15.1.1 Tests About Categorical Data

The Small-Sample Case

Person Brand Preferred

Step 1. State the null and alternative hypotheses.

Step 4. Calculate the value of the test statistic.

Person Brand Preferred Sign

Step 5. Make a decision.

The Large-Sample Case

Step 1. State the null and alternative hypotheses.

Step 3. Determine the rejection and nonrejection regions.

Step 4. Calculate the value of the test statistic.

15.1.2 Tests About the Median of a Single Population

The Small-Sample Case

Step 1. State the null and alternative hypotheses.

Step 4. Calculate the value of the test statistic.

Step 5. Make a decision.

The Large-Sample Case

Solution We use the usual five steps to test this hypothesis.

Step 1. State the null and alternative hypotheses.

Step 4. Calculate the value of the test statistic.

15.1.3 Tests About the Median Difference

The Small-Sample Case

Next we perform the five steps for testing the hypothesis.

Step 1. State the null and alternative hypotheses.

Step 4. Calculate the value of the test statistic.

The Large-Sample Case

Step 1. State the null and alternative hypotheses.

Step 2. Select the distribution to use.

Step 3. Determine the rejection and nonrejection regions.

Step 4. Calculate the value of the test statistic.

Sample Lab A Lab B Sample Lab A Lab B

15.2 The Wilcoxon Signed-Rank Test

The Small-Sample Case

Solution We use the five steps to make the hypothesis test.

Step 1. State the null and alternative hypotheses.

0, 1, or 2 3 or higher Figure 15.7

Step 4. Calculate the value of the test statistic.

Differences Absolute Ranks of Signed

Observed Value of the Test Statistic T

Our example is a right-tailed test. Hence,

The Large-Sample Case

Solution We perform the five steps to conduct this test of hypothesis.

Step 1. State the null and alternative hypotheses.

Step 4. Calculate the value of the test statistic.

Before After Before After Before After

15.3 The Wilcoxon Rank Sum Test for

The Small-Sample Case

Step 1. State the null and alternative hypotheses.

Figure 15.9 51 or lower 52 to 92 93 or higher

Step 4. Calculate the value of the test statistic.

Crimes Rank Crimes Rank

Step 5. Make a decision.

Wilcoxon Rank Sum Test for Small Independent Samples

The Large-Sample Case

Step 2. Select the distribution to use.

Step 4. Calculate the value of the test statistic.

Stress Level Rank Stress Level Rank

c; that is, reject H0 if rs is “too small” or “too large.”

2n1n2 12n1n2 n1 n2 2 21222128212 22 28 22 282