Non-Parametric Statistics: Business Statistics - Naval Bajpai

Copyright© Dorling Kindersley India Pvt Ltd
Chapter
18
Non-Parametric
Statistics
Business Statistics - Naval Bajpai

Learning Objectives

Upon completion of this chapter, you will be able to:
 Analyse nominal as well as ordinal level of data

 Learn relative advantages of non-parametric tests over
parametric tests
 Understand when and how to use the runs test to test
randomness
 Understand when and how to use the Mann–Whitney U
test, the Wilcoxon matched-pairs signed rank test, the
Kruskal–Wallis test, the Friedman test, and Spearman’s
rank correlation

Introduction

 Parametric tests are statistical techniques to test a
hypothesis based on some restrictive assumptions about
the population. Generally, these assumptions are with
respect to the normality of the population and random
selection of samples from the normal population.
Additionally, parametric tests require quantitative
measurement of the sample data in the form of an interval
or ratio scale.
 Non-parametric tests are not dependent upon the
restrictive normality assumption of the population.
Additionally, non-parametric tests can be applied to
nominal and ordinal scaled data. These tests are also
referred to as distribution free statistics (do not require
the population to be normally distributed).

Non-Parametric tests

 Runs test
 Mann–Whitney U test
 Wilcoxon matched-pairs signed rank test
 Kruskal–Wallis test
 Friedman test
 Spearman’s rank correlation

Runs Test for Randomness of Data

 The randomness of the sample can be tested by using the runs
test. A run is defined as the sequence of identical occurrence
of the elements (numbers or symbols), preceded or followed by
different occurrence of the elements or by no elements at all.
 A company wants to send 20 employees (from the Finance and
Marketing departments) for advanced training from a large
population (all the employees of the company). The company’s
administrative officer has selected the following samples randomly
(where F represents selection from the Finance department and M
represents selection from the Marketing department):
Five runs are as shown

Small-Sample Runs Test

 In cases where the sample size is small, the small-sample
runs test is an appropriate choice. The sample is considered
to be small when n1 and n2 are less than or equal to 20,
where n1 is the number of occurrences of Type 1 and n2 is
the number of occurrences of Type 2.
 The null and alternative hypotheses can be stated as below:

Example 18.1 A company wants to send 20 employees selected
randomly from the finance and marketing departments for
advanced training. The company’s administrative officer has
selected random samples as below:
F,F,F,F,M,M,M,M,F,F,F,F,M,M,M,M,F,F,F,F
Test the randomness of the sample.
Solution : n1 = 12 and n2 = 8. Both n1 and n2 are less than

20. Hence, the small-sample runs test is an appropriate
choice.
For n1 = 12 and n2 = 8, from the table (given in the
appendices), the critical value of R for the lower tail is 6 and
the critical value of R for the upper tail is 16. If runs are less
than 6 and more than 16, the decision is to reject the null
hypothesis and accept the alternative hypothesis.

Solution (Example 18.1)

The number of runs are 5 as shown below:
The number of runs 5 is less than the critical value of R

for the lower tail, that is, 6. Hence, the decision is to reject
the null hypothesis and accept the alternative hypothesis.
So, it can be concluded that the observations in the sample
are not randomly generated.
Solved Examples\MINITAB\Ex 18.1.MPJ

Solved Examples\SPSS\Ex 18.1.sav
Solved Examples\SPSS\Output Ex 18.1.spo

Large-Sample Runs Test

For n1 and n2 greater than
20 (or either n1 or n2 is
greater than 20), the tabular
values of run are not
available. Fortunately, the
sampling distribution of R
can be approximated by the
normal distribution with
defined mean and standard
deviation.

Example 18.2 A company has installed a new machine. A quality
control inspector has examined 62 items selected by the machine
operator in a random manner. Good (G) and defective (D) items
are sampled in the following manner:
G,G,G,G,G,G,G,D,D,D,D,G,G,G,G,G,G,D,D,D,D,D,G,G,GG,G,G,G,G,
D,D,D,D,G,G,G,G,G,G,G,G,G,D,D,D,D,D,G,G,G,G,G,G,G,G,D,D,D,D
,D,D
Use alpha = 0.05 to determine whether the machine operator has
selected the sample randomly.



Mann–Whitney U Test

The Mann–Whitney U test (a counterpart of the t test) is used to
compare the means of two independent populations when the normality
assumption of population is not met or when data are ordinal in nature.
Small-Sample U Test
When n1 (number of items in sample 1) and n2 (number of

items in sample 2) are both less than or equal to 10, samples
are considered to be small.
The test statistic U is the smallest of these two U values.

Mann–Whitney U Test (Contd.)

The null and alternative hypotheses for a two-tailed test can
be stated as below:
Example 18.3: he HR manager of a firm has received a

complaint from the employees of the production department
that their weekly compensation is less than the compensation
of the employees of the marketing department. To verify this
claim, the HR manager has taken a random sample of 8
employees from the production department and 9 employees
from the marketing department. The data collected are shown
in Table 18.2.

Table 18.2 : Weekly compensation of the employees of
the production and marketing departments
Use the Mann–Whitney U test to determine whether the firm

offers different compensation packages to employees of the
production and marketing departments. Take alpha = 0.05
for the test.

Using Minitab and SPSS for Mann–Whitney U

Test
 Solved Examples\MINITAB\Ex 18.3.MPJ

 Solved Examples\SPSS\Ex 18.3.sav
 Solved Examples\SPSS\Output Ex 18.3.spo

U Test for Large Samples

 When n1(number of items in sample 1) and n2 (number of
items in sample 2) are both greater than 10, samples are
considered to be large samples. In case of large samples, the
sampling distribution of the U statistic can be approximated
by the normal distribution.
 The z statistic can be computed by using the following formula

Example 18.4: A manufacturing firm claims that it has improved
the saving pattern of its employees, including employees from the
production and quality control departments through some special
initiatives. The company further claims that it provides equal
compensation opportunities to staff from all departments without
any discrimination. Therefore, the savings pattern of all employees
are the same irrespective of departments. To verify the company’s
claim, an investment expert has taken a random sample of size 15
from the production department and a random sample of size 17
from the quality control department. The investment details of
employees from the production and quality control departments are
given in Table 18.4.

Example 18.4 (Contd.)/ Table 18.4: Investment made by 15
randomly selected employees from the production department and
17 randomly selected employees from the quality control department
Use the Mann–Whitney U test, to determine whether the two populations

differ in saving pattern.

Using Minitab and SPSS for Mann–Whitney U

Test


Wilcoxon Matched-pairs Signed Rank Test

 Wilcoxon test is a non-parametric alternative to the t test
for related samples.
 In the Wilcoxon test, when the sample size (number of
pairs) is less than or equal to 15 (n≤ 15), it is treated as a
small sample and when the sample size (number of pairs)
is greater than 15 (n > 15), it is treated as a large
sample.
 The sum of positive differences is denoted by T+ and the
sum of negative differences is denoted by T–. The
Wilcoxon statistic T is defined as the smallest sum of
ranks. Symbolically, Wilcoxon statistic T = Minimum of
(T+, T–)

Wilcoxon Matched-pairs Signed Rank Test

(Contd.)

Wilcoxon Test for Small Samples (n ≤ 15)

Example 18.5: A company is trying to improve the work
efficiency of its employees. It has organized a special
training programme for all employees. In order to assess
the effectiveness of the training programme, the company
has selected 10 employees randomly and administered a
well-structured questionnaire. The scores obtained by the
employees are given in the Table 18.6.

Example 18.5 (Contd.)

Table 18.6: Scores of 10 randomly selected employees before
and after training
At 95% confidence level, examine whether the training programme has

improved the efficiency of employees.

Example 18.5 (Solution)

Table 18.7: Training scores of employees with differences and ranks
for before and after the training programme

Using Minitab and SPSS for Mann–Whitney

U Test


Wilcoxon Test for Large Samples (n > 15)

In case of a large sample (n > 15), the sampling distribution of
T approaches normal distribution with mean and standard
deviation given as below:

Example 18.6: A software company wants to estimate the change in
expenditure of its employees on children’s education in the last five

years. The monthly expenditure of 17 randomly selected employees
on children’s education in 2000 and 2005 is given in Table 18.8.



KRUSKAL–WALLIS TEST

Kruskal–Wallis test is the nonparametric alternative to
one-way ANOVA.
The null and alternative hypotheses for the Kruskal–Wallis test can be stated as below:

Example 18.7: A travel agency wants to know the amount spent
by employees of four different organizations on foreign travel. The
agency’s researchers have taken random samples from the four
organizations. The amount spent is given in Table 18.10. Use the
Kruskal–Wallis test to determine whether there is a significant
difference between employees of organizations in terms of the
amount spent on foreign travel. Use α = 0.05

Example 18.7 (Solution)

Using Minitab and MS Excel for the Kruskal–Wallis

Test
 Solved Examples\Excel\Ex 18.7.xls


Friedman Test

 The Friedman test is the nonparametric alternative to
randomized block design.
 The null and alternative hypotheses in the Friedman test can
be set as

Example 18.8: A two-wheeler manufacturing company
wants to assess the satisfaction level of customers with its
latest brand as against their satisfaction with four other
leading brands. Researchers at the company have selected
8 customers randomly and asked them to rank their
satisfaction levels on a scale from 1 to 5. The results are
presented in Table 18.12. Determine whether there is any
significant difference between the ranking of brands. Use
alpha = 0.05

Table 18.12: Ranking of satisfaction levels of eight randomly
selected customers


Solved Examples\Excel\Ex 18.8.xls
Solved Examples\MINITAB\Ex18.8.MPJ

Spearman’s Rank Correlation

When data is of ordinal level (ranked data), Pearson
correlation coefficient r cannot be applied. In this case,
Spearman’s rank correlation can be used to determine the
degree of association between two variables.
Example 18.9 : A social science researcher wants to find

out the degree of association between sugar prices and
wheat prices. The researcher has collected data relating to
the price of sugar and wheat in 14 randomly selected
months from the last 20 years. How can he compute the
Spearman’s rank correlation from the data provided in
Table 18.14.
Table 18.14: Sugar and wheat prices for 14 randomly

selected months from the last 20 years


Non-Parametric Statistics: Business Statistics - Naval Bajpai

Uploaded by

Copyright:

Available Formats

Non-Parametric Statistics: Business Statistics - Naval Bajpai

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

Non-Parametric Statistics: Business Statistics - Naval Bajpai

Uploaded by

Copyright:

Available Formats

Copyright© Dorling Kindersley India Pvt Ltd

Business Statistics - Naval Bajpai

Copyright© Dorling Kindersley India Pvt Ltd

 Analyse nominal as well as ordinal level of data

Business Statistics - Naval Bajpai

Copyright© Dorling Kindersley India Pvt Ltd

Business Statistics - Naval Bajpai

Copyright© Dorling Kindersley India Pvt Ltd

Business Statistics - Naval Bajpai

Copyright© Dorling Kindersley India Pvt Ltd

Five runs are as shown

Business Statistics - Naval Bajpai

Copyright© Dorling Kindersley India Pvt Ltd

Business Statistics - Naval Bajpai

Solution : n1 = 12 and n2 = 8. Both n1 and n2 are less than

Business Statistics - Naval Bajpai

Copyright© Dorling Kindersley India Pvt Ltd

The number of runs 5 is less than the critical value of R

Solved Examples\MINITAB\Ex 18.1.MPJ

Business Statistics - Naval Bajpai

Copyright© Dorling Kindersley India Pvt Ltd

Business Statistics - Naval Bajpai

Business Statistics - Naval Bajpai

Copyright© Dorling Kindersley India Pvt Ltd

Business Statistics - Naval Bajpai

Copyright© Dorling Kindersley India Pvt Ltd

When n1 (number of items in sample 1) and n2 (number of

The test statistic U is the smallest of these two U values.

Business Statistics - Naval Bajpai

Copyright© Dorling Kindersley India Pvt Ltd

Example 18.3: he HR manager of a firm has received a

Business Statistics - Naval Bajpai

Use the Mann–Whitney U test to determine whether the firm

Copyright© Dorling Kindersley India Pvt Ltd

Copyright© Dorling Kindersley India Pvt Ltd

 Solved Examples\MINITAB\Ex 18.3.MPJ

Business Statistics - Naval Bajpai

Copyright© Dorling Kindersley India Pvt Ltd

Business Statistics - Naval Bajpai

Business Statistics - Naval Bajpai

Use the Mann–Whitney U test, to determine whether the two populations

Copyright© Dorling Kindersley India Pvt Ltd

Copyright© Dorling Kindersley India Pvt Ltd

 Solved Examples\MINITAB\Ex 18.4.MPJ

Business Statistics - Naval Bajpai

Copyright© Dorling Kindersley India Pvt Ltd

Business Statistics - Naval Bajpai

Copyright© Dorling Kindersley India Pvt Ltd

Business Statistics - Naval Bajpai

Copyright© Dorling Kindersley India Pvt Ltd

Business Statistics - Naval Bajpai

Copyright© Dorling Kindersley India Pvt Ltd

At 95% confidence level, examine whether the training programme has

Business Statistics - Naval Bajpai

Copyright© Dorling Kindersley India Pvt Ltd

Business Statistics - Naval Bajpai

Copyright© Dorling Kindersley India Pvt Ltd

 Solved Examples\MINITAB\Ex 18.5.MPJ

Business Statistics - Naval Bajpai