STAT 353: NON-PARAMETRIC STATISTICS

(2,2,3)

S.K APPIAH (PhD)

March 9, 2022

Email: [email protected]

Office: SCB-TF14

Department of Statistics and Actuarial Science

Tel : +233205279926. . . . . . . . . . . . . . . . . . . .
Lecture 1

Learning Objectives
Compare and contrast parametric and non-parametric tests
Identify multiple applications where non parametric approaches are
appropriate
Perform and interpret the Mann Whitney U Test
Perform and interpret the Sign test and Wilcoxon Signed Rank Test
Compare and contrast the Sign test and Wilcoxon Signed Rank Test
Perform and interpret the Kruskal Wallis test
Identify the appropriate non-parametric hypothesis testing procedure
based on type of outcome variable and number of samples.

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 2 / 69
Outline

1. Introduction;
Basic Concept
Comparison of parametric and non-parametric statistics
2. Chi-square Tests
Test for goodness-of-fit
Test for independence/homogeneity
3. Test for randomness
4. Tests for other independent samples
5. Tests for paired samples
6. Efficiency of non parametric tests
7. Mini-Project

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 3 / 69
Introduction

1. Introduction
Statistical methods of inference involving parameters such as mean,
proportion and variance are called parametric or classical methods
These methods are useful when dealing with quantitative data (i.e.
data measured on interval or ratio scales)
Parametric methods (t- and F- tests) are used for comparing
populations are based on the following assumptions.
Sample(s) are drawn are independent and identically distributed (iid).
The sampled populations have equal variances
Sampled populations are normally distributed.

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 4 / 69
 The t-test and F- test used for comparing two or more populations
become unsuitable if these assumptions are not satisfied.
We have another class of statistical methods (non-parametric tests) which
are used to clear doubts about:
The population distributions and to avoid the problem of committing
Type II error; and
Accuracy of the results as some of the observations may be wrongly
recorded, leading to different p-value.

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 5 / 69
Non-parametric tests are designed for qualitative or categorical data
which require fewer or no assumption about the sampled
population(s).
One of the advantages is that the data need not be quantitative but
can be categorical (such as yes or no, defective or non-defective) or
rank data.
Another advantage is that non parametric procedures are usually very
quick and easy to perform.

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 6 / 69
 Many business applications involve opinions or feelings which are
qualitative data where t- and F- tests cannot be applied. For example
comparing the managerial ability of two executives based on
subjective evaluations of trained observers
preferences for a product in a market survey
and evaluating and ranking the instructional abilities of teachers.

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 7 / 69
 A non-parametric method must therefore satisfy at least one of the
following conditions:
The data collected are measured on nominal or ordinal scale.
The data collected are measured on interval or ratio scale where no
assumption can be made about the population probability distribution.
In many cases where a parametric method as well as a non-parametric
method can be applied, the latter is almost as good as or powerful as the
former.

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 8 / 69
Non-parametric methods:
Chi-square tests
Tests for Independent samples (Runs test, Mann-Whitney
U-test,Kruskal-Wallis and Friedman Tests, median test)
Tests for Paired Samples (Sign test, Wilcoxon signed-rank test and
Spearman rank correlation coefficient test)

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 9 / 69
2.0 CHI-SQUARE TESTS
Many types of surveys or experiments result in qualitative data which
can be classified. Data from these experiments are called categorical,
enumerative or count data.
For examples we may classify:
People into income classes: low, middle and high;
Smoking habits:non, former and current smokers.
Recovery from a surgical operation as: well-recovered, recovering and
not recovering.
Consumers reaction in a market survey: satisfied and not satisfied
Industrial process manufactures items that fall into one of the
different quality classes: defective and not defective.

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 10 / 69
Chi-Square tests are designed for the analysis of categorical or enumerative
data. They are:
Test for Goodness-of-fit and
Test for Independence/Homogeneity
Tests for Goodness-of-Fit

The test is performed to determine whether a given categorical data

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 11 / 69
 Hypotheses: The null and alternative hypothesis are given by:
H0 : The sample data drawn can be modelled by a specified
probability distribution, against
H1 : The sample data cannot be modelled by the specified
distribution, at a significance level of α
Test-statistic:

∑
k
(fi − ei )2
x2 =
i=1
ei

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 12 / 69
which is compared with the Chi-square critical value χ2α(df ) where
fi = observed frequency for category i
ei = the expected frequency for category i which should obey the
rule of 5.
df = degrees of freedom = k − 1 − c,
where k is the number of categories and c, the number of unknown
parameters to be estimated.

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 13 / 69
Decision Rule:
Reject H0 if χ2 ≥ χ2α(df )
which means that there is lack of good fit.
Remark (Rule of 5):
This is a rule of thumb, that, X 2 which is discrete has approximately
chi-square distribution which is continuous if n is large and
ei = npi ≥ 5 .

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 14 / 69
Example 1
A die was rolled 600 times with the following results:
Observed(x) 1 2 3 4 5 6
Frequency(f) 90 110 100 105 115 80
Do these results present sufficient evidence that the die is unbiased? Test
using α = 0.05
Solution
H0 : The die is unbiased, that is, the relative frequency for the
1
observed (x) are the same i.e. (p1 = p2 = ... = p6 = )
6
1
H1 : The die is biased, that is, at least one pi ̸= , i = 1, 2, ..., 6 at
6
significance level, α = 0.05.

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 15 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 16 / 69
The test-statistic is computed by:

∑
6
(fi − ei )2
X2 =
i=1
ei

(90 − 100)2 (110 − 100)2 (80 − 100)2

= 8.50

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 17 / 69
ii. The degrees of freedom,

df = k − 1 − c = 6 − 1 − 0 = 5

,
and the critical value at α = 0.05

χ2α(df ) = χ20.05(5) = 11.0705

iv. Test conclusion: The null hypothesis is failed to be rejected since

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 18 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 19 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 20 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 21 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 22 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 23 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 24 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 25 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 26 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 27 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 28 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 29 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 30 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 31 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 32 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 33 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 34 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 35 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 36 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 37 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 38 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 39 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 40 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 41 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 42 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 43 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 44 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 45 / 69
Test for Homogeneity: Results when the column (or row) totals of a
contingency table are predetermined (or said to be fixed), leaving the
other totals to become random variables (i.e. not under the control of
investigator).
Chi-square test for homogeneity is concerned with the question: Are
the samples drawn from homogeneous populations with respect to a
criteria of classification?
The procedure for this test is the same as the test for independence
despite the differences in concept and sampling procedure.

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 46 / 69
Example 1 There is reason to believe that high income families usually
send their children to private schools and low income families often send
their children to public schools. To verify this, 1600 families were selected
at random from a community and following results in the table below
obtained:

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 47 / 69
 The classifications are income and type of school. At a glance, we see
that relatively a greater percentage of high income families (73%),
send their children to private schools than do low income families
(51%). But is such a difference in proportion significant or it is due to
chance?
The chi-square test for independence can be used to answer this
question.
Example 2
The personal manager of a firm attempted to determine whether an
employees educational level had an effect on his/her job performance. To
achieve this an examination was given to a sample of 120 employees.
Instead of using the examination scores, the manager rather rated each
employee performance as high, average or low. The results of the study
are as follows

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 48 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 49 / 69
Does job performance as measured by the examination appear to be
related to an employee educational level? Test, taking α = 0.05
Solution
(i) H0 : an employees job performance is independent of his/her
educational level
H1 : the two classifications (educational level and job performance) are
dependent, at α = 0.05

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 50 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 51 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 52 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 53 / 69
df = (r − 1)(c − 1) = (3 − 1)(3 − 1) = 4 and the critical value at
α = 0.05 is χ20.05(4) = 9.48773
Conclusion: We fail to reject H0 since χ2 < χ20.05(4) and conclude
that there is no evidence of relationship between job performance as
measured by examination and level of education. Thus job
performance is independent of the educational levels of employees.

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 54 / 69
Example 3
A manufacturing company of electronic components has plants in four
cities A, B, C, and D. Over the years the company has gone to great
lengths to discourage the formation of labour unions at these plants
including constructing employee recreational canters and offering better
than average employee benefits. The management, however, suspects that
there is growing interest among the employees in forming a union, and of
particular interest in the union differs among the four plants. The plants
at cities A and D are considerably larger than the other two so the
company obtains a random sample of 200 employees from each of these
two larger plants and 100 from each of the smaller two plants. The results
for the survey are as follows;

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 55 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 56 / 69
(a) Copy and complete the table and test to determine whether employees
interest in labour union is the same in each of the four cities at
(b) What other information can you also deduce from the above test?

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 57 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 58 / 69
This is a chi-square test for homogeneity, and so we proceed as follows:
H0 : The employees interest in labour union is the same in each of
the four cities (populations), against
H1 : The four populations are not homogeneous with respect to the
interest in labour union formation, at α = 0.01

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 59 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 60 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 61 / 69
Conclusion: H0 is rejected since χ2 > χ20.01(6) and this means that
the employees interest in labour union formation differs among the
cities where the plants are sited.
We deduce from the test in (a) that the interest of employees in
labour union depends on where the plants are sited.

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 62 / 69
 . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 63 / 69
May also be given by:
n(ad − bc)2
X2 =
(a + c)(b + d)(a + b)(c + d)

Which has the chi-square distribution,χ2α(1)

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 64 / 69
Self Assessment Questions(Exercise 1)
1. The surrendering rate insurance product is of much concern to insurers.
A random sample 90 live insurance policyholders of an insurance company
is classified according to their intention to surrender and gender, which the
results in the table below. Use a 0.01 level of significance and test the
hypothesis that the intention of policyholders surrendering is independent
of whether the holder is a male or female.

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 65 / 69
2. A study of amount of violence viewed on TV as it relates to the age of
viewer showed the accompanying result for 81 people. Each person in the
study could be classified, according to the persons TV viewing habits, as a
low-violence or high-violence viewer. This results in the table below. Do
the data in the table indicate some form of association between viewing of
violence and age of the age of viewer?

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 66 / 69
3. A study is conducted to assess the effectiveness of a new computerized
system of filling orders in a particular industry. Random samples of 100
customers served via the old system and 100 served via the new system are
selected. Each customer is contacted to determine whether or not the
order was filled satisfactory within two weeks. The table below gives the
results of the study.

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 67 / 69
Test the null hypothesis that the proportion of satisfied customers among
those served by the new system is the same as that among those served by
the old system, at α = 0.05

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 68 / 69
 Thank You

. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . .
S.K APPIAH (PhD) STAT 353: NON-PARAMETRIC STATISTICS March 9, 2022 69 / 69