Test of

Significance of
Difference
Between More
Than Two Means

BLOCK 3
NON-PARAMETRIC STATISTICS

83
Parametric
Statistics

84
UNIT 5 CHI-SQUARE* Chi-Square

Structure
5.0 Objectives
5.1 Introduction
5.2 What is Ch- square?
5.2.1 Assumptions for the Application of χ2Test
5.2.2 χ2Distribution
5.3 Properties of Chi-square Distribution
5.4 Applications of Chi-square Test
5.4.1 Test of Goodness of Fit
5.4.2 Test of Independence
5.4.3 Test of Homogeneity
5.5 Precautions about Using the Chi-square Test
5.6 Let Us Sum Up
5.7 References
5.8 Key Words
5.9 Answers to Check Your Progress
5.10 Unit End Questions

5.0 OBJECTIVES
After reading this unit, you will be able to,

 explain what is chi-square;

 describe for the application of χ2 test;
 explain chi-square distribution;
 discuss the properties of chi-square distribution;
 explain the application of chi-square; and
 discuss the precautions about using the chi-square test.

5.1 INTRODUCTION
In psychology some times the researcher is interested in such questions as to
whether one type of soft drink preferred by the consumer is in any way
influenced by the age of the consumer, and whether? Similarly among the
four colours, red, green, indigo, yellow, is there any significant difference in
the participants in regard to the preference of colours, etc.

*
Dr. Vivek Belhekar, Faculty, Department of Psychology, Bombay University, Mumbai
85
Non-Parametric
Statistics
To approach questions like these, we find the number of persons who
preferred particular colour. Here we have the data in the form of frequencies,
the number of cases falling into each of the four categories of colours. Here
the researcher compares the observed frequencies characterising the choices
of the 4 categories by a large number of individuals. Some prefer red, some
white, some green and some yellow etc. Hypothetically speaking one may
say that all colours are equally preferred and are equal choice for the
individuals. But in reality there may be more number of persons choosing red
or blue than yellow and similarly quite a few may choose white etc. Thus
what is expected as choice which is equal for all colours may be seen
differently when actual choice is made by the persons. Now we note the
number of persons who had chosen red, blue, yellow and white respectively
and note it in the table. Hypothetically we said that there will be the same
number of persons choosing each of the 4 colours. This also we note in the
table. Now we compare and see for each colour what is the difference
between the actual number of choices made as compared to the hypothetical
equal number. The differences thereof are noted. To find out if these
differences are really of statistical significance amongst the persons in regard
to choice of colours, we use a statistical technique called Chi-square. In the
present unit, we will explain what is chi-square and also describe the
assumptions for its application.

5.2 WHAT IS CHI- SQUARE?

This is one of the most important non-parametric statistics. As we see further
that chi-square is used for several purposes.

The term ‗chi‘ (is the Greek letter, it is pronounced ki). The chi-square test
was originally developed by Karl Pearson in 1900 and is sometimes called
the Pearson Chi-square. According to Garrett (1981), the difference between
observed and expected frequencies are squared and divided by the expected
number in each case and the sum of these quotients is chi-square‖. According
to Guilford(1973) ―By definition a χ2 is the sum of ratio (any number can be
summed), each ratio is that between a squared discrepancy of difference and
an expected frequency‖.

On the basis of the above definitions it can be said that the discrepancy
between observed and expected frequencies is expressed in term of a
statistics named chi-square (χ2).

5.2.1 Assumptions for the Application of χ2Test

Before using chi-square as a test statistic to test a hypothesis, the following
assumptions are necessary:

 Data should be in the form of frequencies. Although the chi-square test is

conducted in terms of frequencies or data that can be readily transformed
into frequency, it is best viewed conceptually as a test about proportions.
86
 The chi-square test is applied only to discrete data. However, any Chi-Square
continuous data can be reduced to the categories in such a way that they
can be treated as discrete data and then the application of chi-square is
justified.
 The χ2 is applied when there is quantitative data.
 The observation should be independent. For example let us say that we
wanted to find out the colour preference of 100 females. We select 100
subjects randomly, the preference of one individual is not predictable
from that of any other. If there is a relationship between two variables or
the participants are matched, then chi-square cannot be used to test.
Correlated data like matched pair are not subjected to chi-square
treatment.
 The sample drawn from the population about which inference is to be
made should be random.
 If there are only two cells, the expected frequencies in each cell should
be 5 or more. Because for observation less than 5, the value of χ2 shall
be over estimated, resulting in the rejection of the null hypothesis.
 For more than two cells, no more than 20% of the expected values may
be lesser than 5 and no expected value may be lesser than 1. If more than
20 percent of the cells have expectedfrequencies less than 5 then χ2
should not be applied.
 A sample with a sufficiently large size is assumed. If sample size is small
then χ2 will yield an inaccurate inference. In general, larger the sample
size, the less affected by chance is the observed distribution, and thus the
more reliable the test of the hypothesis.
 The data should be expressed in original units, rather than in percentage
or ratio form. Such precaution helps in comparison of attributes of
interest.

5.2.2 χ2 Distribution

The term non-parametric does not mean that the population distribution under
study has no parameters. All populations have certain parameters which
define their distribution. The sampling distribution of χ2 is called χ2
distribution. Other hypothesis testing procedures, the calculated value of χ2
test statistic is compared with its critical (or table) value to know whether the
null hypothesis is true. The decision of accepting a null hypothesis is based
on how 'close' the sample results are to the expected results. If the null
hypothesis is true, the observed frequencies will vary from their
corresponding expected frequencies according to the influence of random
sampling variation. The calculated value of χ2 will be smaller when
agreement between observed frequencies and expected frequencies are
smaller. The shape of a particular chi-square distribution depends on the
number of degrees of freedom. Let us now see what is degrees of freedom?
Let us say there are 5 scores, as for example, 25,35,45,55,65

87
Non-Parametric
Statistics
The mean here is 45. Let us say that from this mean of 45 we try to see the
difference in each of the other scores. While all the other 4 scores are taken
for the difference, this 45 is not available for other calculations, as from this
Mena=45, all other calculations are made. From this mean we note the
differences in the remaining scores, and thus amongst the 5 scores one score
is not movable or in other words the group of 5 subjects has lost one degree
of freedom. Suppose we go for higher level calculations, then accordingly
there will be more number of items that cannot be moved and thus we may
have greater degrees of freedom. In a 3 × 3 table, we will be using one cell
from the row and one cell from the column for the purpose of chi-square
calculations. Thus the degrees of freedom will be (3 – 1) (3 – 1) = 4. That is
(r- 1) ( k-1 ) (r = row and k = column).

It may also be noted that chi-square assumes positive values only. As such a
chi-square curve starting from the original (zero point) lies entirely to the
right of the vertical axis.

If we draw a curve with degrees of freedom on the x axis and chi-square

values on the y axis, the he shape of a chi-square distribution curve will be
skewed for very small degree of freedom. As the degrees of freedom
increase, the shape also changes. Eventually, for larger degrees of freedom,
the curve looks similar to the curve of a normal distribution. A point worth
nothing is that the total area under a chi-square distribution curve is 1.0 as is
the case in all other continuous distribution curves.

Check Your Progress I

1) State any one assumption for the application of χ2 Test.

……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………

5.3 PROPERTIES OF CHI-SQUARE

DISTRIBUTION
The following properties of chi-square test statistic are to be kept in mind
when we analyse its sampling distribution.

 Chi-square is non-negative in value; it positively valued, because all

discrepancies are squared, both positive and negative discrepancies make
a positive contribution to the value of χ2.
 Chi-square will be zero only in the unusual event that each observed
frequency exactly equals the corresponding expected frequency.

88
 It is not symmetrical; it is skewed to the right as mentioned earlier. The Chi-Square
reason for this is because all the values are positive and hence so instead
of having a symmetrical distribution, we have a distribution all on the
positive side.
 Since density function of χ2 does not contain any parameter of
population, Chi-square test statistic is referred to as a non-parametric
test. Thus chi- square distribution does not depend upon the form of the
parent population.
 There are many chi-square distributions. As with the t-distribution, there
is a different chi-square distribution for the value of each degree of
freedom. If we know the degrees of freedom and the area in the right tail
of a chi-square distribution, we can find the value of chi-square from the
table of chi-square. We give below two examples to show how the value
of chi-square can be obtained from this table.

Example 1: Find the value of chi-square for 10 degree of freedom and an

area of 0.05 in the right tail of the chi-square distribution curve.

Solution: In order to find the required value of chi-square, we first locate 10

in the column for degree of freedom (df) and 0.05 in the top row in table of
chi-square. The required chi-square value is given by the entry at the
intersection of the row for 10 and the column for 0.05. This value is 18.307.

Example 2: Find the value of chi-square for 20 degree of freedom and an

area of 0.10 in the left tail of the chi-square distribution curve.

Solution: This example is different from the above one. Here, the area is in
the left tail of the chi-square distribution curve is given. In such a case, we
have to first find the area in the right tail. This is obtained as follows:

Area in the right tail= 1 - area in the left tail

Area in the right tail = 1 - .10 = .90

Now we can find, in the same manner as used earlier, the value ofχ2. We
locate 20 in the column for df and .90 on top row in the table of chi-square.
The required value of χ2 is 12.443.

Check Your Progress II

1) State any one property of chi-square test statistic are to be kept in mind
when we analyse its sampling distribution.
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………

89
Non-Parametric
Statistics 5. 4 APPLICATIONS OF CHI-SQUARE TEST
Chi-square is used for categorical data that is for data comprising unordered
quantitative categories such as colours, political affiliation etc. The important
applications of χ2test are as follows:

 Test of goodness of fit

 Test of independence
 Test for homogeneity

5.4.1 Test of Goodness of Fit

On several occasion a decision - maker needs to understand whether an actual
sample distribution matches with a known theoretical distribution such as
binomial, poisson, normal and so on. Similarly some time researcher
compares the observed (sample) frequencies characterising the several
categories of the distribution with those frequencies expected according to his
or her hypothesis.

The χ2 test for goodness of fit is a statistical test of how well given data
support an assumption about the distribution of a population. That is, whether
it supports a random variable of interest. In other words, the test determines
how well an assumed distribution fits the given data.

To apply this test, a particular theoretical distribution is first hypothesised for

given population and then the test is carried out to determine whether or not
the sample data could have come from the population of interest with the
hypothesised theoretical distribution. The observed frequencies come from
the observation of sample and the expected frequencies come from the
hypothesised theoretical distribution. The goodness of fit test focusses on the
difference between the observed frequencies and expected frequencies.

This can be explained with the help of following examples.

In a particular study, a researcher wants to study that among four brands of

glucose biscuits, is there a difference in the proportion of consumers who
prefer the taste of each. We formulate the null hypothesis that there is no
differential preference among consumers for the four brand of glucose
biscuits. Here we test the hypothesis that there are equal probability of
individuals preferring each brand i.e. 1/4 = .25. Or if we take 100 persons to
test this hypothesis, 25 each will prefer theoretically the 4 brands of biscuits.

The alternative hypothesis is that a preference exists for the first brand of the
biscuits and the reminder are equally less preferred, or that the first two are
preferred more than the second two and so on.

The use of chi-square tells us whether the relative frequencies observed in the
several categories of our sample frequency distribution are in accordance
with the set of frequencies hypothesised.
90
To test the null hypothesis in the above study, we might allow subjects to Chi-Square

taste each of the four brand of glucose biscuits and then find their preference.
We control possible extraneous influence, such as knowledge of brands name
and order of presentation. Suppose we had as mentioned earlier randomly
selected 100 subjects and that our observed frequencies of preference are as
given in the table below:

Brand A Brand B Brand C Brand D

Observed 20 18 30 32
Frequencies

We calculate the expected frequency for each category by multiplying the

proportion hypothesised to characterise that category in the population by the
sample size. According to null hypothesis, the expected proportionate
preference for each biscuit is 1/4 and the frequencies of preference for each is
therefore 25 (1/4) (100/4) = 25.

In any experiment, we anticipate that the observed frequency of choice will

vary from the expected frequencies. We do not use these differences directly.
One reason is that some differences are positive and some are negative. Thus
they would cancel each other out. To get around this, we square each
difference. But how much variation to expect? Some measures of discrepancy
is required to test the null hypothesis, this can be tested by chi-square. We
calculate the chi-square (the method of computation of chi-square will be
explained in the next unit). If the obtained chi-square value is more than the
critical values at 0.05 or 0.01 level than we reject the null hypothesis and if
obtained value of chi-square is less than the given values then we retain the
null hypothesis.

The obtained value of chi-square depends on the size of the discrepancy

between observed frequency and expected frequencies. Larger the differences
between observed frequencies and expected frequencies the larger will be the
chi-square.

The size of the discrepancy relative to the magnitude of the expected

frequency contribute in the determination of the value of chi-square. For
example, if we toss a number of coins and find out how many times theheads ‘
ortails ‘have appeared. To take a numerical example, when we toss coins in
12 times and 1000 times. Then according to null hypothesis our expected
frequency will be 6, and 500 respectively. If we obtain 11 heads in 12 tosses
the discrepancy is 5. However, if we obtain 505 heads in 1000 tosses, the
discrepancy is also 5.

The value of chi-square depends on the number of discrepancies involved in

its calculation. For example if we use three brands of biscuits not the four,
there would be less discrepancy to contribute to the total of chi-square. This
will influence the degrees of freedom, that is, when the number of 91
Non-Parametric
Statistics
discrepancies are 4 the degrees of freedom will be 3 and when number of
discrepancies are 3 the degree of freedom will be 2. When degrees of
freedom is 4 and obtained chi-square is more than 9.48 then we reject the null
hypothesis at 0.05 level. On the other hand when the degrees of freedom is 4
and obtained chi-square is 7.81 then we retain the null hypothesis at 0.05
level.

5.4.2 Test of Independence

In the above discussion so far, we have considered the application of chi-
square only to ‗one‘ variable case. We have considered testing whether the
categories in an observed frequency distribution differ significantly from one
another. The chi-square test has a much broader use in social science
research; to test whether one observed frequency distribution significantly
differ from another observed frequency. In other words, it can be said that
chi-square is used for the analysis of bivariate frequency.

For example, social researcher is interested in surveying the attitudes of high

school students concerning the importance of getting a college degree. She
questions a sample of 60 senior high school students about whether they
believe that college education is becoming more important, less important or
staying the same and whether male students respond differently from female
students ?

As we know that nominal and ordinal variables are generally presented in the
form of a cross-tabulation. Specifically cross tabulation are used to compare
the distribution of one variable often called dependent variable, across
categories of some other variable the independent variable. In a cross
tabulation the focus is on the difference between group — such as between
males and females — in terms of the dependent variable — for example,
opinion about the changing value of a college education. In the above
example we want to study whether there exists gender differences in belief
regarding the importance of a college degree - are statistically significant.

To study this question, we classify the data (observed frequencies) in a

bivariate distribution. For each variable, the categories are mutually
exclusive. The data is classified in the following table:

More Important Less Important About the Same

Male 25 6 8

Female 10 4 7

Bivariate frequency distribution of two types are known as contingency

tables. From such a table we may inquire what cell frequencies would be
expected if the two are independent of each other in the population. The chi-
square test may be used to compare the observed cell frequencies with those
92
expected under the null hypothesis of independence. If the (fo-fe) Chi-Square

discrepancy are small, chi-square will be small suggesting that the two
variable could be not different. Conversely, a large chi-square points toward a
contingency relationship.

As in the case of the t ratio there is a sampling distribution for chi-square that
can be used to estimate the probability of obtaining a significant chi-square
value by chance alone rather than by actual population difference. The test is
used to study the significance of difference between mean. On the other hand,
chi-square is used to make comparison between frequencies rather than mean
scores. As a result the null hypothesis for the chi-square test states that the
populations do not differ with respect to frequencies of occurrence of a given
characteristic. In general, the null hypothesis of independence for a
contingency table is equivalent to hypothesising that in the population the
relative frequencies for any row (across the categories of the column
variable) are the same for all rows, or that in the population the relative
frequencies for any column (across the categories of the two variables) are
the same for all columns.

If the null hypothesis of independence is true at the population level, we

should expect that random sampling will produce obtained value of chi-
square that are in accord with the tabulated distribution of that statistic. If the
hypothesis is false in any way, the obtained value of chi-square will tend to
be larger than when the alternate hypothesis, stating that there will be a
significant difference between the variables, is true.

In the case of a 2 × 2 contingency table, we can test for a difference between

two frequencies or proportions from independent samples (just as with the 1
× 2 table, where we can test a hypothesis about a single proportion.

5.4.3 Test of Homogeneity

The test of homogeneity is useful in a case when we are interested in
verifying, whether several populations are homogeneous with respect to some
characteristic of interest. For example a movie producer is bringing out a new
movie. In order to develop an advertising strategy he/she wants to determine
whether the movie will appeal most to a particular age group or whether it
will appeal equally to all age groups. We formulate the null hypothesis that
the opinion of all age groups is same about the new movie. Hence, the test of
homogeneity is useful in testing a null hypothesis that several populations are
homogeneous with respect to a character. This test is different from the test
of independence on account of the following reasons:

1) Instead of knowing whether two attributes are independent or not, we

may like to know whether different samples come from the same
population.
2) Instead of taking only one sample for this test two or more independent
samples are drawn from each population.
93
Non-Parametric
Statistics
3) To apply this test first a random sample is drawn from each population,
and then in each sample the proportion falling in each category is
determined. The sample data so obtained is arranged in contingency
table. The procedure for testing of hypothesis is same as for test of
independence.

Check Your Progress III

1) List the the important application of χ2 test.

……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………

5.5 PRECAUTIONS ABOUT USING THE CHI-

SQUARE TEST
In order to use a chi-square test properly, one has to be extremely careful and
keep in mind certain precautions:

1) A sample size should be large enough. If the expected frequencies are

too small, the value of chi-square gets over estimated. To overcome this
problem we must ensure that the observed frequency in any cell of the
contingency table should not be less than 5.
2) When the calculated value of chi-square turns out to be more than the
critical or theoretical value at a predetermined level of significance, we
reject the null hypothesis. In contrast, when the chi-square value is less
than the critical theoretical value, the null hypothesis is not rejected.
However, when the chi-square value turns out to be zero, we have to be
extremely careful to confirm that there is no difference, between the
observed and expected frequencies. Such a situation sometimes arises on
account of faulty method used in the collection of data.

Check Your Progress IV

1) Fill in the blanks

a) When the expected frequencies are too small, the value of chi-square
gets _____________.
b) When the chi-square value is less than the critical theoretical value, the
null hypothesis is ________________________.

94
 Chi-Square
5.6 LET US SUM UP
To summarise, the chi-square test is the most commonly used non-parametric
test. Non parametric techniques are used when a researcher works with
ordinal or nominal data or with a small number of cases representing a highly
asymmetrical underlying distribution. The chi-square test is used as a test of
significance when we have data that are given in frequencies or categories.
The chi-square test is used for two broad purposes. Firstly it is used as a test
of goodness of fit and secondly as a test of independence. As a test of
goodness of fit chi-square tells us whether the frequencies observed among
the categories of a variable are tested to determine whether they differ from a
set of expected hypothetical frequencies. As a test of independence chi-
square is usually applied for testing the relationship between two variables in
two way. First by testing the null hypothesis of independence and second by
computing the value of contingency coefficient, a measurement of
relationship existing between the two variable. When we have less than 5
observed frequencies in any cell then we use the Yate's correction.

5.7 REFERENCES
Kerlinger, Fred N. (1963). Foundation of Behavioural Research, (2nd Indian
reprint) Surjeet Publication, New Delhi.

Garrett, H.E. (1971), Statistics in Psychology & Education, Bombay, Vakils,

Seffer & Simoss Ltd.

Guilford, J.P. (1973), Fundamental Statistics in Psychology & Education,

Newyork, McGraw Hill.

5.8 KEY WORDS

Chi-square distribution: A distribution with degree of freedom as the only
parameter. It is skewed to the right for small degree of freedom.

Chi-square test: A statistical techniques used to test significance in the

analysis of frequency distribution.

Contingency table: A table having rows and column wherein each row
corresponds to a level of one variable and each column to a level of another
variable.

Expected frequencies: The frequencies for different categories which are

expected to occur on the assumption that the given hypothesis is true.

Observed frequencies: The frequencies actually obtained from the

performance of an experiment.

95
Non-Parametric
Statistics 5. 9 ANSWERS TO CHECK YOUR PROGRESS
Check Your Progress I

1) State any one assumption for the application of χ2 test.

In χ2 test, data should be in the form of frequencies. Although the chi-
square test is conducted in terms of frequencies or data that can be
readily transformed into frequency, it is best viewed conceptually as a
test about proportions.

Check Your Progress II

1) State any one property of chi-square test statistic are to be kept in mind
when we analyse its sampling distribution.
Chi-square is non-negative in value; it positively valued, because all
discrepancies are squared, both positive and negative discrepancies make
a positive contribution to the value of χ2.

Check Your Progress III

1) List the the important application of χ2 test.

The important application of χ2 test are test of goodness of fit, test of
independence and test for homogeneity.

Check Your Progress IV

1) Fill in the blanks

a) When the expected frequencies are too small, the value of chi-square
gets over estimated.
b) When the chi-square value is less than the critical theoretical value, the
null hypothesis is not rejected.

5.10 UNIT END QUESTIONS

1) Describe the properties of chi-square test.
2) What do you understand by the goodness of fit test ? Describe with
examples.
3) What is the chi-square test for independence? Explain with examples.
4) What precautions would you keep in mind while using chi-square.

96
UNIT 6 COMPUTATION OF CHI-SQUARE* Computation of
Chi-Square

Structure
6.0 Objectives
6.1 Introduction
6.2 Computation of Chi- square Test
6.2.1 Chi-square as a Test of Goodness of Fit
6.2.2 Testing Hypothesis of Equal Probability
6.2.3 Chi-square as a Test of Independence
6.2.4 2 × 2 Fold Contingency Table
6.3 Chi-square Test when Table Entries are Small (Yates Correction)
6.4 Let Us Sum Up
6.5 References
6.6 Key Words
6.7 Answers to Check Your Progress
6.8 Unit End Questions

6.0 OBJECTIVES
After reading this unit, you will be able to,
 identify chi-square as a test of goodness of fit;
 apply testing of equal probability hypothesis;
 use testing of normal probability hypothesis;
 calculate chi-square as a test of independence in contingency tables; and
 apply chi-square test when table entries are small.

6.1 INTRODUCTION
As we have discussed in last unit that chi-square is a commonly used non-
parametric statistical test. This is used when we have data in the form of
frequencies, ordinal or nominal levels of measurement. Chi-square test is
used for different purposes. In the last unit you were given the conceptual
knowledge of chi-square test and in this unit we will discuss the method of
computation of chi-square.

6.2 COMPUTATION OF CHI- SQAURE

The chi-square test is used for comparing experimentally obtained results
with those to beexpected. The computations of chi-square are discussed as
follows:

*
Dr. Vivek Belhekar, Faculty, Department of Psychology, Bombay University, Mumbai
97
Non-Parametric
Statistics
6.2.1 Chi-square as a Test of Goodness of Fit
Chi-Square goodness of fit test is used to find out how the observed value of
a given phenomena is significantly different from the expected value. In Chi-
Square goodness of fit test, the term goodness of fit is used in order to
compare the observed sample distribution with the expected probability
distribution. Chi-Square goodness of fit test determines how well theoretical
distribution (such as normal, binomial, or Poisson) fits the empirical
distribution. In Chi-Square goodness of fit test, sample data is divided into
intervals. Then the numbers of points that fall into the interval are compared,
with the expected numbers of points in each interval.

In regard to the procedure, set up the hypothesis for chi-square goodness of

fit test, that is set up both null and alternative hypothesis.

a) Null hypothesis: In chi-square, goodness of fit test, the null hypothesis

assumes that there is no significant difference between the observed and
the expected value.
b) Alternative hypothesis: In chi-square, goodness of fittest, the
alternative hypothesis assumes that there is a significant difference
between the observed and the expected value.
Calculate chi-square using the formula and find out for the given degrees
of freedom if chi-square value is significant at 0.05 or 0.01 levels. If so
reject the null hypothesis. If not significant accept the null hypothesis.

6.2.2 Testing Hypothesis of Equal Probability

The Chi-square test is a useful method of comparing experimentally obtained
results with those to be expected theoretically on some hypothesis. The
formula for calculating χ2 is

χ2 =∑[(fo- fe)2]/fe

Where;

fo= observed frequency of a phenomenon oreven which the experimenter is

studying.

fe = expected frequency of the same phenomenon based on “no difference” or

“null” hypotheses.

The use of the above formula can be illustrated by the following example.

Example: An attitude scale designed to measure attitude towards co-

education was administered on 240 students. The data is given in table 6.1.
The observed data is (fo) given in the first row of table 6.1. In the second row
is the distribution of answer to be expected on the basis of null hypothesis
(fe), if each answer is selected equally.

98
 Computation of
Table 6.1: Responses from participants regarding Attitude Chi-Square

Calculations Favourable Neutral Unfavourable Total

fo 70 50 120 240

fe 80 80 80 240

fo- fe -10 -30 40

(fo- fe)2 100 900 1600

(fo- fe)2/fe 100/80 900/80 1600/80

χ2 1.25 11.25 20 2
∑(fo- fe) /fe
= 32.50

The formula of χ2 is
χ2 =∑[(fo- fe)2]/fe

= 1.25 + 11.25 +20

= 32.5

df=(r-1)(c-1)
=(3-1)(2-1)

df = 2

Critical value of χ2 for df= 2 at 0.05 level = 5.59 (refer to statistical table
given at the appendix of Statistics book). Critical value of χ2 at 0.01 level =
9.19 (refer to the table for chi-square given in the appendix of textbook on
statistics) .

The computed value of χ2 , i.e. 32.5 is above the critical values at 0.05 and
0.01significance levels, we conclude that χ2 is significant and we retain the
null hypothesis.

On the other hand if the computed value of χ2 is less than 5.59 or 9.19, then
the null hypothesis is accepted and it may be concluded that there is no
significant difference in the attitude of the participants.

However in our example we have found the chi-square value being greater
than what is given in the table and so we reject the null hypothesis.

Steps for chi-square testing:

1) First set a null of hypothesis.

2) Collect the data and findout observed frequency.
99
Non-Parametric
Statistics
3) Find out the expected frequencies by adding all the observed frequencies
st
divided by number of categories (In I example 240/3=80, in II example
100/5=20).
4) Find out the difference between observed frequencies and expected
frequencies (fo- fe).
5) Find out the square of the difference between observed and expected
frequency (fo- fe)2.
6) Divide it by expected frequency, (fo- fe)2/fe You will get aquotient.
7) Find out the sum of these quotients.
8) Determine the degree of freedom and find out the critical value of χ2
from table.
9) Compare the calculated and table value of χ2 and use the following
decision rule.

Accept null hypothesis if χ2 is less than critical value given in table.

Reject the null hypothesis if calculated value of χ2 is more than what is given
in the table under 0.05 or 0.01 significance levels.

Testing the Divergence of Observed Results from those Expected on the

Hypothesis of a Normal Distribution

Some times the expected frequencies are determined on the basis of the
normal distribution of observed frequencies in the entire population. The
method as to how this is done is explained in the following examples.

Example 1: 180 school teachers were classified into six categories on the
basis of their performance. Do these teachers differ significantly in their
performance as compared to the one expected if performance is supposed to
be distributed normally in our population of school teachers.

Categories 1 2 3 4 5 6 Total

No. of school teachers 15 25 45 50 35 10 180

Here, expected frequencies are to be determined on the basis of the normal

distribution hypothesis. For this purpose we have to divide the base line of
the curve into six equal segments of 1σ each. To mention here, the baseline in
any normal distribution ranges from from -3σ to +3σ. The table below
presents how many cases fall between the different areas.

100
 Computation of
Table 6.2 : Frequencies on the basis of Normal Distribution Chi-Square

Sr. No. Category % of cases falling No. of cases out of

between the area 180

1 + 3.00 σ to 2.00 σ 2.28 3.6

2 + 2.00 σ to +1.00 σ 13.59 25.2

3 + 1.00 σ to M 34.13 61.2

4 M to - 1.00 σ 34.13 61.2

5 -1.00 σ to - 2.00 σ 13.59 25.2

6 - 2.00 σ to -1.00 σ 2.58 3.6

We will now test the null hypothesis by computing the values of χ2 givenin
table.

Table 6.3 : Calculation of Chi-square step by step

1 2 3 4 5 6 Total

fo 15 25 45 50 35 10 180

fe 3.6 25.2 61.2 61.2 25.2 3.6 180

fo- fe 11.4 -0.2 -16.2 -11.2 9.8 6.4

(fo- fe)2 129.96 0.04 262.44 125.44 96.04 40.96

(fo- 36.1 0.001 4.29 2.05 3.81 11.38

fe)2/fe

The formula of χ2 is
χ2=∑[(fo- fe)2]/fe
= 36.1+.001+4.29+2.05+3.81+11.38

=57.63

df =(r-1)(c-1)
=(6-1)(2-1)

df =5

The critical value of χ2 for df= 5 is 11.0700 at 0.05 level and 15.086 at 0.01
level (refer to the table for chi-square given in the appendix of book on
statistics). Obtained values of χ2 is much greater than the critical values.
Hence it can be said that χ2 is significant and we reject the null hypothesis.

101
Non-Parametric
Statistics
Example 2: 200 salesmen were classified into five categories on the basis of
their performance ranging from excellent performance to very poor
performance. Does this evaluation differ significantly from the one expected
from the normal distribution?

The observed frequencies are given in the following table.

Very good Good Average Poor Very poor

30 45 65 40 20

We have to find out the expected frequencies with the help of normal
probability Curve. For this we have to divide the base line of the curve into
five equal segments by 6/5=1.2

Table 6.4: Computation of Expected Frequencies on the basis of Normal

Distribution

Sr. No. Category % of cases falling No. of cases out of 180

between the area

1 + 3.00 σ to 1.8 σ 3.45 6.90 or 7

2 + 1.8 σ to +.60 σ 23.84 47.68 or 48

3 + .60 σ to -.60 σ 45.14 90.28 or 90

4 -.60 σ to - 1.8 σ 23.84 47.68 or 48

5 -1.8 σ to - 3.00 σ 3.45 6.90 or 7

We will now test the null hypothesis by computing the values of χ2 given in
table.

Table 6.5: Calculation of Chi-square step by step

Very Good Average Poor Very Total

good poor

fo 30 45 65 40 20 200

fe 7 48 90 48 7

fo- fe 23 -3 -25 -8 13

(fo- fe)2 529 9 625 64 169

(fo- fe)2/fe 75.57 0.19 6.94 1.33 24.14

χ2=∑[(fo- fe)2]/fe
102 χ2 = 75.57+.19+6.94+1.33+24.14
 =108.17 Computation of
Chi-Square
df =(r-1)(c-1)

= (2-1) (5-1)
df =4

The critical value of χ2 at 4 df is 9.49 at 0.05 level and 13.29 at 0.01 level
(refer to the table for chi-square given in the appendix of textbook on
statistics) therefore obtained values of χ2 is much greater than the critical
values. Hence it can be said that χ2 is significant and we reject the null
hypothesis.

Steps for chi-square testing:

1) Formulate the null hypothesis.

2) Find out the observed frequencies.
3) To find out the expected frequencies first divide the base line of the
normal curve (which is divided into 6 equal σ units) equal to the number
of given categories of observed frequencies. For example in example 1
there are six categories therefore we have to divided the base line of the
curve in 6 units. In example 2 there are 5 categories and we have to
divide the base line of the curve in 5 units. Thus in the first case these
can be obtained by 6/6= 1σ unit in example. In example 2 there are five
categories and 5/6= 1.2 σ unit distance.
4) Now we have to decide the areas of normal curve as shown in table.
5) To decide the number of cases or percentage of cases, refer table of
normal probability. These tables are given in the appendix of books on
Statistics. In example 2, in the first category the area of normal curve is +
3 σ to + 1.8 σ. In table of normal distribution 0.4986 i.e. 49.86% cases
lies between means and 3 σ and 0.4641, that is. 46.41% cases lies in
between mean and 1.80 σ. Therefore 49.86-46.41=3.45% cases are there
in the first category. In the second category the area of normal curve is +
1.80 σ to + 0.6 σ. There are .4641 i.e. 46.41% cases lies between mean
and 1.8 σ and .2257 i.e. 22.57% cases lies in between mean and .6 σ .
Therefore percentage of cases between + 1.80 σ to0.6 σ will be (46.41-
22.57) 23.84.
In the third category the areas of normal curve is + 0.6 σ to - 0.6 σ here
.2257 means 22.57% cases lie on +0.6 σ and 22.57 lies on 0.6σ therefore
(22.57+22.57) 45.14% cases lies in the third category. Because this is
based on normal distributions which is symmetrically divided in two
halves, therefore number of cases in fourth and fifth categories will be
the same as in the second and first category.
6) No. of cases out of total number of cases can be found by dividing the
obtained areas of normal curve by 100 and multiply by total number of
cases. For example, in example 2, 3.45% of the cases falling in area I of
103
Non-Parametric
Statistics
first category is 3.45. Therefore the number of cases out of 200 will be
(3.45 × 200)/100=6.90. In this way find out fe. 7) After calculating the
fe place the values in the table and find out the χ2with the help of given
formula.

6.2.3 Chi-square as a Test of Independence

In addition to testing the agreement between observed frequencies and those
expected from some hypothesis, that is, equal probability and normal
probability, chi-square may also be applied to test the relationship between
variables. In this we test whether two variables are dependent or independent
to each other. The following table is known as contingency table.
Contingency table is a double entry or two way table in which the possession
by a group of varying degree of the characteristics is represented. The same
person for example may be classified as on the basis of sex that is boys and
girls or can be categorised in terms of the colour of their eyes, that is, brown
or black etc. These attributes are noted in terms of the gender factor.

To cite another example, father and sons may be classified with respect to
temperament or achievement and the relationship of the attributes in the
groups may be studied.

In the following table there is a double entry or it is called a two way table in
which 100 boys and 100 girls who express their view on the statement “There
should be same syllabus in all the universities of the country,” in terms of
agree, neutral, disagree. The results are shown in the following table.

Table 6.6: Frequencies of respondents in terms of views expressed

Gender Agree Neutral Disagree Total

Boys 20 30 50 100

Girls 50 20 30 100

Total 70 50 80 200

Do you think that the opinion is influenced by the gender of the students.?

For this we present the data in contingency table form as above. The expected
frequencies after being computed are written within brackets besides the
respective observed frequencies. The contingency table and the process of
computation of expected frequencies is given in the table 6.7.

104
 Computation of
Table 6.7: Contingency table with fo and fe for the data given in table 6.6 Chi-Square

Gender Agree Neutral Disagree Total

Boys 20 (35) 30 (25) 50 (40) 100

Girls 50 (35) 20 (25) 30 (40) 100

Total 70 50 80 200

The method to find out the expected frequency for each cell is given below:

70 × 100/200 = 35 50 × 100/200 = 25 80 × 10/200=40

70 × 100/200 = 35 50 × 100/200 = 25 80 × 100/200=40
The value of χ2may be computed with the help of usual formula,

χ2=∑[(fo-fe)2]/fe

Table 6.8: Computation of χ2 from contingency table 6.7

fo fe fo-fe (fo-fe)2 (fo-fe)2/fe

20 35 -15 225 6.43

30 25 5 25 1

50 40 10 100 2.5

20 35 -15 225 6.43

30 40 -5 25 1

20 40 -10 100 2.5

χ2=19.86

The value of χ2 may be computed with the help of normal formula.

For df = 2, the critical value at 0.05 level is 5.99 and at 0.01 level 9.21. Our
obtained value of χ2 is 19.86 . It is far higher than the table value. Therefore
we reject the null hypothesis and conclude that gender influences the opinion.

Steps for chi-square testing:

1) Formulate the null hypothesis

2) Find out the expected values by the method shown in table.
3) Find out the difference between observed and expected values for each
cell.
4) Square each difference and divide this in each cell by the expected
frequency.
5) Add up these and the sum of these values gives χ2.
105
Non-Parametric
Statistics
6.2.4 2× 2 Fold Contingency Table
When the contingency table is 2 x 2 fold χ2may be calculated without first
computing the four expected frequencies. Example below illustrates the
method.

Example: The mothers of 180 adolescents (some of them were graduates and
others non- graduates) were asked whether they agree or disagree on a certain
aspect of adolescent behaviour. Is the attitude of their mother related to their
educational qualification? Data are arranged in a four fold table below:

Agree Disagree

Graduate mother 30 (A) 50 (B) 80 (A+B)

Non graduate 70 (C) 30 (D) 100 (C+ D)

mother

Total 100 (A +C) 80 (B +D) 180 (A+ B+ C+

D)

In four fold table Chi-square is calculated by the following formula:

χ2= N(AD-BC)2/(A+B)(C+D)(A+C)(B+D)

Substitute for A,B,C,D, in the formula , we have

= 180 (30×30)-(50×70)2/80×100×100x80

=180(2600)2/64000000
=19.01
χ2=19.01

df = (r-1)(c-1)

= (2-1)(2-1)

df = 1

On df = 1, the critical value at 0.05 level is 3.84 and on 0.01 is 6.63 (refer to
the chi-square table given in the appendix of books on Statistics).

The obtained values is much higher therefore χ2is significant at .01 level, we
reject the null hypothesis and conclude that attitude is related with education.

6.3 CHI-SQUARE TEST WHEN TABLE ENTRIES

ARE SMALL (YATES CORRECTION)
When χ2 is computed from a table in which any experimental frequency is
less than 5 then χ2 is not stable. Moreover in 2 × 2 contingency table (df=1)
there are more chances of error. When any observed frequency is less than 5.
106
In these situations we make a correction for continuity known as Yates‟ Computation of
Chi-Square
correction.

To apply the Yates correction we subtract 0.05 from the absolute value of the
difference between observed and expected frequency.

What is the reason for applying Yates correction and what is its effect upon
χ2can be illustrated from the following example.

Example: A student is asked to respond on 10 objective type questions

requiring response in ‘yes’ or ‘no’. 7 times the student says ‘yes’ and 3 times
the student says ‘No’. Is this result different from what would have been
obtained by merely guessing?

Table 6.9: Computation of χ2 with Yates Correction-1

Yes No Total

Fo 7 3 10

Fe 5 5 10

fo- fe 2 -2

Correction 1.5 1.5

(fo- fe- 0.5)

(fo- fe)2 2.25 2.25

(fo- fe)2/fe 0.45 0.45

If we do not apply the Yates correction then the value of χ2 will be different.
This can be illustrated in the following table.

Table 6.10: Computation of χ2 with Yates Correction-1-1

Yes No Total

fo 7 3 10

fe 5 5 10

fo- fe 2 -2

(fo- fe)2 4 4

(fo- fe)2/fe 4/5 = 0.20 4/5 = 0.20

107
Non-Parametric
Statistics
χ2=∑[(fo- fe)2/fe]

= .2+.2 =

0.4

df =(3-1)(2-1)

df=2

For df= 2, the critical values (refer to the table for chi-square given in the
appendix of book on statistics) at 0.05 is 5.991 and on 0.01 is 9.210. In these
examples although both the χ2are not significant but if we do not apply the
correction then the probability of significance of χ2increased. We can
conclude that failure to use the correction causes the probability of its being
called significant considerably increased.

Yates Correction in 2×2 contingency

Example: 60 Students 50 from urban areas and 10 from rural area were
administered attitude scale to measure the attitude towards dating. Some of
them expressed positive attitudes and some expressed negative attitudes.
Whether the attitudes of the students will be effected by their belonging to
rural or urban areas. The data are given in the following table.

Table 6.11: Computation of χ2 with Yates Correction-1-1-1

Positive Negative Total

Urban 20 (A) 30 (B) 50

Rural 30 (C) 7 (D) 10

Total 23 (A +C) 37 (B+D) N= 60 (A+ B+

C+ D)

In this type of 2×2 table you can calculate chi-square in a different method
which is given below:

χ2= N(|AD-BC|- N/2)2/(A+B)(C+D)(A+C)(B+D)

= 60(|20×7-30×3|) -60/2]2/ (50)(10)(23)(37)

= 60(|140-90|-30]2/425500
=60×400/425500

=24000/425500

= .056

Entering the table for chi-square we find that for df= 1 the value of Chi-
square at 0.05 level is be 3.841. The obtained value is less than the table
value and hence we retain the null hypothesis and conclude that the attitude
108
towards dating is not influenced by the person belonging to rural or urban Computation of
Chi-Square
areas.

Check Your Progress I

1) Complete the sentences:

a) The chi-square test is a useful method ……………………………
……………………………………………………………………………
……………………………………………………………………………
b) Contingency table is a double entry or two way table in which ….
……………………………………………………………………………
……………………………………………………………………………
c) When the contingency table is 2 x 2 fold χ2may be calculated …..
……………………………………………………………………………
……………………………………………………………………………
d) To apply the Yates correction …………………………………….
……………………………………………………………………………
……………………………………………………………………………

6.4 LET US SUM UP

The chi-square test is used for two broad purposes. It is used as a test of
goodness of fit and second as a test of independence. As a test of goodness of
fit χ2 tries to determine how the observed frequencies are different from the
frequency expected theoretically by hypothesis of equal probability and
hypothesis of normal probability. The formula of χ2 is χ2=∑[(fo- fe)2/fe]. As
a test of independence χ2 is applied for testing the relationship between two
variables in two ways. In the 2 × 2 contingency table there are four cells
A,B,C,D, and χ2 is calculated with the help of formula, χ2= N(AD-
BC)2/(CA+B)(C+D)(A+C)(B+D). When in any cells entries are less than 5,
Yate’s correction is applied and the formula used to calculate is χ2= N(|AD-
BC|)-N/2/(A+B(C+D)(A+C)(B+D). The vertical lines |AD-BC| mean that the
difference is to be taken as positive.

5.5 REFERENCES
Kerlinger, Fred N. (1963). Foundation of Behavioural Research, (2nd Indian
reprint) Surjeet Publication, New Delhi.
Garrett, H.E. (1971), Statistics in Psychology & Education, Bombay, Vakils,
Seffer & Simoss Ltd.
Guilford, J.P. (1973), Fundamental Statistics in Psychology & Education,
Newyork, McGraw Hill.
109
Non-Parametric
Statistics 6.6 KEY WORDS
Critical value: Table value according to the given df value. 32 Expected
Frequency : The expected scores in a given class.
Expected frequencies: The frequencies for different categories which are
expected to occur on the assumption that the given hypothesis is true.
Observed frequencies: The frequencies actually obtained from the
performance of an experiment.
Yate’s correction: The correction of -.5 in expected frequency.
Contingency table: A table having rows and columns where in each row
corresponds to a level of one variable and each column to a level of another
variable.

6.7 ANSWERS TO CHECK YOUR PROGRESS

Check Your Progress I
1) Complete the sentences:
a) The chi-square test is a useful method of comparing experimentally
obtained results with those to be expected theoretically on some
hypothesis.
b) Contingency table is a double entry or two way table in which the
possession by a group of varying degree of the characteristics is
represented.
c) When the contingency table is 2 x 2 fold χ2 may be calculated without
first computing the four expected frequencies.
d) To apply the Yates correction we subtract 0.05 from the absolute value
of thedifference between observed and expected frequency.

6.8 UNIT END QUESTIONS

1) Describe the properties of chi-square test.
2) What do you understand by the goodness of fit test ? Describe with
examples.
3) What is the chi-square test for independence ? Explain with examples.
4) What precautions would you keep in mind while using chi-square.
5) Write short notes on: Yates correction and expected frequencies in
contingency table.

110
UNIT 7 MANN WHITNEY U TEST* Mann Whitney U
Test

Structure
7.0 Objectives
7.1 Introduction
7.2 Mann Whitney U Test
7.3 Computation of Mann Whitney U Test from Small Sample
7.4 Computation of Mann Whitney U Test from Moderately Large Sample
7.5 Computation of Mann Whitney U Test from Large Sample
7.6 Let Us Sum up
7.7 References
7.8 Key Words
7.9 Answers to Check Your Progress
7.10 Unit End Questions

7.0 OBJECTIVES
After reading this unit, you will be able to

• discuss Mann Whitney U test; and

• compute Mann Whitney U test for small sample, moderately large
sample and large sample.

7.1 INTRODUCTION
A researcher wanted to study if significant difference exists between junior
and senior managers with regard to their work motivation. The researcher
collected data using standardised psychological test for work motivation and
then had to carry out statistical analysis of the data. Initially, the researcher
felt that independent t test cannot be used as the assumptions of parametric
tests were not fulfilled. Therefore, the researcher decided to use Mann
Whitney U test.

In unit 3 of this course, we discussed about t test. t test is parametric

statistical technique that us used in order to measure the difference between
two sample means. When the assumptions of parametric tests that we
discussed in unit 2, are not met, it is not possible for us to use t test. In such a
case the non-parametric counterpart of independent t test that is Mann
Whitney U test can be effectively used in order to measure the difference
between two sample means.

*
Prof. Suhas Shetgovekar, Faculty, Discipline of Psychology, SOSS, IGNOU, Delhi
111
Non-Parametric
Statistics
Thus, in the present unit, we will try to understand what is Mann Whitney U
test and also learn how to compute it.

7.2 MANN WHITNEY U TEST

Mann Whitney U test can be explained as a powerful non-parametric test and
an effective alternative to independent t test. It can be used adequately with
variables that are continuous and discrete and can be effectively used with
small sample. Though in terms of power, it is less powerful compared to
independent t test, but its power is higher than other Non-parametric tests like
median test (that will be discussed in the next unit) and Wilcoxon rank sum
test.

Mann Whitney U test can be applied to find out whether two sample
subgroups have been taken from a same population. It also helps in
determining the distribution of data with regard to its median (Veeraraghavan
and Shetgovekar, 2016).

The assumptions of non-parametric tests are applicable to Mann Whitney U

test. Let us now discuss some of the assumptions of Mann Whitney U test
(Mohanty and Misra, 2016):

1) The observations need to be independent.

2) The dependent variable that is taken in the research needs to display
continuity.
3) The test is used when the two sample subgroups are independent and the
data is ordinal. In case if the data is in interval or ratio form then the
same is converted in to rank order.
4) The sample sub groups need to be independent and not correlated.

Check Your Progress I

1) State any one assumption of Mann Whitney U Test.

……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………

7.3 COMPUTATION OF MANN WHITNEY U

TEST FOR SMALL SAMPLE
A school counsellor wanted to study whether gender difference exists in
students with regard to mathematical ability. The same size was quiet sample
as each of the sample subgroup had less than 8 participants. In such a
112
situation Mann Whitney U test for small sample can be computed. The data is Mann Whitney U
Test
given below, let us learn how Mann Whitney U can be computed in this
context with the help of steps:

Table 7.1: Mathematical Ability Scores of Male and Female Students

Males (A) Females (B)

6 10

13 16

7 12

11 14

N=5 N=5

Step 1: Null hypothesis is to be formulated.

The null hypothesis is formulated as follows:

Ho = There is no significant difference between males and females with

regard to mathematical ability.

Step 2:Arranging the scores obtained by males and females in an ascending

order, that is, from lowest to highest.

The codes for their groups, A for males and B for females also need to
mentioned.

Table 7.1: Scores arranged in Ascending Order

5 6 7 10 11 12 13 14 16 17

A A A B A B A B B B

Step 3: U is computed.

This is done by determining the number of scores for A (males) that are
falling below the scores of B (females). This is done as follows:

For the score in group B of 10, the number of scores from group A that fall
below 10 = 3

For the score in group B of 12, the number of scores from group A that fall
below 12= 4

For the score in group B of 14, the number of scores from group A that fall
below 14 = 5
113
Non-Parametric
Statistics
For the score in group B of 16, the number of scores from group A that fall
below 16 = 5

For the score in group B of 17, the number of scores from group A that fall
below 17= 5

Thus, U = 3 + 4 + 5 + 5 +5 = 22

Let us also compute U'

For the score in group A of 5, the number of scores from group B that fall
below 5 = 0

For the score in group A of 6, the number of scores from group B that fall
below 6 = 0

For the score in group A of 7 , the number of scores from group B that fall
below 7 = 0

For the score in group A of 11, the number of scores from group B that fall
below 11 = 1

For the score in group A of 13, the number of scores from group B that fall
below 13 = 2

U' = 0 +0 +0 +1 + 2= 3

Thus, U = 22 and U' = 3

Further, the total of U and U' needs to be equal to N1 N2.

Thus,

U + U' = N1 N2

Thus,

U' = N1 x N2 - U

= 5 x 5 - 22

= 25-22

=3

The value obtained is 3 and it is same as the value obtained earlier using the
formula.

Step 4: Interpretation of the U value

To interpret the obtained U, the table for U is to be referred to, that can be
found in appendix of books on Statistics. In our example N1 = 5 and N2 = 5.
And the U = 3. The table value in this regard is .579. The U obtained by us is
more than the table value and thus null hypothesis can be accepted and it can
be said that the samples have been drawn from the same population.
114
Check Your Progress II Mann Whitney U
Test
1) List the steps in computation of Mann Whitney U test for small sample.
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………

7.4 COMPUTATION OF MANN WHITNEY U

TEST FOR MODERATELY LARGE
SAMPLES
In the previous section, we discussed about how Mann Whitney U test can be
computed for small sample. In this section, we will discuss how it is
computed for a moderately large sample, that is a sample size that falls
between 9 and 20.
Let us now understand how to compute Mann Whitney U test for moderately
large sample with the help of an example.
A researcher was carrying out a study on how two different teaching
methods, lecture and group discussion have an effect on academic
performance of the students. The first group (Group A) consisted of 11
students and were taught using lecture method and the other group (Group B)
consisted of 9 students and were taught using group discussion method. The
date obtained is given in table 7.3.

Table 7.3: Academic Performance of the Students

Group A Group B
49 51
39 48
65 36
62 37
34 51
45 56
46 59
65 78
65 83
66
85
N= 11 N= 9

115
Non-Parametric
Statistics
Step 1: Formulate the Null hypothesis.

Ho = There will be no significant difference between the two groups with

regard to their academic performance.

Step 2:Ranks are to be given to the combined scores from both groups.

The scores of both the groups are combined and then arranged in an
ascending order and ranks are assigned to them. Rank 1 is allotted to the
lowest score and the next above it will be given a rank 2 and so on. In case
of tied scores, the procedure similar to the one that we discussed in BPCC104
while computing Spearman’s rho is used. Thus, the re are two 51 scores, the
ranks allotted are (9+ 10)/ 2= 9.5 and the next score gets rank 11. Further,
there are three 65 score. Thus, (14 + 15+ 16)/ 3= 15. Thus, the rank assigned
to this score is 15 and the next score gets rank 17.

Table 7.4: Academic Performance of the Students

Group 1 Rank (R1) Group 2 (Audio- Rank (R2)

(Lecture video method)
Method)

49 8 51 9.5 (tied ranks)

39 4 48 7

65 15 (tied ranks) 36 2

62 13 37 3

34 1 51 9.5 (tied ranks)

45 5 56 11

46 6 59 12

65 15 (tied ranks) 78 18

65 15 (tied ranks) 83 19

66 17

85 20

N= 11 ∑R1 = 119 N= 9 ∑R2 = 91

Step 3: Compute the summation of the ranks in each group

Thus R1= 119and R2= 91. (R1 + R2 = 210)

The same can be checked as using a formula as follows:

116 N (N +1)/ 2
 where N = N1 + N2 Mann Whitney U
Test
= 20 (20 +1)/2

= 20 x 21/ 2

= 420/ 2

= 210

The value obtained is 210 and is equal to the value we obtained after adding
R1and R2.

Step 4: Compute U

U can be computed using the following formula

U = N1 N2 + [N1 (N1 + 1)]/ 2 - ∑R1

The formula forU' is as follows:

U' = N1 N2 + [N2 (N2 + 1)/2] - ∑R2

Let us now compute U using the values given in table 7.3

U = N1 N2 + N1(N1 + 1) / 2 - ∑R1

= 11 x 9 + [11 (11 +1)] /2 - 119

= 99+ 11 (12) / 2- 119

= 99 + 132 /2 - 119

= 99 + 66 - 119

= 165- 119 = 46

Step 5: Find out U' using the formula given in the box above.

U' = N1 N2 + [N2 (N2 + 1)/2] - ∑R2

= 11 x 9+ [9(9+1) / 2] – 91
= 99 + (9 x 10) / 2] – 91
= 99+ 90 / 2 – 91
= 99 + 45 – 91
= 144 – 91 = 53

Thus, U = 46 and U' = 53

Further, the total of U and U' needs to be equal to N1 N2.

U' = N1 N2 – U
=11 x 9 - 46
= 99- 46
= 53 117
Non-Parametric
Statistics Step 6: Interpretation of U

In order to interpret the obtained U, we need to refer to the table value that is
given in the table for Mann Whitney U in appendix of any Statistics book.
Referring to such a table, the table value with the large sample as 11 and
small sample as 9 is 18 at 0.01 level of significance and 27 at 0.05 level of
significance. The obtained U is 46 is higher than the table value. Thus, the
null hypothesis is accepted and it can be said that there is no significant
difference in the academic performance with regard to the two teaching
methods.

Check Your Progress III

1) What is the formula to compute U.

……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………

7.5 COMPUTATION OF MANN WHITNEY U

TEST FOR LARGE SAMPLES
A large sample is a sample that is more than 20. The distribution for such a
sample could appear similar to normal distribution and thus, in this case the
formula for z can be employed in order to find out if significant difference
exists between the two groups.

Let us now try to understand the computation of Mann Whitney U test for
large samples with the help of an example.

A researcher wanted to find if significant difference exists in attitude towards

environment of individuals in rural and urban areas. The data collected is
give below in table

118
 Mann Whitney U
Table 7.5: Scores on Attitude tiowards Environment along with their Test
ranks .

Individuals from Rank 1 Individuals Rank 2

rural area from urban
area

42 9 85 28

45 10 109 37

46 11 87 29

95 34 88 30

23 1 90 31

94 33 91 32

33 6 26 2

35 7 100 35

32 5 102 36

40 8 27 3

30 4 110 38

83 27 76 26

72 24 73 25

69 23 47 13

47 13 50 16

59 22 57 21

47 13 50 16

51 18 56 20

54 19 50 16

120 39 121 40

N= 20 ∑R1 =326 N= 20 ∑R2 =494

Step 1:The data collected in tabulated as is give in table 7.5.

The data is to be organised in a table form with scores as obtained by

participants in each group.
119
Non-Parametric
Statistics Step 2: Rank the scores

As was done in previous section. In similar manner, ranks are to be given to

the scores. Rank 1 is given to the lowest score and the last score gets the
highest rank. Tied ranks are also given as was discussed earlier.

Step 3: Ranks for each group are summated.

Thus ∑R1= 326 and ∑R2= 494

R1+ R2= 326 + 494 = 820

This can be checked using the formula as follows:

N (N +1)/ 2
where N = N1 + N2
= 40(40 +1)/2
= 40 x 41/ 2
= 1640/ 2
= 820

The value obtained is 210 and is equal to the value we obtained after adding
R1and R2.

Step 4: Compute U

U can be computed using the following formula

U = N1 N2 + [N1 (N1 + 1)]/ 2 - ∑R1

The formula for U’ is as follows:

U' = N1 N2 + [N2 (N2 + 1)/2] - ∑R2

Let us now compute U using the values given in table 7.3

U = N1 N2 + N1(N1 + 1) / 2 - ∑R1

= 20 x 20+ [20(20 +1)] /2 - 326

= 400+ 20 (21) / 2- 326

= 400 + 420 /2 - 326

= 400 + 210 - 326

= 610- 326= 284

Step 5: Find out U' using the formula given in the box above.

U' = N1 N2 + [N2 (N2 + 1)/2] - ∑R2

= 20 x 20+ [20(20 +1)] /2 - 494

120
 = 400+ 20 (21) / 2- 494 Mann Whitney U
Test
= 400 + 420 /2 - 494

= 400 + 210 - 494

= 610- 494= 116

Thus, U = 284 and U' = 116

Further, the total of U and U' needs to be equal to N1 N2.

U' = N1 N2 – U

=20 x 20 - 284

= 400- 284

= 116

Step 5: Compute z

Now we need to compute z with the help of the following formula:

z = U – [(N1 N2) /2] / √ {(N1)(N2)(N1+N2+1)/ 12}

= 284 – [20x20) / 2] / √ [(20) (20)(20+20+1)/12]

= 284 - [400 /2] / √ [400x 41/12]

= 284 – 200 / √ 1366.67

= 84 / 36.97

= 2.27

In place of U, you could have taken U'. The z value obtained in either case
wold remain same.

z = –U' [(N1N2) /2] / √ {(N1)(N2)(N1+N2+1)/ 12}

= 116– [20x20) / 2] / √ [(20) (20)(20+20+1)/12]

= 116 - [400 /2] / √ [400x 41/12]

= 116 – 200 / √ 1366.67

= 84 / 36.97

= 2.27

Thus, the z value is obtained as 2.27.

Step 6: Interpretation of z value

To interpret the z value we will be required to refer to the table value from
the table for z value that is given in appendix of any Statistics book.
121
Non-Parametric
Statistics
In this context, it can be said that if the z value is >1.96, U is interpreted as
significant at 0.05 level of significance (for two tailed test) and thus null
hypothesis is rejected. If the obtained z value is >2.58, U is interpreted as
significant at 0.01 level of significance (for two tailed test) and thus null
hypothesis is rejected. With regard to one tailed test the z value needs to be is
> 1.64 at 0.05 level of significance and > 2.33 at 0.01 level of significance
order to reject the null hypothesis.

The z value obtained by us is 2.27 which is less than >2.58 for 0.01 level
of significance but of more than 1.96, thus it can be said that U is significant
at 0.05 level and null hypothesis can be rejected.

This indicates that the null hypothesis is rejected and one may conclude that
there is a significant difference in the attitude towards environment of
individuals belonging to rural and urban area of residence.

Check Your Progress IV

1) What is the formula for z?

……………………………………………………………………………

……………………………………………………………………………

……………………………………………………………………………

……………………………………………………………………………

……………………………………………………………………………

7.6 LET US SUM UP

In the present unit, we discussed about Mann Whitney U test. Mann Whitney
U test can be explained as a powerful Non-parametric test and an effective
alternative to independent t test. It can be used adequately with variables that
are continuous and discrete and can be effectively used with small sample.
Though in terms of power, it is less powerful compared to independent t test,
but its power is higher than other non-parametric tests like median test (that
will be discussed in the next unit) and Wilcoxon rank sum test. After
understanding the test, we also discussed about its assumptions. Further, we
learnt to compute Mann Whitney U test for small sample, moderately large
sample and large sample with the help of steps and examples.

7.7 REFERENCES
King, Bruce. M; Minium, Edward.W. (2008).Statistical Reasoning in the
Behavioural Sciences. Delhi: John Wiley and Sons, Ltd.

Mangal, S. K. (2002). Statistics in Psychology and [Link] Delhi: Phi

Learning Private Limited.
122
UNIT 8 MEDIAN TEST * Median Test

Structure
8.0 Objectives
8.1 Introduction
8.2 Median Test
8.3 Computation of One Sample Median Test
8.4 Computation of Median Test for Two Independent Samples
8.5 Computation of Median Test for more than Two Independent Samples
8.6 Let Us Sum up
8.7 References
8.8 Key Words
8.9 Answers to Check Your Progress
8.10 Unit End Questions

8.0 OBJECTIVES
After reading this unit, you will be able to,

• discuss median test and its assumptions; and

• compute one sample median test, compute median test with two
independent samples and median test for more than two independent
samples.

8.1 INTRODUCTION
A researcher wanted to study the median academic scores obtained by
students in 9th standard class of a school. Yet another researcher wanted to
compare the median of scores obtained by individuals in urban and rural
areas on attitude towards environment.

In both these case, the researchers can use median test in order to analyse the
data collected by them. In the previous unit, we discussed about Mann-
Whitney U test. In the present unit, we will discuss about median test that can
be used in similar way as the Mann Whitney U test. We will discuss median
test and its assumptions. And will further explain the computation of one
sample median test. compute median test with two independent samples and
median test for more than two independent samples.

*
Prof. Suhas Shetgovekar, Faculty, Discipline of Psychology, SOSS, IGNOU, Delhi
125
Non-Parametric
Statistics 8.2 MEDIAN TEST
It can be said that median test is a non-parametric alternative of t test. It can
be employed to know if the median of a certain data is same as that of
assumed median in which case one sample median test can be used. It can
also be used to know whether two or more samples that are independent and
equal or not equal with regard to their size, have different median or not.
Thus, medians of two sample sub groups are compared and these samples can
be equal or unequal in size.

Median test is a non-parametric statistic and thus it can be used with

distribution that is normal or not normal. Further it can be used with variables
having either continuous and discontinuous measurements. It can also be
conveniently used with small sample. though when compared with Mann-
Whitney U test, it is less powerful.

We need to remember here that as such the assumptions of Non-parametric

test are applicable here as well. Let us now look at the assumptions of the
median test (Mohanty and Misra, 2016) :

1) The scores in the samples (two or more) can be measured in terms of

ordinal scale of measurement.
2) Every score in the sample occurs in a random manner.
3) Every score is independent of other scores.
4) The samples (two or more) need to be independent and not correlated to
each other.

Check Your Progress I

1) When can the median test be used?

……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………

8.3 COMPUTATION OF ONE SAMPLE MEDIAN

TEST
One sample median test is used in order to know if the median of certain
sample is equal to the population median that is assumed (veeraraghavan and
Shetgovekar, 2016).

Let us now try to understand the computation of one sample median test with
the help of steps and example.
126
An industrial psychologist wanted to know the average years of experience of Median Test

newly appointed junior managers in the organisation. The data collected is

given as follows:

2 3 6 4 3 2 4 6 10 9 7 8 3 2

Step 1: Null hypothesis is to be formulated.

H0 = The median years of experience of junior managers will be the same as

the assumed median years of experience.

Step 2: Assuming median sample years of experience as 5

Step 3: Coding the scores

The scoreshaving lower value than the assumed median are to be given the
code as ‘0’ and the scores having value more than the assumed median

Those having value more than the assumed median will be given a code as
‘1’. The values that are the same as that of the assumed median will not be
considered.

2 3 6 4 3 2 4 6 10 9 7 8 3 2

0 0 1 0 0 0 0 1 1 1 1 1 0 0

Step 4: Rearranging the data

The data is now rearranged in terms of frequency of scores above assumed

median and below assumed median as can be seen below:

Above Assumed Median Below Assumed Median

6 8

Step 5:Frequency expected (fe) is computed using the formula:

N= 14, that is the total number of participants.

fe = 1⁄2x N

= 1/2 x 14

=7

Thus

Frequency observed (fo) Frequency expected (fe)

6 7

8 7
127
Non-Parametric
Statistics
Step 6: Chi Square is computed using the following formula:

χ2 =Σ [( fo- fe)2 ]/ fe

Frequency Frequency fo-fe ( fo- fe)2 ( fo- fe)2 /fe

observed (fo) expected (fe)

6 7 -1 1 0.14

8 7 1 1 0.14

Χ2=0.28

The Χ2 =0.28

Step 7: Interpretation

In order to interpret the obtained chi- square, we need to first know the
degree of freedom that will be

df= (r-1) (k-1)

r means rows

k means columns.

Thus,

= (2-1) (2-1)

=1

Thus, df= 1

Now we can refer to the chi- square table that is given in appendix of any
Statistics book. The critical value of chi square for df= 1, is 6.635 at 0.01
level of significance and 3.841at 0.05 level of significance. The chi- square
obtained by us is 0.28 which is lower than the table value at both the levels.
Thus the null hypothesis is accepted and it can be said that there is no
significsnt difference between the assumed median and the sample median
years of experience of the newly appointed junior managers.

Check Your Progress II

1) List the steps in computation of one sample median test.

……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
……………………………………………………………………………
128
8.4 COMPUTATION OF MEDIAN TEST FOR Median Test

TWO INDEPENDENT SAMPLES

As was mentioned earlier, median test can also be used in order to know
whether two samples that are independent differ in terms of their median or
not. In the present section, we will learn how to compute median test for two
independent samples and we will do so with the help of steps and an
example.

The scores obtained by employees of public and private banks on work

motivation are given as follows

Employees from private 12 13 13 15 17 18 19 21 22 21 11 23

banks

Employees from public 23 12 12 14 14 21 23 21 23 21 20

banks

Step1: Formulate null hypothesis.

The scores obtained by the employees from private and public sector banks
are not different from the common median.

Step 2: Computation of common median

A common median is to be computed for both the groups.

n= n1 + n2

= 12 +11= 23

Median = n=1/ 2th score

= 23 + 1/2

= 12th score

Now we need to arrange the scores from both the groups in an ascending
order and then identify the 12th score

11, 12, 12, 12, 13, 13, 14, 15, 17, 18, 19, 20, 21, 21, 21,21, 21, 22, 23, 23 ,
23, 23

The 12th score is 19. Thus, the common median is obtained as 19.

Step 3: Assignment of signs to the scores.

The scores in each group are assigned positive (+) or negative (-) signs
depending on whether they fall above or below the median. If above median,
positive sign is given and if below or equal to the median, then negative sign
is given.

129
Non-Parametric
Statistics Employees from private banks - - - - - - - + + + - +

Employees from public banks + - - - - + + + + + +

Step 4: 2 x 2 contingency table is created.

Such a table is created so that chi square can be computed.

Employees Positive signs Negative signs Total

Private banks 4 8 12

Public banks 7 4 11

Total 11 12 23

Step 5: Compute chi- square

S[(f0 – fe)2 / fe ]
For that we first need to compute the frequency expected (fe) as follows:
fe for cell 1: (12 × 11)/23 = 132/23 = 5.74
fe for cell 2: (11 × 11)/23 = 121/23 = 5.74
fe for cell 4: (12 × 12)/23 = 144/23 = 6.26
fe for cell 5: (12 × 12)/23 = 132/23 = 5.74
Frequency Frequency fo - fe (fo - fe)2 (fo - fe)2/fe
observed (fo) expected (fe)
4 5.74 -1.74 3.0276 0.53
7 5.26 1.74 3.0276 0.57
8 6.26 1.74 3.0276 0.48
4 5.74 -1.74 3.0276 0.53
2
X = 2.11
The χ2 obtained is 2.11
Calculate the degrees of freedom applying the formula (r-1)(k-1). The df for
the data in the table is = (2-1)(2-1) = 1 × 1 = 1

Step 6: Interpretations

Now we can refer to the chi- square table that is given in appendix of any
Statistics book. The critical value of chi square for df= 1, is 6.635 at 0.01
level of significance and 3.841at 0.05 level of significance. The chi- square
obtained by us is 2.11 which is lower than the table value at both the levels.
Thus the null hypothesis is accepted and its can be said that the scores
obtained by the employees from private and public sector banks are are not
different from the common median.

130
Check Your Progress III Median Test

1) State the formula for chi- square while computing median test for two
independent samples.

8.5 COMPUTATION OF MEDIAN TEST FOR

MORE THAN TWO INDEPENDENT
SAMPLES
Median test can also be computed for more than two independent samples.
The procedure for computation is similar to that of median test for two
independent sample, though the formula used for computation of chi-square
will differ. Let us try to understand the computation with the help of steps
and example.

The scores obtained by three groups (Groups A, B and C) of students on

achievement motivation are given as follows

Group A 4 27 6 4 26 8 9 2 7 3

Group B 3 6 9 12 16 20 21 26 5 29

Group C 4 8 10 14 15 18 23 12 15 17

Step1: Formulate null hypothesis.

The scores obtained by the three groups of students are not different from the
common median.

Step 2: Computation of common median

A common median is to be computed for the three groups.

n= n1 + n2 + n3

= 10 +10 +10 = 30

Median = n=1/ 2th score

= 30 + 1/2

= 15.5th score

Now we need to arrange the scores from the three groups in an ascending
order and then identify the 15.5th score

2, 3, 3, 7, 4, 4, 4, 5, 6, 6, 8, 8, 9, 9, 10, 12, 12, 14, 15,15, 17, 16, 18, 20, 21,
23, 26 26, 27, 29
131
Non-Parametric
Statistics
The median is the 15,5th score , that is an average of 15th and 16th score. In our
example, the 15th score is 10 and t16h score is 12. The average of these scores
would be (10 +12)/2+ 11. Thus, the common median is 11.

Step 3: Assignment of signs to the scores.

The scores in each group are assigned positive (+) or negative (-) signs
depending on whether they fall above or below the median. If above median,
positive sign is given and if below or equal to the median, then negative sign
is given.

Group A - + - - + - - - - -

Group B - - - + + + + + - +

Group C - - - + + + + + + +

Step 4: 2 x 2 contingency table is created.

Such a table is created so that chi square can be computed.

Positive sign Negative sign Total

Group A 2 8 10

Group B 6 4 10

Group C 7 3 10

Total 15 15 n= 30

Step 5: Compute chi- square

The formula used is as follows:

x2 = ∑[(f0 – fe)² /fe ]

For that we first need to compute the frequency expected (fe) as follows:

fefor cell 1: (10 x 15)/ 30 = 150/ 30 = 5

fe for cell 2: (10 x 15)/ 30= 150/ 30 = 5

fefor cell 3: (10 x 15)/ 30= 150/ 30 = 5

fe for cell 4: (10 x 15)/ 30= 150/ 30 = 5

fe for cell 5: (10 x 15)/ 30= 150/ 30 = 5

fefor cell 6: (10 x 15)/ 30 = 150/ 30 = 5

132
 Median Test
Frequency Frequency fo-fe ( fo- fe)2 ( fo- fe)2 /fe
observed (fo) expected (fe)

-3 9 1.8

1 1 0.2

7 5 2 4 0.8

8 5 3 9 1.8

4 5 -1 1 0.2

3 5 -2 4 0.8

Χ2= 5.6

The chi-square obtained is 5.6

Calculate the degrees of freedom applying the formula (r-1)(k-1). The df for
the data in the table is = (3-1)(2-1) = 2 x 1= 2

Step 6 Interpretation

We need to refer to the table value for chi square that is given in any statistics
book.

For df = 2 the table value is 5.99 for 0.05 level of significance and 13.82 at
0.01 level of significance. The chi square obtained by us is 5.6 which is lower
than the table values at both the levels of significance. Thus, the null
hypothesis that the scores obtained by the three groups of students are not
different from the common [Link] accepted.

Check Your Progress IV

1) State the formula for chi- square while computing median test for more
than two independent samples.

8.6 LET US SUM UP

To summarise, median test is a non parametric alternative of t test. It can be
employed to know if the median of a certain data is same as that of assumed
median in which case one sample median test can be used. It can also be used
133
Non-Parametric
Statistics
to know whether two or more samples that are independent and equal or not
equal with regard to their size, have different median or not. Thus, medians
of two sample sub groups are compared and these samples can be equal or
unequal in size. The assumptions for median test were also discussed. The
unit then focused on the computation one sample median test. compute
median test with two independent samples and median test for more than two
independent samples.

8.7 REFERENCES
King, Bruce. M; Minium, Edward.W. (2008).Statistical Reasoning in the
Behavioural Sciences. Delhi: John Wiley and Sons, Ltd.

Mangal, S. K. (2002). Statistics in Psychology and [Link] Delhi: Phi

Learning Private Limited.

Minium, E. W., King, B. M., & Bear, G. (2001).Statistical Reasoning in

Psychology and Education. Singapore: John-Wiley.

Mohanty, B and Misra, S. (2016). Statistics for Behavioural and Social

Sciences. Delhi: Sage.

Veeraraghavan, V and Shetgovekar, S. (2016). Textbook of Parametric and

Non-parametric Statistics. Delhi: Sage.

8.8 KEY WORD

Median test: Median test is a non parametric alternative of t test. It can be
employed to know if the median of a certain data is same as that of assumed
median in which case one sample median test can be used. It can also be used
to know whether two or more samples that are independent and equal or not
equal with regard to their size, have different median or not.

8.9 ANSWERS TO CHECK YOUR PROGRESS

Check Your Progress I

1) When can the median test be used?

Median can be used to know if the median of a certain data is same as
that of assumed median in which case one sample median test can be
used. It can also be used to know whether two or more samples that are
independent and equal or not equal with regard to their size, have
different median or not.

Check Your Progress II

1) List the steps in computation of one sample median test.

The steps in computation of one sample median test are:
134
 Step 1: Null hypothesis is to be formulated. Median Test

Step 2: Assuming median

Step 3: Coding the scores
Step 4: Rearranging the data
Step 5:Frequency expected (fe) is computed
Step 6: Chi Square is computed
Step 7: Interpretation

Check Your Progress III

1) State the formula for chi- square while computing median test for two
independent samples.
∑[(f0 – fe )² / fe ]

Check Your Progress IV

1) State the formula for chi- square while computing median test for more
than two independent samples.
∑[(f0 – fe )² / fe ]

8.10 UNIT END QUESTIONS

1) Explain median test with a focus on its assumptions.
2) Compute one sample median test for the following data:

34 32 22 34 43 45 56 54 56 43 22 36 43 33

3) Explain the computation of median test for two independent samples

with the help of steps and examples.
4) Compute median test for more than two independent samples for the
following dats:

Sample A 4 5 4 6 8 22 14 45 32 2

Sample B 6 7 45 5 6 32 16 6 4 7

Sample C 4 8 3 6 44 34 32 4 10 12

Sample D 3 3 4 6 5 12 12 8 9 13

135
unnecessary calculations and also give rise to type I error. The answer in Kruskal- Wallis
Analysis of
these cases when we have more than two groups (>2 groups) to be compared Variance
is to conduct Analysis of Variance .

9.2 KRUSKAL-WALLIS ANOVA TEST

Let us revise ANOVA before we go on to discuss about Kruskal- Wallis
ANOVA test.

The term Analysis of Variance (for which the acronym ANOVA is often
employed) describes a group of inferential statistical procedures developed
by the British statistician Sir Ronald Fisher. Analysis of Variance is all about
examining the amount of variability in a y (response) variable and trying to
understand where that variability is coming from. One way that you can use
ANOVA is to compare several populations regarding some quantitative
variable, y. The populations you want to compare constitute different groups
(denoted by an x variable), such as political affiliations, age groups, or
different brands of a product. ANOVA is also particularly suitable for
situations involving an experiment where you apply certain treatments (x) to
subjects, and you measure a response (y).

Null hypothesis (H0): Population means are equal. There will be no

difference in the population means.

Alternative hypothesis (H1): Population means are not equal. There will be
difference in the means of the different populations.

The logic used in ANOVA to compare means of multiple groups is similar to

that used with the t-test to compare means of two independent groups. When
one way ANOVA is applied to the special case of two groups, this one way
ANOVA gives identical results as the t-test.

Not surprisingly, the assumptions needed for the t-test are also needed for
ANOVA. We need to assume:

1) random, independent sampling from the k populations;

2) normal population distributions;
3) equal variances within the k populations.

Assumption 1 is crucial for any inferential statistic. As with the t-test,

Assumptions 2 and 3 can be relaxed when large samples are used, and
Assumption 3 can be relaxed when the sample sizes are roughly the same for
each group even for small samples. (If there are extreme outliers or errors in
the data, we need to deal with them first.)

9.2.1 Kruskal- Wallis ANOVA Test

When there are more than two groups or k number of groups to be compared,
ANOVA is utilised but, again since ANOVA is a parametric statistics and
137
Non-Parametric
Statistics
requires assumption of normality as a key assumption, we need to also be
aware of its non-parametric counterpart. The Kruskal- Wallis test compares
the medians of several (more than two) populations to see whether they are
all the same or not. The Kruskal- Wallis test is a non-parametric analogue to
ANOVA. It can be viewed as ANOVA based on rank transformed data.

That is, the initial data are transformed to their associated ranks before being
subjected to ANOVA. In other words, it’s like ANOVA, except that it is
computed with medians and not means. It can also be viewed as a test of
medians.

The null and alternative hypotheses may be stated as:

H0: the population medians are equal

H1: the population medians differ

The Kruskal- Wallis one way Analysis of Variance by ranks (Kruskal, 1952)
and (Kruskal and Wallis, 1952) is employed with ordinal (rank order) data in
a hypothesis testing situation involving a design with two or more
independent [Link] test is an extension of the Mann-Whitney U test
(Test 12) to a design involving more than two independent samples and,
when k = 2, the Kruskal- Wallis one way Analysis of Variance by ranks will
yield a result that is equivalent to that obtained with the Mann-Whitney U
test.

If the result of the Kruskal- Wallis one way Analysis of Variance by ranks is
significant, it indicates there is a significant difference between at least two of
the sample medians in the set of k medians. As a result of the latter, the
researcher can conclude there is a high likelihood that at least two of the
samples represent populations with different median values.

In employing the Kruskal- Wallis one way Analysis of Variance by ranks one
of the following is true with regard to the rank order data that are evaluated:

a) The data are in a rank-order format, since it is the only format in which
scores are available; or
b) The data have been transformed into a rank-order format from an
interval/ratio format, since the researcher has reason to believe that one
or more of the assumptions of the single-factor between-subjects
Analysis of Variance (which is the parametric analog of the Kruskal-
Wallis test) are saliently violated.

It should be noted that when a researcher decides to transform a set of

interval/ ratio data into ranks, information is sacrificed. This latter fact
accounts for why there is reluctance among some researchers to employ non-
parametric tests such as the Kruskal- Wallis oneway Analysis of Variance by
ranks, even if there is reason to believe that one or more of the assumptions
of the single factor between subjects Analysis of Variance have beenviolated.
138
Various sources [Conover (1980, 1999), Daniel (1990), and Marascuilo and Kruskal- Wallis
Analysis of
McSweeney (1977)] note that the Kruskal-Wallis one way Analysis of Variance
Variance by ranks is based on the following assumptions:

a) Each sample has been randomly selected from the population it

represents;
b) The k samples are independent of one another;
c) The dependent variable (which is subsequently ranked) is a continuous
random variable. In truth, this assumption, which is common to many
non-parametric tests, is often not adhered to, in that such tests are often
employed with a dependent variable which represents a discrete random
variable; and
d) The underlying distributions from which the samples are derived are
identical in shape.

The shapes of the underlying population distributions, however, do not have

to be normal.

Maxwell and Delaney (1990) point out that the assumption of identically
shaped distributions implies equal dispersion of data within each distribution.
Because of this, they note that, like the single factor between subjects
Analysis of Variance , the Kruskal- Wallis one way Analysis of Variance by
ranks assumes homogeneity of variance with respect to the underlying
population distribution. Because the latter assumption is not generally
acknowledged for the Kruskal- Wallis one way Analysis of Variance by
ranks, it is not uncommon for sources to state that violation of the
homogeneity of variance assumption justifies use of the Kruskal- Wallis one
way Analysis of Variance by ranks in lieu of the single factor between
subjects Analysis of Variance.

It should be pointed out, however, that there is some empirical research

which suggests that the sampling distribution for the Kruskal- Wallis test
statistic is not as affected by violation of the homogeneity of variance
assumption as is the F distribution (which is the sampling distribution for the
single-factor between- subjects Analysis of Variance ).

One reason cited by various sources for employing the Kruskal- Wallis one
way Analysis of Variance by ranks is that by virtue of ranking interval/ratio
data a researcher can reduce or eliminate the impact of outliers. In t test for
two independent samples, since outliers can dramatically influence
variability, they can be responsible for heterogeneity of variance between two
or more samples. In addition, outliers can have a dramatic impact on the
value of a sample mean.

139
Non-Parametric
Statistics Check Your Progress I

1) List any one assumption of Kruskal- Wallis one way ANOVA.

9.3 COMPUTATION OF KRUSKAL-

WALLISANOVA
Let us us now understand the computation of Kruskal-Wallis one way ANOVA
with the help of steps and example.

Step 1:Rank all the numbers in the entire data set from smallest to largest (using all
samples combined); in the case of ties, use the average of the ranks that the values
would have normally been given.

Step 2:Total the ranks for each of the samples; call those totals T1, T2, . . ., Tk,
where k is the number of groups or populations.

Step 3: Calculate the Kruskal-Wallis test statistic with the help of following
formula:

H=[12/N(N+1)][Σ((ΣR)2 /n)]–3(N+1)

N = the total number of cases

n = the number of cases in a given group

(ΣR)2 = the sum of the ranks squared for a given group of subjects

Step 4: Find the p-value.

Step 5: Make your conclusion about whether you can reject Ho by examining
the p-value.

Example of a Small Sample: In a Study, 12 participants were divided into

three groups of 4 each, they were subjected to three different conditions, A
(Low Noise), B (Average Noise), and C (Loud Noise). They were given a test
and the errors committed by them on the test were noted and are given in the
table below.

140
 Kruskal- Wallis
Participant Condition A Participant Condition B Participant Condition Analysis of
No. (Low Noise) No. No. C (Loud Variance
(Average
Noise) Noise)

1 3 5 2 9 10

2 5 6 7 10 8

3 6 7 9 11 7

4 3 8 8 12 11

The researcher wishes to know whether these three conditions differ amongst
themselves. and there are no assumptions of the probability. To apply
Kruskal- Wallis test, following steps would be taken:

Step 1: Rank all the numbers in the entire data set from smallest to largest
(using all samples combined); in the case of ties, use the average of the ranks
that the values would have normally been given.

Condition Ranks T 1 Condition Ranks T2 Condition Ranks T3

A B C

3 2.5 2 1 10 11

5 4 7 6.5 8 8.5

6 5 9 10 7 6.5

3 2.5 8 8.5 11 12

T1 = 14 T2 = 26 T3 = 38

Step 2: Total the ranks for each of the samples; call those totals T1, T2, . . .,
Tk, where k is the number of populations. T1 =14, T2=26, T3=38

Step3: Calculate H

H = [ 12 / N (N+1) ] [ ((R)2 / n) ] – 3(N + 1)

N = 12
n=4
2 2
(R) = (14+ 26+ 38) =6084
2 2 2
H= [12/ 12 (12 + 1) ] [ (14 /4) + (26 /4) + (38 /4)] – 3 (12+ 1)
H= [12/156] [49 + 169 + 361] – 39
H= (0.076 x 579) – 39
H= 44.525 – 39
H= 5.537
141
Worked Example for a large sample: A state court administrator asked the Kruskal- Wallis
Analysis of
24 court coordinators in the state’s three largest counties to rate their relative Variance
need for training in case flow management on a Likert scale (1 to 7).

1 = no training need

7 = critical training need

Training Need of Court Coordinators

District A District District

B C

3 7 4

1 6 2

3 5 5

1 7 1

5 3 6

4 1 7

4 6

2 4

Step 1: Rank order the total groups’ Likert scores from lowest to highest.

If tied scores are encountered, sum the tied positions and divide by the
number of tied scores. Assign this rank to each of the tied scores.

Scores & Ranks Across the Three Counties

Ratings Ranks Ratings Ranks

1 2.5 4 12

1 2.5 4 12

1 2.5 5 16.5

1 2.5 5 16.5

2 5.5 5 16.5

143
Non-Parametric
Statistics 2 5.5 5 16.5

3 8 6 20

3 8 6 20

3 8 6 20

4 12 7 23

4 12 7 23

4 12 7 23

Calculating the ranks of tied scores

Example: Three court administrators rated their need for training as a 3.

These threescores occupy the rank positions 7, 8, & 9. (7 + 8 + 9) / 3 = 8

Step 2: Sum the ranks for each group and square the sums

District A District B District C

Rating Rank Rating Rank Rating Rank

3 8 7 23 4 12

1 2.5 6 20 2 5.5

3 8 5 16.5 5 16.5

1 2.5 7 23 1 2.5

5 16.5 3 8 6 20

4 12 1 2.5 7 23

4 12 6 20

2 5.5 4 12

4 12

5 16.5

R 67.0 153.5 79.5

( R)2 4489 23562.25 6320.25

Step3: CalculateH
2
H = [ 12 / N (N+1) ] [ ((R) / n) ] – 3(N + 1)
144 H = [ 12 / 24 (24+1) ][4489 / 8 + 23562.25 / 10 + 6320.25 / 6] – 3 (24 + 1)
Non-Parametric
Statistics Group 1 Group 2 Group 3 Group1 Group2 Group3
(Ranks) (Ranks) (Ranks)

3 6 9 1 3.5 10

5 7 10 2 5.5 13

6 8 11 3.5 7.5 15.5

7 9 12 5.5 10 17

8 10 15 7.5 13 18

9 10 10 13

11 15.5

38 61 57 Total 29.5 68 73.5

ANOVA
n1=6 n2=7 n3 =5
X1= 38 X2=61 X3 =57
( X1 )2=264  (X2 )2=551  (X3 )2 =671
SStotal = (264 + 551 + 671) – [(38 + 61+ 57)2 / 18] =134
SSBetween Group = (382/6) + (612/7) + (572/5) – [(38 + 61+ 57)2 / 18]
SSWithin Group = [ 264 - (382/6)] + [ 551- (612/7)] + [ 671- (572/5)] = 63.962

Source of SS df MS F ratio F Test

Variation critical Decision
Value

Between 70.038 2 35.019 8.213 3.68 Reject

Groups H0

Within 63.962 15 4.264

Groups

Total 134.000 17

Kruskal-Wallis H test:

H = [12 / 18(18+1)] [ (29.52/6) + (682/7) + (73.52/5)] – [3 (18+1)]

H= 66.177 – 57 = 9.177

Chi Square for Degrees of freedom 2 (3 – 1) is 5.99, Therefore reject the Ho

146
In both the case, ANOVA or Kruskal-Wallis ANOVA, we will reject the Null Kruskal- Wallis
Analysis of
Hypothesis,and state that the three groups differ. Variance

F ratio as a function of H:The Fisher’s F or F ratio or ANOVA one way

variance is equivalent to H test or Kruskal-Wallis test or Kruskal-Wallis
ANOVA, or ANOVA by rank order. This can also be seen in book by Iman
and Conover (1981).

Where the rank transform statistics states:

F= [{(k-1)/(N-k)} {((N-1)/H)-1}]-1
If We see from the above mentioned example F was 8.213 and H was 9.177
F= [{(3-1)/(18-3)} {((18-1)/9.177)-1}]-1 F= [(2/15) {(17/9.177)-1}]-1
F= [0.133 x (1.852-1)] -1 = 0.1214-1
F= 8.231

Check Your Progress IV

1) Fill in the blanks:

a) The Kruskal- Wallis (KW)ANOVAis the non-parametric equivalent of a
………………………...
b) The ……………………….. is basically an extension of the Wilcoxon-
Mann-Whitney (WMW) 2 sample test, and so has the same assumptions:
1) the groups have the same spreads; and 2) the data distributions have
the same shape.
c) ANOVA compares ……………………….. to indicate the similarity
between the populations KWANOVA compares medians of these
populations.

9.6 LET US SUM UP

Kruskal-Wallis Analysis of Variance (KW ANOVA) is a non-parametric
analogue of ANOVA one way variance for independent sample. Kruskal-
Wallis ANOVA is used when there are more than 2 groups (k > 2). The
assumption of KW ANOVA are that: a) Each sample has been randomly
selected from the population it represents; b) The k samples are independent
of one another; c) The dependent variable (which is subsequently ranked) is a
continuous random variable. In truth, this assumption, which is common to
many non-parametric tests, is often not adhered to, in that such tests are often
employed with a dependent variable which represents a discrete random
variable; and d) the underlying distributions from which the samples are
derived are identical in shape.

The ANOVA or F test and KW ANOVA Or H test are equivalent to each

other and can be appropriately used depending upon the type of population in
question.
147
Non-Parametric
Statistics 9.7 REFERENCES
Daniel, W. W. (1990) Applied Non-parametric Statistics, 2d ed. Boston:
PWS- Kent.
Iman, R. L., and W. J. Conover (1981), Rank transformations as a bridge
between parametric and non-parametric statistics, The American Statistician,
35, 124–129.
Johnson, Morrell, and Schick (1992), Two-Sample Non-parametric
Estimation and Confidence Intervals Under Truncation, Biometrics, 48,
1043-1056.
Leach, C. (1979). Introduction to statistics: A non-parametric approach for
the social sciences. Chichester: John Wiley & Sons
Lehman, E. L. (1975). Non-parametric statistical methods based on ranks.
San Francisco: Holden-Day.
Siegel S. and Castellan N.J. (1988) Non-parametric Statistics for the
Behavioral Sciences (2nd edition). New York: McGraw Hill.
Wampold BE & Drew CJ. (1990) Theory and application of statistics. New
York: McGraw-Hill.

9.8 KEY WORD

Kruskal-Wallis One way Analysis of Variance (KW ANOVA): Kruskal-
Wallis One way Analysis of Variance (KW ANOVA) is a Non-parametric
Analogue of ANOVA one way variance for independent sample. Kruskal-
Wallis ANOVA is used when there are more than 2 groups (k > 2).

9.9 ANSWERS TO CHECK YOUR PROGRESS

Check Your Progress I

1) List any one assumption of Kruskal-Wallis One Way ANOVA.

One of the assumptions of Kruskal-Wallis One Way ANOVA is:
Each sample has been randomly selected from the population it
represents;

Check Your Progress II

1) What is the formula for Kruskal-Wallis test statistic?

H=[12/N(N+1)][Σ((ΣR)2 /n)]–3(N+1)

Check Your Progress III

1) Fill in the blanks:

a) As the values of k and N increase, the chi-square distribution provides a
148
 more accurate estimate of the exact Kruskal-Wallis distribution. Kruskal- Wallis
Analysis of
b) Use of the chi-square distribution for small sample sizes will generally Variance

result in a slight decrease in the power of the test

Check Your Progress IV

1) Fill in the blanks:

b) The Kruskal-Wallis (KW)ANOVAis the non-parametric equivalent of a
one-way ANOVA.
c) The KWANOVA is basically an extension of the Wilcoxon-Mann-
Whitney (WMW) 2 sample test, and so has the same assumptions: 1) the
groups have the same spreads; and 2) the data distributions have the
same shape.
d) ANOVA compares means of different populations to indicate the
similarity between the populations KWANOVA compares medians of
these populations.

9.10 UNIT END QUESTIONS

1) Explain Kruskal–Wallis ANOVA test.
2) A firm wishes to compare four programs for training workers to perform
a certain manual task. Twenty new employees are assigned to the
training programs, with 5 in each program. At the end of the training
period, a test is conducted to see how quickly trainees can perform the
task. The number of times the task is performed per minute is recorded
for each trainee, with the following results:

Observation Program Program Program Program

1 2 3 4

1 9 10 12 9

2 12 6 14 8

3 14 9 11 11

4 11 9 13 7

5 13 10 11 8

Calculate H, and report your results appropriately.

3) Compare between ANOVA and Kruskal-Wall is ANOVA Test

149
Non-Parametric
Statistics

150