MST-004

STATISTICAL
Indira Gandhi
National Open University INFERENCE
School of Sciences

Block

3
TESTING OF HYPOTHESIS
UNIT 9
Concepts of Testing of Hypothesis 5
UNIT 10
Large Sample Tests 25
UNIT 11
Small Sample Tests 63
UNIT 12
Chi-Square and F-Tests 91
Curriculum and Course Design Committee
Prof. K. R. Srivathasan Prof. Rahul Roy
Pro-Vice Chancellor Math. and Stat. Unit
IGNOU, New Delhi Indian Statistical Institute, New Delhi

Prof. Parvin Sinclair Dr. Diwakar Shukla

Pro-Vice Chancellor Department of Mathematics and Statistics
IGNOU, New Delhi Dr. Hari Singh Gaur University, Sagar

Prof. Geeta Kaicker Prof. Rakesh Srivastava

Director, School of Sciences Department of Statistics
IGNOU, New Delhi M.S. University of Baroda, Vadodara

Prof. Jagdish Prasad Prof. G. N. Singh

Department of Statistics Department of Applied Mathematics
University of Rajasthan, Jaipur I.S.M., Dhanbad

Prof. R. M. Pandey Dr. Gulshan Lal Taneja

Department of Bio-Statistics Department of Mathematics
All India Institute of Medical Sciences M.D. University, Rohtak
New Delhi
Faculty members of School of Sciences, IGNOU
Statistics Mathematics
Dr. Neha Garg Dr. Deepika
Dr. Nitin Gupta Prof. Poornima Mital
Mr. Rajesh Kaliraman Prof. Sujatha Varma
Dr. Manish Trivedi Dr. S. Venkataraman

Block Preparation Team

Dr. Ramkishan (Editor) Dr. Diwakar Shukla (Units 1 - 4)
Department of Statistics Department of Mathematics and Statistics
D. A. V. (PG) College Dr. Hari Singh Gaur University, Sagar
C.C. S. University, Merrut
Mr. Prabhat Kumar Sangal (Units 1 - 4)
Dr. Parmod Kumar (Language Editor) School of Sciences, IGNOU
School of Humanities, IGNOU

Course Coordinator: Mr. Prabhat Kumar Sangal

Programme Coordinator: Dr. Manish Trivedi

Block Production
Mr. Sunil Kumar, AR (P),School of Sciences, IGNOU
CRC prepared by Mr. Prabhat Kumar Sangal, School of Sciences, IGNOU

Acknowledgement: I gratefully acknowledge my colleague Mr. Rajesh Kaliraman, Statistics

Discipline, School of Sciences for their great support.

July, 2013
© Indira Gandhi National Open University, 2013
ISBN-978-81-266-
All rights reserved. No part of this work may be reproduced in any form, by mimeograph or any other
means, without permission in writing from the Indira Gandhi National Open University.

Further information on the Indira Gandhi National Open University may be obtained from University’s
Office at Maidan Garhi, New Delhi-110068 or visit University’s website http://www.ignou.ac.in

Printed and published on behalf of the Indira Gandhi National Open University, New Delhi by the
Director, School of Sciences.

Printed at:
TESTING OF HYPOTHESIS
In previous block of this course, we have discussed one part of statistical
inference, that is, estimation and we have learnt how we estimate the unknown
population parameter(s) by using point estimation and interval estimation.
In this block we will focus on the second part of statistical inference which is
known as testing of hypothesis.
In this block, you will study the basic concept of testing of hypothesis and
different kinds of well-known tests. At the end of this block you will be aware
of the idea, procedure and applications of the hypothesis testing. This block
contains four units.
Unit 9: Concepts of Testing of Hypothesis
This unit explains the fundamental aspect relating to testing of hypothesis with
examples. It describes basic concepts and methodologies as hypothesis,
critical region, types of errors, level of significance, general procedure of
testing a hypothesis, concept of p-value, relation between confidence interval
and testing of hypothesis.
Unit 10: Large Sample Tests
This unit explores the procedure of testing the hypothesis based on one sample
and two samples when sample size is large and taken from normal or non-
normal population(s). In this unit, you will learn about Z-test for testing of
hypothesis about mean, difference of two means, proportion, difference of two
proportions, variance and equality of two variances.
Unit 11: Small Sample Tests
This unit is devoted to test a hypothesis based on one sample and two samples
when sample sizes are small. In this unit, you will learn about t-test for testing
of hypothesis about mean and difference of two means. This unit also explores
the idea for testing the hypothesis for equality of two means when samples are
dependent and testing of hypothesis about population correlation coefficient.
Unit 12: Chi-square and F Tests
The last unit of this block describes the chi-square test for population variance
and F-test for equality of two population variances.
Notations and Symbols
X1, X2, …, Xn : Random sample of size n
H0 : Null hypothesis
H1 or HA : Alternative hypothesis
 : Critical (rejection) region
 : Non-rejection region
α : Size of critical region or type-I error or level of significance
 : Type-II error
1‒ : Power of the test
z : Critical value of Z-test at α level of significance

t   ,  : Critical value of t-test with ν degrees of freedom at α level

of significance
2,  : Critical value of χ2-test with ν degrees of freedom at α level
of significance
F 1 , 2 ,  : Critical value of F-test with (ν1, ν2) degrees of freedom at α
level of significance
UNIT 9 CONCEPTS OF TESTING OF
HYPOTHESIS
Structure
9.1 Introduction
Objectives
9.2 Hypothesis
Simple and Composite Hypotheses
Null and Alternative Hypotheses
9.3 Critical Region
9.4 Type-I and Type-II Errors
9.5 Level of Significance
9.6 One-Tailed and Two-Tailed Tests
9.7 General Procedure of Testing a Hypothesis
9.8 Concept of p-Value
9.9 Relation between Confidence Interval and Testing of Hypothesis
9.10 Summary
9.11 Solutions /Answers

9.1 INTRODUCTION
In previous block of this course, we have discussed one part of statistical
inference, that is, estimation and we have learnt how we estimate the unknown
population parameter(s) by using point estimation and interval estimation. In
this block, we will focus on the second part of statistical inference which is
known as testing of hypothesis.
In our day-to-day life, we see different commercials advertisements in
television, newspapers, magazines, etc. such as
(i) The refrigerator of certain brand saves up to 20% electric bill,
(ii) The motorcycle of certain brand gives 60 km/liter mileage,
(iii) A detergent of certain brand produces the cleanest wash,
(iv) Ninety nine out of hundred dentists recommend brand A toothpaste for
their patients to save the teeth against cavity, etc.
Now, the question may arise in our mind “can such types of claims be verified
statistically?” Fortunately, in many cases the answer is “yes”.
The technique of testing such type of claims or statements or assumptions is
known as testing of hypothesis. The truth or falsity of a claim or statement is
never known unless we examine the entire population. But practically it is not
possible in mostly situations so we take a random sample from the population
under study and use the information contained in this sample to take the
decision whether a claim is true or false.
This unit is divided into 11 sections. Section 9.1 is introductory in nature. In
Section 9.2, we defined the hypothesis. The concept and role of critical region
in testing of hypothesis is described in Section 9.3. In Section 9.4, we explored
the types of errors in testing of hypothesis whereas level of significance is
explored in Section 9.5. In Section 9.6, we explored the types of tests in testing
5
Testing of Hypothesis of hypothesis. The general procedure of testing a hypothesis is discussed in
Section 9.7. In Section 9.8, the concept of p-value in decision making about the
null hypothesis is discussed whereas the relation between confidence interval
and testing of hypothesis is discussed in Section 9.9. Unit ends by providing
summary of what we have discussed in this unit in Section 9.10 and solution of
exercises in Section 9.11.
Objectives
After reading this unit, you should be able to:
 define a hypothesis;
 formulate the null and alternative hypotheses;
 explain what we mean by type-I and type-II errors;
 explore the concept of critical region and level of significance;
 define one-tailed and two-tailed tests;
 describe the general procedure of testing a hypothesis;
 concept of p-value; and
 test a hypothesis by using confidence interval.
Before coming to the procedure of testing of hypothesis, we will discuss the
basis terms used in this procedure one by one in subsequent sections.

9.2 HYPOTHESIS
As we have discussed in previous section that in our day-to-day life, we see
different commercials advertisements in television, newspapers, magazines,
etc. and if someone may be interested to test such type of claims or statement
then we come across the problem of testing of hypothesis. For example,
(i) a customer of motorcycle wants to test whether the claim of motorcycle
of certain brand gives the average mileage 60 km/liter is true or false,
(ii) the businessman of banana wants to test whether the average weight of a
banana of Kerala is more than 200 gm,
(iii) a doctor wants to test whether new medicine is really more effective for
controlling blood pressure than old medicine,
(iv) an economist wants to test whether the variability in incomes differ in
two populations,
(v) a psychologist wants to test whether the proportion of literates between
two groups of people is same, etc.
In all the cases discussed above, the decision maker is interested in making
inference about the population parameter(s). However, he/she is not interested
in estimating the value of parameter(s) but he/she is interested in testing a
claim or statement or assumption about the value of population parameter(s).
Such claim or statement is postulated in terms of hypothesis.
In statistics, a hypothesis is a statement or a claim or an assumption about the
value of a population parameter (e.g., mean, median, variance, proportion,
etc.).
Similarly, in case of two or more populations a hypothesis is comparative
statement or a claim or an assumption about the values of population
parameters. (e.g., means of two populations are equal, variance of one
population is greater than other, etc.). The plural of hypothesis is hypotheses.
6
In hypothesis testing problems first of all we should being identifying the claim Concepts of Testing of
or statement or assumption or hypothesis to be tested and write it in the words. Hypothesis
Once the claim has been identified then we write it in symbolical form if
possible. As in the above examples,
(i) Customer of motorcycle may write the claim or postulate the hypothesis
“the motorcycle of certain brand gives the average mileage 60 km/liter.”
Here, we are concerning the average mileage of the motorcycle so let µ
represents the average mileage then our hypothesis becomes µ = 60 km /
liter.
(ii) Similarly, the businessman of banana may write the statement or
postulate the hypothesis “the average weight of a banana of Kerala is
greater than 200 gm.” So our hypothesis becomes µ > 200 gm.
(iii) Doctor may write the claim or postulate the hypothesis “ the new
medicine is really more effective for controlling blood pressure than old
medicine.” Here, we are concerning the average effect of the medicines
so let µ1 and µ2 represent the average effect of new and old medicines
respectively on controlling blood pressure then our hypothesis becomes
µ1 > µ2.
(iv) Economist may write the statement or postulate the hypothesis “ the
variability in incomes differ in two populations.” Here, we are concerning
the variability in income so let 12 and 22 represent the variability in
incomes in two populations respectively then our hypothesis becomes
12  22 .
(v) Psychologist may write the statement or postulate the hypothesis “the
proportion of literates between two groups of people is same.” Here, we
are concerning the proportion of literates so let P1 and P2 represent the
proportions of literates of two groups of people respectively then our
hypothesis becomes P1 = P2 or P1 –P2 = 0.
The hypothesis is classified according to its nature and usage as we will discuss
in subsequent subsections.
9.2.1 Simple and Composite Hypotheses
In general sense, if a hypothesis specifies only one value or exact value of the
population parameter then it is known as simple hypothesis. And if a
A hypothesis which
hypothesis specifies not just one value but a range of values that the population completely specifies
parameter may assume is called a composite hypothesis. parameter(s) of a
As in the above examples, the hypothesis postulated in (i) µ = 60 km/liter is theoretical population
(probability distribution)
simple hypothesis because it gives a single value of parameter (µ = 60) is called a simple
whereas the hypothesis postulated in (ii) µ > 200 gm is composite hypothesis hypothesis otherwise
because it does not specify the exact average value of weight of a banana. It called composite
may be 260, 350, 400 gm or any other. hypothesis.

Similarly, (iii) µ1 > µ2 or µ1 −µ2 > 0 and (iv) 12  22 or 12  22  0 are not
simple hypotheses because they specify more than one value as µ1 −µ2 = 4,
µ1 −µ2 = 7, 12  22  2, 12  22  5 , etc. and (v) P1 = P2 or P1 –P2 = 0 is simple
hypothesis because it gives a single value of parameter as P1 –P2 = 0.
9.2.2 Null and Alternative Hypotheses
As we have discussed in last page that in hypothesis testing problems first of
all we identify the claim or statement to be tested and write it in symbolical
7
Testing of Hypothesis form. After that we write the complement or opposite of the claim or statement
in symbolical form. In our example of motorcycle, the claim is µ = 60 km/liter
then its complement is µ ≠ 60 km/liter. In (ii) the claim is µ > 200 gm then its
complement is µ ≤ 200 gm. If the claim is µ < 200 gm then its complement is
µ ≥ 200 gm. The claim and its complement are formed in such a way that they
cover all possibility of the value of population parameter.
Once the claim and its compliment have been established then we decide of
We state the null and these two which is the null hypothesis and which is the alternative hypothesis.
alternative The thump rule is that the statement containing equality is the null hypothesis.
hypotheses in such a
way that they cover
That is, the hypothesis which contains symbols  or  or  is taken as null
all possibility of the hypothesis and the hypothesis which does not contain equality i.e. contains
value of population  or  or  is taken as alternative hypothesis. The null hypothesis is denoted
parameter.
by H0 and alternative hypothesis is denoted by H1 or HA.
In our example of motorcycle, the claim is µ = 60 km/liter and its complement
is µ ≠ 60 km/liter. Since claim µ = 60 km/liter contains equality sign so we take
it as a null hypothesis and complement µ ≠ 60 km/liter as an alternative
hypothesis, that is,
H0 : µ = 60 km/liter and H1: µ ≠ 60 km/liter
In our second example of banana, the claim is µ > 200 gm and its complement
is µ ≤ 200 gm. Since complement µ ≤ 200 gm contains equality sign so we take
complement as a null hypothesis and claim µ > 200 gm as an alternative
hypothesis, that is,
H0 : µ ≤ 200 gm and H1: µ > 200 gm
Formally these hypotheses are defined as
The hypothesis which we wish to test is called as the null hypothesis.
According to Prof. R.A. Fisher,
“A null hypothesis is a hypothesis which is tested for possible rejection under
the assumption that it is true.”
The hypothesis which complements to the null hypothesis is called alternative
hypothesis.
Note 1: Some authors use equality sign in null hypothesis instead of ≥ and ≤
signs.
The alternative hypothesis has two types:
(i) Two-sided (tailed) alternative hypothesis
(ii) One-sided (tailed) alternative hypothesis
If the alternative hypothesis gives the alternate of null hypothesis in both
directions (less than and greater than) of the value of parameter specified in
null hypothesis then it is known as two-sided alternative hypothesis and if it
gives an alternate only in one direction( less than or greater than) only then it is
known as one-sided alternative hypothesis. For example, if our alternative
hypothesis is H1 : θ ≠ 60 then it is a two-sided alternative hypothesis because its
means that the value of parameter θ is greater than or less than 60. Similarly, if
H1 : θ > 60 then it is a right-sided alternative hypothesis because its means that
the value of parameter θ is greater than 60 and if H1: θ < 60 then it is a
left-sided alternative hypothesis because its means that the value of parameter θ
is less than 60.
In testing procedure, we assume that the null hypothesis is true until there is
sufficient evidence to prove that it is false. Generally, the hypothesis is tested
8
with the help of a sample so evidence in testing of hypothesis comes from a Concepts of Testing of
sample. If there is enough sample evidence to suggest that the null hypothesis Hypothesis
is false then we reject the null hypothesis and support the alternative
hypothesis. If the sample fails to provide us sufficient evidence against the null
hypothesis we are not saying that the null hypothesis is true because here, we
take the decision on the basis of a random sample which is a small part of the
population. To say that null hypothesis is true we must study all observations
of the population under study. For example, if someone wants to test that the
person of India has two hands then to prove that this is true we must check all
the persons of India whereas to prove that it is false we require a person he /
she has one hand or no hand. So we can only say that there is not enough
evidence against the null hypothesis.
Note 2: When we assume that null hypothesis is true then we are actually
assuming that the population parameter is equal to the value in the claim. In our
example of motorcycle, we assume that µ = 60 km/liter whether the null
hypothesis is µ = 60 km/liter or µ ≤ 60 km/liter or µ ≥ 60 km/liter.
Now, you can try the following exercises.
E1) A company manufactures car tyres. Company claims that the average life
of its tyres is 50000 miles. To test the claim of the company, formulate
the null and alternative hypotheses.
E2) Write the null and alternative hypotheses in case (iii), (iv) and (v) of our
example given in Section 9.2.
E3) A businessman of orange formulates different hypotheses about the
average weight of the orange which are given below:
(i) H0:  = 100 (ii) H1 :  >100 (iii) H0 :  ≤ 100 (iv) H1:  ≠ 100
(v) H1:  > 150 (vi) H0:  = 130 (vii) H1:   0
Categorize the above cases into simple and composite hypotheses.
After describing the hypothesis and its types our next point in the testing of
hypothesis is critical region which will be described in next section.

9.3 CRITICAL REGION

As we have discussed in Section 9.1 that generally, null hypothesis is tested by
the sample data. Suppose X1, X2,…, Xn be a random sample drawn from a
population having unknown population parameter . The collection of all
possible values of X1, X2,…, Xn is a set called sample space(S) and a particular
value of X1, X2,…, Xn represents a point in that space. A statistic is a function
In order to test a hypothesis, the entire sample space is partitioned into two of sample observations
(not including
disjoint sub-spaces, say,  and S    . If calculated value of the test parameter). Basically, a
statistic lies in ω , then we reject the null hypothesis and if it lies in ω , then we test statistic is a statistic
do not reject the null hypothesis. The region  is called a “rejection region or which is used in
critical region” and the region  is called a “non-rejection region”. decision making about
the null hypothesis.
Therefore, we can say that
“A region in the sample space in which if the calculated value of the test
statistic lies, we reject the null hypothesis then it is called critical region or
rejection region.”
This can be better understood with the help of an example. Suppose, 100
students appear in total 10 papers two of each in English, Physics, Chemistry,
Mathematics and Computer Science of a Programme. Suppose the scores in
Testing of Hypothesis these papers are denoted by X1, X2, …, X10 and maximum marks =100 for each
paper. For obtaining the distinction award in this Programme, a student needs
to have total score equal to or more than 750 which is a rule.
Suppose, we select one student randomly out of 100 students and we want to
test that the selected student is a distinction award holder. So we can take the
null and alternative hypotheses as
H0 : Selected student is a distinction award holder
H1 : Selected student is not a distinction award holder
For taking the decision about the selected student, we define a statistic
10
T10  X
i 1
i as the sum of the scores in all the 10 papers of the student. The

range of T10 is 0  T10  1000. Now, we divide the whole space ( 0-1000) into
two regions as no-distinction awarded region (less than 750) and distinction
awarded region (greater than or equal to 750) as shown in Fig. 9.1. Here, 750 is
the critical value which separates the no-distinction and distinction awarded
regions.

Tn
Fig. 9.1: Non-rejection and critical regions for distinction award
On the basis of scores in all the papers of the selected student, we calculate the
10
value of the statistic T10   Xi . And calculated value may fall in distinction
i 1
award region or not, depending upon the observed value of test statistic.
For making a decision to reject or do not reject H0, we use test statistic
10
T10  X
i 1
i (sum of scores of 10 papers). If calculated value of test statistic T10

lies in no-distinction awarded region (critical region), that is, T10 < 750 then we
reject H0 and if calculated value of test statistic T10 lies in distinction awarded
region (non-rejection region), that is, T10  750 then we do not reject H0. It is a
basic structure of the procedure of testing of hypothesis which needs two
regions like:
(i) Region of rejection of null hypothesis H0
(ii) Region of non-rejection of null hypothesis H0
The point of discussion in this test procedure is “how to fix the cut off value
750”? What is the justification for this value? The distinction award region
may be like T10  800 or at T10  850 or at T10  900. So, there must be a
scientific justification for the cut-off value 750. In a statistical test procedure it
is obtained by using the probability distribution of the test statistic.
The region of rejection is called critical region. It has a pre-fixed area generally
denoted by , corresponding to a cut-off value in a probability distribution of
test statistic.
The rejection (critical) region lies in one-tail or two-tails on the probability
curve of sampling distribution of the test statistic its depends upon the
alternative hypothesis. Therefore, three cases arise: Concepts of Testing of
Hypothesis
Case I: If the alternative hypothesis is right-sided such as H1 : θ > θ0 or
H1 : θ1 > θ2 then the entire critical or rejection region of size α lies on
right tail of the probability curve of sampling distribution of the test
statistic as shown in Fig. 9.2.

Critical value is a
value or values that
separate the region of
rejection from the non-
rejection region.

Fig. 9.2
Case II: If the alternative hypothesis is left-sided such as H1: θ < θ0 or
H1 : θ1 < θ2 then the entire critical or rejection region of size α lies on
left tail of the probability curve of sampling distribution of the test
statistic as shown in Fig. 9.3.

Fig. 9.3
Case III: If the alternative hypothesis is two sided such as H1 : θ ≠ θ0 or
H1 : θ1 ≠ θ2 then critical or rejection regions of size α/2 lies on both
tails of the probability curve of sampling distribution of the test
statistic as shown in Fig. 9.4.

Fig. 9.4
Now, you can try the following exercise.
E4) If H0: θ = 60 and H1: θ ≠ 60 then critical region lies in one-tail or two-
tails.
Testing of Hypothesis
9.4 TYPE-I AND TYPE-II ERRORS
In Section 9.3, we have discussed a rule that if the value of test statistic falls in
rejection (critical) region then we reject the null hypothesis and if it falls in the
non-rejection region then we do not reject the null hypothesis. A test statistic is
calculated on the basis of observed sample observations. But a sample is a
small part of the population about which decision is to be taken. A random
sample may or may not be a good representative of the population.
A faulty sample misleads the inference (or conclusion) relating to the null
hypothesis. For example, an engineer infers that a packet of screws is sub-
standard when actually it is not. It is an error caused due to poor or
inappropriate (faulty) sample. Similarly, a packet of screws may infer good
when actually it is sub-standard. So we can commit two kinds of errors while
testing a hypothesis which are summarised in Table 9.1 which is given below:
Table 9.1: Type of Errors
Decision H0 True H1 True
Reject H0 Type-I Error Correct Decision
Do not reject H0 Correct Decision Type-II Error

Let us take a situation where a patient suffering from high fever reaches to a
doctor. And suppose the doctor formulates the null and alternative hypotheses
as
H0 : The patient is a malaria patient
H1 : The patient is not a malaria patient
Then following cases arise:
Case I: Suppose that the hypothesis H0 is really true, that is, patient actually
a malaria patient and after observation, pathological and clinical
examination, the doctor rejects H0, that is, he / she declares him / her
a non-malaria-patient. It is not a correct decision and he / she
commits an error in decision known as type-I error.
Case II: Suppose that the hypothesis H0 is actually false, that is, patient
actually a non-malaria patient and after observation, the doctor
rejects H0, that is, he / she declares him / her a non-malaria-patient. It
is a correct decision.
Case III: Suppose that the hypothesis H0 is really true, that is, patient actually
a malaria patient and after observation, the doctor does not reject H0,
that is, he / she declares him / her a malaria-patient. It is a correct
decision.
Case IV: Suppose that the hypothesis H0 is actually false, that is, patient
actually a non-malaria patient and after observation, the doctor does
not reject H0, that is, he / she declares him / her a malaria-patient. It
is not a correct decision and he / she commits an error in decision
known as type-II error.
Thus, we formally define type-I and type-II errors as below:
Type-I Error:
The decision relating to rejection of null hypothesis H0 when it is true is called
type-I error. The probability of committing the type-I error is called size of test,
denoted by  and is given by
 = P [Reject H0 when H0 is true] = P [Reject H0 / H0 is true]

12
We reject the null hypothesis if random sample / test statistic falls in rejection Concepts of Testing of
region, therefore, Hypothesis

α = P [ X   / H0 ]
where X = (X1, X2,…,Xn) is a random sample and ω is the rejection region and
1- = 1-P[Reject H0 / H0 is true]
= P[Do not reject H0 / H0 is true] = P[Correct decision]
The (1-) is the probability of correct decision and it correlates to the concept
of 100(1-)% confidence interval used in estimation.
Type-II Error:
The decision relating to non-rejection of null hypothesis H0 when it is false
(i.e. H1 is true) is called type-II error. The probability of committing type-II
error is generally denoted by  and is given by
 = P[Do not reject H0 when H0 is false]
= P[Do not reject H0 when H1 is true]
= P[Do not reject H0 / H1 is true]
= P[ X  ω / H1 ] where,  is the non-rejection region.
and
1- = 1-P[Do not reject H0 / H1 is true]
= P[Reject H0 / H1 is true] = P[Correct decision]
The (1-) is the probability of correct decision and also known as “power of
the test”. Since it indicates the ability or power of the test to recognize
correctly that the null hypothesis is false, therefore, we wish a test that yields a
large power.
We say that a statistical test is ideal if it minimizes the probability of both types
of errors and maximizes the probability of correct decision. But for fix sample
size,  and  are so interrelated that the decrement in one results into the
increment in other. So minimization of both probabilities of type-I and type-II
errors simultaneously for fixed sample size is not possible without increasing
sample size. Also both types of errors will be at zero level (i.e. no error in
decision) if size of the sample is equal to the population size. But it involves
huge cost if population size is large. And it is not possible in all situations such
as testing of blood.
Depending on the problem in hand, we have to choose the type of error which
has to minimize. For this, we have to look at a situation, suppose there is a
decision making problem and there is a rule that if we make type-I error, we
lose10 rupees and if we make type-II error we lose 1000 rupees. In this case,
we try to eliminate the type-II error, since it is more expensive.
In another situation, suppose the Delhi police arrests a person whom they
suspect is a murderer. Now, policemen have to test hypothesis:
H0: Arrested person is innocent (not murderer)
H1: Arrested person is a murderer
The type-I error is
 = P [Reject H0 when it is true]
That is, suspected person who is actually an innocent will be sent to jail when
H0 rejects, although H0 being a true.
Testing of Hypothesis The type-II error is
 = P [Do not reject H0 when H1 is true]
That is, when arrested person truly a murderer but released by the police. Now,
we see that in this case type-I error is more serious than type-II error because a
murderer may be arrested / punished later on but sending jail to an innocent
person is serious.
Consider another situation, suppose we want to test the null hypothesis
H0 : p  0.5 against H1 : p  0.5 on the basis of tossing a coin once, where p is
the probability of getting a head in a single toss (trial). And we reject the null
hypothesis if a head appears and do not reject otherwise. The type-I error, that
is, the probability of Reject H0 when it is true can be calculated easily(as shown
in Example 1) but the computation of type-II error is not possible because there
are infinitely many alternatives for p such as p = 0.6, p = 0.1, etc.
Generally, strong control on α is necessary. It should be kept as low as
possible. In test procedure, we prefix it at very low level like  = 0.05 ( 5%) or
0.01 (1%) .
Now, it is time to do some examples relating to α and .
Example 1: It is desired to test a hypothesis H 0 :p  p0  1/ 2 against the
alternative hypothesis H1 :p  p1  1/ 4 on the basis of tossing a coin once,
where p is the probability of “getting a head” in a single toss (trial) and
agreeing to reject H0 if a head appears and accept H0 otherwise. Find the value
of  and .
Solution: In such type of problems, first of all we search for critical region.
Here, we have critical region  = {head}
Therefore, the probability of type-I error can be obtained as
 = P[Reject H0 when H0 is true]
 P[X   / H 0 ]= P[Head appears / H 0 ]
1  H0 is trueso we take value 
 P  Head appears  1 
p
2 2  of parameter pgiven in H 0 
Also,
 = P[Do not reject H0 when H1 is true]
 P  X   / H1   P [Tail appears / H1 ]

 P  Tail appears   H1 is trueso we take value 

p
1
4 of parameter p given in H1 
1 3
 1  P  Head appears  1  1  
p
4 4 4
Example 2: For testing H0 :  = 1 against H1 :  = 2, the pdf of the variable is
given by
1
 ; 0x
f x ,    
 0; elsewhere

Obtain type-I and type-II errors when critical region is X  0.4. Also obtain
power function of the test.

14
Solution: Here, we have critical (rejection) and non-rejection regions as Concepts of Testing of
Hypothesis
  X : X  0.4 and   X : X  0.4
We have to test the null hypothesis
H 0 : θ  1 against H1 : θ  2
The size of type-I error is given by
  P  X   / H 0   P  X  0.4 /   1
   

   f  x,  dx   
 P X  a   f  x,  dx
 … (1)
0.4 1  a 
1
Now, by using f x ,    ; 0  x  , we get from equation (1)

 1  1
1
    dx    dx   x 0.4  1  0.4  0.6
 0.4   1 0.4
Similarly, the size of type-II error is given by
  P  X   / H1   P  X  0.4 /   2 

 0.4 1  0.4
1 1 0.4 1
    dx    dx   x 0   0.4  0   0.2
 0    2 0 2 2 2

The power function of the test = 1- = 1- 0.2 = 0.8.

Now, you can try the following exercise.
E5) An urn contains either 4 white and 2 black balls or 2 white and 4 black
balls. Two balls are to be drawn from the urn. If less than two white
balls are obtained, it will be decided that this urn contains 2 white and 4
black balls. Calculate the values of  and .

9.5 LEVEL OF SIGNIFICANCE

So far in this unit, we have discussed the hypothesis, types of hypothesis,
critical region and types of errors. In this section, we shall discuss very useful
concept “level of significance”, which play an important role in decision
making while testing a hypothesis.
The probability of type-I error is known as level of significance of a test. It is
also called the size of the test or size of critical region, denoted by α. Generally,
it is pre-fixed as 5% or 1% level (α = 0.05 or 0.01). As we have discussed in
Section 9.3 that if calculated value of the test statistic lies in rejection(critical)
region, then we reject the null hypothesis and if it lies in non-rejection region,
then we do not reject the null hypothesis. Also we note that when H0 is rejected
then automatically the alternative hypothesis H1 is accepted. Now, one point of
our discussion is that how to decide critical value(s) or cut-off value(s) for a
known test statistic.
If distribution of test statistic could be expressed into some well-known
distributions like Z, 2, t, F etc. then our problem will be solved and using the
probability distribution of test statistic, we can find the cut-off value(s) that
provides us critical area equal to 5% (or 1%).
15
Testing of Hypothesis Another viewpoint about the level of significance relates to the trueness of the
conclusion. If H0 do not reject at level, say,  = 0.05 (5% level) then a person
will be confident that “concluding statement about H0” is true with 95%
assurance. But even then it may false with 5% chance. There is no cent-percent
assurance about the trueness of statement made for H0.
As an example, if among 100 scientists, each draws a random sample and use
the same test statistic to test the same hypothesis H0 conducting same
experiment, then 95 of them will reach to the same conclusion about H0. But
still 5 of them may differ (i.e. against the earlier conclusion).
Similar argument can be made for, say,  = 0.01 (=1%). It is like when H0 is
rejected at  = 0.01by a scientist , then out of 100 similar researchers who
work together at the same time for the same problem, but with different
random samples, 99 of them would reach to the same conclusion however, one
may differ.
Now, you can try the following exercise.
E6) If probability of type-I error is 0.05 then what is the level of significance?

9.6 ONE-TAILED AND TWO-TAILED TESTS

We have seen in Section 9.3 that rejection (critical) region lies at one-tail or
two-tails on the probability curve of sampling distribution of the test statistic its
depend upon the form of alternative hypothesis. Similarly, the test of testing
the null hypothesis also depends on the alternative hypothesis.
A test of testing the null hypothesis is said to be two-tailed test if the alternative
hypothesis is two-tailed whereas if the alternative hypothesis is one-tailed then
a test of testing the null hypothesis is said to be one-tailed test.
For example, if our null and alternative hypothesis are
H 0 :   0 and H1 :   0
then the test for testing the null hypothesis is two-tailed test because the
alternative hypothesis is two-tailed that means, the parameter θ can take value
greater than θ0 or less than θ0.
If the null and alternative hypotheses are
H 0 :   0 and H1 :   0
then the test for testing the null hypothesis is right-tailed test because the
alternative hypothesis is right-tailed.
Similarly, if the null and alternative hypotheses are
H 0 :   0 and H1 :   0
then the test for testing the null hypothesis is left-tailed test because the
alternative hypothesis is left-tailed.
The above discussion can be summarised in Table 9.2.
Table 9.2: Null and Alternative Hypotheses and Corresponding One-tailed and
Two-tailed Tests
Null Hypothesis Alterative Hypothesis Types of Critical Region / Test
Two-tailed test having critical
H0 : θ  θ0 H1 : θ  θ 0 regions under both tails.
Right-tailed test having critical
H 0 :   0 H1 : θ  θ 0 region under right tail only.
Left- tailed test having critical
H 0 :   0 H1 : θ  θ 0 region under left tail only.

16
Let us do one example based on type of tests. Concepts of Testing of
Hypothesis
Example 3: A company has replaced its original technology of producing
electric bulbs by CFL technology. The company manager wants to compare the
average life of bulbs manufactured by original technology and new technology
CFL. Write appropriate null and alternate hypotheses. Also say about one tailed
and two tailed tests.
Solution: Suppose the average lives of original and CFL technology bulbs are
denoted by 1 and 2 respectively.
If company manager is interested just to know whether any significant
difference exists in average-life time of two types of bulbs then null and
alternative hypotheses will be:
H0 : µ1 = µ2 [average lives of two types of bulbs are same]
H1 : µ1  µ2 [average lives of two types of bulbs are different]
Since alternative hypothesis is two-tailed therefore corresponding test will be
two-tailed.
If company manager is interested just to know whether average life of CFL is
greater than original technology bulbs then our null and alternative hypotheses
will be
H0 : µ1 ≥ µ2

H1: µ1 < µ2  average life of CFLtechnology bulbs 

 is greater than average life of original technology 
Since alternative hypothesis is left-tailed therefore corresponding test will be
left-tailed test.
Now, you can try the following exercises.
E7) If we have null and alternative hypotheses as
H0:  = 0 and H1:   0
then corresponding test will be
(i) left-tailed test (ii) right-tailed test (iii) both-tailed test
Write the correct option.
E8) The test whether one-tailed or two-tailed depends on
(i) Null hypothesis (H0) (ii) Alternative hypothesis (H1)
(iii) Neither H0 nor H1 (iv) Both H0 and H1

9.7 GENERAL PROCEDURE OF TESTING A

HYPOTHESIS
Testing of hypothesis is a huge demanded statistical tool by many discipline
and professionals. It is a step by step procedure as you will see in next three
units through a large number of examples. The aim of this section is just give
you flavor of that sequence which involves following steps:
Step I: First of all, we have to setup null hypothesis H0 and alternative
hypothesis H1. Suppose, we want to test the hypothetical / claimed /

17
Testing of Hypothesis assumed value θ0 of parameter θ. So we can take the null and
alternative hypotheses as
H 0 :   0 and H1 :   0 for two-tailed test
H 0 :   0 and H1 :   0 
or  for one-tailed test 
H 0 :   0 and H1 :   0 
In case of comparing same parameter of two populations of interest,
say, 1 and 2, then our null and alternative hypotheses would be
H0 : 1  2 and H1 : 1  2 for two-tailed test
H 0 : 1  2 and H1 : 1  2 
or   for one-tailed test 
H 0 : 1  2 and H1 : 1  2 
Step II: After setting the null and alternative hypotheses, we establish a
criteria for rejection or non-rejection of null hypothesis, that is,
decide the level of significance (), at which we want to test our
hypothesis. Generally, it is taken as 5% or 1% (α = 0.05 or 0.01).
Step III: The third step is to choose an appropriate test statistic under H0 for
testing the null hypothesis as given below:
Statistic  Value of the parameter under H 0
Test statistic 
Standard error of statistic
After that, specify the sampling distribution of the test statistic
preferably in the standard form like Z (standard normal), 2, t, F or
any other well-known in literature.
Step IV: Calculate the value of the test statistic described in Step III on the
basis of observed sample observations.
Step V: Obtain the critical (or cut-off) value(s) in the sampling distribution
of the test statistic and construct rejection (critical) region of size .
Generally, critical values for various levels of significance are
putted in the form of a table for various standard sampling
distributions of test statistic such as Z-table, 2-table, t-table, etc.
Step VI: After that, compare the calculated value of test statistic obtained
from Step IV, with the critical value(s) obtained in Step V and
locates the position of the calculated test statistic, that is, it lies in
rejection region or non-rejection region.
Step VII: In testing of hypothesis ultimately we have to reach at a conclusion.
It is done as explained below:
(i) If calculated value of test statistic lies in rejection region at 
level of significance then we reject null hypothesis. It means
that the sample data provide us sufficient evidence against the
null hypothesis and there is a significant difference between
hypothesized value and observed value of the parameter.
(ii) If calculated value of test statistic lies in non-rejection region at
 level of significance then we do not reject null hypothesis. Its
means that the sample data fails to provide us sufficient
evidence against the null hypothesis and the difference between
hypothesized value and observed value of the parameter due to
fluctuation of sample.
18
Note 3: Nowadays the decision about the null hypothesis is taken with the help Concepts of Testing of
of p-value. The concept of p-value is very important, because computer Hypothesis
packages and statistical software such as SPSS, SAS, STATA, MINITAB,
EXCEL, etc. all provide p-value. So, Section 9.8 is devoted to explain the
concept of p-value.
Now, with the help of an example we explain the above procedure.
Example 4: Suppose, it is found that average weight of a potato was 50 gm
and standard deviation was 5.1 gm nearly 5 years ago. We want to test that due
to advancement in agricultural technology, the average weight of a potato has
been increased. To test this, a random sample of 50 potatoes is taken and
calculate the sample mean (X) as 52gm. Describe the procedure to carry out
this test.
Solution: Here, we are given that
Specified value of population mean = 0 = 50 gm,
Population standard deviation = σ = 5.1 gm,
Sample size = n = 50,
Sample mean = X = 52 gm
To carry out the above test, we have to follow up the following steps:
Step I: First of all, we setup null and alternative hypotheses. Here, we want
to test that the average weight of potato is increased. So our claim is
“average weight of potato has increased” i.e. µ > 50 and its
complement is µ ≤ 50. Since complement contains equality sign so
we can take the complement as the null hypothesis and claim as the
alternative hypothesis, that is,
H0 : µ ≤ 50 gm and H1: µ > 50 gm [Here, θ = µ]
Since the alternative hypothesis is right-tailed, so our test is right-
tailed.
Step II: After setting the null and alternative hypotheses, we fix level of
significance α. Suppose, α = 0.01 (= 1 % level).
Step III: Define a test statistic to test the null hypothesis as
Statistic  Value of the parameter under H 0
Test staistic 
Standard error of statistic
X  50
T
σ/ n
Since sample size is large (n = 50 > 30) so by central limit theorem
the sampling distribution of test statistic approximately follows
standard normal distribution (as explained in Unit 1 of this course),
i.e. T ~ N(0,1)
Step IV: Calculate the value of test statistic on the basis of sample
observations as
52  50 2
T   2.78
5.1/ 50 0.72
Step V: Now, we find the critical value. The critical value or cut-off value for
standard normal distribution is given in Table I (Z-table) in the
Appendix at the end of Block 1 of this course. So from this table, the
critical value for right-tailed test at  = 0.01 is zα = 2.33.
19
Testing of Hypothesis Step IV: Now, to take the decision about the null hypothesis, we compare the
calculated value of test statistic with the critical value.
Since calculated value of test statistic (= 2.78) is greater than critical
value (= 2.33), that means calculated value of test statistic lies in
rejection region at 1% level of significance as shown in Fig. 9.5. So
we reject null hypothesis and support the alternative hypothesis. Since
alternative hypothesis is our claim, so we support the claim.
Thus, we conclude that sample does not provide us sufficient
evidence against the claim so we may assume that the average weight
of potato has increased.
Fig. 9.5 Now, you can try the following exercise.
E9) What is the first step in testing of hypothesis?

9.8 CONCEPT OF p-VALUE

In Note 3 of Section 9.6, we promised that p-value will be discussed in Section
9.8. So, it is the time to keep our promise. Nowadays use of p-value is
becoming more and more popular because of the following two reasons:
 most of the statistical software provides p-value rather than critical value.
 p-value provides more information compared to critical value as far as
rejection or do not rejection of H 0 .
The first point listed above needs no explanation. But second point lies in the
heart of p-value and needs to explain more clearly. Moving in this direction, we
note that in scientific applications one is not only interested simply in rejecting
or not rejecting the null hypothesis but he/she is also interested to assess how
strong the data has the evidence to reject H0. For example, as we have seen in
Example 4 based on general procedure of testing a hypothesis where we tested
the null hypothesis
H0 :  ≤ 50 gm against H1:  > 50 gm
To test the null hypothesis, we calculated the value of test statistic as 2.78 and
the critical value (zα) at  = 0.01 was zα = 2.33.
Since calculated value of test statistic (= 2.78) is greater than critical
(tabulated) value (= 2.33), therefore, we reject the null hypothesis at 1% level
of significance.
Now, if we reject the null hypothesis at this level (1%) surely we have to reject
it at higher level because at α = 0.05, zα = 1.645 and at α = 0.10, zα = 1.28.
However, the calculated value of test statistic is much higher than 1.645 and
1.28, therefore, the question arises “could the null hypothesis also be rejected at
values of α smaller than 0.01?” The answer is “yes” and we can compute the
smallest level of significance (α) at which a null hypothesis can be rejected.
This smallest level of significance (α) is known as “p-value”.
The p-value is the smallest value of level of significance(α) at which a null
hypothesis can be rejected using the obtained value of the test statistic and can
be defined as:
The p-value is the probability of obtaining a test statistic equal to or more
extreme (in the direction of sporting H1) than the actual value obtained when
null hypothesis is true.
The p-value also depends on the type of the test. If test is one-tailed then the p- Concepts of Testing of
value is defined as: Hypothesis

For right-tailed test:

p-value = P[Test Statistic (T) ≥ observed value of the test statistic]
For left-tailed test:
p-value = P[Test Statistic (T) ≤ observed value of the test statistic]
If test is two-tailed then the p-value is defined as:
For two-tailed test:
p-value = 2P T  observed value of the test statistic 

Procedure of taking the decision about the null hypothesis on the basis of
p-value:
To take the decision about the null hypothesis based on p-value, the p-value is
compared with level of significance (α) and if p-value is equal or less than 
then we reject the null hypothesis and if the p-value is greater than  we do not
reject the null hypothesis.
The p-value for various tests can be obtained with the help of the tables given
in the Appendix of the Block 1 of this course. But unless we are dealing with
the standard normal distribution, the exact p-value is not obtained with the
tables as mentioned above. But if we test our hypothesis with the help of
computer packages or softwares such as SPSS, SAS, MINITAB, STATA,
EXCEL, etc. then these types of computer packages or softwares present the p-
value as part of the output for each hypothesis testing procedure. Therefore, in
this block we will also describe the procedure to take the decision about the
null hypothesis on the basis of critical value as well as p-value concepts.

9.9 RELATION BETWEEN CONFIDENCE

INTERVAL AND TESTING OF HYPOTHESIS
In Units 7 and 8 of this course, we have learned about confidence intervals
which were used to estimate the unknown population parameters with certain
confidence. In Section 9.7, we have discussed the general procedure of testing
a hypothesis which has been used in making decision about the specified/
assumed/ hypothetical values of population parameters. Both confidence
interval and hypothesis testing have been used for different purposes but have
been based on the same set of concepts. Therefore, there is an extremely close
relationship between confidence interval and hypothesis testing.
In confidence interval, if we construct (1−α) 100% confidence interval for an
unknown parameter then this interval contained all probable values for the
parameter being estimated and relatively improbable values are not contained
by this interval.
So this concept can also be used in hypothesis testing. For this, we contract an
appropriate (1−α) 100% confidence interval for the parameter specified by the
null hypothesis as we have discussed in Units 7 and 8 of this course and if the
value of the parameter specified by the null hypothesis lies in this confident
interval then we do not reject the null hypothesis and if this specified value
does not lie in this confidence interval then we reject the null hypothesis.
Therefore, three cases may arise:
21
Testing of Hypothesis Case I: If we want to test the null hypothesis H0: θ = θ0 against the alternative
hypothesis H1 : θ ≠ θ0 at 5% or 1% level of significance then we
construct two-sided  1  α 100%  95% or 99% confidence interval
for the parameter θ. And we have 95% or 99% (as may be the case)
confidence that this interval will include the parameter value θ0. If the
value of the parameter specified by the null hypothesis i.e. θ0 lies in
this confidence interval then we do not reject the null hypothesis
otherwise we reject the null hypothesis.
Case II: If we want to test the null hypothesis H0 : θ ≤ θ0 against the alternative
hypothesis H1: θ > θ0 then we construct the lower one-sided
confidence bound for parameter θ. If the value of the parameter
specified by the null hypothesis i.e. θ0 is greater than or equal to this
lower bound then we do not reject the null hypothesis otherwise we
reject the null hypothesis.
Case III: If we want to test the null hypothesis H0: θ ≥ θ0 against the
alternative hypothesis H1: θ < θ0 then we construct the upper one-
sided confidence bound for parameter θ. If the value of the parameter
specified by the null hypothesis i.e. θ0 is less than or equal to this
upper bound then we do not reject the null hypothesis otherwise we
reject the null hypothesis.
For example, referring back to Example 4 of this unit, here we want to test the
null hypothesis
H 0 : µ  50gm against H1 : µ  50gm
This was tested with the help of test statistic
X µ
T  Here,θ  μ 
σ/ n
and we reject the null hypothesis at  = 0.01.
This problem could also have been solved by obtaining confidence interval
estimate of population mean which is described in Section 7.4 of Unit 7.
Here, we are given that
n  50, X  50 and σ  5.1
Since alternative hypothesis is right-tailed, therefore, we construct lower one-
sided confidence bound for population mean.
Since population variance is known so lower one-sided (1−α) 100 %
confidence bound for population mean when population variance is known is
given by

X  z
n
Since we test our null hypothesis at  = 0.01 therefore, we contract 99% lower
confidence bound and for  = 0.01, we have z   z0.01  2.33.
Thus, lower one-sided 99% confidence bound for average weight of potatoes is
5.1
52  2.33  52  1.68  50.32
50
Since the value of the parameter specified by the null hypothesis i.e. µ = 50 is
less than lower bound for average weight of potato so we reject the null
22
hypothesis. Concepts of Testing of
Hypothesis
Thus, we can use three approaches (critical value, p-value and confidence
interval) for taking decision about null hypothesis.
With this we end this unit. Let us summarise what we have discussed in this
unit.

9.10 SUMMARY
In this unit, we have covered the following points:
1. Statistical hypothesis, null hypothesis, alternative hypothesis, simple &
composite hypotheses.
2. Type-I and Type-II errors.
3. Critical region.
4. One-tailed and two-tailed tests.
5. General procedure of testing a hypothesis.
6. Level of significance.
7. Concept of p-value.
8. Relation between confidence interval and testing of hypothesis.

9.11 SOLUTIONS / ANSWERS

E1) Here, we wish to test the claim of the company that the average life of
its car tyres is 50000 miles so
Claim: µ = 50000 miles and complement: µ ≠ 50000 miles
Since claim contains equality sign so we take claim as the null
hypothesis and complement as the alternative hypothesis i.e.
H0 :  = 50000 miles
H1 :   50000 miles
E2) (iii) Here, doctor wants to test whether new medicine is really more
effective for controlling blood pressure than old medicine so
Claim: µ1 > µ2 and complement: µ1 ≤ µ2
Since complement contains equality sign so we take complement as the
null hypothesis and claim as the alternative hypothesis i.e.
H0 : µ1 ≤ µ2
H1 : µ1 > µ2
(iv) Here, economist wants to test whether the variability in incomes
differ in two populations so
Claim: 12  22 and complement: 12  22
Since complement contains equality sign so we take complement as the
null hypothesis and claim as the alternative hypothesis i.e.
H0 : 12  22

H1 : 12  22

23
Testing of Hypothesis (v) Here, psychologist wants to test whether the proportion of literates
between two groups of people is same so
Claim: P1 = P2 and complement: P1 ≠ P2
Since claim contains equality sign so we take claim as the null
hypothesis and complement as the alternative hypothesis i.e.
H0 : P1 = P2
H1 : P1 ≠ P2
E3) Here, (i) and (vi) represent the simple hypotheses because these
hypotheses tell us the exact values of parameter average weight of
orange  as  = 100 and  = 130.
The rest (ii), (iii), (iv), (v) and (vii) represent the composite hypotheses
because these hypotheses do not tell us the exact values of parameter .
E4) Since alternative hypothesis H1 : θ ≠ 60 is two tailed so critical region
lies in two-tails.
E5) Let A and B denote the number of white balls and black balls in the urn
respectively. Further, let X be the number of white balls drawn among
the two balls from the urn then we can take the null and alternative
hypotheses as
H0 : A = 4 & B = 2 and H1: A = 2 & B = 4
The critical region is given by
w  X : X  2  X : X  0,1
Thus,
 = P [Reject H0 when H0 is true]
 P  X  w / H0   PX  0 / H0   PX  1/ H0 
4
C0 2 C0 4
C1 2 C1 1  1 4  2 1 8
 6
 6
   
C2 C2 15 15 15 15
9 3
α 
15 5
Similarly,
 = P [Do not reject H0 when H1 is true]
2
C 2 4 C0 1  1 1
 P  X  w / H1   PX  2 / H1   6
 
C2 15 15
E6) Since level of significance is the probability of type-I error so in this
case level of significance is 0.05 or 5%.
E7) Here, the alternative hypothesis is two-tailed therefore, the test will be
two-tailed test.
E8) Whether the test of testing a hypothesis is one-tailed or two-tailed
depends on the alternative hypothesis. So correct option is (ii).
E9) First step in testing of hypothesis is to setup null and alternative
hypotheses.

24
UNIT 10 LARGE SAMPLE TESTS
Structure
10.1 Introduction
Objectives
10.2 Procedure of Testing of Hypothesis for Large Samples
10.3 Testing of Hypothesis for Population Mean Using Z-Test
10.4 Testing of Hypothesis for Difference of Two Population Means Using
Z-Test
10.5 Testing of Hypothesis for Population Proportion Using Z-Test
10.6 Testing of Hypothesis for Difference of Two Population Proportions
Using Z-Test
10.7 Testing of Hypothesis for Population Variance Using Z-Test
10.8 Testing of Hypothesis for Two Population Variances Using Z-Test
10.9 Summary
10.10 Solutions /Answers

10.1 INTRODUCTION
In previous unit, we have defined basic terms used in testing of hypothesis.
After providing you necessary material required for any test, we can move
towards discussing particular tests one by one. But before doing that let us tell
you the strategy we are adopting here.
First we categories the tests under two heads:
 Large sample tests
 Small sample tests
After that, their unit wise distribution is done. In this unit, we will discuss large
sample tests whereas in Units 11 and 12 we will discuss small sample tests.
The tests which are described in these units are known as “parametric tests”.
Sometimes in our studies in the fields of economics, psychology, medical, etc.
we take a sample of objects / units / participants / patients, etc. such as 70, 500,
1000, 10,000, etc. This situation comes under the category of large samples.
As a thumb rule, a sample of size n is treated as a large sample only if it
contains more than 30 units (or observations, n > 30). And we know that, for
large sample (n > 30), one statistical fact is that almost all sampling
distributions of the statistic(s) are closely approximated by the normal
distribution. Therefore, the test statistic, which is a function of sample
observations based on n > 30, could be assumed follow the normal distribution
approximately (or exactly).
But story does not end here. There are some other issues which need to be
taken care off. Some of these issues have been highlighted by making different
cases in each test as you will see when go through Sections 10.3 to 10.8 of this
unit.
This unit is divided into ten sections. Section 10.1 is introductory in nature.
General procedure of testing of hypothesis for large samples is described in
25
Testing of Hypothesis Section 10.2. In Section 10.3, testing of hypothesis for population mean is
discussed whereas in Section 10.4, testing of hypothesis for difference of two
population means with examples is described. Similarly, in Sections 10.5 and
10.6, testing of hypothesis for population proportion and difference of two
population proportions are explained respectively. Testing of hypothesis for
population variance and two population variances are described in Sections
10.7 and 10.8 respectively. Unit ends by providing summary of what we have
discussed in this unit in Section 10.9 and solution of exercises in Section 10.10.

Objectives
After studying this unit, you should be able to:
 judge for a given situation whether we should go for large sample test or
not;
 Applying the Z-test for testing the hypothesis about the population mean
and difference of two population means;
 Applying the Z-test for testing the hypothesis about the population
proportion and difference of two population proportions; and
 Applying the Z-test for testing the hypothesis about the population variance
and two population variances.

10.2 PROCEDURE OF TESTING OF HYPOTHESIS

FOR LARGE SAMPLES
As we have described in previous section that for large sample size (n > 30),
one statistical fact is that almost all sampling distributions of the statistic(s) are
closely approximated by the normal distribution. Therefore, when sample size
is large one can apply the normal distribution based test procedures to test the
hypothesis.
In previous unit, we have given the procedure of testing of hypothesis in
general. Let us now discuss the procedure of testing a hypothesis for large
samples in particular.
Suppose X1, X2, …, Xn is a random sample of size n (> 30) selected from a
population having unknown parameter  and we want to test the hypothesis
about the hypothetical / claimed / assumed value θ0 of parameter θ. For this, a
test procedure is required. We discuss it step by step as follows:
Step I: First of all, we have to setup null hypothesis H0 and alternative
hypothesis H1. Here, we want to test the hypothetical / claimed /
assumed value θ0 of parameter θ. So we can take the null and
alternative hypotheses as
H 0 :   0 and H1 :   0  for two-tailed test 
H 0 :   0 and H 1 :   0 
or  for one-tailed test 
H 0 :   0 and H 1 :   0 
In case of comparing same parameter of two populations of interest,
say, 1 and 2, then our null and alternative hypotheses would be
H 0 : 1  2 and H1 : 1  2 for two-tailed test 
26
 H 0 : 1  2 and H 1 : 1  2  Large Sample Tests
or   for one-tailed test 
H 0 : 1  2 and H 1 : 1  2 

Step II: After setting the null and alternative hypotheses, we have to choose
level of significance. Generally, it is taken as 5% or 1% (α = 0.05 or
0.01). And accordingly rejection and non-rejection regions will be
decided.
Step III: Third step is to determine an appropriate test statistic, say, Z in case
of large samples. Suppose Tn is the sample statistic such as sample
mean, sample proportion, sample variance, etc. for the parameter 
then for testing the null hypothesis, test statistic is given by
 we know that SE of a statistic is 
Tn  E(Tn ) Tn  E(Tn )  theSD of the sampling distribution 
Z 
SE(Tn ) Var(Tn )  of that statistic 
SE(Tn )  SD(Tn )  Var(Tn ) 

where, E(Tn) is the expectation (or mean) of Tn and Var(Tn) is

variance of Tn.
Step IV: As already mentioned for large samples, statistical fact is that almost
all sampling distributions of the statistic(s) are closely approximated
by the normal distribution as the parent population is normal or non-
normal. So, the test statistic Z will assumed to be approximately
normally distributed with mean 0 and variance 1 as
Tn  E(Tn )
Z ~ N(0,1)
Var(Tn )

By putting the values of Tn, E(Tn) and Var(Tn) in above formula we

calculate the value of test statistic Z. Let z be the calculated value of
test statistic Z.
Step V: After that, we obtain the critical (cut-off or tabulated) value(s) in the
sampling distribution of the test statistic Z corresponding to 
assumed in Step II. These critical values are given in Table-I
(Z-table) at the Appendix of Block 1 of this course corresponding to
different level of significance (α). For convenient some useful critical
values at α = 0.01, 0.05 for Z-test are given in Table 10.1 in this
section. After that, we construct rejection (critical) region of size α in
the probability curve of the sampling distribution of test statistic Z.
Step VI: Take the decision about the null hypothesis based on the calculated
and critical values of test statistic obtained in Step IV and Step V.
Since critical value depends upon the nature of the test that it is one-
tailed test or two-tailed test so following cases arise:
In case of one-tailed test:
Case I: When H 0 :   0 and H 1 :   0 (right-tailed test)

In this case, the rejection (critical) region falls under the right tail of
the probability curve of the sampling distribution of test statistic Z.
Fig. 10.1
Suppose z is the critical value at  level of significance so entire
region greater than or equal to z is the rejection region and less than
z is the non-rejection region as shown in Fig. 10.1.
Testing of Hypothesis If z (calculated value ) ≥ z (tabulated value), that means the
calculated value of test statistic Z lies in the rejection region, then we
reject the null hypothesis H0 at  level of significance. Therefore, we
conclude that sample data provides us sufficient evidence against the
null hypothesis and there is a significant difference between
hypothesized or specified value and observed value of the parameter.
If z < z , that means the calculated value of test statistic Z lies in non-
rejection region, then we do not reject the null hypothesis H0 at 
level of significance. Therefore, we conclude that the sample data
fails to provide us sufficient evidence against the null hypothesis and
the difference between hypothesized value and observed value of the
parameter due to fluctuation of sample.
so the population parameter θ may be 0.
Case II: When H 0 :    0 and H 1 :   0 (left-tailed test)

In this case, the rejection (critical) region falls under the left tail of the
probability curve of the sampling distribution of test statistic Z.
Suppose -z is the critical value at  level of significance then entire
region less than or equal to -z is the rejection region and greater
than -z is the non-rejection region as shown in Fig. 10.2.

Fig. 10.2 If z ≤-z, that means the calculated value of test statistic Z lies in the
rejection region, then we reject the null hypothesis H0 at  level of
significance.
If z >-z, that means the calculated value of test statistic Z lies in the
non-rejection region, then we do not reject the null hypothesis H0 at 
level of significance.
In case of two-tailed test: When H 0 :   0 and H 1 :   0

In this case, the rejection region falls under both tails of the
probability curve of sampling distribution of the test statistic Z. Half
the area (α) i.e. α/2 will lies under left tail and other half under the
right tail. Suppose  zα / 2 and zα / 2 are the two critical values at the
left-tailed and right-tailed respectively. Therefore, entire region less
than or equal to z  / 2 and greater than or equal to zα / 2 are the
rejection regions and between  zα / 2 and zα / 2 is the non-rejection
Fig. 10.3
region as shown in Fig. 10.3.

If z  z / 2 or z   z / 2 , that means the calculated value of test

statistic Z lies in the rejection region, then we reject the null
hypothesis H0 at  level of significance.

If zα/2  z  zα/ 2 , that means the calculated value of test statistic Z

lies in the non-rejection region, then we do not reject the null
hypothesis H0 at  level of significance.
Table 10.1 shows some commonly used critical (cut-off or tabulated) values
for one-tailed test and two-tailed test at different level of significance (α) for
Z-test.
 Table 10.1: Critical Values for Z-test Large Sample Tests

Level of Two-Tailed Test One-Tailed Test

Significance (α)
Right-Tailed Test Left- Tailed Test

α = 0.05 (= 5%) ± zα/2 = ± 1.96 zα = 1.645 − zα = − 1.645

α = 0.01 (= 1%) ± zα/2 = ± 2.58 zα = 2.33 − zα = − 2.33

Note 1: As we have discussed in Step IV of this procedure that when sample

size is large then test statistic follows the normal distribution as the parent
population is normal or non-normal so we do not require any assumption of the
form of the parent population for large sample size but when sample size is
small (n < 30) then for applying parametric test we must require the assumption
that the population is normal as we shall in Units 11 and 12. If this assumption
is not fulfilled then we apply the non-parametric tests which will be discussed
in Block 4 of this course.
Decision making procedure about the null hypothesis using the concept of
p-value:
To take the decision about the null hypothesis on the basis of p-value, the p-
value is compared with given level of significance (α) and if p-value is less
than or equal to  then we reject the null hypothesis and if the p-value is
greater than  we do not reject the null hypothesis.
Since test statistic Z follows approximately normal distribution with mean 0
and variance unity, i.e. standard normal distribution and we also know that
standard normal distribution is symmetrical about Z = 0 line therefore, if z
represents the calculated value of Z then p-value can be calculated as follows:
For one-tailed test:
For H1: θ > θ0 (right-tailed test)
p-value = P[Z  z]
For H1: θ < θ0 (left-tailed test)
p-value = P[Z  z]
For two-tailed test:
For H1:   0
p-value = 2P  Z  z 

These p-values for Z-test can be obtained with the help of Table-I (Z-table)
given in the Appendix at the end of Block 1 of this course (which gives the
probability [0  Z  z] for different value of z) as discussed in Unit 14 of
MST-003.
For example, if test is right-tailed and calculated value of test statistic Z is 1.23
then
p-value = P  Z  z   P  Z  1.23  0.5  P  0  Z  1.23

 0.5  0.3907  From Z-table given in Appendix 

of Block 1of this course. 
= 0.1093
29
Testing of Hypothesis Now, you can try the following exercises.
E1) If an investigator observed that the calculated value of test statistic lies in
non-rejection region then he/she will
(i) reject the null hypothesis
(ii) accept the null hypothesis
(iii) not reject the null hypothesis
Write the correct option.
E2) If we have null and alternative hypotheses as
H0:  = 0 and H1:   0
then the rejection (critical) region lies in
(i) left tail
(ii) right tail
(iii) both tails
Write the correct option.
E3) If test is two-tailed and calculated value of test statistic Z is 2.42 then
calculate the p-value for the Z-test.

10.3 TESTING OF HYPOTHESIS FOR

POPULATION MEAN USING Z-TEST
In previous section, we have discussed the general procedure for Z-test. Now
we are discussing the Z-test for testing the hypothesis or claim about the
population mean when sample size is large. Let population under study has
mean  and variance σ2, where  is unknown and σ2 may be known or
unknown. We will consider both cases under this heading. For testing a
hypothesis about population mean we draw a random sample X1, X2,…,Xn of
size n  30 from this population. As we know that for drawing the inference
about the population mean we generally use sample mean and for test statistic,
we require the mean and standard error of sampling distribution of the statistic
(mean). Here, we are considering large sample so we know by central limit
theorem that sample mean is asymptotically normally distributed with mean 
and variance σ2/n whether parent population is normal or non-normal. That
is, if X is the sample mean of the random sample then
σ2
E  X   µ , Var  X   … (1)
n
But we know that standard error = Variance

σ
 SE  X   Var  X   … (2)
n
Now, follow the same procedure as we have discussed in previous section, that
is, first of all we have to setup null and alternative hypotheses. Since here we
want to test the hypothesis about the population mean so we can take the null
and alternative hypotheses as

30
 Here,    and 0  0  Large Sample Tests
H0 :   0 and H1 :   0 for two-tailed test if we compareit with 
general procedure. 
H0 :   0 and H1 :   0 
or  for one-tailed test 
H0 :   0 and H1 :   0 
When we assume that
For testing the null hypothesis, the test statistic Z is given by the null hypothesis is
true then we are actually
X  E X  assuming that the
Z
SE  X  population parameter is
equal to the value in the
X  µ0 null hypothesis. For
Z  Using equations (1) and (2) and  example, we assume that
σ/ n  under H 0 we assume that µ  µ 0 .  µ = 60 whether the null
hypothesis is µ = 60 or
The sampling distribution of the test statistic depends upon σ2 that it is known µ ≤ 60 or µ ≥ 60.
or unknown. Therefore, two cases arise:
Case I: When σ2 is known
In this case, the test statistic follows the normal distribution with
mean 0 and variance unity when the sample size is the large as the
population under study is normal or non-normal. If the sample size is
small then test statistic Z follows the normal distribution only when
population under study is normal. Thus,
X  µ0
Z ~ N  0, 1 
σ/ n
Case II: When σ2 is unknown
In this case, we estimate σ2 by the value of sample variance (S2)
where,
1 n 2
S2  
n  1 i1
 Xi  X 

X  0
Then become test statistic follows the t-distribution with
S/ n
(n−1) df as the sample size is large or small provided the population
under study follows normal as we have discussed in Unit 2 of this
course. But when population under study is not normal and sample
size is large then this test statistic approximately follows normal
distribution with mean 0 and variance unity, that is,
X  0
Z ~ N  0,1
S/ n
After that, we calculate the value of test statistic as may be the case (σ2 is
known or unknown) and compare it with the critical value given in Table 10.1
at prefixed level of significance α. Take the decision about the null hypothesis
as described in the previous section.
From above discussion of testing of hypothesis about population mean, we note
following point:
(i) When σ2 is known then we apply the Z-test as the population under study
is normal or non-normal for the large sample. But when sample size is

31
Testing of Hypothesis small then we apply the Z-test only when population under study is
normal.
(ii) When σ2 is unknown then we apply the t-test only when the population
under study is normal as sample size is large or small. But when the
assumption of normality is not fulfilled and sample size is large then we
can apply the Z-test.
(iii) When sample is small and σ2 is known or unknown and the form of the
population is not known then we apply the non-parametric test as we will
be discussed in Block 4 of this course.
Following examples will help you to understand the procedure more clearly.
Example 1: A light bulb company claims that the 100-watt light bulb it sells
has an average life of 1200 hours with a standard deviation of 100 hours. For
testing the claim 50 new bulbs were selected randomly and allowed to burn
out. The average lifetime of these bulbs was found to be 1180 hours. Is the
company’s claim is true at 5% level of significance?
Solution: Here, we are given that
Specified value of population mean = 0 = 1200 hours,
Population standard deviation = σ = 100 hours,
Sample size = n = 50
Sample mean = X = 1180 hours.
In this example, the population parameter being tested is population mean i.e.
average life of a bulb (µ) and we want to test the company’s claim that average
life of a bulb is 1200 hours. So our claim is  = 1200 and its complement is
 ≠ 1200. Since claim contains the equality sign so we can take the claim as the
null hypothesis and complement as the alternative hypothesis. So
H 0 :    0  1200 average life of a bulb is 1200 hours 
H 1 :   1200 average life of a bulb is not1200 hours 
Also the alternative hypothesis is two-tailed so the test is two-tailed test.
Here, we want to test the hypothesis regarding mean when population SD
(variance) is known and sample size n = 50(> 30) is large. So we will go for
Z-test.
Thus, for testing the null hypothesis the test statistic is given by
X  0
Z
/ n
1180  1200 20
   1.41
100 / 50 14.14
The critical (tabulated) values for two-tailed test at 5% level of significance are
± zα/2 = ± z0.025 = ± 1.96.
Fig. 10.4 Since calculated value of test statistic Z ( = –1.41) is greater than critical value
(= − 1.96) and less than the critical value (= 1.96), that means it lies in non-
rejection region as shown in Fig. 10.4, so we do not reject the null hypothesis.
Since the null hypothesis is the claim so we support the claim at 5% level of
significance.
Decision according to p-value:
The test is two-tailed, therefore,
p-value = 2P Z  z   2P  Z  1.41 Large Sample Tests

 2  0.5  P  0  Z  1.41  2  0.5  0.4207   0.1586

Since p-value (= 0.1586) is greater than α (= 0.05) so we do not reject the null
hypothesis at 5% level of significance.
Decision according to confidence interval:
Here, test is two-tailed, therefore, we contract two-sided confidence interval for
population mean.
Since population standard deviation is known, therefore, we can use
(1−α) 100 % confidence interval for population mean when population
variance is known which is given by
   
X  z  / 2 n , X  z / 2 n 
 

Here,  = 0.05, we have z /2  z0.025  1.96.

Thus, 95% confidence interval for average life of a bulb is given by

 100 100 
1180  1.96 50 ,1180  1.96 50 
 
or 1180  27.71,1180  27.71 
or 1152.29, 1207.71
Since 95% confidence interval for average life of a bulb contains the value of
the parameter specified by the null hypothesis, that is,    0  1200 so we do
not reject the null hypothesis.
Thus, we conclude that sample does not provide us sufficient evidence against
the claim so we may assume that the company’s claim that the average life of a
bulb is 1200 hours is true.
Note 2: Here, we note that the decisions about null hypothesis based on three
approaches (critical value or classical, p-value and confidence interval) are
same. The learners are advised to make the decision about the claim or
statement by using only one of the three approaches in the examination. Here,
we used all these approaches only to give you an idea how they can be used in
a given problem. Those learners who will opt biostatistics specialisation will
see and realize the importance of confidence interval approach in Unit 16 of
MSTE-004.
Example 2: A manufacturer of ball point pens claims that a certain pen
manufactured by him has a mean writing-life at least 460 A-4 size pages. A
purchasing agent selects a sample of 100 pens and put them on the test. The
mean writing-life of the sample found 453 A-4 size pages with standard
deviation 25 A-4 size pages. Should the purchasing agent reject the
manufacturer’s claim at 1% level of significance?
Solution: Here, we are given that
Specified value of population mean = 0 = 460,

33
Testing of Hypothesis Sample size = n = 100,
Sample mean = X = 453,
Sample standard deviation = S = 25
Here, we want to test the manufacturer’s claim that the mean writing-life (µ) of
pen is at least 460 A-4 size pages. So our claim is  ≥ 460 and its complement
is  < 460. Since claim contains the equality sign so we can take the claim as
the null hypothesis and the complement as the alternative hypothesis. So
H 0 :    0  460 and H 1 :   460

Also the alternative hypothesis is left-tailed so the test is left-tailed test.

Here, we want to test the hypothesis regarding population mean when
population SD is unknown. So we should used t-test for if writing-life of pen
follows normal distribution. But it is not the case. Since sample size is n = 100
(n > 30) large so we go for Z-test. The test statistic of Z-test is given by
X  0
Z
S/ n
453  460 7
   2.8
25 / 100 2.5
We have from Table 10.1 the critical (tabulated) value for left-tailed Z test at
1% level of significance is zα = −2.33.
Since calculated value of test statistic Z (= ‒2.8) is less than the critical value
(= −2.33), that means calculated value of test statistic Z lies in rejection region
Fig. 10.5 as shown in Fig. 10.5, so we reject the null hypothesis. Since the null
hypothesis is the claim so we reject the manufacturer’s claim at 1% level of
significance.
Decision according to p-value:
The test is left-tailed, therefore,

p-value = P  Z  z  P Z  2.8  P Z  2.8  Z is symetrical 

about Z  0 line 

 0.5  P 0  Z  2.8  0.5  0.4974  0.0026

Since p-value (= 0.0026) is less than α (= 0.01) so we reject the null hypothesis
at 1% level of significance.
Therefore, we conclude that the sample provide us sufficient evidence against
the claim so the purchasing agent rejects the manufacturer’s claim at 1% level
of significance.
Now, you can try the following exercises.
E4) A sample of 900 bolts has a mean length 3.4 cm. Is the sample regarded
to be taken from a large population of bolts with mean length 3.25 cm
and standard deviation 2.61 cm at 5% level of significance?
E5) A big company uses thousands of CFL lights every year. The brand that
the company has been using in the past has average life of 1200 hours. A
new brand is offered to the company at a price lower than they are paying
 for the old brand. Consequently, a sample of 100 CFL light of new brand Large Sample Tests
is tested which yields an average life of 1220 hours with standard
deviation 90 hours. Should the company accept the new brand at 5% level
of significance?

10.4 TESTING OF HYPOTHESIS FOR

DIFFERENCE OF TWO POPULATION MEANS
USING Z-TEST
In previous section, we have learnt about the testing of hypothesis about the
population mean. But there are so many situations where we want to test the
hypothesis about difference of two population means or two population means.
For example, two manufacturing companies of bulbs are produced same type
of bulbs and one may be interested to test that one is better than the other, an
investigator may want to test the equality of the average incomes of the peoples
living in two cities, etc. Therefore, we require an appropriate test for testing the
hypothesis about the difference of two population means.
Let there be two populations, say, population-I and population-II under study.
Also let 1 ,  2 and 12 ,  22 denote the means and variances of population-I and
population-II respectively where both 1 and 2 are unknown but 12 and 22 may
be known or unknown. We will consider all possible cases here. For testing the
hypothesis about the difference of two population means, we draw a random
sample of large size n1 from population-I and a random sample of large size n2
from population-II. Let X and Y be the means of the samples selected from
population-I and II respectively.
These two populations may or may not be normal but according to the central
limit theorem, the sampling distribution of difference of two large sample
means asymptotically normally distributed with mean (µ1-µ2) and variance
(σ12 / n1  σ 22 / n 2 ) as described in Unit 2 of this course.
Thus,
E  X  Y   E  X   E  Y   1   2 … (3)

and
12 22
Var  X  Y   Var  X   Var  Y   
n1 n 2
But we know that standard error = Variance

12 22
 SE  X  Y   Var  X  Y    … (4)
n1 n 2
Now, follow the same procedure as we have discussed in Section 10.2, that is,
first of all we have to setup null and alternative hypotheses. Here, we want to
test the hypothesis about the difference of two population means so we can take
the null hypothesis as
 Here, 1  1 and 2   2 
H0 : 1  2 (no difference in means) if we compareit with 
 general procedure. 
35
Testing of Hypothesis or H 0 : 1  2  0 (difference in two means is 0)
and the alternative hypothesis as
H1 : 1  2 for two-tailed test 

H0 : 1  2 and H1 : 1  2 
or  for one-tailed test 
H0 : 1  2 and H1 : 1  2 
For testing the null hypothesis, the test statistic Z is given by

Z
 X  Y   E X  Y 
SE  X  Y 

 X  Y    µ1  µ 2 
or Z  using equations (3) and (4) 
σ12 σ 22

n1 n 2
Since under null hypothesis we assume that µ1 = µ2, therefore, we have
XY
Z
σ12 σ 22

n1 n 2

Now, the sampling distribution of the test statistic depends upon 12 and 22
that both are known or unknown. Therefore, four cases arise:
Case I: When 12 & 22 are known and 12 22  2
In this case, the test statistic follows normal distribution with mean
0 and variance unity when the sample sizes are large as both the
populations under study are normal or non-normal. But when
sample sizes are small then test statistic Z follows normal
distribution only when populations under study are normal, that is,
XY
Z ~ N  0,1 
1 1
σ 
n1 n2
Case II: When 12 & 22 are known and 12  22
In this case, the test statistic also follows the normal distribution as
described in case I, that is,
XY
Z ~ N(0, 1)
σ12 σ 22

n1 n 2

Case III: When 12 & 22 are unknown and 12 22  2

In this case, σ2 is estimated by value of pooled sample variance S2p

where,
1
S2p   n1S12  n 2S22 
n1  n 2  2 
and
36
 n n Large Sample Tests
1 1
2 1 2
2
2
S 
1 
(n1  1) i1
 Xi  X  and S22  
(n 2  1) i1
 Yi  Y 

and test statistic follows t-distribution with (n1 + n2 − 2) degrees of

freedom as the sample sizes are large or small provided populations
under study follow normal distribution as described in Unit 2 of this
course. But when the populations are under study are not normal
and sample sizes n1 and n2are large (> 30) then by central limit
theorem, test statistic approximately normally distributed with mean
0 and variance unity, that is,
XY
Z ~ N  0,1 
1 1
Sp 
n1 n 2
Case IV: When 12 & 22 are unknown and 12  22

In this case, 12 & 22 are estimated by the values of the sample
variances S12 &S22 respectively and the exact distribution of test
statistic is difficult to derive. But when sample sizes n1 and n2 are
large (> 30) then central limit theorem, the test statistic
approximately normally distributed with mean 0 and variance unity,
that is,
XY
Z ~ N(0, 1)
S12 S22

n1 n 2
After that, we calculate the value of test statistic and compare it with the
critical value given in Table 10.1 at prefixed level of significance α. Take the
decision about the null hypothesis as described in Section10.2.
From above discussion of testing of hypothesis about population mean, we note
following point:
(i) When 12 & 22 are known then we apply the Z-test as both the population
under study are normal or non-normal for the large sample. But when
sample sizes are small then we apply the Z-test only when populations
under study are normal.
(ii) When 12 & 22 are unknown then we apply the t-test only when the
populations under study are normal as sample sizes are large or small.
But when the assumption of normality is not fulfilled and sample sizes
are large then we can apply the Z-test.
(iii) When samples are small and 12 & 22 are known or unknown and the
form of the population is not known then we apply the non-parametric
test as we will be discussed in Block 4 of this course.
Let us do some examples based on above test.
Example 3: In two samples of women from Punjab and Tamilnadu, the mean
height of 1000 and 2000 women are 67.6 and 68.0 inches respectively. If
population standard deviation of Punjab and Tamilnadu are same and equal to
5.5 inches then, can the mean heights of Punjab and Tamilnadu women be
regarded as same at 1% level of significance?
37
Testing of Hypothesis Solution: We are given
n1 = 1000, n2 = 2000, X  67.6, Y  68.0 and σ1  σ2  σ  5.5
Here, we wish to test that the mean height of Punjab and Tamilnadu women is
same. If 1 and 2 denote the mean heights of Punjab and Tamilnadu women
respectively then our claim is 1 = 2 and its complement is 1 ≠ 2. Since the
claim contains the equality sign so we can take the claim as the null hypothesis
and complement as the alternative hypothesis. Thus,
H 0 : 1   2 and H1 : 1   2
Since the alternative hypothesis is two-tailed so the test is two-tailed test.
Here, we want to test the hypothesis regarding two population means. The
standard deviations of both populations are known and sample sizes are large,
so we should go for Z-test.
So, for testing the null hypothesis, the test statistic Z is given by
XY
Z
σ12 σ 22

n1 n 2
67.6  68.0  0.4
 2 2

  5.5   5.5   1 1 
 1000  2000  5.5   
 1000 2000 
 
 0.4
  1.88
5.5  0.0387
The critical (tabulated) values for two-tailed test at 1% level of significance are
± zα/2 = ± z0.005 = ± 2.58.
Since calculated value of Z ( = −1.88) is greater than the critical value
(= − 2.58) and less than the critical value (= 2.58), that means it lies in non-
rejection region as shown in Fig. 10.6, so we do not reject the null hypothesis
i.e. we fail to reject the claim.
Decision according to p-value:
Fig. 10.6
The test is two-tailed, therefore,
p-value = 2P  Z  z   2P  Z  1.88

 2  0.5  P  0  Z  1.88  2  0.5  0.4699   0.0602

Since p-value (= 0.0602) is greater than  (  0.01) so we do not reject the null
hypothesis at1% level of significance.
Thus, we conclude that the samples do not provide us sufficient evidence
against the claim so we may assume that the average height of women of
Punjab and Tamilnadu is same.
Example 4: A university conducts both face to face and distance mode classes
for a particular course indented both to be identical. A sample of 50 students of
face to face mode yields examination results mean and SD respectively as:
X  80.4, S1  12.8
and other sample of 100 distance-mode students yields mean and SD of their Large Sample Tests
examination results in the same course respectively as:
Y  74.3, S2  20.5
Are both educational methods statistically equal at 5% level?
Solution: Here, we are given that
n1  50, X  80.4, S1  12.8;

n 2  100 , Y  74.3, S2  20.5

We wish to test that both educational methods are statistically equal. If 1 and
2 denote the average marks of face to face and distance mode students
respectively then our claim is 1 = 2 and its complement is 1 ≠ 2. Since the
claim contains the equality sign so we can take the claim as the null hypothesis
and complement as the alternative hypothesis. Thus,
H 0 : 1   2 and H1 : 1   2
Since the alternative hypothesis is two-tailed so the test is two-tailed test.
We want to test the null hypothesis regarding two population means when
standard deviations of both populations are unknown. So we should go for t-
test if population of difference is known to be normal. But it is not the case.
Since sample sizes are large (n1, and n2 > 30) so we go for Z-test.
For testing the null hypothesis, the test statistic Z is given by
XY
Z
S12 S22

n1 n 2

80.4  74.3 6.1

   2.23
12.8 
2
 20.5 
2 3.28  4.20
 Fig. 10.7
50 100
The critical (tabulated) values for two-tailed test at 5% level of significance are
± zα/2 = ± z0.025 = ± 1.96.
Since calculated value of Z ( = 2.23) is greater than the critical values
(= ±1.96 ), that means it lies in rejection region as shown in Fig. 10.7, so we
reject the null hypothesis i.e. we reject the claim at 5% level of significance.
Decision according to p-value:
The test is two-tailed, therefore,

p-value = 2P  Z  z   2P  Z  2.23

 2  0.5  P 0  Z  2.23  2  0.5  0.4871  0.0258

Since p-value (= 0.0258) is less than  (  0.05) so we reject the null hypothesis
at 5% level of significance.
Thus, we conclude that samples provide us sufficient evidence against the
claim so both methods of education, i.e. face-to-face and distance-mode, are
not statistically equal.
Testing of Hypothesis Now, you can try the following exercises.
E6) Two brands of electric bulbs are quoted at the same price. A buyer was
tested a random sample of 200 bulbs of each brand and found the
following information:
Mean Life (hrs.) SD(hrs.)
Brand A 1300 41
Brand B 1280 46
Is there significant difference in the mean duration of their lives of two
brands of electric bulbs at 1% level of significance?
E7) Two research laboratories have identically produced drugs that provide
relief to BP patients. The first drug was tested on a group of 50 BP
patients and produced an average 8.3 hours of relief with a standard
deviation of 1.2 hours. The second drug was tested on 100 patients,
producing an average of 8.0 hours of relief with a standard deviation of
1.5 hours. Does the first drug provide a significant longer period of relief
at a significant level of 5%?

10.5 TESTING OF HYPOTHESIS FOR

POPULATION PROPORTION USING Z-TEST
In Section 10.3, we have discussed the procedure of testing of hypothesis for
population mean when sample size is large. But in many real world situations,
in business and other areas where collected data are in form of counts or the
collected data are classified into two categories or groups according to an
attribute or a characteristic. For example, the peoples living in a colony may be
classified into two groups (male and female) with respect to the characteristic
sex, the patients in a hospital may be classified into two groups as cancer and
non-cancer patients, the lot of articles may be classified as defective and non-
defective, etc. Here, collected data are available in dichotomous or binary
outcomes which is a special case of nominal scale and the data categorized into
two mutually exclusive and exhaustive classes generally known as success and
failure out comes. For example, the characteristic sex can be measured as
success if male and failure if female or vice versa. So in such situations,
proportion is suitable measure to apply.
In such situations, we require a test for testing a hypothesis about population
proportion.
For this purpose, let X1 , X 2 , ..., X n be a random sample of size n taken from a
population with population proportion P. Also let X denotes the number of
observations or elements possess a certain attribute (number of successes) out
of n observations of the sample then sample proportion p can be defined as
X
p 1
n
As we have seen in Section 2.4 of the Unit 2 of this course that mean and
variance of the sampling distribution of sample proportion are
PQ
E(p)  P and Var(p) 
n
where, Q = 1‒ P.
40
Now, two cases arise: Large Sample Tests

Case I: When sample size is not sufficiently large i.e. either of the conditions
np > 5 or nq > 5 does not meet, then we use exact binomial test. But exact
binomial test is beyond the scope of this course.
Case II: When sample size is sufficiently large, such that np > 5 and nq > 5
then by central limit theorem, the sampling distribution of sample proportion p
is approximately normally distributed with mean and variance as
PQ
E(p) = P and Var(p) = … (5)
n
But we know that standard error = Variance

PQ
 SE (p)  … (6)
n
Now, follow the same procedure as we have discussed in Section 10.2, first of
all we setup null and alternative hypotheses. Since here we want to test the
hypothesis about specified value P0 of the population proportion so we can take
the null and alternative hypotheses as
Here,   P and   P0 
H 0 : P  P0 and H1 : P  P0 for two-tailed test   if we compare it with
0

 general procedure. 

H 0 : P  P0 and H1 : P  P0 
or  for one-tailed test 
H 0 : P  P0 and H1 : P  P0 
For testing the null hypothesis, the test statistic Z is given by
p  E p 
Z
SE  p 
p  P0
Z ~ N  0, 1  under H0  using equations(5) and (6) 
P0 Q 0
n
After that, we calculate the value of test statistic and compare it with the
critical value(s) given in Table 10.1 at prefixed level of significance α. Take
the decision about the null hypothesis as described in Section 10.2.
Let us do some examples of testing of hypothesis about population proportion.
Example 5: A machine produces a large number of items out of which 25%
are found to be defective. To check this, company manager takes a random
sample of 100 items and found 35 items defective. Is there an evidence of more
deterioration of quality at 5% level of significance?
Solution: The company manager wants to check that his machine produces
25% defective items. Here, attribute under study is defectiveness. And we
define our success and failure as getting a defective or non defective item.
Let P = Population proportion of defectives items = 0.25(= P0 )
p = Observed proportion of defectives items in the sample = 35/100 = 0.35
Here, we want to test that machine produces more defective items, that is, the
proportion of defective items (P) greater than 0.25. So our claim is P > 0.25
41
Testing of Hypothesis and its complement is P ≤ 0.25. Since complement contains the equality sign so
we can take the complement as the null hypothesis and the claim as the
alternative hypothesis. So
H 0 : P  P0  0.25 and H 1 : P  0.25

Since the alternative hypothesis is right-tailed so the test is right-tailed test.

Before proceeding further, first we have to check whether the condition of
normality meets or not.
 np  100  0.35  35  5
nq  100   1  0.35   100  0.65  65  5

We see that condition of normality meets, so we can go for Z-test.

So, for testing the null hypothesis, the test statistic Z is given by
p  P0
Z
P0Q 0
n
0.35  0.25 0.10
   2.31
0.25  0.75 0.0433
100
Since test is right-tailed so the critical value at 5% level of significance is
zα = z0.05 = 1.645.
Since calculated value of test statistic Z (= 2.31) is greater than the critical
value (= 1.645), that means it lies in the rejection region as shown in Fig. 10.8.
So we reject the null hypothesis and support the alternative hypothesis i.e. we
Fig. 10.8 support the claim at 5% level of significance.
Decision according to p-value:
The test is right-tailed, therefore,

p-value = P  Z  z  P  Z  2.31

 0.5  P 0  Z  2.31  0.5  0.4896

 0.0104
Since p-value (= 0.0104) is less than  (  0.05) so we reject the null
hypothesis at 5% level of significance.
Thus, we conclude that the sample fails to provide us sufficient evidence
against the claim so we may assume that deterioration in quality exists at 5%
level of significance.
Example 6: A die is thrown 9000 times and draw of 2 or 5 is observed 3100
times. Can we regard that die is unbiased at 5% level of significance.
Solution: Let getting a 2 or 5 be our success, and getting a number other than 2
or 5 be a failure then in usual notions, we have
n = 9000, X = number of successes = 3100, p = 3100/9000 = 0.3444
Here, we want to test that the die is unbiased and we know that if die is Large Sample Tests
unbiased then proportion or probability of getting 2 or 5 is
P = Probability of getting a 2 or 5
= Probability of getting 2 + Probability of getting 5
1 1 1
    0.3333
6 6 3
So our claim is P = 0.3333 and its complement is P ≠ 0.3333. Since the claim
contains the equality sign so we can take the claim as the null hypothesis and
complement as the alternative hypothesis. Thus,
H0 : P  P0  0.3333 and H1 :P  0.3333

Since the alternative hypothesis is two-tailed so the test is two-tailed test.

Before proceeding further, first we have to check whether the condition of
normality meets or not.
 np  9000  0.3444  3099.6  5
nq  9000   1  0.3444   9000  0.6556  5900.4  5
We see that condition of normality meets, so we can go for Z-test.
So, for testing the null hypothesis, the test statistic Z is given by
p  P0
Z
P0Q 0
n
0.3444  0.3333 0.0111
   2.22
0.3333  0.6667 0.005
9000
Since test is two-tailed so the critical values at 5% level of significance are
± zα/2 = ± z0.025 = ±1.96.
Since calculated value of Z (= 2.22) is greater than the critical value (= 1.96),
that means it lies in rejection region, so we reject the null hypothesis i.e. we
reject our claim.
Decision according to p-value:
Since test is two-tailed so

p-value = 2P Z  z   2P  Z  2.22

 2 0.5  P  0  Z  2.22  2  0.5  0.4868  0.0264

Since p-value (= 0.0264) is less than  (  0.05), so we reject the null

hypothesis at 5% level of significance.
Thus, we conclude that the sample provides us sufficient evidence against the
claim so die cannot be considered as unbiased.
Now, you can try the following exercises.
E8) In a sample of 100 MSc. Economics first year students of a University, it
was seen that 54 students came from Science background and the rest are
43
Testing of Hypothesis from other background. Can we assume that 50% of the students are from
Science background in MSc. Economics first year students in the
University at 1% level of significance?
E9) Out of 200 patients who are given a particular injection 180 survived.
Test the hypothesis that the survival rate is more than 80% at 5% level of
significance?

10.6 TESTING OF HYPOTHESIS FOR

DIFFERENCE OF TWO POPULATION
PROPORTIONS USING Z-TEST
In Section 10.5, we have discussed the testing of hypothesis about the
population proportion. In some cases, we are interested to test the hypothesis
about difference of two population proportions of an attributes in the two
different populations or groups. For example, one may wish to test whether the
proportions of alcohol drinkers in the two cities are same, one may wish to test
proportion of literates in a group of people is greater than the proportion of
literates in other group of people, etc. Therefore, we require the test for testing
the hypothesis about the difference of two population proportions.
Let there be two populations, say, population-I and population-II under study.
And also let we draw a random sample of size n1 from population-I with
population proportion P1 and a random sample of size n2 from population-II
with population proportion P2. If X1 and X2 are the number of observations
/individuals / items / units possessing the given attribute in the sample of sizes
n1 and n2 respectively then sample proportions can be defined as
X1 X
p1  and p 2  2
n1 n2
As we have seen in Section 2.5 of the Unit 2 of this course that mean and
variance of the sampling distribution of difference of sample proportions are
E(p1-p2) = P1-P2
and variance
P1Q1 P2Q 2
Var(p 1-p2)  
n1 n2

where, Q1 = 1‒ P1 and Q2 = 1‒ P2.

Now, two cases arise:
Case I: When sample sizes are not sufficiently large i.e. any of the conditions
n1p1  5 or n1q1  5 or n 2 p2  5 or n 2q2  5 does not meet, then we use exact
binomial test. But exact binomial test is beyond the scope of this course.
Case II: When sample sizes are sufficiently large, such that n1p1  5, n1q1  5,
n2p2  5 and n2q2  5 then by central limit theorem, the sampling distribution
of sample proportions p 1 and p 2 are approximately normally as

44
  PQ   PQ  Large Sample Tests
p1 ~ N  P1 , 1 1  and p 2 ~ N  P2 , 2 2 
 n1   n2 

Also, by the property of normal distribution described in Unit 13 of MST-003,

the sampling distribution of the difference of sample proportions follows
normal distribution with mean
E(p1-p2) = E(p1)-E(p2) = P1-P2 … (7)
and variance
P1Q1 P2Q 2
Var(p1-p2) = Var(p 1)+Var(p2)  
n1 n2

That is,
 PQ P Q 
p1  p 2 ~ N P1  P2 , 1 1  2 2 
 n1 n2 

Thus, standard error is given by

P1Q1 P2Q 2
SE  p1  p 2   Var  p1  p 2    … (8)
n1 n2

Now, follow the same procedure as we have discussed in Section 10.2, first of
all we have to setup null and alternative hypotheses. Here, we want to test the
hypothesis about the difference of two population proportions so we can take
the null hypothesis as
 Here, 1  P1 and 
H 0 : P1  P2 (no difference in proportions)   2  P2 if we compare 
 it with general 
 procedure. 

or H 0 : P1  P2  0 (difference in two proportions is 0)

and the alternative hypothesis may be
H1 : P1  P2 for two-tailed test 

H 0 : P1  P2 and H1 : P1  P2 
or  for one-tailed test
H 0 : P1  P2 and H1 : P1  P2 

For testing the null hypothesis, the test statistic Z is given by

 p1  p 2   E  p1  p 2 
Z
SE  p1  p 2 

 p1  p2    P1  P2 
or Z  using equations (7) and (8) 
P1Q1 P2Q2

n1 n2
Since under null hypothesis we assume that P1 = P2 = P, therefore, we have

45
Testing of Hypothesis p1  p 2
Z
1 1 
PQ   
 n1 n 2 
where, Q = 1-P.
Generally, P is unknown then it is estimated by the value of pooled proportion
P̂, where
n p  n 2 p 2 X1  X 2 ˆ  1  Pˆ
Pˆ  1 1  and Q
n1  n 2 n1  n 2
After that, we calculate the value of test statistic and compare it with the
critical value(s) given in Table 10.1 at prefixed level of significance α. Take
the decision about the null hypothesis as described in Section 10.2.
Now, it is time for doing some examples for testing of hypothesis about the
difference of two population proportions.
Example 7: In a random sample of 100 persons from town A, 60 are found to
be high consumers of wheat. In another sample of 80 persons from town B, 40
are found to be high consumers of wheat. Do these data reveal a significant
difference between the proportions of high wheat consumers in town A and
town B ( at α = 0.05 )?
Solution: Here, attribute under study is high consuming of wheat. And we
define our success and failure as getting a person of high consumer of wheat
and not high consumer of wheat respectively.
We are given that
n1 = total number of persons in the sample of town A = 100
n2 = total number of persons in the sample of town B = 80
X1 = number of persons of high consumer of wheat in town A = 60
X2 = number of persons of high consumer of wheat in town B = 40
The sample proportion of high wheat consumers in town A is
X1 60
p1    0.60
n1 100
and the sample proportion of wheat consumers in town B is
X 2 40
p2    0.50
n 2 80
Here, we want to test that the proportion of high consumers of wheat in two
towns, say, P1 and P2, is not same. So our claim is P1 ≠ P2 and its complement
is P1 = P2. Since the complement contains the equality sign, so we can take the
complement as the null hypothesis and the claim as the alternative hypothesis.
Thus,
H 0 : P1  P2  P and H 1 : P1  P2

Since the alternative hypothesis is two-tailed so the test is two-tailed test.

Before proceeding further, first we have to check whether the condition of
normality meets or not.

46
 n1p1  100  0.60  60  5, n1q1  100  0.40  40  5 Large Sample Tests

n 2 p2  80  0.50  40  5, n 2q 2  80  0.50  40  5
We see that condition of normality meets, so we can go for Z-test.
The estimate of the combined proportion (P) of high wheat consumers in two
towns is given by
n1p1  n 2p2 X1  X 2 60  40 5
P̂    
n1  n 2 n1  n 2 100  80 9

ˆ  1  Pˆ  1  5  4
Q
9 9
For testing the null hypothesis, the test statistic Z is given by
p1  p 2
Z
ˆ  1  1 
P̂Q
 n1 n 2 
0.60  0.50 0.10
   1.34
5 4 1 1  0.0745
 
9 9  100 80 
The critical values for two-tailed test at 5% level of significance are ± zα/2
= ± z0.025 = ±1.96.
Since calculated value of Z (=1.34) is less than the critical value (= 1.96) and
greater than critical value (= −1.96), that means calculated value of Z lies in
non-rejection region, so we do not reject the null hypothesis and reject the
alternative hypothesis i.e. we reject the claim.
Decision according to p-value:
Since the test is two-tailed, therefore
p-value = 2 P  Z  z   2P  Z  1.34

 2 0.5  P  0  Z  1.34  2  0.5  0.4099   0.1802

Since p-value (= 0.1802) is greater than  (  0.05) so we do not reject the null
hypothesis at 5% level of significance.
Thus, we conclude that the samples provide us the sufficient evidence against
the claim so we may assume that the proportion of high consumers of wheat in
two towns A and B is same.
Example 8: A machine produced 60 defective articles in a batch of 400. After
overhauling it produced 30 defective in a batch of 300. Has the machine
improved due to overhauling? (Take  = 0.01).
Solution: Here, the machine produced articles and attribute under study is
defectiveness. And we define our success and failure as getting a defective or
non defective article. Therefore, we are given that
X1 = number of defective articles produced by the machine before overhauling
= 60
X2 = number of defective articles produced by the machine after overhauling
= 30

47
Testing of Hypothesis and n1  400, n 2  300,
Let p1 = Observed proportion of defective articles in the sample before the
overhauling
X1 60
   0.15
n1 400
and p2 = Observed proportion of defective articles in the sample after the
overhauling
X2 30
   0.10
n 2 300
Here, we want to test that machine improved due to overhauling that means the
proportion of defective articles is less after overhauling. If P1 and P2 denote the
proportion defectives before and after the overhauling the machine so our claim
is P1 > P2 and its complement P1 ≤ P2. Since the complement contains the
equality sign so we can take the complement as the null hypothesis and claim
as the alternative hypothesis. Thus,
H 0 : P1  P2 and H1 : P1  P2
Since the alternative hypothesis is right-tailed so the test is right-tailed test.
Since P is unknown, so the pooled estimate of proportion is given by
X1  X 2 60  30 90 9 ˆ  1  Pˆ  1  9  61 .
P̂     and Q
n1  n 2 400  300 700 70 70 70
Before proceeding further, first we have to check whether the condition of
normality meets or not.
 n1p1  400  0.15  60  5, n1q1  400  0.85  340  5
n 2 p2  300  0.10  30  5, n 2q 2  300  0.90  270  5
We see that condition of normality meets, so we can go for Z-test.
For testing the null hypothesis, the statistic is given by
p1  p 2
Z
ˆ  1  1 
P̂Q
 n1 n 2 
0.15  0.10 0.05
   1.95
9 61  1 1  0.0256
 
70 70  400 300 
The critical value for right-tailed test at 1% level of significance is
zα = z0.01 = 2.33.
Since calculated value of Z (= 1.95) is less than the critical value (= 2.33) that
means calculated value of Z lies in non-rejection region, so we do not reject the
null hypothesis and reject the alternative hypothesis i.e. we reject the claim at
1% level of significance.
Decision according to p-value:
Since the test is right-tailed, therefore,
p-value = P Z  z  P Z  1.95

48
  0.5  P 0  Z  1.95  0.5  0.4744  0.0256 Large Sample Tests

Since p-value (= 0.0256) is greater than  (  0.01) so we do not reject the null
hypothesis at1% level of significance.
Thus, we conclude that the samples provide us sufficient evidence against the
claim so the machine has not been improved after overhauling.
Now, you can try the following exercises.
E10) The proportions of literates between groups of people of two districts A
and B are tested. Out of the 100 persons selected at random from each
district, 50 from district A and 40 from district B are found literates. Test
whether the proportion of literate persons in two districts A and B is
same at 1% level of significance?
E11) In a large population 30% of a random sample of 1200 persons had blue-
eyes and 20% of a random sample of 900 persons had the same blue-
eyes in another population. Test the proportion of blue-eyes persons is
same in two populations at 5% level of significance.

10.7 TESTING OF HYPOTHESIS FOR

POPULATION VARIANCE USING Z-TEST
In Section 10.3, we have discussed testing of hypothesis for population mean
but when analysing quantitative data, it is often important to draw conclusion
about the average as well as the variability of a characteristic of under study.
For example, a company manufactured the electric bulbs and the manager of
the company would probably be interested in determining the average life of
the bulbs and also determining whether or not the variability in the life of bulbs
is within acceptable limits, the product controller of a milk company may be
interested to know variance of the amount of fat in the whole milk processed
by the company is no more than the specified level, etc. So we require a test for
this purpose.
The procedure of testing a hypothesis for population variance or standard
deviation is similar to the testing of population mean.
For testing a hypothesis about the population variance, we draw a random
sample X 1 , X 2 ,..., X n of size n  30 from the population with mean  and
variance σ2 where,  be known or unknown.
We know that by central limit theorem that sample variance is asymptotically
normally distributed with mean σ2 and variance 2σ4/n whether parent
population is normal or non-normal. That is, if S2 is the sample variance of
the random sample then
2σ 4
E(S2 )  σ 2 and Var(S2 ) = … (9)
n

But we know that standard error = Variance

2 2
 SE  S2   Var  S2   σ … (10)
n
The general procedure of this test is explained in the next page.
49
Testing of Hypothesis As we are doing so far in all tests, first Step in hypothesis testing problems is to
setup null and alternative hypotheses. Here, we want to test the hypothesis
specified value 20 of the population variance 2 so we can take our null and
alternative hypotheses as

H 0 : 2  02 and H1 : 2  02 for two-tailed test 

H0 : 2  02 and H1 : 2  02 
or  for one-tailed test 
H0 : 2  02 and H1 : 2  02 

For testing the null hypothesis, the test statistic Z is given by

S2  E  S2 
Z ~ N  0,1 
SE  S2 

S2  σ20  Usingequations(9) and(10) and 

Z 2 2
2  under H0 : σ  σ0 
σ02
n
After that, we calculate the value of test statistic and compare it with the
critical value given in Table 10.1 at prefixed level of significance α. Take the
decision about the null hypothesis as described in Section 10.2.
Note 2: When population under study is normal then for testing the hypothesis
about population variance or population standard deviation we use chi-square
test which will be discussed in Unit 12 of this course. Whereas when the
distribution of the population under study is not known and sample size is large
then we apply Z-test as discussed above.
Now, it is time to do example based on above test.
Example 9: A random sample of size 65 screws is taken from a population of
big box of screws and measured their length (in mm) which gives sample
variance 9.0. Test that the two years old population variance 10.5 is still
maintained at present at 5% level of significance.
Solution: We are given that
n  65, S2  9.0, 20  10.5
Here, we want to test that the two years old screw length population variance
(σ2) is still maintained at 10.5. So our claim is σ2 = 10.5 and its complement is
σ2 ≠ 10.5. Since the claim contains the equality sign so we can take the claim as
the null hypothesis and complement as the alternative hypothesis. Thus,
H 0 : 2  02  10.5 and H1 : 2  10.5
Since the alternative hypothesis is two-tailed so the test is two-tailed test.
Here, the distribution of population under study is not known and sample size
is large (n > 30) so we can go for Z-test.
For testing the null hypothesis, the test statistic Z is given by
S2  σ20
Z ~ N  0,1 
2 2
σ0
n
50
 9.0  10.5  1.5 1.5 Large Sample Tests
    0.81
2 10.5  0.175 1.84
10.5
65
The critical values for two-tailed test at 5% level of significance are
± zα/2 = ± z0.025 = ±1.96.
Since calculated value of Z (= −0.81) is less than critical value (= 1.96) and
greater than the critical value (= −1.96), that means it lies in non-rejection
region, so we do not reject the null hypothesis i.e. we support the claim.
Thus, we conclude that sample fails to provide us sufficient evidence against
the claim so we may assume that two years old screw length population
variance is still maintained at 10.5mm.
Now, you can try the following exercise.
E12) A random sample of size 120 bulbs is taken from a lot which gives the
standard deviation of the life of electric bulbs 7 hours. Test the standard
deviation of the life of bulbs of the lot is 6 hours at 5% level of
significance.

10.8 TESTING OF HYPOTHESIS FOR TWO

POPULATION VARIANCES USING Z-TEST
In previous section, we have discussed testing of hypothesis about the
population variance. But there are so many situations where we want to test the
hypothesis about equality of two population variances or standard deviations.
For example, an economist may want to test whether the variability in incomes
differ in two populations, a quality controller may want to test whether the
quality of the product is changing over time, etc.
Let there be two populations, say, population-I and population-II under study.
Also let 1, 2 and 12 , 22 denote the means and variances of population-I and
population-II respectively where both 12 and 22 are unknown but 1 and  2 may
be known or unknown. For testing the hypothesis about equality of two
population variances or standard deviations, we draw a random sample of large
size n1 from population-I and a random sample of large size n2 from
population-II. Let S12 and S22 be the sample variances of the samples selected
from population-I and population-II respectively.
These two populations may or may not be normal but according to the central
limit theorem, the sampling distribution of difference of two large sample
variances asymptotically normally distributed with mean (σ12  σ 22 ) and
variance (2σ14 / n1  2σ24 / n2 ) .
Thus,
E S12  S22   E S12   E S22   12  22 … (11)

and

214 224
Var S12  S22   Var S12   Var S22   
n1 n2

51
Testing of Hypothesis But we know that standard error = Variance

2 14 2  42
 SE S12  S22   Var  S12  S22    … (12)
n1 n2

Now, follow the same procedure as we have discussed in Section 10.2, that is,
first of all we have to setup null and alternative hypothesis. Here, we want to
test the hypothesis about the two population variances, so we can take our null
and alternative hypotheses as
H 0 : 12  22  2 and H1 : 12  22 for two-tailed test 
H 0 : 12  22 and H1 : 12  22 
or  for one-tailed test 
H 0 : 12  22 and H1 : 12  22 

For testing the null hypothesis, the test statistic Z is given by

 S12  S22   E  S12  S22  ~ N 0,1
Z  
SE  S12  S22 

or
 S12  S22    σ12  σ 22 
Z  using equations (11) and (12) 
2σ14 2σ 42

n1 n2

Since under null hypothesis we assume that σ12  σ22  σ2 , therefore, we have

S12  S22
Z ~ N  0,1 
 1 1 
σ2 2  
 n1 n 2 
Generally, population variances σ12 and σ 22 are unknown, so we estimate them
by their corresponding sample variances S12 and S 22 as

ˆ 12  S12 and ˆ 22  S 22
Thus, the test statistic Z is given by
S12  S22
Z ~ N  0,1
 2S14 2S42 
 n  n 
 1 2 

After that, we calculate the value of test statistic as may be the case and
compare it with the critical value given in Table 10.1 at prefixed level of
significance α. Take the decision about the null hypothesis as described in
Section 10.2.
Note 3: When populations under study are normal then for testing the
hypothesis about equality of population variances we use F- test which will be
discussed in Unit 12 of this course. Whereas when the form of the populations
under study is not known and sample sizes are large then we apply Z-test as
discussed above.
Now, it is time to do an example based on above test.

52
Example 10: A comparative study of variation in weights (in pound) of Army- Large Sample Tests
soldiers and Navy- sailors was made. The sample variance of the weight of 120
soldiers was 60 pound2 and the sample variance of the weight of 160 sailors
was 70 pound2. Test whether the soldiers and sailors have equal variation in
their weights. Use 5% level of significance.

Solution: Given that

n1  120, S12  60, n 2  160 , S22  70

We want to test that the Army-soldiers and Navy-sailors have equal variation
in their weights. If 12 and 22 denote the variances in the weight of Army-
soldiers and Navy-sailors so our claim is 12  22 and its complement is
12  22 . Since the claim contains the equality sign so we can take the claim as
the null hypothesis and complement as the alternative hypothesis. Thus,

H 0 : 12  22  Army-soldiers and Navy-sailors 

 have equal variation in their weights 

and the alternative hypothesis as

H1 : 12   22  Army-soldiers and Navy-sailors 

 have different variation in their weights 
Here, the distributions of populations under study are not known and sample
sizes are large  n1  120  30, n 2  160  30  so we can go for Z-test.

Since population variances are unknown so for testing the null hypothesis, the
test statistic Z is given by
S12  S22
Z
 2S14 2S24 
 n  n 
 1 2 

60  70

2 2
2   60  2   70 

120 160
10 10
   0.91
60.0  61.25 11.01

The critical values for two-tailed test at 5% level of significance are

± zα/2 = ± z0.025 = ±1.96.
Since calculated value of Z (= −0.91) is less than critical value (= 1.96) and
greater than the critical value (= −1.96), that means it lies in non-rejection
region, so we do not reject the null hypothesis i.e. we support the claim.
Thus, we conclude that samples fail to provide us sufficient evidence against
the claim so we may assume that the Army-soldiers and Navy-sailors have
equal variation in their weights.
Now, you can try the following exercise.

53
Testing of Hypothesis E 13) Two sources of raw materials of bulbs are under consideration by a bulb
manufacturing company. Both sources seem to have similar
characteristics but the company is not sure about their respective
uniformity. A sample of 52 lots from source A yields variance 25 and a
sample of 40 lots from source B yields variance of 12. Test whether the
variance of source A significantly differs to the variances of source B at
 = 0.05?
We now end this unit by giving a summary of what we have covered in it.

10.9 SUMMARY
In this unit we have covered the following points:
1. How to judge a given situation whether we should go for large sample test
or not.
2. Applying the Z-test for testing the hypothesis about the population mean
and difference of two population means.
3. Applying the Z-test for testing the hypothesis about the population
proportion and difference of two population proportions.
4. Applying the Z-test for testing the hypothesis about the population variance
and two population variances.

10.10 SOLUTIONS / ANSWERS

E1) Since we have a rule that if the observed value of test statistic lies in
rejection region then we reject the null hypothesis and if calculated
value of test statistic lies in non-rejection region then we do not reject
the null hypothesis. Therefore in our case, we do not reject the null
hypothesis. So (iii) is the correct answer. Remember always on the
basis of the one sample we never accept the null hypothesis.
E2) Since the test is two-tailed therefore rejection region will lie under both
tails.
E3) Since test is two-tailed, therefore,

p-value = 2P  Z  z   2P  Z  2.42

 2 0.5  P  0  Z  2.42  2  0.5  0.4922   0.0156

E4) We are given that

n  900, X  3.4 cm, μ 0  3.25 cm and σ  2.61 cm
Here, we wish to test that the sample comes from a large population of
bolts with mean (µ) 3.25cm. So our claim is µ = 3.25cm and its
complement is µ ≠ 3.25cm. Since the claim contains the equality sign
so we can take the claim as the null hypothesis and the complement as
the alternative hypothesis. Thus,
H0 : µ  µ0  3.25 and H1 : µ  3.25
Since the alternative hypothesis is two-tailed, so the test is two-tailed
test.
54
 Here, we want to test the hypothesis regarding population mean when Large Sample Tests
population SD is unknown, so we should use t-test if the population of
bolts known to be normal. But it is not the case. Since the sample size is
large (n > 30) so we can go for Z-test instead of t-test as an
approximate. So test statistic is given by
X  0
Z
/ n
3.40  3.25 0.15
   1.72
2.61/ 900 0.087
The critical (tabulated) values for two-tailed test at 5% level of
significance are ± zα/2 = ± z0.025 = ± 1.96.
Since calculated value of test statistic Z (= 1.72) is less than the critical
value (= 1.96) and greater than critical value (= −1.96), that means it
lies in non-rejection region, so we do not reject the null hypothesis i.e.
we support the claim at 5% level of significance.
Thus, we conclude that sample does not provide us sufficient evidence
against the claim so we may assume that the sample comes from the
population of bolts with mean 3.25cm.
E5) Here, we given that
µ 0  1200, n  100, X  1220, S  90
Here, the company may accept the new CFL light when average life of
CFL light is greater than 1200 hours. So the company wants to test that
the new brand CFL light has an average life greater than 1200 hours. So
our claim is  > 1200 and its complement is  ≤ 1200. Since
complement contains the equality sign so we can take the complement
as the null hypothesis and the claim as the alternative hypothesis. Thus,
H0 : µ  µ0  1200 and H1 : µ  1200
Since the alternative hypothesis is right-tailed so the test is right-tailed
test.
Here, we want to test the hypothesis regarding population mean when
population SD is unknown, so we should use t-test if the distribution of
life of bulbs known to be normal. But it is not the case. Since the
sample size is large (n > 30) so we can go for Z-test instead of t-test.
Therefore, test statistic is given by
X  0
Z
S/ n
1220  1200 20
   2.22
90/ 100 9
The critical values for right-tailed test at 5% level of significance is
zα = z0.05 = 1.645.
Since calculated value of test statistic Z (= 2.22) is greater than critical
value (= 1.645), that means it lies in rejection region so we reject the
null hypothesis and support the alternative hypothesis i.e. we support
our claim at 5% level of significance.
55
Testing of Hypothesis Thus, we conclude that sample does not provide us sufficient evidence
against the claim so we may assume that the company accepts the new
brand of bulbs.
E6) Given that
n1  200, X  1300, S1  41;

n 2  200, Y  1280, S2  46
Here, we want to test that there is significant difference in the mean
duration of their lives of two brands of electric bulbs. If 1 and 2
denote the mean lives of two brands of electric bulbs respectively then
our claim is 1 ≠ 2 and its complement is 1 = 2. Since the
complement contains the equality sign so we can take the complement
as the null hypothesis and the claim as the alternative hypothesis. Thus,
H 0 : 1   2 and H1 : 1   2
Since the alternative hypothesis is two-tailed so the test is two-tailed
test.
We want to test the null hypothesis regarding equality of two
population means. The standard deviations of both populations are
unknown so we should go for t-test if population of difference is known
to be normal. But it is not the case. Since sample sizes are large (n1, and
n2 > 30) so we go for Z-test.
So for testing the null hypothesis, the test statistic Z is given by
XY
Z
S12 S22

n1 n 2
1300  1280 20 20
    4.59
 41
2
 46
2
8.41  10.58 4.36

200 200
The critical (tabulated) values for two-tailed test at 1% level of
significance are ± zα/2 = ± z0.005 = ± 2.58.
Since calculated value of test statistic Z (= 4.59) is greater than the
critical values (= ± 2.58), that means it lies in rejection region, so we
reject the null hypothesis an support the alternative hypothesis i.e.
support the claim at 1% level of significance.
Thus, we conclude that samples do not provide us sufficient evidence
against the claim so there is significant difference in the mean duration
of their lives of two brands of electric bulbs.
E7) Given that
n1  50, X  8.3, S1  1.2;

n 2  100 , Y  8.0, S2  1.5

Here, we want to test that the first drug provides a significant longer
period of relief than the other. If 1 and 2 denote the mean relief time
due to first and second drugs respectively then our claim is 1 > 2 and
56
 its complement is 1 ≤ 2. Since complement contains the equality sign Large Sample Tests
so we can take the complement as the null hypothesis and the claim as
the alternative hypothesis. Thus,
H 0 : 1   2 and H1 : 1   2
Since the alternative hypothesis is right-tailed so the test is right-tailed
test.
We want to test the null hypothesis regarding equality of two
population means. The standard deviations of both populations are
unknown. So we should go for t-test if population of difference is
known to be normal. But it is not the case. Since sample sizes are large
(n1, and n2 > 30) so we go for Z-test.
So for testing the null hypothesis, the test statistic Z is given by
XY
Z
S12 S22

n1 n 2

8.3  8.0 0.3

 2 2

 1.2   1.5  0.0288  0.0255

50 100
0.3
  1.32
0.2265
The critical (tabulated) value for right-tailed test at 5% level of
significance is zα = z0.05 = 1.645.
Since calculated value of test statistic Z (= 1.32) is less than the critical
value (=1.645), that means it lies in non-rejection region, so we do not
reject the null hypothesis and reject the alternative hypothesis i.e. we
reject the claim at 5% level of significance.
Thus, we conclude that samples provide us sufficient evidence against
the claim so the first drug does not has longer period of relief than the
other.
E8) Here, students are classified as Science background and other. We
define our success and failure as getting student of Science background
and other respectively. We are given that
n = Total number of students is the sample = 100
X = Number of students from Science background = 54
p = Sample proportion of students from Science background
54
  0.54
100
We want to test whether 50% of the students are from Science
background in MSc. If P denotes the proportion of first year Science
background students in the University. So our claim is P = P0 = 0.5 and
its complement is P ≠ 0.5. Since the claim contains the equality sign so
we can take the claim as the null hypothesis and complement as the
alternative hypothesis. Thus,
57
Testing of Hypothesis H0 = P = P0 = 0.5 (= 50%)
H1 : P ≠ 0.5 [Science background differs to 50%]
Since the alternative hypothesis is two-tailed so the test is two-tailed
test.
Before proceeding further, first we have to check whether the condition
of normality meets or not.
 np  100  0.54  54  5
nq  100   1  0.54   100  0.46  46  5

We see that condition of normality meets, so we can go for Z-test.

For testing the null hypothesis, the test statistic Z is given by
p  P0 0.54  0.50
Z   Q 0  1  P0 
P0Q 0 0.5  0.5 /100
n
0.04
  0.80
0.05
The critical (tabulated) values for two-tailed test at 1% level of
significance are ± zα/2 = ± z0.005 = ± 2.58.
Since calculated value of test statistic Z (= 0.80) is less than the critical
value (= 2.58) and greater than critical value (= −2.58), that means it
lies in non-rejection region, so we do not reject the null hypothesis i.e.
we support the claim at 1% level of significance.
Thus, we conclude that sample fails to provide us sufficient evidence
against the claim so we may assume that 50% of the first year students
in MSc. Economics in the University are from Science background.
E9) We define our success and failure as a patient is survived and not
survived. Here, we are given that
n = 200
X = Number of survived patients who are given a particular injection
= 180
p = Sample proportion of survived patients who are given a particular
injection
X 180
   0.9
n 200
80 80
P0  80%   0.80  Q 0  1  P0   0.20
100 100
Here, we want to test that the survival rate of the patients is more than
80%. If P denotes the proportion of survival patients then our claim is P
> 0.80 and its complement is P ≤ 0.80. Since complement contains the
equality sign so we can take the complement as the null hypothesis and
the claim as the alternative hypothesis. Thus,
H 0 : P  P0  0.80 and H1 : P  0.80

58
 Since the alternative hypothesis is right-tailed so the test is right-tailed Large Sample Tests
test.
Before proceeding further, first we have to check whether the condition
of normality meets or not.
 np  200  0.9  180  5

nq  200   1  0.9   200  0.1  20  5

We see that condition of normality meets, so we can go for Z-test.
For testing the null hypothesis, the test statistic Z is given by
p  P0
Z
P0 Q 0
n
0.9  0.8 0.1
   3.53
 0.8  0.2  0.0283
200
The critical (tabulated) value for right-tailed test at 5% level of
significance is zα = z0.05 = 1.645.
Since calculated value of test statistic Z (= 3.53) is greater than the
critical value (=1.645), that means it lies in rejection region, so we
reject the null hypothesis and support the alternative hypothesis i.e.
support our claim at 5% level of significance.
Thus, we conclude that sample fails to provide us sufficient evidence
against the claim so we may assume that the survival rate is greater than
80% in the population using that injection.
E10) Let X1 and X2 stand for number of literates in districts A and B
respectively. Therefore, we are given that
X1 50
n1  100, X1  50  p1    0.50
n1 100
X 40
n 2  100, X 2  40  p 2  2   0.40
n 2 100
Here, we want to test whether the proportion of literate persons in two
districts A and B is same. If P1 and P2 denote the proportions of literate
persons in two districts A and B respectively then our claim is P1 = P2
and its complement P1 ≠ P2. Since the claim contains the equality sign
so we can take the claim as the null hypothesis and complement as the
alternative hypothesis. Thus,
H 0 : P1  P2  P and H1 : P1  P2

Since the alternative hypothesis is two-tailed so the test is two tailed

test.
The estimate of the combined proportion (P) of literates in districts A
and B is given by
n1p1  n 2 p2 X1  X 2 50  40
P̂     0.45
n1  n 2 n1  n 2 100  100
ˆ  1  Pˆ  1  0.45  0.55
Q
59
Testing of Hypothesis Before proceeding further, first we have to check whether the condition
of normality meets or not.
 n1p1  100  0.50  50  5, n1q1  100  0.50  50  5
n 2 p2  100  0.40  40  5, n 2q 2  100  0.60  60  5
We see that condition of normality meets, so we can go for Z-test.
For testing the null hypothesis, the test statistic Z is given by
p1  p 2
Z
ˆ  1  1 
P̂Q
 n1 n 2 
0.50  0.40 0.10
   1.42
1 1  0.0704
0.45  0.55   
 100 100 
The critical (tabulated) values for two-tailed test at 1% level of
significance are ± zα/2 = ± z0.005 = ± 2.58.
Since calculated value of test statistic Z (= 1.42) is less than the critical
value (= 2.58) and greater than critical value (= −2.58), that means it
lies in non-rejection region, so we do not reject the null hypothesis i.e.
we support our claim at 1% level of significance.
Thus, we conclude that samples fail to provide us sufficient evidence
against the claim so we may assume that the proportion of literates in
districts A and B is equal.
E11) Here, we are given that
n1  1200, p1  30%  0.30

n 2  900, p2  20%  0.20

Here, we want to test that the proportion of blue-eye persons in both the
population is same. If P1 and P2 denote the proportions of blue-eye
persons in two populations respectively then our claim is P1 = P2 and its
complement P1 ≠ P2. Since the claim contains the equality sign so we
can take the claim as the null hypothesis and complement as the
alternative hypothesis. Thus,
H 0 : P1  P2  P and H1 : P1  P2

Since the alternative hypothesis is two-tailed so the test is two tailed

test.
The estimate of the combined proportion (P) of literates in districts A
and B is given by

n1p1  n 2 p2 1200  0.30  900  0.20

P̂    0.257
n1  n 2 1200  900
ˆ  1  Pˆ  1  0.257  0.743
Q

Before proceeding further, first we have to check whether the condition

of normality meets or not.
60
  n1p1  1200  0.30  360  5, n1q1  1200  0.70  840  5 Large Sample Tests

n 2p2  900  0.20  180  5, n 2q 2  900  0.80  720  5

We see that condition of normality meets, so we can go for Z-test.
For testing the null hypothesis, the test statistic Z is given by
p1  p 2
Z
ˆ  1  1 
P̂Q
 n1 n 2 
0.30  0.20 0.10
   5.26
 1 1  0.019
0.257  0.743   
 1200 900 
The critical (tabulated) values for two-tailed test at 5% level of
significance are ± zα/2 = ± z0.005 = ± 1.96.
Since calculated value of test statistic Z (= 5.26) is greater than critical
values (= ±1.96), that means it lies in rejection region, so we reject the
null hypothesis i.e. we reject our claim at 5% level of significance.
Thus, we conclude that samples provide us sufficient evidence against
the claim so the proportion of blue-eyed person in both the population
is not same.
E12) Here, we are given that
n  120, S  7, 0  6
Here, we want to test that standard deviation (σ) of the life of bulbs of
the lot is 6 hours. So our claim is σ = 6 and its complement is σ ≠ 6.
Since the claim contains the equality sign so we can take the claim as
the null hypothesis and complement as the alternative hypothesis. Thus,
H 0 :   0  6 and H1 :   6
Since the alternative hypothesis is two-tailed so the test is two-tailed
test.
Here, the distribution of population under study is not known and
sample size is large (n > 30) so we can go for Z-test.
For testing the null hypothesis, the test statistic Z is given by
S2  σ02
Z ~ N  0,1 
2
σ02
n
2 2


 7  6 
13

13
 2.80
2 2 36  0.129 4.64
6
120
The critical values for two-tailed test at 5% level of significance are
± zα/2 = ± z0.025 = ±1.96.
Since calculated value of Z (= 2.8) is greater than critical values
(= ±1.96), that means it lies in rejection region, so we reject the null
hypothesis i.e. we reject our claim at 5% level of significance.

61
Testing of Hypothesis Thus, we conclude that sample provides us sufficient evidence against
the claim so standard deviation of the life of bulbs of the lot is not 6.0
hours.
E13) Here, we are given that
n1  52, S12  25
n2  40, S22  12
Here, we want to test that variance of source A significantly differs to
the variances of source B. If 12 and 22 denote the variances in the raw
materials of sources A and B respectively so our claim is 12  22 and
its complement is 12   22 . Since complement contains the equality
sign so we can take the complement as the null hypothesis and the
claim as the alternative hypothesis. Thus,
H0 : 12  22 and H1 : 12  22
Since the alternative hypothesis is two-tailed so the test is two-tailed
test.
Here, the distributions of populations under study are not known and
sample sizes are large (n1  52  30, n 2  40  30) so we can go for Z-
test.
Since population variances are unknown so for testing the null
hypothesis, the test statistic Z is given by
S12  S22
Z
 2S14 2S24 
 n  n 
 1 2 

25  12 13
   2.36
2
225 212
2 5.5

52 40
The critical values for two-tailed test at 5% level of significance are
± zα/2 = ± z0.025 = ±1.96.
Since calculated value of Z (= 2.36) is greater than critical values (= ±1.96),
that means it lies in rejection region, so we reject the null hypothesis and
support the alternative hypothesis i.e. we support our claim at 5% level of
significance.
Thus, we conclude that samples fail to provide us sufficient evidence against
the claim so variance of source A significantly differs to the variance of source
B.

62
UNIT 11 SMALL SAMPLE TESTS
Structure
11.1 Introduction
Objectives
11.2 General Procedure of t-Test for Testing a Hypothesis
11.3 Testing of hypothesis for Population Mean Using t-Test
11.4 Testing of Hypothesis for Difference of Two Population Means Using
t-Test
11.5 Paired t-Test
11.6 Testing of Hypothesis for Population Correlation Coefficient Using
t-Test
11.7 Summary
11.8 Solutions /Answers

11.1 INTRODUCTION
In previous unit, we have discussed the testing of hypothesis for large samples
in details. Recall that throughout the unit, we were making an assumption that
“if sample size is sufficiently large then test statistic follows approximately
standard normal distribution”. Also recall two points highlighted in this course,
i.e.
 Cost of our study increases as sample size increases.
 Sometime nature of the units in the population under study is such that they
destroyed under investigation.
If there are limited recourses in terms of money then first point listed above
force us not to go for large sample size when items /units under study are very
costly such as airplane, computer, etc. Second point listed above give an alarm
for not to go for large sample if population units are destroyed under
investigation.
So, we need an alternative technique which is used to test the hypothesis based
on small sample(s). Small sample tests do this job for us. But in return they
demand one basic assumption that population under study should be normal as
you will see when you go through the unit. t,  2 and F -tests are some
commonly used small sample tests.
In this unit, we will discuss t-test in details which is based on the t-distribution
described in Unit 3 of this course. And  2 and F-tests will be discussed in next
unit which are based on χ2 and F-distributions described in Unit 3 and Unit 4 of
this course respectively.
This unit is divided into eight sections. Section 11.1 is described the need of
small sample tests. The general procedure of t-test for testing a hypothesis is
described in Section 11.2. In Section 11.3, we discuss testing of hypothesis for
population mean using t-test. Testing of hypothesis for difference of two
population means when samples are independent is described in Section 11.4
whereas in Section 11.5, the paired t-test for difference of two population
means when samples are dependent(paired) is discussed. In Section 11.6
testing of hypothesis for population correlation coefficient is explained. Unit

63
Testing of Hypothesis ends by providing summary of what we have discussed in this unit in Section
11.7 and solution of exercises in Section 11.8.
Before moving further a humble suggestion to you that please revise what you
have learned in previous two units. The concepts discussed there will help you
a lot to better understand the concepts discussed in this unit.
Objectives
After studying this unit, you should be able to:
 realize the importance of small sample tests;
 know the procedure of t-test for testing a hypothesis;
 describe testing of hypothesis for population mean for using t-test;
 explain the testing of hypothesis for difference of two population means
when samples are independent using t-test;
 describe the procedure for paired t-test for testing of hypothesis for
difference of two population means when samples are dependent or paired;
and
 explain the testing of hypothesis for population correlation coefficient using
t-test.

11.2 GENERAL PROCEDURE OF t-TEST FOR

TESTING A HYPOTHESIS
The general procedure of t-test for testing a hypothesis is similar as Z-test
already explained in Unit 10. Let us give you similar details here.
For this purpose, let X1, X2,…, Xn be a random sample of small size n (< 30)
selected from a normal population (recall the demand of small sample tests
pointed out in previous Section 11.1) having parameter of interest, say, 
which is actually unknown but its hypothetical value, say, 0 estimated from
some previous study or some other way is to be tested. t-test involves following
steps for testing this hypothetical value:
Step I: First of all, we setup null and alternative hypotheses. Here, we want
to test the hypothetical value θ0 of parameter θ so we can take the null
and alternative hypotheses as
H 0 :   0 and H 1 :   0 for two-tailed test 
H 0 :   0 and H 1 :   0 
or  for one-tailed test 
H 0 :   0 and H 1 :   0 

In case of comparing same parameter of two populations of interest,

say, 1, and 2 then our null and alternative hypotheses would be

H 0 : 1   2 and H1 : 1  2 for two-tailed test

H 0 : 1  2 and H1 : 1  2 
or  for one-tailed test 
H 0 : 1  2 and H1 : 1  2 

Step II: After setting the null and alternative hypotheses our next step is to
decide a criteria for rejection or non-rejection of null hypothesis i.e.
64
 decide the level of significance , at which we want to test our null Small Sample Tests
hypothesis. We generally take  = 5 % or 1%.
Step III: The third step is to determine an appropriate test statistic, say, t for
testing the null hypothesis. Suppose Tn is the sample statistic (may be
sample mean, sample correlation coefficient, etc. depending upon )
for the parameter  then test-statistic t is given by
Tn  E(Tn )
t
SE  Tn 

Step IV: As we know, t-test is based on t-distribution and t-distribution is

described with the help of its degrees of freedom, therefore, test
statistic t follows t-distribution with specified degrees of freedom as
the case may be.
By putting the values of Tn, E(Tn) and SE(T n) in above formula, we
calculate the value of test statistic t. Let tcal be the calculated value of
test statistic t after putting these values.
Step V: After that, we obtain the critical (cut-off or tabulated) value(s) in the
sampling distribution of the test statistic t corresponding to  assumed
in Step II. The critical values for t-test are given in Table-II (t-table)
of the Appendix at the end of Block 1 of this course corresponding to
different level of significance (α). After that, we construct rejection
(critical) region of size  in the probability curve of the sampling
distribution of test statistic t.
Step VI: Take the decision about the null hypothesis based on calculated and
critical value(s) of test statistic obtained in Step IV and Step V
respectively. Since critical value depends upon the nature of the test
that it is one-tailed test or two-tailed test so following cases arise:
In case of one-tailed test:
Case I: When H 0 :   0 and H 1 :   0 (right-tailed test)

In this case, the rejection (critical) region falls under the right tail of
the probability curve of the sampling distribution of test statistic t.
Suppose t(ν), is the critical value at  level of significance then entire
region greater than or equal to t(ν), is the rejection region and less
than t(ν), is the non-rejection region as shown in Fig. 11.1.
If tcal ≥ t(ν,), that means calculated value of test statistic t lies in the Fig. 11.1
rejection (critical) region, then we reject the null hypothesis H0 at 
level of significance. Therefore, we conclude that sample data
provides us sufficient evidence against the null hypothesis and there is
a significant difference between hypothesized value and observed
value of the parameter.
If tcal < t(ν,), that means calculated value of test statistic t lies in non-
rejection region, then we do not reject the null hypothesis H0 at 
level of significance. Therefore, we conclude that the sample data
fails to provide us sufficient evidence against the null hypothesis and
the difference between hypothesized value and observed value of the
parameter due to fluctuation of sample.
Testing of Hypothesis Case II: When H 0 :    0 and H 1 :   0 (left-tailed test)
In this case, the rejection (critical) region falls under the left tail of the
probability curve of the sampling distribution of test statistic t.
Suppose - t(ν), is the critical value at  level of significance then
entire region less than or equal to - t(ν), is the rejection region and
greater than - t(ν), is the non-rejection region as shown in Fig. 11.2.
If tcal ≤ − t(ν),, that means calculated value of test statistic t lies in the
rejection (critical) region, then we reject the null hypothesis H0 at 
level of significance.
If tcal >- t(ν),, that means calculated value of test statistic t lies in the
Fig. 11.2 non-rejection region, then we do not reject the null hypothesis H0 at 
level of significance.
In case of two-tailed test:
That is, when H 0 :   0 and H 1 :   0
In this case, the rejection region falls under both tails of the
probability curve of sampling distribution of the test statistic t. Half
the area (α) i.e. α/2 will lies under left tail and other half under the
right tail. Suppose  t (  ),  / 2 and t (  ),  / 2 are the two critical values at
the left- tailed and right-tailed respectively. Therefore, entire region
less than or equal to  t (  ),  / 2 and greater than or equal to t (  ),  / 2 are
Fig. 11.3
the rejection regions and between  t (  ),  / 2 and t (  ),  / 2 is the non-
rejection region as shown in Fig. 11.3.
If tcal ≥ t(ν),/2, or tcal ≤ -t(ν),/2, that means calculated value of test
statistic t lies in the rejection(critical) region, then we reject the null
hypothesis H0 at  level of significance.
And if -t(ν),/2 < tcal < t(ν),/2, that means calculated value of test
statistic t lies in the non-rejection region, then we do not reject the
null hypothesis H0 at  level of significance.
Procedure of taking the decision about the null hypothesis on the basis of
p-value:
To take the decision about the null hypothesis on the basis of p-value, the p-
value is compared with given level of significance (α). And if p-value is less
than or equal to α then we reject the null hypothesis and if p-value is greater
than α then we do not reject the null hypothesis at α level of significance.
Since the distribution of test statistic t follows t-distribution with ν df and we
also know that t-distribution is symmetrical about t = 0 line therefore, if tcal
represents calculated value of test statistic t then p-value can be defined as:
For one-tailed test:
For H1: θ > θ0 (right-tailed test)
p-value = P[t  tcal]
For H1: θ < θ0 (left-tailed test)
p-value = P  t  t cal 
For two-tailed test: For H1:   0
 p-value = 2P  t  t cal  Small Sample Tests

These p-values for t-test can be obtained with the help of Table-II (t-table)
given in the Appendix at the end of Block 1 of this course. But this table gives
the t-values corresponding to the standard values of α such as 0.10, 0.05,
0.025, 0.01 and 0.005 only, therefore, the exact p-values are not obtained with
the help of this table and we can approximate the p-value for this test.
For example, if test is right-tailed and calculated (observed) value of test
statistic t is 2.94 with 9 df then p-value is obtained as:
Since calculated value of test statistic t is based on the 9 df therefore, we use
row for 9 df in the t-table and move across this row to find the values in which
calculated t-value falls. Since calculated t-value falls between 2.821 and 3.250,
which are corresponding to the values of one-tailed area α = 0.01 and 0.005
respectively, therefore, p-value will lie between 0.005 and 0.01, that is,
0.005  p-value  0.01
If in the above example, the test is two-tailed then the two values 0.01 and
0.005 would be doubled for p-value, that is,
0.005  2  0.01  p-value  0.02  2  0.01
Note 1: With the help of computer packages and softwares such as SPSS, SAS,
MINITAB, EXCEL, etc. we can find the exact p-values for t-test.
Now, you can try the following exercise.

E1) If test is two-tailed and calculated value of test statistic t is 2.42 with 15
df then find the p-value for t-test.

11.3 TESTING OF HYPOTHESIS FOR

POPULATION MEAN USING t-TEST
In Section 10.3 of the previous unit, we have discussed Z-test for testing the
hypothesis about population mean when population variance σ2 is known and
unknown.
Recall form these, we have already pointed out that one basic difference
between Z-test and t-test is that Z-test is used when population SD is known
whether sample size is large or small and t-test is used when population SD is
unknown whether sample size is small or large. But in case of large sample size
Z-test is an appropriate of t-test as we did in previous unit. But in practice
standard deviation of population is not known and sample size is small so in
this situation, we use t-test provided population under study is normal.
Assumptions
Virtually every test has some assumptions which must be met prior to the
application of the test. This t-test needs following assumptions to work:
(i) The characteristic under study follows normal distribution. In other
words, populations from which random sample is drawn should be
normal with respect to the characteristic of interest.
(ii) Sample observations are random and independent.
(iii) Population variance σ2 is unknown.
67
Testing of Hypothesis For describing this test, let X1, X2,…, Xn be a random sample of small size
n (< 30) selected from a normal population with mean µ and unknown
variance σ2.
Now, follow the same procedure as we have discussed in previous section,
that is, first of all we setup the null and alternative hypotheses. Here, we
want to test the claim about the specified value 0 of population mean µ so
we can take the null and alternative hypotheses as
 Here,    and 0  0 
if we compareit with 
H0 :   0 and H1 :   0  for two-tailed test  general procedure. 

H 0 :   0 and H1 :   0 
or   for one-tailed test 
H 0 :   0 and H1 :   0 
For testing the null hypothesis, the test statistic t is given by
X  0
t ~ t  n 1 under H 0
S/ n

1 n 1 n 2
where, X  
n i1
X i is the sample mean and S 2
 
n  1 i1
 X i  X  is the

sample variance.
For computational simplicity, we may use the following formulae for X, S2 :
2
1 1    d 
n  1 
X  a   d and S 
2 2
 d  
n n 
 
where, d = (X − a ), ‘a’ being the assumed arbitrary value.
Here, the test statistic t follows t-distribution with (n − 1) degrees of freedom as
we discussed in Unit 3 of this course.
After substituting values of X, S and n , we get calculated value of test
statistic t. Then we look for critical (or cut-off or tabulated) value(s) of test
statistic t from the t-table. On comparing calculated value and critical value(s),
we take the decision about the null hypothesis as discussed in previous section.
Let us do some examples of testing of hypothesis about population mean using
t-test.
Example 1: A manufacturer claims that a special type of projector bulb has an
average life 160 hours. To check this claim an investigator takes a sample of 20
such bulbs, puts on the test, and obtains an average life 167 hours with standard
deviation 16 hours. Assuming that the life time of such bulbs follows normal
distribution, does the investigator accept the manufacturer’s claim at 5% level
of significance?
Solution: Here, we are given that
0  160, n  20, X  167 and S  16
Here, we want to test the manufacturer claims that a special type of projector
bulb has an average life (µ) 160 hours. So claim is µ = 160 and its complement
is µ ≠ 160. Since the claim contains the equality sign so we can take the claim
as the null hypothesis and complement as the alternative hypothesis. Thus,
68
 H 0 :    0  160 and H 1 :   160 Small Sample Tests

Since the alternative hypothesis is two-tailed so the test is two-tailed test.

Here, we want to test the hypothesis regarding population mean when
population SD is unknown. Also sample size is small n = 20(n < 30) and
population under study is normal, so we can go for t-test for testing the
hypothesis about population mean.
For testing the null hypothesis, the test statistic t is given by
X  µ0
t
S/ n
167  160 7
   1.96
16 / 20 3.58
The critical value of the test statistic t for various df and different level of
significance  are given in Table II of the Appendix at the end of the Block
1of this course.
The critical (tabulated) values of test statistic for two-tailed test corresponding
(n-1) = 19 df at 5% level of significance are  t ( n 1), α / 2   t (19 ), 0.025   2.093.

Since calculated value of test statistic t (= 1.96) is greater than the critical
value (= − 2.093) and is less than critical value (= 2.093), that means calculated
value of test statistic lies in non-rejection region as shown in Fig. 11.4. So we
do not reject the null hypothesis i.e. we support the manufacture’s claim at 5%
level of significance.
Fig. 11.4
Decision according to p-value:
Since calculated value of test statistic t is based on 19 df therefore, we use row
for 19 df in the t-table and move across this row to find the values in which
calculated t-value falls. Since calculated t-value falls between 1.729 and 2.093
corresponding to one-tailed area α = 0.05 and 0.025 respectively therefore p-
value lies between 0.025 and 0.05, that is,
0.025  p-value  0.05
Since test is two-tailed so
2  0.025  0.05  p-value  0.10  0.05  2
Since p-value is greater than  (  0.05) so we do not reject the null
hypothesis at 5% level of significance.
Thus, we conclude that sample fails to provide us sufficient evidence against
the null hypothesis so we may assume that the manufacture’s claim is true so
the investigator may accept the manufacturer’s claim at 5% level of
significance.
Example 2: The mean share price of companies of Pharma sector is Rs.70. The
share prices of all companies were changed time to time. After a month, a
sample of 10 Pharma companies was taken and their share prices were noted as
below:
70, 76, 75, 69, 70, 72, 68, 65, 75, 72
Assuming that the distribution of share prices follows normal distribution, test
whether mean share price is still the same at 1% level of significance?
Testing of Hypothesis Solution: Here, we wish to test that the mean share price (µ) of companies of
Pharma sector is still Rs.70 besides all changes. So our claim is µ = 70 and its
complement is µ ≠ 70. Since the claim contains the equality sign so we can
take the claim as the null hypothesis and complement as the alternative
hypothesis. Thus,
H 0 :    0  70 mean share price of companies is still Rs. 70 
H1 :    0  70  mean share price of companies is not still Rs. 70
Since the alternative hypothesis is two-tailed so the test is two-tailed test.
Here, we want to test the hypothesis regarding population mean when
population SD is unknown. Also sample size is small n = 10(n < 30) and
population under study is normal, so we can go for t-test for testing the
hypothesis about population mean.
For testing the null hypothesis, the test statistic t is given by
X  µ0
t ~ t ( n 1) … (1)
S/ n
Calculation for X and S:
S. No. Sample value Deviation d2
(X) d = (X-a), a = 70
1 70 0 0
2 76 6 36
3 75 5 25
4 69 -1 1
5 70 0 0
6 72 2 4
7 68 -2 4
8 65 -5 25
9 75 5 25
10 72 2 4
Total 12 124

The assumed value of a is 70.

From the above calculation, we have
1 1
Xa
n  d  70   12  71.2
10

2
S 
1  2
d 
 d 
2


n 1  n 
 

1   12 2  1  144 
  124    124   12.18
10  1  10  9  10 

S 12.18  3.49

Putting the values in equation (1), we have
71.2  70 1.2
t   1.09
3.49 / 10 1.10
70
The critical (tabulated) values of test statistic for two-tailed test corresponding Small Sample Tests
(n-1) = 9 df at 1% level of significance are  t ( n 1), α / 2   t ( 9 ), 0.005   3.250.

Since calculated value of test statistic t (= 1.09) is less than the critical value
(= 3.250) and greater than the critical value (= ‒3.250), that means calculated
value of t lies in non-rejection region as shown in Fig. 11.5. So we do not
reject the null hypothesis i.e. we support the claim at 1% level of significance.
Decision according to p-value:
Since calculated value of test statistic t is based on 9 df therefore, we use row
for 9 df in the t-table and move across this row to find the values in which Fig. 11.5
calculated t-value falls. Since all values in this row are greater than calculated
t-value 1.09 and the smallest value is 1.383 corresponding to one-tailed area
α = 0.10 therefore p-value is greater than 0.10, that is,
p-value  0.10
Since test is two-tailed so
p-value  2  0.10  0.20
Since p-value (= 0.20) is greater than (  0.01) so we do not reject the null
hypothesis at 1% level of significance.
Thus, we conclude that the sample fails to provide us sufficient evidence
against the claim so may assume that the mean share price is still Rs. 70.
Now, you can try the following exercises.
E2) A tyre manufacturer claims that the average life of a particular category
of his tyre is 18000 km when used under normal driving conditions. A
random sample of 16 tyres was tested. The mean and SD of life of the
tyres in the sample were 20000 km and 6000 km respectively.
Assuming that the life of the tyres is normally distributed, test the claim
of the manufacturer at 1% level of significance using appropriate test.
E3) It is known that the average weight of cadets of a centre follows normal
distribution. Weights of 10 randomly selected cadets from the same
centre are as given below:
48, 50, 62, 75, 80, 60, 70, 56, 52, 77
Can we say that average weight of all cadets of the centre from which
the above sample was taken is equal to 60 kg at 5% level of
significance?

11.4 TESTING OF HYPOTHESIS FOR

DIFFERENCE OF TWO POPULATION MEANS
USING t-TEST
In Section 10.4 of the previous unit, we have discussed Z-test for testing the
hypothesis about difference of two population means under different possibility
of population variances 12 and 22 . Recall from there, we have pointed out that
one basic difference between Z-test and t-test is that, Z-test is used when
standard deviations of both populations are known and t-test is used when
standard deviations of both populations are unknown. But in practice standard
Testing of Hypothesis deviations of both populations are not known, so in real life problems t-test is
more suitable compared to Z-test.
Assumptions
This test works under following assumptions:
(i) The characteristic under study follows normal distribution in both the
populations. In other words, both populations from which random
samples are drawn should be normal with respect to the characteristic of
interest.
(ii) Samples and their observations both are independent to each other.
(iii) Population variances 12 and  22 are both unknown but equal.
For describing this test, let there be two normal populations N 1, 12 and  
N 2 , 22  under study. And we have to draw two independent random samples,
say, X1,X2 ,...,Xn1 and Y1, Y2 ,..., Yn2 of sizes n1 and n2 from these normal
   
populations N 1, 12 and N 2 , 22 respectively. Let X and Y be the means of
first and second sample respectively. Further, suppose the variances of both the
populations are unknown but are equal, i.e., 12  22  2 (say). In this case, σ2
is estimated by value of pooled sample variance S 2p where,
1
S2p   n1  1 S12   n 2  1 S22 
n1  n 2  2

and
n n
1 1
2 1 2
2
S12  
(n1  1) i 1
 X i  X  , S 2
2  
(n 2  1) i1
 Yi  Y 

This can also be written as

1  n1 2
n2
2
  Xi  X     Yi  Y  
2
S 
p
n1  n 2  2  i 1 i 1 
For computational simplicity, use the following formulae for X, Y and S 2p :

1 1
Xa
n1  d1 , Y  b 
n2 d 2 and
2 2
1     d1      d 2   
n1  n 2  2    1     d2 
2 2 2
S 
p  d  
n1 n2
    
where, d1= (X − a ) and d2 = (Y − b), ‘a’ and ‘b’ are the assumed arbitrary
values.
Now, follow the same procedure as we have discussed in Section 11.2, that is,
first of all we have to setup null and alternative hypotheses. Here, we want to
test the hypothesis about the difference of two population means so we can take
the null hypothesis as
 Here, 1  1 and 2   2 
H 0 : 1  2 (no difference in means)  if we compare it with 
 general procedure. 

or H 0 : 1  2  0 (difference in two means is 0)

72
and the alternative hypothesis as Small Sample Tests

H1 : 1  2 for two-tailed test 

H0 : 1  2 and H1 : 1  2 
or  for one-tailed test 
H0 : 1  2 and H1 : 1  2 
For testing the null hypothesis, the test statistic t is given by
XY
t ~ t (n 1 n 2 2) under H0
1 1
Sp 
n1 n 2

After substituting values of X, Y, S p , n1 and n 2 , we get calculated value of test

statistic t. Then we look for critical (or cut-off or tabulated) value(s) of test
statistic t from the t-table. On comparing calculated value and critical value(s),
we take the decision about the null hypothesis as discussed in Section 11.2.
Let us do some examples to become more user friendly with the test explained
above.
Example 3: In a random sample of 10 pigs fed by diet A, the gain in weights
(in pounds) in a certain period were
12, 8, 14, 16, 13, 12, 8, 14, 10, 9
In another random sample of 10 pigs fed by diet B, the gain in weights (in
pounds) in the same period were
14, 13, 12, 15, 16, 14, 18, 17, 21, 15
Assuming that gain in the weights due to both foods follows normal
distributions with equal variances, test whether diets A and B differ
significantly regarding their effect on increase in weight at 5% level of
significance.
Solution: Here, we can test that diets A and B differ significantly regarding
their effect on increase in weight of pigs. If 1 and 2 denote the average gain
in weights due to diet A and diet B respectively then our claim is 1 ≠ 2 and
its complement is 1 = 2. Since complement contains the equality sign so we
can take the complement as the null hypothesis and the claim as the alternative
hypothesis. Thus,
H 0 : 1   2 and H 1 : 1   2

Since the alternative hypothesis is two-tailed so the test is two-tailed test.

Since it is given that the increase in the weight due to both foods follows
normal distributions and population variances are equal and unknown. And
other assumptions of t-test for testing a hypothesis about difference of two
population means also meet. So we can go for this test.
For testing the null hypothesis, the test statistic t is given by
XY
t ~ t (n 1n 22) under H0 … (2)
1 1
Sp 
n1 n 2

73
Testing of Hypothesis Calculation for X, Y and S p :
Diet A Diet B
X d1 = (X−a) d1 2 Y d2 = (Y−b) d2 2
a =12 b = 16
12 0 0 14 −2 4
8 −4 16 13 −3 9
14 2 4 12 −4 16
16 4 16 15 −1 1
13 1 1 16 0 0
12 0 0 14 −2 4
8 −4 16 18 2 4
14 2 4 17 1 1
10 −2 4 21 5 25
9 −3 9 15 −1 1
Total −4 70 −5 65

Here, a =12, b = 16 are assumed values.

From above calculations, we have
1  4 
Xa
n1  d1  12 
10
 11.6,

1  5 
Y  b   d 2  16   15.5
n2 10
2 2
1     d1      d2   
n1  n 2  2    1     d2 
2 2 2
S 
p  d  
n1 n2
    


1 
70 
 42   65   52 
   
10  10  2  10   10 

1
  68.4  62.5   7.27
18
 Sp  7.27  2.70

Putting the values in equation (2), we have

11.6  15.5
t
1 1
2.70 
10 10
 3.90 3.90
   3.21
2 .70  0 .45 1.215
The critical values of test statistic t for two-tailed test corresponding
(n1 + n2 -2) = 18 df at 5% level of significance are
 t ( n 1  n 2  2 ), α / 2   t (18 ), 0.025   2.101.

Since calculated value of test statistic t (= −3.21) is less than critical values
(± 2.101) that means calculated value of test statistic t lies in rejection region,
so we reject the null hypothesis and support the alternative hypothesis i.e.
support our claim at 5% level of significance.
74
Thus, we conclude that samples do not provide us sufficient evidence against Small Sample Tests
the claim so diets A and B differ significantly in terms of gain in weights of
pigs.
Example 4: The means of two random samples of sizes 10 and 8 drawn from
two normal populations are 210.40 and 208.92 respectively. The sum of
squares of the deviations from their means is 26.94 and 24.50 respectively.
Assuming that the populations are normal with equal variances, can samples be
considered to have been drawn from normal populations having equal mean.
Solution: In usual notations, we are given that
n1  10, n 2  8, X  210.40, Y  208.92,
2 2
 X  X   26.94,  Y  Y   24.50

Therefore,
1
S 2p     X  X  2    Y  Y 2 
n1  n 2  2  

1
 26.94  24.50  1  51.44  3.215
10  8  2 16
 Sp  3.215  1.79
We wish to test that both the samples are drawn from normal populations
having the same means. If 1 and 2 denote the means of both normal
populations respectively then our claim is 1= 2 and its complement is
1 ≠ 2. Since the claim contains the equality sign so we can take the claim as
the null hypothesis and complement as the alternative hypothesis. Thus,
H 0 : 1   2 [mean of both populations is equal]
H1 : 1   2 [mean of both populations is not equal]
Since the alternative hypothesis is two-tailed so the test is two-tailed test.
Since it is given that two populations are normal with equal and unknown
variances and other assumptions of t-test for testing a hypothesis about
difference of two population means also meet. So we can go for this test.
For testing the null hypothesis, the test statistic t is given by
XY
t ~ t (n 1 n 2 2) under H0
1 1
Sp 
n1 n 2
210.40  208.92 1.48 1.48
    1.76
1 1 1.79  0.47 0.84
1.79 
10 8
The critical values of test statistic t for two-tailed test corresponding
(n1 + n2 -2) = 16 df at 5% level of significance are
 t ( n1  n 2  2 ), α / 2   t (16 ), 0.025   2.12.

Since calculated value of test statistic t (= 1.76) is less than the critical value
(= 2.12) and greater than the critical value (= ‒2.12), that means calculated
value of test statistic t lies in non-rejection region so we do not reject the null
hypothesis i.e. we support the claim.
75
Testing of Hypothesis Thus, we conclude that samples fail to provide us sufficient evidence against
the claim so we may assume that both samples are taken from normal
populations having equal means.
Now, you can try the following exercises.
E4) Two different types of drugs A and B were tried on some patients for
increasing their weights. Six persons were given drug A and other 7
persons were given drug B. The gain in weights (in ponds) is given
below:

Drug A 5 8 7 10 9 6 −

Drug B 9 10 15 12 14 8 12

Assuming that increment in the weights due to both drugs follows

normal distributions with equal variances, do the both drugs differ
significantly with regard to their mean weights increment at 5% level of
significance?
E5) To test the effect of fertilizer on wheat production, 26 plots of land with
equal areas were chosen. Half of these plots were treated with fertilizer
and the other half were untreated. Other conditions were the same. The
mean yield of wheat on the untreated plots was 4.6 quintals with a
standard deviation of 0.5 quintals, while the mean yield of the treated
plots was 5.0 quintals with standard deviations of 0.3 quintals.
Assuming that yields of wheat with and without fertilizer follow normal
distributions with equal variances, can we conclude that there is
significant improvement in wheat production due to effect of fertilizer
at 1% level of significance?

11.5 PAIRED t-TEST

In the previous section, we have discussed t-test for equality of two population
means in case of independent samples. However, there are so many situations
where two samples are not independent and observations are recorded on the
same individuals or items. Generally, such types of observations are recorded
to assess the effectiveness of a particular training, diet, treatment, medicine,
etc. In such situations, the observations are recorded “before and after” the
insertion of training, treatment, etc. as the case may be. For example, if we
wish to test a new diet on, say, 15 individuals then the weight of the individuals
recorded before diet and after the diet will form two different samples in which
observations will be paired as per each individual. Similarly, in the test of
blood-sugar in human body, fasting sugar level before meal and sugar level
after meal, both are recorded for a patient as paired observations, etc. The
parametric test designed for this type of situation is known as paired t-test.
Now, come to the working principle of this test. This test first of all converts
the two populations into a single population by taking the difference of paired
observations.
Now, instead of two populations, we are left with one population, the
population of differences. And the problem of testing equality of two
population mean reduces to test the hypothesis that mean of the population of
differences is equal to zero.

76
Assumptions Small Sample Tests

This test works under following assumptions:

(i) The population of differences follows normal distribution.
(ii) Samples are not independent.
(iii) Size of both the samples is equal.
(iv) Population variances are unknown but not necessarily equal.
Let (X1, Y1), (X2, Y2), …,(Xn, Yn) be a paired random sample of size n and the
difference between paired observations Xi & Yi be denoted by Di, that is,
D i  X i  Yi for all i  1, 2,..., n
Hence, we can assume that D1, D2, …, Dn be a random sample from normal
population of differences with mean D and unknown variance  2D . This is
same as the case of testing of hypothesis for population mean when population
variance is unknown which is described in Section 11.3 of this unit.
Here, we want to test that there is an effect of a diet, training, treatment,
medicine, etc. So we can take the null hypothesis as
H0: µ1 = µ2 or H0 : µD  µ1  µ2  0
and the alternative hypothesis
H1 : 1   2 or H1 :  D  0 for two-tailed test 
H0: 1  2 and H1 : 1  2 
or  for one-tailed test 
H0: 1  2 and H1 : 1  2 
For testing the null hypothesis, the test statistic t is given by
D
t ~ t  n 1  under H 0
SD / n
2
  n  
1 n 1 n 2
 D 
1  n 2  i1  
n i1
2
where, D   Di and SD    Di  D   n  1  
n  1 i1 i 1
D 
n 
 
 
After substituting values of D, S D and n we get calculated value of test
statistic t. Then we look for critical (or cut-off or tabulated) value(s) of test
statistic t from the t-table. On comparing calculated value and critical value(s),
we take the decision about the null hypothesis as discussed in Section 11.2.
Let us do some examples to become more user friendly with paired t-test.
Example 5: A group of 12 children was tested to find out how many digits
they would repeat from memory after hearing them once. They were given
practice session for this test. Next week they were retested. The results
obtained were as follows:
Child Number 1 2 3 4 5 6 7 8 9 10 11 12
Recall Before 6 4 5 7 6 4 3 7 8 4 6 5
Recall After 6 6 4 7 6 5 5 9 9 7 8 7

77
Testing of Hypothesis Assuming that the memories of the children before and after the practice
session follow normal distributions, is the memory practice session improve the
performance of children?
Solution: Here, we want to test that memory practice session improve the
performance of children. If 1 and 2 denote the mean digit repetition before
and after the practice so our claim is 1 < 2 and its complement is 1 ≥ 2.
Since complement contains the equality sign so we can take the complement as
the null hypothesis and the claim as the alternative hypothesis. Thus,
H 0 : 1   2 and H1 : 1   2

Since the alternative hypothesis is left-tailed so the test is left-tailed test.

It is a situation of before and after. Also, it is given that the memories of the
children before and after the practice session follow normal distributions. So,
population of differences will also be normal. Also all the assumptions of
paired t-test meet so we can go for paired t-test.
For testing the null hypothesis, the test statistic t is given by
D
t ~ t  n 1  under H 0 … (3)
SD / n

where, D and S D are mean and standard deviation of the population of

differences.
Calculation for D and S D :
Child Digit recall D = (X−Y) D2
Number Before (X) After (Y)
1 6 6 0 0
2 4 6 −2 4
3 5 4 1 1
4 7 7 0 0
5 6 6 0 0
6 4 5 −1 1
7 3 5 −2 4
8 7 9 −2 4
9 8 9 −1 1
10 4 7 −3 9
11 6 8 −2 4
12 5 7 −2 4
   14
D   32
D 2

From above calculations, we have

1 1
D
n  D   14    1.17
12
2
1    D  
n  1  
2 2
S 
D D  
n
 

1  14 2  1
  32    15.67  1.42
11  12  11

 SD  1.42  1.19

78
Substituting these values in equation (3), we have Small Sample Tests

 1.17 1.17
t   3.44
1.19 / 12 0.34
The critical value of test statistic t for left-tailed test corresponding (n-1) = 11
df at 5% level of significance is  t ( n 1), α   t (11), 0.05   1.796.

Since calculated value of test statistic t (= −3.44) is less than the critical value
(=−1.796), that means calculated value of t lies in rejection region, so we reject
the null hypothesis and support the alternative hypothesis i.e. support the claim
at 5% level of significance.
Thus, we conclude that samples fail to provide us sufficient evidence against
the claim so we may assume that memory practice session improves the
performance of children.
Example 6: Ten students were given a test in Statistics and after one month’s
coaching they were again given a test of the similar nature and the increase in
their marks in the second test over the first are shown below:
Roll No. 1 2 3 4 5 6 7 8 9 10
Increase in Marks 6 −2 8 −4 10 2 5 −4 6 0

Assuming that increment in marks follows normal distribution. Do the data

indicate that students have gained knowledge from the coaching at 1% level of
significance?
Solution: Here, we want to test that students have gained knowledge from the
coaching. If D denotes the average increment in the marks due to one month’s
coaching then our claim is D < 0 but here we are given increment
Di = (Yi – Xi) instead of Di = (Xi – Yi) so we take our claim is D > 0 and its
complement is D ≤ 0. Since complement contains the equality sign so we can
take the complement as the null hypothesis and the claim as the alternative
hypothesis. Thus,
H 0 :  D  0 and H1 :  D  0

Since the alternative hypothesis is right-tailed so the test is right-tailed test.

It is given that increment in marks after one month coaching follows normal
distribution and population variance is unknown. Also participants are same in
both situations before and after the coaching. And all the assumption of paired
t-test meet so we can go for paired t-test.
For testing H0, the test statistic is given by
D
t ~ t  n 1  under H 0 … (4)
SD / n
Calculation for D and S D :
Roll
1 2 3 4 5 6 7 8 9 10 Total
No.
D 6 −2 8 −4 10 2 5 −4 6 0  D  27
2 2
D 36 4 64 16 100 4 25 16 36 0 D  301

From above calculations, we have

1 1
D
n  D   27  2.7
10
79
Testing of Hypothesis 2
1    D  
n  1  
2 2
S 
D D  
n
 

1   27 2  1
 301    228.1  25.34
10  1  10  9

 SD  25.34  5.03
Substituting these values in equation (4), we have
2.7 2.7
t   1.70
5.03/ 10 1.59
The critical value of test statistic t for right-tailed test corresponding (n-1) = 9
df at 1% level of significance is t (n 1), α  t (9), 0.01  2.821.
Since calculated value of test statistic t (= 1.70) is less than the critical value
(= 2.821), that means calculated value of test statistic t lies in non-rejection
region, so we do not reject null hypothesis and reject the alternative hypothesis
i.e. we reject our claim at 1% level of significance.
Thus, we conclude that sample provides us sufficient evidence against the
claim so students are not gained knowledge from the coaching.
Now, you can try the following exercises.
E6) To verify whether the programme “Post Graduate Diploma in Applied
Statistics (PGDAST)” improved performance of the graduate students
in Statistics, a similar test was given to10 participants both before and
after the programme. The original marks out of 100 (before course)
recorded in an alphabetical order of the participants are 42, 46, 50, 36,
44, 60, 62, 43, 70 and 53. After the course the marks in the same order
are 45, 46, 60, 42, 60, 72, 63, 43, 80 and 65. Assuming that marks of
the students before and after the course follow normal distribution. Test
whether the programme PGDAST has improved the performance of the
graduate students in Statistics at 5% level of significance?
E7) A drug is given to 8 patients and the increments in their blood pressure
are recorded to be 4, 0, 7, −2, 0, −3, 2, 0. Assume that increment in their
blood pressure follows normal distribution. Is it reasonable to believe
that the drug has no effect on the change of blood pressure at 5% level
of significance?

11.6 TESTING OF HYPOTHESIS FOR

POPULATION CORRELATION COEFFICIENT
USING t-TEST
In Unit 6 of MST-002, we have discussed the concept of correlation. Where,
we studied that if two variables are related in such a way that change in the
value of one variable affects the value of another variable then the variables are
said to be correlated or there is a correlation between these two variables.
Correlation can be positive, which means the variables move together in the
same direction, or negative, which means they move in opposite directions.
And correlation coefficient is used to measure the intensity or degree of linear
relationship between two variables. The value of correlation coefficient varies

80
between −1 and +1, where −1 representing a perfect negative correlation, 0 Small Sample Tests
representing no correlation, and +1 representing a perfect positive correlation.
Sometime, the sample data indicate for non-zero correlation but in population
they are uncorrelated ( = 0).
For example, price of tomato in Delhi (X) and in London (Y) are not correlated
in population ( = 0). But paired sample data of 20 days of prices of tomato at
both places may show correlation coefficient (r) ≠ 0. In general, in sample data
r ≠ 0 does not ensure in population  ≠ 0 holds.
In this section, we will know how we test the hypothesis that population
correlation coefficient is zero.
Assumptions
This test works under following assumptions:
(i) The characteristic under study follows normal distribution in both the
populations. In other words, both populations from which random
samples are drawn should be normal with respect to the characteristic of
interest.
(ii) Samples observations are random.
Let us consider a random sample (X1, Y1), (X2, Y2), …, (Xn, Yn) of size n taken
from a bivariate normal population. Let  and r be the correlation coefficients
of population and sample data respectively.
Here, we wish to test the hypothesis about population correlation coefficient
(), that is, linear correlation between two variables X and Y in the population,
so we can take the null hypothesis as
H0 :   0 and H1 :   0 for two-tailed test   Here,    and 0  0 if 
we compare it with general 
procedure given inSection11.2.

H 0 :   0 and H1 :   0 
or   for one-tailed test 
H 0 :   0 and H1 :   0 
For testing the null hypothesis, the test statistic t is given by
r n2
t ~ t  n 2 
1 r2
which follows t-distribution with n – 2 degrees of freedom.
After substituting values of r and n, we find out calculated value of test statistic
t. Then we look for critical (or cut-off or tabulated) value(s) of test statistic t
from the t-table. On comparing calculated value and critical value(s), we take
the decision about the null hypothesis as discussed in Section 11.2.
Let us do some examples of testing of hypothesis that population correlation
coefficient is zero.
Example 7: A random sample of 18 pairs of observations from a normal
population gave a correlation coefficient of 0.7. Test whether the population
correlation coefficient is zero at 5% level of significance.
Solution: Given that
n = 18, r = 0.7
Here, we wish to test that population correlation coefficient () is zero so our
claim is  = 0 and its complement is  ≠ 0. Since the claim contains the
81
Testing of Hypothesis equality sign so we can take the claim as the null hypothesis and complement
as the alternative hypothesis. Thus,
H 0 :   0 and H1 :   0

Since the alternative hypothesis is two-tailed so the test is two-tailed test.

Here, we want to test the hypothesis regarding population correlation
coefficient is zero and the populations under study follow normal distributions,
so we can go for t-test.
For testing the null hypothesis, the test statistic t is given by
r n2
t
1  r2
0.7 18  2 0.7  4 2.8
    3.94
1   0.7 
2
0.51 0.71

The critical value of test statistic t for two-tailed test corresponding (n-2) = 16
df at 5% level of significance are  t ( n  2 ), α / 2   t (16 ), 0.025   2.120.

Since calculated value of test statistic t (= 3.94) is greater than the critical
values (= ± 2.120), that means calculated value of test statistic t lies in rejection
region, so we reject the null hypothesis. i.e. we reject the claim at 5% level of
significance.
Thus, we conclude that sample provides us sufficient evidence against the
claim so there exists a relationship between two variables.
Example 8: A random sample of 15 married couples was taken from a
population consisting of married couples between the ages of 30 and 40. The
correlation coefficient between the IQs of husbands and wives was found to be
0.68. Assuming that the IQs of husbands and wives follow normal distributions
then test that IQs of husbands and wives in the population are positively
correlated at1% level of significance.
Solution: Given that
n = 15, r = 0.68
Here, we wish to test that IQs of husbands and wives in the population are
positively correlated. If  denote the correlation coefficient between IQs of
husbands and wives in the population then the claim is  > 0 and its
complement is  ≤ 0. Since complement contains the equality sign so we can
take the complement as the null hypothesis and the claim as the alternative
hypothesis. Thus,
H 0 :   0 and H1 :   0
Since the alternative hypothesis is right-tailed so the test is right-tailed test.
Here, we want to test the hypothesis regarding population correlation
coefficient is zero and the populations under study follow normal distributions,
so we can go for t-test.
For testing the null hypothesis, the test statistic t is given by
r n2
t
1  r2

82
 0.68 15  2 0.68  3.61 Small Sample Tests
   3.36
1   0.68 
2 0.73

The critical value of test statistic t for right-tailed test corresponding (n-2) = 13
df at 1% level of significance is t ( n  2 ), α  t (13 ), 0.01  2.650.

Since calculated value of test statistic t (= 3.36) is greater than the critical value
(= 2.650), that means calculated value of test statistic t lies in rejection region,
so we reject the null hypothesis and support the alternative hypothesis i.e. we
support our claim at 1% level of significance.
Thus, we conclude that sample fail to provide us sufficient evidence against the
claim so we may assume that the correlation between IQs of husbands and
wives in the population is positive.
In the same way, you can try the following exercise.
E8) Twenty families were selected randomly from a colony to determine
that correlation exists between family income and the amount of money
spent per family member on food each month. The sample correlation
coefficient was computed as r = 0.40. Assuming that the family income
and the amount of money spent per family member on food each month
follow normal distributions then test that there is a positive linear
relationship between the family income and the amounts of money
spent per family member on food each month in colony at 1% level of
significance.
We now end this unit by giving a summary of what we have covered in it.

11.7 SUMMARY
In this unit, we have discussed the following points:
1. Need of small sample tests.
2. Procedure of testing a hypothesis for t-test.
3. Testing of hypothesis for population mean using t-test.
4. Testing of hypothesis for difference of two population means when samples
are independent using t-test.
5. The procedure of paired t-test for testing of hypothesis for difference of two
population means when samples are dependent or paired.
6. Testing of hypothesis for population correlation coefficient using t-test.

11.8 SOLUTIONS /ANSWERS

E1) Since calculated value of test statistic t is based on 15 df therefore, we
use row for 15 df in the t-table and move across this row to find the
values in which calculated t-value lies. Since calculated t-value falls
between 2.131 and 2.602, which are corresponding to the values of
one-tailed area α = 0.025 and 0.01 respectively, therefore p-value will
lie between 0.01 and 0.025, that is, 0.01  p-value  0.025
Since test is two-tailed, therefore, the values are doubled, so
0.02  2  0.01  p-value  2  0.025  0.05
E2) Here, we are given that
83
Testing of Hypothesis n  16,  0  18000, X  20000, S  6000
Here, we want to test that manufacturer’s claim is true that the average
life () of tyres is 18000 km. So claim is µ = 18000 and its complement
is µ ≠ 18000. Since the claim contains the equality sign so we can take
the claim as the null hypothesis and complement as the alternative
hypothesis. Thus,
H0 :  = µ0 = 18000 [average life of tyres is 18000 km]
H1 :  ≠ 18000 [average life of tyres is not 18000 km]
Here, population SD is unknown and population under study is given to
be normal. So we can go for t-test.
For testing the null hypothesis, the test statistic t is given by
X  µ 0 20000  18000 2000
t    1.33
S/ n 6000 / 16 1500
The critical value of test statistic t for two-tailed test corresponding
(n-1) = 15 df at 1% level of significance are ± t(15) ,0.005 = ± 2.947.
Since calculated value of test statistic t (= 1.33) is less than the critical
(tabulated) value (= 2.947) and greater that critical value (= − 2.947),
that means calculated value of test statistic lies in non-rejection region,
so we do not reject the null hypothesis i.e. we support the
manufacture’s claim at 1% level of significance.
Thus, we conclude that sample fails to provide sufficient evidence
against the claim so we may assume that manufacturer’s claim is true.
E3) Here, we want to test that the average weight () of all cadets of the
centre is 60 kg. So our claim is µ = 60 and its complement is µ ≠ 60.
Since the claim contains the equality sign so we can take the claim as
the null hypothesis and complement as the alternative hypothesis. Thus,
H 0 :    0  60  average weight of all cadets is 60 kg 
H1 :   60 average weight of all cadets is not 60 kg 
Since the alternative hypothesis is two-tailed so the test is two-tailed
test.
Here, population SD is unknown and population under study is given to
be normal. So we can go for t-test.
For testing the null hypothesis, the test statistic t is given by
X  µ0
t ~ t ( n 1) under H 0 … (5)
S/ n
Before moving further, first we have to calculate value of X and S.
Calculation for X and S:
Sample value (X) X  X  2
X  X
48 −15 225
50 −13 169
62 −1 1
75 12 144
80 17 289
60 −3 9
70 7 49

84
 56 −7 49 Small Sample Tests
52 −11 121
77 14 196
 X  630  X  X
2
1252

From the above calculation, we have

1 1
X
n
 X  10  630  63
1 2
S2 
n 1
  X  X

1
  1252  139.11
10  1
 S  139.11  11.79
Putting the values in equation (5), we have
63  60
t
11.79 / 10
3
  0.80
3.73
The critical values of test statistic t for two-tailed test corresponding
(n-1) = 9 df at 5% level of significance are ± t(9) ,0.025 = ± 2.262.
Since calculated value of test statistic t (= 0.80) is less than the critical
value (= 2.262) and greater than the critical value (= −2.262), that
means calculated value of test statistic t lies in non-rejection region so
we do not reject H0 i.e. we support the claim at 5% level of
significance.
Thus, we conclude that sample fails to provide sufficient evidence
against the claim so we may assume that the average weight of all the
cadets of given centre is 60 kg.
E4) Here, we want to test that there is no difference between drugs A and B
with regard to their mean weight increment. If 1 and 2 denote the
mean weight increment due to drug A and drug B respectively then our
claim is µ1 = 2 and its complement is µ1 ≠ 2. Since the claim contains
the equality sign so we can take the claim as the null hypothesis and
complement as the alternative hypothesis. Thus,
H 0 : 1   2 effect of both drugs is same 
H 1 : 1   2 effect of both drugs is not same 
Since the alternative hypothesis is two-tailed so the test is two-tailed
test.
Since it is given that increments in the weight due to both drugs follow
normal distributions with equal and unknown variances and other
assumptions of t-test for testing a hypothesis about difference of two
population means also meet. So we can go for this test.
For testing the null hypothesis, the test statistic t is given by

85
Testing of Hypothesis XY
t ~ t ( n 1 n 22) under H0 … (6)
1 1
Sp 
n1 n 2

Assume, a = 8, b = 12 and use short-cut method to find X, Y and Sp .

Calculation for X, Y and S p :

Drug A Drug B

X d1 = (X-a) d1 2 Y d2 = (Y-b) d2 2
a=8 b = 12
5 ‒3 9 9 ‒3 9
8 0 0 10 ‒2 4
7 ‒1 1 15 3 9
10 2 4 12 0 0
9 1 1 14 2 4
6 ‒2 4 8 ‒4 16
12 0 0

d 1  3 d 2
1  19 d 2  4  d22  42
From above calculation, we have
1 1
n1  1
Xa d  8   3   7.5,
6
1 1
Y  b   d 2  12   4   11.43
n2 7
2 2
1     d1      d2   
n1  n 2  2    1     d2 
2 2 2
S 
p  d  
n1   n2  

2 2
1    3    4   
  19    42  
6  7  2  6   7 
   
1
 17.5  39.71  5.20
11
 Sp  5.20  2.28
Putting these values in equation (6), we have
7.5  11.43 3.93 3.93
t    3.07
1 1 2.28  0.56 1.28
2.28 
6 7
The critical values of test statistic t for two-tailed test corresponding
(n1 + n2 -2) = 11 df at 5% level of significance are ± t(11) ,0.025 = ± 2.201.
Since calculated value of test statistic t (= −3.07) is less than the critical
values (= ± 2.201) that means calculated value of test statistic t lies in
rejection region, so we reject the null hypothesis i.e. we reject the claim
at 5% level of significance.

86
 Thus, we conclude that samples provide us sufficient evidence against Small Sample Tests
the claim so drugs A and B differ significantly. Any one of them is
better than other.
E5) Here, we are given that
n1  13, X  4.6, S1  0.5,

n 2  13, Y  5.0, S2  0.3

Therefore, the pooled variance S 2p can be calculated as

1
S2p   n1  1 S12   n 2  1 S22 
n1  n 2  2

1 12   0.5  2  12   0.3 2 


13  13  2  
1
  3.00  1.08  0.17
24
 Sp  0.17  0.41

We want to test that the there is significant improvement in wheat

production due to fertilizer. If 1 and 2 denote the mean wheat
productions without and with the fertilizer respectively then our claim
is 1 < 2 and its complement is 1 ≥ 2. Since complement contains the
equality sign so we can take the complement as the null hypothesis and
the claim as the alternative hypothesis. Thus,
H0 : 1  2 and H1 : 1  2
Since the alternative hypothesis is left-tailed so the test is left-tailed
test.
Since it is given that yield of wheat with and without fertilizer follow
normal distributions with equal and unknown variances and other
assumptions of t-test for testing a hypothesis about difference of two
population means also meet. So we can go for this test.
For testing the null hypothesis, the test statistic t is given by
XY
t ~ t (n 1n 2 2) under H0
1 1
Sp 
n1 n 2
4.6  5.0  0.4  0.4
     2.5
1 1 0.41  0.39 0.16
0.41 
13 13
The critical values of test statistic t for left-tailed test corresponding
(n1 + n2 -2) = 24 df at 1% level of significance is − t(24), 0.01 = −2.492.
Since calculated value of test statistic t (= −2.5) is less than the critical
value (= −2.492), that means calculated value of test statistic t lies in
rejection region, so we reject the null hypothesis and support the
alternative hypothesis i.e. we support our claim at 1% level of
significance.

87
Testing of Hypothesis Thus, we conclude that samples fail to provide us sufficient evidence
against the claim so we may assume that there is significant
improvement in wheat production due to fertilizer.
E6) Here, we want to test whether the programme PGDAST has improved
the performance of the graduate students in Statistics. If 1 and 2
denote the average marks before and after the programmed so our claim
is 1 < 2 and its complement is 1 ≥ 2. Since complement contains the
equality sign so we can take the complement as the null hypothesis and
the claim as the alternative hypothesis. Thus,
H0 : 1  2 and H1 : 1  2
Since the alternative hypothesis is left-tailed so the test is left-tailed
test.
It is a situation of before and after. Also, the marks of the students
before and after the programme PGDAST follow normal distributions.
So, population of differences will also be normal. Also all the
assumptions of paired t-test meet. So we can go for paired t-test.
For testing the null hypothesis, the test statistic t is given by
D
t ~ t  n 1  under H 0 … (7)
SD / n
Calculation for D and S D :

Participant Marks D = (X−Y) D2

Before (X) After (Y)

1 42 45 −3 9
2 46 46 0 0
3 50 60 −10 100
4 36 42 −6 36
5 44 60 −16 256
6 60 72 −12 144
7 62 63 −1 1
8 43 43 0 0
9 70 80 −10 100
10 53 65 −12 144
2
 D  70 D  790

From above calculation, we have

1 1
n
D D   70    7
10
2
1    D 
n 1 
2 2
S 
D  D  
n 



1
790 
 702  1
    300  33 . 33
9 10  9

 SD  33.33  5.77

88
 Putting the values in equation (7), we have Small Sample Tests

7.0
t  3.83
5.77 / 10
The critical value of test statistic t for left-tailed test corresponding
(n-1) = 9 df at 5% level of significance is − t(9), 0.05 = −1.833.
Since calculated value of test statistic t (= −3.83) is less than the critical
(tabulated) value (= −1.833), that means calculated value of test statistic
t lies in rejection region, so we reject the null hypothesis and support
the alternative hypothesis i.e. we support our claim at 5% level of
significance.
Thus, we conclude that samples fail to provide us sufficient evidence
against the claim so we may assume that the participants have
significant improvement after the programme “Post Graduate Diploma
in Applied Statistics (PGDAST)”.
E7) Here, we want to test that the drug has no effect on change in blood
pressure. If D denotes the average increment in the blood pressure
before drug then our claim is D = 0 and its complement is D ≠ 0.
Since the claim contains the equality sign so we can take the claim as
the null hypothesis and complement as the alternative hypothesis. Thus,
H 0 :  D  1 2  0  the drug has no effect 

H1: µ D  0  the drug has an effect 

It is given that increment in the blood pressure follows normal
distribution. Also patients are same in both situations before and after
the drug. And all the assumption of paired t-test meet. So we can go for
paired t-test.
For testing H0, the test statistic is given by
D
t ~ t  n 1  under H 0 … (8)
SD / n
Calculation for D and S D :
Patient
1 2 3 4 5 6 7 8 Total
Number
D 4 0 7 −2 0 −3 2 0 D  8
 D  D 3 −1 6 −3 −1 −4 1 −1
2 2
 D  D 9 1 36 9 1 16 1 1  D  D  74

From above calculation, we have

1 1
D
n  D  8 1
8
1 2 1
S2D 
n 1
  D  D    74  10.57
7
 SD  10.57  3.25
Putting the values in test statistic, we have

89
Testing of Hypothesis 1 1
t   0.87
3.25 / 8 1.15
The critical value of test statistic t for two-tailed test corresponding
(n-1) = 7 df at 5% level of significance are ± t(7), 0.025 = ± 2.365.
Since calculated value of test statistic t (= 0.87) is less than the critical
value (= 2.365) and greater than the critical value (=−2.365) that means
calculated value of test statistic t lies in non-rejection region, so we do
not reject the null hypothesis i.e. we support the claim at 5% level of
significance.
Thus, we conclude that samples fail to provide us sufficient evidence
against the claim so we may assume that the drug has no effect on the
change of blood pressure of patients.
E8) We are given that
n = 20, r = 0.40
and we wish to test that there is a positive linear relationship between
the family income and the amounts of money spent per family member
on food each month in colony. If  denote the correlation coefficient
between the family income and the amounts of money spent per family
member then the claim is  > 0 and its complement is  ≤ 0. Since
complement contains the equality sign so we can take the complement
as the null hypothesis and the claim as the alternative hypothesis. Thus,
H0 :   0 and H0 :   0
Since the alternative hypothesis is right-tailed so the test is right-tailed
test.
For testing the null hypothesis, the test statistic t is given by
r n2
t
1  r2
0.40 20  2 0.40  4.24
   1.84
1   0.40 
2 0.92

The critical value of test statistic t for right-tailed test corresponding

(n-2) = 18 df at 1% level of significance is t ( n  2 ), α  t (1 8 ), 0.01  2.552.

Since calculated value of test statistic t (= 1.84) is less than the critical
value (= 2.552), that means calculated value of test statistic t lies in non-
rejection region, so we do not reject the null hypothesis and reject the
alternative hypothesis i.e. we reject our claim at 1% level of significance.
Thus, we conclude that sample provide us sufficient evidence against the
claim so there is no positive linear correlation between the family income
and the amounts of money spent per family member on food each month
in colony.

90
UNIT 12 CHI-SQUARE AND F-TESTS
Structure
12.1 Introduction
Objectives
12.2 Testing of Hypothesis for Population Variance Using χ2-Test
12.3 Testing of Hypothesis for Two Population Variances Using F-Test
12.4 Summary
12.5 Solutions / Answers

12.1 INTRODUCTION
Recall from the previous unit when we test the hypothesis about the difference
of means of two populations, t-test needs an assumption of equality of
variances of two populations under study. Other than this there are situations
where we want to test the hypothesis about the variances of two populations.
For example, an economist may want to test whether the variability in incomes
differ in two population. In such situations, we use F-test when the populations
under study follow the normal distributions.

Similarly, there are many other situations where we need to test the hypothesis
about the hypothetical or specified value of the variance of the population
under study. For example, the manager of the electric bulbs company would
probably be interested whether or not the variability in the life of bulbs is
within acceptable limits, the product controller of a milk company may be
interested in the variance of the amount of fat in the whole milk processed by
the company is no more than the specified level. In such situations, we use
χ2-test when the population under study follows the normal distribution.
This unit is divided into five sections. Section 12.1 is described the need of χ2
and F-tests. χ2-test for testing the hypothesis about the population variance is
discussed in Section 12.2. And F-test for equality of variances of two
populations is discussed in Section 12.3. Unit ends by providing summary of
what we have discussed in this unit in Section 12.4 and solution of exercises in
Section 12.5.
Objectives
After studying this unit, you should be able to:

 describe the testing of hypothesis for population variance; and

 explain the testing of hypothesis for two population variances.

12.2 TESTING OF HYPOTHESIS FOR

POPULATION VARIANCE USING χ2-TEST
In Section 11.3 of previous unit, we have discussed testing of hypothesis for
population mean when the characteristic under study follows the normal
distribution but when analysing quantitative data, it is often important to draw
conclusion about the average as well as the variability of a characteristic under
study. For example, if a company manufactured the electric bulbs then the
91
Testing of Hypothesis manager of the company would probably be interested in determining the
average life of the bulbs and also determining whether or not the variability in
the life of bulbs is within acceptable limits, the product controller of a milk
company may be interested in the variance of the amount of fat in the whole
milk processed by the company is no more than the specified level, etc.
The procedure of testing a hypothesis for population variance or standard
deviation is similar to the testing of population mean. The basic difference is
that here we use chi-square test instead of t-test because here the sampling
distribution of test statistic follows the chi-square distribution.
Assumptions
This test works under the following assumptions:
(i) The characteristic under study follows normal distribution. In other
words, populations from which random sample is drawn should be
normal with respect to the characteristic of interest.
(ii) Sample observations are random and independent.
Let X1 , X 2 ,..., X n be a random sample of size n drawn from a normal
population with mean  and variance σ2, where  and σ2 are unknown.
The general procedure of this test is explained below in detail:
As we are doing so far in all tests, first step in hypothesis testing problems is to
setup null and alternative hypotheses. Here, we want to test the claim about the
hypothesized or specified value σ02 of population variance σ 2 so we can take
our null and alternative hypotheses as
H 0 : 2  02 and H1 : 2  02 for two-tailed test 
H0 : 2  02 and H1 : 2  02 
or  for one-tailed test
H0 : 2  02 and H1 : 2  02 
For testing the null hypothesis, the test statistic χ2 is given by
2

χ 2

X i X

(n  1)S 2
~ χ 2  n 1  under H 0 ... (1)
σ 02 σ 02
1 2
where, S2 
n  1
X X
Here, the test statistic χ2 follows chi square distribution with (n − 1) degrees of
freedom as we have discussed in Unit 3 of this course.
After substituting values of n, S and 02 , we get calculated value of test
statistic. Let 2cal be the calculated value of test statistic χ2.
Obtain the critical value(s) or cut-off value(s) in the sampling distribution of
the test statistic χ2 and construct rejection (critical) region of size . The critical
value of the test statistic χ2 for various df and different level of significance 
are given in Table III of the Appendix given at the end of the Block 1of this
course.
After doing all the calculation discussed above, we have to take the decision
about rejection or non-rejection of the null hypothesis. The procedure of taking
the decision about the null hypothesis is explained is the next page:
92
For one-tailed test: Chi-square and F-Tests

Case I: When H0 :2  20 and H1 : 2  02 (right-tailed test)

In this case, the rejection (critical) region falls under the right tail of
the probability curve of the sampling distribution of test statistic χ2.
Suppose  2( ),  is the critical value at  level of significance where,
ν = n ‒1, so entire region greater than or equal to  2( ),  is the rejection
region and less than  (2 ),  is the non-rejection region as shown in Fig.
12.1.
If 2cal  (2 ),  , that means calculated value of test statistic lies in the
rejection (critical) region, then we reject the null hypothesis H0 at 
level of significance. Therefore, we conclude that sample data
provides us sufficient evidence against the null hypothesis and there
is a significant difference between hypothesized value and observed
value of the population variance σ2.
If 2cal  (2 ),  , that means calculated value of test statistic lies in non-
rejection region, then we do not reject the null hypothesis H0 at  Fig. 12.1
level of significance. Therefore, we conclude that the sample data
fails to provide us sufficient evidence against the null hypothesis and
the difference between hypothesized value and observed value of the
population variance σ2 due to fluctuation of sample.
Case II: When H0 :2  02 and H1 : 2  02 (left-tailed test)
In this case, the rejection (critical) region falls under the left tail of the
probability curve of the sampling distribution of test statistic χ2.
Suppose  2( ), (1 ) is the critical value at  level of significance then
entire region less than or equal to  2( ), (1 ) is the rejection (critical)
region and greater than  (2 ),(1 ) is the non-rejection region as shown
in Fig. 12.2.
If  2cal   (2 ),(1 ) , that means calculated value of test statistic lies in
rejection (critical) region, then we reject the null hypothesis H0 at 
Fig. 12.2
level of significance.
For  2cal   (2 ), (1 ) , that means calculated value of test statistic lies in
the non-rejection region, then we do not reject the null hypothesis H0
at  level of significance.
For two-tailed test:
When H0 :2  20 and H1 : 2  02
In this case, the rejection region falls under both tails of the
probability curve of sampling distribution of the test statistic χ2 and
half the area (α) i.e. α/2 of rejection (critical) region lies at left tail
and other half on the right tail. Suppose  2( ),(1 / 2) and 2( ),  / 2 are the
two critical values at the left-tailed and right-tailed respectively on
pre-fixed -level of significance. Therefore, entire region less than or
equal to  2( ),(1 / 2) and greater than or equal to (2 ),  / 2 are the rejection
Testing of Hypothesis (critical) regions and between  2( ),(1 / 2) and  2( ),  / 2 is the non-
rejection region as shown in Fig. 12.3.
If  2cal   (2 ),  / 2 or 2cal  (2 ),(1 / 2) , that means calculated value of test
statistic χ2 lies in rejection (critical) region, then we reject the null
hypothesis H0 at  level of significance.
If  (2 ),(1 / 2)   cal
2
  (2 ),  / 2 , that means calculated value of test
statistic χ2 lies in non-rejection region, then we do not reject the null
Fig. 12.3
hypothesis H0 at  level of significance.
Procedure of taking the decision about the null hypothesis on the basis of
p-value:
You have done it in so many test, it has become a routine thing for you. You
know that to take the decision about the null hypothesis on the basis of p-value,
the p-value is compared with level of significance (α) and if p-value is less than
or equal to α then we reject the null hypothesis and if the p-value is greater than
α we do not reject the null hypothesis.
For χ2-test, p-value is defined as:
For one-tailed test:
For H1 : 2  02 (right-tailed test)
p-value = P  2  cal
2


For H1 : 2  02 (left-tailed test)

p-value = P  2  cal
2


For two-tailed test: H1 : 2  20

For two-tailed test the p-value is approximated as
p-value = 2P  2  cal
2

The p-value for χ2-test can be obtained with the help of the Table-III
(χ2-table) given in the Appendix at the end of Block 1 of this course. Similar to
t-test, this table gives the χ2 values corresponding to the standard values of α
such as 0.995, 0.99, 0.10, 0.05, 0.025, 0.01, etc only. therefore, the exact
p-value is not obtained with the help of this table and we can approximate the
p-value for this test.
For example, if test is right-tailed and calculated (observed) value of test
statistic χ2 is 25.10 with 12 df then p-value is calculated as:
Since test statistic is based on the 12 df therefore, we use row for 12 df in the
χ2-table and move across this row to find the values in which calculated
χ2-value falls. Since calculated χ2-value falls between 23.24 and 26.22,
corresponding to one-tailed area α = 0.025 and 0.01 respectively, therefore
p-value lies between 0.01 and 0.025, that is,
0.01  p-value  0.025
If in the above example the test is two-tailed then the two values 0.01 and
0.005 would be doubled for p-value, that is,
2  0.01  0.02  p-value  0.05  2  0.025
Note 1: With the help of computer packages and softwares such as SPSS, SAS, Chi-square and F-Tests
MINITAB, EXCEL, etc. we can find the exact p-value for χ2-test.
Let us do some examples to become more user friendly with the test explained
above.
Example 1: The variance of a certain dimension article produced by a machine
is 7.2 over a long period. A random sample of 20 articles gave a variance 8. Is
it justifiable to conclude that variability has increased at 5% level of
significance assuming that the measurement of dimension article is normally
distributed?
Solution: Here, we are given that
Sample size = n = 20
Sample variance = S2 = 8
Specified value of population variance under test = σ20  7.2
Here, we want to test that variability of dimension article produced by a
machine has increased. Since variability is measured in terms of variance (  2 )
so our claim is  2  7.2 and its complement is  2  7.2. Since complement
contains the equality sign so we can take the complement as the null hypothesis
and the claim as the alternative hypothesis. Thus,
H0 : 2  20  7.2
H 1 :  2  7.2  variability of dimension article has increased 
Since the alternative hypothesis is right-tailed so the test is right-tailed test.
Here, we want to test the hypothesis about the population variance and sample
size is small n = 20(< 30). Also we are given that measurement of dimension
article follows normal distribution so we can go for χ2 test for population
variance.
So, test statistic is given by
 n  1  S2
χ2  2
~ χ 2 n 1  under H0
σ
19  8
  21.11
7.2
The critical (tabulated) value of test statistic χ2 for right-tailed test
corresponding (n-1) = 19 df at 5% level of significance is
 2(n 1),    (19),0.05
2
= 30.14.
Since calculated value of test statistic (= 21.11) is less than the critical
(tabulated) value (= 30.14), that means calculated value of test statistic lies in
non-rejection region as shown in Fig. 12.4, so we do not reject the null
hypothesis and reject the alternative hypothesis i.e. we reject our claim at 5%
level of significance.
Fig. 12.4
Decision according to p-value:
Since test statistic is based on 19 df therefore, we use row for 19 df in the χ2-
table and move across this row to find the values in which calculated χ2-value
falls. Since calculated χ2-value falls between 11.65 and 27.20, corresponding to
one-tailed area α = 0.90 and 0.10 respectively therefore p-value lies between
0.10 and 0.90, that is,
0.10  p-value  0.90
Testing of Hypothesis Since p-value is greater than  ( 0.05) so we do not reject the null hypothesis
at 5% level of significance.
Thus, we conclude that sample provides us sufficient evidence against the
claim so we may assume that the variability of dimension article produced by a
machine is not increased.
Example 2: The12 measurements of the same object on an instrument are
given below:
1.6, 1.5, 1.3, 1.5, 1.7, 1.6, 1.5, 1.4, 1.6, 1.3, 1.5, 1.5
If the measurement of the instrument follows normal distribution then carry out
the test at 1% level of significance that variance in the measurement of the
instrument is less than 0.016.
Solution: Here, we are given
Sample size = n = 12
Specified value of population variance under test 20  0.016
Here, we want to test that the variance (σ 2 ) in the measurements of the
instrument is less than 0.016. So our claim is  2  0.016 and its complement
is  2  0.016. Since complement contains the equality sign so we can take
complement as null hypothesis and claim as the alternative hypothesis. Thus,
H0 : 2  02  0.016 and H1 : 2  0.016
Since the alternative hypothesis is left-tailed so the test is left-tailed test.
Here, we want to test the hypothesis about the population variance and sample
size is small n = 12(< 30). Also we are given that the measurement of the
instrument follows normal distribution so we can go for χ2-test for population
variance.
For testing the null hypothesis, the test statistic χ2 is given by
2

χ 2

 X i X
~ χ 2 n 1  under H 0 ... (2)
2
σ 0
2
Calculation for   Xi  X  :
X X  X X  X
2

1.6 0.1 0.01

1.5 0 0
1.3 -0.2 0.04
1.5 0 0
1.7 0.2 0.04
1.6 0.1 0.01
1.5 0 0
1.4 -0.1 0.01
1.6 0.1 0.01
1.3 -0.2 0.04
1.5 0 0
1.5 0 0
  18
X 0
 X  X
2
 0.16

96
From above calculation, we have Chi-square and F-Tests

1 1
n
X X  18  1.5
12
2
Putting the values of  X  X  and σ 02 in equation (2), we have
2

χ 2

 X i X0.16

 10
2
σ 0 0.016
The critical value of test statistic χ2 for left-tailed test corresponding
(n-1) = 11 df at 5% level of significance is (2n 1),(1 )   2(11),0.95 = 4.57.

Since calculated value of test statistic (= 10) is greater than the critical value
(= 4.57), that means calculated value of test statistic lies in non-rejection region
as shown in Fig. 12.5 so we do not reject the null hypothesis and reject the
alternative hypothesis i.e. we reject the claim at 5% level of significance.
Thus, we conclude that sample provide sufficient evidence against the claim so
the variance in the measurement of the instrument is not less than 0.016.
In the same way, you can try the following exercises. Fig. 12.5

E1) An ambulance agency claims that the standard deviation in the length of
serving times is less than 15 minutes. Investigator suspects that this
claim is wrong and takes a random sample of 20 serving times which
has a standard deviation of 17 minutes. Assume that the service time of
the ambulance follows normal distribution. Test at  = 0.01, is there
enough evidence to reject the agency’s claim?
E2) A cigarette manufacturer claims that the variance of nicotine content of
its cigarettes is 0.62. Nicotine content is measured in milligrams and is
normally distributed. A sample of 25 cigarettes has a variance of 0.65.
Test the manufacturer’s claim at 5% level of significance.

12.5 TESTING OF HYPOTHESIS FOR TWO

POPULATION VARIANCES USING
F-TEST
In Section 12.1 we have already mentioned that before applying t-test for
difference of two population means, one of the requirements is to check the Some author uses this
equality of variances of two populations. This assumption can be checked with test as the name
the help of F-test for two population variances. This F-test is also important in “homoscedastic test”.
a number of contexts. For example, an economist may want to test whether the Because two
populations with equal
variability in incomes differ in two populations, a quality controller may want variances are called
to test whether the quality of the product is changing over time, etc. homoscedastic. This
word is derived from
Assumptions two wards, homo
The assumptions for F-test for testing the variances of two populations are: means “the same” and
scedastic means
1. The populations from which the samples are drawn must be normally “variability”.
distributed.
2. The samples must be independent of each other.
Now, we come to the general procedure of this test.
Testing of Hypothesis Let X1 , X2 ,..., X n1 be a random sample of size n1 from a normal population
with mean 1 and variance σ12. Similarly, Y1 ,Y2 ,..., Yn2 be a random sample of
size n2 from another normal population with mean 2 and variance σ22. Here,
we want to test the hypothesis about the two population variances so we can
take our alternative null and hypotheses as
H 0 : 12  22  2 and H1 : 12  22  for two-tailed test 
H 0 : 12  22 and H1 : 12  22 
or  for one-tailed test 
H 0 : 12  22 and H1 : 12  22 
For testing the null hypothesis, the test statistic F is given by
S12
F ~ F( n1 1, n 2 1) under H0 … (3)
S22
1 2 1 2
where, S12 
n1  1
  X  X  and S22 
n2 1
  Y  Y .

For computational simplicity, we can also write

2 2
2 1  2   X  2 1 
2   Y 
S 
1
n1  1 
 X  n  and S2  n  1   Y  n 
1 2 2
   
The test statistic F follows F-distribution with ν1   n1  1  and ν 2   n 2  1 
degrees of freedom as discussed in Unit 4 of this course.
After substituting the values of S12 and S22 , we get calculated value of test
statistic. Let Fcal be calculated value of test statistic F.
Obtain the critical value(s) or cut-off value(s) in the sampling distribution of
the test statistic F and construct rejection (critical) region of size . The critical
values of the test statistic F for various df and different level of significance 
are given in Table IV (F-table) of the Appendix at the end of Block 1of this
course.
After doing all calculations discussed above, we have to take the decision about
rejection or non rejection of the null hypothesis. This is explained below:
In case of one-tailed test:
Case I: When H0 :σ12  σ22 and H1 : 12  22 (right-tailed test)
In this case, the rejection (critical) region falls at the right side of the
probability curve of the sampling distribution of test statistic F.
Fig. 12.6
Suppose F( 1 , 2 ),  is the critical value of test statistic F with
(ν1 = n1 – 1, ν2 = n2 – 1) df at  level of significance so entire region
greater than or equal to F( 1 , 2 ),  is the rejection (critical) region and
less than F( 1 , 2 ),  is the non-rejection region as shown in Fig. 12.6.

If Fcal  F( 1 , 2 ),  , that means calculated value of test statistic lies in

rejection (critical) region, then we reject the null hypothesis H0 at 
level of significance. Therefore, we conclude that samples data
 provide us sufficient evidence against the null hypothesis and there is Chi-square and F-Tests
a significant difference between population variances.
If Fcal  F( 1 , 2 ),  , that means calculated value of test statistic lies in
non-rejection region, then we do not reject the null hypothesis H0 at 
level of significance. Therefore, we conclude that the samples data
fail to provide us sufficient evidence against the null hypothesis and
the difference between population variances due to fluctuation of
sample.
Case II: When H0 :σ12  σ22 and H1 : 12  22 (left-tailed test)
In this case, the rejection (critical) region falls at the left side of the
probability curve of the sampling distribution of test statistic F.
Suppose F( 1 , 2 ), (1) is the critical value at  level of significance then
entire region less than or equal to F( 1 , 2 ),(1) is the rejection(critical)
region and greater than F( 1 , 2 ),(1) is the non-rejection region as shown
in Fig. 12.7.
If Fcal  F( 1 , 2 ),(1 ) , that means calculated value of test statistic lies in
rejection (critical) region, then we reject the null hypothesis H0 at  Fig. 12.7
level of significance.
If Fcal  F( 1 , 2 ),(1 ) , that means calculated value of test statistic lies in
non-rejection region, then we do not reject the null hypothesis H0 at 
level of significance.
Note 2: F-table mentioned above gives only the right tailed critical values with
different degrees of freedom at different level of significance. The left tailed
critical value of F-test can always be obtained by the formula given below (as
described in Unit 4 of this course):
1
F 1 , 2 ,1 
F 2 , 1 , 
In case of two-tailed test:
When H0 :σ12  σ22 and H1 : 12  22
In this case, the rejection (critical) region falls at both sides of the
probability curve of the sampling distribution of test statistic F and
half the area(α) i.e. α/2 of rejection (critical) region lies at left tail and
other half on the right tail.
Suppose F( 1 , 2 ),(1 / 2) and F( 1 , 2 ),  / 2 are the two critical values at the
left-tailed and right-tailed respectively on pre-fixed  level of
significance. Therefore, entire region less than or equal to F( 1 , 2 ), (1 / 2)
and greater than or equal to F( 1 , 2 ),  / 2 are the rejection (critical)
regions and between F( 1 , 2 ),(1 / 2) and F( 1 , 2 ),  / 2 is the non-rejection
Fig. 12.8
region as shown in Fig. 12.8.
If Fcal  F( 1 , 2 ),  / 2 or Fcal  F( 1 , 2 ), (1 / 2) , that means calculated value of
test statistic lies in rejection(critical) region, then we reject the null
hypothesis H0 at  level of significance.
Testing of Hypothesis If F( 1 , 2 ),(1 / 2)  Fcal  F( 1 , 2 ),  / 2 , that means calculated value of test
statistic F lies in non-rejection region, then we do not reject the null
hypothesis H0 at  level of significance.
Procedure of taking the decision about the null hypothesis on the basis of
p-value:
To take the decision about the null hypothesis on the basis of p-value, the p-
value is compared with level of significance (α) and if p-value is less than or
equal to α then we reject the null hypothesis and if the p-value is greater than α
we do not reject the null hypothesis.
For F-test, p-value can be defined as:
For one-tailed test:
For H1 : 12  22 (right-tailed test)
p-value = P  F  Fcal 

For H1 : 12  22 (left-tailed test)

p-value = P  F  Fcal 

For two-tailed test: H1 : 12  22

For two-tailed test the p-value is approximated as
p-value = 2P  F  Fcal 
The p-value for F-test can be obtained with the help of Table-IV (F-table)
given in the Appendix at the end of Block 1 of this course. Similar to t-test or
χ2-test, this table gives F values corresponding to the standard values of α such
as 0.10, 0.05, 0.025 and 0.01 only. Therefore with the help of F-table, we
cannot find the exact p-value. So we can approximate p-value for this test.
For example, if test is right-tailed and calculated (observed) value of test
statistic F is 2.65 with 24 degrees of freedom of numerator and 14 degrees of
freedom of the denominator then we can find the p-value as:
Since test statistic is based on (24, 14) df therefore, we move across all the
values of tabulated F corresponding to (24, 14) df at α such as 0.10, 0.05, 0.025
and 0.01 and find the values in which calculated F-value falls. Since we have F
tabulated with (24, 14) df at α = 0.10, 0.05, 0.025 and 0.01 as
α= 0.10 0.05 0.025 0.01
F(24, 14), α 1.94 2.35 2.79 3.43

Since calculated F-value (= 2.65) falls between 2.35 and 2.79, corresponding
to one-tailed area α = 0.05 and 0.025 respectively therefore p-value lies
between 0.0025 and 0.05, that is,
0.025  p-value  0.05
If in the above example the test is two-tailed then the two values 0.025 and
0.05 would be doubled for p-value, that is,
0.05  p-value  0.10
Note 3: With the help of computer packages and software such as SPSS, SAS,
MINITAB, EXCEL, etc. we can find the exact p-value for F-test.
Let us do some examples based on this test.

100
Example 3: The following data relate to the number of items produced in a Chi-square and F-Tests
shift by two workers A and B for some days:
A 26 37 40 35 30 30 40 26 30 35 45
B 19 22 24 27 24 18 20 19 25

Assuming that the parent populations are normal, can it be inferred that B is
more stable (or consistent) worker compared to A?
Solution: Here, we want to test that worker B is more stable than worker A.
As we know that stability of data is related to variance of the data. Smaller
value of the variance implies data that it is more stable. Therefore, to compare
stability of two workers, it is enough to compare their variances. If 12 and 22
denote the variances of worker A and worker B respectively then our claim is
12   22 and its complement is 12  22 . Since complement contains the
equality sign so we can take the complement as the null hypothesis and the
claim as the alternative hypothesis. Thus,
H0 : 12  22

H1 : 12   22  worker B is more stable than worker A 

Since the alternative hypothesis is right-tailed so the test is right-tailed test.

Here, we want to test the hypothesis about two population variances and
sample sizes n1 = 11(< 30) and n2 = 9 (< 30) are small. Also populations under
study are normal and both samples are independent so we can go for F-test for
two population variances.
For testing the null hypothesis, test statistic is given by
S12
F ~ F n1 1, n 2 1  … (4)
S22

1 n1 2 1 n2 2
2
where, S1  
n1  1 i1
 Xi  X  and S2
2  
n2  1 i1
 Yi  Y 

Calculation for S12 and S22 :

Items X  X  2 Items Y  Y 2

Produced by
X  X  Produced by
Y  Y
A   X  34  B   Y  22 
( Variable X) (Variable Y)

26 −8 64 19 −3 9
37 3 9 22 0 0
40 6 36 24 2 4
35 1 1 27 5 25
30 −4 16 24 2 4
30 −4 16 18 −4 16
40 6 36 20 −2 4
26 −8 64 19 −3 9
30 −4 16 25 3 9
35 1 1
45 11 121
Total = 374 0 380 198 0 80

101
Testing of Hypothesis
Therefore, we have
1 1
X   X   374  34
n1 11
and
1 1
Y   Y   198  22
n2 9
Thus,
1 2 1
S12 
n1  1
  X  X    380  38
10
1 2 1
S22 
n2  1
  Y  Y    80  10
8
Putting the value of S12 and S 22 in equation (4), we have
38
F  3.8
10
The critical (tabulated) value of test statistic F for right-tailed test
corresponding (n1-1, n2 -1) = (10, 8) df at 1% level of significance is
F( n 1, n 1),   F(10,8), 0.01 = 5.81.
1 2

Since calculated value of test statistic (= 3.8) is less than the critical value
Fig. 12.9 (= 5.81), that means calculated value of test statistic lies in rejection region as
shown in Fig. 12.9, so we do not reject the null hypothesis and reject the
alternative hypothesis i.e. we reject the claim at 1% level of significance.
Thus, we conclude that samples provide us sufficient evidence against the
claim so worker B is not more stable (or consistent) worker compared to A.
Example 4: Two random samples drawn from two normal populations gave
the following results:

Sample Size Mean Sum of Squares of Deviation

from the Mean
Sample I 9 59 26
Sample II 11 60 32

Test whether both samples are from the same normal populations?
Solution: Since we have to test whether both the samples are from same
normal population, therefore, we will test two hypotheses separately:
(i) Two population means are equal, i.e. H 0 : µ1  µ 2

(ii) Two population variances are equal, i.e. H0 : σ12  σ22

Since sample sizes are small and populations under study are normal so two
means will be tested using t-test whereas two variances will be tested using
F-test. But t-test is based on the prior assumption that both population
variances are same, therefore, first we apply F-test and later the t- test (when
F- test accepts equality hypothesis).
Given that
2
n1  9, X  59,  X  X   26
 2 Chi-square and F-Tests
n 2  11, Y  60,  Y  Y   32
Therefore,
1 2 1
S12 
n1  1
  X  X 
9 1
 26  3.25

1 2 1
S22 
n2  1
  Y  Y 
11  1
 36  3.60

First we want to test that the variances of both normal populations are equal so
our claim is 12   22 and its complement is 12  22 . Thus, we can take the null
and alternative hypotheses as
H0 : 12  22 and H1 : 12  22
Since the alternative hypothesis is two-tailed so the test is two-tailed test.
For testing this, the test statistic F is given by
S12
F ~ F n1 1,n 2 1 
S22
3.20
  0.88
3.65
The critical (tabulated) value of test statistic F for two-tailed test corresponding
(n1-1, n2 -1) = (8, 11) df at 5% level of significance are F( n1 1, n 2 1),  / 2 
1 1 1
F(8,11),0.025  3.66 and F( n 1, n 1),(1 / 2)     0.24.
1 2
F(n2 1,n1 1),  / 2  F(11,8),0.025 4.24

Since calculated value of test statistic (= 0.88) is less than the critical value
(= 3.66) and greater than the critical value (= 0.24), that means calculated value
of test statistic F lies in non-rejection region so we do not reject the null
hypothesis i.e. we support the claim.
Thus, we conclude that both samples may be taken from normal populations
having equal variances.
Now, we test that the means of two normal populations are equal so our claim
is 1  2 and its complement is 1  2. Thus, we can take the null and
alternative hypotheses as
H0 : 1  2 and H1 : 1  2
The test statistic is given by
XY
t … (5)
1 1
Sp 
n1 n 2
1   X  X 2   Y  Y 2 
where, S2p 
n1  n 2  2  

1
 26  32  1  58  3.22
9  11  2 18

103
Testing of Hypothesis
Sp  3.22  1.79

Putting the values of X,Y,Sp ,n1 and n 2 in equation (5), we have,

59  60 1 1
t     1.23
1 1 1.79  0.45 0.81
1.79 
9 11
The critical values of test statistic t for (n1 + n2 -2) = 18 df at 5% level of
significance for two-tailed test are ± t(18) ,0.025 = ± 2.101.
Since calculated value of test statistic t (= −1.23) is less than the critical value
(= 2.101) and greater than the critical value (= ‒2.101), that means calculated
value of test statistic t lies in non-rejection region so we do not reject the null
hypothesis i.e. we support the claim.
Thus, we conclude that both samples may be taken from normal populations
having equal means.
Hence, overall we conclude that both samples may come from the same normal
populations.
Now, you can try the following exercises.
E3) Two sources of raw materials are under consideration by a bulb
manufacturing company. Both sources seem to have similar
characteristics but the company is not sure about their respective
uniformity. A sample of 12 lots from source A yields a variance of 125
and a sample of 10 lots from source B yields a variance of 112. Is it
likely that the variance of source A significantly differs to the variance
of source B at significance level  = 0.01?
E4) A laptop computer maker uses battery packs of two brands, A and B.
While both brands have the same average battery life between charges
(LBC), the computer maker seems to receive more complaints about
shorter LBC than expected for battery packs of brand A. The computer
maker suspects that this could be caused by higher variance in LBC for
brand A. To check that, ten new battery packs from each brand are
selected, installed on the same models of laptops, and the laptops are
allowed to run until the battery packs are completely discharged. The
following are the observed LBCs in hours:
Brand A 3.2 3.7 3.1 3.3 2.5 2.2 3.2 3.1 3.2 4.3

Brand B 3.4 3.6 3.0 3.2 3.2 3.2 3.0 3.1 3.2 3.2

Assuming that the LBCs of both brands follows normal distribution,

test the LBCs of brand A have a larger variance that those of brand B
at 5% level of significance.
We now end this unit by giving a summary of what we have covered in it.

12.4 SUMMARY
In this unit, we have discussed the following points:
1. Testing of hypothesis for population variance using χ2-test.
2. Testing of hypothesis for two population variances using F-test.
104
 Chi-square and F-Tests
12.5 SOLUTIONS / ANSWERS
E1) Here, we are given that
σ0  15, n  20, S  17
Here, we want to test the agency’s claim that the standard deviation ()
of the length of serving times is less than 15 minutes. So our claim is
 < 15 and its complement is  ≥ 15. Since complement contains the
equality sign so we can take complement as null hypothesis and claim
as the alternative hypothesis. Thus,
H 0 :   0  15

H1 :   15 SD of the length of serving 

 times is less than 15 minutes 
Since the alternative hypothesis is left-tailed so the test is left-tailed
test.
Here, we want to test the hypothesis about the population standard
deviation and sample size is small n = 20(< 30). Also we are given that
the service time of the ambulance follows normal so we can go for χ2
test for population variance.
The test statistic is given by
 n  1  S2
χ2  2
~ χ 2 n 1 
σ
2
19   17.0 
  24.40
 15.0 2
The critical value of test statistic χ2 for left-tailed test corresponding
(n-1) = 19 df at 1% level significance is (2n 1), (1 )  2(19),0.99 = 7.63.
Since calculated value of test statistic (= 24.40) is greater than the
critical value (= 7.63), that means calculated value of test statistic lies in
non-rejection region so we do not reject the null hypothesis and reject
the alternative hypothesis i.e. we reject our claim at 1% level of
significance.
Thus, we conclude that sample provides us sufficient evidence against
the claim so agency’s claim that the standard deviation () of the length
of serving times is less than 15 minutes is not true.
E2) Here, we are given that
σ20  0.62, n  25, S2  0.65
Here, we want to test the cigarette manufacturer’s claims that the
variance (2) of nicotine content of its cigarettes is 0.62 milligrams. So
claim is 2 = 0.62 and its complement is  ≠ 0.62. Since claim contains
the equality sign so we can take claim as null hypothesis and
complement as the alternative hypothesis. Thus,
H 0 : σ 2  σ 02  0.62 and H1 : σ 2  0.62
Since the alternative hypothesis is two-tailed so the test is two-tailed
test.

105
Testing of Hypothesis Here, we want to test the hypothesis about the population variance and
sample size is small n = 25(< 30). Also we are given that the nicotine
content of its cigarettes follows normal distribution so we can go for χ2
test for population variance.
The test statistic is given by
 n  1  S2
χ2  ~ χ 2 n 1 
σ2
24  0.65
  25.16
0.62
The critical (tabulated) values of test statistic χ2 for two-tailed test
corresponding (n-1) = 24 df at 5% level of significance are
 2( n 1),  / 2   (224),0.025  39.36 and  2( n 1), (1 / 2)   (224), 0.975 =12.40.
Since calculated value of test statistic (= 25.16) is less than the critical
value (= 39.36) and greater than the critical value (= 12.40), that means
calculated value of test statistic lies in non-rejection region, so we do
not reject the null hypothesis i.e. we support the claim at 5% level of
significance.
Thus, we conclude that sample fails to provide sufficient evidence
against the claim so we may assume that manufacturer’s claim that the
variance of the nicotine content of the cigarettes is 0.62 milligram is
true.
E3) Here, we are given that
n1  12, S12  125, n 2  10, S22  112
Here, we want to test that variance of source A significantly differs to
the variances of source B. If 12 and 22 denote the variances in the raw
materials of sources A and B respectively so our claim is 12  22 and
its complement is 12  22 . Since complement contains the equality sign
so we can take the complement as the null hypothesis and the claim as
the alternative hypothesis. Thus,
H0 : 12  22 and H1 : 12  22
Since the alternative hypothesis is two-tailed so the test is two-tailed
test.
Here, we want to test the hypothesis about two population variances
and sample sizes n1 = 12(< 30) and n2 = 10 (< 30) are small. Also
populations under study are normal and both samples are independent
so we can go for F-test for two population variances.
For testing this, the test statistic is given by
S12
F ~ F n1 1,n 2 1 
S22
125
  1.11
112
The critical (tabulated) value of test statistic F for two-tailed test
corresponding (n1-1, n2 -1) = (11, 9) df at 5% level of significance are
F( n1 1, n 2 1),  / 2  F(11,9),0.025  3.91 and

106
 1 1 1 Chi-square and F-Tests
F( n1 1, n 2 1),(1 / 2)     0.28.
F(n2 1,n1 1),  / 2  F(9, 11),0.025 3.59
Since calculated value of test statistic (= 1.11) is less than the critical
value (= 3.91) and greater than the critical value (= 0.28), that means
calculated value of test statistic lies in non-rejection region, so we do
not reject the null hypothesis and reject the alternative hypothesis i.e.
we reject the claim at 5% level of significance.
Thus, we conclude that samples provide us sufficient evidence against
the claim so we may assume that the variances of source A and B is
differ.
E4) Here, we want to test that the LBCs of brand A have a larger variance
than those of brand B. If 12 and 22 denote the variances in the LBCs
of brands A and B respectively so our claim is 12  22 and its
complement is 12  22 . Since complement contains the equality sign
so we can take the complement as the null hypothesis and the claim as
the alternative hypothesis. Thus,
H0 : 12  22 and H1 : 12  22
Since the alternative hypothesis is right-tailed so the test is right-tailed
test.
Here, we want to test the hypothesis about two population variances
and sample sizes n1 = 10(< 30) and n2 = 10 (< 30) are small. Also
populations under study are normal and both samples are independent
so we can go for F-test for two population variances.
For testing the null hypothesis, test statistic is given by
S12
F ~ F n1 1, n 2 1  … (6)
S22
2
1 2 1    X 
where, S 
n1  1 
2
1  X  X   n  1   X  n  and
2

1  1 
2
1 2 1    Y  .
S22 
n2  1 
 Y  Y  n 1
 Y 2

n2 

2  
2 2
Calculation for S1 and S2 :
LBCs of Brand A X2 LBCs of Brand A Y2
(X) (Y)
3.7 13.69 3.6 12.96
3.2 10.24 3.2 10.24
3.3 10.89 3.2 10.24
3.1 9.61 3.0 9.00
2.5 6.25 3.0 9.00
2.2 4.84 3.2 10.24
3.1 9.61 3.2 10.24
3.2 10.24 3.1 9.61
4.3 18.49 3.2 10.24
3.2 10.24 3.1 9.61
Total = 3.18 104.1 31.8 101.38
107
Testing of Hypothesis
From the calculation, we have
1 1
X
n1  X   31.8  3.18,
10
1 1
Y
n2  Y   31.8  3.18
10

Thus,
2
1    X  1  31.8 2 
n1  1  
2 2
S   X    104.10 
1
n1  9  10 
 
1
  2.98  0.33
9
2
1    Y    1 101.38   31.8 2 
n2  1  
S22   Y 2
 
n 2  9  10 
 
1
  0.26  0.03
9
Putting the values of S12 and S22 in equation (6), we have
0.33
F  1.1
0.03
The critical (tabulated) value of test statistic F for right-tailed test
corresponding (n1-1, n2 -1) = (9, 9) df at 1% level of significance is
F( n1 1, n2 1),   F(9,9),0.01 = 5.35.

Since calculated value of test statistic (= 1.1) is less than the critical
value (= 5.35), that means calculated value of test statistic lies in non-
rejection region so we do not reject the null hypothesis and reject the
alternative hypothesis i.e. we reject the claim at 1% level of
significance.
Thus, we conclude that samples provide us sufficient evidence against
the claim so variance in LBCs of brand A is not greater than variance in
LBCs of brand B.

108