CHAPTER FIVE

SAMPLING AND SAMPLING DISTRIBUTION

Learning Objectives
When you have completed this chapter, you will be able to:

Define sample, sampling and sampling distribution.

Understand and identify the different types of sampling
techniques in statistics.

Understand and calculate the sampling distribution of the
different types of sample Statistics.

Use and Understand the central Limit Theorem.

5.1 Introduction
In statistics, Sampling plays vital role. Because most of the time we are occupied
with a class of problems that involve an attempt to say something about the
properties of a large group of objects, given information on a relatively small subset
of them. This is done, especially in Least Developed Countries (LDCs), because
there is no resource to undertake census or complete enumeration there. So the
major motivation for examining a sample rather than the whole population is that
the collection of complete information on the latter would typically be prohibitively
expensive. Even in circumstances where sufficient resources are apparently
available to contact the whole population, it may be preferable to devote these
resources to just a subset of the population in the hope that such a concentration
of effort will produce more accurate measurements. But if we take a sample from a
population, the eventual aim is to make statements that have some validity for the
population at large. Therefore, it is important that the sample be representative of
the whole population.
Generally, the overall purpose of this chapter is to introduction and equip students
with the concepts of sample, sampling and sampling distribution.

5.2 Introduction to Sampling

At the out set of this course, we have defined what sample and population mean.
Restating again, population refers to all items that have been chosen for study. That
is the larger parent group is called population and sample refers to a portion or
subset of the population selected.

Examples of populations that might be of interest are:

a) The incomes of all families living in Mekelle.
b) The annual yields of all stocks traded on the American Stock Exchange.
c) The costs of all claims for automobile accident insurance coverage received
by Awash Insurance Company in a given year.
d) The number of miles per gallon achieved by all cars of a particular model.
Sometimes it is possible and practical to examine every person or item in the
population we wish to study, which we call complete enumeration or census,

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 1

though may be preferable to devote our resources to just a subset of the
population.
Sampling: - is the process of selecting samples (or part of the items from the
population) from of population of interest for critical investigation.

Check yourself
1. Why sampling?
2. What is the difference and similarity between Sampling
And Census (complete enumeration).

Mathematically, we can describe samples and population by using measures such as

the mean, median, mode, and standard deviation. When these terms describe the
characteristics of a sample, they are called statistics. When they describe the
characteristics of a population, they are referred parameters.
Example: If the mean CGPA of all first year economics students is 3.32. In
this case, 3.32 is the characteristics of the population “All 1st year
economics students at Adi-Haki Campus” and is termed as population
parameter. On the other hand, if we say that the mean CGPA of the a
sample of 50 first year economics students at Adi-Haki campus is 3.32, we
are using 3.32 to describe a characteristic of the sample of 50 “First year
economics students at Adi-Haki campus”. In this case, 3.32 would be a
sample statistics.

If we are convinced that the sample statistics are accurate estimate of the
population characteristics, we could use sample statistics to estimate the
population parameter without measuring the entirety of the items under study.
In order to be consistent, Statisticians use lower case Roman letters to denote
sample statistics, and Greek or capital letters to denote population parameters.

Table 5.1: Summary of definitions and characteristics of population and sample

Population Sample
Definition Collection of all items being dealt in Part (sub-set) of the
a study population
Characteristic Population Parameter Sample Statistics
Symbol • Population size = N • Sample size = n
• Population mean = µ • Sample mean = x
• Population variance = σ2 • Sample variance = S2
• Population standard • Sample standard
deviation = σ deviation = S

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 2

5.2.1 Types of Sampling
In statistics, there are two methods of selecting samples from a population:

A. Random or probability sampling

B. Non-random, non-probability or judgment sampling

A. Probability (Random) Sampling- is sampling when all items (i.e., each

element) in the population have a chance of being chosen in the sample and
the probability of each element of the population included in the sample is
known. There are several probability sampling techniques. The following are
the commonly used probability sampling techniques in statistical
investigation:

1. Simple Random Sampling

2. Systematic Random Sampling
3. Stratified Sampling
4. Cluster Sampling
5. Multi- Stage Sampling

1. Simple Random Sampling: - is selecting samples so that each possible sample

has an equal chance of being picked, and each element in the population has the
same probability of being included in the sample and is independent of whether
some other element is chosen. In order to use simple random sampling the
elements in the population from which samples are drawn should be homogeneous.

Example:
Suppose that a restaurant has four branches (N,S, E and W) and that it wants to
select samples of two branches at a time in order to evaluate the operation of the
branches. Using simple random sampling, six different samples of size 2 that can be
drawn from the population (i.e., the four branches). These six samples are (NS);
(NE); (NW); (SE); (SW); and (EW).
The probability of each sample is 1/6 to be selected from the population and the
probability of an element in the sample is ½.

Some Definitions:
Finite population: means that the population has limited size,
that is to say, there is a whole number (N) that tells us how many items are there
in the population.

With Replacement: is immediately putting back an element picked, before the

second element is chosen.
Infinite population: is a population in which it is theoretically impossible to
observe all the elements.

N.B: The numbers of different possible samples of size n that can be drown
without replacement from a population of N elements equals:

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 3

 N!
NC
n=
( N − n)!n!

Example: Suppose that a population has five elements

(N = 5) 3, 6, 9, 12 and 15.
If we draw samples of 3 (n = 3) by applying sampling without replacement,
we will have NCn= 5C3 = 10 different possible sample
For practical purposes, we use the term infinite population when we are talking
about a population that could not be enumerated in a reasonable period of time.

How Statisticians actually do Random Sampling?

In cares where the number of elements in the population is small, they number
each element, record the number on a (slip) piece of paper, place it in a hat or
bowl where it is mixed well, and finally draw n-piece of paper from the hot or
bowl.
This procedure is straight forward but when the numbers of elements are large it
becomes cumbersome. And Statisticians usually use the table of random digits (i.e.
using a Table of Random Digits).

2. Systematic Sampling: - is method of selecting sample in which an element in

the sample is obtained by taking every Kth element on a list of all elements in the
population. To determine which of the first K elements is chosen, a number from
1 to K is chosen at random.

Illustration:- Suppose that there are 1000 resident or households in one Kebelle
with different income levels. If the Statistician /researcher has the list of all
households randomly listed and wants to study the income disparity in that kebelle
by taking 50 samples. Since there are 1000 households the sampling can be
accomplished by taking every 20th household on the list. To determine which of
the first 20 element to begin with the statician/researcher can randomly chose a
number from 1 to 20 Once this number is chosen (let’s say 3), then the statician
selects the 3rd, 23rd, 33rd, 43rd, … households from the list. Such kind of sampling is
systematic sampling.

Often systematic sampling is regarded as identical as the simple random sampling.

This is true only if the elements of the population are in random order on the list.
This means the elements of the population on the list are not in a sort of
periodicity or any other type of pattern on the list.

3. Stratified Sampling: - is sampling in which the population is divided in to

strata and random sample is taken from the elements in each stratum.
The basic idea in formulating strata is to sub divide the population in to a relatively
homogenous groups with in the strata, and subdivide the population so that
relatively greater variations or heterogeneity exists with regard to the
characteristics measured between strata (or sub divisions).

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 4

Illustration: If a researcher wants to deal with the income inequality situation in
Mekelle city. The researcher can divide the households in to different groups. As
follows:
o Civil Servant
o Merchant
o Petty Traders & local drink sellers

4. Cluster Sampling: - is sampling in which one divides the elements in the

population in to a number of clusters or groups. One then begins by choosing at
random a sample of these clusters, after which all or a simple random sample of the
elements in each chosen cluster is selected. Some times, this is referred as two
stage cluster sampling.

Illustration: Still taking the study of the income disparity condition in Mekelle. In
this case, the Mekelle city will be classified by locality (i.e., in to Northern, southern
part of Mekelle, etc). Once the city is classified in to various clusters, randomly
some of the clusters (i.e., locality in our case) will be chosen and the researcher can
take all elements in the cluster or randomly selects elements from the chosen
cluster. This depends on the cost and other considerations.

Comparison of Stratified and Cluster Sampling

Most of the time students face difficulty in differentiating stratified and cluster
sampling. The main distinguishing criteria of stratified from cluster sampling is that
in the case of stratified sampling the population is divided in to well-defined groups,
where each group has homogeneity with in itself but wider heterogeneity (or
variation) among the groups. In the case of cluster sampling, the situation is the
reveres for stratified sampling (i.e., the different clusters are homogeneous but
elements in each cluster are heterogeneous).
o In statistical inference the assumption is that the samples are selected
using simple random sampling (other probability sampling technique
attempt to approximate the simple random sampling).
o Very few so-called random samples are truly random. Why?

5. Multi-Stage Sampling:- The four methods we have covered so far viz. simple,
stratified, systematic and cluster, are the simplest random sampling strategies. In
most real applied social research, we would use sampling methods that are
considerably more complex than these simple variations. The most important
principle here is that we can combine the simple methods described earlier in a
variety of useful ways that help us address our sampling needs in the most efficient
and effective manner possible. When we combine sampling methods, we call this
multi-stage sampling.

For example, consider the idea of sampling Amhara region residents for face-to-
face interviews. Clearly we would want to do some type of cluster sampling as the
first stage of the process. We might sample townships or census tracts throughout

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 5

the region. But in cluster sampling we would then go on to measure everyone in
the clusters we select. Even if we are sampling census tracts we may not be able to
measure everyone who is in the census tract.

So, we might set up a stratified sampling process within the clusters. In this case,
we would have a two-stage sampling process with stratified samples within cluster
samples. Or, consider the problem of sampling students in grade schools. We might
begin with a national sample of school districts stratified by economic status and
educational level. Within selected districts, we might do a simple random sample of
schools.

Within schools, we might do a simple random sample of classes or grades. And,

within classes, we might even do a simple random sample of students. In this case,
we have three or four stages in the sampling process and we use both stratified and
simple random sampling. By combining different sampling methods we are able to
achieve a rich variety of probabilistic sampling methods that can be used in a wide
range of social research contexts.

B. Non-probability (Non-random/Judgment):- sampling is a sampling

methodology where personal knowledge and opinion play major role in identifying
which elements of the population are to be included in the sample, and the
probability of an element from the population to be included in the sample is not
known.
Just like the probability sampling, the following are the non-probability sampling
techniques of selecting a sample that will be discuss later:
1. Accidental, Haphazard or Convenience Sampling
2. Purposive Sampling
3. Modal Instance Sampling
4. Expert Sampling
5. Quota Sampling
6. Snowball Sampling

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 6

1. Convenience Sampling:- One of the most common methods of non-probability sampling
goes under Convenience Sampling. This is the category of the traditional "man on the street" (of
course, now it's probably the "person on the street") interviews conducted frequently by
television news programs to get a quick (although non representative) reading of public opinion.
It is argued that most researchers use, in their process of data collection, primarily convenience
sampling. In many research contexts, we sample simply by asking for volunteers. Clearly, the
problem with all of these types of samples is that we have no evidence that they are
representative of the populations we are interested in generalizing to and in many cases we
would clearly suspect that they are not.

2. Purposive Sampling: - In purposive sampling, we sample with a purpose in mind. We usually

would have one or more specific predefined groups we are seeking. For instance, have you ever
run into people in a mall or on the street who are carrying a clipboard and who are stopping
various people and asking if they could interview them? Most likely they are conducting a
purposive sample (and most likely they are engaged in market research). They might be looking
for Ethiopian females between 30-40 years old. They size up the people passing by and anyone
who looks to be in that category they stop to ask if they will participate. One of the first things
they're likely to do is verify that the respondent does in fact meet the criteria for being in the
sample. Purposive sampling can be very useful for situations where you need to reach a targeted
sample quickly and where sampling for proportionality is not the primary concern. With a
purposive sample, you are likely to get the opinions of your target population, but you are also
likely to overweight subgroups in your population that are more readily accessible.

All of the methods that follow can be considered subcategories of purposive sampling methods.
We might sample for specific groups or types of people as in modal instance, expert, or quota
sampling. We might sample for diversity as in heterogeneity sampling. Or, we might capitalize on
informal social networks to identify specific respondents who are hard to locate otherwise, as in
snowball sampling. In all of these methods we know what we want -- we are sampling with a
purpose.

3. Modal Instance Sampling: - In statistics, the mode is the most frequently occurring value in
a distribution. In sampling, when we do a modal instance sample, we are sampling the most
frequent case, or the "typical" case. In a lot of informal public opinion polls, for instance, they
interview a "typical" voter.

There are a number of problems with this sampling approach. First, how do we know what the
"typical" or "modal" case is? We could say that the modal voter is a person who is of average age,
educational level, and income in the population. But, it's not clear that using the averages of these
is the fairest (consider the skewed distribution of income, for instance). And, how do you know
that those three variables -- age, education, income -- are the only or event the most relevant for
classifying the typical voter? What if religion or ethnicity is an important discriminator? Clearly,
modal instance sampling is only sensible for informal sampling contexts.

4. Expert Sampling: - Expert sampling involves the assembling of a sample of persons with
known or demonstrable experience and expertise in some area. Often, we convene such a
sample under the support of a "panel of experts." There are actually two reasons you might do

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 7

expert sampling. First, it would be the best way to elicit the views of persons who have specific
expertise. In this case, expert sampling is essentially just a specific sub-case of purposive sampling.
But the other reason you might use expert sampling is to provide evidence for the validity of
another sampling approach you've chosen.

For instance, let's say you do modal instance sampling and are concerned that the criteria you
used for defining the modal instance are subject to criticism. You might convene an expert panel
consisting of persons with acknowledged experience and insight into that field or topic and ask
them to examine your modal definitions and comment on their appropriateness and validity. The
advantage of doing this is that you aren't out on your own trying to defend your decisions i.e. you
have some acknowledged experts to back you. The disadvantage is that even the experts can be,
and often are, wrong.

5 .Quota sampling:-In quota sampling, you select samples non-randomly according to some
fixed quota. There are two types of quota sampling: proportional and non proportional. In
proportional quota sampling you want to represent the major characteristics of the population by
sampling a proportional amount of each. For instance, if you know the population has 40%
women and 60% men, and that you want a total sample size of 100, you will continue sampling
until you get those percentages and then you will stop.

So, if you have already got the 40 women for your sample, but not the sixty men, you will
continue to sample men but even if legitimate women respondents come along, you will not
sample them because you have already "met your quota." The problem here (as in much
purposive sampling) is that you have to decide the specific characteristics on which you will base
the quota will it be by gender, age, education race, religion, etc.

Non-proportional quota sampling is a bit less restrictive. In this method, you specify the minimum
number of sampled units you want in each category. Here, you're not concerned with having
numbers that match the proportions in the population. Another term for this is sampling for
diversity. In many brainstorming or nominal group processes (including concept mapping), we
would use some form of heterogeneity sampling because our primary interest is in getting broad
spectrum of ideas, not identifying the "average" or "modal instance" ones.

In effect, what we would like to be sampling is not people, but ideas. We imagine that there is a
universe of all possible ideas relevant to some topic and that we want to sample this population,
not the population of people who have the ideas. Clearly, in order to get all of the ideas, and
especially the "outlier" or unusual ones, we have to include a broad and diverse range of
participants. Heterogeneity sampling is, in this sense, almost the opposite of modal instance
sampling.

6. Snow Ball Sampling:-In snowball sampling, you begin by identifying someone who meets the
criteria for inclusion in your study. You then ask them to recommend others who they may know
who also meet the criteria. Although this method would hardly lead to representative samples,
there are times when it may be the best method available. Snowball sampling is especially useful
when you are trying to reach populations that are inaccessible or hard to find. For instance, if you
are studying the homeless, you are not likely to be able to find good lists of homeless people

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 8

within a specific geographical area. However, if you go to that area and identify one or two, you
may find that they know very well who the other homeless people in their vicinity are and how
you can find them.

Check your Self

List and explain the different types of probability sampling.

List and explain the different types of Non-probability sampling.

5.3 The Concept of Sampling Distribution

So far we have examined how samples can be taken from population. Using one of the already
discussed sampling techniques if we take several samples from a population, the statistics we
would compute for each sample need not be the same and most likely would vary from sample to
sample because each selected sample do not likely to contain the same elements.

Illustration: - Suppose that a population has five elements

(N = 5) 3, 6, 9, 12 and 15. If we draw samples of
3 (n = 3) by applying sampling without replacement, we will have
NCn= 5C3 = 10 different possible sample. The following may be the elements in the
sample.
Possible Samples: 3, 6,9 3,6,12 3,6,15 3,9,12
3,9,15 3,12,15 6,9,12 6,9,15
6,12,15 9,12,15
For each sample we can compute the mean, proportion, variance, and standard deviations
(i.e., the sample statistics). The following table reveals the mean value for each sample.

Samples Mean ( x i ) we know that the population mean is

3, 6, 9 6 given as
3, 6, 12 7

3, 6, 15 8 µ=
∑x =
45
= 9
N 5
3, 9, 12 8
3, 9, 15 9 The population mean value ( µ ) varies from
3, 12, 15 10 some of the sample means ( x i ). This
6, 9, 12 9 leads us in to concept of sampling
6, 9, 15 10 distribution.
6, 12, 15 11
9, 12, 15 12
∑x= 90

Sampling Distribution: - is a probability distribution of all the values of Sample statistics.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 9

To understand how close the sample values (statistics) to the population parameter we need to
understand the properties of the sampling distribution of sample statistics. And we have sampling
distribution of the mean, proportion and variance which we will discuss here.

5.3.1 Sampling Distribution of the mean

It is the probability distribution of all possible sample means of size n that could be taken from a
population of size N. To illustrate, we have taken samples, each with a size of 3, from a
population of 10 elements in the above example and the computed mean values for each sample
with its associated probability is referred as sampling distribution of the mean.

Sample mean ( x ) Probability P ( x )

6 0.1
7 0.1
8 0.2
9 0.2
10 0.1
11 0.1
12 0.1

The above table is sampling distribution of the mean. In this section our objective is to describe
the characteristics of the sampling distribution of the mean and the shape of the sampling
distribution.
The concept of sampling distribution is helpful in letting us to make probability statements about
the error involved when the sample mean ( x ) is used to estimate the population mean (µ). That
is, the practical value of the sampling distribution of the mean can be used to provide probability
information about the sampling ERROR

Characteristics/properties of the sampling distribution of the mean

1. The expected value of the sample mean, E ( x ). (Or which is the mean of the sample
means is equal to the population mean.
Algebraically µ ( x ) = E ( x ) = µ
2. Standard Deviation of the sample mean δ x . Give the population mean (µ)., population
standard deviation (δ), the sample size (n) and population size (N); the standard deviation
of the sample mean is given as follows:

δ
δx = - - - - - - - - - For infinite population.
n

δ N −n
δx = -------- For finite population.
n N −1
With Simple random Sampling the value of Standard deviation of the mean depends on
whether the population is finite or infinite.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 10

 ►A population is said to be infinite when it is not possible to list or count all the elements
included in the population, (i.e., when the elements are unlimited). Or, in the cases when
the elements in the population are limited, the population may be considered as infinite
when the sample size is small and as a rule of thumb, statisticians consider the population
as infinite when n ≤ 5% of N.
►A population is said to be finite when n > 0.05 N.
N −n
The value is referred as finite population correction factor.
N −1

3. The sampling distribution of the mean is normally distributed regardless of the population from
which it is drawn
The aforementioned are some of the basic common properties of the sampling distribution the
mean. Next one examines the shape of the sampling distribution of the mean, which is already
stated as the third characteristics of the sampling distribution of the mean, when the population
from which the samples drawn is normal or non-normal.

I. Sampling Distribution of the mean from normal population

Whenever the population has a normal probability distribution, the sampling distribution of the
mean ( x ) is a normal probability distribution for any sample size.
In this case E ( x ) = µ ( x ) = µ and
δ n N −n n
δx = , for < 0.05 and δ x = δ , For > 0.05
n N N −1 N

a. Sampling distribution of the mean from Unknown or

Non-normal Population
In the preceding case, we have seen that when the population is normal the sampling distribution
is also normal. Yet researchers, decision markers and the like are not only confronted with
normal population, they may face non-normal or unknown population distributions. How does
the sampling distribution of the mean react in such situations? Let’s illustrate:
Illustration: Suppose that the data in the table below indicated the growth rate of some East
Asian countries.

Table 5.2 Foreign Direct Investment (FDI) of East Asian Countries in millions of USD
Country Philippines Indonesia Taiwan [Link] Singapore
FDI in millions 3 3 7 9 14
of USD
Population Mean FDI in millions of USD ( µ) = 36/5 =7.2

Considering the data we agree that the population may not be normal, because there are only 5
elements involved in the population hence too small to be approximated by a normal distribution.
Let us draw samples of size 3, compute the sample means ( x ) and list them; and calculate the
mean of sampling distribution ( µ x ) . This is done and put in table below.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 11

Table 5.3 Possible Sample means of the countries’ FDI in millions of USD
Samples Sample Data Sample Mean
TSKS 7 + 9 + 14 10.00
PSKS 3 + 9 + 14 8.67
PTS 3 + 7 + 14 8.00
ITS 3 + 7 + 14 8.00
I S KS 3 + 9 + 14 8.67
P TSK 3+7+9 6.33
IT SK 3+7+9 6.33
PIS 3 + 3 + 14 6.67
PIT 3+3+7 4.33
PIS 3+3+9 5.00

µx =
∑x i
=
72
= 7 .2 = µ
NCn 10

From the table we recognize that even in a case in which the population is not normal, the mean
of the sampling distribution ( µ x ) is still equal to the population mean.
Probability distribution figures.

Figure: 5.1 Probability distribution of FDI of East Asian Countries.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 12

Looking to the figure above, as the sample size increases, the distribution looks a little more like
the bell shape of the normal distribution. If we could have sufficient time and space and to
repeatedly undertake the above process by enlarging the population size so as to take different
size of samples (i.e., large sample size), doing so will end up with the sampling distribution
approaching normality, regardless of the shape of the population distribution. This leads us to
one of the important theorems in inferential statistics.

Central Limit Theorem (CLT)

It is one of the most important theorems in statistics. In selecting simple random samples of size
n from a population of N elements, the sampling distribution of the sample mean ( x ) can be
approximated by a normal probability distribution as the sample size becomes large. The
significance of the central limit theorem is that it permits us to use sample statistics to make
inferences about population parameters without knowing anything about the shape of the
frequency distribution of that population other than what we can get from the sample. That is the
crucial additional information provided by the central Limit theorem is that whatever the
distribution of the Xi (a certain random variable), provided that σ 2 is finite, as a number of
terms n in the sum becomes large, the distribution of Z tends to the standard normal
distribution.

Central Limit Theorem

Let X1, X2,…,Xn be n independent random variables having identical distributions with mean µ
and standard deviation σ 2 . Denote by X and X , respectively, the sum and average of these
random variables. Then as n becomes large, the distribution of
X − nµ x− µ
Z= = tends to normal distribution
nσ 2 σ/ n

Examples:
1. A population of 100 elements has a mean of 19.2 and standard deviation of [Link] is the
mean and standard deviation of the sampling distribution of the mean for samples of size
25?
Solution:
µ x = µ ⇒ µ x = µ = 19.2
δ N −n n 25
δx = , Since > 0.05 ⇒ > 0.05
n N −1 N 100
1 100 − 25 1 75
δx = . = = 0.174
25 100 − 1 5 99
Check Your Self
Interpret δ x = 0.174 of the above result.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 13

 2. A library Checks out an average of µ = 320 books per day, with a standard deviation of
δ = 75 books. Consider a sample of 30 days of operation, with x being the sample mean
number of books checked out per day. What is the probability that the sample mean for
30 days will be between 300 and 340 books?

Solution:
Given of µ = 320 δ = 75
Required: P (300 < x < 340)?
This is the case of sampling distribution of x . Since the population is normal, the
sampling distribution of x is also normal. So, using the normal probability distribution
way of computing probability, P (300 < x < 340) should be converted in to standard
normal probability distribution.
Given
δ 75 75
x 1 = 300 , x 2 = 340 , and δ x = = = = 13.6
n 30 5.5

Required: P (300 < x < 340)

Solution:
x1 − µ 300 − 320 − 20
z1 = = = = − 1 . 47
δ x 13 . 6 13 . 6
x2 − µ 340 − 320 20
z1 = = = = 1 . 47
δ x
13 . 6 13 . 6

300 µ x =320 340

The standard normal probability table given us an area of 0.4292 corresponding to a Z-
value of -1.47, and it gives an area of 0.4292 for a Z-value 1.47. By summing these two
together; we get 0.8584 as total probability that the sample mean will be between 300 and
340 books per day.

3. The distribution of annual earnings of all economics graduates with zero year experience
is skewed negatively, as shown in figure (a) below. This distribution has a mean of 19,000
Birr and a standard deviation of 2,000 Birr. If we draw a random sample of 30 fresh
economic graduates, what is the probability that their earnings will average more than
19,750 Birr annually?
Solution
Given
µ = 19,000 σ = 2,000
Required: P( x > 19,750)?

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 14

 δ = 2000

X
µ = 19,000
(a
.
In order to answer this question, first let’s calculate the standard error of the mean ( x )
δ 2,000 2000
δx = = = 365.16 Birr
n 30 5.477
Next, let’s convert the random variable in to standard normal probability
value (Z)
Z= x - µ x = 19750
19,750 − 19000 750
δx Z= = = 2.05
365.16 365.16
Then, Z = 2.05 corresponds to area equal to 0.4798; we are interested to area above Z = 2.05.
This is obtained by taking the difference between the area to the right of Z = 0, which is 0.5 and
the area between Z = 0 and Z = 2.05 that is 0.4798. Thus, the area to the right of Z = 2.05 is
(0.5 – 0.4798) = 0.0202. Thus, P( x > 19,750) = 0.0202. This is the area to the right of Z= 2.05 of
the normal distribution graph shown below.
In this problem’s case we assumed that the sampling distribution of the mean is normal, using the
central limit the area since n = 30.

4. If the number of miles per gallon achieved by all cars of a particular Model has mean of 25 and
standard deviation of 2, what is the probability that, for a random sample of 20 such cars, average
miles per gallon will be less than 24? Assume that the population distribution is normal.
Solution:
Let x denote the sample mean. Then we need to find
24 − µ X
P( x < 24) = p[ Z < ]
σX

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 15

 σX 2
Where µ x = 25 is the population mean and σ x = = = 0.4472
n 20
24 − 25
Hence P ( X <24) = P [Z< ]
0.4472
= P [Z< − 2.24 ]
Where Z has a standard normal distribution. Therefore, P ( x <24) =P(Z < - 2.24) = 0.0125
Summary

A sampling distribution is a probability distribution of sample statistics.

If X ∼ N (µ,δ); then x ∼ N (µ x ,δ x )
Applying the central limit theorem,
δ

If Xi α N (µ, δ); then as n ≥ 30, x ∼ N (µ x , ).
n

5.3.2 Sampling Distribution of Proportion

We saw in our earlier discussion that if n independent trials, each with probability of success p,
are carried out, then the total number of success, X, obeys a binomial distribution. A common
problem arises when the parameter p is unknown. For instance, we may want to determine the
proportion of an elaborate intending to vote for a particular candidate for office or the
proportion of an electorate intending to vote for a particular candidate for office or the
proportion of a magazine’s readership likely to be in the market for a specific product. In cases of
this kind it is natural to base inference on the proportion of successes in the sample.
Definition:- Let X be the number of successes in a binomial sample of n observations, where the
probability of success is p. (In most applications, the parameter p is the proportion of members
of a large population possessing a characteristics of interest.) Then the proportion of successes:
X
P = is the sample proportion
n
Example
Consider a population of N = 5 given numbers 3, 6, 9, 12, and 15. Let’s take even numbers, the
proportion of even numbers is 2/5 = 0.4. Consider a samples of size 3 (n = 3) that are drawn
from the population the samples, sample proportions are given in table 5.3 below.
Table: 5.3 Different samples drawn from a population
Samples Sample Proportion ( P )
3, 6, 9 1/3
3, 6, 12 2/3
3, 6, 15 1/3
3, 9, 12 1/3
3, 9, 15 0/3
3, 12, 15 1/3
6, 9, 12 2/3
6, 9, 15 1/3
6, 12, 15 2/3
9, 12, 15 1/3

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 16

Given table 5.3 above, we can construct the probability distribution of the sample proportions as
shown in table 5.4 below.

Table: - 5.4 Probability Distribution of sample proportion ( P )

Sample proportion ( P ) 0/3 1/3 2/3

Probability: P( P ) 0.1 0.6 0.3

Given table 5.4, probability distribution of sample proportion ( P ) is the

Sampling distribution of the proportion

Sampling distribution of the proportion is the probability distribution of all possible values of the
sample proportion ( P ).
If necessitates the understanding of the properties of sampling distribution of the proportion
( P ): the mean value of ( P ), standard deviation of ( P ) and the shape or form of the sampling
distribution of ( P ).
Properties of the sampling distribution of the proportion ( P ).
1. The expected value of the sample proportion E( P ) is equal to the population
proportion, P.
Symbolically: E ( P ) = P

Where E ( P ) = is the expected value of the random variable ( P ).

P = is the population proportion
From the above example,
E ( P ) = Pr( P1 ) P 1 + Pr( P2 ) P 2 + Pr( P3 ) P 3
0 1 2
= 0.1( ) + 0.6( ) + 0.3( )
3 3 3
= 0 + 0 .2 + 0 .2
= 0 .4
Thus, E ( P ) = P
2. Just as with the standard deviation of the sample means ( δ ), the standard deviation of
x
the sample proportion ( δ ) also depends on whether the population is finite or infinite.
P
It follows that the standard deviation of the sample proportion is:
N −n p ( 1 − P)
δP = --- for finite population (i.e., n > 0.05 N)
N −1 n
p ( 1 − P)
δP = --- for finite population (i.e., n ≤ 0.05 N)
n
Where δ P is the standard deviation of ( P )
P is any given population proportion.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 17

3. When the population from which the samples are drawn is normal, the form of
the sampling distribution of the sample proportion is also normal. In the other
case, when the population is non-normal or unknown applying the central limit
theorem the sampling distribution of the sample proportion ( P ) can be
approximated by normal probability distribution provided that these two
conditions are fulfilled:
np ≥ 5 and n (1 – p) ≥ 5
Where n is the sample size, and
P is any given population proportion.
Examples:
1. A population proportion is 0.4. A simple random sample of size 200 will be taken and the
sample proportion P will be used to estimate the population proportion. What is the
probability that the sample proportion will be with in ± 03 of the population proportion?
Solution
Given: P = 0.4 n = 200
Required: P (0.37 < P < 0.43)?
To solve this problem, first let’s compute σ P .
p ( 1 − P) 0 .4 ( 0 .6 )
δP = = ≈ 0.035
n 200
P1 − P 0.37 − 0.4 ,For P 1=0.3
Z1 = =
δP 0.035

− 0.03
Z1 = = − 0.857
0.035

-0.857 0 0.86

P (-0.86 < Z < 0.86) = P (0.37 < P < 0.43)

= 2 (0.3051) = 0.6102.
2. A corporation receives 100 applications for a position from recent college graduates in
Economics. Assuming that these applicants can be regarded as a random sample of all such
graduates, what is the probability that between 25% and 35% of them are women if 30% of all
recent college graduates in Economics are women?

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 18

 Solution:
Denote by p x the proportion of women in the sample of n =100 applicants. The population
proportion of women in the sample of n =100 applicants is
p =0.30. Then we have to find
0.25 − p PX − p
P(0.25< p X <0.35) = P ( < )
σP σP
Where
p (1 − p ) (0.3)(0.7)
σP = = = 0.046
n 100
The required probability is then
0.25 − 0.30 PX − P 0.35 − 0.30
P( < < ) = P (−1.09 < Z < 1.09)
0.046 σP 0.046
Where, to a good approximation, the random variable Z has a standard normal distribution.
We have
P (0.25< PX <0.35) = Fz(1.09) – Fz(-1.09)
= Fz(1.09) – [1-Fz(1.090]
=2Fz(1.09)-1
Where Fz(z) is the cumulative distribution function of the standard normal random variable.
From a table Fz(1.09) = 0.8621, so that
P (0.25< PX <0.35) = 2(0.8621) -1 = 0.7242

5.3.3. Sampling distribution of Variance

In our introduction chapter we discussed about inferential statistics which is concerned with the
problem of making inference about a population parameter based on sample information. One of
the population parameter that will be inferred from the sample information is the population
variance. We now turn our attention to the population variance distribution from sample
information.
Suppose that a random sample of n observations is draw from a population with unknown mean
µx and unknown variance σ 2x, the sample members being denoted as X1, X2, …, Xn. Now, the
population variance is the expectation.
σ 2x = E [X - µx ]2
So that an obvious quantity to look at would be the average of (Xi - µx)2 over the n sample
members. However, the populations mean µx is unknown, so that in practice this quantity can not
be calculated. It is natural then to replace the unknown µx by the sample mean X , and to
consider the average of (Xi- X ) 2. In fact, as we noted in chapter 2, the sample variance is defined
as:

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 19

 n

∑
i = 1
( Xi − X ) 2

= ≤
2
S x
n − 1
, for n 30

or

∑
i = 1
( Xi − X ) 2

= >
2
S x
n
, for n 30

Definition
Let X1, X2, …, Xn be a random sample from a population. The quantity
n
n

∑ ( Xi
i =1
− X )2
= n ≤ 30
2
S x
n −1
, for

or

∑ ( Xi
i =1
− X )2
= n > 30
2
S x
n
, for
2 2
S is called the variance. [Again we distinguish between the random variable S x and
x
specific values it can take. Thus, if the actual sample observed is X1, X2, …, Xn, then
2
the realization of S x
is by using the above formula.

Its square root, χ10 S x is called the Sample Standard Deviation.

2

At the first, the use of (n – 1) rather than n as the divisor in our definition of the sample variance
may be rather surprising. The motivation for this formulation is that, if the sample variance is
defined in this way1, it can be shown that the mean of its sampling distribution is the true
population variance; that is,
E (Sx) = σ
2 2
x

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 20

1
Proof:
Here, we want to proof that the sampling distribution of the sample variance is the population
variance. We begin by finding the expectation of the sum of squares of the sample members
about their mean, i.e. the expectation of
n n

∑ ( X i − X ) 2 = ∑ [( X i − µ X ) − ( X − µ X )] 2
i =1 i =1
n
= ∑ [( X i − µ X ) 2 − 2( X − µ X )( X i − µ X ) + ( X − µ X ) 2 ]
i =1
n n
= ∑ ( X i − µ X ) − 2( X − µ X ) ∑ ( X i − µ X ) + ∑ ( X − µ X ) 2
2

i =1 i =1
n
= ∑ ( X i − µ X ) 2 − 2n ( X − µ X ) 2 + n( X − µ X ) 2
i =1
n
= ∑ ( X i − µ X ) 2 − n( X − µ X )
i =1
Taking expectations then gives
n n
E[∑ ( X − X ) 2 ] = E[∑ ( X i − µ X ) 2 ] − nE[( X − µ X ) 2 ]
i =1 i =1
n
= ∑ E[( X i − µ X ) 2 − nE[( X − µ X ) 2 ]
i =1

Now, the expectation of each ( ( X i − µ X ) 2 is the population variance σ 2 X , and the expectation
of ( ( X − µ X ) 2 is the variance of the sample mean, i.e. σ 2 X n .Hence, we have
n
nσ 2 X
E[ [∑ ( X i − X ) 2 ] = nσ 2 X −
= (n − 1)σ 2 X
i =1 n
Finally, for the expected value of the sample variance, we have
1 n
∑
n − 1 i =1
(X i − X )2 ]
n
1
[ E [
E ( S x ) = E [ n − 1 i =1
2 ∑ (X i − X )2 ]

1
= .(n − 1)σ 2 X
n −1
=σ x
2

This is the result we set out to establish.

The conclusion that the expected value of the sample variance is the population variance is quite
general. However, in order to characterize further the sampling distribution we need to know
more about the underlying population distribution. In many practical applications the assumption
that the population distribution is normal is not unreasonable. In this case it can be shown that
the random variable

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 21

 n

(n − 1) S 2 ∑ (X
i =1
i − X )2
= S x2 =
X

σ σ
2 2
X
x

has a distribution known as the χ distribution (Chi-Square distribution) with (n- 1) degrees of
2

freedom.(N.B: The chi-square distribution with v degrees of freedom is the distribution of the sum of
squares of v independent standard normal random variables.)
The chi-square family of distributions is frequently employed in statistical analysis. The
distributions are defined only for positive values of a random variable, which is appropriate in the
present context since a sample variance can not be negative. The density function, which is
illustrated in the figure below which is asymmetric. A specific member of the chi-square family is
characterized by a single parameter, referred to as the number of degrees of freedom, for
which the symbol v is typically used. If a random variable has a χ 2 distribution with v degrees of

χ
2
freedom, it will be denoted . The mean and variance of this distribution are equal to the
v
number of degrees of freedom and twice the number of degrees of freedom, respectively; that is,

Figure 5.2 probability functions of the chi-square distribution with v= 4, 6, and 8 degree
of freedom

χ χ
2 2
E( ) = vi Var ( ) = 2v
v v

χ
2
In the present context, the random variable (n – 1) S2x/ σ 2
x has (n-1) distribution, so that
v

(n − 1) S 2 x
its mean is E ( ) = n −1
σ 2X

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 22

Hence, we have
(n − 1) E ( S 2 x )
= n −1
σ 2X

So that
E ( S x2 ) = σ x2

As before. To get the variance of S x2 , we have

2σ X
2

Var ( S x2 ) = = 2(n-1)
(n − 1)
2σ x2
So that Var ( S x2 ) =
(n − 1)
The properties of the X2 distribution can therefore be used to find the variance of the sampling
distribution of the sample variance. (It should be repeated that this result holds only when the parent
population is normal)

χ
2
The parameter v of the distribution is known as the number of degrees of freedom. To
v
understand this terminology, let us look at the sample variance. It involves the sum of the squares
of the quantities: ( X 1 − X ), ( X 2 − X ) K ( X n − X )

Thus, these n pieces of information are employed to calculate the sample variance. However,
they are not independent pieces of information, since they must sum to 0 (as follows from the
definition of X). Hence, if we know any (n – 1) of the ( X i − X ) , we can calculate the other one
from the first (n-1). For example, since

∑ (X
i =1
i − X) = 0

it follows that:
n −1
Xn - X = ∑ (X
i =1
i − X)

The n quantities (Xi – X ) are equivalent to a set of (n-1) independent pieces of information. The
situation can be thought of as follows: we want to make inference about the unknown σ 2x. If the
population mean µx were known, this inference could be based on the sum of squares of
(X1 – µx); (X2 – µx); …. ; (Xn – µx)
These quantities are independent of one another, and we would say that we have n degrees of
freedom for the estimation of σ 2x. However, since the unknown population mean must be
replaced in practice by its estimate X , one of these degrees of freedom is used up and we are left
with the equivalent of (n -1) independent observations for use in making inference about the
population variance. It is then said that (n- 1) degrees of freedom are available.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 23

We frequently need to find values of the cumulative distribution function for a χ 2 random
variable. Such problems are often phrased in terms of the determination of cutoff point
corresponding to particular specifies probabilities. For instance, if a random variable has a χ10
2

distribution we may require the number K for which

P( χ10 < K) = 0.90

2

Or, equivalently,
P ( χ10 >K) = 0.10
2

The distribution function of the chi-square random variable is available in the Appendix of many
statistical books and from these tables these cutoff points can be read directly. For the χ10
2

random variable, it can be seen from a table that the P( χ10 >K) = 0.10, then K= 15.99. This
2

probability is shown as an area under the density function of the random variable in the following
Figure.

Figure: probability (0.9) that a chi-square random variable with 10 degrees of freedom is less than
15.99
Sampling distribution of the sample variance

2
Let S x denote the sample variance for a random sample of n observations from a
population with variance σ 2x. Then
has mean σ 2x; that is,
2
i. The sampling distribution of S x

E ( S x ) = σ 2x
2

ii. The variance of the sampling distribution of S2x depends on the underlying
population distribution. If that distribution is normal, then
2 2σ 4 X
Var ( S x ) =
n −1
iii. If the population distribution is normal, then
(n − 1) S 2 x
is distributed as χ 2 ( n −1)
σ 2
X

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 24

Suppose that we take a random sample from a population and want to make some inferential
statements about the population variance. Given an assumption of normality in the underlying
population, the chi-square distribution can be used; as illustrated in the following example.

Example:
A manufacturer of canned peas is concerned that the mean weight of the product be close to the
advertised weight. In addition, he does not want too much variability in the weights of the cans of
peas; otherwise, a large proportion will differ markedly from the advertised weight. Assume that
the population distribution of weights is normal. If a random sample of twenty cans is checked,
find the numbers K1, and K2 such that
S 2x S 2x
P( 2 < K1) = 0.05; P ( 2 > K1) = 0.05;
σ X σ X

We have
S 2x  (n − 1) S x2 
0.05 = P ( 2 < K1) , P  < (n − 1) K 1 
σ X  σx
2

= P[ χ ( n −1) < (n-1) K1]
2

Where n = 20 is the sample size and X2 (n-1) is a chi-square random variable with
(n – 1) = 19 degrees of freedom. Then
0.05 = P( χ 192 < 19K1) or 0.95 = P ( χ 192 > 19K1)
From χ 2 table, we therefore have

19K1 = 10.12
So that K1 = 0.533
The conclusion then is that the probability is 0.05 that the sample variance will be less than 53.3%
of the population variance.
We also require the number K2 such that
S 2x
0.533 = P 2 > K2)
σ X

Equivalently, we have
 (n − 1) S x2 
0.05 = P  > (n − 1) K 2 
 σx
2

= P[ χ ( n −1) > (n-1) K2]
2

Hence, since n = 20,

0.05 + P( χ 192 > 19K2)
Then, from X2 table, it follows that
19K2 = 30.14
So that K2 = 1.586
This implies that the probability is .05 that the sample variance will be more than 58.6% bigger
than the population variance.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 25

These probabilities are shown in the figure below as areas under the probability density function
of the χ 192 distribution.
It should be emphasized that the technique of the above example is less universally applicable in
practice than are those of earlier sections of this chapter. The assumption that the distribution of
the population being sampled is normal is more critical here. We have seen how probability
statements can be made about both the sample mean and sample variance when sampling from a
normal distribution. However, the latter will typically be far more affected than the former by any
departures from normality in the distribution being sampled. In making probability statements
about the sample mean, the Central Limit Theorem ensures that, for moderately large samples,
modest departures from normality in the sampled population have only minor influence on the
validity of any derived probability statements. As a result, we say that inference based on the
sample mean is robust to departures from assumed normality in the parent population, whereas
inference based on the sample variance is not.
Nevertheless, it often happens in practice that the population variance is of direct interest to an
investigator. It must be kept in mined that, if only a moderate number of sample observations are
available, serious departures from normality in the parent population can severely invalidate the
conclusions of analyses based on the technique described in this section. The cautions analyst will,
therefore, be rather tentative in making inference in these circumstances.

Figure: probability is 0.05 that a chi- square random variable with 19 degrees of freedom is less
than 10.12, and also that this random variable is bigger than 30.14

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 26

 CHAPTER SIX
STATISTICAL INFERENCE ESTIMATION
Learning objective

To show how point estimates can be calculated for the population parameters.

To describe how confidence intervals are computed for the population parameters.

To indicate how you can determine how large a complete must be in order for an
estimate to have a desired level of accuracy.

One of the principal objectives of statistical investigation is to make reasonable estimates. The
over all objective of this chapter is to describe how statisticians or professionals that use
statistical tools go about doing statistical estimates.

Introduction
Every one makes estimates. When you are ready to cross a street, you see a car approaching to
you on the street; you estimate the speed of any car that is approaching, the distance between
you and that car, and your own speed. Having made these quick estimates, you decide whether
to wait, walk, or run.
So far we have covered discussions on the concepts of probability theory and sampling
distribution that forms foundation for statistical inference. In this chapter we begin to explore the
possibility of making inferential statements about a population, based on the information
contained in a random sample.

Statistical Inference: - is a branch of statistics which is concerned with using probability

concepts to deal with uncertainty in decision making. Statistical inference is based on estimation
and hypothesis testing, and any inference drawn about the population will, of necessarily, is based
on sample Statistics i.e. on functions of the sample information. The choice of appropriate
statistics will depend on which population parameter is of interest. The true parameter will be
unknown, and one objective of sampling could be to estimate its value, which we shall discuss in
this and the following chapter.

Estimation: - is a method that enables us to estimate, with reasonable accuracy, the population
parameter. In reality, to calculate the population parameter is extremely difficult or an impossible
goal, hence we need to make estimates and some times make a statement about the error that
will likely accompany the estimate.
The procedure of marking estimation is to have random sample of size n from the know
probability distribution, compute sample statistics and use it as an estimate of the population
parameters. This procedure of estimation can be categorized in to two: Point Estimation and
Interval Estimation.

6.1 Point Estimation

Point Estimate is a single number, which is used to estimate an unknown population parameter.
If, a research head of the national bank of Ethiopia (NBE) would make a point estimate saying,
“our current date indicates that the country is growing at 7% on average per year.”

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 27

 Some Terminologies of Estimation
An estimator: - is a sample statistics used to estimate a population parameter. And it can be
defined as an estimator of a population parameter is a random variable that depends on the
sample information, and whose realizations provide approximations to this unknown parameter.
Example: The sample mean, can be an estimator of the population mean,µ.
An estimate: - is a specific observed value of a statistic or it is a specific realization of that
random variable. We can get an estimate by taking a sample and computing the value taken by
our estimator in that sample.
To clarify the distinction between the terms Estimator and Estimate, consider the estimation of
the mean income of all families in a neighborhood, based on a random sample of the mean
income of all families in a neighborhood, based on a random sample of twenty families. It seems
reasonable to base our conclusions on the sample mean income, so we say that the estimator of
the population mean is the sample mean. Suppose that, having obtained the sample, we find that
the average income of the families in the sample is $29,356. Then the estimate of the population
mean family income is $29,[Link] illustration suppose that we want to compute the mean
age of a student from a sample of this section and find it to be 19 years. If we use this specific
value to estimate the age of students in the campus, the value 19 years would be an estimate.

6.1.1 Point Estimates for population Parameters

Point estimates for population parameters based on certain criteria, staticians frequently use to
choose among estimators. The following are the standard estimators used by staticians to
estimate these parameters.
Sample Mean X : - is the most common estimator of the population mean. The sample mean
( X ) is unbiased and consistent. Moreover, it can be shown that if the population is normal the
sample mean is the most efficient unbiased estimator available. For these reasons the sample
mean is generally the preferred estimator of the population mean.
Sample Variance ( σ 2 ) & Standard Deviation ( σ ):- the sample variance an unbiased and
consistent estimator of the population variance. It is relatively efficient as compared to other
estimators. Its equal root, the sample standard deviation, is generally used as an estimator of the
population standard deviation is also relatively efficient.

Sample proportion P : - is an unbiased, consistent, and relatively efficient estimator of the

population proportion. Due to this fact, it is generally the preferred estimator of the population
proportion.

6.1.2 Computing Point Estimates

The sample mean( X ) formula:
n

∑X
n =1
i
X= Where ( X ) is sample mean
n
Σ is summation
Xi values of random variables
n is sample size
The sample variance and standard deviation

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 28

 n

∑ (X
n =1
i − X )2
S2 = ------- Sample variance formula
n −1
n

∑ (X
n =1
i − X )2
S= ------ Sample standard deviation formula
n −1
The Sample Proportion P
It is calculated by taking elements in the sample that have the same characteristics; we can use
P this estimator of P
Example
Price- earnings ratios for a random sample of ten stocks traded on the Addis Ababa stock
Exchange on December 27 2006 were:
10; 16; 5; 10; 12; 8; 4; 6; 5; 4
Find point estimates of the population mean, variance and standard deviation and of the
proportion of stocks in the population for which the price – earning ratio exceeded 8.5.
Solution
To find the first three of these sample quantities, we show the calculations in tabular form;

i Xi X 2i
1 10 100
2 16 256
3 5 25
4 10 100
5 12 144
6 8 64
7 4 16
8 6 36
9 5 25
10 4 16
Sum 80 782

We then have
n =10; ∑ X i = 80 ; ∑ X i = 782
2

Hence, the sample mean is

1 80
X = ∑ Xi = =8
n 10
This is our point estimate of the population mean.
A point estimate of the population variance is provided by
1
Sx = (∑ i X i − n X 2 )
2 2

n −1

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 29

 782 − (10)(8) 2
= 15.78
9
For the population standard deviation, the point estimate is
S X = S 2 = 15.78 = 3.97
Finally, in the sample, the number of stocks for which the price –earning ratio exceeds 8.5 is
x = 4. Hence, our point estimate of the population proportion is
X 4
PX = = = 0 .4
n 10
Shortcoming of Point Estimate
A point estimate is always insufficient, because it is either right or wrong. If you take the earlier
example that the research head of NBE states the point estimate that Ethiopia is growing on
average at 7% per year, is wrong. You do not know how wrong it is and you can not be certain
of the estimate’s reliability. If you learn that it is off by only 0.5%, you would accept 7% as a good
estimate of the country’s pace of growth rate. But if the estimate is off by 3%, you would reject
it as an estimate of the country’s growth rate condition. Thus, a point estimate is much more
useful if it is accompanied by an estimate of the error that might be involved.
Note: Estimation of population parameter by using a point sample statistics has some problems;
since we are estimating a large population parameter by using a single number which is either
false or true. As a result the choice of point estimator has been based on intuitive plausibility.
Meaning we have to consider various desirable properties of point estimators. This provides a
framework within which a particular choice can be evaluated and alternatives examined. At the
outset it must be stated that no single mechanism exists for the determination of a uniquely point
estimator in all circumstances. What is available instead is a set of criteria under which particular
estimators can be evaluated. We will see at the end of this chapter some of these desirable
properties.

Interval Estimation
An Interval Estimate is a range of values used to estimate a population parameter. It indicates the
error in two ways: By the extent of its range and the probability of the true population
parameter lying with in that range. Instead of relying on the point estimate alone, we may
construct an interval around the point estimator, say with in two or three standard error of the
mean on either side of the point estimator such that this interval has, for instance 0.95 probability
of including the true parameter value.
Assume that we want to find out how “close” is and estimator X to µ. For this purpose, we try
to find out two positive values U and X such that the probability that the random variable
(X − U and X + U ) contains the true µ is 1 - α. i.e.
P( X − U ≤ µ ≤ X + U ) = 1 − α
0 ≤ α ≤ 1 and 0 ≤ 1 − α ≤ 1
Where: α is the level of significance
1 – α: is confidence coefficient
Such an interval is known as confidence Interval. The random variables ( X − U ) and
( X + U ) are the lower and the upper confidence limit (or critical values) respectively.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 30

6.2.1Calculating Confidence Interval Estimate of a Population Mean:
Normal Population with known δ
Suppose we have a normal population whose mean and standard deviation are µ and δ then the
sampling distribution of the mean will be normal with mean of µ X = µ and
δ X −µ
δX = . For sampling distribution of the mean the standard normal variable is Z = ,
n δX
then the interval estimate of a population mean (the confidence interval for a population mean) is:
X − Zα δ X ≤ µ X
≤ X + Zα δ X .
2 2

Where Z α is the value of standard normal variable that is exceeded with a probability of α /2
2

or Z α is Z value providing an area of α/2 in the upper tail of the standard normal probability
2
distribution.

Example 1:
A normal infinite population has a standard deviation of 10. A random sample of size 25 has a
mean of 50. Construct a 95% confidence interval of the population mean?
Solution:
Given δ = 10 n = 25
χ = 50 95% confidence interval.
Required: X − Z α δ X ≤ µ X
≤ X + Zα δ X ?
2 2
1st find the standard error of the mean i.e.,
δ 10 10
δx = = =2
n 5 5
Since the population is infinite.
2 We have that α = level of significance = 0.05
nd

1 − α = confidence coefficient = 0.95

α 0.05
Find = = 0.025, the
2 2
P ( 0 to Z α ) = Z 0.025 = 1.96
2

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 31

3rd the confidence interval estimation for the population mean (µ) is:
χ − Zα / 2 δ χ ≤ µ ≤ χ + Zε / 2 δ χ
50 - (1.96) 2 ≤ µ ≤ 50 + (1.96)2
46.08 ≤ µ ≤ 53.92
P (46.08 µ ≤ 53.92) = 0.95.
In this case we may say that “we are 95% confident that the population mean lies with in 46.08
and 53.92.” This statement does not mean that the chance is 0.95 that the population mean of all
the random variables falls with in the interval established from this one sample. In stead, it means
that if we select many random samples of the same size and if we compute a confidence interval
for each of these samples, then in about 95 percent of these cases, the population mean will lie
with in that interval.

Values of Z α /2 for the most commonly used confidence

levels

Confidence α α /2 Z α /2
Level

90% 0.1 0.05 1.645

95% 0.05 0.025 1.96
99% 0.01 0.005 2.576
Example 2:
The mean annual income of Ethiopian Air lines Workers (EAL) workers is supposed to be
24,000 Birr. Assume that this estimate was based on a sample of 250 airline workers and the
population standard deviation was 5000 Birr.
a) Compute the 95% confidence interval for the population mean?
b) Construct the 90% confidence interval for the population mean?

Solution:
Given χ = 24,000 Birr n = 250
δ = 5,000 Birr
a) α = 0.05
σ 5000
1 − α = 0.95, σ X = =
n 250
Z α = Z 0.025 = 1.96
2

The 95% confidence interval is given as:

24,000 – (1.96) (316.23) ≤ µ ≤ 24,000 + (1.96) (316.23)
23,380.19 ≤ µ ≤ 24,619.81
b) α = 0.1 α /2 = 0.1/2 = 0.05
1 – X = 0.9
Z α /2 = Z0.05 = 1.64

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 32

 The 90% confidence interval is given as:
24,000 – (1.64) (316.23) ≤ µ ≤ 24,000 + (1.64) (316.23)
23,481 ≤ µ ≤ 24,518.62
δ δ
χ − Zα ≤ µ χ Zα
2 n 2 n
Note that:

A narrower confidence interval is more precise

Larger samples give more precise estimates

Small variance leas to more precise estimates

Lower confidence coefficients allow us to construct more precise estimate

6.2.2 Calculating Confidence Interval Estimate of a Population Mean:

Normal Population with δ Unknown
When the population standard deviation is unknown, we use the sample standard deviation, S, as
an estimate of δ. The sample standard deviation is given by:

S =
∑ ( Xi − χ ) 2

n −1
Thus, the standard deviation of the sampling distribution of the sample means,
δ X ------- is given by:
S
δX = -------- For infinite population
n
In this case, the construction of confidence interval estimate depends up on whether the sample
size is larger or small:
Case 1. When the sample size is large and unknown δ
(A sample size is large when n ≥ 30)
Confidence interval estimate for population mean (µ) is given by:
S
X ± Zx/2
n
Case 2: when the sample size is small and unknown δ (A sample size is small when
n < 30).
The confidence interval estimate of µ is:
S S S
χ ± Zx/2 , χ − tx/2 ≤ µ χ + tx/2 .
n n n

The t-distribution is a family of similar probability distributions, with a specific

t- distribution depending on a parameter known as the degrees of freedom. As the number of
degrees of freedom in creases, the difference between the t-distribution and standard normal
probability becomes smaller and smaller, and the t-distribution will have les dispersion.
The t-distribution is symmetrical, bell-shaped and has zero as its mean.
The interval estimate of the population mean for the small-sample case with δ unknown is given
S S
by: X − t x / 2 ≤ µ X ≤ X + tx/2 .
n n

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 33

Where t x / 2 - is the t-value providing an area of α/2 in the upper tail of a t-distribution with n – 1
degrees of freedom, and S - is sample standard deviation.

Degrees of freedom: - are the number of values we can choose freely. Assume that we are
dealing with two sample values a and b, and we know that they have a mean of 18, symbolically,
the situation is
A+ B
= 18
2
How can we find what values a and b take on this situation? The answer is that a and b can be
any two values whose sum is 36, because 36 ÷ 2 = 18. Suppose we learn that a has a value of 10.
Then, b will have the value of 26, because if a = 10, then 10 + b = 18 10 + b = 36, ∴ b = 26
This example shows that when there are two elements in a sample and we know the sample
mean of these two elements, we are free to specify only one of the elements, because the other
element will be determined by the fact that the two elements sum to twice the sample mean.
Staticians say, “We have one degree of freedom.”
In our context, the degree of freedom is related to the sample standard deviation. There are n-
values of X i − X , involved in computing ∑ ( X i − X ) 2 , which are X 1 − X , X 2 − X , ……
X n − X , and we know that ∑ ( X i − X ) = 0 for any data set.
Therefore, if we known n – 1 values of, the remaining value can be determined exactly by using
the condition that the sum of X − X value must be zero. Thus, there are n – 1 values degrees
of freedom or n – 1 values that can independently determine. This n – 1 is associated with
∑ (X i − X )2.

Example 1:- In the testing of a new production method, 18 employees were selected randomly
and asked to try the new method. The sample mean production rate for the 18 employees was
80 parts per hour and the sample standard deviation was 10 parts per hour. Provide 90% and
95% confidence intervals for the population mean production rate for the new method, assuming
the population has a normal probability distribution.
Solution: - Given n = 10 χ = 80
S = 10 construct CI for 90% and 95% CI?
S 10
δX = = = 2.36
n 18
At 10% level of significance
t x / 2 = t 0.05 = 6.314
µ = X ± tx/ 2 δ X
µ = 80 ± 6.314( 2.36)
65.1 ≤ µ ≤ 94.9

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 34

 Check your self
Sales personnel for Beer factory are required to submit weekly reports listing the customer
contacts made during the week. A sample of 61 weekly contact reports showed a mean of 22.4
customer contacts per week for the sales personnel. The sample standard deviation was five
contacts. Develop a 95% confidence interval for the sale personnel?

6.2.3 Determining the Sample Size in Estimating Population Mean

Some sampling error will arise because we have not studied the whole population. Whenever
we sample, we always miss some helpful information about the population. If we want a high
level of precision (that is, if we want to be quite sure of our estimate). We have to sample
enough of the population to provide the required information. Sampling error is controlled by
selecting a sample size that is adequate. How is this adequate sample size determined for any
specified level of precision or confidence? A method to determine an adequate sample size is
dealt below.
δ
The confidence interval for the population mean is: X ± Zx/2 . In this case the sampling
n
δ δ
error is Z x / 2 or less, so, let’s designate by E, which is the maximum tolerable
Zx/2
n n
sampling error for some specified level of confidence (1 – α ).
δ
E = Zx/2
n
Squaring both sides
2
 δ 
E =  Z x / 2
2

 n
 Z 2 x / 2δ 2 
E = 
2

 n 
 Z 2 x / 2δ 2 
n =  2

 E 
Z δ 
2

n = x/2 
 E 
Where E is the maximum sampling error at some level of precision (1 – α ).

Example: The CSA of Ethiopia has past data that indicate the interview time for a consumer
opinion study has standard deviation of 6 minutes.
(a) How large a sample should be taken if the authority desires a 98% level of precision
that the mean interval time to be with in 2 minutes or less?
(b) Assume that the sample size recommended in (a) above is taken and that the mean
interview time for the sample is 32 minutes. What is the 98% confidence interval
estimate for the mean interview time for the population?
Solution: Given: δ = 6 confidence level = 0.98 = 1 - α
Level of significance = α = 0.02

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 35

 Z α /2 = Z0.01 = 2.32
 Z . δ   2.32 x 6 
2 2

n =  α/2  =  = (6.96) = 48.44 ≅ 49

2
(a)
 E   2 
(b) µ = X ± Zα /2 δ X
X − Zα δ X ≤ µ X
≤ X + Zα δ X
2 2

 6   6 
32 −  2.32  ≤ µ ≤ 32 +  2.32 ⋅ 
 49   49 
32 - 2 ≤ µ ≤ 32 + 2
30 ≤ µ ≤ 34

Test your self

1. Why Sampling Error arises?
2. If you take a sample that consists of the entire population, what is the sampling error?

6.3 Computing Confidence Interval for Population Proportion

From the previous chapter we know that the sampling distribution of sample proportion P has a
pq
mean of P and a standard deviation of for infinite population.
n
In this section we shall briefly discuss how we construct a confidence interval for the unknown
value of P; we use the estimation P that compute the sample standard error of the proportion
S P and the use S P as an estimator for σ P , finally the interval estimate (CI) of population
proportion will be:

P ± Z α /2 δ P , δ P = S P
P (1 − P )
P - Z α /2 S P ≤ P ≤ P + Zx/2 S P
n
P (1 − P ) P (1 − P )
P - Z α /2 ≤ P ≤ P + Z α /2
n n

Examples
1. A survey conducted by Ethiopia Economic Policy Research Institute (EEPRI) shows that 47% of
all investors included in the sample invested in agricultural sector.
Compute a 95% confidence interval estimate for the proportion of the population investors who
invested in agricultural sector assuming that a sample of size 250 investors was used in the study?
0.4081 ≤ P ≤ 0.5319

Solution:
Given: P = 0.47 n = 250

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 36

 1 - P = 0.53 x = 0.05 Zx/2 = Z0.025 = 1.96

 .47(0.53)   .47(0.53) 
Thus, 0.47 – 1.96 ⋅  ≤ P ≤ 0.47 + 1.96 ⋅
 

 250   250 
0.47 – (1.96 x 0.0316) ≤ P ≤ 0.47 + (1.96 x 0.0316)
Interpretation: of all the investors taken as a sample, the proportion of the investors that
invest in agricultural sectors varies between 40% and 53%.

2. A random sample of 400 members of the labour force in four regional states of Ethiopia
showed 32 week unemployed. Construct a 95% confidence interval for the proportion of
unemployed labor force in there four regions.
Solution:
32
Given: P = = 0.08
400

Interpretation
The unemployed labour force with a confidence interval of 95% varies between 5.3% and 10.67%.

6.3.1 Determining the Sample Size for Estimating Population Proportion

The rational for determining the sample size in estimating the population proportion is the same
as we discussed for the case of population mean estimation. In this case the adequate sample size
for a specified level of precision can be determined as follows:
Given the confidence interval for the population proportion;
P +Z α S P
2

The values Z α S P in confidence interval for the population proportion is the maximum sampling
2
error involved with , α ,a specified level of precision, and let represent it by E.
E = Zα SP
2

P (1 − P
E=
n
P (1 − P )
E = Zα
2
n
P (1 − P )
E 2 = Z 2α
2 n
P (1 − P )
n = Z 2α
2
E2
Note: - In most cases, the desired maximum sampling error (E) is given as 0.01 or less.
In some statistics text books, the sample size is determined by the following way:

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 37

 P (1 − P )
n = Z 2α =
2
E2
In such situation, planning value for the population proportion ‘P’ is required. Practically, this
planning value can be chosen by one of the following procedures:
I) Use the sample proportion from previous sample of the some units
II) Undertake a pilot study to select a preliminary sample of units. The sample proportion
from this sample can be used as the planning value of P.
III) Use judgment or a “best guess” for the value of P.
IV) If none of the preceding alternative apply, use P = 0.5

Example 1. Suppose we want to estimate a population proportion with ± 0.04 and a confidence
coefficient of 90%. What sample size should be taken when P = 0.5.

Solution:
Z α = Z 0.05 = 1.64
Given E = 0.04 2

1 − P = 0 .5
P = 0.5
p (1 − P ) (1.64) 2 (0.5)(0.5)
n = Z 2α . = = 420.25 = 421
2 n (0.04) 2
Check yourself
How large a sample of investors should be used in example 1 above if it is desired to be at 95%
confidence level that the sampling error is 5% or less?

6.4 Confidence interval for difference of means of two normal populations

An important problem in statistical inference deals with the comparisons of the two population
means. As one illustration, a company might receive shipment of the chemical from their two
suppliers and be concerned a bout the difference a bout the means level of then impurity present
in the chemicals from the two sources of supply.

As another example, a farmer may consider the use of the two alternative fertilizers, his interest
being in the difference between the resulting mean crop yields per acre. In order to compare
population mean, a random sample is drawn from the two populations, and inference a bout the
difference between the population means is based on the population results. The appropriate
method for analyzing this information depend on the procedure used in selecting the samples.
We will consider the following two very common sampling schemes:
(a) MATCHED PAIRS: in this scheme, the sample members are chosen in pairs, one from each
population. The idea is that, a part from the factor under study, the members of these
pairs should resemble one another as closely as possible, so that the comparison of
interest can be directly made. For instance, suppose we want to measure the effectiveness
of the speed-reading course. One possible approach would be to record the number of
words per minute read by sample of student before taking the course ,and compare with
the result for the same students after completing the course, and compare with results

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 38

 for the same students after completing the course. In this case each pair of observations
consists of “before” and “after” measurements on a single student.
(b) INDEPENDENT SAMPLES in this scheme, samples are drawn independently from the two
populations of interest, so that the membership of one sample is not influenced by that of
another. In the example of the company that receives shipments of chemical from two
suppliers, we might choose independent random samples of batches from each supplier
and measure the impurity levels of each batch sampled.
Whichever sampling method is used, our objective in this section is to obtain confidence intervals
for the difference between the two population means.

6.4.1 Confidence Intervals Based On Matched Pairs

Suppose, in general, that we obtain sample of n matched pairs of observation, denoted
(x1 , y1 ), (x 2, y 2 ), (x n, y 2 ),..., (x n , y n ),
from populations with means µ x and µ y. Thus, X1, X2, X3…Xn
…. denote the observations from the population with mean µ x and Y1, Y2 Y3,…Yn denote the
observations from the population with mean µ y. The table below shows miles per gallon figures
obtained for a random sample of eight cars from each of two different models. The sample cars
were paired, and each member of a particular pair was drive over the same route by the same
driver, so that variability between drivers and routes could be eliminated from the comparisons.
d
The table also shows the differences, i , between these miles per gallon figures. These differences
represent a random sample from a population whose mean is ( µ x - µ y.), the difference between
the population means for two car models. Based on the information in the table, the sample
mean and variance of the differences in achieved miles per gallon can be calculated, we have, for
the mean,
1 n
∑ d i 1 (6.2 ) = 0.775
n
d = i =1 =8
And, for the variance,
Table: Miles per gallon achieved by matched pairs of cars

X-cars Y- Differences
cars
I xi yi di d 2i

1 19.4 19.6 -0.2 0.04

2 18.8 17.5 1.3 1.69
3 20.6 18.4 2.2 4.84
4 17.6 17.5 0.1 0.01
5 19.2 18.0 1.2 1.44
6 20.9 20.0 0.9 0.81
7 18.3 18.8 -0.5 1.25
8 20.4 19.2 1.2 0.44
Sums 6.2 10.52

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 39

 S2d= 1  ∑ d 2 − −2 
n

nd 
n − 1  i =1 i 

1
[10.52 − (8)(0.775)2]0.816
=
7

So that the observed sample standard deviation is

Sd = 0.816 = 0.903
We are now in the position of requiring a confidence interval for a population mean
(µx -µy), given a random sample (the value of the differences di) from that population. If the
population distribution is assumed to be normal, the procedure developed under confidence
interval computation for mean is immediately applicable because the differences in matched
pairs constitute a random sample from a population whose mean is the quantity we are
trying to estimate.

Confidence intervals for difference between means: matched pairs

Suppose we have a random sample of n matched pairs of observations from
distributions with means µx and µy. let d and s denote the observed sample mean and
standard deviation for the n differences di=xi –yi. if the population distribution of the
differences is assumed to be normal, then a 100(1- α ) % confidence interval for (µx-µy)
is given by

t n −1.a Sd t n −1 , a S d
d+ 2 2
< µ X − µY < d +
n n
Where tn-1,α/2= is that number for which
a
P(tn-1>tn,a/2)= 2
and the random variable tn-1 has a student’s distribution with (n-1) degrees of freedom.

In the automobile mileage example, we have found

d = 0.775; Sd=0.903; n =8
For a 99%confidance interval, so that
tn-1,a/2= t7,.005=3.499 from a table.
Hence, we substitute to obtain the 99% confidence interval for the difference between the
population means:

t n −1.a Sd t n −1 , a S d
d+ 2
< µ X − µY d + 2
n n

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 40

that is
(3.499)(0.903) (3.499)(0.903)
0.775 − < µ X − µ Y < 0.775 +
8 8
Or -0.342 < µx-µy <892

We therefore find that, based on the data of the above table, a 99% confidence interval for the
difference in population mean miles per gallon for these two types of automobile ranges from -
0.342 to [Link] the interval includes 0, the sample evidence against the conjecture that the
population means are the same in not very strong.

6.4.2 Confidence intervals based on independent samples

We now consider the case where independent samples, not necessarily of equal size, are taken
from the two populations of interest.
Suppose we have a random sample of nx observations from a population with mean µ X and
variance σ X , and an independent random sample of ny observation from a population with
2

mean µY and variance σ Y . Let the respective sample means be X and Y .

2

As a first step, we examine the situation where the two population distributions are normal with
known variances. Since the object of interest is the difference between the two population
means, it is natural to base inference on the difference between the corresponding sample means.
This random variable has mean
E ( X - Y ) = E ( X ) – E ( Y ) = µ X + µY
and, since the samples are independent, variance
σ 2X σ 2Y
+
Var( X - Y ) = Var( X ) +Var ( Y ) = n X nY
Furthermore, it can be shown that its distribution is normal. It therefore follows that the
random variable
( X + Y ) − (µ X − µ Y )
Z=
σ 2X σ 2Y
+
nX nY
Has a standard normal distribution. An argument parallel to that of confidence interval
computation for the mean of a normal population can then be used to obtain confidence
intervals for the difference between the population means. Since this interval requires
knowledge of true population variance, it is rarely of much direct use.
However, as was the case in confidence interval computation for the mean of a normal
population its range of applicability is greatly extended when the sample sizes are large, as
indicated in the box.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 41

 Confidence intervals for Difference between Means:
Independent Samples (Known Variances or Large Sample Sizes)
Suppose we have independent random samples of nX and nY observations from
normal distributions with means µ X and µY and variances σ x and σ y. If the
2 2

observed sample means are X and Y , then a 100 (1- α ) %confidence interval for (x-y) is
given by
σ 2X σ 2Y σ 2X σ2
( X − Y ) − Za + < µ X − µY < ( X − Y ) + Z a +
2 nX nY 2 nX nY

Za
Where is that number for which
2

a
Za
P(Z> 2 ) = 2 and the random variable Z has a standard normal distribution.
If the sample sizes nX and nY are large, then to a good approximation, a 100(1- α )%
confidence interval for ( µ X - µY ) is obtained by replacing the population variances in the
previous expression by the corresponding observed sample variances S2x and S2y. For large
sample sizes, this approximation will typically remain adequate even if the
population distributions are not normal.

Example1. A professor, teaching two sections of Statistics for Economists course, organized
quizzes differently in the two sections. In one section quizzes, based on pre-assigned reading,
were given on the first day of discussion of a topic. In the other section quizzes were given after
each chapter was completed. A common examination was set for the two sections, which each
contained 40
students. For the first group, the mean score was 143.7, and the standard deviation was 21.2,
while for the second the mean and standard deviation were 131.7 and 20.9.
If we can regard these students as independent random samples from the populations of all
students who might be exposed to these two approaches to quizzes, find a 99% confidence
interval for the difference between the population mean scores.

Solution:
For the pre-quiz group we have
X = 143.7; nx = 40; S2x = (21.2)2 = 449.44
and for the post-quiz group
Y = 131.7; ny = 40; S2y = (20.9)2 = 436.81
Since the sample sizes are quite large, we can use the sample variances in place of the
population variances, in the formula given above, to find confidence intervals for the difference
between the two population means. These intervals then take the form
2
( X − Y ) − Za S 2 X nX + S Y < µ X − µY < ( X − Y ) + Z a S 2 X nX + S 2Y nY
2 nY 2

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 42

Where, for a 99% interval,
Z α /2= Z0.005 =2.575
The interval needed is then
(143.7-131.7) – (2.575) 449.44 / 40 + 436.81 / 40 < µ X - µY < (143.7-131.7) + (2.575)
449.44 / 40 + 436.81 / 40 which is

-0.1< ( µ X - µY ) <24.1
This 99% confidence interval for the difference in population mean scores just includes zero,
suggesting that the evidence in the data against the conjecture that the two population means are
the same is not overwhelmingly strong.

2. Researchers interested in verbal responses to survey questions have found that not only the
responses themselves, but also subjects’ judgment time in responding to the question, contain
useful information. Typically the data analyzed are the reciprocals of judgment time, and are called
the certainty values. Thus, the larger the time taken to produce a response, the smaller the
certainty value. A random sample13 of 143 people were contacted by telephone and asked:
“Assuming that it’s convenient for you, do you expect to get a swine flu shot?” One hundred of
these subjects answered “yes.” Their mean certainty value was 0.76, and the standard deviation
was 0.50. For the 43 subjects answering “no,” the mean certainty value was 1.41 and the standard
deviation was [Link] by µ X the population mean of those answering “yes” and by µY
the population mean for the “no” group, find a 95% confidence interval for ( µ X - µY ).
Solution:
Again, since the sample sizes are large, we can use the sample variances in place of the population
variances and obtain intervals from
σ 2X σ 2Y σ 2X σ2
( X − Y ) − Za + < µ X − µY < ( X − Y ) + Z a +
2 nX nY 2 nX nY
Where: X = 0.76 nX = 100 Sx= 0.50
Y = 1.41 nY = 43 Sy =0.72 and for a 95% confidence interval,
Z α /2= Z0.025 =1.96
The interval is then
(0.50) 2 (0.720 2 (0.50) 2 (0.72) 2
(0.76 − 1.141) − (1.96) + < µ X − µY < (0.76 − 1.41) + (1.96) +
100 43 100 43 or

-0.89< ( µ X - µY ) >-0.41
This interval includes only negative values, indicating that those who say they will get a shot are
less certain of their answers, on the average, than those who say they will not, in the population
at large. The figure shows this confidence interval, together with 80%, 90% confidence intervals
for the difference in the population means.
We now have to consider the case where the sample sizes are not large, and a confidence
interval is needed for the difference between the means of two normal populations based on
independent random samples from the two populations.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 43

In fact, when the population variances are unknown, there is considerable difficulty in attacking
this general problem. However, in one special case, where it can be assumed

In the above figure 80%, 90%, 95% and 99% confidence intervals for difference in population
means based on data of Example 2 that the two population variance are equal, a fairly
straightforward method is available.
Suppose again that we have independent random samples of nx and ny observations from normal
populations with mean µ X -and µY , and that the populations have a common (unknown)
variance σ 2 . Inference about the population means is, as before, based on the difference ( X - Y )
between the two sample means. This random variable has a normal distribution with mean ( X -
Y ) and variance
Var ( X - Y ) = Var ( X ) + Var( Y )
σ2 σ2
= +
nX nY
1 1
= σ 2( + )
n X nY
n X + nY
= σ 2( )
n X nY
It therefore follows that the random variable
( X − Y ) − (µ X + µY )
Z =
n + nY (a)
σ 2( X )
n X nY

Has a standard normal distribution. However, this result cannot be used as it stands because the
unknown population variance is involved. Since this variance is common to the two populations,
the two sets of sample information can be pooled together to estimate it. The estimator used is

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 44

 (nx − 1) S 2 X + (nY − 1) S 2 X
S = 2

(n X + nY − 2)
2 2
Where S x and S y are the two sample variances.
Reporting the unknown σ 2 by its estimator s2X in equation (a) gives the random variable
( X − Y ) − ( µ X − µY )
S2 =
n + nY
S( ( X )
nX nY
It can be shown that this random variable obeys the student’s t distribution with (nx+ny-2) degree
of freedom. Given this result, confidence intervals for the difference between the population
means can be obtained through an argument similar to that used in confidence interval
computation of a normal population.

Confidence intervals for Difference between the Means of two

Normal Populations: Independent Samples, Population Variance Equal
Suppose we have independent random samples of nx and ny observations from normal
distributions with means ux and uy and a common variance. If the observed sample means are X
and Y , and S2x, S2y, then a 100 (1- α ) % confidence interval for ux – uy is given by
nX + nY n + nX
( X − Y ) − tn x + nY − 2,α / 2 S < µ X − µY < ( X + Y ) + tn X + nY − 2,α / 2 S X
nX nY nX nY
Where
(nx − 1) S 2 X + (nY − 1) S 2 X
S = 2

(n X + nY − 2)
t
And n x + nY − 2,α / 2 is that number for which

α
P (t n X + n Y − 2 , > (t n X + n Y − 2 , α / 2
)= 2
Where the random variable tnt +ny-2 has a student’s t distribution with
(nx +ny -2) degree of freedom.

Example:
In a study of the effect of planning on the financial performance of banks, a random sample of six
“partial formal planners” showed mean annual percentage increase in net income of 9.972 and a
standard deviation of 7.470. An independent random sample of nine banks with no formal
planning system had a mean annual percentage increase in net income of 2.098 and a standard
deviation of 10.834. Assuming the two population distributions are normal with the same
variance, find a 90% confidence interval for the difference between their means.
Solution:
We have, with x referring to the “partial formal planners,” and y to those with no formal
planning.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 45

 nx =6; X = 9.972; Sx = 7.470
ny = 9; Y = 2.098; Sy = 10.834
The estimate of the common population variance is then
(nx − 1) S 2 X + (nY − 1) S 2 X
S =
2

(n X + nY − 2)
= (5) (7.470)2 + (8)(10.834)2 = 93.693
13
So that S = 93.693 = 9.680
The interval required is of the form
n X + nY n + nX
( X − Y ) − t n x + nY − 2,α / 2 S < µ X − µ Y < ( X + Y ) + t n X + nY − 2,α / 2 S X
n X nY n X nY
Where, for a 90% confidence interval, α = 0.10, so that
tn x + nY − 2,α / 2 = t13, 0.05 = 1.771 from a table.
Hence, the 90% confidence interval for the difference between the population mean percentage
increases in net incomes is
6+9 6+9
(9.972 − 2.098) − (1.771)(9.680( < µ X − µY < (9.972 − 2.098) + (1.771)(9.680) or -
54 54
1.161 < µ X - µY <16.909
Our 90% confidence interval for the difference between population mean annual percentage
increases in net income for these two groups of banks includes 0. This suggests that the evidence
in the data against the conjecture that the two population means are the same is not strong.

6.5 Confidence Interval for Difference of Two Population Proportions

(Large Samples)
In our discussion of confidence interval computation of population proportion, we derived
confidence intervals for a single population proportion. Often we are interested in comparing
two proportions. For instance, we might want to compare the proportion of football players who
succeed in graduating with the proportion of nonetheless who graduate. In this section we show
how to obtain confidence intervals for the different population proportions when independent
large samples are taken from the two populations.

Suppose that a random sample on nx observations from a populations with proportion Px of

“successes” yields sample proportion P x, and that an independent random sample of ny
observations from a population with proportion Py of “successes” produces sample proportion
P y. Since our concern is with the population difference (Px – Py), it is natural to examine the
random variable ( P x – P y). This has mean
E( P x – P y) = E( P x) – E( P y) = Px – Py
and, since the samples are taken independently, variance

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 46

 Var ( P X − P X ) = Var ( P X ) + ( PY )
………………. (b)

PX (1 − PX ) PY (1 − PY )
= +
nX nY

Furthermore, if the sample sizes are large, the distribution of this random variable is
approximately normal, so that subtracting its mean and dividing by its standard deviation gives a
standard normal random variable. Moreover, for large sample sizes, this approximation remains
good when the unknown population proportions in equation (b) are replaced by the
corresponding sample quantities. Thus, to a good approximation, the random variable
( PX − PY ) − ( PX − PY )
Z=
P X (1 − P X ) P Y (1 − P Y )
[ +
nX nY
has a standard normal distribution.
This result allows the derivation of confidence intervals for the difference between the two
population’s proportions when the sample sizes are large, as shown in the box.

Confidence Intervals for the Difference between Proportions (Large Samples)

Let P x denote the observed proportion of successes in a random sample of nx observations from
a population with proportion px successes, and P y the proportion of successes observed in an
independent random sample from a population with proportion py successes. Then, if the sample
sizes are large, a 100(1 - α ) % confidence interval for (px – py) is given by:
P (1 − P ) P (1 − P X ) P (1 − P ) P (1 − P X )
(P X − PY ) − Z a + < (px – py)< ( P X − P Y ) − Z a +
2
nX nY 2
nX nY
Where za/2 is that number for which
P (z>za/2) = α / 2
and the random variable Z has a standard normal distribution

We now illustrate this procedure with two examples.

Example:
1. In a study of the effectiveness of premiums given as an inducement to open bank accounts, a
random sample of 200 accounts attracted by premiums (either cookware or a calculator)
contained 79% that were retained over a 6-month period. An independent random sample of 200
accounts not attracted by premiums contained 89% that were retained over a 6-month period. If
the respective population proportions are denoted px and py, fine a 90% confidence interval for
(px – py).
Solution:
From the sample information, we have

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 47

 nx = 200; P x = 0.79; nx = 200; P y = 0.89
For a 90% confidence interval, α = 0.10, so that
Za/2= z.05 = 1.645
Substituting these values into the formula for the confidence interval gives:

(0.79)(0.21) (0.89)(0.11))
(0.79 – 0.89) – (1.645) +
200 200
(0.79)(0.21) (0.89)(0.11))
< Px – Py < +
200 200
or
-0.16 < Px – Py < -0.04
The figure below also shows 80%, 95% and 99% confidence intervals for the difference between
the two population proportions. Notice that none of these intervals contains the difference 0 in
population proportions. Thus, the data strongly suggest that accounts attracted by premiums are
less likely to be held for as long as 6 months than those not attracted by premiums.

2. Frequently populations are surveyed by mail questionnaires, and investigations are anxious to
obtain as high a response rate as possible. A questionnaire, printed on a single sheet, front and
back, was sent to a random sample of 220 households, of

This figure shows 80%, 90%, 95% and 99% confidence intervals for difference in population proportions,
using data of Example 1 above.
which 36% responded. The same questionnaire, printed on two sheets, front only, was sent to an
independent random sample of 220 households, and the achieved response rate was 30%. Find a
95% confidence interval for the difference between the two population proportions responding.

Solution:
The sample values are
nx = 220; Px = 0.36; ny = 220; Py = 0.30
For a 95% confidence interval, α = 0.05, and so
Za/2 = Z.025 = 1.96

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 48

Substituting into the formula above then yields the interval
(0.36)(0.64) (0.30)(0.70))
(0.36 – 0.30) – (1.96) +
220 220
(0.36)(0.64) (0.30)(0.70))
<Px – Py < (0.36 – 0.30) + (1.96) +
220 220
That is -0.03 < Px – Py < 0.15
The 95% confidence interval for the difference between these two population proportions
contained zero. This indicates that the evidence in the data against the proposition that the two
population proportions are the same is not very strong.

6.6 Estimating sample sizes for research activities

The ever increasing demand for research has created a need for an efficient method of
determining the sample size needed to be representative of a given population. In the article
“small sample technique,” the research division of the National Education Association of Texas A.
and M. University has published a formula for determining sample size. Regrettably a table has not
been available for ready, easy reference which could have been constructed using the following
formula.
I. Formula for determining Sample size when population size
is known
χ 2 NP(1 − P)
S= 2
d ( N − 1) + χ 2 P (1 − P )
Where S = required sample size
χ 2 = the table value of Chi-square for 1 degree of freedom at the desired
significance level.
N = the population size.
P = the population proportion (assumed to be 0.50 since this would provide
the maximum sample size).
d = the degree of accuracy expressed as a proportion of (0.05).
Here are some selected values of Chi-Square @ degrees of freedom of 1 for desired confidence
levels: 0.10 = 2.71 , 0.05 = 3.84 , 0.01= 6.64, 0.001 = 10.83

II. Formula for determining Sample size when population size

is unknown
RANGE 2
( )
Sample size, S = 2
ACCURACY LEVEL 2
( )
CONFIDENCELREVEL
Where:-Confidence levels is given for selected values of α :

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 49

 α α /2
0.10 level 1.28 1.64
0.05 level 1.68 1.96
0.01 level 2.33 2.58
0.001 level 3.09 3.29

Accuracy Levels = [RANGE ] X [Desired Level of Accuracy expressed as a proportion]

No calculations are needed to use the following table. For example, one may wish to know the
sample size required to be representative of the opinions of 9000 high school teachers relative to
merit pay increase. To obtain the required sample enter the table at N= 9000. The sample size
representative of teachers in this example is 368. This table is applicable to any defined
population.
The relationship between sample size and total population is also illustrated in the figure that
follows. It should be noted that as the population increases the sample increases at a diminishing
rate and remains relatively constant at slightly more than 380 cases.

Table for Determining Sample Size from A Given Population

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 50

 N S N S N S
10 10 220 140 1200 291
15 14 230 144 1300 297
20 19 240 148 1400 302
25 24 250 152 1500 306
30 28 260 155 1600 310
35 32 270 159 1700 313
40 36 280 162 1800 317
45 40 290 165 1900 320
50 44 300 169 2000 322
55 48 320 175 2200 327
60 52 340 181 2400 331
65 56 360 186 2600 335
70 59 380 191 2800 338
75 63 400 196 3000 341
80 66 420 201 3500 346
85 70 440 205 4000 351
90 73 460 210 4500 354
95 76 480 214 5000 357
100 80 500 217 6000 361
110 86 550 226 7000 364
120 92 600 234 8000 367
130 97 650 242 9000 368
140 103 700 248 10000 370
150 108 750 254 15000 375
160 113 800 260 20000 377
170 118 850 265 30000 379
180 123 900 269 40000 380
190 127 950 274 50000 381
200 132 1000 278 75000 382
210 136 1100 285 1000000 384

Note: N is population Size

S is sample size

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 51

Sample Size Vs Total Population
400
300
sample
200
100
0

0 200000 400000 600000 800000 1000000

population

The figure assumes standard Error = 0.05

6.8 Properties of Estimators

There are various methods with which we may obtain estimates of the parameters of economic
relationships. How are we to decide whether an estimate is ‘good’ or whether it is better than
another obtained from a different method?
We need some criteria for judging the goodness of the estimate. The criteria or properties for a
good estimator may be different for small and large samples.

I. Small sample properties

a) Unbiased ness: - An estimator is said to be unbiased if the expected value of
the estimator is equal to the true population parameter, i.e. if E ( X ) = µ .
Biased E ( X − µ ) ≠ 0
Unbiased E ( X − µ ) = 0
Where X is an estimator of µ and µ is a population parameter.
b) Least variance (Minimum variance) - Best Estimator- An estimator is best when
it possesses the smallest variance as compared to any other estimator which
is obtained by various methods.
c) Efficiency, Efficient estimator: - An estimator is efficient when it occupies both
the aforementioned properties (Unbiased ness, Minimum variance)
d) Minimum Mean Square Estimator (MSE):- An estimator is a minimum, MSE, if it has
the smallest mean square error defined as the expected value of the square differences
around the true population parameter.

e) Sufficiency, Sufficient Estimator:-An estimator is said to be sufficient if it utilizes

all the information a sample contains about the true parameter that is it must use
all the observations of a sample
f) Best, Linear and Unbiased Estimator (BLUE):- An estimator is BLUE if it is best,
linear ,and Unbiased.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 52

An estimator is linear if it is a linear function of the sample observation.
For instance, sample mean ( X ) is a linear estimator

X=
∑ X i = 1 ( x + X + X + ... + X )
1 2 3 n
n n
We see that X is a linear function of all observations in the sample.

II. Large Sample Properties (Asymptotic Properties)

a) Asymptotic Unbiased: - An estimator is asymptotically unbiased if
lim E ( X ) = µ
n→∞
b) Consistency: An estimator is consistence if it satisfies the following conditions:
i. The estimator must be asymmetrically unbiased
ii. The variance of the estimator must approach to zero as n approaches
infinity.
lim var( X ) → 0
n→∞
c) Asymptotic Efficiency:- An estimator is asymptotically efficient if:
i. The estimator is consistent
ii. The estimator has a smaller asymptotic variance as compared to
any consistent estimator.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 53

 CHAPTER SEVEN
HYPOTHESIS TESTING
7.1 Concepts of Hypothesis Testing

When a sample is drawn from a population, the evidence obtained can be used to make
inferential statements about the characteristics of the population. As we have seen, one
possibility is to estimate the unknown population parameters through the calculation of point
estimates or confidence intervals. Alternatively, the sample information can be employed to
assess the validity of some conjecture, or hypothesis, that an investigator has formed about the
population. Examples of situations of this kind are as follows:
1. A manufacturer who produces boxes of cereal claims that, on average, the contents weigh at
least 20 ounces. In order to check this claim, the contents of random sample of boxes can be
weighed, and inference based on the sample results.
2. A company receiving a large shipment of parts may want to accept delivery only if no more
than 5% of the parts is defective. The decision on whether to take delivery might be based on
a check of a random sample of these parts.
3. An instructor is interested in the value of regularly administered quizzes in a statistics course.
She uses these quizzes in one section of the course, but not in another. At the end of the
course she compares the average performances of students in the two sections on the final
examination, in order to check here hypothesis that the quizzes raise average performance.
4. a political scientist wants to know if a tax reform proposal appeals equally to men and
women. In order to check whether this is so he obtains the options of randomly selected
samples of males and females.

The examples given here have a common theme. A hypothesis is formed about some population,
and conclusions about the merits of this hypothesis are to be formed on the basis of sample
information. In this section we introduce a general framework for approaching such problems.
Specific procedures are then developed in the following section.

To keep our discussion quite general, let us denote the population parameter of interest (for
example, the population mean, variance, or proportion) by =. Suppose that some hypothesis has
been formed about this parameter, and that this hypothesis will be believed unless sufficient
contrary evidence is produced. This can be thought of as a maintained hypothesis. In the language
of statistical hypothesis testing, it s called a null hypothesis.

For example, we weight, in the absence of evidence to dispute it, believe the manufacturer’s claim
that, on average, the contents of its boxes of cereal weigh at least 20 ounces. When sample
information is collected, this hypothesis is put in jeopardy, or tested. If the hypothesis is not true,
then some alternative must be true and, in carrying out a hypothesis test, the investigator
formulates an alternative hypothesis against which the null hypothesis is tested. For the cereal
manufacturer we could test the null hypothesis that the mean contents weight is at least 20
ounces against the alternative hypothesis that the mean weight is less than 20 ounces. The null
hypothesis will be denoted Ho and the alternative hypothesis H1.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 54

A hypothesis, whether null or alternative, might specify just a single value, say X 0 for the
population parameter µ 0 In that case the hypothesis is said to be simple. A convenient shorthand
notation would read, for example,
Ho; µ 0 = 0
For “the null hypothesis is that the population parameter µ 0 = is equal to the specific value X 0 .”
for instance, in the fourth example listed previously, the political scientist might begin his
investigation with the simple null hypothesis that the difference between the proportions of men
and women in the population who favor the tax reform proposal is 0.

On the other hand, a hypothesis could specify a range of values for the unknown population
parameter. Such a hypothesis is said to be composite. And will hold true for more than one
value of the population parameter. For instance, the null hypothesis that mean weight of boxes of
cereal is at least 20 ounces is composite. The hypothesis is true for any population mean weight
greater than or equal to 20 ounces.

In many applications, a simple null hypothesis, say Ho; µ 0 =o, is tested against a composite
alternative. In some cases, only alternatives on one side of the null hypothesis are of interest. For
example, we might want to test this null hypothesis against the alternative hypothesis that the
true value of µ 0 is bigger than X 0 , which we can write.
H1:> 0
Conversely, the alternative of interest might be
H1:< o
Such alternative hypotheses are called one-sided alternatives. Another possibility is that we
want to test this simple null hypothesis against the very general alternative that the true of = is
something other than =o, that is.
H1; ≠o
This is referred to as a two-sided alternative.
The specification of appropriate null and alternative hypotheses is problem specific To illustrate,
we return to our earlier examples:
1. Let = denote the population mean weight (in ounces) of cereal per box. The null
hypothesis is that this mean is at least 20 ounces. So we have the composite null
hypothesis
H0; ≥ 20
The obvious alternative is that the true mean weight is less than 20 ounces, that is,
H1; < 20
2. A company intends to accept delivery of parts unless it has evidence to suspect that more
than 5% are defective. Let P denote the population proportion of defectives. The null
hypothesis here is that this proportion is at most 0.05, that is,
H0= = < 0.05

On the basis of sample information, this hypothesis is tested against the alternative
H1= = > 0.05
The null hypothesis, then, is that the shipment of parts is of adequate quality overall, while
the alternative is that it is not.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 55

 3. Suppose an instructor conjectures that the regular administration of quizzes in class
makes no difference to the average scores on the final examination. Let = denote the
difference between the population mean scores for section with and without regular
quizzes. The null hypothesis is then the simple null hypothesis
H 0; = = 0
However, she may suspect the possibility that quizzes lead to an increase in
average performance, and thus would want to test the null hypothesis against
the alternative hypothesis.
H 1; = > 0
4. A political scientist might hold, as a working hypothesis, the view that the tax reform
proposal is equally appealing to men and women. If = is the difference between the two
population proportions in favor of the proposal, then the null hypothesis is
H 0; = = 0
If the political scientist has no good reason to suspect the bulk of support comes from
one population rather than the other. This null hypothesis would be tested against
the two-sided alternative hypothesis.
H 1; = ≠ 0
Having specified a null and alternative hypothesis and collected sample information, a decision
concerning the null hypothesis must be made. The two possibilities are to accept the null
hypothesis, or reject it in favor of the alternative. In order to reach one of these conclusions,
some decision rule, based on the sample evidence, has to be formulated. In subsequent sections
we will discuss specific decision rules, noting for now only that their general form is often fairly
obvious. Suppose, for instance, that a random sample of ten boxes of cereal is taken and their
contents weighed. If the sample mean weight is much less than 20 ounces we might aspect the
validity of the null hypothesis that the population mean is at least 200 ounces. The mean was “too
low.” Thus, in testing a null hypothesis about a population mean, it is plausible that our conclusion
will be based on the value observed for the sample mean. All other things equal, the greater the
difference between the sample mean and the values postulated by the null hypothesis about a
population mean, the more suspicious would we be of the truth of that hypothesis.

If all that is available is a sample from population, then the population parameters will not be
precisely known. Accordingly, it cannot be known for sure adopted, there is some chance of
reaching an erroneous conclusion about the population parameter of interest. In fact, as indicated
in table 7.1, either of two possible kinds of error could be made. There are two possible states of
nature-either the null hypothesis is true or it is false. One error that could be made, called a
Type I error is the rejection of a true null hypothesis. If the decision rule is such that the
probability of rejecting the null hypothesis hen it is true is ά, then is ά said to be the significance
level of the test. Since the null hypothesis must either be accepted or rejected, it follows that the
probability of accepting the null hypothesis when it is true is (1- ά). The other possible error,
called a Type II error, arises when a false null hypothesis is accepted. Suppose that, for a
particular decision rule, the probability of making such an error when the null hypothesis is false
is denoted β. Then, the probability of rejecting a false null hypothesis is (1- β), which is called the
power of the test.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 56

We will illustrate these ideas by reference to one of our earlier examples. Consider, again, the
problem of the political scientist trying to determine whether a tax reform proposal appeals
equally to men and women.

The null hypothesis is that, in the population, the proportion of men in favor of this proposal is
the same as the proportion of women. This null hypothesis is to be tested against the alternative
that the two population proportions differ. In order to test the null hypothesis, independent
Table 7.1 states of nature and decision on null hypothesis, with associated probabilities of making
the decisions, given the particular states of nature random samples of men and women are taken,

States of Nature

Null hypothesis true Null hypothesis false

Decision of Accept correct decision probability = 1- ά Type II
error probability = β
Null hypothesis
Reject Type I error probability = ά Correct decision
probability = 1- β
(ά is called significance level) (1- β is called power)

and the views of the sample members are solicited. It is natural to base inference about the null
hypothesis on the difference between the sample proportions of men and women in favor of the
proposal. If this difference is large, the null hypothesis of equality of the population proportions
would be rejected; otherwise, this null hypothesis would be accepted. Let P, denote the sample
proposal. If this difference is large, the null hypothesis of equality of the population proportions
would be rejected; otherwise, this null hypothesis would be accepted. Let P x denote the sample
proportion of men, and P y the sample proportion of women, in favor of the tax reform
proposal. Then a possible decision rule is
Reject H0 if ( p x − Py ) > 0.05 or ( Px − P y ) < -0.05

Now suppose that, in fact, the null hypothesis that the two population proportions favoring the
proposal are equal is true. It nevertheless could happen that the sample proportions differ by
more than 0.05 so that, according to our decision rule, the null hypothesis would be rejected. In
that case, a Type I error would have been made. The probability of this occurring (when the null
hypothesis is true) is the significance level ά. On the hand, suppose that the null hypothesis is
false and that, in fact, the population proportion of men and women in favor of the proposal are
not the same. It may still be the vase that the two sample proportions differ by less than 0.05.
Then according to our decision rule, the null hypothesis would be accepted, and a Type II error
would have been made. The probability of making such an error will depend on just how different
are the two population proportions.

We would be less likely, for given sample sizes, to accept the null hypothesis if 80% of men and
20% of women favored the proposal than if these percentage were 55% and 45%.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 57

Ideally, of course, we would like to have the probabilities of both types of error be as small has
possible. However, there is clearly a trade-off between the two. Once a sample has been taken,
any adjustment to the decision rule that makes it less likely to reject a true null hypothesis will
inevitably render it more likely to accept this hypothesis when it is false. Specially, suppose we
want to test, on the basis of a random sample, the null hypothesis that the true mean weight of
the contents of boxes of cereal is at least 20 ounces. Given a specific sample size- say, n = 30
observations – we might adopt the decision rule that the null hypothesis is rejected if the sample
mean weight is less than 18.5 ounces. Now, it is easy to find a decision rule for which the
probability of Type I error is lower. If we modify our decision rule to “Reject null hypothesis if
sample mean weight is less than 18 ounces” this objective will have been achieved. However,
there s a price to be paid, using the modified decision rule, we will be more likely to accept the
null hypothesis, whether it is true or false. Thus, in decreasing the type I error probability, we
have increased the type II error probability. The only way of simultaneously lowering both error
probabilities would be to obtain more information about the true population mean, by taking a
larger sample. Typically what is done in practice is to fix at some desired level the probability of
making a type I error, that is; the significance level is fixed. This then determines the appropriate
decision is illustrated in figure 7.1.

To illustrate this sequence, consider again the problem of testing based on a sample of thirty
observations, whether the true mean weight of boxes of cereal is at least 20 ounces given a
decision rule. We could determine the probabilities of Type.

Investigator chooses Decision rule Probability of

significance level (probability is determined type II error
of type I error) follows

Figure 7.1 Consequences of fixing the significance level of a test

Type I and type II errors associated with the test. However, in fact, we proceed by first fixing the
type I error probability. Suppose, for example, that we want to ensure that the probability of
rejecting the null hypothesis when it is true is at most 0.05. We can do this by choosing an
appropriate number, K, in the decision rule. “Reject the null hypothesis if the sample mean is less
that K ounce.” (We will discuss in the next section how this can be done.) Once the number K
is chosen, the type II error probabilities can be computed, using procedures to be discussed in
section 7.9.
We have seen that, since the decision rule is determined by the particular significance level
chosen, the concept of power plays no direct part in the decision as to whether to reject a null
hypothesis. However, calculations of power, stemming from particular significance level choices,
provide the investigator with valuable information about the properties of the decision rule.
Often an investigator has some flexibility in the choice of the number of sample observations to
take. For a given significance level, the bigger the sample size, the higher will be the power of the
test. In deciding how big the sample should be, the analyst must balance the benefits from
increased power against the costs acquiring additional sample information. Another important use

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 58

of power calculations arises when we have available two or more possible tests for analyzing the
same problem. If the decision rules associated with these tests are determined so that each has
the same significance level, for a given sample size, then it is natural to prefer the procedure with
the smallest probability of type II error-that is, the procedure with the highest power.

In sections 7.2 -7.8, we show how, for given significance levels, decision rules can be formulated
for some important classes of hypothesis testing problems, we will return in section7.9 to a
consideration of the power of test. For convenience, the new terminology introduced in this
section is summarized in the accompanying box.

Some terminologies in hypothesis testing

Null Hypothesis (Ho) : A maintained hypothesis that is held to be

true until sufficient evidence to the country is obtained.

Alternative Hypothesis (H1) : A hypothesis against which the null

hypothesis is tested, and which will be held to be true if the null is held
false.
Simple Hypothesis : A hypothesis that specifies a single value for a
population parameter
Composite Hypothesis: A hypothesis that specifies a range of value for
a population parameter
One Sided Alternative : An alternative hypothesis involving all possible
values of a population parameter on either one side or the other of (that is,
either greater than or less than) the value specified by a simple null
hypothesis.
Two Sided Alternative : An alternative hypothesis involving all possible
values of a population parameter other than the value specified by a simple
null hypothesis.
Hypothesis Test Decisions: A decision rule is formulated, leading the
investigator to either accept or reject the null hypothesis on the basis of
sample evidence.
Type I Error: the rejection of a true null hypothesis
Type Ii Error : the acceptance of a false mull hypothesis
Significance Level : the probability of rejecting a null hypothesis which is
true (this probability is some times expressed as a percentage, so that a test
of significance level ά is referred to as a 100 ά% level test.)
POWER: The probability of rejecting a null hypotheses which is false

The terminology “accept” and “reject” for the possible decisions about a null hypothesis is
commonly used in formal summaries of the outcomes of particular tests. However, these terms
do not adequately reflect the asymmetry of the status of the null and alternative hypotheses, or
the consequences of a procedure in which the significance level of fixed, and the probability of
Type II error is not controlled. As we have already noted, the null hypothesis has the status of a
maintained hypothesis a hypothesis that will be held true, unless the data contain sufficient contrary
evidence. Moreover, in fixing a significance level, generally at some small probability, we are

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 59

ensuring that the chance is low that a true null hypothesis will be rejected. In such a setup, we are
not likely with only a modest amount of data to be in a position to reject a null hypothesis, unless
it is wildly in error. As we have seen, as the number of sample observations increases, so does
the chance of our being able to detect a false null hypothesis. Thus, in “accepting” a null
hypothesis, we are not necessarily saying a great deal in its favor. A more accurate, though more
pedantic, statement of the position might be “The data available do not provide enough evidence
for rejection of the null hypothesis, given that we want to fix at ά the probability of rejecting a
null hypothesis that is true.” For this reason, some writers prefer the phrase “the null hypothesis
is not rejected” rather than “the null hypothesis is accepted.” We will continue to use “accept,”
as an efficient way of expressing this idea, but it is important that this interpretation of the phrase
be kept in mind. The position is rather similar to that prevailing in a court of law, where the
defendant is, at the outset, deemed innocent, and the burden is on the prosecution to present
sufficiently strong contrary evidence to secure a verdict of “guilty.” In the classical hypothesis
testing framework, the null hypothesis is, in the same sense, initially held to be true. The burden
of persuading us otherwise rests on the sample data.
In the following sections of this chapter we will present tests of a number of specific hypotheses.

7.2 Tests of the Mean of a Normal Distribution: Population Variance

Known
We introduce the methodology of classical hypothesis testing by considering the case where a
random sample of n observations, X1, X2 ... Xn, from a normal distribution with mean µ and
variance σ 2, is available. The objective is to test hypotheses about the unknown population
mean. Initially it will be assumed that the population variance is known. Later we will see that this
assumption and that of normality can be relaxed when the number of sample observations is
large.
We begin with the problem of testing the simple null hypothesis that the population mean is
equal to some specified value, µo. This hypothesis is denoted
Ho: µ = µo
Suppose that the alternative hypothesis of interest is that the population mean exceeds this
specified value; that is,
H 1: µ > µ o
It is natural to base tests of the population mean on the sample mean X in particular; one would
doubt the truth of the null hypothesis, as opposed to this alternative, if the observed sample
mean was greatly in excess of µo. We require the format of a test with some preassigned
significance level ά. That is, we want a decision rule such that the probability of rejecting the null
hypothesis, when it is fact true, is ά. The basis for such a test lies in the fact that the random
X − µ0
variable. > Zα
σ
n
Follows a standard normal distribution; that is, the sample distribution of the sample mean is
normal, with mean µ and standard deviation σ /√n. When the null hypothesis is true, µ is equal to
µo, so that the random variable Z has a standard normal distribution.
Now, the null hypothesis is to be rejected if the sample mean greatly exceeds the value µo
hypothesized for the population mean. Thus, Ho will be rejected if a high value for the random

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 60

variable 7.2.1 is observed. We want to fix at ά the probability of rejecting the null hypothesis
when it is true. As in the previous chapter, we denote by Zά that number for which
P(Z> Zά) = ά
It then follows that, when the null hypothesis is true, the probability that the random variable
7.2.1 is bigger than Zά is ά. Hence, denoting the observed sample mean by X , suppose we adopt
the following decision rule:

X − µ0
Figure7.2 the probability density of > Z α when the null hypothesis Ho: µ = µo is true, and the
σ
n
decision rule for testing Ho against the alternative H1: µ >µo at significance level ά is:
X − µ0
Reject Ho if > Zα
σ
n
Then the probability of rejecting Ho when it is true will be ά, so that ά is the significance level of
the test based on this decision rule. This situation is illustrated in figure 7.2, which shows the
sampling distribution of the random variable 7.2.1 when the null hypothesis is true, through a
graph of its probability density function. The figure shows the value Zά, which is such that the
probability of its being exceeded, when the null hypothesis is true, is significance level ά of the
test. It follows that the probability of a sample result in the corresponding rejection region,
shown as the shaded area in the figure, must be ά when the null hypothesis is correct.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 61

A Test Of The Mean Of A Normal Population:
Population Variance Known
Suppose we have a random sample of n observations from a normal population with mean
µ and known variance σ 2 . if the observed sample mean is X , then a test with significance
level ά of the null hypothesis
Ho: µ = µo
against the alternative
H1 : µ > µ o
is obtained from the decision rule
X − µ0
Reject Ho if > Zα
σ
n
where Zά is that number for which
P(Z > Zά) = ά
and Z is a standard normal random variable

Example:
1. When a process producing ball bearings is operating correctly, the weights of the ball bearings
have a normal distribution with mean 5 ounces and standard deviation 0.1 ounce. An adjustment
has been made to the process, and the plant manager suspects this has raised the mean weight of
ball bearing produced, leaving the standard deviation unchanged. A random sample of sixteen ball
bearings is taken, and their mean weight is found to be 5.038 ounces. Test at significance levels
0.05 and 0.10 (that is, at 5% and 10% levels) the null hypothesis that the population mean weight
is 5 ounces against the alternative that it is bigger.
Denoting by µ the population mean weight (in ounces), we want to test
Ho: µ = µo = 5
against H 1: µ>5
The decision rule is to reject Ho in favor of H1 if
X − µ0
> Zα
σ
n
From the statement of the example, we have
X = 5.038; µo = 5; σ = 0.1; n= 16

X − µ0 5.038 − 5
So that = = 1.52
σ 0 .1
n 16

For a 5% level test, we find from a table:

Z0.05 = 1.645
Since 1.52 does not exceed 1.645, we fail to reject the null hypothesis at the 5% level of
significance; that is, the null hypothesis is accepted at this significance level. In other words, if w3e

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 62

use a test that ensures that the probability of rejecting the null hypothesis when it is in fact true is
0.05, the sample data do not contain enough evidence to allow rejection of that hypothesis.
For a 10% level test, we have
Z0.10 = 1.28
Since 1.52 is bigger than 1.28, the null hypothesis is rejected at the 10% level of significance. To
this extent, then, there is some evidence in the data to suggest that the true mean weight
exceeds 5 ounces.
Let us pause to consider what is meant by the rejection of a null hypothesis. In example1 the
hypothesis that the population mean weight is 5 ounces was rejected by a test with significance
level 0.1. This certainly does not mean that we have proved that the true mean exceeds 5-
ounces. Given only sample information, it will never be possible to be certain about a population
parameter. Rather, we might view the data as having cast some doubt on the truth of the null
hypothesis. If that hypothesis were true, then the observed value
X − µ0
= 1.52 Would represent a single observation drawn from a standard normal
σ
n
population. In testing hypotheses, we are really asking how likely it would be to observe such an
extreme value if the null hypothesis were in fact true. In example1, we saw that the probability of
observing a value bigger than 1.28 is 0.1. Hence, in rejecting the null hypothesis we are saying
either that the null hypothesis is false, or that we have observed an unlikely event – one that
would occur only with the probability specified by the significance level. This is the sense in which
the sample information has aroused doubt about the null hypothesis.
Notice that in example 1 the null hypothesis was rejected at significance level 0.10, but was not
rejected at the lower level 05.05. In lowering the significance level, we are reducing the
probability of rejecting a true null hypothesis, and therefore modifying the decision rule to make
it less likely that the null hypothesis will be rejected, whether or not it is true.
Obviously the lower the significance level at which a null hypothesis can be rejected, the greater
the doubt cast on its truth. Rather than testing hypotheses at preassigned levels of significance,
investigators often determine the smallest level of significance at which a null hypothesis can be
rejected.

Definition
The smallest significance level at which a null hypothesis can be rejected
is called the probability- value, or P-value, of the test.

X − µ0
In example 1 we found = 1.52
σ
n

Therefore, according to our decision rule, the null hypothesis is rejected for any significance level
ά for which Zά is less than 1.52. From a table we find that, when Zά is 1.52, ά is equal to 0.0643.
This, then, is the P– value of the test the implication is that the null hypothesis can be rejected at
all levels of significance

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 63

 Figure 7.3 rejection regions for testing Ho: µ= µo for significance levels. 10, .0643, 0.05

higher than 6.43% this is illustrated in figure 9.3, which shows the correspondence between the
significance level of the test, ά, and the corresponding value Zά, which enters the decision true.
Suppose that, in place of the simple null hypothesis, we had wanted to test the composite null
hypothesis
H 0: µ < µ o
against the alternative
H 1: µ > µo
at significance level ά. For the decision rule developed in the case of the simple null hypothesis,
we saw that if the population mean is precisely µo, then the probability of rejecting the null
hypothesis is ά. For this same decision rule, if the true population mean is anything less than µo
we would be even less likely to reject the null hypothesis. Hence, use of this decision rule in the
present context guarantees a probability of at most ά of rejecting the composite null hypothesis
when it is true.
A Test of the mean of a normal distribution (variance known):
composite Null and Alternative hypotheses
The appropriate procedure for testing, at significance level , the null hypothesis
H0: µ < µo
Against the alternative hypothesis
H 1: µ > µo
is precisely the same as when the null hypothesis is Ho: µ = µo

Consider, now, the problem of testing the simple null hypothesis

H0: µ = µo
Against the alternative hypothesis
H1 : µ< µo

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 64

 X − µ0
Figure 7.4 the probability density function of Z = when the null hypothesis
σ
n
Ho: µ=µo is true, and the decision rule for testing Ho against the alternative H1: µ< µo
at significance level ά

In this circumstance, doubt would be cast on the null hypothesis if the sample mean were a good
deal lower than the hypothesized population mean. Once again, if the null hypothesis were true,
the random variance 7.2.1 would follow a standard normal distribution. To achieve a test with
significance level, we only need to note that
P(Z< Z ά) = ά
If Z is a standard normal random variable: Hence, if X is the observed sample mean, the
appropriate decision rule is:
X − µ0
Reject Ho if < −Z α
σ
n
This is illustrated in figure 7.4, which should be compared with figure 7.2. Clearly the former is
simply the mirror image of the latter.
Using an analogous argument to that developed earlier, we can see that this decision rule
continues to be appropriate if, in place of the simple null hypothesis, we have the composite
hypothesis Ho:µ>µo
With the same alternative hypothesis

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 65

Tests of the mean of a normal distribution: population variance known
Suppose w have a random sample of n observations from a normal population with mean µ and
known variance σ 2. if the observed sample mean is X , a test with significance level ά of either null
hypothesis
Ho: µ= µo Ho: µ>µo
against the alternative
H1: µ<µo
is obtained from the decision rule

X − µ0
Reject Ho if < −Z α
σ
n

We now consider the test of the null hypothesis

H 0: µ= µo
Against the two-sided alternative
H 1: µ ≠ µo
It is assumed here that the investigator has no strong reason for suspecting departures on one
side rather than the other side of the hypothesized population mean. The null hypothesis would
then be doubted if the observed sample mean were either much higher or much lower than µo.
Once again, if the null hypothesis is true, the random variable 7.2.1 has a standard normal
distribution. To obtain a test with significance level ά, note that, under the null hypothesis,
P (Z > Zά/2) = ά/2 and P (Z< -Zά/2) = ά/2
Hence, the probability that Z either exceeds Zά/2 or less than - Zά/2 is ά. It therefore follows
that a test of level ά is obtained from the decision rule:
X − µ0
Reject Ho if is either bigger than -Zά/2
σ
n

This is illustrated in figure7.5, from which we see that the region of sample outcomes for which
the null hypothesis is rejected is divided into two parts. The upper part of the region
corresponds to observed values of the sample mean greatly in excess of the hypothesized
population mean, and the lower part to values of the sample mean that are substantially below µo

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 66

Test for the mean of normal distribution against two-sided alternative; population
variance known
Suppose we have a random sample of n observations from a normal population with mean µ and
known variance σ 2. if the observed sample mean is X , then a test with significance level ά of the
null hypothesis
H0 = µ= µo
against the two-sided alternative
H0 = µ= µo
is oobtained from the decision rule

Reject H0 if X − µ 0 > Z a or X − µ 0 < − Z a

σ 2
σ 2
n n

X − µ0
Figure7.5 the probability density function of Z = when the null hypothesis Ho: µ= µo is true,
σ
n
and the decision rule for testing Ho against the alternative H1 : µ≠ µo at significance level ά
The reader has probably noticed the similarity between the developments of procedures for
determining confidence intervals and testing hypotheses. A review of the material in section 8.2
will clarify the relationship. The null hypothesis Ho: µ= µo is rejected against the two –sided
alternative H1: µ≠ µo at significance level ά if and only if the 100 (1- ά) % confidence interval for µ
does not contain µo.
Example:
A drill, as part of an assembly line operation, is used to drill holes in sheet metal. When the drill
is functioning properly, the diameters of these holes have a normal distribution with mean 2
inches and standard deviation 0.06 inch. Periodically, to check that the drill is functioning
properly, the diameters of a random sample of holes are measured. Assume that the standard
deviation does not vary. A random sample of nine measurements yield mean diameter 1.95
inches. Test the null hypothesis that the population mean is 2 inches against the alternative that it
is not. Use a 5% significance level and also find the P-value of the test.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 67

Solution:
Let µ denote the population mean diameter (in inches). Then we require to test
H 0: µ= µo=2
Against
H 1: µ≠ 2
The decision rule is to reject Ho in favor of H1 if
X − µ0 X − µ0
> Zα / 2 < −Z α / 2
σ σ
n n
We have
X = 1.95; µo = 2; σ =0.06; n= 9
And so
X − µ0 1.95 − 2
= = − 2 .5
σ 0.06
n 9
For a 5% level test, ά = 0.05, and Zά/2 = Z0.025 = 1.96. Then, since – 2.50 is less than -1.96, the null
hypothesis is rejected at the 5% significance level.
In fact, according to the decision rule, the null hypothesis will be rejected for any significance level
ά for which - Zά/2 is bigger than -2.50. From a table, we see that when Zά/2 is 2.50, ά/2 is equal
to 0.006. Hence, ά = 0.0124. This is the P- value of the test, implying that the null hypothesis can
be rejected against the two-sided alternative at any level of significance greater than 1.24%. This
certainly casts substantial doubt on the hypothesis that the drill is functioning correctly.
Until now we have dealt only with the (generally unrealistic) case where the population variance
is known. However, if the available number of sample observations is large, the tests can readily
be modified to deal with an important class of practical problems. As we discussed in the
previous chapter, we are indebted to the central limit theorem, which allows us to conclude that,
for large samples, the sampling distribution of the sample mean will be approximately normal,
even though the population distribution is not normal.

Tests For The Mean Large Sample Sizes

Suppose we have a random of n observations from a population with mean µ and variance σ 2. If
the sample size n is large, the test procedures developed for the case where the population variance
is known can be employed when it is unknown, replacing σ 2 by the observed sample variance S2x.
Moreover, these procedures remain approximately valid even if the population distribution is not
normal.

Example:
It might be suspected that the firms most likely to attract take-over bids are those that have been
achieving relatively poor returns. One measure o such performance is through “abnormal
returns,” which average 0 over all firms. A random sample of 88 firms for which cash tender
offers had been made showed abnormal returns with a mean of -0.0029 and a standard deviation
of 0.0169 in the period from 24 months to 4 months prior to the take-over bids. Test the null

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 68

hypothesis that the mean abnormal returns for this population is 0 against the alternative that it is
negative.
Solution:
Let µ denote the population mean abnormal returns. Then we want to test
H0: µ= µo = 0
against
H 1: µ < µ o
The decision rule is to reject Ho in favor of H1 if
X - µo < -Zά
Sx/√n
Here we have
X = -0.0029; µo = 0; Sx = 0.0169; n= 88
And hence
X - µo = -0.0029 = -1.61
Sx/√n 0.0169/√88
According to our decision rule, the null hypothesis is rejected for any significance level ά for
which –Zά is bigger than -1.61. From a table, we see that when Zά is 1.61, ά is equal to 0.0537.
Hence, the null hypothesis is rejected at any significance level bigger than 5.37%.Thus, the
probability of observing sample mean abnormal returns as low as or lower than those actually
observed would be 0.0537 if the true mean abnormal returns for all firms attracting take-over
bids were 0. The data suggest quite strongly that, on average, abnormal returns are lower for
such firms.
The conclusion s illustrated in figure 7.6, which shows the distribution of the decision rule
criterion under the null hypothesis, with the lower-tail area probability corresponding to the
observed values.

Figure 7.6 conclusion of the test in the above example; the null hypothesis Ho: µ = µo is rejected against
the alternative H0: µ < µo at significance levels greater than .0537

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 69

7.3 Tests of the Mean of Normal Distribution: Population Variance Unknown
In this section we gain consider the problem where a random sample of n observation is take
from a normal population, and it is required to test hypotheses about the population mean µ.
However, the population variance is no longer assumed known. If the sample size is not large, the
procedures discussed at the end of section 7.2 are no longer appropriate. Nevertheless, valid
tests can be derived. It was seen there that, if the sample mean and variance are denoted X and
S2x, the random variable
X − µ
t n −1 =
sX
n
follows a student’s distribution with (n-1) degrees of freedom. Following precisely the line of
argument adopted in section 7.2, with the student’s t distribution now playing the same role as
the standard normal distribution, we can obtain valid tests, as indicated in the box.

Tests Of The Mean Of A Normal Distribution: Population Variance Unknown

suppose we have a random sample of n observations from a normal population with mean µ. if
the observed sample mean and standard deviation are X and Sx, then the following tests have
significance level ά:
i) To test either null hypothesis
H0: µ= µo or H0: µ< µo
against the alternative
H1: µ>µo = 0
the decision rule is
Reject H0 if X − µ 0 > tn – 1,ά
sX
n

ii) To test either null hypothesis

H0: µ= µo or H0: µ> µo
the decision rule is
Reject H0 if X − µ 0 < tn – 1,ά
s X
n

iii) To test the null hypothesis

H0: µ= µo
against the two-sided alternative
H1: µ≠µo

the decision rule is

Reject H0 if X − µ 0 > tn – 1,ά/2 or X − µ 0 < tn – 1,ά/2
s X sX
n n

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 70

Here, tn – 1.ά is that number for which
P (tn – 1 > tn – 1.ά) = ά
Where the random variable tn – 1 follows a student’s t distribution with (n-1) degrees of
freedom.

Figure 7.7 illustrates the setup of the test against a two-sided alternative. The probability density
function is now that of the student’s t- distribution and this figure is the analogue of figure 7.5,
which related to the case where the population variance was known. In an obvious way, tests
against one-sided alternative hypotheses can be viewed pictorially in a manner analogous to
figures 7.2 and 7.4.

Figure 7.7 the probability density function of tn – 1 = X − µ 0

s X
n
When the null hypothesis Ho: µ= µo is true, and the decision rule for testing H0 against the alternative
H1: µ≠µo at significance level ά

Example:
A real chain knows that, on average, sales in its stores are 20% higher in December than in
November. For the current year, a random sample of six stores was selected. Their percentage
December sales increases were found to be:
19.2; 18.4; 19.8; 20.2; 20.4; 19.0
Assuming a normal population distribution, test the null hypothesis that the true mean
percentage sales increase is 20, against the two-sided alternative, at the 10% significance level.
Solution:
Letting µ denote the population mean percentage increase in sales in December, we want to test
the null hypothesis.
H0: µ= µo = 20
against the alternative
H1: µ ≠20

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 71

The decision rule is to Reject Ho if favor of H1 if

The sample mean and variance are obtained by using the computations in the accompanying table.
We have for the sample mean,

X=
∑ X i = 117 = 19.6
n 6

Xi Xi2
19.2 368.64
18.4 338.56
19.8 392.04
20.2 408.04
20.4 416.16
19.0 316.00
Sum=117 2,284.44

and for the sample variance,

2
Sx2 =∑xi– n X = 2,284.44 – (6)(19.5) = 0.588
n -1 5
So that the sample standard deviation is
Sx= √0.588 = 0.767
We then have
X - µo = 19.5 – 20 = -1.597
Sx/√n 0.767/√6
Since a test of significance level ά = 0.10 is required, we have from a table,
tn – 1,ά/2 = t5.05 = 2.015
Thus, since –1.597 lies between - 2.015 and 2.015, the null hypothesis that the true mean
percentage increase is 20 is accepted at the 10% level. The evidence in the data against this
hypothesis is not terribly strong.

7.4 Tests of the Variance of a Normal Distribution

In this section, we develop procedures for testing the population variance σ 2, based on a random
sample of n observations from a normal population.
It is natural to base these tests on the sample variance Sx2. the basis for developing particular tests
lies in the fact that the random variable.
X2n-1 = (n- 1) Sx2
σ2
follows a chi-square distribution with (n-1) degrees of freedom.(N.B once again, we caution the
reader that the validity of this result rests crucially on the assumption that the population
distribution is normal). If the null hypothesis is that the population variance is equal to some
specified value σ 0 2, that is,

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 72

 H 0: σ 2 = σ 0 2

then when this hypothesis is true the random variance

X2n-1 = (n- 1) Sx2
σ 2 ----------- - ---------------------- (7.4.1)

obeys a chi-square distribution with (n-1) degrees of freedom. Tests of hypotheses about the
variance of a normal population are then based on the sample value observed for the equation
7.4.1. If the alternative hypothesis is that the true variance exceeds σ 0 2, we would be suspicious
of the null hypothesis is the observed sample variance was much bigger than σ 0 2. Hence, the null
hypothesis would be rejected if a high value of 7.4.1 we observed. Conversely, if the alternative is
that the population variance is less than the value specified by the null hypothesis, the null
hypothesis would be rejected for low values of 7.4.1. Finally, for the two-sided alternative that
the population variance differs from σ 0 2, we would want to reject the null hypothesis on
observing either unusually high or unusually low values of 7.4.1.
The rationale for the development of appropriate tests now follows the same pattern as in
section 7.2

Figure 7.8 some probability for the chi-square distribution

We denote by X2v,ά that number which is exceeded with probability ά by a chi-square random
variable with a degrees of freedom. Hence,
P (X2v > X2v,ά) = ά
It then follows that
P (X2v < X2vά) = ά
and that
P (X2v > X2v, ά/2 or X2v < X2v,1 – ά/2) = ά
These probabilities are shown in figure 7.8. Tests for the normal variance then follow as indicated
in the box.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 73

Tests Of The Variance Of A Normal Population

Suppose we have a random sample of n observations from a normal population with variance
σ 2. if the observed sample variance is Sx2, then the following tests have significance level ά:
i) to test either null hypothesis
Ho: σ 2 = σ 2o or Ho: σ 2 < σ 2o
against the alternative /
H1: σ 2 > σ 2o
the decision rule is
reject Ho if ( n-1) Sx2 >X2n-1,ά
σ 2o
ii) To test either null hypothesis
Ho: σ 2 = σ 02 or Ho: σ 2 > σ 2o
against the alternative /
H1: σ 2 < σ 2o
the decision rule is
reject Ho if ( n-1) Sx2 < X2n-1,1-ά
σ 02
iii) To test the null hypothesis
Ho: σ 2 = σ 2o
against the two-sided alternative
H 1: σ 2 ≠ σ 2o

Figure 7.9 the probability density function of X2n-1 = (n – 1) S2x/∂2o when the null hypothesis Ho: σ 2 = σ 2
0 is
true, and the decision rule for testing Ho against the alternative H1: σ 2 ≠ σ 2o at significance level ά

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 74

the decision rule is
Reject Ho if = (n- 1) Sx2 >X2n-1,ά/2 or (n- 1) Sx2 < X2n-1,1-ά/2
σ 2 σ2
here X2n-1,ά is that number for which
P( 2
X n-1 > X2n-1,ά) = ά
Where the random variable X2n-1 follows a chi-square distribution with (n-1) degrees of
freedom.
The decision rule for the test against the two-sided alternative is illustrated in figure 7.9.

Example:
In order to meet established standards, it is important that the variance of the percentage
impurity levels in consignments of a chemical not exceed 4.0. A random sample of twenty
consignments had a sample variance of 5.62 in impurity level percentage. Test the null hypothesis
that the population variance is not more than 4.0.
Solution:
Let σ 2 denote the population variance of impurity concentrations. The null hypothesis
Ho: σ 2 < σ 20 = 4.0
Is to be tested against
H1: σ 2 > 4.0
Based on the assumption that the population distribution is normal, the decision rule, for a test of
significance level ά, is to reject Ho in favor of H1 if
(n- 1) S2x < X2n-1, ά
σ2
From the statement of the example we have
S2x = 5.62; n= 20; σ 2o = 4.0
Hence,
(n- 1) S2x = (19) (5.62) = 26.695
σ 20 4.0
For a 10% level test, ά = 0.10 and we see from table 5 in the appendix that the corresponding
cutoff point of the chi-square distribution with (n-1) = 19 degrees of freedom is
X219,.10 = 27.20
Therefore, since 26.695 is not bigger than 27.20, the null hypothesis can not be rejected at the
10% level. Hence, the data do not contain terribly strong evidence against the hypothesis that the
population variance in impurity level percentages is at most 4.0.

7.5 Tests of the Population Proportion (Large Samples)

In many practical problems we want to test hypotheses about the proportion of members of a
large population passing some particular attribute. Inference about the population proportion is
based on the proportion of individuals in a random sample who possess the attribute of interest.
Denoting by p the population proportion, and by P X the proportion in a random sample of n
observations, we know that if the sample size is large, then to a good approximation the random
variable.
p− p
Z=
P(1 − P)
n

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 75

has a standard normal distribution. That the sampling distribution for the sample proportion is
approximately normal when the sample size is large follows, as we have noted in previous
chapters, as a result of the central limit theorem.
If the null hypothesis is that the population proportion is equal to some specific value P0, it
follows that when this hypothesis is true, the random variable.
p − p0
Z=
p 0 (1 − P0 )
n
follows a standard normal distribution. We can now deduce appropriate tests for the population
proportion, as described in the box.

Testing of the population proportion (large sample sizes)

Suppose we have a sample of n observations from a population, a proportion p of whose
members possess a particular attribute. Then, if the number of sample observations is large(
and the observed sample proportion is p x, the following tests have significance level ά:
i) To test either null hypothesis
Ho: P= Po or Ho: P< Po
against the alternative
H1: P > Po
the decision rule is
p X − P0
Reject Ho if > Za
P0 (1 − p 0 )
ii) To test either null hypothesis
Ho: P= Po or Ho: P>Po
against the alternative
Ho: P< Po

( PX − P0 )
Fig 7.10 The probability density function of Z = when the null hypothesis H0:
P0 (1 − P0 ) / n
P =P0 is true, and the decision rule for testing H0 against the alternative H1: P<P0 at a significance level ά

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 76

the decision rule is
p − P0
Reject Ho if < Zά
P0 (1 − p 0)
n
iii) To test the null hypothesis
Ho: P= Po
against the two-sided alternative
H1: P ≠ Po
the decision rule is

P − P0 P − P0
Reject Ho if > Z a or < −Z a
P0 (1 − P0 ) 2 P0 (1 − P0 ) 2
n n
Here, as previously, Zά is that number for which
P (Z > Zά) = ά
Where the random variable Z has a standard normal distribution.

The decision rule for the second of these tests is illustrated in figure 7.10.

Example:
Forecasts of corporate earnings per share are made on a regular basis by many financial analysts.
In a random sample of 600 forecasts, it was found that 382 of these forecasts exceed the actual
out come for earnings. Test against a two-sided alternative the null hypothesis that the population
proportion of forecasts that are higher than actual outcomes is 0.05. (This is the hypothesis we
would expect to be true if there were no overall tendency for financial analysts to be either
unduly optimistic or unduly pessimistic about earnings prospects.)
Solution:
Let P denote the population proportion of forecasts that are above actual out comes. We want
to test
Ho: P= Po = 0.50
Against H1: P ≠ 0.50
The decision rule is to reject. Ho in favor of the alternative if
P − P0 P − P0
> Z a or < −Z a
P0 (1 − P0) 2 P0 (1 − P0 ) 2
n n
We have
Po = 0.50; n = 600; p 0 = 382 = 0.637
600
P − P0 0.637 − 0.50
Then = = 6.71
P0 (1 − P0 ) (0.50)(0.50)
n 600

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 77

Referring from a table 3, we see that such an outcome would be extremely unlikely if the null
hypothesis were indeed true. For instance, for a 0.2% level test, ά = 0.002, so that
Zά/2 = Z.001 = 3.09
Since 6.71 is much bigger than 3.09, the null hypothesis is clearly rejected even at such a low
significance level. The evidence against the hypothesis that the population proportion of forecasts
that exceed outcome is 0.50 is really overwhelming. It is clear from these data that there is an
overall tendency for the proportion of financial analysts’ forecasts of future corporate earnings
that exceed the actual values to differ from 0.50.

7.6 Tests for the Difference between Two Means

In this section we examine the case where random samples are available from two populations,
and the quantity of interest is the difference between the two population means. In developing
procedures for testing hypotheses, the appropriate methodology once again depends on the
manner in which the samples are taken. As we discussed in section 6.5 we need to consider
separately the cases of matched pairs and independent samples.

Tests Based on matched pairs

Here we assume that a random sample of n matched pairs of observations is obtained from
populations with means µx and µy. The actual sample observations will be denoted (X1, Y1),
(X2, Y2), ….. (Xn, Yn).
In a study aimed at assessing the relationship between a subject’s brain activity while watching a
television commercial, and the subject’s subsequent ability to recall the contents of the
commercial, the data in Table 7.2 were obtained. Subjects were shown commercials for two
brands of each of ten products. For each commercial, the ability to real 24 hours later was
measured, and each member of a pair of commercials was then designated “high recall” or “low
recall.” The table shows an index of the total amount of brain activity of subjects while watching
these commercials. If µx denotes the population mean brain activity for the high recall
commercials

Table 7.2 Brain Activities of Subjects Watching Ten Pares of Television

Commercials
Product High Recall Low Recall Differences
I Xi Yi di di 2
1 137 53 84 7,056
2 135 114 21 441
3 85 81 2 4
4 125 86 39 1,521
5 47 34 13 169
6 46 66 -20 400
7 114 89 25 625
8 157 113 44 1,936
9 57 88 -31 961
10 144 111 33 1,089
Sum 210 14,202

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 78

and µy the population mean for the low recall commercials, then the differences di represent a
random sample of ten observations from a population with mean (µx – µy). If the population
distribution of these differences can be assumed normal, the methods of section 7.3 are
immediately applicable for testing hypothesis about (µx – µy). For this particular example, a
natural test is of the null hypothesis of no difference in mean brain activity levels,
Ho: µx – µy = 0
against the alternative that, on average, brain activity is grater for the high recall commercials,
that is,
H 1: µ x – µ y > 0
More generally, we can test the hypothesis that the difference ( µx – µy) is equal to any specified
value D0 using the procedure given in the box.

Tests Of The Difference Between Population Means: Matched Pairs

Suppose we have a random sample of n matched pairs of observations from distributions with
means µx and µy . Let d and Sd denote the observed sample mean and standard deviation for
the n differences (Xi – Yi).If the population distribution of the differences is normal, then the
following tests have significance level ά:
i) To test either null hypothesis
Ho: µx – µy = D0 or Ho: µx – µy < D0
against the alternative
Ho: µx – µy > D0
the decision rule is
d − D0
reject Ho if > tn – 1,ά
Sd
n

ii) To test either null hypothesis

H o: µ x – µ y = D 0 or Ho: µx – µy > D0
against the alternative
Ho: µx – µy < D0
the decision rule is
d − D0
Reject Ho if < tn – 1,ά
Sd
n

iii) To test the null hypothesis

Ho: µx – µy = D0
against the alternative
Ho: µx – µy ≠ D0
the decision rule is

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 79

 d − D0 d − D0
Reject Ho if < -tn – 1,ά/2 or if > tn – 1,ά/2
Sd Sd
n n
Here tn – 1,ά is that number for which
P(tn-1 > tn – 1,ά) = ά
Where the random variable tn-1 follows a student’s distribution with (n-1) degrees of freedom.
When we want to test the null hypothesis that the two population means are equal,
we set D0 = 0 in the formulas.

We now return to our example of brain activity of subjects watching television commercials;
from table 7.2, the sample mean of the differences is

n
d = ∑ di = 1 (210) = 21.0
i=1 10
The sample variance is
n
Sd = 1 (∑ di2 – nd2)
2

n – 1 i=1
= 1 [14,202 – (10)(21.0)2] = 1,088
9
So that the sample standard deviation is
Sd = √1,088 = 32.98
We want to test the null hypothesis
Ho: µx – µy = Do = 0
against the alternative
Ho: µx – µy > 0
d − D0 21.0
The test is based on or = =2.014
Sd 32.98 / 10
n

This quantity must be compared with tabulated values of the student’s t distribution with
(n- 1) = 9 degrees of freedom. From a table, we have for 5% level and 2.5% level tests,
t9, .05 = 1.833 and t9, .025 = 2.262
Hence, the null hypothesis of equality of the population means can be rejected at the 5% level,
but not at the 2.5% level of significance. We see then that the data of table 7.2 contain much
evidence suggesting that, on the average, brain activity were the same for these two is higher for
the high recall than for the low recall group. If, in fact, the mean brain activity as extreme or
more extreme than actually obtained would be between .025 and .05

Tests Based On Independent Samples

Suppose now that we have a random sample of nx observation from a normal population with
mean µx and variance σ 2x and an independent random sample of ny observations from a Norman
population with mean µy and variance σ 2y .

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 80

In section 8.7 we saw that, if the sample means are denoted X & Y, then random variable
( X − Y ) − (µ X − µY )
Z=
σX2 σY 2
−
nX nY
has a standard normal distribution. If the two population variances are known, tests for the
difference between the population means can be based on this result, using the same arguments
as before. Moreover (thanks to the central limit theorem), if the two sample sizes are large, the
result continues to a good approximation when the sample variances replace the population
variances, even when the population distributions are not normal. This allows the derivation of
tests of wide applicability, as summarized in the box.

Tests For The Difference Between Population Means: Independent Samples

(Known Variances Or Large Sample Sizes)

Suppose we have independent random samples of nx and ny observations from normal

distributions with means µx and µy and variances and σ 2x + σ 2y. if the observed sample means
are X and Y, then the following tests have significance level ά:
i) To test either null hypothesis
Ho: µx – µy = D0 Or Ho: µx – µy < D0
against the alternative
H1: µx – µy > D0
the decision rule is
( X − Y ) − (µ X − µY )
Reject H0 if >-Za
σX2 σY 2
−
nX nY
ii) To test either null hypothesis
Ho: µx – µy = D0 or Ho: µx – µy > D0
against the alternative H1: µx – µy < D0
the decision rule is
( X − Y ) − (µ X − µY )
Reject H0 if <-Za
σX2 σY 2
−
nX nY
iii) Te test the null hypothesis Ho: µx – µy = D0
against alternative H 1: µ x – µ y ≠ D o
The decision rule is
The decision rule is
( X − Y ) − (µ X − µY ) ( X − Y ) − (µ X − µY )
Reject Ho if <-Za/2 or > Za/2
σX 2
σY 2
σX 2
σY 2

− −
nX nY nX nY

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 81

 If the sample sizes nx are large, (this approximation is generally satisfactory when each
sample contains at least thirty observations),then to a good approximation, tests of
significance level ά for the difference between population means are obtained by replacing
the population variances by the observed sample variances, S2x and S2y. For large sample
sizes, these approximations remain good even when the population distributions are not
normal.

Example:
1. The international banking crisis of 1974, involving the failure of the franking national bank in
New York, led the Federal Reserve System to guarantee the international as well as the domestic
deposits of the bank. It might be hypothesize that this “Franklin Message” would lead to a
decrease in the risk premium attached to large American Banks’ deposits. (Risk premium here is
taken to be measured by the excess of secondary market certificate of deposit rates over
Treasury bill yields.) For 48 months before the “Franklin Message,” the mean risk premium was
0.899, and the variance was 0.247. For 48 months after the message, the mean and variance were
0.703 and 0.320. If µx and µy denote respectively the means before and after the message, test the
null hypothesis Ho: µx – µy = 0
Against the alternative Ho: µx – µy > 0

Solution:
Assume that the data can be regarded as independent random samples from the two populations.
The decision rule is to reject Ho in favor of H1.

In this example,
X= 0.899; S2x= 0.247; nx = 48; Y = 0.703; S2y = 0.320; ny= 48
So that
X −Y 0.0899 − 0.703
= = 1.80 From a table, we find that the value of ά corresponding
2 2
S X S Y 0.247 0.320
+ + to Zά = 1.80 is 0.0359. Hence, the null hypothesis can
nx nY 48 48 be rejected at all levels of significance greater than
3.59%. Hence, were the null hypotheses of equality of population means true, the probability of
observing a sample result as extreme or more extreme than that found would be 0.0359. This
represents pretty strong evidence against the null hypothesis of equality of these means,
suggesting rather a decrease in the mean risk premium after the “Franklin Message.”
We will now treat the case where the sample sizes are not large. If it can be assumed that the
two population variances are equal, then tests can be based on the result our discussion of
confidence intervals for the difference between the means of two normal population that the
random variable.
( X − Y ) − ( µ X − µY )
t=
n + nY
S X
n X nY
has a student’s t distribution with (nx + ny -2) degrees of freedom, where

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 82

 S2 = (nx – 1) S2x + (ny – 1)S2y
(nx + ny – 2)
is an estimator of the common variance. The appropriate hypothesis tests on the difference
between the population means are given in the box.

Tests For The Difference Between The Means Of Two Normal Populations:
Independent Samples, Population Variances Equal
Suppose we have independent random samples of nx and ny observations from normal
distributions with means µ X and µ Y and a common variance. If the observed sample variances
are S2x and S2y, an estimate of the common population variance is provided by
S2 = (nx – 1) S2x + (ny – 1)S2y
(nx + ny – 2)
Then, if the observed sample means are X and Y , the following tests have significance level ά:
i) To test either null hypothesis
Ho: µx – µy = D0 or Ho: µx – µy < D0
against the alternative H1: µx – µy > D0
the decision rule is
X − Y − D0
Reject Ho if > t n x + nY − 2, a
n x − nY
S
nx n y
ii) To test either null hypothesis
Ho: µx – µy = D0 or Ho: µx – µy > D0
against the alternative
H1: µx – µy < D0
the decision rule is
X − Y − D0
Reject Ho if < −t n x + nY − 2, a
n − nY
S x
nx n y
iii) To test the null hypothesis
Ho: µx – µy = D0
against the alternative
H1: µx – µy < D0
the decision rule is
X − Y − D0 X − Y − D0
Reject Ho if < −t n x + nY − 2, a or > t n x + nY − 2, a
n − nY n − nY
S x S x
nx n y nx n y

Here tnx+ny-2,ά is that number for which

P(tnx+ny-2 > tnx+ny-2,ά) = ά
where tnx+ny-2 has a student’s t distribution with (nx + ny -2) degrees of random.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 83

2. In a student aimed at finding early warning signals of business failure, a random sample of
twenty-three failed retail firms showed mean return on assets 3 year previously of 0.058, and
sample standard deviation 0.055. An independent random sample of twenty-three nonfailed retail
firms showed mean return of 0.146 and standard deviation 0.058 for the same period. If µx and µy
denote the population means for failed and nonfailed firms, respectively, test the null hypothesis
H0: µx- µy =0
Against the alternative
H1: µx- µy
Solution:
Assume the two population distributions are normal and have the same variance:
The decision rule is to reject H0 favor of H1 if

X −Y
< −t nx + nY − 2,a
n x − nY
S
nx n y
For these data, we have
X = 0.058; Sx= 0.055; nx = 23; Y = 0.146; Sy= 0.058; ny = 23
Hence
S2 = (nx -1) S2x + (ny – 1) S2y
nx + ny - 2
= (22) (0.055)2 + (22) (0.058)2 = 0.0031945
23 + 23 – 2
So that S = √0.0031945 = 0.0565
X −Y 0.058 − 0.146
Then = = −5.282
n x − nY 23 + 23
S 0.0565
nx n y 23 2
For a 0.5% level test, we have by interpolation from table 6, for the student’s t distribution with
(nx + ny – 2) = 44 degrees of freedom,
t44, .005 = 2.695
Then, since -5.282 is much less than -2.695, the null hypothesis is overwhelmingly rejected even
at this level of significance. The data cast considerable doubt on the hypothesis that the two
population means are equal. Rather, they suggest very strongly that the population mean return
on assets is lower for failed than for nonfailed retail firms.
The test just discussed and illustrated is based on an assumption that the two population
variances are equal. In fact, it is possible to develop tests that are valid when this assumption does
not hold. However, these will not be discussed further here.

7.7 Tests for the Difference between Two Population Proportions (Large Sample)
We turn now to the problem of comparing two population proportions. As we discussed in
chapter 6,Suppose that a random sample of nx observations from a population with proportion
px “successes” gives a sample proportion PX , and that an independent random sample of ny
observations from a population with proportion py ”successes” yields sample proportion PY .

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 84

 In our discussion of interval estimation of the difference between population proportions we saw
that, if the numbers of sample observations are large, then to a very good approximation, the
random variable
( p X − PY ) − ( Px − PY )
Z=
PX (1 − PX ) PX (1 − PY )
+
nX nY
has a standard normal distribution, by virtue of the central limit theorem.
Suppose that we want to test the hypothesis that the population proportions px and py are equal.
If their common value is denoted p0, then we have under this hypothesis that

( p X − PY ) − ( Px − PY )
Z =
PX (1 − PX ) PX (1 − PY )
+
nX nY
p X − PY
= (7.7.1)
n + nY
P0 (1 − P0 )( X )
n X nY

follows to a good approximation a standard normal distribution.

Finally, the unknown common proportion p0 in expression 7.7.1 can be estimated by the pooled
estimator p0 given by
n x PX + nY P Y
P0 =
n X + nY
Replacing the unknown p0 in 7.7.1 gives a random variable that has distribution close to the
standard normal, provided the sample sizes are large. This result forms the basis for our tests, as
indicated in the box.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 85

Testing The Equality Of Two Population Proportions (Large Samples)
Let PX denote the proportion of successes in random sample of nx observations
from a population with proportion px successes observed in an independent random sample of ny
observations from a population with proportion py successes. If it is hypothesized that the population
proportions are equal, an estimate of the common proportion is given by
n x PX + nY P Y
P0 =
n X + nY
Then ,if the sample sizes are large (the approximation is generally adequate if each sample contains at
least forty observation), the following tests have significance level α :
(i) To test either null hypothesis
Ho: px –py=0 or Ho: px –py ≤0
against the alternative
H1: px –py > 0
the decision rule is
p X − PY
Reject Ho if > Zα
n X + nY
P 0 (1 − P 0 )( )
n X nY
(ii) To test either null hypothesis
Ho: po -py=0 or H0: px-py >0

against the alternative H1: px –py < 0

the decision rule is
p X − PY
Reject Ho <- Z α
n + nY
P 0 (1 − P 0 )( X )
n X nY
(iii) To test the null hypothesis Ho: px –py =0
against alternative H1: px –py ≠ 0
the decision rule is
p X − PY
Reject H0: <- Z α / 2
n + nY
P0 (1 − P0 )( X )
n X nY
p X − PY
Or > Zα / 2
n + nY
P0 (1 − P0 )( X )
n X nY

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 86

Example:
In a study designed to determine important criteria in selecting a retail pharmacy, a random
sample of 200 people, aged 18-30, contained 36 who said price was the most important criterion.
An independent random sample of 118 people, aged 50 years or older, showed 29 indicating
price as the most important criterion. Denoting by Px and PY the respective population
proportion, test the null hypothesis
Ho: px – py =0
against the alternative
H1: px – py ≠ 0
Solution:

The decision rule is to reject Ho in favor of H1 if

p X − PY p X − PY
<- Z α / 2 or > Zα / 2
n + nY n + nY
P0 (1 − P0 )( X ) P0 (1 − P0 )( X )
n X nY n X nY
For these data, we have
P x= 36 = 0.180; nx= 200; P y = 29 = 0.246; ny =118
200 118

Hence,

n X PX + nY PY
= = (200)(0.180)+(118)(0.246) /200+118 =0.204
n X + nY
Then
p X − PY 0.180 − 0.246
= = -1.41
n + nY 200 + 118
P 0 (1 − P 0 )( X ) (0.204)(0.796)[ ]
n X nY (200)(118)

The value of α /2 corresponding to z α /2= 1.41 is, from a table α /2=.0793,sothat α =.1586.
Hence, the null hypothesis can be rejected only at significance levels higher than 15.86% .The
evidence against the hypothesis that the population proportions viewing price as the most
important criterion are the same in these two age groups is very strong.

7.8 Testing the Equality of the Variances of Two Normal Populations

One of tests developed in section 7.6 for the comparison of population means depends on an
assumption of equality of the two population variance. While in many practical applications such
an assumptions is reasonable, it is prudent to use the available data to test its validity. In addition,
it sometimes happens when comparing population distributions that population variances ate of
interest in their own right, and that we wish t compare them.
In this section, we consider the problem where independent random samples from two normal
population s are available, and it is required to tenses the equality of the population variance. To
develop such a test, another probability distribution must be. Let s2x be the sample variance for a
random sample of nx observations from a normal population with variance σ 2x and s2y be the

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 87

sample variance from an independent random sample of ny observations from a normal
population with variance σ 2y then the random variable

F= S X / σ X
2 2
(7.8.1)
S 2Y σ 2Y

Follows a distributions known as the F distribution ( Formally this distribution is defined as the
distribution followed by the ratio of two independent Chi-square variables , each divided by its associated
degree of freedom). This family of distribution is widely used in statistical analysis. A particular
member of the family is distinguished by two values-the degrees of freedom associated with the
numerator and with the denominator. In the preset context, recall that the degrees of freedom
associated with the sample variance s2xis (nx-1), and that with s2y is (ny-1). The random variable
7.8.1 then has an F distribution with numerator degrees of freedom (nx-1 and denominator
degrees of freedom (ny-1).
The F distribution has an asymmetric probability density function, defined only for nonnegative
values. This density function is illustrated in figure 7.11.

The F Distribution
Suppose that independent random samples of nx=and ny observations are taken from two
normal population with variances σ 2x and σ 2y .If the sample variances are s2xand s2y,then the
random variable
S X /σ 2 X
2

F= 2
S Y / σ 2Y
has an F distribution with numerator degrees of freedom (nx – 1) and denominator degrees of
freedom (ny – 1). An F distribution with numerator degrees of freedom V1 and denominator
degrees of freedom V2 will be denoted FV1, [Link] denoted by FV1, V2,ά that number for which
P(FV1 V2 > FV1, V2,ά) = ά

Figure 7.11 probability density function of the distribution with 6numerator degrees of freedom and 4
denominator degrees of freedom; the probability is ά that F6,4 exceeds F6,4,ά

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 88

The cutoff points FV1 V2.ά, for ά equal to 0.05 and 0.01, are provided from a Chi-square the. For
example, for 10 numerator degrees of freedom and 20 denominator degrees of freedom, we see
from the table
F10, 20, 05 = 2.35 and F10, 20,01 = 3.37
Hence,
P (F10, 20>2.35) = 0.05 and P (F10, 20>3.37) = 0.01
In practical applications, provided we arrange matters so that the larger sample variance appears
in the numerator, these are the only cutoff points necessary to test the hypothesis of equality of
population variances. When the population variances are equal, it follows from7.8.1 that the
random variable.
S 2X
F= 2
S Y
Obeys an Fnx -1,ny- 1 distribution. Appropriate hypothesis tests are described in the box.

Tests For The Equality Of The Variances Of Two Normal Populations

Let S2x and S2y be observed sample variances from independent random samples of nx and ny
observations from normal populations with variances σ 2x and σ 2y. If S2x is bigger than S2y, then
the following tests have significance level ά:
i) To test either null hypothesis
Ho: σ 2x = σ 2y or H1: σ 2x < σ 2y
against the alternative σ x > σ y
2 2

the decision rule is

Reject Ho if S2x > Fnx -1,ny- 1, ά
S2y
ii) To test the null hypothesis
Ho: σ 2x = σ 2y
against the alternative Ho: σ 2x ≠ σ 2y
the decision rule is
Reject Ho if S2x > Fnx -1,ny- 1, ά

S2y
Here, Fnx -1,ny- 1, ά is that number for which
P (Fnx -1,ny- 1, ά > Fnx -1,ny- 1, ά) = ά
where Fnx -1,ny- 1 has an F distribution with numerator degrees of freedom (nx -1)
denominator degrees of freedom (ny -1).

Example:
It is hypothesized that the market share of a corporation should vary more in an industry with
active price competition than in one with duopoly and tacit collusion. In a study of the steam
turbine generator industry, it was found that in 4 years of active price competition, the variance
of general electric’s market share was 114.0895. In the following 7 years, in which there was
duopoly and tacit collusion, this variance was 16.0780. If the two population variances are
denoted σ 2x and σ 2y test

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 89

 Ho σ 2x = σ 2y
Against
H1 σ 2x = σ 2y
Solution:
Assume that the data can be regarded as independent random samples from two normal
distributions.
The decision rule is to reject Ho in favor of H1 if
S2x > Fnx -1,ny- 1, ά
S2y

From the statement of the example, we have

nx = 4; S2x = 114.0895; ny = 7; S2y = 16.0780

and so S2x = 114.0895 = 7.10

S2y 16.0780
Since the degrees of freedom are (nx – 1) = 3 and (ny – 1) = 6, the cutoff points for 5% level and
1% level tests are
F3, 6,0.05 = 4.76 and F3,6,0.01 = 9.78 ,from a table.
Hence, the null hypothesis can be rejected at the 5% level, but not at the 1% level of
significance. The evidence is pretty strong against the hypothesis that variability in market
shares is the same in pretty strong against the hypothesis that variability in market shares is the
same in an industry with price competition as in one with duopoly and tacit collusion. It appears
that the data strongly support the alternative hypothesis of more variability in market shares
under price competition.

7.9 Measuring the Power of a Test

In sections 7.2 – 7.8, we have concentrated on the development of tests at a particular
significance level. That is, we have developed decision rules for which the probability of marking
a type I error- rejecting the null hypothesis when it is true is fixed at some reassigned value. As
noted in section 7.1, a decision rule of this kind will necessarily imply some probability of
making a Type II error accepting a null hypothesis which is false. Moreover, it is often important
to know what are the probabilities of making this kind of error, so that, if a null hypothesis is
accepted, we will have an assessment of how likely such a decision would be when that
hypothesis is false.
In this section we consider for the first time the characteristics of some of our tests when the
null hypothesis is true. In particular, we show how the power can be calculated for tests of the
mean of a normal distribution when the variance is known, and for tests of the population
proportion.

7.9.1 Tests of the Mean of a Normal Distribution: Population Variance Know

Suppose we have a random sample of n observations from a normal population with unknown
mean µ and known variance σ 2, and we want to test the null hypothesis
Ho: µ = µo
Against the alternative
H 1: µ > µo

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 90

 We saw in section 7.2 that a test of significance level ά is obtained from the decision rule
Reject Ho If X - µo > Zά
σ /√n
Hence, the probability that the null hypothesis will be rejected is
X − µ0
P( > Za
σ
n
This probability is, of course, equal to the significance level ά when the null hypothesis is true.
We now wish to determine its value when that hypothesis is false.
Assume, then, that the null hypothesis is false, and that the population mean is in fact some
number µ1, bigger than µ1 , in that case, the random variable
Z = X- µ
σ /√n
has a standard normal distribution. This fact follows us to calculate the power of the test. We
have, when the population mean is µ1,
Power = 1-β
= p(Null hypothesis rejected when it is false)
= P ( X - µo > Zά)
σ /√n
Z σ
= p( ( X > µ 0 + a )
n
X − µ1 µ 0 − µ1
= P( > + Za )
σ n σ n
µ 0 − µ1
= P( Z > + Za )
σ n
where Z is a standard normal random. This probability can be calculated using tables of the
standard normal cumulative distribution function, once µ1,ά, n and Z a are specified.

Example:
In example 7.1 we tested the null hypothesis that the population mean weight (in ounces) of ball
bearings was
Ho: µ = µo =5
against the alternative
H1: µ >5
The population standard deviation was σ = 0.1, and the test was based on n =16 observation.
The test was carried out at significance level ά = 0.05, so that
Zά = Z.05=1.645
We now determine the probability that our decision rule will reject the null hypothesis when
the rule mean weight is µ1 = [Link] power is then
1 – β = p (Z> µo+ µ1 + Zά)
σ /√n

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 91

 =P (z > 5.00-5.02 +1.645)
0.1/√16
=P (Z> 0.845) = 0.1991
From a table(This probability is obtained by interpolating between the z values 0.85 in table) .
Thus, the probability is 0.1991 that our decision rule will reject the null hypothesis when the true
mean weight is 5.02 ounces.
These power calculations are illustrated in figure 7.12, which shows the probability density
functions of the sample mean when the
population mean is 5 and 5.02.

Figure 7.12 sampling distributions of sample mean in example 7.11 for population mean µ0 =5 µ1=5.02.
Figures show calculation of power,1-β,corresponding to significance level ά= .05 for testing H0 : µ =5
against H1: µ> 5;power is evaluated at µ=5.02

Recall that a 5% level test rejects the null hypothesis when

X- µ0 > 1.645
σ /√n or, with µ0=5, σ =0.1, and n =16,when
X − 5 >1.645
0.1/4
This is equivalent to requiring
X > 5.045
Thus, when the null hypothesis is true, the probability that the sample mean exceeds 5.041 is
0.05. This is show in part (a) of figure 7.12. Part (b) of the figure shows the density function of the
sampling distribution of the sample mean when the population mean is 5.02 . It differs from part
(a) of the figure in being shifted to the right by an amount 0.02 the difference between the means
5.02 and 5. The shaded area in this figure shows the probability that the sample mean exceeds
5.041 when the population mean is 5.02. This is the power, evaluated at that point, as calculated
previously.
In a similar manner, such probabilities can be calculated for any value of µ[Link] powers are
shown in the table and graphed in Figure 7.13.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 92

µ1 5.00 5.01 5.02 5.03 5.04 5.05
Power .05 .1066 .1991 .3282 .4820 .6387
µ1 5.00 5.02 5.08 5.09 5.10 5.11
Power .7749 .8760 .9400 .9747 .9908 .9971

Figure 7.13 power function for example 9.11; test of H0: µ=5.00 against H1: µ>5.00(ά =0.05, σ =0.1,
n=16)

Figure7.14 power functions for test of H0: µ=5.00 against H1: µ>5.00(ά=0.05, σ = 0.1), shown for
sample sizes 4,9,16

We note that the power function has the following features:

(i) Everything else equal, the further is the true mean µ1 from the hypothesized mean µ0, the
greater the power of the test. This illustrated in figure 7.13, and means that we are more likely to
detect large than small discrepancies from the hypothesized mean.
(ii) Everything else equal, the smaller the significance level of the test, the smaller the power. In
other words, reducing the probability of a Type I error will increase the probability of a Type II
error.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 93

(iii) Everything else equal, the large the population variance, the lower the power of the test. We
ate less likely to detect small departures from the hypothesized mean when there is greater
variability in the population.
(iv) Everything else equal, the large the sample size, the greater the power of the test. Again, this
intuitively plausible. The more information obtained from the population, the greater the chance
of detecting any departure from the null hypothesis. This is illustrated in figure 7.14 .Together
with the power function derived in example 7.11 for samples of n=16 observation, this figure
shows also the power function for the same test based on n=4 and n=9 observations. Note that,
for every value of the population mean higher than the hypothesized mean the greater the
number of observations.
Using arguments similar to those given previously. We can also calculate the power of a test
against the two-sided alternative hypothesis. However, rather than pursuing that case here, we
will illustrate these calculate these calculations for tests on a population proportion.

7.9.2 Tests of the Population Proportion (Large Sample)

Assume that a random sample of n observations is taken from a population congaing a population
p of “successes,” and that the null hypothesis
H 0 : p = p0
is to be tested against the two-side alternative
H 1: P = P0
It was seen in section 7.5 that a test of significance level ά is obtained from the decision rule
P − P0
> Za PX − P0
Reject H0 if or < −Z a 2
P0 (1 − P0 ) n 2
P0 (1 − P0 ) n
Therefore, the probability that the null hypothesis will be rejected is
P − P0
> Za PX − P0
P[ or < −Z a 2 ]
P0 (1 − P0 ) n 2
P0 (1 − P0 ) n

P − P0
> Z a ] + P[ PX − P0
P[ < −Z a 2 ]
P0 (1 − P0 ) n 2
P0 (1 − P0 ) n
Against, this probability is equal to the significance level ά when the null hypothesis is true.
Suppose now that the null hypothesis is false and that the population proportion of successes is
p1, which differs from p0 .In that case, provided the sample size is large, we know that to a good
approximation, the random variable
PX − P1
Z=
P1(1 − P1 ) n
has a standard normal distribution. The power of the test can then be found as
P − P0
> Za + PX − P0
Power = 1-β= p [ < −Z a 2 ]
P0 (1 − P0 ) n 2
P0 (1 − P0 ) n

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 94

 P0 (1 − p 0 ) P (1 − P0 )
=P[ P X < P0 − Z a 2 ] + p[ P X > P0 + Z a 2 0 ]
n n

PX − P1 P0 − P1 P0 (1 − P0 ) / n
=P[ < − Za 2
P1 (1 − p1 ) n P1 (1 − P1 ) / n P1 (1 − P1 ) / n
PX − P1 P0 − P1 P0 (1 − P0 ) / n
+ > + Za 2 ]
P1 (1 − p1 ) n P1 (1 − P1 ) / n P1 (1 − P1 ) / n
P0 − P1 P0 (1 − P0 ) / n
=P [Z< − Za 2 ]
P1 (1 − P1 ) / n P1 (1 − P1 ) / n
P0 − P1 P0 (1 − P0 ) / n
+ P[ Z > + Za 2 ]
P1 (1 − P1 ) / n P1 (1 − P1 ) / n
Where Z is standard normal random variable. This result allows the calculation of power a
function of p1 =and n.

Example:
In Example 7.6 we tested the null hypothesis that population of earnings forecasts by financial
analysts that exceed the actual outcome is
H0: P= P0 =0 .50
against the alternative
H1 : P ≠ 0 .50

The test was based on a random sample of n=600 observations. For a test at significance level
ά= 0.05, we have
Zά/2 =Z.205= 1.96
We now determine the probability that, for a 5% level test, the null hypothesis will be rejected
when the rule population proportion is P1=0.52.
Substitution into the formula gives the power
1-β = P{[Z < 0.05 -0.52 ] - [1.96 √(0.50)(0.50)/600 ] ] + p [z > 0.05 -0.52 ]
√(0.52)(0.48)/600 √(0.52)(0.48)/600 √(0.52)(0.48)/600
+ [1.96 √(0.50)(0.50)/600 ] }
√(0.52)(0.48)/600
= p (Z< 2.94)+p(Z>.98)
= 0.0016+0.1635=0.1651 , from a table.
So, the probability is 0.1651 that this decision rule will reject the null hypothesis when the
population proportion is 0.52.
Similarly this probability can be calculated for any population proportion p1.
Figure 7.15 shows the power function for this example. Because the alternative hypothesis is two
side, the power function differs in shape from that of figure 7.13

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 95

Figure 7.15 power function for the above example; test of H0 : p=0.50 against
H1 : P0≠ 0.50(ά=0.05,n=600)
Here we are considering possible values of the population proportion ob either side of the
hypothesized value,0.05. As we see, the probability of rejecting the null hypostasis when it is false
increases the further is the true population proportion from the hypothesis value.

7.10 Some Comments on Hypothesis Testing

Classical hypothesis testing methods is probably used more frequently in empirical investigations
than any other statistical technique. The relative simplicity of the procedures, together with the
frequency of real-world problems for which competing hypotheses exist, render hypothesis
testing extremely appealing. Nevertheless, some caution is necessary in interpreting test results.
Perhaps of paramount concern is the crucial role played by the hull hypothesis in the hypothesis
testing framework. In a typical investigation, the significance level that is, the probability of
rejecting the null hypothesis when it is true- will be set at some low level. Evidence is then
gathered from the population to put this hypothesis in jeopardy. However, we might fail to reject
a drastically false null hypothesis simply because very little sample information is available, or
because the test employed has low power. This may be perfectly appropriate if the null
hypothesis is indeed special and the investigator is, for some reason, reluctant to abandon it. Such
might be the case, for example, if rejection of a null hypothesis would lead to costly modifications
in a production process. However, the special status of the null hypothesis is very often neither
warranted nor appropriate. In such circumstances a more symmetric treatment of competing
hypotheses would be desirable. One obvious possibility would be to take into account the actual
costs (if they could be quantified) of making both Type1 errors and Type 11 errors in deciding
between competing hypotheses.
On some occasions; very large amounts of sample information are available, and the opposite
problem arises the null hypothesis is put on very considerable jeopardy indeed. It is then
important to distinguish between the use of the word significant, as it is used in satirical
significance testing, compared with its dictionary definition. Suppose very large numbers of
households are sampled from two cities. One occasionally meets such statements as: “the
difference between the sample average annual incomes for the two cities is $2.37, this is
significant;” presumably what is meant by this statement is that the null hypothesis of equality of
the population means can, on the basics of such large samples, be reject at the usual significance
levels. Nevertheless, the best estimate available of the difference between the population mean
annual incomes is a mere $2.37, which is of no practical significance whatever!

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 96

The test developed in this chapter has typically been based on an assumption about the
underlying population distribution. Often we have assumed that distribution to be normal. When
such an assumption fails to hold, the procedures are not strictly valid. Specially, the true
significance levels will differ from those derived on the basis of the normal, it is important to ask
how badly the significance levels might be affected by no normality in the parent population. If the
effect is relatively small, the tests are said to be robust to departures from normality. Generally
speaking, it is known that tests on population means are robust. However, test (such as those of
section 7.8) that compare population variances are not.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 97

Sample Exam Question

Mekelle University
Faculty of Business & Economics
Department of Economics

Econ 232: Statistics For Economists Final Exam

Time allowed: 300 Hours

INSTRUCTIONS

⇒ Work out the following the following questions and show your steps clearly and neatly.
⇒ Legible & neat works credited more.
⇒ Make sure that the Exam contains four pages, two parts and ten questions.
⇒ Attempt all the questions.

To err is human, but costly!

Part I. Discussion Questions
Instruction l: Answer the following Question.

[Link] Distribution also called the Normal Probability Distribution

happens to be the most useful theoretical distribution for a continuous
random variable and it is called in particular a cornerstone of Modern Statistical
investigation/statistical Project in Business and Economics. Why Normal
distribution occupies such a central place in Modern Statistical
investigation?

2. Please explain what the following concepts are (In 3 lines)

a) confidence interval
b) central limit theorem
c) systematic random sampling
d) judgment sampling
e) Small population correction factor

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 98

Part II. Work out Questions
Instruction II: Work out the following questions on a separate answer Sheet
provided neatly

3.A certain local investor has $10,000 two invest and two agriculture investment opportunities,
each requiring a minimum of $[Link] return per $100 from the first alternative can be
represented by a random variable, X, having the following probability function:

X -5 20
P(X) 0.4 0.6

The Return per $100 from the second is given by the random variable, y, whose probability
function is:
Y 0 25
P(Y) 0.6 0.4
The random variable X and Y are independent. And the investor has the following three possible
strategies:
I. $10,000 in the first investment opportunity
II. $10,000 in the second investment opportunity
III. $ 5000 in each investment.
Given this answer the following:
a. What will be the mean and variance of return from each strategy?
b. Give your opinion about which strategy is promising for the investor? Why? Would you
necessarily advise the investor to adopt this strategy?

4.A certain Brewery’s beer bottles are not always filled to its capacity. The brewery
advertises that its bottles contain, on average 12 ounces of beer with a standard
deviation of 0.4 ounces. If a sample of 100 bottles was taken from the production
line, what is the probability of observing a sample mean of 11.9 ounces or less?

[Link] years UNESCO report indicated that the per-capita income of LDCs, including
Ethiopia, distributed Normally with mean annual income of $ 120 and standard
deviation of $20. But an Economist from African Economic Commission (AEC)
challenged this report. According to the Economist, there is/was an absolute
poverty in LDCs, especially, in Ethiopia, where most of the citizens earn mean
annual income below Absolute Poverty Line, $100. To disproof the UNESCO report,
he collect the following 32 sample Ethiopians` annual incomes by using
Area (Cluster) sampling method. The following data shows the result of the
sampling.

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 99

 180, 145, 100, 135, 90, 340, 118, 240
95, 150, 280, 200, 110, 190, 210, 105
250, 300, 99, 108, 100, 110, 104, 109
130, 125, 290, 120, 350, 108, 102, 112

Using the above information deduce whether there is an absolute poverty in

Ethiopia or not. Based on the above sample result, if there is What type of Policy
Measures must be taken to eliminate absolute?

[Link] studies indicated that the reason for most new products market failures is
associated with the setting of price higher than the consumers willingness to pay.
To build this hypothesis a random sample of 89 new product market failures were
taken and the Managers associated with this assessment indicated that the main
cause for Market failure for the sample is 18.2%. Find a 95% confidence interval
for the proportion of all market failures for which high price is held to be the main
cause.

7. From 5000 new extension package adopters 600 are observed not to benefit from
the extension package. And 50 interviewers are sent to 50 different villages to
collect data on 100 farmers each. After reaching on the site they use a random
number to sample the farmers. How many of the interviewers are expected to
find 5 to 8 farmers who do not benefit from the extension package. (Makes sure
that you explain why you use a given theoretical distribution).

8. A town has 50,000 population and small studies show that 30% of the population
expects inflation to be much higher next year but not the rest. But we want to
conform it by study which includes 1000 people. So we first get the list of
all individuals, we gave them an identification number and we pick them by lottery
with out replacement. What are the chances that our sample will include 400 to
500 individuals which does not expect prices to rise? (Makes sure that you
explain why you use a given theoretical distribution)

9. If the mean wheat price is 90 Birr and it has a standard deviation of 20 Birr and
the prices are highly skewed to the right. How large a sample size is need to be,
if we want with 90% confidence that sample average price to be in range of 80
to [Link] interpret the result and the confidence interval.

[Link] distribution of income among 100 individuals in a given area is normal and
20 of them are randomly selected and the mean and standard deviation is found
to be 200 Birr and 100 Birr, what is the 90% confidence interval for the
population mean income. Could you compare this result with the point estimate
of income?

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 100

 Bonus questions: Attempt any (only) two of the three bonuses

1. A Company receives a shipment of 20 items. Because inspection of each

individual item is expensive, it has a policy of checking a random sample of
6 items from such a shipment, accepting delivery if no more than one
sampled item is defective. What is the probability that a shipment with 5
defective items will be accepted?

2. In package with 10 products 2 products are defected. A firm wants to make

sure that with 70% probability that it have to reject the defected package,
how large should be the sample size in order to achieve such 70% chance
of rejecting the package with defect.

3. Before 10 years the annual income of Ethiopian farmers was estimated to be

Birr 686 per Household. Suppose that, The Ethiopian Development Research
Institute (EDRI) believes that this value has increased since then. To test
this belief (hypothesis), (EDRI) randomly samples 23 farm households across
the country and records the annual income of these farm households. The data for each
farm household are shown (in Birr) below.
750, 680, 845, 1100, 735, 800,
845, 1150, 780, 800, 910, 790,
950, 600, 850, 900, 1000, 900
900, 850, 990, 1200, 850
Test the EDRI belief (theory) that farmers are now earning more than Birr 686
per year at 5% level of significance.
4. What is degree of freedom for a statistician?
GOOD LUCK!

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 101

Revised by: Achamyeleh Tamiru, MU, FBE, Department of Economics 102