Solved Problems and Exercises. Part 2

Гумбат Наби оглы Алиев
СТАТИСТИКА
Задачи и упражнения. Часть 1
Humbat Nabi oglu Aliyev
STATISTICS
SOLVED PROBLEMS AND EXERCISES. PART 2
Подписано в печать 07.02.2006
Формат 60 х 84 1/16. Бумага тип. Ризограф.

Усл.печ.15,1 л. Уч.-изд. 15,3 л. Тираж 500.
Заказ № 700
Цена договорная
Издание Казахской головной архитектурно-строительной академии

Издательский дом КазГАСА «Строительство и архитектура»
480043, г.Алматы, ул. К.Рыскулбекова, 28
Süleyman Demirel University
Faculty of Economics
H.N.Aliyev
STATISTICS
Solved problems and exercises
Part 2
Almaty – 2006
ББК 60.6
A 36
Рекомендовано к печати Ученым Советом Университета имени

Сулеймана Демиреля.
Aliyev H.N..
А 36 Statistic. Solved problems and exercises. Part 2. –
Алматы, 2006 - 245с.
ISBN № 9965-9605-7-7
ББК 60.6
0702000000
A
00 (05)-05
ISBN № 9965-9605-7-7
© Aliyev H.N., 2006

© Университет имени
Сулеймана Демиреля, 2006
Exercises
1. Explain which of the following is a two tailed test, a left tailed test, or a right tailed test.
a) H 0 :   25, H 1 :   25
b) H 0 :   134 , H 1 :   134
c) H 0 :   16 , H 1 :   16
Show the rejection and nonrejection regions for each of these cases by drawing a sampling distribution
curve for the sample mean, assuming that sample size is large in each case.
2. Consider H 0 :   35, against H 1 :   35 .
a) What type of error would you make if the null hypothesis is actually false and you fail to reject it?
b) What type of error would you make if the null hypothesis is actually true and you reject it?
3. For each of the following rejection regions, sketch the sampling distribution for z and indicate the
location of rejection region.
a) z  2.05 ; b) z  2.75 ; c) z  1.28 ;
d) z  2.13 ; f) z  2.575 or z  2.575 ;
g) z  1.82 or z  1.82
4. Write the null hypothesis and alternative hypothesis for each of the following examples. Determine if
each is a case of a two tailed, a left tailed, or a right tailed test.
a) To test whether or not the mean price of houses in a certain city is greater than $ 45 000.
b) To test if the mean number of hours spent working per week by students who hold jobs is different
from 18 hours.
c) To test whether the mean life of a particular brand of auto batteries is less than 28 days.
d) To test if the mean amount of time taken by all workers to do a certain job is more than 45 minutes.
f) To test the mean age of all managers of companies is different from 40 years.
g) To test the mean time for an airline passenger to obtain his or her luggage, once luggage starts coming
out the conveyer belt, is less than 180 seconds.
CONTENTS
Chapter 1. Hypothesis testing ……………………………..7

1.1. Introduction…………………………………………………...7
1.12. Concepts of hypothesis testing ……………………………...7
1.13. The null and alternative hypothesis………………………… 8
1.14. Tails of the test…………………………………………………...11
Exercises…………………………………………………………..14
1.2. Tests of the mean of a normal distribution: Population variance
known …………………………………………………..………………..15
unknown (Large sample size)……………………………………..18
Exercises………………………………………………..…………21
1.4. Hypothesis testing using the p –value approaches …………..23
Exercises…………………………………………………………..25
unknown. Small samples ………………………………………….26
Exercises………………………………………………………… 29
1.6. Tests of the population proportion (Large sample) ……….… 32
Exercises…………………………………………………………...34
1.7. Tests of the variance of a normal distribution……………… 36
Exercises………………………………………………………… 38
1.8. Tests for the difference between two population means…… ..40
1.8.1. Tests based on paired samples………………………………40
Exercises…………………………………………………………...43
1.8.2. Tests based on independent samples (Known variance
or large sample size)……………………………………………… 44
Exercises………………………………………………………… ..47
1.8.3. Tests based on independent samples (Population variances
are unknown and equal)…………………………………………...49
Exercises……………………………………………………………52
1.9. Tests for the difference between two population proportions
(Large samples) …………………………………………………….55
Exercises…………………………………………………………….59
Chapter 2. Some nonparametric tests..…..……………………...61
2.1. Introduction…………………………………………………….61
2.2.1. The Sign test for paired or matched samples………….… ….62
2.2.2. The sign test: Normal approximation (Large samples) ..…… 65
3
Exercises……………………………………………………………67
2.3. The Wilcoxon signed test ………………………….………….70
2.3.1. The Wilcoxon signed test for paired samples
(small sample size)………………………………………………....70
(large sample size)………………………………...……………….72
Exercises……………………………………………………………74
2.4. The Mann-Whitney test………………………………………..77
Exercises……………………………………………………………82
Chapter 3. Simple linear regression…..………….…………..84
3.1. Introduction ……………………………………………………84
3.2. The scatter diagram……………….…………………….……...84
3.3. Correlation analysis…………………………………………….86
3.3.1. Hypothesis test for correlation ……………………………….89
Exercises…………………………………………………………… 91
3.4. Spearman rank correlation …………………………………......93
Exercises…………………………………………………………… 95
3.5. The linear regression model…………………………………… 96
3.5. 1. Least squares coefficient estimators…………………… …. 98
3.5.2. Least square procedure …...………………………………… .99
3.5.3. Interpretation of a and b ……………………...……………...102
3.5.4. Assumptions of the regression model ……………………….103
Exercises…………………………………………………………….104
3.6. The explanatory power of a linear regression equation ……….106
3.6.1. Coefficient of determination R 2 ………………………..…...109
3.6.2. Estimation of model error variance …… …………………110
Exercises………………………………………………………… 111
3.7. Statistical inference: Hypothesis tests and confidence
intervals…………………………………………………………….113
3.7.1. Hypothesis testing about  ………………………………....114
3.7.2. Confidence intervals for the population regression slope  ..116
Exercises…………………………………………………………….117
3.8. Using the regression model for prediction a particular
value of y……………………………………………………………120
Exercises………………………………………………………… .123
Chapter 4. Multiple regression analysis …………………….125
4.1. Introduction…………………………………………………….125
4
4.2. Multiple regression model……………………………… ……125
4.3. Standard assumptions for the multiple regression models …..127
4.4. The explanatory power of a multiple regression equation…..…127
4.4.1. Estimation of error variance distribution…………….………127
4.4.2. The coefficient of determination……………………………..128
4.4.3. Adjusted coefficient of determination………………………..129
4.4.4. Predictions from the multiple regression models…………….130
Exercises…………………………………………………………….130
4.5. Computer solution of multiple regressions …………………….132
4.6. Confidence interval for individual coefficients …………….….138
Exercises…………………………………………………………….139
4.7. Test of hypothesis about individual coefficients ……………….143
4.8. Tests on sets of regression parameters………………………….145
Exercises……………………………………………………………. 146
4.9. Dummy variables in the regression models…………………….150
Exercises…………………………………………………………….155
Chapter 5. Analysis of variance (ANOVA) …………………..159
5.1. Introduction………………………..…………………………...159
5.2. One-way analysis of variance …….…………………………...159
Exercises…………………………………………………………….168
5.3. The Kruskal-Wallis test …………………….………………….171
Exercises…………………………………………………………….173
5.4. Two-way analysis of variance …….…………………………...175
Exercises…………………………………………………………….182
Chapter 6. Statistical quality control ………………………185
6.1. Introduction……………………………………………………..185
6.2. Variation ……………………………………….………………185
6.3. Control charts …………………………………………………..186
6.3.1. Control charts for means and standard deviations…………….187
6.4. Interpretation of control charts………………………………… 192
Exercises……………………………………………………………..193
6.5. Control charts for proportions …………...……………………..195
Exercises……………………………………………………………..197
6.6. Control charts for number of occurrences: c-chart ……………..199
Exercises……………………………………………………………..201
Chapter 7. Time series analysis and forecasting …..…..… 203
7.1. Introduction to index numbers ……...………………………….203
7.1.1. Price index for a single item (Simple index number)…………203
5
7.1.2. Unweighted aggregate price index…………………………....203
7.1.3. A weighted aggregate price index…………………………… 205
7.1.4. A weighted aggregate quantity index…………………………206
7.2. Commonly used index numbers………………………………...206
7.3. Deflating a series by price indexes……………………………...207
Exercises……………………………………………………………..209
7.4 A nonparametric test for randomness……………………………211
7.4.1. The runs test for the small sample sizes………………………211
7.4.2. The run test for the large sample sizes………………………..212
Exercises…………………………………………………………….213
7.5. Components of time series……………………………………...216
7.6. Moving averages………………………………………………..217
Exercises…………………………………………………………….220
7.7. Exponential smoothing…………………………………………221
Exercises…………………………………………………………….224
7.8. Double exponential smoothing (Holt-Winters exponential
forecasting model)…………………………………………….……..226
Exercises…………………………………………………………….230
Appendix………………………………………………………… ....232
References………………………………………………………….. 253
6
Chapter 1
Hypothesis testing
1.1. Introduction
Inferential statistics consists of methods that use sample results to

help make decisions or predictions about a population. The point and interval
estimation procedures are forms of statistical inference. Another type of
statistical inference is hypothesis testing. In hypothesis testing we begin by
stating a hypothesis about a population characteristic. This hypothesis, called
the null hypothesis, is assumed to be true unless sufficient evidence can be
found in a sample to reject it. The situation is quite similar to that in a
criminal trial. The defendant is assumed to be innocent; if sufficient evidence
to the contrary is presented, however, the jury will reject this hypothesis and
conclude that the defendant is guilty.
In statistical hypothesis testing, often the null hypothesis is an
assumption about the value of a population parameter. A sample is selected
from the population, and a point estimate is computed. By comparing the
value of the point estimate to the hypothesized value of the parameter we
draw a conclusion with respect to whether or not there is a sufficient
evidence to reject the null hypothesis. A decision is made and often a
specific action is taken depending upon whether or not the null hypothesis
about the population parameter is accepted or rejected.
1.1.2. Concepts of hypothesis testing
Let us consider example about coffee cans. A company may claim that, on
average, its cans contain 100 grams of coffee. A government agency may
want to test whether or not such cans contain, on average, 100 grams of
coffee.
Suppose we take a sample of 50 cans of the coffee under investigation. We
then find out that the mean amount of coffee in these 50 cans is 97 grams.
Based on these results, can we state that on average, all such cans contain
less than 100 grams of coffee and that the company is lying to the public?
Not until we perform a test of hypothesis. The reason is that the mean

x  97 grams is obtained from the sample. The difference between 100
grams (the required amount for the population) and 97 grams (the observed
7
average amount for the sample) may have occurred only because of the
sampling error. Another sample of 100 cans may give us a mean of 105
grams. Therefore, we make a test of hypothesis to find out how large the
difference between 100 grams and 97 grams is and to investigate whether or
not this difference has occurred as a result of chance alone. If 97 grams is the
mean of all cans and not for only 100 cans, then we do not need to make a
test of hypothesis. Instead, we can immediately state that the mean amount
of coffee in all such cans is less than 100 grams. We perform a test of
hypothesis only when we are making a decision about a population
parameter based on the value of a sample statistic.
1.1.3. The null and alternative hypothesis
We will begin our general discussion by using  to denote a

population probability distribution parameter of interest, such as the mean,
variance, or proportion. Our discussion begins with a hypothesis about the
parameter that will be maintained unless there is strong contrary evidence. In
statistical language it is called the null hypothesis.
For example, we might initially accept company’s claim that on
average, the contests of the cans weight at least 100 grams. Then after
collecting sample data this hypothesis can be tested. If the null hypothesis is
not true, then some alternative must be true. In carrying out a hypothesis test
the investigator defines an alternative hypothesis against which the null
hypothesis is tested.
For this coffee cans example a likely alternative is that on average
can’s weights are less than 100 grams. These hypotheses are chosen such
that one or the other must be true. The null hypothesis will be denoted as
H 0 and the alternative hypothesis as H 1 .
Definition: A null hypothesis is a claim (or statement) about a population
parameter that is assumed to be true until it is declared false.
Definition: An alternative hypothesis is a claim about population
parameter that will be true if the null hypothesis is false.
Our analysis will be designed with the objective of seeking strong evidence
to reject the null hypothesis and accept the alternative hypothesis. We will
only reject the null hypothesis when there is a small probability that the null
hypothesis is true. Thus rejection will provide strong evidence against
H 0 and in favor of the alternative hypothesis, H 1 . If we fail to reject
8
H 0 then either H 0 is true or our evidence is not sufficient to reject H 0 and
hence accept H 0 . Thus we will be more comfortable with our decision if we
reject H 0 and accept H 1 .
A hypothesis, whether null or alternative, might specify a single
value, say  0 , for the population parameter  . In that case, the hypothesis is
said to be a simple hypothesis designated as
H 0 :  0
That is read as, “The null hypothesis is that the population parameter  is
equal to the specific value  0 ”.
Alternatively, a range of values might be specified for unknown parameter.
We define such hypothesis as a composite hypothesis, and it will hold true
for more than one value of the population parameter. In many applications, a
simple null hypothesis, say
H 0 :  0
is tested against a composite alternative. One possibility would be to test the
null hypothesis against the general two-sided hypothesis
H1 :    0
In other cases, only alternatives on one side of the null hypothesis are of
interest. For example, a government agency would be perfectly happy if the
mean weight of coffee cans greater than 100 grams. Then we could write the
null hypothesis as
H 0 :  0
and the alternative hypothesis of interest might be
H1 :    0
We call these hypothesis one- sided composite alternatives.
Example:
A company intends to accept the product unless it has evidence to suspect
that more than 10% of products are defective. Let  denote the population
proportion of defectives. The null hypothesis is that the proportion is less
than 0.1, that is
H 0 :   0 .1
and the alternative hypothesis is
9
H 1 :   0.1
The null hypothesis is that the product is of adequate quality overall, while
the alternative is that the product is not adequate quality. In this case the
product would only be rejected if there is strong evidence that there are more
than 10% defectives.
Once we have specified a null hypothesis and alternative hypothesis and
collected sample data, a decision concerning the null hypothesis must be
made. We can either accept the null hypothesis or reject it in favor of the
alternative. For good reasons many statisticians prefer not to use the term
“accept the null hypothesis” and instead say “fail to reject”. When we accept
or fail to reject the null hypothesis, then either the hypothesis is true or our
test procedure was not strong enough to reject and we have committed an
error. When we use the term accept a null hypothesis that statement can be
considered shorthand for failure to reject.
From our discussion of sampling distributions, we know that the
sample mean is different from the population mean. With only a sample
mean we can not be certain of the value of the population mean. Thus the
decision rule we adopt will have some chance of reaching an erroneous
conclusion. One error we call Type I error. Type I error is defined as the
rejection of the null hypothesis when the null hypothesis is true. We will see
that our decision rules will be defined so that the probability of rejecting a
true null hypothesis, denoted as  , is “small”. The probability,  , is
defined as the significance level of the test. Since the null hypothesis is
either accepted or rejected, it follows that the probability of accepting the
null hypothesis when it is true is (1   ) . The other possible error, called
Type II error, arises when false null hypothesis is accepted. We say that for
a particular decision rule, the probability of making such an error when the
null hypothesis is false is denoted  . Then, the probability of rejecting a
false null hypothesis is (1   ) which is called the power of test.
10
Type I error
A type I error occurs when a true null hypothesis is rejected. The value
 represents the probability of committing this type of error, that is
  P( H 0 is rejected / H 0 is true)
The value  represents the significance level of the test.
Type II error
A Type II error occurs when a false null hypothesis is not rejected. The
value  represents the probability of committing a Type II error,
that is
  P( H 0 is not rejected / H 0 is false)
The value (1   ) is called the power of the test. It represents the
probability of not making a Type II error.
1.1.4. Tails of the test
In statistics, the rejection region for a hypothesis testing problem can be on

both sides with non rejection region in the middle, or it can be on the left
side or in the right side of the non rejection region. A test with two rejection
regions is called a two tailed test, and a test with one rejection region is
called a one tailed test. The one tailed test is called a left tailed test if the
rejection region is in the left tail of the distribution curve, and a right tailed
test if the rejection region is in the right tail of the distribution curve.
a) A two tailed test
Example:
The mean family size in a particular country was 3.75 in 1990. We want to
check whether or not this mean has changed since 1990. The mean family
size has changed if it has either increased or decreased during this period.
This is an example of two tailed test. Let  be the current mean family size
for all families. We write the null and alternative hypothesis for this test as
H 0 :   3.75 (The mean family size has not changed)
H 1 :   3.75 (The mean family size has changed)
11
This shaded This shaded
area is  / 2 area is  / 2
_
x
  3.75
Rejection Non rejection Rejection
region region region
Fig.1.1
As shown in Figure 1.1, a two tailed test has two rejection regions, one in
each tail of the distribution curve.
b) A left tailed test
Reconsider the example of the mean amount of coffee can produced by

company. The company claims that these cans, on average, contain 100
grams of coffee. However, if these cans contain less than the claimed amount
of coffee, then the company can be accused of cheating. Suppose that the
government agency wants to test whether the amount of coffee can is less
than 100 grams. Note that the key phrase this time is less than, which
indicates a left tailed test. Let  be the mean amount of coffee in all cans.
The null and alternative hypothesis for this test are written as
H 0 :   120 grams (The mean is not less than 120 grams)
H 1 :   120 grams (The mean is less than 120 grams)
In this case, we can also write the null hypothesis as H 0 :   120 grams .
This will not affect the result of the test as long as the sign in H 1 is less than.
When the alternative hypothesis has a less than (<) sign, as in this case, the
test is always left tailed. In a left tailed test the rejection region is always in
the left tail of the distribution curve, as shown in Figure 1.2.
12
Shaded area is 
  120
Rejection Nonrejection region
region
Fig.1.2
c) A right tailed test
Suppose that mean monthly income of all households was 45 500 tg in 2001.
We want to test if current income of all households is higher than 45 500 tg.
The key phrase in this case is higher than, which indicates a right tailed test.
Let  be the mean income of all households.
We write the null and alternative hypothesis for this test as
H 0 :   45500 (The current income is not higher than 45 500 tg)
H 1 :   45500 (The current income is higher than 45 500 tg)
In this case, we can also write the null hypothesis as H 0 :   45500 , which
states that current mean income is either equal to or less than 45 500 tg.
Again, the result of the test will not be affected whether we use an equal to
(=) or a less or equal to (  ) sign in H 0 as long as the alternative hypothesis
has a greater than (>) sign.
_
x
  45500
Nonrejection region Rejection region
Fig.1.3
13
When an alternative hypothesis has a greater than (>) sign, the test is always
right tailed. As shown in the Fig. 1.3, in a right tailed test, the rejection
region is in the right tail of the distribution curve. The area of this rejection
region is equal to  , the significance level. We will reject H 0 if the value of
_
x obtained from the sample falls in the rejection region. Otherwise, we will
not reject H 0 .
Remark: Note that the null hypothesis always has an equal to (=) or a less
or equal to () or a greater than or equal to () sign and the alternative
hypothesis always has a not equal to () or a greater than (>) or a less than
(<) sign.
Exercises
1. Explain which of the following is a two tailed test, a left tailed test, or a
right tailed test.
a) H 0 :   25, H 1 :   25
b) H 0 :   134 , H 1 :   134
c) H 0 :   16 , H 1 :   16
Show the rejection and nonrejection regions for each of these cases by
drawing a sampling distribution curve for the sample mean, assuming that
sample size is large in each case.
2. Consider H 0 :   35, against H 1 :   35 .
a) What type of error would you make if the null hypothesis is actually false
and you fail to reject it?
b) What type of error would you make if the null hypothesis is actually true
and you reject it?
3. For each of the following rejection regions, sketch the sampling
distribution for z and indicate the location of rejection region.
a) z  2.05 ; b) z  2.75 ; c) z  1.28 ;
d) z  2.13 ; f) z  2.575 or z  2.575 ;
g) z  1.82 or z  1.82
14
4. Write the null hypothesis and alternative hypothesis for each of the
following examples. Determine if each is a case of a two tailed, a left tailed,
or a right tailed test.
a) To test whether or not the mean price of houses in a certain city is greater
than $ 45 000.
b) To test if the mean number of hours spent working per week by students
who hold jobs is different from 18 hours.
c) To test whether the mean life of a particular brand of auto batteries is less
than 28 days.
d) To test if the mean amount of time taken by all workers to do a certain job
is more than 45 minutes.
f) To test the mean age of all managers of companies is different from 40
years.
g) To test the mean time for an airline passenger to obtain his or her luggage,
once luggage starts coming out the conveyer belt, is less than 180 seconds.
1.2. Tests of the mean of a normal distribution:

Population variance known
In this and following sections we will present specific procedures for

developing and implementing hypothesis test procedures with applications to
business and economic problems.
We are given a random sample of n observations from a normal population
_
with mean  and known variance  2 . If the observed sample mean is x ,
then the test statistic is
_
x 
T .S .  z 
/ n
and we can use the following tests with significance level  .
1. To test either null hypothesis
H 0 :    0 or H 0 :    0 against the alternative
H1 :   0
the decision rule is
Reject H 0 if T .S .  z
15
H 0 :   0 or H 0 :    0 against the alternative
H1 :   0
Reject H 0 if T .S .   z
3. To test the null hypothesis
H 0 :    0 against the two sided alternative
H1 :   0
Reject H 0 if T .S .  z / 2 or T .S .   z / 2 ,
where z / 2 is the number for which
P ( Z  z / 2 )   / 2
and Z is the standard normal distribution.
A statistical test of hypothesis procedure contains the following five steps:
1. State the null and alternative hypothesis
2. Select the distribution to use
3. Determine the rejection and nonrejection regions
4. Calculate the value of the test statistic
5. Make a decision.
Example:
A manufacturer of detergent claims that the content of boxes sold weigh on
average at least 160 grams. The distribution of weights is known to be
normal, with standard deviation of 14 grams. A random sample of 16 boxes
yielded a sample mean weight of 158.9 grams. Test at the 10% significance
level the null hypothesis that the population mean is at least 160 grams.
Solution:
_
Let  be the mean average of all boxes and x be the corresponding mean
for the sample.
_
n  16 ;   14 ; x  158.9
The significance level is  is 0.1. That is, the probability of rejecting the
null hypothesis when it is actually is true should not exceed 0.1. This is the
probability of making a Type I error. We perform the test of hypothesis
using the five steps as follows.
16
Step 1. State the null and alternative hypothesis
We write the null and alternative hypothesis as
H 0 :   160 grams
H 1 :   160 grams
Step 2. Select the distribution to use
_
x 
Since population standard deviation is known we will use z  .
/ n
Step 3. Determine the rejection and nonrejection regions
The significance level is 0.1. The < sign indicates that the test is left tailed.
We look for 0.9 from in the standard normal distribution table, (Table 1 of
Appendix). The value of z is  1.28 . (Fig. 1.4).
Step 4. Calculate the value of the test statistic
The decision to reject or not to reject the null hypothesis will depend on
whether the evidence from the sample falls in the rejection or nonrejection
_
region. If the value of the sample mean x falls in rejection region, we
reject H 0 . Otherwise we do not reject the null hypothesis. To locate the
_ _
position of x  158.9 on the sampling distribution curve of x in Figure 1.4
_
we first calculate z value for x  158.9 . This is called the value of the test
statistic.
_
x  158.9  160
T .S .  z    0.31
/ n 14 / 16
  0.1
_
x
  160
Rejection region Nonrejection region
z
-1.28 0
Fig.1.4
17
Step 5. Make a decision
In the final step we make a decision based on the value of the test statistic
_
T .S.  z for x in previous step. This value of z  0.31is not less than the
critical value of z  1.28 , and it falls in the nonrejection region. Hence we
accept H 0 and conclude that based on sample information, it appears that the
mean weight of all boxes is greater than 160 grams.
By accepting the null hypothesis we are stating that the difference between
_
the sample mean x  158.9 and the hypothesized value of the population
mean   160 is not too large and may occurred because of the chance or
sampling error. There is a possibility that the mean weight is less than 160
grams, by the luck of the draw, we selected a sample with a mean that is not
too far from required mean of 160 grams.

Population variance unknown (Large sample size)
When the population standard deviation is unknown, we simply estimate

 with the value of the sample standard deviation s . We must consider
separately the large sample ( n  30 ) and small sample size ( n  30) cases.
If the sample size n is large, the test procedure developed for the case when
population variance is known can be employed when it is unknown,
replacing  2 by the observed sample variance s . All the hypotheses and
decision rules are stated in the same way as before (i.e. when  2 is known).
Example:
When a machine that is used to make bolts is working properly, the mean
length of these bolts 2.5 cm. However, from time to time this machine falls
out of alignment and produces bolt that have a mean length of either less
than or 2.5 cm or more than 2.5 cm. When this happens, the process is
stopped and the machine is adjusted. To check whether or not the machine is
producing bolts with a mean length of 2.5 cm, the quality control department
at the company takes a sample of bolts each week and makes a test of
hypothesis. One such a sample of 49 bolts produced a mean length of
2.49 cm and a standard deviation of 0.021 cm. Using the 5% significance
level, can we conclude that the machine needs to be adjusted?
18
Solution:
_
Let  be the mean length bolts made on this machine and x be the
corresponding mean for the sample.
_
n  49 ; x  2.49 cm; s  0.021 cm
The mean length of all bolts is supposed to be 2.5 cm. The significance level
is  is 0.05. That is, the probability of rejecting the null hypothesis when it
is actually is true should not exceed 0.05.
We are testing to find whether or not the machine needs to be adjusted. The
machine will need an adjustment if the mean length of these bolts is either
less than 2.5 cm or more than 2.5.
H 0 :   2.5 cm (The machine does not need adjustment)
H 1 :   2.5 cm (The machine needs an adjustment)

_
Because the sample size is large (n >30), the sampling distribution of x is
_
x 
(approximately) normal. Consequently we will use z  to make the
s/ n
test.
The significance level is 0.05. The  sign indicates that the test is two tailed
with two rejection regions, one in each tail of the normal distribution curve
_
of x . Because the total area of both rejection regions is 0.05 (the significance
level), the area of the rejection region in each tail is 0.025.
These areas are shown in Fig.1.5. To find the z values for these critical
points, we look for 0.975 in the standard normal distribution table.
Hence, the z values of the two critical points as shown in Fig.1.5, are  1.96
and 1.96.
19
 / 2 =0.025  / 2 =0.025
_
x
  2.5
Reject H 0 Reject H 0
Do not
reject H 0
z
-1.96 1.96
Fig.1.5

_
The value of x from the sample is 2.49. As  is not known, we calculate the
z value as follows
_
x  2.49  2.5
T .S .  z    3.33
s/ n 0.003
z  3.33 is the value of the test statistic.
The value of z  3.33 is less than the critical value of z  1.96 , and it falls
in the rejection region in the left tail. Hence we reject H 0 and conclude that
based on sample information, it appears that the mean length of all bolts
produced on this machine is not equal to 2.5 cm. Therefore, the machine
needs to be adjusted
By rejecting the null hypothesis we are stating that the difference between
_
the sample mean x  2.49 and the hypothesized value of the population
mean   2.5 is too large and may not have occurred because of the chance
or sampling error. This difference seems to be real and, hence the mean
length of bolts is different from 2.5 cm. Note that the rejection of the null
hypothesis does not necessarily indicate that the mean length of bolts is
definitely different from 2.5 cm. It simply indicates that there is strong
evidence (from sample) that the mean length of bolts is not equal to 2.5 cm.
20
There is a possibility that the mean length of bolts equal to 2.5 cm. If so, we
have wrongfully rejected the null hypothesis H 0 . This is Type I error and
probability of making such an error in this case is 0.05.
Exercises
1. Make the following tests of hypotheses.

_
a) H 0 :   25 ; H 1 :   25 ; n  81 ; x  28 ; s  3;   0.01
_
b) H 0 :   12 ; H 1 :   12 ; n  45 ; x  11 ;   4.5 ;   0.05
_
c) H 0 :   40 ; H 1 :   40 ; n  100 ; x  46 ; s 7;   0.1
2. Consider H 0 :   100 ; against the two sided alternative H1 :   100 .
a) A random sample of 64 observations produced a sample mean of 98 and a
standard deviation of 12. Using   0.01 , would you reject the null
hypothesis?
b) Another random sample of 64 observations taken from the same
population produced a sample mean of 104 and a standard deviation of 10.
Using   0.01 , would you reject the null hypothesis?
Comment on the results of parts a) and b).
3. A survey showed that people with a bachelor’s degree earned average of
$2116 a year in 2001. A sample of 900 persons with a bachelor’s degree
taken recently by a researcher showed that the persons in this sample earned
on average of $2345 a year with a standard deviation of $210. Test at 5%
significance level whether people with a bachelor’s degree currently earn an
average of $2116 against the alternative that it is more than $2116 in a year.
4. The manufacturer of a certain brand of auto batteries claims that the
mean life of these batteries is 45 month. A consumer protection agency that
wants to check this claim took a random sample of 36 such batteries and
found the mean life for this sample is 43.75 month with a standard deviation
of 4 month. Using the 2.5% significance level, test the manufacturer claim
against the alternative that the mean life of batteries is less than 45 month.
5. A random sample of 100 observations from a population with standard
deviation 60 yielded a sample mean of 110.
a) Test the null hypothesis that   100 against the alternative hypothesis
that   100 using   0.05 . Interpret the results of the test.
21
b) Test the null hypothesis that   100 against the alternative hypothesis
that   100 using   0.05 . Interpret the results of the test.
c) Compare the results of the two tests you conducted. Explain why the
results differ.
6. In a random sample of 250 observations, the mean and standard deviation
are found to be 169.8 and 31.6, respectively. Is the claim that  larger than
169 substantiated by these data at the 10% level of significance?
7. From records, it is known that the duration of treating a disease by a
standard therapy has a mean of 15 days. It is claimed that a new therapy can
reduce the treatment time. To test this claim, the new therapy is tried on 70
patients, and from the data of their times to recovery, the sample mean and
standard deviation are found to be 14.6 and 3.0 days, respectively.
Perform the hypothesis test using a 2.5% level of significance.
8. Suppose that you are to verify the claim that   20 on the basis of a
random sample of size 70, and you know that   5.6 .
_
a) If you set the rejection region to be x  21.31 , what is the level of
significance of your test?
_
b) Find the numerical value of c so that the test x  c has a 5% level of
significance.
Answers
1. a) T .S.  9.00 ; reject H 0 ; b) T.S.  1.49 ; do not reject H 0 ;c) T .S.  8.57;
reject H 0 ; 2. a) T.S.  1.33 ; do not reject H 0 ; b) T .S.  3.20 ; reject H 0 ;
3. T.S.  32.71; reject H 0 ; 4. T.S.  1.87 ; accept H 0 ; 5. a) z  1.67 ;
reject H 0 ; b) z  1.67 ; accept H 0 ; 6. T .S.  0.4 ; accept H 0 ;7. z  1.12 ;
H 0 is not rejected at   0.025 ; 8. a)   0.025 ; b) c=21.10.
22
1.4. Hypothesis testing using the p –value approaches
In previous section, the value of the significance level  was

selected before the test performed. Sometimes we may prefer not to
predetermine  . Instead, we may want to find a value such that a given null
hypothesis will be rejected for any  greater than this value and it will not be
rejected for any  smaller than this value. In this approach, we calculate the
p-value for the test, which is defined as the smallest level of significance at
which the given null hypothesis is rejected.
Definition:
The p-value is the smallest significance level at which the null hypothesis is
rejected.
Using the p-value approach, we reject the null hypothesis if
p- value  
and we do not reject the null hypothesis if
p- value  
Steps necessary for calculating the p-value for a test of hypothesis
1. Determine the value of the test statistic T .S.  z corresponding to the result
of the sampling experiment.
2.
a) If the test is one- tailed, the p-value is equal to the tail area beyond z in the
same direction as the alternative hypothesis. Thus, if the alternative
hypothesis is of the form >, the p- value is the area to the right of, or above,
the observed z value. Conversely, if the alternative is of the form <, the
p- value is the area to the left of, or below, the observed z value. (Fig.1.6;1.7)
p-value
p-value
z
0 z 0
Test statistic
Test statistic
Fig.1.6. Left tailed test Fig.1.7. Right tailed test
23
b) If the test is two tailed, the p-value is equal to twice the area beyond the
observed z-value in the direction of the sign of z. That is, if z is positive, the
p-value is twice the area to the right of, or above, the observed z- value.
Conversely, if z is negative, the p-value is twice the area to the left of, or
below, the observed z-value. (See Fig.1.8)
p/2 p/2
Test statistic Test statistic

Fig.1.8. Finding the p-value for a two tailed test
Example:
The management of Health club claims that its members lose an average of
10kg or more within the first month after joining the club. A random sample
of 36 members of this health club was taken and found that they lost an
average of 9.2 kg within the first month of membership with standard
deviation of 2.4kg. Find the p- value for this test.
Solution:
Let  be the mean weight lost during the first month of membership by all
_
members and x be corresponding mean for the sample.
H 0 :   10 (The mean weight lost is 10kg or more)

H 1 :   10 (The mean weight lost is less than 10kg)
Because the sample size is large we use the normal distribution to make the
test and calculate p-value.
Step 3. Calculate the p-value.
The < sign in the alternative hypothesis indicates that test is left tailed. The
p- value is given by the area in the left tail of the sampling distribution curve
24
_ _
of x where x is less than 9.2. To find this area, we first find the z value for
_
x  9.2 as follows
_
x  9.2  10
T .S .  z    2.00
s/ n 2.4 / 36
_ _
The area to the left of x  9.2 under the sampling distribution of x is equal to
the area under the standard normal curve to the left of z  2.00 . The area to
the left of z  2.00 is 0.0228. Consequently,
p  value  0.0228
Thus, based on the p- value of 0.0228 we can state that for any
 (significance level) greater than 0.0228 we will reject the null hypothesis
and for any  less than 0.0228 we will accept the null hypothesis.
Suppose we make the test for this example at   0.01 . Because   0.01 is
less than p-value of 0.0228, we will not reject the null hypothesis. Now
suppose we make the test at   0.05 . Because   0.05 is greater than the p-
value of 0.0228, we will reject the null hypothesis.
Exercises
1. Find the p-value for each of the following hypothesis tests

_
a) H 0 :   18 ; H 1 :   18 ; n  50 ; x  20 ; s  5;
_
b) H 0 :   15 ; H 1 :   15 ; n  80 ; x  13.2 ; s  5.5 ;
_
c) H 0 :   38 ; H 1 :   38 ; n  35 ; x  40.6 ; s  7.2
2. Consider H 0 :   29 ; against the alternative H 1 :   29 .
A random sample of 60 observations taken from this population produced a
sample mean of 31.4 and a standard deviation of 8.
a) Calculate the p-value.
b) Considering the p-value of part a), would you reject the null hypothesis if
the test were made at the significance level of 0.05?
c) Considering the p-value of part a), would you reject the null hypothesis if
the test were made at the significance level of 0.01?
25
3. In a given situation, suppose H 0 was rejected at   0.05 . Answer the
following questions as “yes”, “no”, or “can’t tell” as the case may be.
a) Would H 0 also be rejected at   0.02 ?
b) Would H 0 also be rejected at   0.10 ?
c) Is the p-value smaller than 0.05?
4. In a problem of testing H 0 :   75 against H 1 :   75 , the following
sample quantities are recorded.
_
n  56; x  77.04; s  6.80
a) State the test statistic and find the rejection region with   0.05 .
b) Calculate the test statistic and draw a conclusion with   0.05 .
c) Find the p-value and interpret the results.
Answers
1. a) 0.0046; b) 0.0017 ;c) 0.0162; 2. a) 0.0204; b) yes, reject H 0 ; c) no, do

_
x  75
not reject H 0 ;3. a) can’t tell; b) yes; c) no; 4. a) T .S .  Z  ;
s/ n
Z  1.645 ; b) T .S.  2.24 , H 0 is rejected at   0.05 ; c) 0.0125;

Population variance unknown. Small samples
Many times the size of a sample that is used to make test of hypothesis about
 is small, that is, n  30 . If the population is (approximately) normally
distributed, the population standard deviation  is not known and the sample
size is small ( n  30 ), then the normal distribution is replaced by the
Student’s t distribution to make a test of hypothesis about  . In such a case
the random variable
_
x 
t n 1 
s/ n
has a Student’s t distribution with (n  1) degrees of freedom.
26
_
The value of test statistic t for the sample mean x is computed as
_
x 
T .S .  t n 1 
s/ n
and we can use the following tests with significance level  .
H 0 :    0 or H 0 :    0 against the alternative
H1 :   0
Reject H 0 if T .S .  t n 1,
H 0 :    0 or H 0 :    0 against the alternative
H1 :   0
Reject H 0 if T .S .  t n 1,
H 0 :    0 against the two sided alternative
H1 :   0
Reject H 0 if T .S .  t n 1, / 2 or T .S .  t n 1, / 2 ,
Here, t n 1,  is the number for which
P(t n 1  t n 1, )  
where the random variable t n 1 follows a Student’s t distribution with
(n  1) degrees of freedom.
Example:
The company that produces auto batteries claims that its batteries are good,
for an average, for at least 64 days. A consumer protection agency tested 15
such batteries to check this claim. It found the mean life of these 15 batteries
to be 62 days with a standard deviation of 3 days. At the 5% significance
level, can you conclude that the claim of the company is true? Assume that
the life of such a battery has and approximate normal distribution.
27
Solution:
_
Let  be the mean life of all batteries and x be the corresponding mean for
the sample. Then from the given information,
_
n  15 ; x  62 days; s  3 days
The mean life of all batteries is supposed to be at least 64 days. The
significance level is  is 0.05. That is, the probability of rejecting the null
hypothesis when it is actually is true should not exceed 0.05.
H 0 :   64 days (The mean life is at least 64 days)
H 1 :   64 days (The mean life is less than 64 days)
The sample size is small ( n  15 ), and the life of a battery is approximately
normally distributed. Since population standard deviation is unknown, we
use the Student’s t distribution to make the test.
The significance level is 0.05. The  sign in the alternative test indicates that
the test is left tailed with the rejection region in the left tail of the t
distribution curve.
Area in the left tail=   0.05
Degree of freedom= n  1  15  1  14
From the Student’s t distribution table (Table 2 of Appendix), the critical
value of t for 14 degrees of freedom and an area 0.05 in the left tail
is  1.761 . (Fig.1.9).
Rejection region Nonrejection region
  0.05
t
-1.761 0
Fig.1.9
28
As  is not known, and sample size is small, we calculate the t value as
follows
_
x  62  64
T .S .  t    2.50
s / n 3 / 15
The value of T.S.  t  2.50 is less than the critical value of t  1.761 , and
it falls in the rejection region. Therefore, we reject H 0 and conclude that the
sample mean is too small compared to 62 days (company’s claimed value
of  ) and the difference between the two may not be attributed to chance
alone. We can conclude that the mean life of company’s batteries is less than
62 days.
Remark: The conclusion of a t-test can also be strengthened by reporting
the significance probability (p- value) of the observed statistic. Since the t
table provides only a few selected percentage points, we can get an idea
about the p-value but not its exact determination. For instance, the data in
example above gave an observed value T.S.  t  2.50 with degree of
freedom=14. Scanning the t table for (n  1)  14 , we notice that that 2.50
lies between t 0.025 and t 0.010 . Therefore, the p-value of t  2.50 is higher than
0.025 but not as great as 0.010.
Exercises
1. For each of the following examples of tests of hypothesis about  , show

the rejection and nonrejection regions on the t distribution curve.
a) A two tailed test with   0.2 and n  14
b) A left tailed test with   0.005 and n  23
c) A right tailed test with   0.025 and n  14
2. Consider the null hypothesis H 0 :   45 about the mean of a population
that is normally distributed. Suppose a random sample of 20 observations is
taken from this population to make this test. Using   0.05 show the
rejection and nonrejection regions and find critical value(s) for t for
a) left tailed test; b) two tailed test; c) right tailed test
29
3. Consider H 0 :   40 versus H 0 :   40 for a population that is normally
distributed.
a) A random sample of 16 observations taken from this population produced
a sample mean of 45 and a standard deviation of 5. Using   0.025 , would
you reject the null hypothesis?
population produced a sample mean of 41.9 and a standard deviation of 7.
Using   0.025 , would you reject the null hypothesis?
Comment on the result of parts a) and b).
4. Assuming that respective populations are normally distributed, make the
following hypothesis tests.
_
a) H 0 :   60 ; H 1 :   60 ; n  14 ; x  56 ; s 9 ;   0.05
_
b) H 0 :   35 ; H 1 :   35 ; n  24 ; x  29 ; s  5.4 ;   0.005
_
c) H 0 :   47 ; H 1 :   47 ; n  18 ; x  51 ; s 6;   0.001
5. A business school claims that students who complete a three month course
of typing course can type on average, at least 1200 words an hour.
A random sample of 25 students who completed this course typed, on
average, 1130 words an hour with a standard deviation of 85 words. Assume
that the typing speeds for all students who complete this course have an
approximate normal distribution.
Using the 5% significance level, can you conclude that the claim of the
business school is true?
6. The supplier of home heating furnaces of a new model claims that the
average efficiency of the new model is at least 60. Before buying these
heating furnaces, a distributor wants to verify the supplier’s claim is valid.
To this end, the distributor chooses a random sample of 9 heating furnaces of
a new model and measures their efficiency. The data are
63; 72; 64; 69; 59; 65; 66; 64; 65
Determine the rejection region of the test with   0.05 . Apply the test and
state your conclusion.
7. A past study claims that adults spend an average of 18 hours a week on
leisure activities. A researcher wanted to test this claim. He took a sample of
10 adults and asked them about the time they spend per week on leisure
activities. Their responses (in hours) were as follows
30
14; 25; 22; 38; 16; 26; 19; 23; 41; 33
Assume that the time spent on leisure activities by all adults is normally
distributed. Using the 5% significance level, can you conclude that the claim
of earlier study is true?
8. According to the department of Labor, private sector workers earned, on
average $354.32 a week in 2001. A recently taken random sample of 400
private sector worker showed that they earn, on average, $362.50 a week
with a standard deviation of $72. Find p-value for the test with an alternative
hypothesis that the current wean weekly salary of private sector workers is
different from $354.32.
9. A manufacturer of a light bulbs claims that the mean life of these bulbs is
at least 2500 hours. A consumer agency wanted to check whether or not this
claim is true. The agency took a random sample of 36 such bulbs and tested
them. The mean life for the sample was found to be 2447 hours with a
standard deviation of 180 hours.
a) Do you think that the sample information supports the company’s claim?
Use   2.5% .
b) What is the Type I error in this case? Explain. What is the probability of
making this error?
c) Will your conclusion of part a) change if the probability of making a
Type I error is zero?
10. Given the eight sample observations 31, 29, 26, 33, 40, 28, 30, and 25,
test the null hypothesis that the mean equals 35 versus the alternative that it
does not. Let   0.01 .
Answers
1.a) reject H 0 if t  1.350 or t  1.350 ;b) reject H 0 if t  2.819 ;c)reject

H 0 if t  2.160 ; 2. a) reject H 0 if t  1.729 ; b) reject H 0 if either t  2.093
or t  2.093 ; c) reject H 0 if t  1.729 ; 3. a) T.S.  t  4.00 ; reject H 0 ;
b) T.S.  t  1.086 ;accept H 0 ;4.a) T.S.  1.663 ;accept H 0 ;b) T.S.  5.443 ;
reject H 0 ; c) T.S.  2.828 ; accept H 0 ;5. T.S.  4.118 ;reject H 0 ;
6. t  1.860 ; T .S.  4.26 ; H 0 is rejected at   0.05 ;7. T.S.  2.692 ;
reject H 0 ;8. T .S.  2.27 ; p  value  0.0232; 9.a) T.S.  z  1.77 ;accept H 0 ;
b) 0.025; c) no; 10. T.S.  2.85 ; H 0 is not rejected.
31
1.6. Tests of the population proportion (Large sample)
Often we want to conduct test of hypothesis about a population proportion.

This section presents the procedure to perform tests of hypothesis about the
population proportion. p for large samples ( n  40) . The procedure to make
such tests is similar in many respects to the one for the population mean  .
^
The value of the test statistic T .S.  z for the sample proportion p computed
as
^
p p0
T .S .  z 
p 0 (1  p 0 ) / n
^
where p -is the sample proportion, and the value of p 0 used in this formula
is the one used in the null hypothesis.
Then, if the number of sample observations is large and observed proportion
^
is p , the following tests have significance level  :
H 0 : p  p0 or H 0 : p  p0 against the alternative
H 1 : p  p0
Reject H 0 if T .S.  z

H 0 : p  p0 or H 0 : p  p0 against the alternative
H 1 : p  p0
Reject H 0 if T .S.   z

H 0 : p  p0 against the two sided alternative
H 1 : p  p0
Reject H 0 if T .S.  z / 2 or T .S.   z / 2
32
.
Once again, z  is the number for which
P ( Z  z )  
and Z is the standard normal distribution.
Example:
Mr. A and Mr. B are running for local public office in a large city. Mr. A
says that only 30% of the voters are in favor of a certain issue, a law to sell
liquor on Sundays. Mr. B doubts A’s statement and believes that more than
30% favor such legislation. Mr. B pays for an independent organization to
make a study of this situation. In a random sample 400 voters, 160 favored
the legislation. What conclusions should the polling organization report to
Mr. B?
Solution:
^
Let p 0 be proportion of all people who favor such legislation and p the
corresponding sample proportion. Then from given information,
^ 160
n  400 ; p0  0.30 ; p  0.40 . Let   0.05 .
400
The null and alternative hypotheses are as follows
H 0 : p  p0  0.30
H 1 : p  0.30
The decision rule is to reject the null hypothesis in favor of alternative if
T .S.  z
  0.05 ;  / 2  0.025 .
P( Z  z / 2 )  P(Z  z 0.025 )  0.025 .
P(Z  z 0.025 )  F ( z 0.025 )  0.975 and
z / 2  z 0.025  1.645
From the given information we calculate the value of test statistic as
^
p p0 0.40  0.30
T .S .  z    4.36
p 0 (1  p 0 ) / n 0.30  0.70 / 400
Since 4.36  1.645 we reject H 0 . We make conclusion that more than 30% of
voters are in favor of a law to sell liquor on Sundays.
33
Exercises
1. Make the following hypothesis tests about p.

^
a) H 0 : p  0.45 ; H 1 : p  0.45 ; n  100 ; p  0.48 ;   0.10
^
b) H 0 : p  0.72 ; H 1 : p  0.72 ; n  700 ; p  0.65 ;   0.05
^
c) H 0 : p  0.30 ; H 1 : p  0.30 ; n  200 ; p  0.34 ;   0.01
2. Consider H 0 : p  0.70 versus H 0 : p  0.70 .
a) A random sample of 600 observations produced a sample proportion equal
to 0.67. Using   0.01 , would you reject the null hypothesis?
population produced a sample proportion of 0.76. Using   0.01 , would you
reject the null hypothesis?
Comment on the result of parts a) and b).
3. A food company is planning to market a new type of ice cream. Before
marketing this ice cream, the company wants to find what percentage of the
people like it. The company’s management has decided that it will market
this ice cream only if at least 35% of people like it. The company’s research
department selected a random sample of 400 persons and asked them to test
this ice cream. Of these 400 persons, 128 said they liked it.
a) Testing at 2.5% significance level, can you conclude that the company
should market this yogurt?
b) What will your decision be in part a) if the probability of making a Type I
error is zero?
4. A mail order company claims that at least 60% of all orders are mailed
within 48 hours. The quality control department took a sample of 500 orders
and found that 310 of them were mailed within 48 hours of the placement of
the orders. Testing at 1% significance level, can you conclude that the
company’s claim is true?
5. Let p=proportion of adults in a city who required a lawyer in the past year.
a) Determine the rejection region for   0.05 level test of H 0 : p  0.25
against H 1 : p  0.25 .
b) If 65 persons in a random sample of 200 required lawyer services, what
does the test conclude?
34
6. A magazine claims that 25% of its readers are university students. A
random sample of 200 readers is taken and 42 of these readers are university
students. Use   0.10 level of significance to test the validity of the
magazine’s claim.
7. Suppose that in order to test the hypothesis that p  0.6 against the
alternative that p  0.6 ,we decide to obtain a sample of size 100 and reject
H 0 if we obtain fewer than 48 successes.
a) What is the approximate size of the Type I error?
b) If the value of p is really 0.5, what is the size of Type II error?
8. An educator wishes to test H 0 : p  0.3 against H 1 : p  0.3 , where
p-proportion of football players who graduate university in four years.
a) State the test statistic and the rejection region having   0.05 .
b) If 19 out of a random sample of 48 players graduated in four years, what
does the test conclude? Also evaluate p-value.
9. The president of a company that produces national brand coffee claims
that 40% of the people prefer to buy national brand coffee. A random sample
of 700 people who buy coffee showed that 252 of them buy national brand
coffee.
a) Using   0.01 , can you conclude that the percentage of people who buy
national brand coffee is different from 40%?
b) Find the p-value for the test. Using this p-value, would you reject the null
hypothesis at   0.05 ? What if   0.02 ?
Answers
1. a) T .S.  z  0.60 ; do not reject H 0 ; b) T.S.  4.12 ; reject H 0 ;

c) T.S.  1.23 ; do not reject H 0 ; 2. a) T.S.  1.60 ; do not reject H 0 ;
b) T.S.  3.21 ; reject H 0 ; 3. a) T.S.  1.26 ; do not reject H 0 ; b) do not
reject H 0 ; 4. T .S.  0.91; accept H 0 ; 5. a) z  1.645 ; b) T .S.  2.45 ; reject
H 0 ; 6. T.S.  1.31 ; accept H 0 ; 7. a) about 0.0071 b)approximately 0.6554;
8. a) z  1.645 ; b) T.S.  1.45 ; accept H 0 for   0.05 ; p  value  0.0735;
9. a) T.S.  2.16 ; do not reject H 0 ;b) p-value=0.0308; reject H 0 at
  0.05 ; do not reject H 0 at   0.02
35
1.7. Tests of the variance of a normal distribution
In addition to the need for tests based on the sample mean and sample
proportion, there are a number of situations where we want to determine if
the population variance is a particular value or set of values. The basis for
developing particular tests lies in the fact that the random variable
(n  1)  s 2
 n21 
2
follows a Chi-square distribution with (n  1) degrees of freedom.
The value of the test statistic  n21 is calculated as
(n  1)  s 2
T .S .   n21 
2
We are given a random sample of n observations from a normally distributed
population with variance  2 . If we observe the sample variance s 2 , then the
following tests have significance level  :
H 0 :  2   02 or H 0 :  2   02 against the
alternative
H 1 :  2   02
Reject H 0 if T .S .   n21,

H 0 :  2   02 or H 0 :  2   02 against the
alternative
H 1 :  2   02
Reject H 0 if T .S .   n21,1

H 0 :  2   02 against the two sided alternative
36
H 1 :  2   02
Reject H 0 if T .S .   n21, / 2 or T .S .   n21,1 / 2
where  n21 is a Chi-square random variable and P (  n21   n21, )   .
Example:
Variance of yearly earnings of all state employees for all 40 states is
$49000 square dollars. A sample of 29 employees selected from state A
produced a variance of their earnings equal to $600 000 square dollars. Test
at 5% significance level if the variance of yearly earnings of state employees
in state A is different from $490 000 square dollars. Assume that the yearly
earnings of all state employees in state A have an (approximate) normal
distribution.
Solution:
From the given information,
n  29 ;   0.05 ; s 2  600000
The null and alternative hypotheses are
H 0 :  2  49000
H 1 :  2  49000
We use Chi square distribution to use. The decision rule is
Reject H 0 if T .S .   n21, / 2 or T .S .   n21,1 / 2
 / 2  0.025 ; 1   / 2  0.975 ;   n  1  29  1  28
Then from Table 3 of appendix we obtain
 n21, / 2   28
2
, 0.025  44.461 and  n 1,1 / 2   28, 0.975  15 .308
2 2
The value of the test statistic is

(n  1)  s 2 (29  1)  (600000 )
T .S .   n21    34.286
2 490000
The value of the test statistic 34.286 is between the two critical values of
 n21 , 15.308 and 44.461, and falls in the nonrejection region. Consequently
we fail to reject H 0 and conclude that the population variance of yearly
earnings of all employees in state A is not different from 490000 square
dollars.
37
Exercises
1. A sample of 24 observations selected from a normally distributed
population produced a sample variance of 12.
a) Write the null hypothesis and alternative hypothesis, and decision rule to
test if the population variance is different from 10.
b) Using   0.05 , find the critical values of  n21 . Show the rejection and
nonrejection regions on a Chi-square distribution curve.
c) Using the 5% significance level, will you reject the null hypothesis stated
in part a)?
2. A sample of 25 observations selected from a normally distributed
population produced a sample variance of 18. Using the 2.5% significance
level, test hypothesis if the population variance is less than 25.
3. Usually people do not like waiting in line for service for a long time.
A bank management does not want the variance of the waiting time for her
customers to be higher than 4.0 square minutes. A random sample of 25
customers taken from this bank gave the variance of the waiting times equal
to 7.9 square minutes. Test at 1% significance level if the variance of the
waiting time for all customers at this bank is higher than 4.0 square minutes.
Assume that the waiting time for all customers is normally distributed.
4. Test H 0 :   10 against H 1 :   10 with   0.05 in each case
25 _
a) n  25; i 1
( x i  x) 2  4016
b) n  15 ; s  12
5. A sample of seven observations taken from a population produced the
following data
10; 8; 13; 15; 6; 8; 13
Assuming that the population from which this sample is selected is normally
distributed, test at 2.5 significance level if the population variance is
different from 10.
6. A drug manufacturer requires that the variance for a chemical contained in
the bottles of certain type of drug should not exceed 0.03 square grams. A
sample of 25 such bottles gave the variance for this chemical as 0.06 square
grams. Test at the 1% significance level if the variance of this chemical in all
such bottles exceeds 0.03 square grams. Assume that the amount of this
chemical in all such bottles is (approximately) normally distributed. Find and
interpret p-value of this test.
38
7. A random sample of ten students was asked, in hours, for time they spent
studying in the week before final exams. The data are as follows:
28; 57; 42; 35; 61; 39; 55; 46; 49; 38
Assuming that the population distribution is normal, test at 5% significance
level against two sided alternative the null hypothesis that the population
standard deviation is 10 hours
8. Company claims that its employees earns a mean of at least $40 000 in a
year and that the population standard is no more than $6 000. Earnings of a
random sample of nine employees of this company produced
9 9 _
i 1
x i  333 and 
i 1
( xi  x) 2  312
where x i are measured in thousands of dollars and population distribution

can be assumed to be normal. Test at 10% significance level the null
hypothesis that the population standard deviation is at most $6 000.
9. State whether each of the following statement is true or false
a) The significance level of a test is the probability that the null hypothesis is
false.
b) A Type I error occurs when a true null hypothesis is rejected.
c) A null hypothesis is rejected at the 0.025 level, but is accepted at the 0.01
level. This means that p-value of the test lies between 0.01 and 0.025.
d) If a null hypothesis is rejected against an alternative at the 5% level, then
using the same data, it must be rejected against that alternative at the 1%
level.
f) If a null hypothesis is rejected against an alternative at the 1% level, then
using the same data, it must be rejected against that alternative at the 5%
level.
g) The p- value of a test is the probability that the null hypothesis is true.
Answers
1. b) reject H 0 if  2
23  38.08 or  23
2
 11.69 ; b) T .S .   23
2
 27.6 ; do not
reject H 0 ; 2. T .S .   24
2
 17.280 ; do not reject H 0 ;3. T .S .   24
2
 47.400 ;
reject H 0 ; 4. a) T.S.  40.16 ; reject H 0 ; b) T.S.  20.16 ; accept H 0 ;
5. T.S.  6.571 ; do not reject H 0 ; 6. T.S.  48.00 ; reject H 0 ; reject H 0 for
  0.0005 ; 7. T .S .   2  9.999 ; accept H 0 ; 8. T .S.  8.67 ; accept H 0 ;
9. a) false; b) true; c) false; d) false; f) true; g) false.
39
1.8. Tests for the difference between two population means
1.8.1. Tests based on paired samples
Suppose that a random sample of n matched pairs of observations is obtained

from populations with means  x and  y . The observations will be denoted
by ( x1 , y1 ), ( x 2 , y 2 ),........ .( x n , y n ) .
Let
n
_ d
i 1
i
d
n
n _ n _
i 1
(d i  d ) 2 
i 1
d i2  n(d ) 2
and s_  
d n 1 n 1
denote the observed sample mean and standard deviation for the n
differences d i  xi  y i . Let us denote difference between two population
means by D0   x   y . In this case test statistic will be calculated as
_
d  D0
T .S . 
s_ / n
d
If the population differences is a normal distribution, then the following tests
have significance level 

H 0 :  x   y  D0 or H 0 :  x   y  D0
against the alternative
H 1 :  x   y  D0
H 0 :  x   y  D0 or H 0 :  x   y  D0
40
H 1 :  x   y  D0
Reject H 0 if T .S .  t n 1,
H 0 :  x   y  D0 against the two sided alternative
H 1 :  x   y  D0
Reject H 0 if T .S .  t n 1, / 2 or T .S .  t n 1, / 2
Here, t n 1, is the number for which
P(t n 1  t n 1, )  
where the random variable t n 1 follows a Student’s t distribution with
(n  1) degrees of freedom.
Remark: When we want to test the null hypothesis that the two population
means are equal, we set D0  0 .
Example:
A medical researcher wishes to determine if a pill has the undesirable side
effect of reducing the blood pressure of the user. The study involves
recording the initial blood pressures of 7 college age adults. After they use
the pill regularly for three month, their blood pressures are again recorded.
The researcher wishes to draw inferences about the effect of the pill on blood
pressure from the information given in table
Before x i 64 71 68 66 73 62 70
After y i 60 66 66 69 63 57 62
Do the data substantiate the claim that use of the pill reduces the blood
pressure? Use   0.01 . Assume that the population of paired differences
has a normal distribution.
Solution:
Let d be the difference between the pressures before and after using pills.
d=before –after= xi  yi
The necessary calculations are shown in the following table
41
Before After Difference d2
d
64 60 4 16
71 66 5 25
68 66 2 4
66 69 -3 9
73 63 10 100
62 57 5 25
70 62 8 64
 d  31  d  243
2
_
The values of d and S d are calculated as follows:
_
d
 d  31  4.43
n 7
1   
d
1
Sd   n  (d ) 2  
2
(243  7  4.43 2 )  4.198 .
n  1   6
Let  x be the mean blood pressure for all adults before and  y -after using
the pill.
H 0 :  x   y  0 (no difference)
against
H 1 :  x   y  0 ( mean decreases)
The decision rule is that

_
d  D0 4.43  0
T .S .    2.792
s_ / n 4.198 / 7
d
t n1,  t 6,0.01  3.14
Since 2.792<3.14, we accept H 0 and make conclusion at the level 0.01 that
using pills does not affect blood pressure.
42
Exercises
1. Perform the following tests of hypothesis assuming that the population of

paired differences are normally distributed.
_
a) H 0 :  d  D0  0 ; H 1 :  d  D0  0 ; n  9 ; d  6.7 ; s d  2.5 ;   0.10
_
b) H 0 :  d  D0  0 ; H 1 :  d  D0  0 ; n  22 ; d  14.3 ; s d  6.4 ;   0.05
_
c) H 0 :  d  D0  0 ; H 1 :  d  D0  0 ; n  17 ; d  9.3 ; s d  4.8 ;   0.01
2. It is claimed that an industrial safety program is effective in reducing the
loss of working hours due to factory accidents. The following data are
collected concerning the weekly hours due to accidents in nine plants both
before and after the safety program is installed
Before x i 90 86 72 65 44 52 46 38 43
After y i 85 87 70 62 44 53 42 35 46
Do the data substantiate the claim? Use   0.05 .

Assume that the population of paired differences is (approximately)
normally distributed.
3. A company claims that the course it offers significantly increases the
writing speed of secretaries. The following table gives the scores of 8
secretaries before and after they attended this course
Before x i 81 75 89 91 65 70 90 69
After y i 97 72 93 110 78 69 115 75
Using the 5% significance level, can you conclude that attending this course
increases the writing speed of secretaries?
4. A random sample of nine employees was selected to test for the
effectiveness of hypnosis on their job performance. The following table
gives the job performance ratings (on a scale of 1 to 4, with 1 being the
43
lowest and 4 being the highest) before and after these employees tried
hypnosis.
Before x i 2.3 2.8 3.1 2.7 3.4 2.6 2.8 2.5

After y i 2.6 3.2 3.0 3.5 3.7 2.4 2.9 2.9
Test at the 5% significance level if there is an improvement in the job

performances of employees due to hypnosis.
Answers
1. a) T.S.  8.040 reject H 0 ;b) T .S.  10.847 ;reject H 0 ; c) T.S.  7.989

reject H 0 ; 2. T.S.  1.48 ; do not reject H 0 ;3. T.S.  2.807 ; reject H 0 ;
4. T.S.  2.236 ; accept H 0 .
1.8.2. Tests based on independent samples

(Known variance or large sample size)
Let us consider the case where we have independent random samples from
two normally distributed populations. The first population has mean  x and
variance  x2 and we obtain a random sample of size n x . The second
population has mean  y and variance  y2 and we obtain a random sample of
size n y .
_ _
We know that if the sample means are denoted x and y , then the random
variable
 
_ _
( x y)   x   y
Z
 x2  y2

nx ny
44
has a standard normal distribution. If the population variances are known,
tests for the difference between the population means can be based on this
_ _
result. The value of the test statistic z for ( x y ) is computed as
 
_ _
( x y)   x   y
T .S .  z 
 x2  2

y
nx ny
and the following tests have a significance level 
H 0 :  x   y  D0 or H 0 :  x   y  D0
H 1 :  x   y  D0
H 0 :  x   y  D0 or H 0 :  x   y  D0
H 1 :  x   y  D0
H 1 :  x   y  D0
Remark: If the sample sizes are large ( n x  30; n y  30 ) then a good
approximation at significance level  can be made if the population
variances  x2 and  y2 are replaced by the sample variances s x2 and s 2y .
In addition the central limit theorem leads to good approximations even if
the populations are not normally distributed.
45
Example:
According to the Bureau of Labor Statistics, last year university instructors
earned an average $440 per month and college instructors earned an average
of $420 per month. Assume that these mean earnings have been calculated for
samples of 400 and 600 instructors taken from the two populations,
respectively. Further assume that the standard deviations of monthly earnings
of the two populations are $50 and $63, respectively. Test at 1% significance
level if the mean monthly earnings of the two groups of the instructors are
different.
Solution:
From the information given above,
_
n x  400 ; x  440 ;  x  50 ;
_
n y  600 ; y  420 ;  y  63 ;
where the subscript x refers to university instructors and y-to college
instructors. Let
 x = mean monthly earnings of all university instructors
 y = mean monthly earnings of all college instructors.
We are to test if the two population means are different. The null and
alternative hypotheses are
H 0 :  x   y  0 (the monthly earnings are not different)
H 1 :  x   y  0 (the monthly earnings are different).
The decision rule is
First of all we find the value of z / 2 . Since  / 2  0.005 , the value of z / 2 is
(approximately) 2.58 and  z / 2  2.58 .
The value of the test statistic T .S.  z is computed as follows:
 
_ _
( x y)   x   y (440  420)  (0)
T .S .  z    5.57 .
 x2  y2 50 2 63 2
 
nx ny 400 600
46
5.57  2.58 and the value of test statistic T .S.  z  5.57 falls in the rejection
region, we reject the null hypothesis H 0 . Therefore, we conclude that the
mean monthly earnings of the two groups of instructors are different.
Note that we can not say for sure that two means are different. All we can
say is that the evidence from the two samples is very strong that the
corresponding population means are different.
Exercises
1. The following information is obtained from two independent samples

selected from two populations
_
n1  155; x  5.58; s x  1.62
_
n 2  190; y  4.80; s y  1.52
Test at the 1% significance level if the two population means are the same
against the alternative that they are different.
2. Daily wage is $13.62 for transportation workers and $11.61 for factory
workers. Assume that these two estimates are based on random samples of
1000 and 1200 workers taken, respectively, from the two populations. Also
assume that the standard deviations of the two populations are $1.85 and
$1.40, respectively.
a) Test at the 5% significance level if the mean daily wage of transportation
workers and factory workers are the same against the alternative that it is
higher for transportation workers.
b) What will your decision be in part a) if the probability of making a Type I
error is zero. Explain.
3. A consulting firm was asked by a large insurance company to investigate
if business majors were better salespersons. A sample of 40 salespersons
with a business degree showed that they sold an average of 10 insurance
policies per week with a standard deviation of 1.80. Another sample of 45
salespersons with a degree other than business showed that they sold an
average of 8.5 insurance policies per week with a standard deviation of 1.35.
Using the 1% significance level, can you conclude that person with a
business degree are better salespersons than those who have a degree in
another area?
47
4. The management at the bank A claims that the mean waiting time for all
customers at its branches is less than that at the bank B, which is main
competitor. They took a sample of 200 customers from the bank A and found
that they waited an average of 4.60 minutes with a standard deviation of 1.2
minutes before being served. Another sample of 300 customers taken from
the bank B showed that these customers waited an average of 4.85 minutes
with a standard deviation of 1.5 minutes before being served.
a) Test at the 2.5% significance level if the claim of the management of the
bank A is true.
b) Calculate the p-value. Based on this p-value, would you reject the null
hypothesis if   0.01? What if   0.05 ?
5. A production line is designed on the assumption that the difference in
mean assembly times for two operations is 5 minutes. Independent tests for
the two assembly operations show the following results:
Operation A Operation B
n1  100 n 2  50
_ _
x  14.8 minutes y  10.4 minutes
s x  0.8 minutes s y  0.6 minutes
For   0.02 , test the hypothesis that the difference between the mean
assembly times is 5 minutes.
6. An investigation was carried out to determine if women employees are as
well paid as their male counterparts. Random samples of 75 males and 64
females are selected. Their mean salaries were 45 530 and 44 620, standard
deviations were 780 and 750, correspondingly. If you were to test the null
hypothesis that the mean salaries are equal against the two sided alternative,
what would be the conclusion of your test with   0.05 ?
7. For a random sample of 125 state companies, the mean number of job
changes was 1.91 and the standard deviation was 1.32. For a random sample
of 86 private companies, the mean number of job changes was 0.21 and the
standard deviation was 0.53. Test the null hypothesis that the population
means are equal against the alternative that the mean number of job changes
is higher in state companies than for private companies.
48
Answers
1. T .S.  z  4.56 ;reject H 0 ;2.a) T.S.  28.27 ; reject H 0 ;b)do not reject H 0 ;
3. T .S.  4.30 ; reject H 0 ; 4. a) T.S.  2.06 ; reject H 0 ;b) p-value=0.0197;
do not reject H 0 at   0.01? ; reject H 0 at   0.05 ; 5. T.S.  5.15 ;
reject H 0 ; 6. T .S.  7 ; reject H 0 ; 7. T.S.  13 ; reject H 0 at any level.
1.8.3. Tests based on independent samples

(Population variances are unknown and equal)
Many times it may not be possible to take large samples from

populations to make inferences about the difference between two population
means. This section discusses how to test a hypothesis about the difference
between two population means when samples are small
(n x  30) , ( n y  30) and independent. Our main assumption in this case is
that the two populations from which the two samples are drawn are
(approximately) normally distributed. If this assumption is true, and we
know the population variances, we can still use the normal distribution to
make inferences about (  x   y ) when samples are small and independent.
However, we usually do not know the population variances  x2 and  y2 . In
such cases, we replace the normal distribution by the Student’s t distribution
to make inferences about (  x   y ) for small and independent samples. In
this section we will make one more assumption that the variances of the two
populations are equal. When the variances of the two populations are equal,
we can use  2 for both  x2 and  y2 . Since  2 is unknown, we replace it by
its point estimator s 2p , which is called pooled sample variance.
Now assume that we have independent random samples of size n x and n y
observations from normally distributed populations with means  x and
 y and a common variance. The sample variances s x2 and s 2y are used to
compute a pooled variance estimator
49
(n x  1)  s x2  (n y  1)  s 2y
s 2p 
(n x  n y  2)
_ _
The value of the test statistic z for ( x y ) is computed as
 
_ _
( x y)   x   y
T .S .  t 
s 2p s 2p

nx ny
and the following tests have a significance level 
H 0 :  x   y  D0 or H 0 :  x   y  D0
H 1 :  x   y  D0
Reject H 0 if T .S .  t n x  n y  2,
H 0 :  x   y  D0 or H 0 :  x   y  D0
H 1 :  x   y  D0
Reject H 0 if T .S .  t n x  n y  2,
H 1 :  x   y  D0
Reject H 0 if T .S .  t nx  n y  2, / 2 or T .S .  t n x  n y  2, / 2
Here, t nx  n y 2, is the number for which
P(t nx  n y  2  t nx  n y  2, )  
50
where the random variable t nx  n y 2, follows a Student’s t distribution with
(n x  n y  2) degrees of freedom.
Example:
A sample of 12 cans of Brand A diet soda gave a mean number of calories of
22 per can with a standard deviation of 2 calories. Another sample of 15 cans
of Brand B diet soda gave the mean number of calories of 24 per can with a
standard deviation of 3 calories. At the 1% significance level, are the mean
number of calories per can different for these two brands of diet soda?
Assume that the calories per can of diet soda are normally distributed for
each of the two brands and that the variances for the two populations are
equal.
Solution:
Let  x and  y be the mean number of calories per can for diet soda of Brand
_ _
A and Brand B, respectively, and let x and y be the means of respective
samples. From the given information,
_
n x  12 ; x  22 ; sx  2 ;
_
n y  15 ; y  24 ; sy  3
The significance level is   0.01 .
We are to test for the difference in the mean number of calories per can for
two brands. The null and alternative hypotheses are
H0 : x  y  0 ( the mean number of calories are not different)
H1 :  x   y  0 ( the mean number of calories are different)
Reject H 0 if T .S .  t nx  n y  2, / 2 or T .S .  t n x  n y  2, / 2
t nx  n y  2, / 2  t1215 2,0.005  t 25,0.005  2.787 and  t 25,0.005  2.787 .
The pooled estimate is
(n x  1)  s x2  (n y  1)  s 2y (12  1)  2 2  (15  1)  3 2
s 2p    6.8
(n x  n y  2) (12  15  2)
The test statistic is then computed as
51
 
_ _
( x  y )   x   y (22  24)  (0)
T .S .  t   1.98
s 2p s 2p 6.8 6.8

 12 15
nx ny
_ _
Because the value of test statistic T.S.  t  1.98 for ( x y ) falls in the
nonrejection region (Fig.1.10), we fail to reject the null hypothesis.
Consequently we conclude that there is no difference between the mean
number of calories per can for the two brands of diet soda. The difference in
_ _
x and y observed for two samples may have occurred due to sampling error
only.
Reject H 0 Do not reject H 0 Reject H 0
 / 2 =0.05  / 2 =0.05
t
-2.787 0 2.787
Fig.1.10
Exercises
1. The following information was obtained from two independent samples

selected from two normally distributed populations with unknown but equal
variances
_
n x  20 ; x  33.75 ; s x  5.25 ;
_
n y  23 ; y  28.50 ; s y  4.55
a) Test at 1% significance level if the two population means are different.
b) Test at 5% significance level if  x is different than  y .
52
2. The following summary statistics are recorded for independent random
samples from two normally distributed populations with equal variances
Sample 1 Sample 2
n1  9 n2  6
_ _
x  16.18 y  4.22
s1  1.54 s 2  1.37
Test the null hypothesis 1   2  10 against the alternative that
1   2  10 with   0.01 .
3. Salary surveys of marketing and management majors show the following
starting annual salary data
Marketing majors management majors
n1  14 n 2  16
_ _
x  $14800 x 2  $14300
s1  $1000 s 2  $1400
Consider the test of the hypothesis that the mean annual salaries are the same
for both majors. For   0.05 can you conclude that a difference exists in the
mean annual salary for the two majors?
4. A professor took two samples, one of 21 males and another of 15 females
from university students who were enrolled in business statistics at the same
university. He found that the mean score of male students in a mid-term
examination in statistics was 75.3 with a standard deviation of 6.4, and the
mean score of female students was 78.3 with a standard deviation of 7.3.
Assume that the scores of all male and all female students are normally
distributed with equal but unknown standard deviations.
Test at the 2.5 significance level if the mean score in business statistics for
all male and female students are the same against the alternative that male
students have lower score than that for all female students.
5. The management of a supermarket wanted to investigate if the male
customers spend less money on average, than the female customers. A
sample of 16 male customers who shopped at this supermarket showed that
they spent an average of $55 with a standard deviation of $12.50. Another
sample of 22 female customers who shopped at the supermarket showed that
they spent an average of $63 with a standard deviation of $14.5. Assume that
the amounts of money spent at this supermarket by all male and female
53
customers are normally distributed with equal but unknown population
variance. Test at the 5% significance level if the mean amount spent by all
male and female customers are the same against the alternative that male
customers at this supermarket spend less than that of female customers.
6. A bank has two branches. The quality department wanted to check if the
customers are equally satisfied with the service provided at these two
branches. Randomly selected customers asked to measure the satisfaction of
services (on scale of 1 to 11, 1 being the lowest and 11 being the highest).
A random sample of six customers from the branch A produced following
data:
9.50; 8.60; 8.59; 6.50; 4.79; 4.29
An independent random sample of six customers selected from the branch B
produced following data:
10.21; 9.66; 7.67; 5.12; 4.88; 3.12
Stating any assumptions you need to make, test against two sided alternative
the null hypothesis that the two populations mean satisfaction index for all
customers for the two branches are the same.
_ _ _
7. Given that n1  14 , x  22 ,  ( xi  x) 2  30 , and n2  13 , y  18 ,
_
(y i  y ) 2  24 . Test H 0 : 1   2 against H 1 : 1   2 with   0.05 .
8. A researcher wants to test the mean GPA (grade point averages) of all
male and all female university students. She took a random sample of 28
male students and 24 female students. She found that GPA’s of the two
groups to be 2.62 and 2.74, respectively, with the corresponding standard
deviations equal to 0.43 and 0.38. Test at the 5% significance level if the
mean GPA’s of the two populations are equal against two sided alternative.
Assume that the GPA’s of all male and female students are normally
distributed with equal but unknown standard deviations.
Answers
1. a) T.S.  t  3.514 ; reject H 0 ; b) T.S.  t  3.514 ; reject H 0 ;
2. T.S.  t  2.52 ; H 0 is not rejected; 3. T.S.  t  1.11 ; accept H 0 ;
4. T.S.  1.308 ; accept H 0 ; 5. T.S.  1.778 ; reject H 0 ; 6. We assume that
the values are normally distributed with equal variance; T.S.  0.183 ; Fail to
reject H 0 at 20% significance level; 7. T.S.  7.071 ; reject H 0 ;
8. T.S.  1.058 ; accept H 0 .
54
1.9. Tests for the difference between two population proportions
(Large samples)
Next we will develop procedures for comparing two population proportions.

We will consider standard model with a random sample of n x observations
^
with proportion p x ”successes” and an independent random sample of n y
^
observations from population with proportion p y ”successes”.
We know that for large samples, proportions can be approximated as
normally distributed random variables and as a result
^ 
 
^
 p x  p y   px  p y
Z  
p x (1  p y ) p y (1  p y )

nx ny
has a standard normal distribution.
We want to test the hypothesis that the population proportions p x and p y are
equal. Denote their common value by p 0 , then the value under this
hypothesis
^ ^
(px  py )
Z
p 0 (1  p 0 ) p 0 (1  p 0 )

nx ny
follows a good approximation a standard normal distribution.
Finally, unknown common proportion p 0 can be estimated by a pooled
estimator defined as
^ ^
nx  p x  n y  p y x1  y1
p0  or p 0 
nx  n y nx  n y
where x1 and x 2 are number of “successes” in n x and n y , respectively.
55
Which of these formulas is used to calculate p 0 depends on whether the
^ ^
values of x1 and y1 or the values of p x and p y are known.
Testing equality of two population proportions:
We are given independent samples of size n x and n y with proportion of

^ ^
successes p x and p y . When we assume that the population proportions are
equal, an estimate of the common proportion is
^ ^
nx  p x  n y  p y x1  y1
p0  or p 0  .
nx  n y nx  n y
 ^ ^ 
For large sample sizes  n  p q  9  the value of the test statistic z for is
 
computed as
^ ^
( p x  p y )  ( px  p y )
T .S .  z 
p 0 (1  p 0 ) p 0 (1  p 0 )

nx ny
Then the following tests have significance level  :

H 0 : px  p y  0 or H 0 : px  p y  0
H1 : p x  p y  0
H 0 : px  p y  0 or H 0 : px  p y  0
H1 : p x  p y  0
56
H 0 : px  p y  0 against the two sided alternative
H1 : p x  p y  0
Example:
A company is planning to buy a few machines. Company is considering two
types of machines, but will buy all of the same type. The company selects
one machine from each type and uses for a few days. A sample of 900 items
produced on machine A showed that 55 of them were defective. A sample of
700 items produced on machine B showed that 41 of them were defective.
Testing at 1% significance level, can we conclude based on the information
from these samples that the proportions of the defective items produced on
the two machines are different?
Solution:
Let p x be the proportion of all items in all items produced on machine A,
and p y be the proportion of all items in all items produced on machine B.
^ ^
Let p x and p y be the corresponding sample proportions. Let x1 and x 2 be the
number of defective items in two samples respectively.
Machine A: n x  900 ; x1  55
Machine B: n y  700 ; y1  41
The two sample proportions are calculated as follows:
^ x 55
px  1   0.0611 ;
n x 900
^ y 41
py  1   0.0586
n y 700
H 0 : px  p y  0 ( the two proportions are equal)
H1 : p x  p y  0 ( the two proportions are different)
57
Let us check if the sample sizes are large:
 ^ ^

 n x  p x  q x  9   900  0.0611  0.9389  51.63  9
 
 ^ ^

 n y  p y  q y  9   700  0.05860  0.9414  38.62  9
 
Since the samples are large and independent we apply the normal
distribution to make a test.
The pooled sample proportion is
x  y1 55  41
p0  1   0.06
n x  n y 900  700
The value of the test statistics is
^ ^
( p x  p y )  ( px  p y ) (0.0611  0.0586 )
T .S .  z    0.2089 .
p 0 (1  p 0 ) p 0 (1  p 0 ) 0.06  0.94 0.06  0.94
 
nx ny 900 700
Let us find the value of z / 2 .
  0.01 ;  / 2  0.005
Fz ( z / 2 )  Fz ( z 0.005 )  0.995
z 0.005  2.58 and  z 0.005  2.58
The value of the test statistic T.S.  z  0.2089 falls in the nonrejection
region. Consequently, we fail to reject the null hypothesis. As a result, we
can conclude that proportions of defective items produced by two machines
are not different.
58
Exercises
1. The following information is obtained from two independent samples

selected from two populations
^ ^
n x  750 ; p x  0.56 ; n y  600 ; p y  0.61 ;
Test at the 1% significance level if p x is equal p y , against the alternative
that it is less.
2. A sample of 600 observations taken from the first population
gave x1  320 . Another sample of 700 observations taken from the second
population gave y1  370 . Show the rejection and nonrejection regions on
^ ^
the sampling distribution of px  py for H 0 : px  p y against the
alternative H 1 : p x  p y using significance level of 2.5%. Will you reject the
null hypothesis?
3. According to the statistics, 65% of single women and 80% of single men
own cars. Assume that these estimates are based on random samples of 1700
singles women and 1900 single men. At the 1% significance level, can you
conclude that the proportion of single women who own cars is the same as
the proportion of men who own cars against the alternative that it is less than
the proportion of men who own cars? Also find p-value.
4. The management of a supermarket wanted to investigate if the percentage
of men and women who prefer to buy national brand products over the store
brand products are different. A sample of 600 men shoppers at the
company’s supermarkets showed that 246 of them prefer to buy national
brand products over the store brand products. Another sample of 700 women
shoppers showed that 266 of them prefer to buy national brand products over
the store brand products. Testing at the 5% significance level, can you
conclude that the proportion of all men and all women shoppers at these
supermarkets who prefer to buy national brand products over the store brand
products are equal?
5. A sample of 500 male registered voters showed that 57% of them are in a
favor of higher taxes on wealthy people. Another sample of 400 female
registered voters showed that 54% of them are in favor of higher taxes on
wealthy people. Test at the 1% significance level if the proportion of all male
59
voters who are in favor of higher taxes on wealthy people is not different
from that of female voters against two-sided alternative.
6. A medical researcher investigates if the smoking results in wrinkled skin
around the eyes. By observing 150 smokers and 250 nonsmokers, the
researcher finds that 95 of the smokers and 103 of the nonsmokers have
prominent wrinkles around their eyes. Do these data substantiate the belief
that prominent wrinkles around eyes are more prevalent among smokers than
nonsmokers? Answer by calculating p-value.
7. In a comparative study of two new drugs, A and B, 120 patients treated
with drug A and 150 patients with drug B, and the following results were
obtained
Drug A Drug B
Cured: 52 88
Not cured: 68 62
Total: 120 150
Do these results demonstrate statement that these two drugs have the same
effect against the alternative that higher cure rate with drug A? Test at
  0.05 .
8. According to a 2001 survey, 48% of managers “would choose the same
career if they were starting over again”. In a similar survey conducted 10
years ago, 60% of managers said that they “would choose the same career if
they were starting over again”. Assume that the 2001 survey is based on a
sample of 800 managers and the one done 10 years ago included 600
managers. Test at the 5% significance level if the proportion of all managers
who “would choose the same career if they were starting over again” has not
changed against the alternative that it decreased during the past 10 years.
Answers
1. T.S.  z  1.85 ; accept H 0 ; 2. T .S.  0.17 ; accept H 0 ;3. T.S.  9.80 ;

reject H 0 ; can reject H 0 at virtually any level; 4. T.S.  z  1.10 ; do not reject
H 0 ; 5. T .S.  0.90 ; accept H 0 ; 6. T .S.  4.28 ; p-value=0.0002; claim is
strongly supported;7. H 0 : p A  p B ; H 1 : p A  p B ; T.S.  2.52 ; reject H 0 ;
8. T.S.  4.45 ; reject H 0 .
60
Chapter 2
Some nonparametric tests
2.1. Introduction
When the methods of statistical inference are based upon the

assumption that the population has a certain probability distribution, such as
the normal, the resulting collection of statistical tests and procedures is
referred to as parametric methods. In this chapter we will consider several
statistical procedures that do not require knowledge of the form of the
probability distribution form which the measurements come. The methods of
statistical inference we will study here are called nonparametric methods.
Since nonparametric methods do not require assumptions about the form of
the population distribution they are often referred to as distribution free
methods.
From this discussion we see that one reason for using nonparametric
methods is that in some situations there is insufficient knowledge about the
form of the population distribution. Thus assumptions necessary for use of
parametric tests can not be made.
A second reason for using nonparametric methods concerns data
measurement. Nonparametric methods are often applied to rank order or
preference data. Preference data are the type of data generated when people
express preference for one product over another, one service over another,
etc. Parametric procedures can not be applied with these data, but
nonparametric ones can.
This chapter presents an introduction to some of the commonly used
statistical procedures that can be classified as nonparametric or distribution
free methods. The emphasis will be on the type of problems that can be
solved, how the statistical calculations are made, and how appropriate
conclusions can be developed to assist management in the decision-making
process.
61
2.2. 1. The Sign test for paired or matched samples
In Chapter 1 we considered z and t statistics for testing hypothesis

about a population mean. For both of them, the sample was selected at
random from a normal distribution. The question is: How can we conduct a
test of hypothesis when we have a small sample from a nonnormal
distribution?
The Sign test is a relatively simple and most frequently employed
nonparametric procedure for testing hypothesis about the central tendency of
a nonnormal probability distribution. The sign test is used in studies to
identify if consumer preference exists for one of two products.
Suppose that paired or matched samples are taken from a population, and the
differences equal to 0 are discharged, leaving n observations. The Sign test
can be used to test the null hypothesis that the population median of the
differences is 0. Let “+” indicates a positive difference, and “-“ indicates a
negative difference. If the null hypothesis were true, our sequence of “+” and
“-“ differences could be regarded as a random sample from a population in
which the probabilities for “+” and “-“ were each 0.5. In that case, the
observations would constitute a random sample from a binomial population
in which the probability of “+” was 0.5. Thus, if p denotes the true
population proportion of “+”’s in the population (that is, the true
proportion of positive differences), the null hypothesis is simply
The Sign test is then based on the fact that the number of positive
observations, S, in the sample has a binomial distribution (with p  0.5
under the null hypothesis).
H 0 : p  0.5
Sign test for paired samples
Suppose that paired random samples are taken from a population and the
differences equal to 0 are ignored. Calculate the difference for each pair and
record the sign of this difference. The Sign test is used to test:
H 0 : p  0.5
where p-is the proportion of nonzero observations in the population that are
positive. The test statistic S for the Sign test for paired samples is simply
S = the number of pairs with positive difference
62
and S has a binomial distribution with p  0.5 and n  the number of
nonzero differences.
After determining the null and alternative hypotheses and finding a test
statistic, the next step is to determine the p-value and to draw conclusions
based on a decision rule.
Determining p- value for a Sign test
The p-value for a Sign test is found using the binomial distribution with
n  the number of nonzero differences, S = the number of pairs with positive
differences and p  0.5 .
1. For right tailed test,
H 1 : p  0.5 , p-value = P( x  S )
2. For left tailed test,
H 1 : p  0.5 , p-value = P( x  S )
3. For two tailed test,
H 1 : p  0.5 , 2  ( p  value)
Example:
In the study 8 individuals were asked to rate on a scale from 1 to 10 the test
of products of two brands: Brand A and Brand B. The scores of the test
comparison are shown in the following table
N Brand A Brand B
1 5 7
2 3 10
3 4 8
4 9 6
5 8 8
6 5 7
7 6 5
8 9 6
Do the data indicate an overall tendency to prefer the Brand B to the
Brand A?
63
Solution:
First of all, let us calculate differences
N Brand A Brand B Difference (A-B) Sign of difference

1 5 7 -2 -
2 3 10 -7 -
3 4 8 -4 -
4 9 6 3 +
5 8 8 0 0
6 5 7 -2 -
7 6 5 1 +
8 9 6 3 +
We are discarding those who rated the brands equally. In this example the
values for fifth person is omitted in future analysis and the effective sample
size is reduced to n  7 . The only sample information on which our test is
based is that three of the seven tasters preferred the brand A. Hence, the
value of the Sign test is S  3 .
Let p-denotes the true proportion of “+”s in the population. Then the null
hypothesis is
H 0 : p  0.5 There is no overall tendency to prefer one Brand to the other
A one tailed test is used to determine if there is an overall tendency to prefer

the Brand B to the Brand A. The alternative hypothesis is that in the
population, the majority of preferences are for Brand B. The alternative
hypothesis is expressed as
H 0 : p  0.5 Majority prefer the Brand B
The next step is the finding the p-value. If we denote by P(x) the probability
of observing x “successes” (“+”s) in n  7 binomial trials, each with
probability of success 0.5, then the cumulative binomial probability of
observing tree or fewer “+”s can be obtained using binomial formula
p  value  P( x  3)  P( x  0)  P( x  1)  P( x  2)  P( x  3) 
 C 07 (0.5) 0 (0.5) 7
 C17 (0.5)1 (0.5) 6  C 27 (0.5) 2 (0.5) 5  C 37 (0.5) 3 (0.5) 4 
 0.0078  0.0547  0.1641  0.2734  0.5000
64
For this example p- value is 50%. We are unable to reject the null hypothesis
and conclude that data is not sufficient to suggest that population have a
preference for Brand B. Since the p-value is the smallest significance level at
which the null hypothesis can be rejected, for this example, the null
hypothesis can be rejected at 50% or higher. It is unlikely that one would be
willing to accept such a high significance level. Again, we conclude that the
data is not statistically significant to recommend that Brand B is preferred by
majority.
2.2.2. The sign test: Normal approximation (Large samples)
As a consequence of the central limit theorem, the normal distribution can be

used to approximate the binomial distribution if the sample size is large
Experts differ on the exact definition of “large”. We suggest that the normal
approximation is acceptable if the sample size exceeds 20. With large n, the
binomial distribution with p  0.5 is close to the normal distribution with
mean n / 2 and standard deviation   n  p  q  0.25  n  n / 4 .
The test statistic is
S  n/2 S  n/2
T .S .  z  
 n/4
where
S-is the number of positive signs,
n-is the number of nonzero sample observations.
The null hypothesis to be tested is that the proportion p- of nonzero
observations in the population that are positive is 0.5; that is
H 0 : p  0.5 against the alternative
H 1 : p  0.5
H 0 : p  0.5 against the alternative
H 1 : p  0.5
65

H 0 : p  0.5 against the two sided alternative
H 1 : p  0.5
Example:
In a TV commercial, filmed live, 100 persons tested two brands of coffee:
say brand A and brand B and each selected their favorite. 56 persons
preferred coffee of brand A, 40 preferred coffee of brand B, and four
expressed on preference. Test at the 5% significance level the null
hypothesis that for the population, there is no overall preference for coffee of
brand A over the brand B.
Solution:
From the information given above we obtain that
S  56 ; n  96
To test if there is no overall preference in this population for one brand of
coffee over the other, the hypotheses are
H 0 : p  0.5 (People have no preference for either brand of coffee)
H 1 : p  0.5 (People have preference for one brand of coffee)
S n/2 S  n/2
Reject H 0 if   z / 2 or  z / 2
n/4 n/4
S  n / 2 56  96 / 2
  1.633
n/4 96 / 4
  0.05 ; z / 2  1.96 and  z / 2  1.96 .
Since  1.633 is not less than  1.96 we fail to reject the null hypothesis. And
at 5% significance level we accept that there is no preference for either brand
of coffee.
From the standard normal distribution it follows that the approximate
p  value  0.1032 or 10.32%.
66
Hence the null hypothesis can be rejected at all significance levels greater
than 10.32%.
Exercises
1. Two computer specialists estimated the amount of computer memory (in

gigabytes) required by five different offices.
Office Specialist A Specialist B
1 4.2 6.1
2 6.3 6.7
3 3.1 3.0
4 2.2 2.9
5 7.2 10.9
Use the Sign test to test the null hypothesis that two specialists estimations
are the same against the alternative that specialist B estimates higher than
specialist A.
2. In a test of two chocolate chip cookie recipes, 13 out of 18 subjects
favored recipe A. Using the sign test, find the significance probability when
H 1 states that recipe A is preferable.
3. A firm attempting to determine if a difference exists in two manufacturing
methods. A sample of 10 workers was selected, and each worker completed
the production task using each of the two production methods.
Worker Method 1(minutes) Method 2(minutes)
1 10.2 9.5
2 9.6 9.8
3 9.2 8.8
4 10.6 10.1
5 9.9 10.3
6 10.2 9.3
7 10.6 10.5
8 10.0 10.0
9 10.7 10.2
10 10.9 10.2
Use the sign test and perform the null hypothesis that there is no overall
preference for one method over the other.
67
4. A social researcher interviews 25 newly married couples. Each husband
and wife are independently asked the question: “How many children would
you like to have?” The following data are obtained
Answer of Answer of
Couple Husband Wife Couple Husband Wife
1 3 2 14 2 1
2 2 2 15 3 2
3 2 1 16 2 2
4 2 3 17 0 0
5 5 1 18 1 2
6 0 1 19 2 1
7 0 2 20 3 2
8 1 3 21 4 3
9 2 2 22 3 1
10 3 1 23 0 0
11 4 2 24 2 3
12 1 2 25 2 2
13 3 3
Use the Sign test with   0.05 to test against two sided alternative the null
hypothesis that, for the population of families no difference in opinions
between husbands and wives.
5. A random sample of 80 sale managers was asked to predict whether next
year’s sale would be higher than, lower than, or about the same as in the
current year. The results are shown below. Test the null hypothesis that the
opinion of managers is evenly divided on the question against a two sided
alternative.
Prediction Number
Higher 37
Lower 28
About the same 15
6. Of a random sample of 120 university students, 67 expected to achieve a
better GPA than last year, 48 expected a lower GPA than last year, and
5 expected about the same GPA. Do these data present strong evidence that,
for population of students they are divided evenly on the expectations,
against the alternative that more expect a lower GPA compared with last
year?
68
7. Of a random sample of 150 university instructors, 62 believed that
student’s skills in solving problems increased over the last decade, 54
believed these skills had deteriorated and 4 saw no change. Evaluate the
strength of the sample evidence suggesting that, for all university instructors,
teachers are divided evenly on the issue against the alternative that more
teachers believe that student’s skills in solving problems have improved.
8. In a coffee taste test 48 individuals stated a preference for one of two well-
known brands. Results showed 28 favoring brand A, 16 favoring brand B,
and 4 undecided. Use the sign test with   0.10 to test the null hypothesis
that there is no difference in the preferences for the two brands of coffee
against a two sided alternative.
Answers
1. p- value = 0.1874 or 18.74%;2. p-value = 0.0482 or 4.82%;

3. p– value = 0.1798 or 17.98%; 4. T .S.  0.94 ; accept H 0 ; 5. T .S.  3.39 ;
p-value = 0.06%; 6. T.S.  1.77 ;p-value = 3.84%;7. T .S.  0.74 ;
p-value = 22.96%; 8. T.S.  1.81 ; reject H 0 .
69
2.3. The Wilcoxon signed test

(small sample size)
One disadvantage of the sign test is that it takes into account only a
very limited amount information-namely, the signs of the differences. The
Wilcoxon signed rank test provides a method to use information about the
magnitude of the differences between matched pairs. It is still a distribution
free test. Like many nonparametric test, it is based on ranks.
Table 2.1
Worker Method Method Difference Absolute Rank Rank
I II value of (+) (-)
difference
1 10.2 9.5 0.7 0.7 8
2 9.6 9.8 -0.2 0.2 2
3 9.2 8.8 0.4 0.4 3.5
4 10.6 10.1 0.5 0.5 5.5
5 9.9 10.3 -0.4 0.4 3.5
6 10.2 9.3 0.9 0.9 10
7 10.6 10.5 0.1 0.1 1
8 10.0 10.0 0 0 --
9 11.2 10.6 0.6 0.6 7
10 10.7 10.2 0.5 0.5 5.5
11 10.6 9.8 0.8 0.8 9
49.5 5.5
To demonstrate the use of the Wilcoxon signed ranked test let us consider a
manufacturing firm that is attempting to determine if a difference exists in
two production methods. A sample of 11 workers was selected, and each
worker completed the production task using each of the two production
methods. Each worker in the sample provides a pair of observations, as
shown in Table2.1. Table 2.1 also provides the difference in the completion
times. A positive value indicates that Method I require more time, and a
negative value indicates that Method II require more time. The statistical
question is whether or not the data indicate that the methods are significantly
70
different in terms of completion times. Thus the null and alternative
hypothesis can be written as
H 0 : The two populations of task completion times are identical
H 1 : The two populations of task completion times are not identical
As with the sign test, we ignore any difference of “0”, so sample size in
example above is reduced to n  10 . The nonzero absolute differences are
then ranked in ascending order of magnitude. That is, the smallest absolute
value 0.1 is given a rank of “1”. If two or more values are equal, they are
assigned the average of the next available ranks. In example above, absolute
value of difference-0.4 occurs twice. The rink assigned to them is therefore
the average of ranks 3 and 4-that is 3.5. The next absolute value-0.5 occurs
twice. The rank assigned to them is therefore the average of ranks 5 and 6-
that is 5.5. The next absolute value is assigned rank 7, and so on.
The ranks for positive and negative differences are summed separately. The
smaller of these sums is the Wilcoxon Signed Rank Statistic T.S.
Hence T.S.=5.5.
We will now suppose that the population distribution of the paired
differences is symmetric. The null hypothesis to be tested is that the center of
this distribution is 0. In example above, we are assuming that differences in
the task completion times have a symmetric distribution, and we want to test
whether that distribution is centered on 0-that is no difference between task
completion times.
Cutoff points for the distribution of this random variable are given in
Appendix (Table 4) for tests against a one sided alternative that the
population distribution of the paired differences is specified either to be
centered on some number bigger than 0 or to be centered on some number
less than 0. For sample size, n , the table shows, for selected probabilities  ,
the number T such that P(T  T )   . In other words, the null hypothesis
is rejected if T .S. is less than or equal to the corresponding number in the
Table4.
In example above, T .S.  5.5 . For n =10 we find that the null hypothesis will
be rejected for any significance level greater than   0.005 .
71
Steps in the Wilcoxon Signed Rank test for paired samples
1. Calculate the differences

2. Discard (ignore) any difference of “0”
3. Find absolute value of differences
4. Rank the absolute value of differences in ascending order of
magnitude. Rank positive and negative differences in two different
columns
4. Assign tied absolute differences (if any ties) the average of the ranks
they would receive if they were unequal but occurred in successive order
5. Find separately sum of ranks of positive and negative differences.
6. The smaller of the two sums is the Wilcoxon Signed Rank Statistic.
7. Reject the null hypothesis if the value of the test statistic is less than or
equal to the value in Appendix table 3.

(large sample size)
When the number of n nonzero differences in the sample is large (n >20), the
normal distribution provides a good approximation to the distribution of the
Wilcoxon statistic T under the null hypothesis that the population differences
are centered on 0.
Let T denote the smaller of the rank sums.
With increasing sample size of n (n>20) nonzero differences, the null
hypothesis is that the population differences are centered on 0, Wilcoxon
Signed Rank test has mean and variance given by
n(n  1)
E (T )  T 
4
and
n(n  1)(2n  1)
Var(T )   T2 
24
For large n, the distribution of the random variable, Z, is approximately
standard normal where
T  T
Z
T
72
If the number of nonzero differences is large and T is the observed value of
the Wilcoxon Signed test statistic, then the following tests have significance
level  ,
1. If the alternative hypothesis is one sided, reject the null hypothesis if
T  T
  z
T
2. If the alternative hypothesis is two sided, reject the null hypothesis if
T  T
  z / 2
T
Example:
A random sample of 38 students who had just completed courses in statistics
and accounting was asked to rate each in terms of level of interest, on a scale
from one (very uninteresting) to ten (very interesting). The 38 differences in
the pairs of ratings were calculated and the absolute differences ranked. The
smaller of the rank sums, which was for those finding accounting the more
interesting, was 278. Test at 5 % significance level the null hypothesis that
the population of students would rate these courses equally against the
alternative that the statistics course is viewed as the more interesting.
Also find the p-value.
Solution:
From the given information
n  38 ; T  278
The mean and variance of the Wilcoxon statistic are
n(n  1) 38  (38  1)
T    370.5
4 4
n(n  1)(2n  1) 38  39  77
 T2    4754.75
24 24
So the standard deviation is
 T  68.95
According to the condition, the null and alternative hypothesis can be written
as
H 0 : both courses rated equally interesting
H 1 : statistics course rated more interesting
If T is the observed value of the test statistic, the null hypothesis is rejected
against one sided alternative if
73
T  T
  z
T
Here, the value of T is T  278 and the value of test statistic is
T   T 278  370 .5
  1.34
T 68.95
  0.05 ; Fz ( z 0.05 )  0.95 ; z 0.05  1.65 ;  z 0.05  1.65
Since  1.34 is not less than  1.65 we fail to reject H 0 , and accept it.
The value of  corresponding to z  1.34 is, from Table 1 of the
Appendix, (1  0.9099)  0.0901 . Then the null hypothesis can be rejected at
all significance levels greater than 9.01%. The data contain modest evidence
suggesting that statistics course is more interesting.
Exercises
1. Two critics rate the service at six award winning restaurants on a

continuous 0 to 10 scale. Apply Wilcoxon signed rank test with   0.05 if
there is no difference between the critics’ ratings?
Restaurant Critic 1 Critic 2
1 6.2 8.4
2 5.3 5.8
3 7.5 7.1
4 7.4 7.0
5 4.3 5.1
6 9.8 9.9
2. Two computer specialists estimated the amount of computer memory
(in gigabytes) required by five different offices
Office Specialist A Specialist B
1 5.7 6.1
2 6.4 6.8
3 3.2 3.1
4 2.0 2.9
5 8.1 12.3
Apply Wilcoxon signed rank test with   0.05 to test the null hypothesis
that there is no difference between estimations against a two sided
alternative.
74
3. Twelve customers were asked to estimate the selling price of two models
of refrigerators. The estimates of selling price provided by the customers are
shown below:
Customer Model A Model B
1 $650 $900
2 760 720
3 740 690
4 700 850
5 590 920
6 620 800
7 700 890
8 690 920
9 900 1000
10 500 690
11 610 700
12 720 700
Use these data and test at the 0.05 level of significance to determine if there
is no difference in the customers’ perception of selling price of the two
models.
4. A certain brand of microwave oven was priced at 12 stores in two
different cities.
These data are presented below:
District A District B
18 500 16 700
16 000 20 500
12 000 23 000
20 000 17 500
19 000 22000
17 000 21 000
16 500 21 500
19 000 19 500
15 500 17 000
16 000 23 000
17 500 21 000
18 000 22 000
Use a 0.05 level of significance and apply the Wicoxon signed rank test to
test whether o not prices for the microwave oven are the same in the two
cities.
75
5. The company is interested in the impact of the newly introduced quality
management program on job satisfaction of workers. A random sample of 34
workers was asked to assess level of satisfaction on a scale from 1 to 10 two
month before the program. These same sample members were asked to make
this assessment again two month after the introduction of the program. The
34 differences in the pairs of ratings were calculated and absolute differences
ranked. The smaller of the rank sums, which was for those more satisfied
before the introduction of the program, was 178. What can be concluded
from these findings?
6. A random sample of 90 members was taken. Each sample member was
asked to assess the amounts of time in a month spent watching TV and the
amounts of time in a month spent reading. The 90 differences in times spent
were then calculated and their absolute differences ranked. The smaller of
the of the rank sums, which was for watching TV, was 1680. Test the null
hypothesis that the population amounts of time spent on watching TV and
reading divides equally against the alternative that watching TV takes more
amounts of time.
7. Suppose you wish to test hypothesis that two treatments, A and B, are
equivalent against the alternative that the responses for A tend to be larger
than those of B. If the number of pairs equals 25, and smaller of the rank of
the absolute differences is 273, then what would you decide? Use   5%
then find p-value for the test and interpret it.
8. An experiment was conducted to compare two print types, A and B, to
determine whether type A is easier to read. A sample of 22 persons was
given the same material to read. First they read the material printed with type
A, then read the same material printed with type B. The times necessary for
each person to read the materials (in seconds) were
Type A: 95;122;101;99;108;122;135;127;119;127;99;98;97;96;112;97;100;
116; 111;117;102;103
Type B: 110;102;115;112;120;117;119;127;137;119;99;100;102;103;118;
99;89;97;112;116; 178; 94.
Do the data provide sufficient evidence to indicate that print type A and print
type B are the same for reading against the alternative that print type A is
easier to read? Test using   0.05 .
76
Answers
1. T .S.  7 ; accept H 0 ; 2. T.S.  1 ; reject H 0 ;3. T .S.  6 ; reject H 0 ;
4. T .S.  3 ; reject H 0 virtually at any levels;5. T.S.  2.04 ;p-value = 4.12%;
6. T.S.  1.48 ; reject H 0 at levels higher than 6.94%;7. T .S.  2.97 ; accept
H 0 at any levels; 8. T.S.  0.71 ; reject H 0 .
2.4. The Mann-Whitney test
Suppose two independent random samples are to be used to compare

two populations. We may be unwilling to make assumptions about the form
of the underlying population probability distributions or we may be unable to
obtain exact values of the sample measurements. If the data can be ranked in
order of magnitude for either of these situations, the Mann-Whitney test
(sometimes called Mann-Whitney U test) can be used to test the hypothesis
that the probability distributions associated with the two populations are
equivalent.
Assume that apart from any possible differences in central location, that the
two population distributions are identical. Suppose that n1 observations are
available from the first population and n 2 observations from the second
population. The two samples are pooled and the observations are ranked in
ascending order, with ties assigned the average of the next available ranks.
Let R1 denote the sum of the ranks from the first population. The Mann-
Whitney statistic is
n  (n  1)
U  n1  n2  1 1  R1
2
In testing the null hypothesis that the central locations of the two population
distributions are the same, we assume that the two population distributions
are identical. It can be shown that if the null hypothesis is true, the random
variable U has mean
n n
E (U )  U  1 2
2
and variance
n  n  (n1  n2  1)
Var(U )   U2  1 2
12
77
Then for large sample sizes (both at least 10), the distribution of the random
variable,
U  U
Z
U
is well approximated by the standard normal distribution.
Decision rules for the Mann-Whitney test
Suppose that two population distributions are identical, apart from

any possible differences in central location. In testing the null hypothesis the
two population distributions have the same central location, the following
test have significance level  :
H 0 : Two population distributions have the same central location
1. If the alternative hypothesis is one sided hypothesis that the location of
population 1 is higher than the location of population 2, the decision rule is
U  U
Reject H 0 if   z
U
2. If the alternative hypothesis is one sided hypothesis that the location of
population 1 is lower than the location of population 2, the decision rule is
U  U
Reject H 0 if  z
U
3. If the alternative hypothesis is two sided hypothesis that the two
population distributions differ, the decision rule is
U  U U  U
Reject H 0 if   z / 2 or  z / 2
U U
Example:
Let us demonstrate the methodology of the Mann-Whitney test by using it
conduct a test on the population of account balances at two branches of some
Bank. Data collected from two independent simple random samples, one
from each branch, are shown in Table 2.2.
78
Table2.2
Branch 1 Branch 2
Sampled Account Sampled Account
Account balance account balance
1 1 095 1 885
2 955 2 850
3 1 200 3 915
4 1 195 4 950
5 925 5 800
6 950 6 750
7 805 7 865
8 945 8 1 000
9 875 9 1 050
10 1 055 10 935
11 1 025
12 975
The first step in the Mann- Whitney test is to rank the combined (pooled)
data from the two samples from low to high. Using the combined set of 22
observations shown in Table 2.2, the lowest value of $750(item 6 of
sample2) is ranked number 1. Continuing the ranking, we have
Account balance Item Rank
750 6 of sample 2 1
800 5 of sample 2 2
805 7 of sample 1 3
…… ……………. …
1 195 4 of sample1 21
1 200 3 of sample 1 22
Item 6 of sample 1 and item 4 of sample 2 both have the same account
balance, $950. We could give one of these items a rank 12 and the other a
rank 13, but this could lead to an erroneous conclusion. In order to avoid this
difficulty the usual treatment for tied data values is to assign each value the
rank equal to the average of the ranks associated with the tied items. Thus
the tied observations of $950 are both assigned ranks of 12.5. Table 2.3
shows the entire data set with the rank of each observation.
79
Table2.3
Branch 1 Branch 2
Sampled Account Sampled Account
Account balance Rank account balance Rank
1 1 095 20 1 885 7
2 955 14 2 850 4
3 1 200 22 3 915 8
4 1 195 21 4 950 12.5
5 925 9 5 800 2
6 950 12.5 6 750 1
7 805 3 7 865 5
8 945 11 8 1 000 16
9 875 6 9 1 050 18
10 1 055 19 10 935 10
11 1 025 17
12 975 15________________________________
Sum of ranks 169.5 83.5
The next step in the Mann-Whitney test is to sum the ranks for each sample.
These sums are shown in Table 2.3. The test procedure can be based upon
the sum of the ranks for either sample. In the following discussion we use
the sum of the ranks for the sample from branch 1. We will denote this sum
by R1 . Thus, in our example R1  169.5 .
The value observed for the Mann-Whitney test is
n  (n  1) 12  13
U  n1  n2  1 1  R1  12  10   169.5  28.5
2 2
Since two samples are selected from identical populations and n1 and n 2
each is 10 or greater, the sampling distribution of U can be approximated by
a normal distribution with mean
n n 12  10
E (U )  U  1 2   60
2 2
and variance
n  n  (n1  n2  1) 12  10  23
Var(U )   U2  1 2   230
12 12
80
Suppose that we want to test the null hypothesis that the central locations of
the distributions of account balance are identical against the two-sided
alternative for   0.05 . The decision rule is to reject the null hypothesis if
U  U U  U
  z / 2 or  z / 2
U U
Here
U  U 28.5  60
  2.08
U 230
z / 2  z 0.025  1.96 and  z 0.025  1.96
Since -2.08 is less than -1.96, we reject the null hypothesis that two
population account balances are identical. Thus we conclude that two
populations are not identical. The probability distribution of account
balances at branch 1 is not the same as that at branch 2.
Now, from Table1 of the Appendix, the value of  / 2 corresponding to a
value (-2.08) is 0.0188, so the corresponding  is 0.0376
p  value  2  (1 - Fz (test statistics))  2(1 - 0.9812)  0.0376
The null hypothesis will be rejected for any significance level higher than
3.76%. Thus, these data do not contain strong evidence against the
hypothesis that the central locations of accounts at two branches are the
same. There is very strong support that two branches account balances are
not identical.
81
Exercises
1. Starting salaries were recorded for ten recent business administration
graduates at each of two well-known universities. Use   0.1 and test for
the difference in the starting salaries from the two universities is zero against
the alternative that starting salaries are higher for the university A.
University A University B
Student Monthly salary ($) Student Monthly salary ($)
1 890 1 1 000
2 950 2 1 020
3 1 200 3 1 140
4 1 150 4 1 000
5 1 300 5 975
6 1 350 6 925
7 990 7 900
8 1 050 8 1 025
9 1 400 9 1 075
10 1 450 10 930
2. The following data show product weights for items produced on two
production lines
Line 1: 13.6; 13.8; 14.0; 13.9; 13.4; 13.2; 13.3; 13.6; 12.9; 14.4
Line 2: 13.7; 14.1; 14.2; 14.0; 14.6; 13.5; 14.4; 14.8; 14.5; 14.3; 15.0; 14.9
Test that the difference between the product weights for the two lines is zero
against the alternative that product weights of second line is higher.
Use   0.10 . Also find p-value.
3. A random sample of 14 male students and an independent random sample
of 16 female students were asked to write essays at the conclusion of a
writing course. Their grades were recorded below:
Male: 75; 80; 60; 80; 95; 100; 65; 70; 75; 60; 50; 55; 90; 95
Female: 85; 70; 90; 100; 95; 67; 50; 50; 67; 83; 78; 62; 43; 97; 89; 73
Test the 5% significance level null hypothesis that, in the aggregate the male
and female students are equally ranked, against a two-sided alternative. Also
find p-value.
4. For a random sample of 12 management department gradates and 14
economics department graduates were asked their starting salaries. Those
salaries were then ranked from 1 to 26. The following rankings resulted
Management: 2; 6; 7; 1; 11; 20; 8; 14; 21; 12; 4; 26
Economics: 13; 3; 17; 25; 5; 9; 10; 24; 15; 23; 16; 22; 18; 19
Analyze the data using the Mann-Whitney test, and comment on the results.
82
5. Starting salaries of graduates from two leading universities were
compared. Independent random samples of 40 from each university were
taken, and the 80 starting salaries were pooled and ranked. The sum of the
ranks for students from one of these universities was 1450. Test the null
hypothesis that the central locations of the population distributions are
identical against two sided alternative.
6. A stock market analyst produced at the beginning of the year a list of
stocks to buy and another list of stocks to sell. For a random sample of ten
stocks from the “buy list”, percentage returns over the year were as follows:
10.6; 5.2; 12.8; 16.2; 10.6; 4.3; 3.1; 11.7; 13.9; 11.3
For an independent random sample of ten stocks from the “sell list”,
percentage returns over the year were as follows:
-2.6; 6.1; 9.9; 11.3; 2.3; 3.9; -2.3; 1.3; 7.9; 10.8
For   0.05 use the Mann-Whitney test to interpret these data. Also find
and interpret p-value.
Answers
1. T .S.  ; reject H 0 ;2. T .S.  ; reject H 0 ; p-value = 0.3%;
3. T .S.  ; accept H 0 ; 4. T .S.  ; p- value =12.36%; H 0 will be rejected
at all levels higher than 12.36%; 5. T .S.  ; p-value = 0.101; H 0 will be
rejected at any level higher than 10.1%; 6. T .S.  ; reject H 0 at 5%;
p- value = 2.58%.
83
Chapter 3
Simple linear regression
3.1. Introduction
In day-to-day decisions-making situations, businesspersons and

economists frequently draw conclusions and make recommendations based
on the relationship between two variables. For example, a marketing
manager may project sales volume based upon observed relations between
advertising expenditures and sales volume. Although in some instances the
manager will rely on his or her intuition as to how the variables are related,
the safest approach, by far, is to collect data on the two variables and then
evaluate their relationship statistically. These relationships are expressed
mathematically as
y  f (x)
where the function may follow linear and nonlinear forms.
3.2. The scatter diagram
As a first step in determining if a relationship exists between two variables

we could plot or graph the available data for the two variables. Suppose that
a sales manager has recorded containing data on annual sales and years of
experience. The information is given in the following table:
Salesperson 1 2 3 4 5 6 7 8 9 10
Years 1 3 4 4 6 8 10 10 11 13
of experience
Annual sales 80 97 92 102 103 111 119 123 117 136
($1000’s)
Let us plot these data on a graph with years of selling experience on the
horizontal axis and annual sales on the vertical axis. We now have a scatter
diagram. It is given this name because the plotted points are “scattered”
over the graph or diagram. The scatter diagram for these data is shown in
Figure 3.1.
84
y
140
Annual sales ($1000’s)
130
120
110
100
90
80
70 x
1 2 3 4 5 6 7 8 9 10 11 12 13
Years of experience
Fig.3.1. Scatter diagram of annual sales and years of experience
In regression analysis statisticians commonly will classify a variable as an

independent or a dependent variable. The classification is used to indicate
which variable is doing the predicting or explaining (independent variable)
and which variable is being predicted or explained (dependent variable). In
our example, the years of selling experience is referred to as the independent
variable. It is used to predict the sales volume, or dependent variable.
Does the scatter diagram in Fig. 3.1 allow us to draw conclusions?
It gives us an overview of the data. It indicates that in this case there is a
good chance that the variables are related. In fact, it appears that the
relationship between these two variables may be approximated by a straight
line or linear function.
85
3.3. Correlation analysis
We will introduce some statistical measures that provide greater precision

for describing relationships.
Let X and Y be a pair of random variables, with means  x and  y , and
variances  x2 and  y2 . As a measure of the association between these
variables, we introduced the covariance, defined as
n _ _

i 1
( xi  X )( y i  Y )
Cov ( x, y )  S xy 
n 1
_ _
where x i and y i are the observed values, X and Y are the sample means, and
n is the sample size.
A positive value of the covariance indicates a direct or increasing linear
relationships and a negative value of covariance indicates a decreasing linear
relationship. Positive association indicates that the high values of X tend to
be associated with high values of Y and low X with low Y. When there is a
negative association, so that high values of X are associated with low values
of Y and low X with high Y, the covariance is negative. If there is no linear
association between X and Y, their covariance is 0.
Another measure of the relationship between two variables is the correlation
coefficient. In this section we will consider the simple linear correlation, for
short linear correlation, which measures the strength of the linear association
between two variables.
Definition:
The simple linear correlation, denoted by, rxy , measures the strength of the
linear relationship between two variables for a sample and is calculated as
n _ _
Cov ( x, y )
 (x
I 1
i  x)( y i  y )
rxy   (3.1)
sx  s y n _ n _
 (x
i 1
i  x)  2
(y
i 1
i  y) 2
An equivalent expression is
86
n _ _
x y
I 1
i i nx y
rxy  (3.2)
 n 2 _   n 2 _ 

 xi  n  ( x) 2   
 i 1
 
  i 1
y i n  ( y ) 2 

1. The sample correlation coefficient ranges from  1 to  1 with,

a) rxy  1 indicates a perfect positive linear relationship;
b) rxy  0 indicates no relationships between X and Y
c) rxy  1 indicates a perfect decreasing linear relationship between
X and Y.
2. Positive correlations indicate positive or increasing linear relationship
with values closer to +1, indicating data points closer to a straight line, and
closer to 0, indicating greater deviations from a straight line.
3. Negative correlations indicate negative or decreasing linear relationship
with values closer to -1, indicating data points closer to a straight line, and
closer to 0, indicating greater deviations from a straight line.
If rxy  1 , it is said to be a case of perfect positive linear correlation. In such
cases, all points in the scatter diagram lie on a straight line that slopes
upward from left to right, if rxy  1 , the correlation is said to be a perfect
negative linear correlation. In this case, all points in a scatter diagram fall on
a straight line that slopes downward from left to right.
If the correlation between two variables is positive and close to 1,
we say that the variables have a strong positive linear correlation. If the
correlation between two variables is positive but close to 0, then the
variables have a weak positive linear correlation. On the other hand, if the
correlation between two variables is negative and close to 1, then the
variables are said to have a strong negative linear correlation. Also, if the
correlation between two variables is negative and close to 0, there exists a
weak negative linear correlation between the variables.
Example:
An economist is interested in the relationship between food expenditure and
income. Calculate the sample correlation coefficient for the data recorded on
monthly incomes and food expenditure of seven households.
87
Household 1 2 3 4 5 6 7
Income (100’s of $) 35 49 21 39 15 28 25
Food expenditure (100’s of $) 9 15 7 11 5 8 9
Solution: The sample means are

_
x
 xi 212
  30.29 ;
_
y
y i

64
 9.14
n 7 n 7
The sample correlation coefficient can be calculated either by (3.1) or (3.2)

It is more convenient to use (3.2) to calculate correlation coefficient:
Necessary calculations of the sample correlation for the data are set out in
the following table 3.1
Table 3.1
Food
Household Income expenditure xi  y i x i2 y i2
( xi ) ( yi )
1 35 9 315 1225 81
2 49 15 735 2401 225
3 21 7 147 441 49
4 39 11 429 1521 121
5 15 5 75 225 25
6 28 8 224 784 64
7 25 9 225 625 81
Sums 212 64 2150 7222 646
Hence, the sample correlation is:

n _ _
x y
I 1
i i nx y
rxy  
 n _
2 
n _
2
 
 xi  n  ( x)   
 i 1
2
 
  i 1

y i  n  ( y ) 
2

2150  7  (30.29)  (9.14) 212.05
   0.96
(7222  7  (30.29) 2 )  (646  7  (9.14) 2 ) 221.25
88
The sample correlation, 0.96, indicates very strong positive relationships
between monthly income and food expenditure. The high value of monthly
income tends to be associated with the higher value of food expenditure.
3.3.1. Hypothesis test for correlation
The sample correlation coefficient rxy is useful as a descriptive measure of

the strength of linear association in a sample. We can also use the correlation
coefficient to test the null hypothesis that there is no linear association in the
population between a pair of random variables; that is
H0 :   0
We can show that when the null hypothesis is true and the random variable
have a joint normal distribution then the random variable
rxy  n  2
t
1  rxy2
follows a Student’s t distribution with (n-2) degrees of freedom. The
following tests of the null hypothesis
H0 :   0
have a significance level of  :
1. To test H 0 against the alternative
H1 :   0
reject H 0 if T .S .  t n  2,
2. To test H 0 against the alternative
H1 :   0
reject H 0 if T .S .  t n  2,
3. To test H 0 against the two sided alternative

H1 :   0
reject H 0 if T .S .  t n 2, / 2 or T .S .  t n  2, / 2
89
rxy  n  2
where T .S .  , and t n 2, is the number for which
1  rxy2
P(t n 2  t n 2, )  2
where the random variable t n2 follows a Student’s t distribution with (n -2)
degrees of freedom.
Example:
A sample data set produced the following information
n  10 ; 
xi  66 ; 
yi  588 ; 
xi yi  2244 ;
 xi2  396 ; and   58734

yi2
Find the sample correlation, and test against a two sided alternative the null
hypothesis that the population correlation is 0. Take   0.05 .
Solution: Denoting by  the population correlation, we want to test
H0 :   0
against the two sided alternative
H1 :   0
reject H 0 if T .S .  t n 2, / 2 or T .S .  t n  2, / 2
Firstly, let us find the value of sample correlation coefficient
2244  12  5.5  49 990
rxy    0.996
(396  12  (5.5) )  (58734  12  (49) )
2 2 993.69
rxy  n  2  0.996  10  2 2.817
T .S .     31.5
1  rxy
2
1  (0.996 ) 2 0.0894
t n 2, / 2  t 8,0.025  2.306 and  t 8,0.025  2.306

Since  31.5  2.306 we reject H 0 . Virtually for any level of  we reject
hypothesis that there is no association between x and y. These data contain
very strong evidence of positive (linear) association between x and y.
90
Exercises
1. For the data set
x 0 1 6 3 5
y 4 3 0 2 1
a) Construct a scatter diagram

b) Guess the sign and value of the correlation coefficient
c) Calculate the correlation coefficient.
2. A sample data set produced the following information.
n  460 ; 
xi  9880 ; 
yi 1456 ; x y i i  85080 ;
 xi2  485870 ; and  yi2 135675

Calculate the linear correlation coefficient rxy .
3. Calculations from a data set of n  48 pairs of (x, y) values have provided
the following results
_ _ _ _
 ( xi  x) 2  260 .2 ;  ( y i  y ) 2  403.7 ;  ( xi  x)( y i  y )  298 .8
Calculate the linear correlation coefficient.
4. The following table gives the experience (in years) and monthly salaries
(in hundred of dollars) of nine randomly selected secretaries
Experience 14 3 5 6 4 9 18 5 16
Monthly salary 22 12 15 17 15 19 24 13 27
a) Do you expect the experience and monthly salaries to be positive or

negatively related?
b) Compute the correlation coefficient.
c) Test at the 10% significance level, against a two sided alternative, the null
hypothesis that the population correlation coefficient is zero.
5. The following data were collected regarding the starting monthly salary
and the grade point average (GPA) for students who had obtained a degree in
business administration and economics:
GPA 2.6 3.4 3.6 3.2 3.5 2.9

Monthly salary ($) 900 1200 600 1100 1400 1000
91
a) Develop a scatter diagram for the above data.
b) Compute the sample correlation coefficient between grade point average
and salary.
c) Test at the 5% significance level the null hypothesis that the population
correlation coefficient is zero against the alternative that it is positive.
6. The management of a supermarket wanted to check the effect of the
number of broadcast on TV on the gross sales at the store. The management
experimented for eight weeks by broadcasting a different number of
commercials each week on TV. The following table gives the number of
commercials during each week and the gross sales (in 1000’s of dollars)
Number 22 16 28 12 30 19 24 32
of commercials
Gross sales 3.64 3.12 4.08 2.84 3.98 3.55 4.02 4.38
per week
a) Compute the sample correlation coefficient between number of broadcasts

and gross sales.
b) Test the null hypothesis that number of broadcasts and gross sales are
uncorrelated in the population against the alternative that population
correlation is positive.
Answers
1. b) high negative correlation: c) -0.992; 2. 8.99; 3. 0.92; 4. a) positively;
b) 0.95; c) T.S.  8.051 ; fail to reject H 0 ; 5. b) 0.114; c) T .S.  0.23 ; reject
H 0 ; 6. a) 0.96; b) T .S.  8.40 ; reject H 0 at virtually any level.
92
3.4. Spearman rank correlation
Suppose that a random sample ( x1 , y1 ), ( x2 , y 2 ),........ , ( x n , y n ) of n pairs of
observations is taken. If x i and y i are each ranked in ascending order and the
sample correlation of these ranks is calculated, the resulting coefficient is
called the Spearman rank correlation coefficient. If there are no tied
ranks, an equivalent formula for computing this coefficient is
n
6 d
i 1
i
2
rs  1 
n( n  1) 2
where the d i are the differences of the ranked pairs:

d i  rank ( xi )  rank ( y i )
Remark: Spearman rank correlation shares the properties of rxy that
 1  rs  1 and that values near +1 indicate a tendency for the larger values
of X to be paired with the larger values of Y.
The following test of the null hypothesis H 0 of no association in the
population
H 0 : x and y are independent
have significance level  :
1. To test against the alternative of positive association, the decision rule is
Reject H 0 if rs  rs ,
2. To test against the alternative of negative association, the decision rule is
Reject H 0 if rs   rs ,
3. To test against the two sided alternative of some association, the decision
rule is
Reject H 0 if rs  rs , / 2 or rs  rs , / 2
The table of critical values of the Spearman rank correlation coefficient is
given in the Table 5 of the Appendix.
Example:
Scores that 8 salesmen made on a test that measures their aggressiveness (x ),
and their sales in thousands of dollars for their second year with a certain
company (y ).
x (aggressiveness) 30 17 35 28 42 25 19 34
y (sales) 35 31 40 46 50 32 33 42
93
a) Find and interpret Spearman rank correlation
b) Test the null hypothesis that aggressiveness and sales are independent
again the alternative that they are positively correlated. Take   0.05 .
Solution:
a) First of all, let us rank separately x and y in ascending order. These two
rank appear in third and fourth columns of the following table 3.2
Table 3.2
x y Rank x i Rank y i d i  xi  y i d i2
30 35 4 5 1 1
17 31 8 8 0 0
35 40 2 4 -2 4
28 46 5 2 3 9
42 50 1 1 0 0
25 32 6 7 -1 1
19 33 7 6 1 1
34 42 3 3 0 0
sum 16
The differences between ranks and squared differences between ranks are
shown in the last two columns of the table. Substituting the values n  8 and
 d i2  16 into formula for Spearman rank correlation, we obtain
n
6 d
i 1
i
2
6  16
rs  1  1  1  0.19  0.81
n( n  1)
2
8  63
It means that there exists strong positive correlation between aggressiveness
and sales volume.
b) The null and alternative hypotheses are
H 0 : x and y are independent
H 1 : x and y are positively correlated
Reject H 0 if rs  rs ,
For a sample of size n=8, and   0.05 ,
rs ,  r8,0.05  0.643
Since 0.81>0.643 we reject H 0 , and accept the alternative hypothesis that x
and y are positively correlated.
94
Exercises
1. Specify the rejection region for Spearman’s nonparametric test for rank
correlation in each of the following cases
a) H 0 :   0; H1 :   0; n  10;   0.05
b) H 0 :   0; H1 :   0; n  20;   0.025
c) H 0 :   0; H 1 :   0; n  30;   0.01
2. Compute Spearman’s rank correlation coefficient for each of the
following pairs of sample observations
a) b)
x 33 61 20 19 40 x 5 20 15 10 3
y 26 36 65 25 35 y 80 83 91 82 87
3. A random sample of nine pairs of observations are recorded on two

variables, x and y.
x 19 27 15 35 13 29 16 22 17
y 12 19 7 25 11 10 16 13 18
Do the data provide sufficient evidence to indicate that  , the rank
correlation between x and y, differs from zero? Test using   0.05 .
4. Two expert wine testers were asked to rank six brands of wine. Their
rankings are shown in the table.
Brand Expert 1 Expert 2

A 6 5
B 5 6
C 1 2
D 3 1
E 2 4
F 4 3
Do the data present sufficient evidence to indicate a positive correlation in
the rankings of the two experts?
5. Refer to the data of Exercise 6 of the previous section. Find Spearman’s
rank correlation coefficient, and use it to test, against two sided alternative,
95
the null hypothesis of no association in the population between this pair of
random variables.
6. Refer to the data of Exercise 5 of the previous section. Find Spearman’s
rank correlation coefficient, and use it to test the null hypothesis that these
quantities are uncorrelated in the population against the alternative that
population correlation is negative.
Answers
1. a) rs  0.648 or rs  0.648 ; b) rs  0.450 ; c) rs  0.432 ;2. a) rs  0.4 ;

b) rs  0.2 ; 3. rs  0.48 ; accept H 0 ; 4. rs  0.66 ; reject H 0 virtually at any
level; 5. rs  0.93 ; reject H 0 virtually at any level; 6. rs  0.143 ; can not
reject H 0 at 5% level.
3.5. The linear regression model
Let us return to the example of an economist investigating the

relationship between food expenditure and income. What factors or variables
does a household consider when deciding how much money should be spent
on food every week or every month? Certainly, income of household is one
factor. Many other factors, say, the size of household, the preferences and
tests of household members, are some of the variables that will influence a
household’s decision about food expenditure. These variables are called
independent variables because they are all vary independently and they
explain the variation in food expenditure among different households. In
other words, these variables explain why different households spend
different amounts of money on food. Food expenditure is called the
dependent variable because it depends on the independent variables.
Studying the effect of two or more independent variables on a dependent
variable using regression analysis is called multiple regression. If we
choose only one (usually the most important) independent variable and study
the effect of that single variable on a dependent variable, it is called a simple
regression. Thus, simple regression includes only two variables: one
independent and one dependent.
Definition: A regression model is a mathematical equation that describes
relationship between two or more variables. A simple regression model
includes only two variables: one independent and one dependent. The
96
dependent variable is the one being explained and the independent variable
is the one used to explain the variation in the dependent variable.
The relationship between two variables in a regression analysis is expressed
by a mathematical equation called a regression equation or model.
A regression equation that gives a straight line relationship between two
variables is called a linear regression model; otherwise, it is called a
nonlinear regression model. In this chapter we will consider only linear
regression model.
In a regression model, the independent variable is usually denoted by x and
the dependent variable is usually denoted by y. Simple linear regression
model is written as
y    x (1)
In model (1),  gives the value of y for x  0 , and  gives the change in y
due to a change of one unit in x. This model simply states that y is
determined exactly by x and for a given value of x there is one and only one
value of y. For example, if y is food expenditure and x is income, then model
(1) would state that food expenditure is determined by income only and that
all households with the same income will spend the same amount on food.
But as mentioned above, food expenditure is determined by many variables,
only one of which is included in model (1). In reality, different households
with the same income spend different amounts of money on food because of
the differences in size of the household, their preferences and tastes. Hence,
to take these variables into consideration and make model complete, we add
another term to the right side of model (1). This term is called the error
term. It is denoted by  (Greek letter epsilon). The complete regression
model is written as
y i      xi   i (2)
Equation (2) is called the population (or true) regression line of y on x.

In equation (2)  and  are the population model coefficients and 
is a random error term .
Population data are difficult to obtain. As a result, we almost always use

sample data to estimate model (2). The estimated regression model is given
by the equation
yi  a  b  xi  ei
97
where a and b are estimated values of the coefficients and e is the difference
between the predicted value of y on the regression line, defined as
^
y i  a  b  xi
^
and the observed value y i . The difference between y i and y i for each value
of x is defined as the residual
^
ei  y i  y i  y i  ( a  b  x i )
Thus for each observed value of x there is a predicted value of y from the
estimated model and an observed value. The difference between the
observed and predicted values of y is defined as the residual. The residual,
ei , is not the model error,  , but is the combined measure of the model
error and errors in estimating, a and b, and in turn the errors in estimating the
predicted value.
3.5. 1. Least squares coefficient estimators
The population regression line is useful theoretical construct, but for

applications we need to determine an estimate of the model using available
data. Suppose that we have n pairs of observations ( x1 , y1 ), ( x 2 , y 2 ),...
....( xn , y n ) . We would like to find the straight line that best fits these points.
To do this we need estimators of unknown coefficients  and  of the
population regression line.
We obtain the coefficient estimators, a and b using equations derived by
using the least squares procedure. As shown in Figure 3.2 there is a
^
deviation, ei between the observed, y i and the predicted value, y i , on the
^
estimated regression equation for each value of x , where ei  yi  y i .
98
^
y ( xi , y i )
ei
^
( x1 , y1 ) ( xi , y i )
e1
+ ( x1 , y1 )
x
x1 xi
Fig. 3.2
Some of the ei will be positive and some negative. We then compute a

mathematical function that represents the effect of squaring all of the
residuals and computing the sum of the squared residuals. This function-
whose left side is labeled SSE –includes the coefficients, a and b. The
quantity SSE is defined as the “Error Sum of Squares”. The coefficient
estimators a and b are selected as the estimators that minimize the Error Sum
of Squares.
3.5.2. Least square procedure
The least square procedure obtains estimates of the linear equation

coefficients, a and b , in the model
^
y i  a  b  xi
by minimizing the sum of the squared residuals ei
 
^
SSE  ei2  ( yi  y i ) 2
The coefficients a and b are chosen so that the quantity
 
SSE  ei2  ( yi  (a  bxi )) 2
is minimized. It can be shown that the resulting estimates are
99
n _ _ n _ _
 (x
i 1
i  x)( y i  y ) x y
i 1
i i nx y
b n
 n
_ _

i 1
( xi  x) 2 i 1
xi2  n  ( x) 2
_ _
and a  y b  x
_ _
where x and y are the respective sample means.
The line
^
y a b x
is called the sample regression line or the least squares regression line of
y on x.
Example:
Find the least squares regression line for the data on incomes (in hundreds of
dollars) and food expenditures of seven households given in the table below.
Household 1 2 3 4 5 6 7
Income x 35 49 21 39 15 28 25
Food expenditure y 9 15 7 11 5 8 9
Use income as an independent variable and food expenditure as a dependent

variable.
Solution:
^
We are to find the values of a and b for the regression model y i  a  b  xi .
The following table shows the calculations required for the computations of
a and b.
Using data from the table 3.3 we find
_ _
212 64
x  30.2857 ; y  9.1429
7 7
100
Table3.3
Food
Household Income expenditure xi  y i x i2
( xi ) ( yi )
1 35 9 315 1225
2 49 15 735 2401
3 21 7 147 441
4 39 11 429 1521
5 15 5 75 225
6 28 8 224 784
7 25 9 225 625
Sums 212 64 2150 7222
n _ _

i 1
xi y i  n  x  y
2150  7  (30.2857 )  (9.1429 )
b   0.2642
n _
7222  7  (30.2857 ) 2
x i 1
2
i  n  ( x) 2
_ _
a  y  b  x  9.1429  (0.2642 )  (30.2857 )  1.1414
^
Thus, our estimated regression model y  a  b  x is
^
y  1.1414  0.2642  x
This regression line is called the least squares regression line. It gives the
regression of food expenditure on income.
Using this estimated model, we can find the predicted value of y for a
specific value of x. For example, suppose that we randomly select a
household whose monthly income is $3500 so that x  35 (x denotes income
in hundred of dollar in our example). The predicted value of food
expenditure for this household is
^
y  1.1414  0.2642  35  $10.3884 hundred
In other words, based on our regression line, we predict that a household
with a monthly income of $3500 is expected to spend $1038.84 per month
on food.
101
3.5.3. Interpretation of a and b
a) Interpretation of a
Consider a household with zero income. Using the estimated regression line
obtained above, the predicted value of y for x  0 is
^
y  1.1414  0.2642  0  $1.1414 hundred
Thus, we can state that a household with no income is expected to spend
$114.4 per month on food. We should be very careful while making this
interpretation of a. In example of seven households, the incomes vary from a
minimum of $1500 to a maximum of $4900. Hence, our regression line is
valid only for the values of x between 15 and 49. If we predict y for a value
of x outside this range, the prediction usually will not hold true. Thus, since
x  0 is outside the range of household incomes that we have in the sample
data, the prediction that a household with zero income spends $114.14 per
month on food does not carry much credibility.
b) Interpretation of b
The value of b in a regression model gives the change in y (dependent
variable) due to a change of one unit in x (independent variable).
^
For example, by using the regression line y  1.1414  0.2642  x
^
when x  30 ; y  1.1414  0.2642  30  9.0674
^
when x  31 ; y  1.1414  0.2642  31  9.3316
^
Hence, when x increased by one unit, from 30 to 31, y increased
by 9.3316  9.0674  0.2642 , which is the value of b. Because of unit of
measurement in hundred of dollars, we can state that, on average, a $100
increase in income will cause a $26.42 increase in food expenditure. We can
also state that, on average, a $1 increase in income of household will
increase the food expenditure by $0.2642.
Note that when b is positive, an increase in x will lead to an increase in y and
decrease in x will lead to a decrease in y. Such a relationship between x and y
is called a positive linear relationship. On the other hand, if the value of b is
negative, an increase in x will cause a decrease in y and a decrease in x will
cause an increase in y. Such a relationship between x and y is called a
negative linear relationship.
102
The values of y- intercept and slope calculated from sample data on x and y
are called estimated values of  and  and denoted by a and b. Using a and
b we can write estimated model as
^
y a b x
^
where y (read as y hat) is the estimated or predicted value of y for a given
value of x.
3.5.4. Assumptions of the regression model
Like any other theory, the linear regression analysis is also based on certain
assumptions. Consider the population regression model
y i      xi   i
There are four assumptions made about this model.
Assumption1: The random error term  has a mean equal to zero for each x.
In other words, among all households with the same income, some spend
more than predicted food expenditure; others spend less than predicted food
expenditure. Some of positive errors equal to the sum of negative errors so
that the mean of errors for all households with the same income is zero.
Assumption 2: The errors associated with different observations are
independent. According to this assumption, the errors for any two
households are independent. All households decide independently how much
spend on food.
Assumption 3: For any given x , the distribution of errors is normal. In other
words, food expenditure for all households with the same income are
Assumption 4: The distribution of population errors for each x has the same
(constant) standard deviation, which is denoted by   . This assumption
indicates that the spread of points around the regression line is similar for all
x values.
103
Exercises
1. Plot the following straight lines. Give the values of the y-intercept and
slope for each of these lines and interpret them. Indicate whether each of the
lines gives a positive or negative relationships between x and y .
a) y  53  7 x ; b) y  75  6 x
2. The following information is obtained from a sample data
10 10 10 10
n  10 ; i 1
x i  100 ;  i 1
y i  220 ; 
i 1
xi y i  3680 ; x
i 1
2
i  1140
Find the estimated regression line.

3. Computing from a data set of (x, y) values we obtained the following
summary statistics
_ _ 14 _
n  14 ; x  3.5 ; y  5.1 ;  (x
i 1
i  x) 2  10.82 ;
14 _ _ 14 _

i 1
( xi  x)( y i  y )  2.677 ; 
i 1
( y i  y ) 2  2.01
Obtain the equation of the estimated regression line.

4. Given the five pairs of ( x, y ) values,
x 0 1 6 3 5
y 4 3 0 2 1
a) Construct a scatter diagram

b) Calculate the least squares estimates a and b.
c) Determine the fitted line and draw the line on the scatter diagram.
5. A researcher took a sample of 36 electronic companies and found the
following relationship between x and y where x is the amount of money (in
thousands of dollars) spent on advertising by a company during a year and y
represents the total gross sales (in thousands of dollars) of that company in
that year.
^
y  5.6  22.5  x
a) An electronic company spent 2000$ on advertising during a year. What
are its expected gross sales for that year?
104
b) Suppose five electronic companies spent 2000$ each on advertising
during that year. Do you expect these five companies to have the same actual
gross sales for that year? Explain.
6. An economist wanted to determine whether or not the amount of phone
bills and income of households are related. The following table gives
information on the monthly incomes (in hundreds of dollars) and monthly
telephone bills (in dollars) for a random sample of 10 households
Income 16 45 36 32 30 13 41 15 36 40
Phone bill 35 78 102 56 75 26 130 42 59 85
a) Find the regression line with income as an independent variable and the
amount of the phone bill as a dependent variable.
b) Give an interpretation of the values of a and b calculated in part a.
c) Estimate the amount of the monthly phone bill for a household with a
monthly income of $2500.
7. An auto manufacturing company wanted to investigate how the price of
one of its car models depreciates with age. The research department at the
company took a sample of 9 cars of this model and collected the following
information on the ages (in years) and prices (in hundreds of dollars) of these
cars.
Age 8 3 7 10 3 5 6 9
Price 16 74 38 21 98 56 49 30
a) Construct a scatter diagram for these data. Interpret your results.

b) Find the regression line with price as a dependent variable and age as an
independent variable.
c) Give a brief interpretation of the values of a and b calculated in part b.
d) Predict the price of a 4 year-old car of this model.
e) Estimate the price of a 19-year-old car of this model. Comment on this
finding.
8. Construct a scatter diagram for the data in the following table
x 0.5 1 1.5
y 2 1 3
a) Plot the following two lines on your scatter diagram
1) y  3  x and 2) y 1  x
105
b) Which of these lines would you choose to characterize the relationship
between x and y? Explain
c) Show that the sum of errors for both of these lines equals 0.
d) Which of these lines has smaller SSE ?
e) Find the least squares regression line for the data and compare it to two
lines described in part a.
Answers
^ ^ ^
2. y  83.714  10.571  x ;3. y  4.225  0.247  x ;4.c) y  3.845  0.615  x ;
^
5. a)$50.6 thousand; b)different amounts;6.a) y  2.3173  2.1869  x ;
^
c) $56.99; 7. y  111  9.84  x ; 8. b) The second line; d) The second line;
^
e) y 1  x .
3.6. The explanatory power of a linear regression equation
In Figure 3.3 it is shown that the deviation of an individual y value from its
mean can be
^
y  a  bx
_ ^
y SST yi  y ei  y  y i SSE
^ _
y y SSR
_
y
_
x xi
Fig.3.3
106
partitioned into deviation of the predicted value from the mean and the
deviation of the observed value from the predicted value
_ ^ ^ _
yi  y  ( yi  y)  ( y i  y)
We square each side of the equation-because the sum of deviations about the
mean is equal to zero-and sum the results over all n points
n _ n n _
(y (y  ( y  y)
^ ^
i  y) 2  i  y) 2  i
2
i 1 i 1 i 1
Some of you may note the squaring of the right- hand side should include the
cross product of the two terms in addition to their squared quantities. It can
be shown that the cross predicted term goes to zero. This equation is
expressed as
SST  SSR  SSE
We see that the total variability-SST- consists of two components-SSR-the
amount of variability explained by the regression equation- named
“Regression Sum of Squares” and –SSE-random or unexplained deviation of
points from the regression line-named “Error Sum of Squares”. Thus
n _
Total sum of squares: SST  (y
i 1
i  y) 2
n _ n _
  (x
^
Regression Sum of Squares: SSR  ( y  y) 2  b 2  i  x) 2
i 1 i 1
n n n
   (e )
^
Error Sum of Squares: SSE  ( yi  y) 2  ( y i  (a  bx i )) 2  i
2
i 1 i 1 i 1
For a given set of observed values of the dependent variables, y, the SST is
fixed as the total variability of all observations from the mean. We see that in
the partitioning larger values of SSR and hence smaller value of SSE indicate
a regression equation that “fits” or comes closer to the observed data. This
partitioning is shown graphically in Figure 3.3.
Example:
Let us find SST, SSR and SSE for the data on incomes and food expenditure.
Using calculation given in the table 3.3 we find the value of total sum of
squares as
107
2
 7 
n 7

 y i 
2
y i2     646  64  60.8571
_
SST  (y
i 1
i  y)  2

i 1
i 1
n 7
Table3.4
x y ^
y2 ei _ _
ei
2
y xi  x ( xi  x) 2
35 9 10.3884 81 -1.3884 4.7143 22.2246 1.9277

49 15 14.0872 225 0.9128 18.7143 350.225 0.8332
21 7 6.6896 49 0.3104 -9.2857 86.2242 0.0963
39 11 11.4452 121 -0.4452 8.7143 75.9390 0.1982
15 5 5.1044 25 -0.1044 -15.286 233.653 0.0109
28 8 8.5390 64 -0.5390 -2.2857 5.2244 0.2905
25 9 7.7464 81 1.2536 -5.2857 27.9386 1.5715
646 801.429 4.9283
The error sum of squares SSE is given in the sum of the eights column in
Table 3.4. Thus,
n n
  (e )
^
SSE  ( yi  y) 2  i
2
 4.9283
i 1 i 1
The regression sum of squares can be found from SST  SSR  SSE .
Thus
SSR  SST  SSE  60.8571  4.9283  55.9288 .
The value of SSR can also be computed by using the formula.(Check!!)
n _ n _
 
^
SSR  ( y  y) 2  b 2  ( xi  x) 2 .
i 1 i 1
The total sum of squares SST is a measure of the total variation in food
expenditures, SSR is the portion of total variation explained by the regression
model (or by income), and the error sum of squares SSE is the portion of
total variation not explained by the regression model.
108
3.6.1. Coefficient of determination R 2
If we divide both side of the equation

SST  SSR  SSE
by SST, we obtain
SSR SSE
1 
SST SST
We have seen that the fit of the regression equation to the data is improved
SSR
as SSR increases and SSE decreases. The ratio provides a descriptive
SST
measure of the proportion or percent of the total variability that is explained
by the regression model. This measure is called the coefficient of
determination-or more generally R 2 .
SSR SSE
R2  1
SST SST
The coefficient of determination is often interpreted as the percent of
variability in y that is explained by the regression equation. We see that
R 2 increases directly with the spread of the independent variable.
R 2 can vary from 0 to 1 since SST is fixed and 0  SSE SST . A larger
R 2 implies a better regression, everything else being equal.
 
Interpretation of R 2 : About 100  R 2 % of the sample variation in y
(measured by the total sum of squares of deviations of the sample y values
_
about their mean y ) can be explained by using x to predict y in the straight
line model.
Example:
Calculate the coefficient of determination for the data on monthly incomes
and food expenditures of seven households.
Solution:
From earlier calculations
SSR  55.9288 and SST  60.8571
Hence,
SSR 55.9288
R2    0.92
SST 60.8571
109
We can state that 92% of the variability in y is explained by linear
regression, and the linear model seems very satisfactory in this respect. In
other words, we can state that 92% of the total variation in food expenditures
of households occurs because of the variation in their incomes, and the
remaining 8% is due to other variables, like differences in size of the
household, preferences and tastes and so on.
3.6.2. Estimation of model error variance
When we consider income and food expenditures, all households with the
same income are expected to spend different amounts on food.
Consequently, the random error  i will have different values for these
^ 2
households. The variance  i measures the spread of these errors around the
^ 2
population regression line. Note that  i denotes the variance of errors for
^ 2
the population. However, usually  i is unknown. In such cases, it is
estimated by s e2 , which is the standard deviation of errors for the sample
data.
An estimator for the variance of the population model error is
n
^ 2 e SSE
i 1
2
i
 e  s e2  
n2 n2
Division by ( n  2 ) instead of (n  1) results because the simple regression
model uses two estimated parameters, a and b , instead of one.
The formula for SSE is
n n
 (y
^
SSE  (ei ) 2  i  y) 2
i 1 i 1
If we introduce the following notations
n
n _ n
( x ) i
2
SS xx   (x
i 1
i  x) 2  x
i 1
2
i  i 1
n
110
n
n _ n
( y ) i
2
SS yy  (y
i 1
i  y) 2  y
i 1
2
i  i 1
n
n n
n _ _ n
(  xi )( y ) i
SS xy  i 1
( xi  x)( y i  y )  
i 1
xi y i  i 1
n
i 1
where SS stands for “sums of squares”, then

2
SS xy
SSE  SS yy  .
SS xx
Exercises
1. The following information is obtained from a sample data set

12 12 12
n  12 ; 
i 1
xi  66 ;  i 1
y i  588 ; x y
i 1
i i  2244 ;
12 12

i 1
xi2  396 ; and yi 1
2
i  58734
Find the values of s e2 and R 2 .

2. A sample data set produced the following information
460 460 460
n  460 ; 
i 1
x i  3920 ;  i 1
y i  2650 ; x y
i 1
i i  26570 ;
460 460
x
i 1
2
i  48530 ; and yi 1
2
i  39347
Find the values of s e2 and R 2 .

3. Computing from a data set of (x, y) values produced the following
summary statistics
_ _
n  14 ; x  3.5 ; y  2.32 ;
SS xx  10.82 ; SS xy  2.677 ; SS yy  1.035
a) Obtain equation of the best fitting straight line.
^ 2
b) Estimate  e .
111
4. Computing from a data set of (x, y) values produced the following
summary statistics
_ _
n  14 ; x  1.2 ; y  5.1 ;
SS xx  14.10 ; SS xy  2.31 ; SS yy  2.01
Determine the proportion of variation in y that is explained by linear
regression.
5. A calculation shows that SS xx  10.1 , SS yy  16.5 , and SS xy  9.3 ,
determine the proportion of variation in y that is explained by linear
regression.
6. The following table lists the sizes of offices (in hundreds of square meters)
and the rents (in dollars) paid for those offices.
Size of offices 22 17 19 28 35 24
Monthly rent 710 590 730 880 1080 820
^
a) Find the regression line y  a  bx with the size of an office as an
independent variable and monthly rent as a dependent variable.
b) Give a brief interpretation of the values of a and b.
c) Predict the monthly rent for the office with 2400 square meters.
d) One of the offices is 2600 square meters and its rent is $850. What is the
predicted rent for this office? Find the error for this office.
e) Compute the standard deviation of errors.
f) Calculate the coefficient of determination. What percentage of the
variation in monthly rents explained by the sizes of the offices? What
percentage of this variation is not explained?
7. Refer to exercise 7 of previous chapter. The following table which gives
the ages (in years) and prices (in hundred of dollars) of eight cars of specific
model, is reproduced from that exercise.
Age 8 3 7 10 3 5 6 9
Price 16 74 38 21 98 56 49 30
a) Calculate the standard deviation of errors.

b) Compute the coefficient of determination and give a brief interpretation of
it.
112
Answers
^
1. 22.2; 0.99; 2.50.06; 0.04; 3. a) y  1.454  0.247  x ; b) s e2  0.031 ;
^
4. 0.188; 5. 0.5190; 6. a) y  194  25.1  x e)40.2; f) 0.953;
; 7. a) 11.39; b) 0.856
3.7. Statistical inference: Hypothesis tests and confidence intervals
One of the main purposes for determining a regression line is to find the true
value of the slope  of the population regression line. However, in almost
all cases, the regression line is estimated using sample data. Then based on
the sample regression line, inferences are made about the population
regression line. The slope b of a sample regression line is a point estimator
of the slope  of the population regression line. The different sample
regression lines estimated for different samples taken from the same
population will give different values of b. If only one sample is selected,
then the value of b will depend on which elements are included in the
sample. Thus, b is a random variable and it possesses a probability
distribution called a sampling distribution.
Assume that assumptions 3.5.4 are hold. Then b is an unbiased estimator of
 and has a population variance
 2  2
 b2  n _
 n _
i 1
( xi  x) 2 x
i 1
2
i  n  ( x) 2
and unbiased estimator of  is provided by

b
2
2
s s e2 s e2
sb2  n
e
 n

_ _ SS xx
 (x
i 1
i  x) 2 x
i 1
2
i  n  ( x) 2
In applied regression analysis we first would like to know if there is a

relationship. We see that if  is zero then there is no relationship- y would
not continuously increase or decrease with increase in x .
113
3.7.1. Hypothesis testing about 
Let  be a population regression slope and b its least square estimate based
on n pairs of sample observations. Assume that assumptions 3.5.4 hold and
also assume that the errors  i are normally distributed. Then the random
variable
b
t
sb
is distributed as Student’s t distribution with (n  2) degree of freedom.
If we use notation
b
T .S .  t 
sb
for the test statistic then the following tests have a significance level 
H 0 :    0 or H 0 :    0
H1 :    0
Reject H 0 if T .S .  t n  2,
H 0 :    0 or H 0 :    0
H1 :    0
Reject H 0 if T .S .  t n  2,
3. To test null hypothesis
H 0 :   0
H1 :    0
Reject H 0 if T .S .  t n 2, / 2 or T .S .  t n  2, / 2
114
Remark1: To test the hypothesis that x does not determine y linearly and
there is no linear relationship, we will test the null hypothesis that the slope
of the regression line is zero, that is H 0 :    0  0 ; the alternative
hypothesis that H 1 :    0  0 means x determines y linearly;
H 1 :    0  0 means x determines y positively; H 1 :    0  0 means x
determines y negatively.
Remark2: The null hypothesis does not always have to be   0 .We may
test the null hypothesis that  is equal to a value different from zero.
Example:
Test at the 5% significance level if the slope of the population regression line
for the example on incomes and food expenditure of seven households is
positive.
Solution:
From earlier calculations we have
n7; b  0.2642 and s e  0.9922
s e2 0.9856
sb2  n
  0.001229 ; and sb  0.0350 .
_ 801.429
 (x
i 1
i  x) 2
We are to test whether or not slope  of the population regression line is

positive. The two hypotheses are
H 0 :    0 (Slope is zero)
H 1 :    0 (Slope is positive)
reject H 0 if T .S .  t n  2, .
b   0.2642  0
T .S .  t    7.549
sb 0.0350
The significance level is 0.05. Therefore,
t n2,  t 5,0.05  2.015
The value of the test statistic T.S.  7.549 is greater than the critical value of
t  2.015 and it falls in the rejection region. Hence, we reject the null
hypothesis and conclude that x (income) determines y (food expenditure)
115
positively. That is, food expenditure increases with an increase in income
and it decreases with a decrease in income.
3.7.2. Confidence intervals for the population regression slope 
We can derive confidence intervals for the slope  of the population

regression line by using coefficient b and variance estimators we have
developed.
If the assumptions 3.5.4 hold, and if the regression errors,  i , are normally
distributed, then 100(1   )% confidence interval for the population
regression slope  is given by
b  t n 2, / 2  sb    b  t n  2, / 2  sb
where t n 2, / 2 is the number for which P(t n 2  t n 2, / 2 )   / 2 and the
random variable t n2 follows Student’s t distribution with (n  2) degrees of
freedom.
Example:
Construct a 95% confidence interval for  for the data on incomes and food
expenditures of seven households.
Solution:
From earlier calculations we have
n  7 ; b  0.2642 ; s e  0.9922 and sb  0.0350
The confidence level is 95%. So
100(1   )%  95%
1    0.95
  0.05
 / 2  0.025
t n 2, / 2  t 5,0.025  2.571
The 95% confidence interval for  is
b  t n 2, / 2  sb    b  t n  2, / 2  sb
0.2642  2.571  0.0350    0.2642  2.571  0.0350
0.17    0.35
Thus, we are 95% confident that slope  for the population regression line is
between 0.17 and 0.35.
116
Exercises
1. The following information is obtained for a sample of 16 observations
taken from a population
^
SS xx  340.700 ; s e  1.951 ; and y  12.45  6.32  x
a) Make a 99% confidence interval for  .
b) Using a significance level of 0.025, test the null hypothesis that  is zero
against the alternative that  is positive.
c) Using a significance level of 0.01, can you conclude that  is zero against
the alternative that it is different from zero?
d) Using a significance level of 0.02, test whether  is different from 4.50.
2. The following information is obtained for a sample of 100 observations
taken from a population. (Note that because n  30 , we can use the normal
distribution to make a confidence interval and test a hypothesis about  )
^
SS xx  524.884 ; s e  1.464 ; and y  5.48  2.50  x
a) Make a 98% confidence interval for 
b) Test at the 2% significance level whether  is zero against the alternative
that it is positive.
c) Can you conclude that  is zero? Use   0.01 .
d) Using a significance level of 0.01, test whether  is 1.75 against the
alternative that it is greater than 1.75.
3. Refer to exercise 7 of previous chapter. The following table which gives
the ages (in years) and prices (in hundred of dollars) of eight cars of specific
model, is reproduced from that exercise.
Age 8 3 7 10 3 5 6 9
Price 16 74 38 21 98 56 49 30
a) Construct a 95% confidence interval for  .

b) Test at the 5% significance level if  is zero against the alternative that it
is negative.
117
4. The following table gives the experience (in years) and monthly salaries
(in thousands of tenge) of nine randomly selected secretaries
Experience 14 3 5 6 4 9 18 5 16
Monthly salary 22 12 15 17 15 19 24 13 27
a) Find the least squares regression line with experience as an independent

and monthly salary as dependent variables.
b) Construct a 95% confidence interval for  .
c) Test at the 2.5% significance level if  is zero against the alternative that
it is positive.
5. The data on the size of six offices (in hundreds of square meters) and the
monthly rents (in dollars) paid by firms for those offices are reproduced
below from exercise 6 of the previous section.
Size of offices 22 17 19 28 35 24
Monthly rent 710 590 730 880 1080 820
a) Construct a 99% confidence interval for  . You can use the calculations
made in exercise 6 of previous section here.
b) Test at the 5% significance level the null hypothesis that  is zero against
the alternative that it is different from zero.
6. The following data give information on the ages (in years) and the number
of breakdowns during the past year for a sample of six machines at a large
company.
Age 9 14 18 15 10 11
Number of breakdowns 34 46 52 64 42 44
^
a) Find the least squares regression line y  a  b  x
b) Give a brief interpretation of the values a and b.
c) Compute and interpret R 2 .
d) Compute the standard deviation of errors.
e) Construct a 98% confidence interval for  .
f) Test at the 2.5% significance level the null hypothesis that  is zero
against the alternative that it is positive.
118
7. The following table gives information on the temperature in a city and
volume of the ice cream (in thousands) sold at the supermarket for a random
sample of eight days during the summer.
Temperature 22 16 28 12 30 19 24 32
Ice cream sold 3.64 3.12 4.08 2.84 3.98 3.55 4.02 4.38
^
a) Find the least squares regression line y  a  b  x . Take temperature as an
independent variable and volume of ice cream sold as a dependent variable.
b) Give a brief interpretation of the values a and b.
c) Compute and interpret R 2 .
d) Compute the standard deviation of errors.
e) Construct a 95% confidence interval for  .
f) Test at the 1% significance level the null hypothesis that  is zero against
the alternative that it is positive.
Answers
1.a) 6.01 to 6.63;b) T .S.  t  59.792; reject H 0 ;c) T .S.  t  59.792; reject H 0
d) T .S.  t  17.219; reject H 0 ;2.a)2.35 to 2.65;b) T .S.  z  39.12; reject H 0 ;
c) T .S.  z  39.12; reject H 0 ; d) T .S.  z  11.74; reject H 0 ; 3. a) -15.53 to -
^
7.17; b) T .S.  t  6.645; reject H 0 ;4. a) y  10.4986  0.8689  x ; b) 0.5559
to 1.1819; c) T .S.  t  8.323; reject H 0 ; 5. a) 12.34 to 37.92 ; b)
^
T .S.  t  4.604; reject H 0 6. a) y  1.4337  0.8916  x ; c) R 2  0.94 ;
d) s e  0.9285 ; e) 0.5708 to 1.2124; f) T .S.  t  9.356; reject H 0 ;
^
7. a) y  2.0680  0.0714  x ; c) R 2  0.92 ;d) s e  0.1537 ; e) 0.0511 to
0.0917; f) T.S.  t  8.602 reject H 0 .
119
3.8. Using the regression model for prediction a particular value of y
The second major use of a regression model is to predict a particular value of

y for a given value of x, say x 0 . For example, we may want to predict the
food expenditure of a randomly selected household with a monthly income
of $3000. In this case, we are not interested in the mean food expenditure of
all households with a monthly income of $3000 but in the food expenditure
of one particular household with a monthly income of $3000. This predicted
value of y is denoted by y p . To predict a single value of y for x  x 0 from
^
estimated sample regression line, we use the value of y as a point estimate of
^
y p . Using the estimated regression line, we find y for x =30 as
^
y  1.1414  0.2642  (30)  9.0674
Thus, based on our regression line, the point estimate for the food
expenditure of a given household with a monthly income of $3000 is
$906.74 per month.
Different regression lines estimated by using different samples of seven
households each taken from the same population will give different values of
the point estimator for the predicted value of y for x =30.Hence , a
confidence interval constructed for y p based on one sample will give a more
reliable estimate of y p than will a point estimate. The confidence interval
constructed for y p is more commonly called a prediction interval.
Suppose that the population regression model is
y i      xi   i (i  1,2,......... , n  1)
and that the standard regression assumptions hold, and that the  i are
normally distributed. Let a and b be the least squares estimates of  and  .
It can be shown that the following are 100(1   )% intervals:
1. For the forecast of the single value resulting for y n 1 at a given x n 1 , the
prediction interval is
120
 
 _
2 
^ 1 ( x  x) 
y n 1  t n  2, / 2  1   n n 1 s
 n _  e

 i 1

( xi  x) 
2

2. For the forecast of the conditional expectation, E ( y n1 / x n1 ) , the

confidence interval is
 
 _
2 
^ 1 ( x  x) 
y n 1  t n  2, / 2    n n 1 s
n _  e



i 1
( xi  x) 
2

where
n
_ x
i 1
i
^
x and y n 1  a  b  x n 1 .
n
Example:
For the data on incomes and food expenditures of seven households, find
a) 99% prediction interval for the predicted food expenditure for a single
household with a monthly income of $3500;
b) Obtain a 99% confidence interval for the expected food expenditure for all
households with a monthly income of $3000.
Solution:
a) The point estimate of the predicted food expenditure for x  35 is given by
^
y  1.1414  0.2642  (35)  10.3884
100(1   )%  99%
  0.01
 / 2  0.005
t n 2, / 2  t 5,0.005  4.032
Using data from the previous chapters
_
s e  0.9922 ; x  30.2857 ; and SS xx  801.4286
121
Hence, the 99% prediction interval for y p for x  35 is
 
 _
2 
^ 1 ( x  x) 
y n 1  t n  2, / 2  1   n n 1 s
 n _  e

 i 1
( xi  x) 
2

 1 (35  30.2857 ) 2 
10.3884  4.032  1     0.9922 
 7 801.4286 
 10.3884  4.3284  6.0600 to 14.7168
Thus, with 99% confidence we can state that the predicted food expenditure
of a household with a monthly income of $3500 is between $606.00 and
$1471.68.
b) Once again, the point estimate of the expected food expenditure for
x  35 is
^
y  1.1414  0.2642  (35)  10.3884
Hence, the 99% confidence interval for E ( y n1 / 35) is
 
 _
2 
^ 1 ( x  x) 
y n 1  t n  2, / 2    n n 1 s
n _  e



i 1
( xi  x) 
2

 1 (35  30.2857 ) 2 
10.3884  4.032      0.9922 
7 801.4286 
 10.3884  1.6523  8.7361 to 12.0407
Thus, with 99% confidence we can state that the mean food expenditure for
all households with monthly income of $3500 is between $873.61 and
$1204.07.
As we can observe, the interval in part a) 606.00 to 1471.68 is much wider
than the one for the mean value of y for x  35 calculated in part b)
873.61 to1204.04 . This is always true. The prediction interval for
predicting a single value of y is always larger than the confidence interval for
estimating the mean value of y for a certain value of x.
122
Exercises
1. Construct a 99% confidence interval for the mean value of y and a 99%
prediction interval for the predicted value of y for the following
^
a) y  3.25  80  x for x  15 given se  0.954 ;
_
x  18.52 ; SS xx  144.65 ; and n  10
^ _
b) y  27  7.67  x for x  12 given s e  2.46 ; x  13.43 ;
SS xx  369.77 ; and n  10
2. Refer to Exercise 4 of the previous section. Construct a 90% confidence
interval for the mean monthly salary of secretaries with 10 years of
experience. Construct a 90% prediction interval for the monthly salary of a
randomly selected secretary with 10 years of experience.
3. Refer to Exercise 6 of the previous section. Construct a 95% confidence
interval for the mean number of breakdowns for all cars which are 16 years
old. Determine a 95% prediction interval for y p for x  16 .
4. The following data give information on the lowest cost price (in dollars)
and the average attendance (thousand) for the past year for eight football
teams
Ticket price 3.6 3.3 2.8 2.6 2.7 2.9 2.0 2.6
Attendance 24 21 22 22 18 13 9 6
a) Taking ticket price as an independent variable and attendance as a

dependent variable, estimate the regression of attendance on the ticket price.
b) Interpret the slope of the estimated regression line.
c) Find and interpret the coefficient of determination.
d) Find and interpret a 90% confidence interval for the slope of the
population regression line.
e) Find a 90% confidence interval for expected number of attendance for
which the price of ticket is 20.
5. A sample of 25 employees at a production plant was taken. Each
employee was asked to assess his or her own job satisfaction (x), on scale
from 1 to 10. In addition, the number of days absent (y) from work during
the last year were found for these employees. The sample regression line
123
^
y  13.6  1.2  x
was estimated by least squares for these data. Also found that
_ 25 _
x  6.0 ;  (x
i 1
i  x) 2  130 ; SSE  80.6
a) Test at the 1% significance level against the appropriate one sided

alternative the null hypothesis that job satisfaction has no linear effect on
absence.
b) A particular employee has job satisfaction level 4. Find a 90% confidence
interval for the number of days this employee would be absent from work in
a year.
Answers
1. a) 13.871 to 16.629; b) 11.765 to 18.735; 2. 18.108 to 20.267; 15.838 to

^
22.537; 3. 4. a) y  2.029  0.0464  x ; c) R 2  0.4194 ; d)
0.003    0.0928 ; e) 2.6525 to 3.2615; 5. a) T.S.  t  7.303 ; reject H 0 ;
b) 5.4798 to 12.1202.
124
Chapter 4
Multiple regression analysis
4.1. Introduction
In Chapter 3 we showed how regression analysis could be used to

develop an equation that would estimate the relationship between two
variables. Recall that we limited our discussion to the development of a
linear relationship between the two variables, or what is commonly referred
to as simple linear regression. There are many important situations, however,
where the underlying relationship between two variables can not be
explained adequately with a straight-line relationship. In addition, the most
of real world problems require the consideration of more than one
independent variable in order to predict the dependent variable. In this
chapter we discuss how multiple regression analysis can be used to handle
such situations.
4.2. Multiple regression model
Usually a dependent variable is affected by more than one independent

variable. When we include two or more independent variables in a regression
model, it is called a multiple regression model.
A multiple regression model with y as a dependent variable and x1 , x2 as
independent variables is written as
y    1  x1   2  x2  
where the numbers  ,  1 , and  2 must be estimated from sample data.
More generally, a multiple regression model with y as a dependent variable
and x1 , x2 , x3 ,....... and x k as independent variables is written as
y    1  x1   2  x2   3  x3  ....   k  xk   (1)
where the numbers  represents the constant term and 1 ,  2 ,  3 ,..... k are
the regression coefficients of an independent variables x1 , x2 , x3 ,.......
and x k , respectively.
In (1) if each of the independent variables is set to 0, it follows that
E ( y / x1  0; x 2  0;...... x k  0)  
125
Thus,  is expected value of the dependent variable when every independent
variable takes value 0. Frequently this interpretation does not carry practical
interest and often leads to meaningless.
The interpretation of the coefficients 1 ,  2 ,  3 ,..... k is extremely
important. For example,  1 is expected increase in y resulting from 1 unit
increase in x1 when the values of the other independent variables remain
constant. In general,  i is expected increase in the dependent variable
resulting from a 1-unit increase in the independent variable x i when the
values of the other independent variables remain constant.
If model (1) is estimated using sample data, which is usually the case, the
estimated regression model is written as
^
y  a  b1  x1  b2  x2  b3  x3  ......  bk  xk (2)
In model (2) a, b1 , b2 , b3 ,......and bk are the sample statistics, which are the
point estimators of  , 1 ,  2 ,  3 ,.....and  k , respectively.
In model (1) y denotes the actual values of the dependent variable. In model
^
(2), y denote the predicted or estimated values of the dependent variable.
^
The difference between y and y gives the error of prediction.
The method of fitting multiple regression of least squares model is similar to
that of fitting the linear regression model: method of least squares. That is,
we choose the estimated model
^
y  a  b1  x1  b2  x2  b3  x3  ......  bk  xk
that minimizes
 ( y  y)
^
SSE  2
.
126
4.3. Standard assumptions for the multiple regression models
Like the simple linear regression model, the multiple regression model is
also based on certain assumptions.
Consider the multiple regression model
y    1  x1   2  x2   3  x3  ....   k  xk  
The following assumptions are often made:
Assumption 1: For any given set of values of x1 , x2 , x3 ,....... and x k , the
random error  has a normal probability distribution with mean equal to 0
and variance equal to  2 .
Assumption 2: The errors associated with different sets of values of
independent variables are independent.
Assumption 3: The independent variables are not linearly related. If any of
them is linearly related, then we can eliminate one of the variables by
making substitution and reduce the number of independent variables.
Assumption 4: It is not possible to find a set of numbers c0 , c1 , c 2 ,........c k ,
such that
c0  c1  x1  c2  x2  c3  x3  ....  ck  xk  0
4.4. The explanatory power of a multiple regression equation
4.4.1. Estimation of error variance
The variance of errors (also called the variance of the estimate) for the
multiple regression model
y    1  x1   2  x2   3  x3  ....   k  xk  
is denoted by  e2 . However, when sample data are used to estimate
multiple regression model (1), the variance of errors, denoted by s e2 , is an
unbiased estimate of the  e2 . The formula for calculating s e2 is as follows
n
e
i 1
2
i
SSE
s e2  
n  K 1 n  K 1
where
n-is the sample size
127
K-is the number of independent variables included in the model.
The positive square root of the variance s e is also called the standard error
of the estimate.
4.4.2 The coefficient of determination
Multiple regression uses independent variables to explain the behaviour of

the dependent variable. Part of the variability in the dependent variable can
be explained by its linear association with the independent variables. We
will develop a measure of the proportion of the variability in the dependent
variable that can be explained by the multiple regression.
Let the multiple regression model fitted by least squares be
^
y i  a  b1  x1i  b2  x 2i  b3  x3i  ......  bk  x ki  ei  y i  ei
where a, b1 , b2 , b3 ,......and bk are the least squares estimates of the population
regression model and ei ’s are the residuals from the estimated regression
model.
The model variability can be partitioned into the components
SST  SSR  SSE
where
n _
Total sum of squares: SST  
i 1
( yi  y) 2
n n
(y e
^
Error sum of squares: SSE  i  yi )2  2
i
i 1 i 1
n _
 ( y  y)
^
Regression sum of squares: SSR  i
2
i 1
The coefficient of determination for a multiple regression model, usually
called the coefficient of determination, is denoted by R 2 and is defined as
the proportion of the total sample sum of squares SST that is explained by
the multiple regression model.
SSR SSE
R2  1
SST SST
It tells us how good the multiple regression model is and how well the
independent variables included in the model explain the dependent variable.
128
The value of the coefficient of determination R 2 always lies in the range 0 to
1, that is
0  R2 1
4.4.3 Adjusted coefficient of determination
There is a potential problem with using R 2 as an overall measure of the

quality of a fitted equation. The value of R 2 generally increases as we add
more and more explanatory variables to the regression model. Therefore, by
adding a large number of variables to our regression model (even if they are
not included in the model) we can make the value of R 2 very close to 1.
Such a value of R 2 will be misleading, and it will not represent the true
explanatory power of the regression model. To eliminate this shortcoming
of R 2 , it is preferable to use the adjusted coefficient of determination,
_ 2 _ 2
which is denoted by R . The value of R may increase, decrease, or
stay the same as we add more explanatory variables to our regression
model. If a new variable added to the regression model contributes
_ 2
significantly to explain the variation in y, then R increases; otherwise it
_ 2
decreases. The value of R is calculated as follows
_ 2
n 1
R  1  (1  R 2 )  or
n  K 1
_ 2
SSE /( n  K  1)
R 1
SST /( n  1)
129
4.4.4 Predictions from the multiple regression models
Given that the population regression model

y    1  x1   2  x2   3  x3  ....   k  xk  
holds and assume that the standard regression assumptions are valid. Let
a, b1 , b2 ,.......bk be the least squares estimates of the model coefficients
 , 1 ,  2 ,...... k , correspondingly. Then given a new observation of a data
^
point, x1,n 1 ; x 2,n 1 ;…. x k ,n 1 the best linear unbiased forecast of y n 1 is
^
y n 1  a  b1  x1,n 1  b2  x 2,n 1  ......  bk  x k ,n 1
Remark: It is very risky to calculate forecasts that are based on x i values
outside the range of the data used to estimate the model coefficients, because
we have not included these points to the linear model.
Exercises
1. The regression model yi    1  x1i   2  x2i   i was fitted to a data
set obtained from 20 runs of an experiment in which two predictors x1i and
x 2i were observed along with the response y i . The least squares estimates
were
a  4.21 ; b1  11.37 ; b2  0.513
Predict the response for
a) x1  8 ; x 2  30
b) x1  8 ; x 2  50
2. The following model was fitted to a sample of 25 families in order to
explain household milk consumption
yi    1  x1i   2  x2i   i
where
y i -milk consumption, in liters per week
x1i -weekly income, in hundreds of dollars
x 2i -family size
The least squares estimates of the regression parameters were
a  0.30 ; b1  2.32 ; b2  1.41
a) Interpret the estimates b1 and b 2
130
b) Is it possible to provide a meaningful interpretation of the estimate a ?
3. The following model was fitted to a sample of 20 students using data
obtained at the end of the education year. The aim was to explain students’
weight gains.
yi    1  x1i   2  x2i   3  x3i   i
where
y i -weight gained, in kilograms, during the academic year
x1i -average number of meals eaten per week
x 2i -average number of exercise per week, ( in hours)
x 3i -average number of beers consumed per week
The least squares estimates of the regression parameters were
a  12.9 ; b1  4.5 ; b2  6.3 ; b3  3.14
a) Interpret the estimates b1 , b2 and b 3
b) Is it possible to provide a meaningful interpretation of the estimate a ?
4. In the study of exercise 2, where the least squares estimates were based on
25 sets of sample observations, the following data were found
SST  160.6 and SSR  80.3
a) Find and interpret the coefficient of determination.
b) Find the adjusted coefficient of determination.
5. In the study of exercise 3, sample of 20 observations were used to
calculate the least squares estimate. The regression sum of squares and error
sum of squares were found to be
SST  82.6 and SSE  49..3
a) Find and interpret the coefficient of determination.
b) Find the adjusted coefficient of determination.
6. A multiple linear regression was fitted to a data set obtained from 27 runs
of an experiment, in which four predictors x1 , x 2 , x3 , and x 4 were observed
along with the response y. The following results were obtained:
a  5.46; b1  2.35 ; b2  18.4 ; b3  0.91 ; b4  6.2 ;
SSR  920.60 ; SSE  78.92
a) Predict response for x1  14; x 2  0.6; x3  5; x 4  5.2
b) Estimate the error standard deviation 
c) What proportion of the y variability I explained by the fitted regression?
131
Answers
1. a) 79.78; b) 69.52; 4. a) 0.5; b) 0.45; 5. a) 0.4; b) 0.52; 6. a) 66.17;
b) 1.89; c) 0.92.
4.5 Computer solution of multiple regressions
Usually the calculations for a multiple regression model are made by using
statistical software package for computers, such as MINITAB, instead of
using the formula manually. In this chapter we will analyze the multiple
regression models using MINITAB statistical software. The solutions
obtained using other packages can be interpreted the same way.
Remark:
To use MINITAB menu follow the following instructions
1. Select Stat>Regression
2. Select Response column
3. Select Predictors columns
4. Click OK.
Example1:
Suppose that we want to find the effect of driving experience and the number
of driving violations on auto insurance premiums. A random sample of 10
drivers insured with a company and having similar auto insurance policies
was selected. Table 4.1 lists the yearly auto insurance premiums (in dollars)
paid by these drivers, y, their driving experience ( x1 , in years), and the
number of driving violations that each of them has committed during the past
five years.
Table 4.1
y x1 x2
74 5 2
50 6 1
97 4 6
57 11 3
99 3 1
35 19 0
40 15 1
49 13 2
101 2 8
42 10 3
132
Use a computer package to perform a regression analysis using model
yi    1  x1i   2  x2i   i
and answer the following questions:
a) Write the estimated regression equation;
b) Interpret the meaning of the estimated regression coefficients;
c) What are the values of the variance and standard deviation of errors, the
coefficient of determination, and the adjusted coefficient of determination?
d) What is the predicted auto insurance premium paid per month by a driver
with seven years of experience and four driving violations?
Solution:
Using MINITAB, we first enter the data of y, x1 , x 2 in three different
columns and then use the regression command. The computer executes a
multiple regression analysis. We focus our attention on the principal aspects
of the output as shown in Figure 4.1
Figure 4.1
The regression equation is
Y= 87.9 - 3.39 X1 + 2.33 X2
Predictor Coef St. dev. T P
Constant 87.92 13.96 6.30 0.000
X1 -3.3869 0.9930 -3.41 0.011
X2 2.327 2.264 1.03 0.338
S = 13.77 R-SQ = 78.4% R-SQ(adj) = 72.3%
Analysis of Variance
Source DF SS MS F P
Regression 2 4824.5 2412.3 12.72 0.005
Residual Error 7 1327.9 189.7
Total 9 6152.4
Source DF SEQ. SS
X1 1 4624.1
X2 1 200.4
We now proceed to interpret the results in Figure 4.1 and use them to make
further statistical inferences.
a) The equation of the fitted linear regression is
133
^
y  87.9  3.39  x1  2.33  x 2
From this equation,
a  87.9 ; b1  3.39 ; b2  2.33
We can also read the values of these coefficients from the column labeled
COEF in the MINITAB solution of Figure 4.1.
Notice that in this column the coefficients appear with more digits after the
decimal point. With these coefficient values, we can write the estimated
regression equation as
^
y  87.92  3.3869  x1  2.327  x 2
b) The value of a  87.92 is in the estimated regression equation gives the
^
value of y for x1  0 and x 2  0 . It means that a driver with no experience
and no driving violations is expected to pay an auto insurance premium of
$87.92 per year. This is the technical interpretation of a.
The value of b1  3.3869 in the estimated regression model gives the
^
change in y for a one unit change in x1 when x 2 is held constant. Thus, we
can state that a driver with one extra year of experience but with the same
number of violations is expected to pay $3.3869 less for the auto insurance
premium per year.
^
The value of b2  2.327 in the estimated regression model gives change in y
for a one unit change in x 2 when x1 is held constant. Thus, we can state that
a driver with one extra driving violation but with the same years of driving
experience is expected to pay $82.327 more per year for the auto insurance
premium.
SSE 1327 .9
b)  e2 is estimated by s e2    189.7 , so s e  13.77 .
n  K 1 7
The values of the standard deviation of errors, the coefficient of
determination and the adjusted coefficient of determination are also given in
MINITAB solution. From Figure 4.1 we obtain
_ 2
s  s e  13.77 ; R-SQ  R 2  78.4% ; R-SQ (adj)  R =72.3%
The value of R 2  78.4% tells us that the two independent variables included
in our model explain 78.4% of the variation in the dependent variable.
134
_ 2
The value of R  72.3% is the value of the coefficient of determination
adjusted for degrees of freedom. It states that when adjusted for degrees of
freedom, the two independent variables explain 72.3% of the variation in the
dependent variable.
c) To predict auto premium paid per year by a driver with seven years of
experience and four driving violations, we substitute x1  7 and x 2  4 in the
estimated regression model
^
y  87.92  3.3869  x1  2.327  x 2  87.92  3.3869  7  2.327  4  73.5197
^
Note that this value of y is a point estimate of the predicted value of y,
which is denoted by y p .
Remark:
In figure 4.1 there is portion of solution in the end reproduced below,
Source DF SEQ. SS
X1 1 4624.1
X2 1 200.4
which we have not used in any of the examples. From figure 4.1 we have
SSR  4824.5
If we estimate the simple linear regression of y on x1 ,
y    1  x1  
the value of SSR will be 4624.1, which is the value in the row of X1 and the
column labeled SEQ.SS. That is, x1 alone will reduce SST by 4624.1
Then, if we add x 2 to model above, the SST will further be reduced by 200.4,
which is the value in the row of X2 and the column labeled SEQ.SS.
The sum of the two numbers in the column of SEQ.SS is
4624.1  200.4  4824.5
which is the value of SSR in the Figure 4.1.
Example2:
We are interested in studying the blood pressure y of males in relation to
weight x1 and age x 2 . Sample of 10 male was selected. The data set listed
below:
135
y x1 x2
120 76 60
160 84 45
134 95 37
149 99 46
153 74 49
164 83 70
130 92 38
170 110 54
148 80 28
125 79 19
Use a computer package to perform a regression analysis using model
yi    1  x1i   2  x2i   i
Solution:
Using MINITAB, we first enter the data of y, x1 , x 2 in three different
columns and then use the regression command. The computer executes a
multiple regression analysis. We focus our attention on the principal aspects
of the output as shown in Table 4.2
Table 4.2
Y = 81.2 + 0.493 X1 + 0.474 X2

Constant 81.17 43.50 1.87 0.104
X1 0.4929 0.4749 1.04 0.334
X2 0.4741 0.3641 1.30 0.234
S = 16.32 R-SQ = 30.2%

Source DF SS MS F P
Regression 2 805.2 402.6 1.51 0.285
Total 9 2670.1
Source DF SEQ. SS
X1 1 353.4
X2 1 451.8
136
We now proceed to interpret the results in table 4.2 and use them to make
further statistical inferences.
a) The equation of the fitted linear regression is
^
y  81.2  0.493  x1  0.474  x2
This means that the mean blood pressure increases by 0.493 if weight x1
increases by 1 kilogram and age x 2 remains fixed.
Similarly, a 1-year increase in age with the weight held fixed will increase
the mean blood pressure by 0.474.
b) The estimated regression coefficients and the corresponding estimated
standard errors are
a  81.17 estimated standard error S.E.(a)  43.50
b1  0.4929 estimated standard error S.E.(b1 )  0.4749
b2  0.4741 estimated standard error S.E.(b2 )  0.3641
Further, the error standard deviation  estimated by s  16.32 with
degrees of freedom  n  (number of variables ) - 1  10 - 2 - 1  7 .
These results are useful in interval estimation and hypothesis tests about the
regression coefficients.
c) In Table 4.2, the result " R  SQ  30.2%" or R 2  0.302 tells us that
30.2% of the variability of y is explained by the fitted multiple regression of
y on x1 and x 2 . The analysis of variance shows the decomposition of the
_
total variability  ( y  y) 2
 2670 .1 into the two components
2670 .1  805 .2  1864.9

Total variablit y Variabilit y explained Residual or
of y by the regression unexplaine d variabili ty
Thus,
805.2
R2   0.302
2670.1
1864.9
and  2 is estimated by s 2   266.41 , so s  16.32 .
7
137
4.6. Confidence interval for individual coefficients
The values of a, b1 , b2 ,......and bk obtained by estimating model (1) using

sample data give the point estimates of  , 1 ,  2 ,.......and  k , respectively,
which are the population parameters. Using the values of the sample
statistics a, b1 , b2 ,......and bk , we can make 100  (1   )% confidence intervals
for the corresponding population parameters  , 1 ,  2 ,.......and  k ,
respectively.
If the population errors,  i , are normally distributed, then
100  (1   )% confidence interval for the regression coefficients  i , are
given by
bi  t n  K 1, / 2  sbi   i  bi  t n  K 1, / 2  sbi
where t n  K 1, / 2 is the number for which

P(t n K 1  t nk 1, / 2 ) 
2
and the random variable t n K 1 follows a Student’s t distribution with
(n  K 1) degrees of freedom.
Example:
Determine a 90% confidence interval for  1 (the coefficient of experience)
for the multiple regression of auto insurance premium on driving experience
and the number of driving violations. Use the MINITAB solution of Fig. 4.1.
Solution:
The portion of the solution is shown below

Constant 87.92 13.96 6.30 0.000
X1 -3.3869 0.9930 -3.41 0.011
X2 2.327 2.264 1.03 0.338
From the given information we obtain

n  10 ; b1  3.3869 ; and s b1  0.9930
The confidence level is 90%. So,
t n  K 1, / 2  t1021,0.05  t 7,0.05  1.895
138
The 90% confidence interval for  1 is
b1  t n  K 1, / 2  s b1  1  b1  t n  K 1, / 2  s b1
 3.3869  1.895  0.9930  1  3.3869  1.895  0.9930
 5.269  1  1.505
Thus, the 90% confidence interval for  1 is  5.269 to  1.505 . That is, we
can state with 90% confidence that for one extra year of driving experience,
the yearly auto insurance premium decreases by an amount between $1.505
and $5.269.
Exercises
1. In the study of exercise 2 of the previous chapter, where the sample

regression was based on 25 observations, the estimated standard errors were
s b1  0.089 ; s b2  0.45 ;
a) Find 90% and 95% confidence intervals for  1
b) Find 95% and 99% confidence intervals for  2
2. In the study of exercise 3 of the previous chapter, where the sample

regression was based on 20 observations, the estimated standard errors were
s b1  0.18 ; s b2  0.64 ;
Find 90%, 95% and 99% confidence intervals for  1
3. Following is the MINITAB solution for a regression of y on x1 and x 2 .
Y = 12.4 + 0.24 X1 + 0.036 X2

Constant 12.410 5.234 2.12 0.060
X1 0.2415 0.0345 7.47 0.000
X2 0.0362 0.024 -1.60 0.140
S = 1.76 R-SQ = 97.8% R-SQ(adj) = 96.6%
139
Source DF SS MS F P
Regression 2 849.65 424.83 130.10 0.000
Total 12 882.30
Source DF SEQ. SS
X1 1 841.25
X2 1 8.40
Using the MINITAB solution, answer the following questions for the
population regression model y    1  x1   2  x2  
a) Write the estimated regression equation.
b) Write the values of a , b1 , b 2 and explain the meaning of these estimated
regression coefficients.
c) Write the values of the standard deviation of the coefficients of a , b1 , b 2 .
d) What are the values of the variance and standard deviations of errors, the
coefficient of determination, the adjusted coefficient of determination, SST,
SSR, SSE, MSR, and MSE?
e) What is the predicted value of y for x1  74 and x 2  140 ?
f) Construct a 99% confidence interval for the coefficient of x1 in the
population regression model.
g) Make a 95% confidence interval for the coefficient of x 2 in the population
regression model.
h) Determine a 90% confidence interval for  , the constant term in the
4. The following is the MINITAB solution for a regression of y on x1 , x 2
and x 3 .
Regression Analysis: Y versus X1, X2, X3

Y = 22.2 + 0.203 X1 - 0.0499 X2 - 0.216 X3
Predictor Coef SE Coef T P
Constant 2.212 4.602 4.83 0.003
X1 0.20276 0.06171 3.29 0.017
X2 -0.04991 0.04166 -1.20 0.276
140
X3 -0.21648 0.04836 -4.48 0.004
S = 1.016 R-SQ = 98.6% R-SQ(adj) = 97.8%
Source DF SS MS F P
Regression 3 422.21 140.74 136.44 0.000
Total 9 428.40
Source DF SEQ SS
X1 1 401.42
X2 1 0.12
X3 1 20.67
population regression model y    1  x1   2  x2   3  x3  
b) Write the values of a , b1 , b 2 , b 3 and explain the meaning of these
estimated regression coefficients.
c) Write the values of the standard deviation of the coefficients of a , b1 , b 2 ,
b3 .
d) What are the values of the variance and standard deviations of errors, the
coefficient of determination, the adjusted coefficient of determination, SST,
SSR, SSE, MSR, and MSE?
e) What is the predicted value of y for x1  33 , x 2  50 and x3  60 ?
regression model.
h) Determine a 99% confidence interval for x 3 , in the population regression
model.
5. In a study of revenue generated by national lotteries, the following
regression equation was fitted to a data from 26 countries with lotteries:
y  30.29  0.0354  x1  0.9734  x 2  340 .9524  x3
(0.00652) (0.3210) (225.78)
141
R 2  0.56%
where
y  dollars of net revenue per capita per year generated by lottery;
x1  mean per capita personal income of the country
x 2  number of hotel, motel, resort rooms per thousand of people
x3  spendable revenue per capita per year generated by legalized gambling
The numbers in parentheses below the coefficient estimates are the
corresponding estimated standard errors.
a) Interpret the estimated coefficient on x1 , x 2 and x 3 .
b) Find and interpret a 90% confidence interval for the coefficient on x 2 , in
the population regression.
c) Find and interpret a 99% confidence interval for the coefficient on x 3 , in
the population regression.
Answers
1. a) 2.167 to 2.473; 2.135 to 2.504; b) 0.48 to 2.34; 0.142 to 2.678; 2. 4.19
^
to 4.814; 4.12 to 4.88; 3.98 to 5.02; 3. a) y  12.4  0.24  x1  0.036  x 2 ;
b) a  12.410; b1  0.2415; b2  0.0362; c) s a  5.234; s b1  0.0345 ;
_ 2
s b2  0.024 ; d) s e2  3.0976 ; s e  1.76 ; R 2  97.8% ; R  96.6% ;
SST  882.30; SSR  849.65; SSE  32.65; MSR  424.83; MSE  3.27;
e) y p  35.2; f) 0.132 to 0.348; g) -0.017 to 0.089; h) 2.93 to 21.89;
^
4. a) y  22.2  0.203  x1  0.0499 x2  0.216 x3 ;b) a  2.212 ; b1  0.20276;
b2  0.04991 ; b3  0.21648 ;c) s a  4.602 ; s b1  0.06171 ; s b2  0.04166 ;
_ 2
s b3  0.04836 ;d) s e2  1.032 ; s e  1.016 ; R 2  98.6% ; R  97.8% ;
SST  428.40; SSR  422.21; SSE  6.19; MSR  140.74; MSE  1.03;
e) y p  13.444; f) 0.051 to 0.355; g) -0.132 to 0.031; h) -0.394 to -0.038;
5. b) 0.422 to 1.524 c) -977.42 to 295.52.
142
4.7. Test of hypothesis about individual coefficients
We can make a test of hypothesis about any of the  i coefficients of model
y    1  x1   2  x2   3  x3  ....   k  xk  
Using the same procedure that we used to make a test of hypothesis about 
for a simple regression model in previous chapter. The only difference is the
degrees of freedom, which are equal to (n  K  1) for a multiple regression.
In this case the value of the test statistic t for bi is calculated as
b  i
T .S .  t  i
s bi
The value of  i is substituted from the null hypothesis.
If the regression errors  i are normally distributed and the standard
regression assumptions hold, then the following hypothesis tests have
significance level 
H 0 :  i   0 or H 0 :  i   0
H1 :  i   0
Reject H 0 if T .S .  t n  K 1,
H 0 :  i   0 or H 0 :  i   0
H1 :  i   0
Reject H 0 if T .S .  t n  K 1,
3. To test null hypothesis
H 0 : i  0
H1 :  i   0
Reject H 0 if T .S .  t n  K 1, / 2 or T .S .  t n  K 1, / 2
Remark: In most cases we are interested in the null hypothesis H 0 :  i  0 .
143
Example:
For example 1 of the section 4.5, using 1% significance level, can you
conclude that the slope of the number of driving violations in regression
model is 0 against the alternative that it is positive? Use the MINITAB
solution given in Figure 4.1.
Solution:
x 2 is the number of driving violations committed during the past five years.
The portion of the solution is reproduced below
Constant 87.92 13.96 6.30 0.000
X1 -3.3869 0.9930 -3.41 0.011
X2 2.327 2.264 1.03 0.338
We are to test the following null and alternative hypotheses
H0 : 2  0
H1 :  2  0
reject H 0 if T .S .  t n  K 1,
From solution we obtain that t  T.S.  1.03 . It also can be found as
b   i 2.327  0
T .S .  t  i   1.03
s bi 2.264
d . f .  n  K  1  10  2  1  7
t n  K 1,  t 7,0.01  2.998
Since 1.03  2.998 , we accept the null hypothesis. Consequently, we
conclude that the slope of x 2 in regression model is zero. That is, the number
of driving violations is not significant and an increase (or decrease) in the
number of driving violations does not affect the auto insurance premium.
Remark:
Note that the observed value of test statistic T (test statistic t) is obtained
from the MINITAB solution only if the null hypothesis is H 0 :  2  0 .
However, if the null hypothesis is that  2 is equal to a number other than
zero, then the t value obtained from the MINITAB solution is no longer
valid. In this case observed value of the test statistic will be calculated as
b  2
T .S .  t  2
s b2
144
4.8. Tests on sets of regression parameters
In the previous section we developed a hypothesis test for individual

regression parameters. There are situations, where we are interested in the
effect of the combination of several variables. Now we will perform a test of
hypothesis with the null hypothesis that the coefficients of all independent
variables in the regression model are equal to zero and the alternative
hypothesis that the coefficients of all independent variables are not zero. For
the multiple regression model
y    1  x1   2  x2   3  x3  ....   k  xk  
the two hypotheses for such test are written as
H 0 : 1   2   3  .......   k  0
H 1 : at least one  i  0
A test of hypothesis for this case is performed by using the F distribution.
The F distribution has 2 degrees of freedom- df for numerator and - df for
denominator. Table 6 in Appendix lists the values of F for F distribution.
The value of the test statistic T.S.  F can be obtained from the computer
solution, or it can also be calculated by using formula
SSR / K
T .S .  F  or
SSE /( n  K  1)
MSR
T .S .  F 
MSE
where MSR stands for the mean square regression and MSE for the mean
square error.
SSR SSE
MSR ; MSE 
K (n  K  1)
In the end, to test the null hypothesis
H 0 : 1   2   3  .......   k  0
against the alternative hypothesis
H 1 : at least one  i  0
the decision rule for a significance level  is
reject H 0 if T .S .  F  FK ,n  K 1,
145
where FK ,n K 1, is the number for which P ( FK ,n  K 1  FK ,n  K 1, )   and
FK ,n K 1 follows an F distribution with numerator degrees of freedom K and
denominator degrees of freedom (n  K  1) .
Example:
Using 5% significance level, can you conclude that the coefficients of all
independent variables in the example 4.1 are equal to zero? Use the
MINITAB solution shown in Figure 4.1
Solution:
The two hypotheses are
H 0 : 1   2  0
H 1 : at least one   0
The portion of the solution is reproduced below
Source DF SS MS F P
Regression 2 4824.5 2412.3 12.72 0.005
Total 9 6152.4
From the portion of MINTAB solution we obtain
MSR  2412.3 ; MSE  189.7
and the value of the test statistic is T .S.  F  12.07
FK ,n  K 1,  F2,1021,0.05  F2,7,0.05  4.74
Because the value of the test statistic T .S.  F  12.07 greater than 4.74 , it
falls in the rejection region. Consequently, we reject the null hypothesis and
conclude that at least one of the two  ’s is different from zero.
Exercises
1. Refer to the study on milk consumption, described in exercise 2 and 4

after the section 4.4.4. Test the null hypothesis that 1   2  0
2. Refer to the study on weight gains, described in exercises 3 and 5 after the
section 4.4.4. Test the null hypothesis that 1   2   3  0
146
3. The following is the MINITAB solution for a regression of y on x1 , x 2
and x 3 .

Y = 51.6 – 0.0599 X1 + 0.0850 X2 - 0.00477 X3

Constant 51.61 11.25 4.59 0.000
X1 -0.05993 0.01526 -3.93 0.003
X2 0.08497 0.02875 2.96 0.014
X3 -0.004773 0.008707 -0.55 0.596
S = 0.8995 R-SQ = 99.7% R-SQ(adj) = 99.6%
Source DF SS MS F P
Regression 3 2909.91 969.97 1198.86 0.000
Total 13 2918.00
Source DF SEQ SS
X1 1 2901.82
X2 1 7.85
X3 1 0.24
population regression model y    1  x1   2  x2   3  x3  
b) Write the values of a , b1 , b 2 , b 3 and explain the meaning of these
estimated regression coefficients.
c) What are the values of the standard deviation of errors, the coefficient of
determination, the adjusted coefficient of determination, SST, SSR, SSE,
MSR, and MSE?
d) Write the values of the standard deviation, the value of test statistic, and
the p-value for each of the coefficients of a , b1 , b 2 , b 3 .
147
e) What is the predicted value of y for x1  310 , x2  260 and x3  180 ?
regression model.
h) Make a 98% confidence interval for the coefficient of x 3 in the population
regression model.
i) Determine a 95% confidence interval for  , the constant term in the
j) Using the 5% significance level, test the null hypothesis that the
coefficient of x1 in the population regression model is zero against the
alternative that it is negative.
k) Using the 1% significance level, can you conclude that the coefficient of
x 2 in the population regression model is zero against the alternative that it is
positive?
l) At the 2.5% significance level, test if the coefficient of x 3 in the
population regression model is zero against the alternative that it is negative.
m) Using the 5% significance level, can you conclude that the coefficients of
all independent variables in the population regression model are equal to
zero?
4. The Corporation has a large number of restaurants through the country.
The research department wanted to find if the sales of the restaurants depend
on the size of the population within a certain area surrounding the restaurants
and the mean income of households in those areas. They collected
information on these variables for 10 restaurants. The following table gives
information on the monthly sales (in thousands of dollars) of these
restaurants, the population (in thousands) within 10 kilometers of the
restaurants, and means monthly income (in hundreds of dollars) of the
households of those areas.
Sales 18 28 16 20 13 29 34 23 19 28
Population 22 16 33 19 47 70 30 45 77 41
Income 39 51 28 32 28 37 42 27 20 18
148
Using MINITAB (or any other statistical software package), find the
regression of sales on a population and income. Using solution, answer the
following questions.
b) Explain the meaning of the estimates of the constant term and regression
coefficients of the population and income.
c) What are the values of the standard deviation of errors, the coefficient of
determination, the adjusted coefficient of determination?
d) What are the value of the total sum of squares? What portion of SST is
explained by our regression model? What portion of SST is not explained by
our regression model?
e) What is the predicted sales for a restaurant with 52 thousand people living
within 10 km surrounding it and $3600 mean monthly income of households
living in those areas?
f) Construct a 95% confidence interval for the coefficient of income.
g) Using the 5% significance level, test the null hypothesis that the
coefficient of population in regression model is zero against the two-sided
alternative.
h) Using the 1% significance level, can you conclude that the coefficients of
both independent variables in the population regression model are equal to
zero?
Answers
1. T .S.  11; reject H 0 virtually at any level; 2. T .S.  3.06; reject H 0 at 5%
^
level; 3. a) y  51.6  0.0599  x1  0.0850  x2  0.0048  x3 ; b) a  51.61 ;
b1  0.05993; b2  0.08497 ; b3  0.004773 ;c) s e  0.8995 ; R 2  99.7%
_ 2
R  99.6% ; SST  2918.00; SSR  2909.91; SSE  8.09; MSR  969.97;
MSE  0.81; d) s a  11.25 ; ta  4.59; pa  0.000 ; s b1  0.01526 ; t b1  3.93;
p b1  0.003 ; s b2  0.02875 ; t b2  2.96; p b2  0.014 ; s b3  0.008707 ;
t b3  0.55; p b3  0.596 ; e) y p  54.2724; f)  0.09393 to  0.02593 ; g) -
0.00614 to 0.17608; h) -0.028839 to 0.019293; i) 26.545 to 76.675; j)
T.S.  3.927 ; reject H 0 ; k) T.S.  2.955 ; reject H 0 ; l) T.S.  0.548 ;
accept H 0 ;m) T.S.  F  1198.86 ; reject H 0
149
4.9. Dummy variables in the regression models
In the discussion of multiple regression we have assumed that the

independent variables, x i , have existed over a range and contained many
different values. All independent variables we have considered were
quantitative. We may include in regression model a variable that is
qualitative. Such a variable contains different categories instead of numerical
values. We will introduce independent variable that will take only two
values: 0 and 1. This structure is commonly defined as a “dummy variable”,
and we will see that it provides a valuable tool for applying multiple
regression to situations involving categorical variables.
Let us consider a simple regression equation
y    1  x1
Now suppose that we introduce a dummy variable, x 2 , that has values 0 and
1 and the resulting equation becomes
y    1  x1   2  x 2
When x 2  0 in this equation the constant is  , but when x 2  1 the constant

is    2 . Thus we see that the dummy variable shift the linear relationship
between y and x1 by the value of the coefficient  2 .
The number of dummy variables in a regression model is equal to the
number of categories minus 1. For instance, if a variable contains two
categories, then we introduce one dummy variable in the regression model
for this variable. If a qualitative variable contains three categories, we will
introduce two dummy variables and so on.
The following example shows how a dummy variable is used in regression
model.
Example:
Refer to example 1. Following table reproduces the data from that example
with additional column that contains information for each of the 10 drivers.
150
Yearly Driving Number of violations
premium experience (past 5 years) Gender
y x1 x2
74 5 2 Male
50 6 1 Female
97 4 6 Female
57 11 3 Female
99 3 1 Female
35 19 0 Male
40 15 1 Female
49 13 2 Female
101 2 8 Male
42 10 3 Male
Using MINITAB, find the regression of yearly auto insurance premium on

the years of experience, the number of driving violations, and the gender of
drivers. Answer the following questions
b) Explain the meaning of the estimated regression coefficient of the
independent variable gender.
c) What is the predicted auto insurance premium paid per year by a male
driver with 14 years of driving experience and 3 driving violations?
d) What is the predicted auto insurance premium paid per year by a female
driver with 14 years of driving experience and 3 driving violations?
e) Construct a 99% confidence interval for the coefficient of gender.
f) Using 1% significance level, test the null hypothesis that the coefficient of
gender is zero.
Solution:
Gender is not a quantitative variable, it is a qualitative variable. So, we will
use a dummy variable for it in regression model. Let
x1  driving experience (in years)
x 2  number of driving violations (during past 5 years)
We can denote dummy variable by x 3 . Also we can denote it by letter D.
Suppose
0 if a driver is a male
D
1 if a driver is a female
151
In this case, our population regression model becomes
y    1  x1   2  x2   3  D  
Assuming values of 0 and 1 to male and female respectively, we rewrite the
data
Yearly Driving Number of violations
premium experience (past 5 years) Gender
y x1 x2
74 5 2 0
50 6 1 1
97 4 6 1
57 11 3 1
99 3 1 1
35 19 0 0
40 15 1 1
49 13 2 1
101 2 8 0
42 10 3 0
The following figure shows the MINITAB solution
Regression Analysis: y versus X1, X2, D

y = 84.5 - 3.32 X1 + 2.56 X2 + 3.57 D
Predictor Coef St.dev. T P

Constant 84.54 17.58 4.81 0.003
X1 -3.318 1.078 -3.08 0.022
X2 2.559 2.502 1.02 0.346
D 3.573 9.829 0.36 0.729
S = 14.72 R-Sq = 78.9% R-Sq(adj) = 68.3%
Source DF SS MS F P
Regression 3 4853.1 1617.7 7.47 0.019
152
Total 9 6152.4
Source DF Seq SS
X1 1 4624.1
X2 1 200.4
D 1 28.6
a) The estimated regression equation is

^
y  84.5  3.32  x1  2.56  x 2  3.57  D (1)
We also can use column labeled COEF and write the estimated regression
equation as
^
y  84.54  3.318  x1  2.559  x 2  3.573  D (2)
b) The coefficient of the variable gender is b3  3.57 . It indicates that the
female drivers pay, an average, $3.573 more than male drivers with similar
driving experiences and the same number of driving violations.
In fact, by using dummy variable D in our regression model, we have
estimated two regression models: one for male drivers, and another for the
female drivers. Since for male drivers D  0 , after substituting it into the
estimated regression model we find the estimated model for male drivers as
^
y  84.54  3.318  x1  2.559  x 2
For female drivers D  1 . Substituting it in the regression model, we obtain
the estimated regression model for the female drivers
^
y  84.54  3.318  x1  2.559  x2  3.573  (1) 
 88.113  3.318  x1  2.559  x 2
We see that, the constant term for female drivers is 3.573 greater than that
for the male drivers’ model. Thus, on average, female drivers pay a yearly
auto insurance premium that is $3.573 more than the yearly auto insurance
premium paid by male drivers with similar driving experiences and the same
number of driving violations.
c) To find the predicted auto insurance premium for a male driver with 14
years of driving experience and 3 driving violations, we substitute x1  14 ,
x 2  3 , and D  0 in the estimated regression model (2),
153
^
y  84.54  3.318  14  2.559  3  3.573  0  45.765  $45.765
Thus, a male driver with 14 years of driving experience and 3 driving
violations is expected to pay a yearly auto insurance premium of $45.765.
d) To find the predicted auto insurance premium for a female driver with 14
years of driving experience and 3 driving violations, we substitute x1  14 ,
x 2  3 , and D  1 in the estimated regression model (2),
^
y  84.54  3.318  14  2.559  3  3.573  1  49.338  $49.338
Thus, a female driver with 14 years of driving experience and 3 driving
violations is expected to pay a yearly auto insurance premium of $49.338.
e) We are to make a 99% confidence interval for  3 . From the given
information and from the MINITAB solution we obtain
n  10 ; b3  3.573 ; and s b3  9.829
t n  K 1, / 2  t1031,0.005  t 6,0.005  3.707
So, from
b3  t n  K 1, / 2  s b3   3  b3  t n  K 1, / 2  s b3
a 99% confidence interval for  3 is
3.573  3.707  9.829   3  3.573  3.707  9.829
 32.863   3  40.009
Thus, the 99% confidence interval for  3 is -$32.863 to $40.009. We can
state with 99% confidence that female drivers pay somewhere between
$32.863 less than to $40.009more than male drivers with similar values for
the x1 and x 2 variables.
f) We are to test whether or not the coefficient  3 of gender in model (1) is
zero. The two hypotheses are
H 0 : 3  0
H1 : 3  0
Reject H 0 if T .S .  t  t n  K 1, / 2 or T .S .  t  t n  K 1, / 2
From MINITAB solution we find that the value of test statistic is
T.S.  t  0.36
t n  K 1, / 2  t1031,0.005  t 6,0.005  3.707 and
154
 t n  K 1, / 2  3.707
Since 0.36 is not greater than 3.707, the value of test statistic falls in the non
rejection region. Consequently, we fail to reject the null hypothesis and
conclude that  3 in regression model is not different from zero. That is, the
variable gender has no effect on the auto insurance premiums paid by
drivers.
Remark:
The number of dummy variables used for qualitative variable in a regression
model is one less than the number of categories for that variable. For
example, we may want to investigate influence of quarters. Because the
variable quarter is a qualitative variable, we will use dummy variables to
represent it in our regression model. Since there are 4 quarters in a year, we
will use 3 dummy variables. Let D1  be the dummy variable for the first
quarter, D2  be the dummy variable for the second quarter and D3  be the
dummy variable for the third quarter. Then
D1  1 for the first quarter, and zero for other quarters
D2  1 for the second quarter, and zero for other quarters
D3  1 for the third quarter, and zero for other quarters
If our regression model consists of two independent variables x1 and x 2 ,
then we will estimate regression model as
y    1  x1   2  x2   3  D1   4  D2   5  D3  
Exercises
1. The following model was fitted to data on 28 insurance companies.

y  6.85  0.12  x1  2.13  D
;
(0.006) (0.532)
where
y  price-earning ratio
x1  size of insurance company assets, in millions of dollars
D  dummy variable, taking the value 1 for foreign companies,
and 0 for national companies
a) Interpret the estimated coefficient on the dummy variable
155
b) Test against a two-sided alternative the null hypothesis that the true
coefficient on the dummy variable is zero. Take   0.05
2. The following model was fitted, to explain the selling prices of home, to a
sample of 815 sales.
y  1264  48.18  x1  3382  x 2  3219  x3  2005  x 4 _ 2
R  0.86
(0.91) (515) (947) (768)
where
y  selling price of home, in thousands of dollars
x1  square meters of living area
x 2  size of garage, in square of meters
x3  dummy variable taking the value 1 if the house has a fireplace,
and 0 otherwise
x 4  dummy variable taking the value 1 if the house has a wood
floors, and 0 otherwise
a) Interpret the estimated coefficient of x 3 .
b) Interpret the estimated coefficient of x 4 .
c) Find a 95% confidence interval for the impact of fireplace on a selling
price, all other being equal.
d) Test the null hypothesis that type of flooring has no impact on selling
price, against the alternative that, all other things equal, house with wood
floors have a higher selling price than other flooring.
3. The following MINITAB solution was obtained for the regression model
y    1  x1   2  x2   3  x3   4  D  
for a sample data set.

y = 22.3 + 0.190 X1 - 0.0533 X2 - 0.190 X3 - 1.93 D

Constant 22.315 2.802 7.96 0.001
X1 0.19019 0.03776 5.04 0.004
X2 -0.05325 0.02538 -2.10 0.090
X3 -0.19033 0.03046 -6.25 0.002
D -1.9349 0.5785 -3.34 0.020
156
S = 0.6183 R-Sq = 99.6% R-Sq(adj) = 99.2%
Source DF SS MS F P
Regression 4 426.49 106.62 278.87 0.000
Total 9 428.40
Source DF Seq SS
X1 1 401.42
X2 1 0.12
X3 1 20.67
D 1 4.28
Using MINITAB solution, answer the following questions.

b) Explain the meaning of b 4 obtained by estimating the given regression
model.
c) What is the predicted value of y for x1  45 ; x 2  40 ; x3  60 ; and D  0 ?
d) Construct a 99% confidence interval for the coefficient of D .
e) Using the 5% significance level, test the null hypothesis that the
coefficient of D is zero against two-sided alternative.
4. The salaries of workers are expected to be depending on the number of
years they have spent in school and their work experience. The following
table gives information on annual salaries (in thousands of dollars), number
of years of studying, and years of experience for 10 persons. It also includes
information on gender. In the table, M represents males and F refers to
females.
Salary 30 22 21 45 36 39 17 22 18 19
Studying 18 16 15 22 20 20 14 16 12 14
Experience 8 7 6 15 14 16 2 4 3 4
Gender F F M M F F M M F M
Using MINITAB (or any other statistical software package), find the
regression of salary on studying, experience, and gender. Then answer the
following questions.
157
b) Explain the meaning of the estimated regression coefficient of the dummy
variables.
c) By estimating the regression model with gender as a dummy variable, you
have actually estimated two regression models-one for males and the other
for females. Write these two regression equations.
d) How much salary is a male worker with 18 years of studying and 7 years
of work experience expected to earn?
e) How much salary is a female worker with 18 years of studying and 7
years of work experience expected to earn?
f) Determine a 95% confidence interval for the coefficient of dummy
variable.
g) Using the 5% significance level, can you conclude that female workers
are paid lower salaries than male workers?
Answers
1. a) All else being equal, expected price-earning ratio is higher by 2.13

million of dollars for foreign companies; b) T .S.  t  4 ; reject H 0 ;
2. a) All else being equal, expected selling price is higher by $3219 if house
has a fireplace; b) All else being equal, expected selling price is higher by
$2005 if house has a wood floor; c) 1363   3  5075 ;d) T.S.  2.611 ;can
^
reject H 0 at 0.5%; 3. a) y  22.3  0.190  x1  0.0533  x2  0.190  x3 
 1.93  D ; c) 17.32375; d) -0.4025 to 4.2625; e) T.S.  t  3.34 ; reject H 0 .
158
Chapter 5
Analysis of variance (ANOVA)
5.1. Introduction
In Chapter 1 we discussed how to test whether or not the means of

two populations are equal. Recall that the test involved the selection of an
independent random sample from each of the populations. In this chapter we
will discuss a statistical procedure for determining whether or not the means
of more than two populations are equal. The technique that we will be
introduced is called the analysis of variance (ANOVA) procedure.
5.2. One-way analysis of variance
This section discusses the one-way analysis of variance procedure

to make tests comparing means often is called one-way analysis, because we
will analyze only one factor. Sometimes we may analyze the effects of two
factors. This is called two-way analysis of variance.
Suppose that we have independent samples of sizes n1 , n2 , n3 ,......n K
observations selected randomly from K populations. Assume that population
means are 1 ,  2 ,  3 ,...... K . The one-way analysis of variance procedure
designated to test the null hypothesis
H 0 : 1   2   3  ......   K (All population means are equal)
against the alternative hypothesis
H 1 : At least one of the population has different mean
The application of one way analysis of variance requires that the following
assumptions hold:
1. The populations from which the samples are selected are (approximately)
normal.
2. The populations from which the samples are selected have the same
variance (or standard deviation).
3. The samples selected from different populations are independent.
Suppose that from K populations samples of sizes n1 , n2 , n3 ,......n K are
selected (Figure 5.1)
159
Figure 5.1
Block POPULATION (GROUP)
1 2 …. K
x11 x 21 …. x K1
x12 x 22 …. xK 2
…. …. …. ….
…. …. …. …..
x1n1 x 2 n2 …. x Kn K
1) The first step is to calculate the sample mean for the K groups of
_ _ _
observations. These sample means will be denoted as x 1 , x 2 ,....... x K .
In general
ni
_
x
j 1
ij
xi 
ni
where n i denotes the number of observations in i th group.
2) The second step is to find overall mean of the all sample observations,
__
denoted x , and defined as
K ni
_
 x
i 1 j 1
ij
x
n
where n denotes the total number of sample observations
K
n   ni
i 1
An equivalent expression for overall mean is
K _
__  ni x i
x i 1
n
3) In third step, we consider variability within-groups. To measure
variability in the any group, we calculate the sum of squared deviations of
160
the observations about their sample means. Within-groups variability will be
denoted by SS. For example, for the first group the sum of squared
_
deviations of the observations about their sample mean x 1 is
n1 _
SS1   ( x1 j  x1 ) 2
j 1
_
For the second group, whose sample mean is x 2 , we calculate
n2 _
SS 2   ( x2 j  x 2 ) 2
j 1
and so on.
4) In fourth step, we find total within-groups variability, denoted SSW. That
is
SSW  SS1  SS 2  .......  SS K
or
K ni _
SSW   ( xij  x i ) 2
i 1 j 1
5) Now we need a measure of variability between groups. It is based on the
discrepancies between the individual group means and the overall mean.
Total between-groups sum of squares denoted SSG , and defined as
K _ _
SSG   ni  ( x i  x) 2
i 1
6) As a last step, we calculate the sum of squared discrepancies of all the
sample observations about their overall mean. This is called the total sum of
squares, denoted SST , expressed as
K ni _
SST   ( xij  x) 2
i 1 j 1
It can be shown, that the total sum of squares is the sum of the within-groups
and between-groups sum of squares, that is
SST  SSW  SSG
Testing the equality of population means is based on the assumption that K

populations have equal variances (or standard deviations).
161
If H 0 : 1   2   3  ......   K
is true, each of the SSW and SSG can be used as the basis for estimate of the
common population variance. To obtain these estimates, the sums of squares
must be divided by the corresponding numbers of degrees of freedom.
SSW divided by (n  K ) results estimate called the within-groups mean
square, denoted MSW, so that
SSW
MSW 
nK
SSG divided by ( K  1) results estimate called the between-groups mean
square, denoted MSG, so that
SSG
MSG 
K 1
The test of null hypothesis is based on the ratio of the mean squares
MSG
F
MSW
If this ratio is close to 1, there would be little cause to doubt the null
hypothesis of equality of population means.
Summary
We define the following sums of squares:
K ni _
Within-groups: SSW   ( x
i 1 j 1
ij  x) 2
K _ _
Between groups: SSG   ni  ( x i  x ) 2
i 1
K ni _
Total: SST  
i 1 j 1
( xij  x) 2
We define the mean squares as follows:

SSW
Within-groups: MSW 
nK
SSG
Between groups: MSG 
K 1
The null hypothesis is
H 0 : 1   2   3  ......   K
162
MSG
Reject H 0 if T .S.  F   FK 1,n K ,
MSW
where FK 1, n  K , is the number for which P ( FK 1,n  K  FK 1,n  K , )   and
FK 1,n  K follows an F distribution with numerator degrees of freedom (K-1)
and denominator degrees of freedom (n  K ) . (Table 6 of Appendix).
For convenience, these calculations are often recorded in a table called a
one-way analysis of variance table or ANOVA table, shown below
(Table5.1):
Table 5.1
Source of Sum of Degrees of Mean F ratio
variation squares freedom squares
Between groups SSG ( K  1) MSG MSG
Within groups SSW (n  K ) MSW MSW
Total SST (n  1)
Example:
A company buys thousands of light bulbs every year. The company is
considering three brands of light bulbs to choose from. Before the company
decides which light bulbs to buy, it wants to investigate if the mean life of
the three types of light bulbs is the same. The research department selects
randomly a few bulbs of each type and tested them. Table lists number of
hours (in thousands) that each of the bulbs in each brand survived before
being burned out.
Brand I Brand II Brand III
22 18 27
23 23 24
26 22 20
27 21 21
22 23
At the 5% significance level, test the null hypothesis that the mean life of
bulbs for each of these three brands is the same.
163
Solution:
There are 3 groups: Brand I, Brand II, and Brand III:

n1  5 ; n2  4 ; n3  5 ; n  14
1) Let us calculate mean of each group
__
22  23  26  27  22 120
x1    24
5 5
__
18  23  22  21 84
x2    21
4 4
__
27  24  20  21  23 115
x3    23
5 5
2) Overall mean is
K _
__ n
i 1
i xi
5  24  4  21  5  23 319
x    22.79
n 14 14
3) In the first group, sum of squared deviations is
5 _
SS1  (x
j 1
1j  x1 ) 2  (22  24) 2  (23  24) 2  (26  24) 2 
 (27  24 ) 2  (22  24 ) 2  4  1  4  9  4  22
Similarly,
4 _
SS2  (xj 1
2j  x 2 ) 2  (18  21) 2  (23  21) 2  (22  21) 2 
 (21  21) 2  9  4  1  0  14
and
5 _
SS3  (x
j 1
3j  x 3 ) 2  (27  23) 2  (24  23) 2  (20  23) 2 
 (21  23) 2  (23  23) 2  16  1  9  4  0  30

4) SSW  SS1  SS 2  .......  SS K  22  14  30  66
5) Now, let us calculate between group variability
164
K _ _
SSG  n
i 1
i  ( x i  x) 2  5  (24  22.79) 2 
 4  (21  22.79) 2  5  (23  22.79) 2  7.32  12.82  0.22  20.36 .

6) SST  SSW  SSG  66  20.36  86.36
Within-groups mean square is obtained as

SSW 66
MSW   6
n  K 14  3
Between-groups mean square is obtained as
SSG 20.36
MSG    10.18
K 1 3 1
H 0 : 1   2   3
The decision rule is reject H 0 if
MSG
T .S.  F   FK 1,n K ,
MSW
MSG 10.18
The value of test statistic is F    1.70
MSW 6
FK 1,n K ,  F2,11,0.05  3.98 .
Since 1.70 is not greater than 3.98 we fail to reject H 0 . We accept
hypothesis that mean of all three populations are equal, in other words, there
is no difference for company which brand to choose.
In the end, substituting the values of various quantities in Table 5.1, we write
an ANOVA table for our example as

Between groups 20.36 2 10.18 10.18
Within groups 66 11 6  1.70
6
Total 86.36 13
165
Remark1: An alternative formula for SSB and SSW are
2
 n 
 T12 T22 T32

 xi 
  i 1 
SSB      ....  
 n1 n2 n3  n
n
 T12 T22 T32 
SSW   x 2
i    ....
i 1  n1 n2 n3 
where
Ti  the sum of the values in sample i
n
x
i 1
i  the sum of the values in all samples  T1  T2  T3  ....
n
x
i 1
2
i  the sum of the squares of the values in all samples.
Example:
Consider the following data obtained for two samples selected from two
populations
Sample I Sample II
9 4
3 1
7 1
8 6
8
Set out the analysis of variance table for these data.

Solution:
T1  9  3  7  8  27
T2  4  1  1  6  8  20
9
x
i 1
i  T1  T2  27  20  47
n1  4 ; n2  5 ; n  n1  n2  9
166
n
x
i 1
2
i  9 2  3 2  7 2  8 2  4 2  12  12  6 2  8 2  321
Substituting all the values in the formula for SSG and SSW, we obtain
2
 n 
 T12 T22 T32
 xi 
  i 1 
SSG      ....   
 n 1 n 2 n 3  n
 27 2 20 2  47 2
     16.81
 4 5  9
n T 2 T 2 T 2 
SSW   xi2  1  2  3  ....  
i 1  n1 n2 n3 
 27 2 20 2 
 321      58.75
 4 5 
Hence, the variance between samples MSG and the variance within samples
MSW are
SSG 16.81
MSG    16.81
K 1 2 1
SSW 58.75
MSW    8.39
nK 92
We write an ANOVA table for our example as

Between groups 16.81 1 16.81 2.00
Within groups 58.75 7 8.39
Total 75.56 8
Remark2:
To use MINITAB menu follow the following instructions:
1. Select Stat>ANOVA>One-way (Unstacked)
2. Select data columns
3. Click OK
167
Exercises
1. The following ANOVA table, based on information obtained for four
samples selected from four independent populations that are normally
distributed with equal variances, has a few missing values

Between groups ?
Within groups 16 8.245 F  5.67
?
Total 19
a) Complete the analysis of variance table.

b) Using   0.05 , test the null hypothesis that the means of the four
populations are equal against the alternative hypothesis that the means of the
four populations are not equal.
2. Respond to each of the following questions using this partially completed
one-way ANOVA table

Between groups 3 F ?
Within groups 405
Total 888 31
a) How many different populations are being considered in this analysis?
b) Fill in the ANOVA table with missing values.
c) State the appropriate null and alternative hypothesis.
d) What conclusions should be reached regarding the null hypothesis?
Test using an   0.05 .
3. Three samples randomly selected from three independent populations that
are normally distributed with equal variances produced the following data.
Sample I Sample II Sample III

32 45 47
28 43 32
40 38 43
36 44 37
39 33 41
168
a) Set out the analysis of variance table for these data.
b) Test at a 1% significance level, the null hypothesis that the means of these
three populations are equal.
4. Consider the following data obtained for two samples selected at random
from two populations that are independent and normally distributed with
equal variances
Sample I Sample II
29 37
31 27
27 36
28 20
25

b) Test at a 5% significance level, the null hypothesis that the means of these
three populations are equal.
5. A company hired three new salespersons with degrees in mathematics,
economics and marketing. The company wants to check if the fields of study
have any effect on the mean number of sales made by these salespersons.
The following table lists the number of items sold by these three
salespersons during certain randomly selected days.
Person with Person with Person with

Math. degree economics degree marketing degree
7 2 3
8 5 1
6 3 1
11 1 2
9 2 6
13 5 5

b) Using the 5% significance level, can you reject the null hypothesis that the
mean number of items sold per day by all salespersons with degrees in each
of these three areas is the same?
169
6. A consumer agency that wanted to compare drying times for paints made
by three companies tested a few samples of paints from each of these
companies. The following table lists the drying times (in minutes) for these
samples of paints
Company A Company B Company C

43 58 44
52 64 49
42 62 50
46 54 57
41 52 43
50 62 40
55 46

b) Using the 1% significance level, test the null hypothesis that the mean
drying times for paints of these companies are equal.
Answers
1. b) T .S.  F  5.67 ; reject H 0 ; 2. a) 4; d) T .S.  F  11.13; reject H 0 ;
3. a)

Between groups 94.5 2 47.3
Within groups 333.2 12 27.8 F  1.70
Total 427.7 14
b) accept H 0 ;
4. a)
Within groups 214.0 7 30.6 F  0.29
Total 222.9 8
b) accept H 0 ;
170
5. a)
Within groups 70.00 15 4.67 F  15.43
Total 214.00 17
b) reject H 0 ;
6. a)
Within groups 481.3 17 28.3 F  10.10
Total 1053.0 19
b) reject H 0 .
5.3. The Kruskal-Wallis test
One-way analysis of variance test is based on an assumption that the

underlying population has a normal distribution. If the population
distribution is not normal, it is possible to develop a nonparametric
alternative to the one-way analysis of variance. This nonparametric test is
known as the Kruskal-Wallis test. Like the most of nonparametric the
Kruskal-Wallis test is based on the ranks of the sample observations.
Suppose that we have independent random samples of sizes
n1 , n2 , n3 ,......., n K observations, selected from K populations. Let
n  n1  n2  n3  .......  n K
Denote the total number of observations. The null hypothesis is
H 0 : 1   2   3  ......   K
To apply Kruskal-Wallis test it is necessary to take following steps:
1. Pool together all sample observations.
2. Rank all of pooled sample observations in ascending order.
3. Denote by R1 , R2 ,....... RK the sums of ranks for the K samples
4. Calculate the value of the test statistic
12 K
Ri2
W   3(n  1)
n(n  1) i 1 ni
171
A test of significance level  is given by the decision rule
Reject H 0 if W   K2 1,
where  K2 1, is the number that is exceed with probability  by a
 2 random variable with ( K  1) degrees of freedom.
Example:
The following table gives the response time (in minutes) of three fire
companies in a city for certain randomly selected incidents after a fire was
reported.
Company A Company B Company C

1.6 1.4 0.8
0.8 2.6 1.3
2.7 0.9 1.7
1.2 3.5 0.9
3.4 1.2 1.1
1.9 1.5 0.7
4.3 2.1
Perform at the 5% significance level the Kruskal-Wallis test to test the null
hypothesis that the mean response time for each of these fire companies for
all fire incidents are the same.
Solution:
First of all we pool all sample observation together and rank them in
ascending order. The following table illustrates this procedure
Company Rank Company Rank Company Rank

A B C
1.5 11 1.4 10 0.8 2.5
0.8 2.5 2.6 16 1.3 9
2.7 17 0.9 4.5 1.7 13
1.2 7.5 3.5 19 0.9 4.5
3.4 18 1.2 7.5 1.1 6
1.9 14 1.6 12 0.7 1
4.3 20 2.1 15
Rank sums 90 69 51
172
H 0 : 1   2   3
Reject H 0 if W   K2 1,
12 K
Ri2
W   3(n  1) 
n(n  1) i 1 ni
12  90 2 69 2 512 
      3  (20  1)  66.35  63  3.35
20  (20  1)  7 6 7 
 K2 1,   22,0.05  5.99
Since, 3.35 is not greater than 5.99, we fail to reject the null hypothesis. And
we accept that the mean response time for each of these fire companies for
all fire incidents is the same.
Exercises
1. A manufacturer of wall papers is considering three alternative colors: red,

green and blue. To check whether such consideration has an effect on sales,
21 stores are chosen. Red color papers are sent 7 of them, green color papers
are sent 9 of them, and blue color papers are sent to the remaining 5. After a
few days, a check is made on the number of sales in each store. The results
are shown below
Red Green Blue

44 53 84
38 77 29
58 29 46
77 65 85
60 89 35
29 34
58 43
62
35
173
At the 5% significance level perform a Kruskal-Wallis test of the null
hypothesis that the population mean sales levels are identical for three wall
paper colors. Also find the p-value.
2. A study was conducted in which samples were selected independently
from four populations. The sample size from each population was 21. The
data were converted to ranks. The sum of the ranks for the data from each
sample is
Sample A Sample B Sample C Sample D

Sum of ranks 642 784 458 1361
a) State the appropriate null and alternative hypothesis if you wish to

determine whether the populations have equal means.
b) At the 2.5% significance level perform a Kruskal-Wallis test.
3. The following summary data have been collected from three samples
selected from three populations
n1  20 n2  25 n3  35
R 1  1660 R 2  1150 R3  1350
Based on these data, what can be concluded about the means for three
populations? Apply Kruskal-Wallis test at an   0.01.
4. Given the following data:
Group 1 Group 2 Group 3 Group 4

21 28 16 22
27 26 15 23
26 21 18 19
22 29 20 17
25 30 15 20
30 25
23
Use the Kruskal-Wallis procedure to test the null hypothesis that the mean
values for all four populations are the same. What conclusion should be
reached using a significance level of 0.10? Also find the p-value.
174
5. Suppose as a part of your job you are responsible for installing emergency
lighting in a series of buildings. Bids have been received from four
manufacturers of battery-operated emergency lights. The costs are about
equal, so the decision will be based on the length of time the lights last
before failing. A sample of five lights from each manufacturer has been
tested, and values (time in hours) recorded for each manufacturer
Type I Type 2 Type 3 Type 4

1025 1222 1121 989
1122 1250 1201 987
1250 1390 1190 1087
1023 1426 1122 1121
1130 1322 1390 1200
Using   0.01, what conclusion should you reach about the mean length of
time the lights last before failing for the four manufacturers? Explain.
Answers
1. T.S.  0.0376 ; accept H 0 ; 2. T.S.  7.74 ; accept H 0 ; 3.

T.S.  206.58 ; reject H 0 ; 4. b) T.S.  14.95 ; reject H 0 ; 5. T .S.  2.22 ; fail
to reject H 0 .
5.4. Two-way analysis of variance
In section 5.2 we introduced one-way ANOVA for testing hypothesis

involving three or more population means. This ANOVA method is
appropriate as long as we are interested in analyzing one factor at a time and
we select independent random samples from the populations. There are
situations in which another factor affects the observed response in a one-way
design.
Suppose that we have K groups and H blocks. We will use x ij to denote the
sample observations corresponding to the i th group and the j th block, as it
shown in Figure 5.2. There are total
nKH
number of observations.
175
Our aim is to test the null hypothesis that all group means are equal, and the
null hypothesis that all block means are equal.
To develop these tests we need to set two-way ANOVA table.
BLOCK GROUP
1 2 …… K
1 x11 x 21 ……. x K1
2 x12 x 22 ……. xK 2
. … … ……. …..
. …. …. ….. …
H x1H x2H ……. x KH
Figure 5.2
To set this table we will use following steps:

1) Find sample mean for each group. For the mean of the i th group we use
_
notation x i , defined as
H
_
x
j 1
ij
x i  ; (i  1,2,3......, K )
H
2) Find sample mean for each block. The mean of the j th block we use
_
notation x  j , defined as
K
_ x
i 1
ij
x j  ; ( j  1,2,3......, H )
K
3) Find the overall mean of the sample observations. The overall mean
_
denoted x , defined as
K H K _ H _
_
 x
i 1 j 1
ij  x i x
j 1
j
i 1
x  
n K H
176
4) Find between groups sum of squares, denoted SSG, defined as
K _ _
SSG  H   (x
i 1
i  x) 2
5) Find between blocks sum of squares, denoted SSB , defined as

H _ _
SSB  K   (x
j 1
j  x) 2
6) Find the error sum of squares, denoted SSE , defined as

K H _ _ _
SSE   ( x
i 1 j 1
ij  x i   x j   x) 2
7) Find the total sum of squares, denoted SST , defined as

K H _
SST  
i 1 j 1
( xij  x) 2
It can be shown that

SST  SSG  SSB  SSE
8) We define the following mean squares
SSG
Between-groups: MSG 
K 1
SSB
Between-blocks: MSB 
H 1
SSE
Error: MSE 
( K  1)( H  1)
9) We define two F ratios
MSG MSB
and
MSE MSE
We will use ratios above to test the null hypothesis about equality of
population blocks and population groups.
1) The null hypothesis H 0 that the K population group means are the same is
provided by the decision rule
MSG
Reject H 0 if  FK 1,( K 1)(H 1),
MSE
177
2) The null hypothesis H 0 that the H population block means are the same is
provided by the decision rule
MSB
Reject H 0 if  FH 1,( K 1)(H 1),
MSE
where, F 1 , 2 , is the number exceeded with probability  by a random
variable following an F distribution with numerator degrees of freedom
 1 and denominator degrees of freedom  2 .
It is very convenient to summarize the calculations in tabular form, called a
two-way analysis of variance table or ANOVA table, shown below
(Table5.2):
Table 5.2
Source of Sum of Degrees of Mean F
variation squares freedom squares ratios
SSG
Between groups SSG ( K  1) MSG 
K 1 MSG
SSB MSE
Within blocks SSB ( H  1) MSB 
H 1 MSB
SSB MSE
Error SSE ( K  1)(H  1) MSE 
( K  1)(H  1)
Total SST (n  1)
Exercise:
Four drivers tested three types of cars for fuel consumptions. The
accompanying table shows fuel consumptions of cars
Block (Drivers) Group (Cars)
A B C
1 22 24 26
2 21 25 22
3 19 20 23
4 18 19 21
a) Set out the two-way analysis of variance table.
b) Test the null hypothesis that the population mean fuel consumption is the
same for all three types of cars. Take   0.05 .
c) Test the null hypothesis that population values of mean fuel consumption
are the same for each driver. Take   0.05 .
178
Solution:
a)
1) Let us find sample mean for each group
H
_
x
j 1
ij
x i  (i  1,2,3)
H
_
22  21  19  18 80
x1    20
4 4
_
24  25  20  19 88
x 2    22
4 4
_
26  22  23  21 92
x 3    23
4 4
2) Find sample mean for each block.
K
_ x
i 1
ij
x j  ; ( j  1,2,3,4)
K
_
22  24  26 72 _
21  25  22 68
x 1    24 ; x 2    22.67
3 3 3 3
_
19  20  23 62 _
18  19  21 58
x 3    20.67 ; x 4    19.33
3 3 3 3
3) Find the overall mean of the sample observations.

K H K _ H _
_

j 1 i 1
xij  x i x
j 1
j
i 1
x  
n K H
K _
_ x
i 1
i
20  22  23
x   21.67
K 3
4) Find between groups sum of squares
K _ _
SSG  H  i 1
( x i  x) 2 

 4  (20  21.67)  (22  21.67) 2  (23  21.67) 2  18.67
2

179
5) Find between blocks sum of squares
H _ _
SSB  K  
j 1
( x  j  x) 2  3  ((24  21.67) 2  (22.67  21.67) 2 
 (20.67  21.67) 2  (19.33  21.67) 2 )  38.71
6) Let us find total sum of squares

K H _
SST   ( x
i 1 j 1
ij  x) 2  (22  21.67) 2  (24  21.67) 2 
 (26  21.67) 2  (21  21.67) 2  (25  21.67) 2  (22  21.67) 2 

 (19  21.67) 2  (20  21.67) 2  (23  21.67) 2  (18  21.67) 2 
 (19  21.67) 2  (21  21.67) 2  68.67
7) By subtraction we obtain error sum of squares
SSE  SST  SSG  SSB  68.67  18.67  38.71  11.29
8) For the fuel consumption data, the mean squares are
SSG 18.67
MSG    9.34
K 1 3 1
SSB 38.71
MSB    12.90
H 1 4 1
SSE 11.29
MSE    1.88
( K  1)( H  1) (3  1)(4  1)
9) We define two F ratios
MSG 9.34
  4.97
MSE 1.88
MSB 12.90
and   6.86
MSE 1.88

Between groups 18.67 2 9.34 4.97
Within blocks 38.71 3 12.90 6.86
Error 11.29 6 1.88
Total 68.67 11
180
b) We can write the null hypothesis that the population means fuel
consumption is the same for all three types of cars as
H 0 : 1   2   3
MSG
Reject H 0 if  FK 1,( K 1)(H 1),
MSE
FK 1,( K 1)(H 1),  F2,6,0.05  5.14
Since the value of test statistic 4.97 is not greater than 5.14, we fail to reject
the null hypothesis. Therefore, we accept the hypothesis that the fuel
consumptions are the same for all types of cars.
c) We write the null hypothesis of equality of the population values of mean
fuel consumption for all four drivers as
H 0 : 1   2   3   4
MSB
Reject H 0 if  FH 1,( K 1)(H 1),
MSE
FH 1,( K 1)(H 1),  F3,6,0.05  4.76
Since the value of test statistic 6.86 is greater than 4.76, we reject the null
hypothesis. Therefore, we accept the hypothesis that the fuel consumptions
are not the same for each driver age class. In other words, fuel consumption
of car depends on driver’s habit.
Remark1:
1. Select Stat>ANOVA>Two-way
2. Enter Response variable
3. Enter row factor
4. Enter column factor
5. Click OK.
181
Remark2:
For example above the MINITAB instruction is shown below
C1 C2 C3
Driver Car Fuel consumption
1 1 22
1 2 24
1 3 26
2 1 21
2 2 25
2 3 22
3 1 19
3 2 20
3 3 23
4 1 18
4 2 19
4 3 21
In this case the row factor is “Car” and the column factor is “driver”
Exercises
1. An ANOVA was performed and the following partially completed

ANOVA table is available:

Between groups 7 26
Within blocks 14
Error 57.900
Total 402.100
a) Complete the analysis of variance table.
b) How many populations are being tested?
c) How many blocks were used in this analysis of variance?
d) Using an alpha equal to 0.05, test the null hypothesis that the population
means are equal for all groups.
e) Using an alpha equal to 0.05, test the null hypothesis that the population
means are equal for all blocks.
182
2. Three analysts were asked to predict earnings growth over the coming
year for four companies producing cars. Their forecasts (in percentage
increase in earnings) are given below
Analysts Cars companies

A B C D
1 7 11 10 13
2 8 9 13 12
3 9 13 16 15
a) Set out analysis of variance table

b) For   0.01 test the null hypothesis that the population mean growth
forecasts are the same for all car companies.
c) For   0.05 test the null hypothesis that the population mean growth
forecasts are the same for all analysts.
3. A soft drink producer wants to compare the effects on sales of can colors:
red, yellow, and blue. He selects four regions and in three stores in each
region, each to sell one color cans. The following table shows number of
sales (in hundreds of cans) at the end of the experiment.
Regions Can colors

Red Yellow Blue
Region I 37 28 25
Region II 29 33 22
Region III 31 29 27
Region IV 23 26 26
a) Set out analysis of variance table

b) Test the null hypothesis that the population mean sales are the same for
each can color. Use p-value approach.
c) Test the null hypothesis that the population mean sales are the same for all
four regions. Use p-value approach.
4. Three real estate agents were each asked to assess the values of five
houses. The results, in thousands of dollars, are shown below.
183
House Agents
A B C
I 200 210 220
II 190 192 196
III 180 195 205
IV 160 182 194
V 170 171 185
a) Set out analysis of variance table.
b) Test at 5% significance level the null hypothesis that population mean
valuations are the same for the three real estate agents.
Answers
MSG MSB
1. b) 8; c) 15; d)  44.06 ; reject H 0 ; e)  19.63 ; reject H 0 ;
MSE MSE
2. a)
Between car comp. 54.00 3 18.00 10.29
Within analysts 22.17 2 11.08 6.33
Error 10.50 6 1.75
Total 86.67 11
b) T.S.  10.29 , reject H 0 ; c) accept H 0 ;
3. a)

Between colors 56.0 2 28.0 1.71
Within regions 42.0 3 14.0 0.86
Error 98.0 6 16.3
Total 196.0 11
4. a)
Between agents 1000.0 2 500.0 14.79
Within houses 2285.3 4 571.3 16.90
Error 270.7 8 33.8
Total 3556.0 14
184
Chapter 6
Statistical quality control
6.1. Introduction
In manufacturing process, there is always some variation in the

items manufactured. For example, even though a machine is designed to cut
a piece of pipe a certain length, the lengths of the pipes cut will not all be
exactly equal. There are two causes of variation: normal or chance of
variation; and variation that is due to human or mechanical reasons in the
manufacturing process. If the variation is due to human or mechanical
reasons in the process, this causes defective parts, it must be detected at an
early stage and corrected.
A real life example concerns the weights of canned food. In this
case, a food producer wishes to provide assurance that the minimum weight
of its canned product is being met. Food producers have a tendency to fill
cans with more food than the can’s label indicates, out of fear of being
caught with under weight cans by the state inspectors.
The question is, “how does one determine (with the variability in the
filling process) with assurance that the minimum weight of all cans is being
achieved and at the same time, not lose money by overfilling each can? This
chapter will explain a technique called statistical quality control, which can
be used to answer this question.
6.2. Variation
One of the fundamental principles of the statistical thinking is that variation

exists in all process. It is important to understand variation in order to predict
the future performance of the process. There are two causes of variation:
1) common causes
2) assignable causes
Common causes of variation (also called uncontrollable causes) are those
causes that are random in occurrence and are happens during all process.
Management, (not workers), are responsible for these causes.
Assignable causes of variation (also called special cases) are the results of
external sources, that is, sources that are outside of the system. These causes
185
can and must be detected, and corrective actions must be taken to remove
them from the process. Not taking actions will increase variation and lower
the quality.
Definition:
A production process is called stable (in-control) if all assignable causes are
removed; thus, variation results only from common causes.
6.3. Control charts
Figure 6.1 illustrates the general format of a process control chart. The upper
and lower control limits define the normal operating region for the process.
Upper control limit

UCL
Average
LCL
Lower control limit
Figure 6.1. Process control chart format
The horizontal axis reflects the passage of time, or order of production. The
vertical axis corresponds to the variable of interest. There are number of
different types of process control charts. We will consider the most
commonly used process control charts
_
x  chart
_
s  chart
p  chart
c  chart
186
6.3.1 Control charts for means and standard deviations
Let us consider a production process that yields an output whose

characteristic of interest can be measured on a continuum. It is necessary to
set up a quality control scheme for that process. It can be done by taking,
over time, a sequence of small samples of output. Often samples of four or
five are taken. The frequency of sample observations depends on the
characteristics of production process. Management is often interested in both
average performance of production process and the variability in process.
Sample means and standard deviations are used to track process
performance. Three measures used in the development of control charts for
means and standard deviations. They are: overall mean, the average sample
standard deviation, and the process standard deviation.
Assume that a sequence of K samples, each of n observations, is selected
_
over time from the production process. The sample means denoted x i for
_
i  1,2,3,......K can be graphed on an x  chart. The average of these
sample means is called the overall mean of all the sample observations
___
__ K _
x  x /K
i 1
i
The sample standard deviations denoted si for i  1,2,3,......K can be graphed

on an s  chart. The average sample standard deviation is
_ K _
s s / K
i 1
i
The process standard deviation,  is the standard deviation of the

population from which the samples were drawn, and it must be estimated
from the sample data.
Since the sample standard deviation si is based on n observations, it can be
shown that
E ( si )  c 4  
where c 4 is the number that can be computed as a function of the sample size
n. It follows that
_
E ( s)  c 4  
187
and hence that an unbiased estimate of the process standard deviation is
given by
^ _
  s/ c 4
The value of c 4 -control chart factor, can be found in Table 6.1. Table 6.1
lists values of c 4 , corresponding sample sizes from two to ten. It also
contains factors for other control charts that will be discussed throughout this
chapter.
Table6.1 Factors for control charts
N C4 A3 B3 B4
2 0.7879 2.659 0 3.267
3 0.8862 1.954 0 2.568
4 0.9213 1.628 0 2.266
5 0.9400 1.427 0 2.089
6 0.9515 1.287 0.030 1.970
7 0.9594 1.182 0.118 1.882
8 0.9650 1.099 0.185 1.815
9 0.9690 1.032 0.239 1.761
10 0.9727 0.975 0.284 1.716
_
To determine control limits for x -charts, we assume that the process has
been operating at a constant level of performance over the whole observation
period and, assume that all sample observations have been drawn from the
same normal distribution.
The sampling distribution is centered on the overall mean, and the value of
the overall mean determines the central line, called center line. Then, if
three-standard error limits are to be used, the control limits are
___ ___ ___
__ ^ __ _ __ _
x  3   / n  x  3  s/( c 4  n )  x  A3  s
where
A3  3 /( c 4  n )
188
_
Control chart for x  means
_
The x  chart is a time plot of the sequence of sample means.
The center line is
___
__
CL_  x
x
In addition, there are three-standard error control limits.
The lower control limit is
___
__ _
LCL_  x  A3  s
x
The upper control limit is
___
__ _
UCL_  x  A3  s
x
where the values of A3 are given in Table 6.1.
Example:
The accompanying table shows sample means and sample standard
deviations for a sequence of 10 samples of seven observations on a quality
characteristic of a product
_
Sample x s
1 145.2 2.3
2 139.2 3.1
3 146.3 2.1
4 138.2 1.9
5 141.2 2.4
6 144.3 2.2
7 140.1 3.1
8 139.9 2.3
9 145.5 2.7
10 143.3 2.8
_
a) Find the center line and lower and upper control limits for an x  chart.
_
b) Draw the x  chart.
189
Solution:
a) First of all, let us find overall mean and average of the sample standard
deviations are
___
145.2  139.2  .....  143.3

__
x   142.32
10
_
2.3  3.1  ....  2.8
s  2.49
10
The sample size is seven. So from table 6.1 we obtain A3  1.182 .
The central line is
___
__
CL_  x  142.32
x
___
__ _
LCL_  x  A3  s  142.32  1.182  2.49  139.38
x
___
__ _
UCL_  x  A3  s  142.32  1.182  2.49  145.26
x
_
b) Each of the individual sample means are plotted on x  chart in Figure 6.1.
Three of the values fall out side of the control limits and there seems a great
cause for concern. It is necessary to take action to correct the production
process.
145.26 UCL
`
144.32
`
142.32 CL
`
1 2 3 4 5 6 7 8 9 10
140.32
139.38 LCL
_
Figure 6.2. Control chart for x
190
Control chart for s  standard deviations
The s  chart is a time plot of the sequence of sample standard deviations.
The center line is
_
CL s  s
_
LCL s  B3  s
_
UCL s  B 4  s
where the values of B3 and B 4 are given in Table 6.1.
Example: Refer to the data of example above.
a) Find the center line and lower and upper control limits for an s  chart.
b) Draw the s  chart and discuss its features
Solution:
_
2.3  3.1  ....  2.8
a) s  2.49
10
The sample size is seven. So from table 6 of the appendix we obtain
B3  0.118 and B4  1.882
The center line is

_
CL s  s  2.49
_
LCL s  B3  s  0.118  2.49  0.29
_
UCL s  B 4  s  1.882  2.49  4.69
b) We plot each of the individual standard deviations on a control chart with
center line CL s  2.49 , lower control limit LCL s  0.29 , and upper control
limit, UCL s  4.69 . The s-chart will look like Figure 6.3.
191
4.87 UCL
3.49
3 6 8
2.49 CL
1 2 4 5 7 9 10
1.49
0.29 LCL
Figure 6.3. Control chart for s

The observed sample standard deviations are fall between lower and upper
control limits. From this point of view, the process is under control. But
since, the mean production process is out of control, the general production
process must be corrected.
6.4. Interpretation of control charts

Once control charts have been developed, they can be used to determine
whether the process is in control or out of control. The control charts are
used to provide signals that something has changed. There are four primary
signals indicating that a process might be out of control:
1. One or more points outside the upper or lower control limits
2. Nine or more points in a row above (or below) the center line
3. Six or more consecutive points moving in the same direction (increasing
or decreasing).
These signals reduce to variation called assignable-cause variation. This
variation is not random and is due to defects or problems in the
manufacturing process, such as operators using the machine incorrectly, raw
materials changing, etc. Assignable cause variations must be corrected in
order to maintain quality in the production process.
Remark:
1. Select Stat>Control charts>Select Xbar S
2. Enter variable location (for example, C1)
3. Enter subgroup size
4. Click OK.
192
Exercises
1. Data were collected on a quantitative measure with a sample of 6
observations for a sequence of thirty samples. 30 samples were collected,
and following results were found
___
__ _
x  42.3 ; s  4.2
_
a) Find the center line and lower and upper control limits for an x  chart.
b) Find the center line and lower and upper control limits for an s  chart.
2. Weights of samples of canned fruit were measured. Results were available
for a sequence of thirty samples, each of seven observations. The overall
mean of the sample observations was 192.6 grams, and the average sample
standard deviation was 5.42.
a) Use an unbiased estimator to find an estimate of the process standard
deviation.
_
b) Find the center line and lower and upper control limits for an x  chart.
c) Find the center line and lower and upper control limits for an s  chart.
3. The accompanying table shows sample means and sample standard
deviations for a sequence of 14 samples of eight observations on a quality
characteristic of a product
Sample x s
1 146.4 4.37
2 152.8 6.79
3 150.6 3.17
4 149.2 4.71
5 150.6 4.98
6 150.4 6.28
7 151.1 6.20
8 152.9 6.97
9 147.2 4.28
10 154.3 7.29
11 151.8 3.1
12 149.9 5.31
13 146.7 4.73
14 152.1 6.12
193
a) Find the overall mean of the sample observations.
b) Find the average sample standard deviation.
c) Use an unbiased estimator to find an estimate of the process standard
deviation.
_
d) Find the center line and lower and upper control limits for an x  chart.
_
e) Draw the x  chart and discuss its features.
f) Find the center line and lower and upper control limits for an s  chart.
g) Draw the s  chart and discuss its features.
4. Ten samples, each consisting of five automobile batteries, are tested for
strength. The means and sample standard deviations are given here.
Sample Mean Standard deviation

1 12.2 2.1
2 12.5 1.3
3 12.3 1.5
4 11.9 2.1
5 11.8 1.2
6 11.2 1.1
7 12.1 1.3
8 12.0 1.3
9 12.2 1.8
10 11.8 2.0
a) Find the overall mean of the sample observations.
b) Find the average sample standard deviation.
c) Use an unbiased estimator to find an estimate of the process standard
deviation.
_
d) Find the center line and lower and upper control limits for an x  chart.
_
e) Draw the x  chart and discuss its features.
f) Find the center line and lower and upper control limits for an s  chart.
g) Draw the s  chart and discuss its features.
194
Answers
1.a) CL_  42.3 ; LCL_  36.89 ; UCL_  47.71 ; b) CL s  4.2 ; LCLs  0.13 ;
x x x
^
UCL s  8.27 ;2.a)   5.65 ;b) CL_  192.6 ; LCL_  186.20 ; UCL_  199.00 ;
x x x
c) CL s  5.42 ; LCL s  0.64 ; UCLs  10.19 ; 3. a) 150.43; b) 5.307;
^
c)   5.50 ; d) CL_  150.43 ; LCL_  144.60 ; UCL_  156.26 ;
x x x
f) CLs  5.307 ; LCLs  0.98 ; UCLs  9.63 ; 4. a) 12.00; b) 1.57; c) 1.67;
d) CL_  12.00 ; LCL_  9.76 ; UCL_  14.24 ; f) CL s  1.57 ; LCL s  0.00 ;
x x x
UCL s  3.28 ;
6.5. Control charts for proportions
Now let us consider situations, where individual product items will

be checked to have conformed or not to have conformed to specifications.
Again, a sequence of samples is taken over time to control product quality.
Our interest is the proportion of nonconforming, or defective, items in each
sample. It is clear that, it is desirable that this proportion be as small as
possible, and any increasing trend over time should cause concern.
The p  chart is used to monitor the proportion of defective items. One
important difference between p  chart and charts of previous section is that
here much larger sizes are necessary. Any competently engineered
production process is not going to generate a large proportion of
nonconforming items. Therefore, to get a reasonable result, a relatively large
sample size is necessary.
Suppose that a sequence of K samples, each of n observations, is taken from
production process. The proportions of sample members not conforming to
conforming can be determined as
^ x
pi  i  1,2,3,....., K
ni
where x  number of not conforming items;
n  sample size
If the samples are the same size, the average of the sample proportions is
the overall proportion of nonconforming items. This is
195
^
_ n

pi
p
i 1 K
If sample sizes are not equal, then overall proportion of nonconforming

items is defined as
n

^
ni  p i
_
i 1
p n
ni 1
i
^
We know that individual sample proportions p i have sampling distribution
_
with mean estimated by p and standard deviation (standard error) given by
_ _
^ p (1  p )
p
n
_
Similar to the x  chart and s  chart, three-standard error limits will be used
for p  chart.
Control chart for p  proportions
The p  chart is a time plot of the sequence of sample proportions of
nonconforming items.
The central line is
_
CL p  p
_ _
_
p(1  p)
LCL p  p  3 
n
_ _
_
p(1  p)
UCL p  p  3 
n
196
Remark1:
If the lower control limit gets a negative value, which is impossible, we set
the lower control limit at zero.
Remark2:
_
Interpretation of p  chart is similar to the interpretation of x  chart and
s  chart.
Remark3:
1. Select Stat>Control charts>Select P
2. Enter variable location (for example, C1)
3. Enter subgroup size (200, 300 etc.). If sample sizes vary, corresponding
sample sizes should be in column (for example, C2)
4. Click OK.
Exercises
1. In the study of production process, 25 samples, each of 300 observations

were taken. The average of the sample proportions of nonconforming items
was 0.064. Find the center line and upper and lower control limits for p-
chart.
2. Data were collected from a process in which the factor of interest was
whether the finished item contained a particular attribute. A total of 30
samples were selected. The common sample size was 100 items. The total
number of nonconforming items was 270. Based on these data, compute the
central line, the upper and lower control limits for p-chart.
3. Samples of bowling balls are selected and screened to see whether there
are any defects on their surfaces. The number of defective balls recorded for
each sample is given here
Sample Size Number of defective balls
1 200 10
2 200 8
3 200 9
4 200 32
5 200 15
6 200 35
7 200 38
8 200 10
197
a) Find the center line and upper and lower control limits for p-chart.
b) Draw the p-chart and discuss its features.
4. Baseball caps are manufactured and checked to see whether the logos are
properly printed on them. The number of defective logos per sample is
shown below
Sample Size Number of defective balls

1 100 8
2 100 6
3 100 5
4 100 27
5 100 9
6 100 10
5. Proportions of nonconforming items in a sequence of 12 samples, each
400 observations, are given below
Sample ^ Sample ^
p p
1 0.061 7 0.068
2 0.060 8 0.036
3 0.043 9 0.064
4 0.051 10 0.056
5 0.037 11 0.046
6 0.042 12 0.039
Answers
1. CL p  0.064 ; LCL p  0.022 ; UCL p  0.106 ;2. CL p  0.090 ;
LCL p  0.004 ; UCL p  0.176 ;3.a) CL p  0.098 ; LCL p  0.035 ;
UCL p  0.161 ;4. a) CL p  0.108 ; LCL p  0.015 ; UCL p  0.202 ;
5. a) CL p  0.050 ; LCL p  0.017 ; UCL p  0.083 ;
198
6.6. Control charts for number of occurrences: c-chart
The p-chart just discussed is used when you select sample of items
and you determine the number of the sample items that possesses a specific
attribute of interest. Each item either has or does not have that attribute. In
practice often we meet the situations that involve attribute data but differ
from the p-chart. Each sampling unit could have one or more of the
attributes of interest. Number of attributes of interest, called the number of
occurrences , counts number of imperfections per item over time. This is
called a c-chart.
As with the other control charts studied in this chapter, some general
notations used for control charts for number of occurrences are necessary.
Assume that, a sequence of K items is inspected over time, and for
each item, the number of occurrences of some event, such as imperfections,
is recorded. These numbers of occurrences denoted ci for i  1,2,....K .
The sample mean of occurrences is
_ K

ci
c
i 1 K
A 3-sigma (3 standard deviation) control chart for the number of occurrences
(c-chart) can be constructed in the usual way:
Control chart for c- number of occurrences
The c  chart is a time plot of the number of occurrences over the time.
The central line is
_
CLc  c
_ _ _
LCL c  c  3  c if c  9
_
LCLc  0 if c  9
_ _
UCLc  c  3  c
199
Example:
Handheld calculators are manufactured and checked for defects. If a
calculator is not defective, it is packaged and shipped to a retail store. Any
defective calculators are repaired before they are shipped. Twelve of the
defective calculators are checked for the number of defects per calculator.
The numbers of defects per calculator are:
6, 3, 2, 5, 6, 7, 4, 3, 7, 8, 9 and 5
a) Find the central line and lower and upper limits for c-chart.
b) Draw the c-chart and discuss its features.
Solution:
_
a) First of all, let us find the mean number of defects per calculator, c .
_ K
6  3  2  5  6  7  4  3  7  8  9  5 65
K 
ci
c   5.42
i 1 12 12
_
CLc  c  5.42
The lower and upper control limits are
_
LCLc  0 since c  9
_ _
UCLc  c  3  c  5.42  3  5.42  12.40
b) Figure 6.4 represents c-chart
12.40 UCL
8.40
2 3 7 8
5.42 ` CL
1 4 5 6 9 10 11 12
2.40
0.00 LCL
Figure 6.4. Control chart for c
200
Since, all points fall within the upper and lower control limits, the process is
in control. That is, in the defective calculators, the number of defects per
calculator is not excessive.
Remark:
1. Select Stat>Control charts>Select C
2. Enter variable location (C1; C2; etc.)
3. Enter subgroup size (200, 300 etc.)
4. Click OK.
Exercises
1. Sixty minute cassette tapes are checked for defects. The number of defects
in each of 8 tapes is shown below
Tape 1 2 3 4 5 6 7 8
Number of defects 1 2 1 1 1 3 6 2
2. Workers are painting a large apartment building. An inspector checks

several walls for paint defects. The number of defects per wall is shown
below
Sample 1 2 3 4 5 6 7 8 9 10
Number of defects 6 12 9 8 8 6 12 10 11 8
3. A reader has very carefully read local paper for 15 weeks. For each
Sunday’s edition he has counted the number of typographical and spelling
errors. The results are shown below
201
Week Errors Week Errors
1 13 9 12
2 15 10 13
3 14 11 21
4 17 12 9
5 12 13 17
6 20 14 19
7 7 15 22
8 14
a) Find the sample mean number of errors for these 15 weeks.

b) Find the central line and lower and upper limits for c-chart.
c) Draw the c-chart and discuss its features.
4. In a large market, the number of complaints in a week was recorded over

16 weeks. The results are shown in accompanying table
Week Errors Week Errors

1 17 9 23
2 11 10 15
3 15 11 10
4 22 12 16
5 21 13 17
6 29 14 22
7 8 15 23
8 18 16 20
a) Find the sample mean number of complaints per week.

b) Find the central line and lower and upper limits for c-chart.
c) Draw the c-chart and discuss its features.
Answers
1. a) CLc  2.125; LCLc  0 ; UCLc  6.498 ;2. a) CLc  9.00; LCLc  0 ;
UCLc  18 ; 3. a)15; b) CL c  15; LCLc  3.381 ; UCLc  26.62 ;
4. a) 17.94; b) CLc  17.94; LCLc  5.232; UCLc  30.64 .
202
Chapter 7
Time series analysis and forecasting
7.1. Introduction to index numbers
7.1.1 Price index for a single item (Simple index number)
While analyzing time series data, decision maker often compares one
value measured at one point in time with other values measured at
different points in time. For example, a student may wish to compare
book price in 2005 with prices in previous years. A common procedure
for making relative comparisons is to begin by determining a base
period index to which all other data values can be compared. This kind
of index is called a price index for a single item. To form a price
index, one time period is chosen as a base and the price for all other
periods are expressed as a percentage of the base period price. If we
denote the price in the base period by p 0 , and the price in another
period by p1 , then, the price index for the another period is
p 
100   1 
 p0 
For example, the price of the house in 2001 was $73.000 and in 2004
the price was $121.000. If we take 2001 as a base period, then
p0  $73.000 and p1  $121.000 . The price index is
p  121 .000
100   1   100   165 .75%
 0
p 73.000
It means that the price of house increased by 65.75% in 2004
compared with 2001.
7.1.2 Unweighted aggregate price index
While price index for a single item can be used to identify price
changes of a single item, we often are interested in the general price
change for a group of items taken as a whole. An unweighted
aggregate price index can be developed by simply summing the unit
203
prices in the time of interest and dividing this sum by the sum of the
unit prices in the base year. Suppose that we have K number of items.
Let
p0i  denote the price of the i th item in base period
p1i  denote the price of the i th item in period t
The unweighted aggregate price index in period t, denoted I t , is given
by
 K 



p1i 

I t  100   iK1 


 i 1
p 0i 

Example:
The following table lists prices of three different types of cars in
different years
Year Car types

A B C
2000 15.000 32.000 25.000
2001 16.000 34.000 26.000
2002 18.000 35.000 31.000
2003 21.000 37.000 32.000
2004 23.000 39.000 35.000
th
Use 2000 year as a base year, find and interpret unweighted
aggregate price index.
Solution:
Year Car types
A
p1
B
p2
C
p3
Sum
100 
p1 
p0 
2000 15.000 32.000 25.000 72.000 100%
2001 16.000 34.000 26.000 76.000 105.56%
2002 18.000 35.000 31.000 84.000 116.67 %
2003 21.000 37.000 32.000 90.000 125 %
2004 23.000 39.000 35.000 97.000 134.47%
204
For example, 116.67 % indicates that as a group, prices for the three
types of cars in 2002 year have increased by 16.67 % since 2000 year.
7.1.3 A weighted aggregate price index
The philosophy behind the weighted aggregate index is that each item
in the group should be weighted according to its importance. In most
cases the quantity of usage provides the best measure of importance.
Suppose that we have a group of K items. Let
qoi  be the quantity of i th item in the base period;
p0i  be the price of i th item in the base period;
p1i  be the price of i th item in the period of t
The Laspeyres price index for the period t is given by
 K 


q 0i  p1i 

I p  100   iK1 
 

 i 1
q 0i  p 
0i 

Example:
The company sells beer, wine, and soft drinks. Prices and quantity are
shown below
Quantity Unit price ($)
Item (bottles) 2000 2004
Beer 35 000 5.1 5.7
Wine 5 000 35.0 37.0
Soft drink 50 000 2.80 3.5
Compute a weighted aggregate price index for the company sales in
2004, with 2000 as the base period.
Solution:
From the information
q01  35000 ; q 02  5000 ; q03  50000
p01  5.1 ; p02  35.0 ; p03  2.80
p11  5.7 ; p12  37.0 ; p13  3.5
205
5.7  35000  37.0  5000  3.5  50000 559500
Ip   100   100  113.37%
5.1  35000  35.0  5000  2.80  50000 493500
It means in year 2004, at the prices prevailing, the total income of

company from sales in the base period would have been increased by
13.37%.
7.1.4 A weighted aggregate quantity index
Suppose that we have a group of K items. Let

poi  be the price of i th item in the base period;
p1i  be the price of i th item in the base period;
q0i  be the price of i th item in the period of t
The Laspeyres quantity index for the period t is given by
 K 


 p 0i  q1i 

I q  100   iK1 

 
 i 1
p 0i  q 0i 

7.2. Commonly used index numbers
a) Consumer Price Index
To most of us, inflation has come to mean increased prices and

less purchasing power of dollar. The Consumer Price Index (CPI)
attempts to measure the overall changes in retail prices of goods and
services. The CPI uses a “market basket” of goods and services used
by a typical wage earner living in a certain area. The CPI based on
items grouped into seven categories, including food, housing, clothing,
transportation, medical care, entertainment, and miscellaneous items.
206
b) Producer Price Index
The Producer Price Index (PPI), measures the monthly
changes in prices in primary markets in the US. The index is based on
prices for the transaction of each product in nonretail markets. All
commodities sold in commercial transactions in these markets are
represented, including those imported for sale. One of the common
uses of this index is as leading indicator of the future trend of
consumer prices and the cost of living An increase in the PPI reflects
producer price increases that will eventually be passed on to the
consumer through higher retail prices.
c) Dow Jones Average
The Dow Jones averages are indexes which are designed to

show price trends and movements on the New York Stock Exchange.
The best known of the Dow Jones indexes is the Dow Jones Industrial
Average, which is based on common-stock prices of 30 industrial
stocks. It is a weighted average of these stock prices where the weighs
are revised from time to time to adjust for stock splits and switching of
companies in the index.
7.3. Deflating a series by price indexes
Many business and economic series reported over time, such

as company sales, industry sales, and inventories, are measured in
dollar amounts. These time series often show an increasing growth
pattern over time which is generally interpreted as showing an increase
in the physical volume associated with these activities. In periods
where price changes are significant the changes in the dollar amounts
may be very misleading unless we are able to adjust the time series to
eliminate the price-change effect. Whenever we remove the price-
increase effect from the time series, we say we are deflating the
series.
In the area of personal income and salaries we often hear discussions
concerning issues such as “real salaries” or the “purchasing power”.
These concepts are based on the notion of deflating salary index. For
example, the following table shows monthly salary income of factory
workers for the past 5 years.
207
Year Salary ($) CPI (2000 base)
2000 490 100
2001 540 105
2002 585 113
2003 640 122
2004 700 138
At first glance, we see sharply increasing trend in monthly salaries,
with excellent growth from $490 to $700. Should the factory workers
be pleased with this growth in salary? Perhaps yes, but on the other
hand, if the cost of living has increased just as fast as salary, maybe the
answer is no. If we can compare purchasing power of the $490 in 2000
with the purchasing power of the $700 in 2004, we will have a better
idea of the relative improvement in salaries. Table above also shows
the Consumer Price Index (CPI) for the period 2000-2004. Here we
use 2000 as the base for CPI. With these data we will see how the CPI
can be used to deflate the index of monthly salaries. In effect we will
be removing the consumer price increases from power of salaries in an
attempt to measure the change in purchasing power of the wages.
The calculations used to deflate the salaries are not difficult.
The deflated series is found by dividing the monthly salary in each
year by the corresponding value of CPI
Year Deflated salary

2000 ($490/100) ∙100=$490
2001 ($540/105) ∙100=$514.3
2002 ($585/113) ∙100=$517.7
2003 ($640/122) ∙100=$524.6
2004 ($700/138) ∙100=$507.2
What does deflated series of salaries tells us about the “real salary” or
“purchasing power” of workers during 2000-2004? In terms of 2000
dollars, the monthly salary has risen from $490 to $507.2 or
approximately 3.5% In fact, after we remove the price increase effect
we see that factory workers are doing little more than keeping even
with the inflationary price increases of the period. Thus, we see that
the advantage of using price indexes to deflate a series is that we have
a clearer picture of the real dollar changes that are occurring.
208
Exercises
1. Consider the following revenue data for the past 8 years

Year Revenue (in millions of $)
1 18.2
2 22.6
3 25.2
4 32.6
5 38.6
6 39.7
7 40.2
8 41.3
Use year 1 as a base year to construct a relative index showing how
revenues have increased.
2. The accompanying table shows monthly salaries over 5 years for
three types of employees in a small company
Year Economist Manager Clerk
1 450 420 320
2 490 510 350
3 520 550 390
4 560 600 400
5 600 650 420
Take year 1 as a base. In that year there were 20 economists, 15
managers and 5 clerks.
a) Find the unweighted index of monthly salary rates.
b) Find the Laspeyres index for monthly salary rates.
3. Company produces three types of items: A, B, and C. The
beginning-year cost per item, the ending year cost per item, and the
number of items sold in the beginning-year period are shown below:
Items Beginning Ending year Number of

year cost cost items sold
A 2.50 3.95 25
B 8.75 9.90 15
C 0.99 0.95 60
a) Compute the price index for single items.
209
b) Compute a weighted aggregate Laspeyres price index. What is your
interpretation of this index value?
4. Total personal income for the 5 years 1995 to 1999 as follows
Year Total personal income (In millions of dollars)
1995 1200
1996 1450
1997 1650
1998 1800
1999 2050
Use following Consumer Price Index below
Year CPI
1995 100
1996 110
1997 118
1998 122
1999 130
to deflate the personal income series. What has been the percent
increase in “real personal income” from 1995 to 1999? Sketch actual
and real personal incomes and interpret your results.
5. The following table reports total inventories of all companies for
the 5 years 1990 to 1994 as follows
Year Total inventories (In billions of dollars)
1990 155
1991 163
1992 178
1993 198
1994 227
Use following Producers Price Index below
Year CPI
1990 100
1991 105
1992 113
1993 122
1994 138
to deflate this series. Sketch the real and actual total inventories and
interpret your findings.
210
Answers
1. 100%; 124.18%; 138.46%; 179.12%; 212.09%; 218.13%; 220.88%;
226.92%; 2. a) 100%; 113.45%; 122.69%; 131.09; 140.34%; b) 100%;
113.61%; 121.89%; 131.36%; 141.12%; 3. a) 158%; 113%; 96%; b)
120%; 4. 1200; 1318.2; 1398.3; 1475.4; 1576.9; 5. 155; 155; 158; 162;
164.
7.4 A nonparametric test for randomness
In the process of analyzing time series the first step is to

consider a test for randomness. We will consider the runs test, which
is a nonparametric test.
7.4.1. The runs test for the small sample sizes
To demonstrate this test, we will consider series with even number of

observations. Let us consider a series of 14 observations. The data are
shown below
N Value N Value
1 99 8 90
2 92 9 109
3 100 10 121
4 76 11 119
5 123 12 89
6 27 13 84
7 78 14 129
First of all we write data in ascending order and find the median.
th th
 n  1 15  7 th  8 th 92  99
Median     2    95.5
 2    2 2
The run test developed here separates the observations into a subgroup
above the median and a subgroup below the median. Then letting a
“  ”denote observations above the median and a “–“ denote
observations below the median we find the following pattern over the
sequence

211
This sequence consists of a run of one “  ”, followed by a one run of
one “-“, a run of one “ +”, a run of one “-“, a run of one “-“, a run of
three “-“, a run of three “+”, a run of two “-“ , and one run of “+”. In
total there are R  9 runs.
The null hypothesis is that the series is a set of random variables. The
table 7 in the Appendix gives the smallest significance level against
which this null hypothesis can be rejected against the alternative of
positive association between adjacent observations, as a function of
R and n.
If the alternative hypothesis is two-sided hypothesis on randomness,
the significance level must be doubled if it is less than 0.5.
Alternatively, if the significance level,  , read from table is greater
than 0.5, the corresponding significance level for the test against the
two sided alternative is 2(1   ) .
In our case, n  14 , and R  9 . From table in the appendix we see that
for n  14 observations, the probability under the null hypothesis of
finding 9 or fewer runs is 0.791. Therefore, the null hypothesis of
randomness can only be rejected against the alternative hypothesis of
positive association between adjacent observations at the 79.1%
significance level. We have not found strong evidence to reject the null
hypothesis that series are randomness.
7.4.2 The run test for the large sample sizes
If the sample size is large (n  20) , the distribution of the runs under
the null hypothesis can be approximated by a normal distribution.
Under the null hypothesis the random variable
n
R  1
Z 2
n  2n
2
4( n  1)
has a standard normal distribution. In formula above, R, defines the
number of runs, as the number of sequences above or below the
median.
We want to test the null hypothesis
H 0 : The series is random
212
1) If the alternative hypothesis is positive association between adjacent
observations, the decision rule is
n
R  1
Reject H 0 if Z  2   z
n  2n
2
4(n  1)
2) If the alternative hypothesis is that series are nonrandom, then the
decision rule is
n n
R  1 R  1
Reject H 0 if Z  2   z / 2 or Z  2  z / 2
n  2n
2
n  2n
2
4(n  1) 4( n  1)
Remark:
1. Select Stat>Nonparametrics>Runs test
2. Enter time series variable (for example, C1)
3. Select  Aboveand below
4. Insert value of the median
5. Select “generate forecasts”
6. Click OK.
Exercises
1. The following table shows country’s industrial production index
over 14 years.
Year Index Year Index

1 65 8 83
2 74 9 88
3 80 10 98
4 86 11 100
5 89 12 99
6 87 13 104
7 89 14 112
Test this series for randomness using the run test.
213
2. The following table shows 24 annual observations on sale of certain
brand of product
Year Sales Year Index

1 857 13 995
2 690 14 1234
3 710 15 987
4 839 16 653
5 858 17 345
6 791 18 674
7 768 19 980
8 478 20 945
9 658 21 783
10 751 22 342
11 723 23 456
12 567 24 610
Use the large- sample variant of the runs test to test this series for
randomness.
3. The following table shows annual return on a stock market index
over 14 years.
Year Return % Year Return %

1 -8.9 8 3.5
2 5.6 9 32.6
3 18.4 10 19.5
4 34.5 11 3.4
5 -7.8 12 16.6
6 23.4 13 32.1
7 19.6 14 0.3
Test for randomness using runs test.
214
4. The table shows earnings per share of a company over a period of
28 years.
Year Earnings Year Earnings Year Earnings

1 49.3 11 19.6 21 9.5
2 34.5 12 25.7 22 19.5
3 18.3 13 29.1 23 45.2
4 23.6 14 37.5 24 76.6
5 37.7 15 48.3 25 72.1
6 31.3 16 42.1 26 67.7
7 17.9 17 36.6 27 87.3
8 12.2 18 30.1 28 46.5
9 18.0 19 21.3
10 21.9 20 19.3
Use the large-sample variant of the runs test to test this series for
randomness.
Answers
1. Median=88.5; R  6 ; p  value  2  (0.209)  0.481 ; Fail to reject
H 0 at 10% level; 2. Median = 737; R  10 ; Z  1.2523;
p-value = 1  0.8944  0.1056 ; Reject H 0 at level above 10.56%;
3. Median = 17.5; R  9 ; p- value = 0.791; Fail to reject H 0 at 10%
level; 4. Median = 30.7; R  7 ; Z  3.0813 ; p- value =
= (1  0.999)  0.001 ; Reject H 0 at 0.01%.
215
7.5. Components of time series
A critical aspect of managing any company is planning for

future. In fact, the long-run success of any company is closely related
to how well management is able to foresee the future and develop
appropriate strategies.
Let us suppose for a moment that we have been asked to provide
quarterly estimates of the sales volume for a particular product during
the coming 1-year period. We will certainly want to review the actual
sales data for the product in past periods. From these historical data we
can identify the general level of sales and determine whether or not
there is a long-term trend such as an increase or decrease in sales
volume over time. The historical sales data referred to here is called a
time series.
The classical time series model has been developed in an attempt to
explain the pattern or behavior of the data in a time series. The
classical model is based on the assumption that four separate
components
-Trend (T)
-Cyclical (C)
-Seasonal (S)
-Irregular (I)
taken together cause the time series to assume specific values. By
analyzing each of these four components separately, we hope to
identify the effect each of these components has had on the time series
in the past.
a) Trend component
Trend is defined as the long-term movement in a time series. In other
words, an increase or decrease in the values of a variable occurring
over a period of several years gives a trend.
b) Cyclical component
It is fact that most of time series show alternating sequences of points
below and above the trend line. The regular pattern of sequences of
points above and below the trend line is attributable to the cyclical
component of the time.
c) Seasonal component
Seasonal component in a time series are defined as the movement that
occur in a time series within one-year period. Many business activities,
216
such as production and sales, exhibit seasonal patterns over different
time periods of a year. For example, a manufacturer of snow removal
equipment and heavy clothing expects low sales activity in the spring
and summer months, with peak sales occurring in the fall and winter
months.
d) Irregular component
The random or chance variations in a time series are referred to as the
irregular component. For example, strikes and natural disasters such
as storms and earthquakes can cause unpredicted irregular movement
in the time series. Since this component accounts for the random
variability in the time series, it is unpredictable. Thus we can not
attempt to predict its impact on the time series in advance.
7.6. Moving averages
Sometimes the irregular component in time series may be so

large that it creates difficulties in interpretation of the time plot series.
In such cases we reduce this problem by using moving averages. We
can smooth any irregularities using moving averages, based on the
idea that any large irregular component at any point will have a
smaller effect if we average the point with its immediate neighbors.
This procedure is called a simple centered (2m  1)  point moving
average.
Let x1 , x 2 ,........, x n be n observations in a time series. A smoothed
series can be obtained by using a simple centered (2m  1)  point
moving averages
m

1
X t*  xt  j (t  m  1, m  2,......, n  m)
2m  1 J   m
For instance, if we want to find 3-point moving averages, then solve
2m  1  3
and find m  1 . If m  1 , then the first available data will be X 2* .
General X t* in this case is
x  xt  xt 1
X t*  t 1
3
If we set m  2 , then a 5-point moving averages will be formed as
217
xt 2  xt 1  xt  xt 1  xt  2
X t*  .
5
Example:
The following data show the sales over the past six years. Compute a
simple centered 3-point moving averages to smooth data
Year Sales
1999 2169
2000 3678
2001 2789
2002 4783
2003 1280
2004 2379
Solution:
Since we need to find 3-point moving averages then (2m  1)  3 , and
m 1.
x  xt  xt 1
Then X t*  t 1
3
Using formula above, we obtain
x  x2  x3 2169  3678  2789
X 2*  1   2878.67
3 3
x  x3  x4 3678  2789  4783
X 3*  2   3750
3 3
x  x4  x5 2789  4783  1280
X 4*  3   2950.67
3 3
x  x5  x6 4783  1280  2379
X 5*  4   2814
3 3
The original data and smoothed data are given below:
Year Sales X t*
1999 2169 --
2000 3678 2878.67
2001 2789 3750
2002 4783 2950.67
2003 1280 2814
2004 2379 --
218
Figure 7.1
The original data and smoothed data are graphed in Figure 7.1.
Remark:
1. Select Stat>Time series>Moving averages
3. Enter the moving average length
4. Click results and select summary table and results table
5. Click OK.
219
Exercises
1. The following table gives the gross domestic product (in billions of
dollars) of the country for the years 1990 through 1997
Year GDP
1990 1768.4
1991 1974.1
1992 2488.6
1993 3030.6
1994 3405.0
1995 4038.7
1996 4268.6
1997 4900.4
Compute a simple centered 3-point moving average for the GDP. Plot
the smoothed series and comment on your results.
2. The following table shows the year-end price of gold (in dollars)
over 10 consecutive years.
Year Price ($)

1 120
2 135
3 147
4 220
5 256
6 289
7 312
8 350
9 430
10 535
Compute a simple centered 5-point moving averages for the gold price
data.
Draw a time plot of the smoothed series and comment on your results.
220
3. The table shows earnings per share of a corporation over a period of
14 years.
Year Earnings Year Earnings
1 49.2 8 43.2
2 34.7 9 56.2
3 23.6 10 34.2
4 45.7 11 35.8
5 34.8 12 43.2
6 53.2 13 28.9
7 23.7 14 36.5
Compute a simple centered 7-point moving average series for the

corporate earnings. Draw a time plot of the smoothed series.
Answers
1. *; 2077.03; 2497.77; 2974.73; 3491.43; 3904.10; 4402.57; *; 2. *;

*; 175.6; 209.4; 244.8; 285.4; 327.4; 383.2; *; *; 3. *; *; *; 37.84;
36.99; 40.06; 41.57; 40.16; 41.36; 37.89; 39.71; *; *; *;
7.7. Exponential smoothing
Exponential smoothing is a forecasting technique that uses a

weighted average of past time series values to compute a smoothed
value that serves as a forecast for the next period in the time series.
The simple exponential smoothing model that we consider here is
based on the assumption that there is no significant trend or seasonal
effect in the data.
Suppose that a time series x1 , x2 ,........, xt is given. Our aim is to
predict the unknown future values xt 1 , xt  2 ,..... .The simple
exponential smoothing method of forecasting can be used to predict
future values of xt 1 , xt  2 ,.....
To apply this method we accept
^
x1  x1 .
^ ^
Then use x t    x t 1  (1   )  xt ;
0    1 ; t  2,3,4,......, n
221
where  is a smoothing constant whose value is fixed between 0
and 1. And standing at time n , we obtain forecasts of future values,
x t  h of the series by
^ ^
x t h  x h ; h  1,2,3,.....
Example:
The following table shows the price of a share of common stock for a
well-known computer firm over the past 8 weeks. The price shown is
the closing price on the same day of the week for 8 consecutive weeks.
Week Stock price

1 50
2 53
3 49
4 50
5 42
6 57
7 52
8 57
Use the method of simple exponential smoothing to obtain forecasts of

stock price over the next three weeks. Use a smoothing constant of
  0.4 . Graph the observed time series and the forecasts.
Solution:
As mentioned above, we let the smoothed value of the time series for
the first period equal the actual first value of the time series. So,
^
x1  x1  50
To illustrate the nature of the computation, we will use a smoothing
constant of   0.4 . Thus the smoothed value for period two becomes
^ ^
x 2    x1  (1   )  x2  0.4  50  0.6  53  51.8
The smoothed value for period 3 becomes
^ ^
x 3    x 2  (1   )  x3  0.4  51.8  0.6  49  50.12
Continuing this process in the end we obtain the following complete
set of smooth values shown in the table:
222
Week Stock price Smoothed time
series value
1 50 50.00
2 53 51.80
3 49 50.12
4 50 50.05
5 42 45.22
6 57 52.29
7 52 52.12
8 57 55.05
Now, let us use the results of exponential smoothing to develop a

forecast of the stock price for the 9 th ,10 th , and 11th weeks. The
assumption of the simple exponential smoothing is that the smoothed
value of the time series at one period provides the best estimate of the
time series for the next periods. Thus, the simple exponential
smoothing model forecast of the stock price for the following 3 weeks
is $55.05.
^ ^ ^
x 9  $55.05 ; x10  $55.05 ; x11  $55.05
Figure 7.2 shows the plot of smoothed values for the time series.
Price per share
60
55
50
Smoothed time series
45
1 2 3 4 5 6 7 8 9 10 11
Week
Figure 7.2. Exponential smoothing of stock price time series
with smoothing constant   0.4
223
Remark:
1. Select Stat>Time series>Single exponential smoothing
3. Select  Use
4. Insert (1   )
6. Number of forecasts: Insert an integer to indicate number of
forecasts you want.
7. Starting from origin: Enter a positive integer to specify a starting
point for the forecasts. For example, if you specify 4 forecasts and 10
for the origin, Minitab computes forecasts for periods 11, 12, 13, and
14, based on the level and trend components at period 10. If you leave
this space blank, Minitab generates forecasts from the end of the data.
8. Select Options
9. Select graphics, outputs
10. Enter 1 to the window “Use average of __ observations”
11. Click OK.
Exercises
1. The following time series shows the sales of a particular product
over the past 12 month.
Month Sales Month Sales

1 105 7 145
2 135 8 140
3 120 9 100
4 105 10 80
5 90 11 100
6 120 12 110
Use   0.3 to compute the simple exponential smoothing to obtain

forecasts of sales over the next three months. Graph the observed time
series and the forecasts.
224
2. The following table gives the gross domestic product (in billions of
dollars) of the country for the years 1990 through 1997.
Year GDP
1990 1768.4
1991 1974.1
1992 2488.6
1993 3030.6
1994 3405.0
1995 4038.7
1996 4268.6
1997 4900.4
Use the method of simple exponential smoothing, with a smoothing

constant of   0.7 , to obtain forecasts of GDP for the next two years.
Graph the observed time series and the forecasts.
3. The following table shows the year-end price of gold (in dollars)
over 10 consecutive years.
Year Price ($)

1 120
2 135
3 147
4 220
5 256
6 289
7 312
8 350
9 430
10 535
Use the method of simple exponential smoothing, with a smoothing
constant of   0.6 , to obtain forecasts of the price of gold in the next
four years. Graph the observed time series and the forecasts.
Answers
^ ^ ^ ^ ^ ^ ^
1. x13  x14  x15  106.053 ; 2. x 9  x10  3869.67 ; 3. x11  x12 
^ ^
 x13  x14  425.369 .
225
7.8. Double exponential smoothing
(Holt-Winters exponential smoothing forecasting model)
Many business forecasting procedures are based on extensions

of simple exponential smoothing. Double exponential smoothing (also
called the Holt-Winters exponential smoothing procedure) allows for
trend (seasonality also) in time series. In double exponential
smoothing, a second smoothing constant,  , is included to account
for the trend. We will consider a nonseasonal time series.
^
We denote the observed value as x t and x t as the estimate of the level.
The trend estimate is represented as Tt . In Holt-Winters exponential
smoothing forecasting model these two variables will be estimated as
^ ^
x t    ( x t 1  Tt 1 )  (1   )  xt ; (0    1)
^ ^
Tt    Tt 1  (1   )  ( x t  x t 1 ) ; (0    1)
where  and  are smoothing constants whose values lie between 0
and 1.
To apply double exponential smoothing, we begin the computations by
setting
^
T2  x 2  x1 and x 2  x2
Then the above equations are applied for t  3,4,......., n . Standing at
time n ,
We obtain forecasts of future values, x n  h , of the series by
^ ^
x n  h  x n  h  Tn ; (h  1,2,3,....)
Example:
The sales manager needs to determine a monthly forecast for the
number of men’s sweaters that will be sold so he can order an
appropriate amount of packing boxes. Data for the past 8 months are
given below:
226
Month Sales
1 145
2 165
3 175
4 149
5 167
6 156
7 176
8 195
Develop Holt-Winters double exponential smoothing model using

  0.2 and   0.3 as smoothing constants to forecast sales for the
next three months.
Solution:
The initial estimates of level and trend in month 2, are
^
x 2  x 2  165 and T2  x2  x1  165  145  20
This smoothing application will use   0.2 and   0.3 and the
equations
^ ^
x t  0.2  ( x t 1  Tt 1 )  0.8  xt ;
^ ^
Tt  0.3  Tt 1  0.7  ( x t  x t 1 )
Then for t  3 :
^ ^
x 3  0.2  ( x 2  T2 )  0.8  x3  0.2  (165  20)  0.8  175  177
and in addition
^ ^
T3  0.3  T2  0.7  ( x 3  x 2 )  0.3  20  0.7  (177  165)  14.4
Then for t  4 :
^ ^
x 4  0.2  ( x 3  T3 )  0.8  x4 
 0.2  (177  14.4)  0.8  149  157.48
and in addition
^ ^
T4  0.3  T3  0.7  ( x 4  x 3 )  0.3  14.4  0.7  (157.48  177 )  9.344
227
For t  5 :
^ ^
x 5  0.2  ( x 4  T4 )  0.8  x5 
 0.2  (157.48  9.344)  0.8  167  163.23
^ ^
T5  0.3  T4  0.7  ( x 5  x 4 )  0.3  (9.344)  0.7  (163.23  157.48)  1.22
For t  6 :
^ ^
x 6  0.2  ( x 5  T5 )  0.8  x6  0.2  (163.23  1.22)  0.8  156  157.7
^ ^
T6  0.3  T5  0.7  ( x 6  x 5 )  0.3  1.22  0.7  (157.7  163.23)  3.51
For t  7 :
^ ^
x 7  0.2  ( x 6  T6 )  0.8  x7  0.2  (157.7  3.51)  0.8  176  171.64
^ ^
T7  0.3  T6  0.7  ( x 7  x 6 )  0.3  (1.04)  0.7  (171.64  157.7)  8.7
For t  8 :
^ ^
x 8  0.2  ( x 7  T7 )  0.8  x8  0.2  (171.64  8.7)  0.8  195  192.2
^ ^
T8  0.3  T7  0.7  ( x 8  x 7 )  0.3  8.7  0.7  (192.2  171.64)  17.00
In general for h periods forecasting is
^ ^
x n  h  x n  h  Tn
The most recent level and trend estimates are
^
x 8  192.2 ; T8  17.00
Then the forecasts for the next three months are
^
x 9  192.2  1  17.00  209.2
^
x10  192.2  2  17.00  226.2
^
x11  192.2  3  17.00  243.2
The results of these calculations are shown below:
228
Month Sales ^ MINITAB
xt solution
1 145 -- 146.150
2 165 165 161.457
3 175 177 174.503
4 149 157.48 156.590
5 167 163.23 163.157
6 156 157.7 157.823
7 176 171.64 171.735
8 195 192.2 192.106
The last column of the table above contains MINITAB solution.
According to MINITAB solution the predictions are
^ ^ ^
x 9  209.004 ; x10  225.902 ; x11  242.800
The values calculated by the MINITAB program differ slightly from
those in the third column of the table above. The MINITAB
procedures will generally provide slightly better forecasts compared to
the more simplified procedure we have shown. The observed time
series and forecasts are shown in Figure 7.3.
Figure 7.3
229
Remark:
1. Select Stat>Time series>Double exponential smoothing
3. Select  Use
4. Enter 1   -for level
5. Enter 1   - for trend
7. Enter number of forecasting
8. Enter number of starting point for forecasting
9. Select Options
10. Select graphics, outputs
11. Click OK.
Exercises
1. The table shows the sales of a particular brand of automatic

dishwasher detergent in a store over a period of 7 weeks.
Week Sale
1 22
2 23
3 19
4 16
5 21
6 28
7 19
Use the Holt-Winters procedure, with smoothing constants   0.4 and
  0.5 to obtain forecasts over the next 5 weeks.
2. The following table shows manufacturing monthly earnings over 12
months.
Month Earnings Month Earnings

1 350 7 420
2 370 8 400
3 340 9 350
4 370 10 420
5 390 11 360
6 410 12 450
230
Use the Holt-Winters procedure, with smoothing constants   0.3 and
  0.4 to obtain forecasts for the next 3 months. Graph the data and
forecasts.
3. The following table shows percentage profit of a company over a
period of 8 years.
Year Profit margin

1 7.6
2 6.4
3 6.9
4 7.9
5 8.1
6 7.0
7 8.2
8 6.2
Find forecasts for the next three years, using the Holt-Winters
procedure, with smoothing constants   0.7 and   0.6 . Graph the
data and forecasts.
Answers
1. 22.31; 22.50; 22.68; 22.87; 23.06; 2. 449.01; 474.56; 500.12; 3.

7.29; 7.23; 7.17.
231
APPENDIX
232
Table 1: Cumulative distribution function of the standard normal distribution
z F (z ) z F (z ) z F (z ) z F (z ) z F (z ) z F (z )
.00 .5000
.01 .5040 .21 .5832 .41 .6591 .61 .7291 .81 .7910 1.01 .8438
.02 .5080 .22 .5871 .42 .6628 .62 .7324 .82 .7939 1.02 .8461
.03 .5120 .23 .5910 .43 .6664 .63 .7357 .83 .7967 1.03 .8485
.04 .5160 .24 .5948 .44 .6700 .64 .7389 .84 .7995 1.04 .8508
.05 .5199 .25 .5987 .45 .6736 .65 .7422 .85 .8023 1.05 .8531
.06 .5239 .26 .6026 .46 .6772 .66 .7454 .86 .8051 1.06 .8554
.07 .5279 .27 .6064 .47 .6803 .67 .7486 .87 .8078 1.07 .8577
.08 .5319 .28 .6103 .48 .6844 .68 .7517 .88 .8106 1.08 .8599
.09 .5359 .29 .6141 .49 .6879 .69 .7549 .89 .8133 1.09 .8621
.10 .5398 .30 .6179 .50 .6915 .70 .7580 .90 .8159 1.10 .8643
.11 .5438 .31 .6217 .51 .6950 .71 .7611 .91 .8186 1.11 .8665
.12 .5478 .32 .6255 .52 .6985 .72 .7642 .92 .8212 1.12 .8686
.13 .5517 .33 .6293 .53 .7019 .73 .7673 .93 .8238 1.13 .8708
.14 .5557 .34 .6331 .54 .7054 .74 .7704 .94 .8264 1.14 .8729
.15 .5596 .35 .6368 .55 .7088 .75 .7734 .95 .8289 1.15 .8749
.16 .5636 .36 .6406 .56 .7123 .76 .7764 .96 .8315 1.16 .8770
.17 .5675 .37 .6443 .57 .7157 .77 .7794 .97 .8340 1.17 .8790
.18 .5714 .38 .6480 .58 .7190 .78 .7823 .98 .8365 1.18 .8810
.19 .5753 .39 .6517 .59 .7224 .79 .7852 .99 .8389 1.19 .8830
.20 .5793 .40 .6554 .60 .7257 .80 .7881 1.00 .8413 1.20 .8849
233
1.21 .8869 1.46 .9279 1.71 .9564 1.96 .9750 2.21 .9864 2.46 .9931
1.22 .8888 1.47 .9292 1.72 .9573 1.97 .9756 2.22 .9868 2.47 .9932
1.23 .8907 1.48 .9306 1.73 .9582 1.98 .9761 2.23 .9871 2.48 .9934
1.24 .8925 1.49 .9319 1.74 .9591 1.99 .9767 2.24 .9875 2.49 .9936
1.25 .8944 1.50 .9332 1.75 .9599 2.00 .9772 2.25 .9878 2.50 .9938
1.26 .8962 1.51 .9345 1.76 .9608 2.01 .9778 2.26 .9881 2.51 .9940
1.27 .8980 1.52 .9357 1.77 .9616 2.02 .9783 2.27 .9884 2.52 .9941
1.28 .8997 1.53 .9370 1.78 .9615 2.03 .9788 2.28 .9887 2.53 .9943
1.29 .9015 1.54 .9382 1.79 .9633 2.04 .9793 2.29 .9890 2.54 .9945
1.30 .9032 1.55 .9394 1.80 .9641 2.05 .9798 2.30 .9893 2.55 .9946
1.31 .9049 1.56 .9406 1.81 .9649 2.06 .9803 2.31 .9896 2.56 .9948
1.32 .9066 1.57 .9418 1.82 .9656 2.07 .9808 2.32 .9898 2.57 .9949
1.33 .9082 1.58 .9429 1.83 .9664 2.08 .9812 2.33 .9901 2.58 .9951
1.34 .9099 1.59 .9441 1.84 .9671 2.09 .9817 2.34 .9904 2.59 .9952
1.35 .9115 1.60 .9452 1.85 .9678 2.10 .9821 2.35 .9906 2.60 .9953
1.36 .9131 1.61 .9463 1.86 .9686 2.11 .9826 2.36 .9909 2.61 .9955
1.37 .9147 1.62 .9474 1.87 .9693 2.12 .9830 2.37 .9911 2.62 .9956
1.38 .9162 1.63 .9484 1.88 .9699 2.13 .9834 2.38 .9913 2.63 .9957
1.39 .9177 1.64 .9495 1.89 .9706 2.14 .9838 2.39 .9916 2.64 .9959
1.40 .9192 1.65 .9505 1.90 .9713 2.15 .9842 2.40 .9918 2.65 .9960
1.41 .9207 1.66 .9515 1.91 .9719 2.16 .9846 2.41 .9920 2.66 .9961
1.42 .9222 1.67 .9525 1.92 .9726 2.17 .9850 2.42 .9922 2.67 .9962
1.43 .9236 1.68 .9535 1.93 .9732 2.18 .9854 2.43 .9925 2.68 .9963
1.44 .9251 1.69 .9545 1.94 .9738 2.19 .9857 2.44 .9927 2.69 .9964
234
1.45 .9265 1.70 .9554 1.95 .9744 2.20 .9861 2.45 .9929 2.70 .9965
Table 1: Cumulative distribution function of the standard normal distribution (continue)
z F (z ) z F (z ) z F (z ) z F (z ) z F (z ) z F (z ) z F (z )
2.71 .9966 2.91 .9982 3.11 .9991 3.31 .9995 3.51 .9998 3.71 .9999 3.91 1.0000
2.72 .9967 2.92 .9982 3.12 .9991 3.32 .9996 3.52 .9998 3.72 .9999 3.92 1.0000
2.73 .9968 2.93 .9983 3.13 .9991 3.33 .9996 3.53 .9998 3.73 .9999 3.93 1.0000
2.74 .9969 2.94 .9984 3.14 .9992 3.34 .9996 3.54 .9998 3.74 .9999 3.94 1.0000
2.75 .9970 2.95 .9984 3.15 .9992 3.35 .9996 3.55 .9998 3.75 .9999 3.95 1.0000
2.76 .9971 2.96 .9985 3.16 .9992 3.36 .9996 3.56 .9998 3.76 .9999 3.96 1.0000
2.77 .9972 2.97 .9985 3.17 .9992 3.37 .9996 3.57 .9998 3.77 .9999 3.97 1.0000
2.78 .9973 2.98 .9986 3.18 .9993 3.38 .9996 3.58 .9998 3.78 .9999 3.98 1.0000
2.79 .9974 2.99 .9986 3.19 .9993 3.39 .9997 3.59 .9998 3.79 .9999 3.99 1.0000
2.80 .9974 3.00 .9986 3.20 .9993 3.40 .9997 3.60 .9998 3.80 .9999
2.81 .9975 3.01 .9987 3.21 .9993 3.41 .9997 3.61 .9998 3.81 .9999
2.82 .9976 3.02 .9987 3.22 .9994 3.42 .9997 3.62 .9999 3.82 .9999
2.83 .9977 3.03 .9988 3.23 .9994 3.43 .9997 3.63 .9999 3.83 .9999
2.84 .9977 3.04 .9988 3.24 .9994 3.44 .9997 3.64 .9999 3.84 .9999
2.85 .9978 3.05 .9989 3.25 .9994 3.45 .9997 3.65 .9999 3.85 .9999
2.86 .9979 3.06 .9989 3.26 .9994 3.46 .9997 3.66 .9999 3.86 .9999
2.87 .9979 3.07 .9989 3.27 .9995 3.47 .9997 3.67 .9999 3.87 .9999
2.88 .9980 3.08 .9990 3.28 .9995 3.48 .9997 3.68 .9999 3.88 .9999
2.89 .9981 3.09 .9990 3.29 .9995 3.49 .9998 3.69 .9999 3.89 1.0000
2.90 .9981 3.10 .9990 3.30 .9995 3.50 .9998 3.70 .9999 3.90 1.0000
235
Table 2: Cut-off point of Student’s t distribution
 
0.100 0.050 0.025 0.010 0.005
1 3.078 6.314 12.706 31.821 63.657
2 1.886 2.920 4.303 6.965 9.925
3 1.638 2.353 3.182 4.541 5.841
4 1.533 2.132 2.776 3.747 4.604
5 1.476 2.015 2.571 3.365 4.032
6 1.440 1.943 2.447 3.143 3.707
7 1.415 1.895 2.365 2.998 3.499
8 1.397 1.860 2.306 2.896 3.355
9 1.383 1.833 2.262 2.821 3.250
10 1.372 1.812 2.228 2.764 3.169
11 1.363 1.796 2.201 2.718 3.160

12 1.356 1.782 2.179 2.681 3.055
13 1.350 1.771 2.160 2.650 3.012
14 1.345 1.761 2.145 2.624 2.977
15 1.341 1.753 2.131 2.602 2.947
16 1.337 1.746 2.120 2.583 2.921
17 1.333 1.740 2.110 2.567 2.898
18 1.330 1.734 2.101 2.552 2.878
19 1.328 1.729 2.093 2.539 2.861
20 1.325 1.725 2.086 2.528 2.845
21 1.323 1.721 2.080 2.518 2.831

22 1.321 1.717 2.074 2.508 2.819
23 1.319 1.714 2.069 2.500 2.807
24 1.318 1.711 2.064 2.492 2.797
25 1.316 1.708 2.060 2.485 2.787
26 1.315 1.706 2.056 2.479 2.779
27 1.314 1.703 2.052 2.473 2.771
28 1.313 1.701 2.048 2.467 2.763
29 1.311 1.699 2.045 2.462 2.756
30 1.310 1.697 2.042 2.457 2.750
40 1.303 1.684 2.021 2.423 2.704

60 1.296 1.671 2.000 2.390 2.660
 1.282 1.645 1.960 2.326 2.576
236
Table 3: Chi- square distribution
 
.995 .990 .975 .950 .900 .100 .050 .025 .010 .005
1 0.04393 0.03157 0.03982 0.02393 0.0158 2.71 3.84 5.02 6.63 7.88
2 0.0100 0.0201 0.0506 0.103 0.211 4.61 5.99 7.38 9.21 10.60
3 0.072 0.115 0.216 0.352 0.584 6.25 7.81 9.35 11.34 12.84
4 0.207 0.297 0.484 0.711 1.064 7.78 9.49 11.14 13.28 14.86
5 0.412 0.554 0.831 1.145 1.61 9.24 11.07 12.83 15.09 16.75
6 0.676 0.872 1.24 1.64 2.20 10.64 12.59 14.45 16.81 18.55
7 0.989 1.24 1.69 2.17 2.83 12.02 14.07 16.01 18.48 20.28
8 1.34 1.65 2.18 2.73 3.49 13.36 15.51 17.53 20.09 21.96
9 1.73 2.09 2.70 3.33 4.17 14.68 16.92 19.02 21.67 23.59
10 2.16 2.56 3.25 3.94 4.87 15.99 18.31 20.48 23.21 25.19
11 2.60 3.05 3.82 4.57 5.58 17.28 19.68 21.92 24.73 26.76
12 3.07 3.57 4.40 5.23 6.30 18.55 21.03 23.34 26.22 28.30
13 3.57 4.11 5.01 5.89 7.04 19.81 22.36 24.74 27.69 29.82
14 4.07 4.66 5.63 6.57 7.79 21.06 23.68 26.12 29.14 31.32
15 4.60 5.23 6.26 7.26 8.55 22.31 25.00 27.49 30.58 32.80
16 5.14 5.81 6.91 7.96 9.31 23.54 26.30 28.85 32.00 34.27
17 5.70 6.41 7.56 8.67 10.09 24.77 27.59 30.19 33.41 35.72
18 6.26 7.01 8.23 9.39 10.86 25.99 28.87 31.53 34.81 37.16
19 6.84 7.63 8.91 10.12 11.65 27.20 30.14 32.85 36.19 38.58
20 7.43 8.26 9.59 10.85 12.44 28.41 31.41 34.17 37.57 40.00
21 8.03 8.90 10.28 11.59 13.24 29.62 32.67 35.48 38.93 41.40
22 8.64 9.54 10.98 12.34 14.04 30.81 33.92 36.78 40.29 42.80
23 9.26 10.20 11.69 13.09 14.85 32.01 35.17 38.08 41.64 44.18
24 9.89 10.86 12.40 13.85 15.66 33.20 36.42 39.36 42.98 45.56
25 10.52 11.52 13.12 14.61 16.47 34.38 37.65 40.65 44.31 46.93
26 11.16 12.20 13.84 15.38 17.29 35.56 38.89 41.92 45.64 48.29
27 11.81 12.88 14.57 16.15 18.11 36.74 40.11 43.19 46.96 49.64
28 12.46 13.56 15.31 16.93 18.94 37.92 41.34 44.46 48.28 50.99
29 13.12 14.26 16.05 17.71 19.77 39.09 42.56 45.72 49.59 52.34
30 13.79 14.95 16.79 18.49 20.60 40.26 43.77 46.98 50.89 53.67
40 20.71 22.16 24.43 26.51 29.05 51.81 55.76 59.34 63.69 66.77
50 27.99 29.71 32.36 34.76 37.69 63.17 67.50 71.42 76.15 79.49
60 35.53 37.48 40.48 43.19 46.46 74.40 79.08 83.30 88.38 91.95
70 43.28 45.44 48.76 51.74 55.33 85.53 90.53 95.02 100.4 104.2
80 51.17 53.54 57.15 60.39 64.28 96.58 101.9 106.6 112.3 116.3
90 59.20 61.75 65.65 69.16 73.29 107.6 113.1 118.1 124.1 128.3
100 67.33 70.06 74.22 77.93 82.36 118.5 124.3 129.6 135.8 140.2
237
Table 4: Cutoff point of the distribution of the Wicoxon test
statistic
n 
0.005 0.01 0.025 0.05 0.10
4 0 0 0 0 1
5 0 0 0 1 3
6 0 0 1 3 4
7 0 1 3 4 6
8 1 2 4 6 9
9 2 4 6 9 11
10 4 6 9 11 15
11 6 8 11 14 18
12 8 10 14 18 22
13 10 13 18 22 27
14 13 16 22 26 32
15 16 20 26 31 37
16 20 24 30 36 43
17 24 28 35 42 49
18 28 33 41 48 56
19 33 38 47 54 63
20 38 44 53 61 70
238
Table 5: Cutoff point of the distribution of the Spearman’s
rank correlation coefficient
n 
0.05 0.025 0.01 0.005
5 .900 - - -
6 .829 .886 .943 -
7 .714 .786 .893 -
8 .643 .738 .833 .881
9 .600 .683 .783 .833
10 .564 .648 .745 .794
11 .523 .623 .736 .818

12 .497 .591 .703 .780
13 .475 .566 .673 .745
14 .457 .545 .646 .716
15 .441 .525 .623 .689
16 .425 .507 .601 .666

17 .412 .490 .582 .645
18 .399 .476 .564 .625
19 .388 .462 .549 .608
20 .377 .450 .534 .591
21 .368 .438 .521 .576

22 .359 .428 .508 .562
23 .351 .418 .496 .549
24 .343 .409 .485 .537
25 .336 .400 .475 .526
26 .329 .392 .465 .515

27 .323 .385 .456 .505
28 .317 .377 .448 .496
29 .311 .370 .440 .487
30 .305 .364 .432 .478
239
Table 2: Cutoff points for the F distribution
  0.05
 2  denominato r  1  numerator
1 2 3 4 5 6 7 8 9 10 12 15 20 24 30 40 60 120 
1 161.4 199.5 215.7 224.6 230.2 234.0 236.8 238.9 240.5 241.9 243.9 245.9 248.0 249.1 250.1 251.1 252.2 253.3 254.3
2 18.51 19.00 19.16 19.25 19.30 19.33 19.35 19.37 19.38 19.40 19.41 19.43 19.45 19.45 19.46 19.47 19.48 19.49 19.50
3 10.13 9.55 9.28 9.12 9.01 8.94 8.89 8.85 8.81 8.79 8.74 8.70 8.66 8.64 8.62 8.59 8.57 8.55 8.53
4 7.71 6.94 6.59 6.39 6.26 6.16 6.09 6.04 6.00 5.96 5.91 5.86 5.80 5.77 5.75 5.72 5.69 5.66 5.63
5 6.61 5.79 5.41 5.19 5.05 4.95 4.88 4.82 4.77 4.74 4.68 4.62 4.56 4.53 4.50 4.46 4.43 4.40 4.36
6 5.99 5.14 4.76 4.53 4.39 4.28 4.21 4.15 4.10 4.06 4.00 3.94 3.87 3.84 3.81 3.77 3.74 3.70 3.67
7 5.59 4.74 4.35 4.12 3.97 3.87 3.79 3.73 3.68 3.64 3.57 3.51 3.44 3.41 3.38 3.34 3.30 3.27 3.23
8 5.32 4.46 4.07 3.84 3.69 3.58 3.50 3.44 3.39 3.35 3.28 3.22 3.15 3.12 3.08 3.04 3.01 2.97 2.93
9 5.12 4.26 3.86 3.63 3.48 3.37 3.29 3.23 3.18 3.14 3.07 3.01 2.94 2.90 2.86 2.83 2.79 2.75 2.71
10 4.96 4.10 3.71 3.48 3.33 3.22 3.14 3.07 3.02 2.98 2.91 2.85 2.77 2.74 2.70 2.66 2.62 2.58 2.54
11 4.84 3.98 3.59 3.36 3.20 3.09 3.01 2.95 2.90 2.85 2.79 2.72 2.65 2.61 2.57 2.53 2.49 2.45 2.40
12 4.75 3.89 3.49 3.26 3.11 3.00 2.91 2.85 2.80 2.75 2.69 2.62 2.54 2.51 2.47 2.43 2.38 2.34 2.30
13 4.67 3.81 3.41 3.18 3.03 2.92 2.83 2.77 2.71 2.67 2.60 2.53 2.46 2.42 2.38 2.34 2.30 2.25 2.21
14 4.60 3.74 3.34 3.11 2.96 2.85 2.76 2.70 2.65 2.60 2.53 2.46 2.39 2.35 2.31 2.27 2.22 2.18 2.13
15 4.54 3.68 3.29 3.06 2.90 2.79 2.71 2.64 2.59 2.54 2.48 2.40 2.33 2.29 2.25 2.20 2.16 2.11 2.07
16 4.49 3.63 3.24 3.01 2.85 2.74 2.66 2.59 2.54 2.49 2.42 2.35 2.28 2.24 2.19 2.15 2.11 2.06 2.01
17 4.45 3.59 3.20 2.96 2.81 2.70 2.61 2.55 2.49 2.45 2.38 2.31 2.23 2.19 2.15 2.10 2.06 2.01 1.96
18 4.41 3.55 3.16 2.93 2.77 2.66 2.58 2.51 2.46 2.41 2.34 2.27 2.19 2.15 2.11 2.06 2.02 1.97 1.92
19 4.38 3.52 3.13 2.90 2.74 2.63 2.54 2.48 2.42 2.38 2.31 2.23 2.16 2.11 2.07 2.03 1.98 1.93 1.88
20 4.35 3.49 3.10 2.87 2.71 2.60 2.51 2.45 2.39 2.35 2.28 2.20 2.12 2.08 2.04 1.99 1.95 1.90 1.84
21 4.32 3.47 3.07 2.84 2.68 2.57 2.49 2.42 2.37 2.32 2.25 2.18 2.10 2.05 2.01 1.96 1.92 1.87 1.81
22 4.30 3.44 3.05 2.82 2.66 2.55 2.46 2.40 2.34 2.30 2.23 2.15 2.07 2.03 1.98 1.94 1.89 1.84 1.78
23 4.28 3.42 3.03 2.80 2.64 2.53 2.44 2.37 2.32 2.27 2.20 2.13 2.05 2.01 1.96 1.91 1.86 1.81 1.76
240
1 2 3 4 5 6 7 8 9 10 12 15 20 24 30 40 60 120 
24 4.26 3.40 3.01 2.78 2.62 2.51 2.42 2.36 2.30 2.25 2.18 2.11 2.03 1.98 1.94 1.89 1.84 1.79 1.73
25 4.24 3.39 2.99 2.76 2.60 2.49 2.40 2.34 2.28 2.24 2.16 2.09 2.01 1.96 1.92 1.87 1.82 1.77 1.71
26 4.23 3.37 2.98 2.74 2.59 2.47 2.39 2.32 2.27 2.22 2.15 2.07 1.99 1.95 1.90 1.85 1.80 1.75 1.69
27 4.21 3.35 2.96 2.73 2.57 2.46 2.37 2.31 2.25 2.20 2.13 2.06 1.97 1.93 1.88 1.84 1.79 1.73 1.67
28 4.20 3.34 2.95 2.71 2.56 2.45 2.36 2.29 2.24 2.19 2.12 2.04 1.96 1.91 1.87 1.82 1.77 1.71 1.65
29 4.18 3.33 2.93 2.70 2.55 2.43 2.35 2.28 2.22 2.18 2.10 2.03 1.94 1.90 1.85 1.81 1.75 1.70 1.64
30 4.17 3.32 2.92 2.69 2.53 2.42 2.33 2.27 2.21 2.16 2.09 2.01 1.93 1.89 1.84 1.79 1.74 1.68 1.62
40 4.08 3.23 2.84 2.61 2.45 2.34 2.25 2.18 2.12 2.08 2.00 1.92 1.84 1.79 1.74 1.69 1.64 1.58 1.51
60 4.00 3.15 2.76 2.53 2.37 2.25 2.17 2.10 2.04 1.99 1.92 1.84 1.75 1.70 1.65 1.59 1.53 1.47 1.39
120 3.92 3.07 2.68 2.45 2.29 2.17 2.09 2.02 1.96 1.91 1.83 1.75 1.66 1.61 1.55 1.50 1.43 1.35 1.25
 3.84 3.00 2.60 2.37 2.21 2.10 2.01 1.94 1.88 1.83 1.75 1.67 1.57 1.52 1.46 1.39 1.32 1.22 1.00
  0.01
 2  denominato r  1  numerator
1 2 3 4 5 6 7 8 9 10 12 15 20 24 30 40 60 120 
1 4052 4999 5403 5625 5764 5859 5928 5982 6022 6056 6106 6157 6209 6235 6261 6287 6313 6339 6366
2 98.50 99.00 99.17 99.25 99.30 99.33 99.36 99.37 99.39 99.40 99.42 99.43 99.45 99.46 99.47 99.47 99.48 99.48 99.50
3 34.12 30.82 29.46 28.71 28.24 27.91 27.67 27.49 27.35 27.23 27.05 26.87 26.69 26.60 26.50 26.41 26.32 26.22 26.13
4 21.20 18.00 16.69 15.98 15.52 15.21 14.98 14.80 14.66 14.55 14.37 14.20 14.02 13.93 13.84 13.75 13.65 13.56 13.46
5 16.26 13.27 12.06 11.39 10.97 10.67 10.46 10.29 10.16 10.05 9.89 9.72 9.55 9.47 9.38 9.29 9.20 9.11 9.02
6 13.75 10.92 9.78 9.15 8.75 8.47 8.26 8.10 7.98 7.87 7.72 7.56 7.40 7.31 7.23 7.14 7.06 6.97 6.88
7 12.25 9.55 8.45 7.85 7.46 7.19 6.99 6.84 6.72 6.62 6.47 6.31 6.16 6.07 5.99 5.91 5.82 5.74 5.65
8 11.26 8.65 7.59 7.01 6.63 6.37 6.18 6.03 5.91 5.81 5.67 5.52 5.36 5.28 5.20 5.12 5.03 4.95 4.86
9 10.56 8.02 6.99 6.42 6.06 5.80 5.61 5.47 5.35 5.26 5.11 4.96 4.81 4.73 4.65 4.57 4.48 4.40 4.31
10 10.04 7.56 6.55 5.99 5.64 5.39 5.20 5.06 4.94 4.85 4.71 4.56 4.41 4.33 4.25 4.17 4.08 4.00 3.91
11 9.65 7.21 6.22 5.67 5.32 5.07 4.89 4.74 4.63 4.54 4.40 4.25 4.10 4.02 3.94 3.86 3.78 3.69 3.60
12 9.33 6.93 5.95 5.41 5.06 4.82 4.64 4.50 4.39 4.30 4.16 4.01 3.86 3.78 3.70 3.62 3.54 3.45 3.36
241
1 2 3 4 5 6 7 8 9 10 12 15 20 24 30 40 60 120 
13 9.07 6.70 5.74 5.21 4.86 4.62 4.44 4.30 4.19 4.10 3.96 3.82 3.66 3.59 3.51 3.43 3.34 3.25 3.17
14 8.86 6.51 5.56 5.04 4.69 4.46 4.28 4.14 4.03 3.94 3.80 3.66 3.51 3.43 3.35 3.27 3.18 3.09 3.00
15 8.68 6.36 5.42 4.89 4.56 4.32 4.14 4.00 3.89 3.80 3.67 3.52 3.38 3.29 3.21 3.13 3.05 2.96 2.87
16 8.53 6.23 5.29 4.77 4.44 4.20 4.03 3.89 3.78 3.69 3.55 3.41 3.26 3.18 3.10 3.02 2.93 2.84 2.75
17 8.40 6.11 5.18 4.67 4.34 4.10 3.93 3.79 3.68 3.59 3.46 3.31 3.16 3.08 3.00 2.92 2.83 2.75 2.65
18 8.29 6.01 5.09 4.58 4.25 4.01 3.84 3.71 3.60 3.51 3.37 3.23 3.08 3.00 2.92 2.84 2.75 2.66 2.57
19 8.18 5.93 5.01 4.50 4.17 3.94 3.77 3.63 3.52 3.43 .3.30 3.15 3.00 2.92 2.84 2.76 2.67 2.58 2.49
20 8.10 5.85 4.94 4.43 4.10 3.87 3.70 3.56 3.46 3.37 3.23 3.09 2.94 2.86 2.78 2.69 2.61 2.52 2.42
21 8.02 5.78 4.87 4.37 4.04 3.81 3.64 3.51 3.40 3.31 3.17 3.03 2.88 2.80 2.72 2.64 2.55 2.46 2.36
22 7.95 5.72 4.82 4.31 3.99 3.76 3.59 3.45 3.35 3.26 3.12 2.98 2.83 2.75 2.67 2.58 2.50 2.40 2.31
23 7.88 5.66 4.76 4.26 3.94 3.71 3.54 3.41 3.30 3.21 3.07 2.93 2.78 2.70 2.62 2.54 2.45 2.35 2.26
24 7.82 5.61 4.72 4.22 3.90 3.67 3.50 3.36 3.26 3.17 3.03 2.89 2.74 2.66 2.58 2.49 2.40 2.31 2.21
25 7.77 5.57 4.68 4.18 3.85 3.63 3.46 3.32 3.22 3.13 2.99 2.85 2.70 2.62 2.54 2.45 2.36 2.27 2.17
26 7.72 5.53 4.64 4.14 3.82 3.59 3.42 3.29 3.18 3.09 2.96 2.81 2.66 2.58 2.50 2.42 2.33 2.23 2.13
27 7.68 5.49 4.60 4.11 3.78 3.56 3.39 3.26 3.15 3.06 2.93 2.78 2.63 2.55 2.47 2.38 2.29 2.20 2.10
28 7.64 5.45 4.57 4.07 3.75 3.53 3.36 3.23 3.12 3.03 2.90 2.75 2.60 2.52 2.44 2.35 2.26 2.17 2.06
29 7.60 5.42 4.54 4.04 3.73 3.50 3.33 3.20 3.09 3.00 2.87 2.73 2.57 2.49 2.41 2.33 2.23 2.14 2.03
30 7.56 5.39 4.51 4.02 3.70 3.47 3.30 3.17 3.07 2.98 2.84 2.70 2.55 2.47 2.39 2.30 2.21 2.11 2.01
40 7.31 5.18 4.31 3.83 3.51 3.29 3.12 2.99 2.89 2.80 2.66 2.52 2.37 2.29 2.20 2.11 2.02 1.92 1.80
60 7.08 4.98 4.13 3.65 3.34 3.12 2.95 2.82 2.72 2.63 2.50 2.35 2.20 2.12 2.03 1.94 1.84 1.73 1.60
120 6.85 4.79 3.95 3.48 3.17 2.96 2.79 2.66 2.56 2.47 2.34 2.19 2.03 1.95 1.86 1.76 1.66 1.53 1.38
 6.63 4.61 3.78 3.32 3.02 2.80 2.64 2.51 2.41 2.32 2.18 2.04 1.88 1.79 1.70 1.59 1.47 1.32 1.00
242
Table 7. Cumulative distribution function of the runs test statistic
n R
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
6 .100 .300 .700 .900 1.00
8 .029 .114 .371 .629 .886 .971 1.00
10 .008 .040 .167 .357 .643 .833 .960 .992 1.00
12 .002 .013 .067 .175 .392 .608 .825 .933 .987 .998 1.00
14 .001 .004 .025 .078 .209 .383 .617 .791 .922 .975 .996 .999 1.00
16 .000 .001 .009 .032 .100 .214 .405 .595 .786 .900 .968 .991 .999 1.00 1.00
18 .000 .000 .003 .012 .044 .109 .238 .399 .601 .762 .891 .956 .988 .997 1.00 1.00 1.00
20 .000 .000 .001 .004 .019 .051 .128 .242 .414 .586 .758 .872 .949 .981 .996 .999 1.00 1.00 1.00
243
References
1.Paul Newbold, William L. Carlson, Betty M. Thorne, “Statistics

for business and economics”, Pearson education, Inc., upper
saddle river, New jersey, 2003.
2.David R. Anderson, Dennis J. Sweeney, Thomas A.Williams, “
Introduction to statistics” West Publishing company, St. Paul,
Minnesota., 1981.
3. H. T. Hayslett, MS, : Statistics”, British library cataloguing in
publication data, 1981.
4. Paul Newbold, “Statistics for business and economics”
Prentice – Hall Inc., 1995.
5. James T.McClave, P.George Benson; Therry Sincich, “
Statistics for business and economics”, Prentice-Hall, Inc., 1998.
6. David F.Groebner, Patric W. Shannon, Phillip C.Fry, Kent
D.Smith, Business statistics, Pearson education, Inc., upper
saddle river, New jersey, 2005.
244

Solved Problems and Exercises. Part 2

Uploaded by

Copyright:

Available Formats

Solved Problems and Exercises. Part 2

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Solved Problems and Exercises. Part 2

Uploaded by

Copyright:

Available Formats

Гумбат Наби оглы Алиев

Подписано в печать 07.02.2006

Формат 60 х 84 1/16. Бумага тип. Ризограф.

Издание Казахской головной архитектурно-строительной академии

Рекомендовано к печати Ученым Советом Университета имени

© Aliyev H.N., 2006

Chapter 1. Hypothesis testing ……………………………..7

Inferential statistics consists of methods that use sample results to

1.1.2. Concepts of hypothesis testing

1.1.3. The null and alternative hypothesis

We will begin our general discussion by using  to denote a

1.1.4. Tails of the test

In statistics, the rejection region for a hypothesis testing problem can be on

b) A left tailed test

Reconsider the example of the mean amount of coffee can produced by

1.2. Tests of the mean of a normal distribution:

In this and following sections we will present specific procedures for

1.3. Tests of the mean of a normal distribution:

When the population standard deviation is unknown, we simply estimate

Step 2. Select the distribution to use

Step 4. Calculate the value of the test statistic

1. Make the following tests of hypotheses.

In previous section, the value of the significance level  was

Steps necessary for calculating the p-value for a test of hypothesis

Test statistic Test statistic

H 0 :   10 (The mean weight lost is 10kg or more)

1. Find the p-value for each of the following hypothesis tests

1. a) 0.0046; b) 0.0017 ;c) 0.0162; 2. a) 0.0204; b) yes, reject H 0 ; c) no, do

1.5. Tests of the mean of a normal distribution:

Rejection region Nonrejection region

1. For each of the following examples of tests of hypothesis about  , show

1.a) reject H 0 if t  1.350 or t  1.350 ;b) reject H 0 if t  2.819 ;c)reject

Often we want to conduct test of hypothesis about a population proportion.

2. To test either null hypothesis

3. To test the null hypothesis

1. Make the following hypothesis tests about p.

1. a) T .S.  z  0.60 ; do not reject H 0 ; b) T.S.  4.12 ; reject H 0 ;

2. To test either null hypothesis

3. To test the null hypothesis

The value of the test statistic is

where x i are measured in thousands of dollars and population distribution

Suppose that a random sample of n matched pairs of observations is obtained

1. To test either null hypothesis

The decision rule is that

1. Perform the following tests of hypothesis assuming that the population of

Do the data substantiate the claim? Use   0.05 .

Before x i 2.3 2.8 3.1 2.7 3.4 2.6 2.8 2.5

Test at the 5% significance level if there is an improvement in the job

1. a) T.S.  8.040 reject H 0 ;b) T .S.  10.847 ;reject H 0 ; c) T.S.  7.989

1.8.2. Tests based on independent samples

1. The following information is obtained from two independent samples

1.8.3. Tests based on independent samples

Many times it may not be possible to take large samples from

Reject H 0 Do not reject H 0 Reject H 0

1. The following information was obtained from two independent samples

Next we will develop procedures for comparing two population proportions.

Testing equality of two population proportions:

We are given independent samples of size n x and n y with proportion of

Then the following tests have significance level  :