# Chapter 1 – Descriptive Statistics

Data
Data are very important for scientific study, and statistics is a discipline that deals with the
collection, presentation and analysis of data. In this chapter we are going to study how we can
summarize and describe a set of data. When we study a set of data we need to identify the
following important characteristics of the dataset.
• Primary and secondary data. When the data are collected by us it is called primary data.
We always have the individual values of the data. When the dataset is collected by others,
it is called secondary data. Sometimes the data is grouped into a table, and is called
grouped data.
• Population and sample data. Population refers to the totality of elements in which we are
interested. Suppose we want to study the salary of Hong Kong people, our population
includes all those persons who work in Hong Kong. However as the population is so big
that it is not practical and economical to collect salary data of all the working people, we
always select randomly only a subset of the population and the data is sample.
• Discrete and continuous data. It is important to identify whether the data is continuous or
discrete. For example data on the number of persons in a household is discrete, and data
on salary is continuous. Different statistical techniques are used for handling discrete or
continuous data.
Frequency Distribution
Statistical data obtained by means of census, sample surveys or experiments usually consist of
raw, unorganized sets of numerical values. Before these data can be used as a basis for inferences
about the phenomenon under investigation or as a basis for decision, they must be summarized
and the pertinent information must be extracted.
Example 1
A random sample of 100 households in a town was selected and their monthly town gas
consumption (in cubic metres) in last month were recorded as follows:
55 82 83 109 78 87 95 94 85 67
80 109 83 89 91 104 90 103 67 52
107 78 86 29 72 66 92 99 60 75
88 112 97 88 49 62 70 66 88 62
72 85 81 78 77 41 105 92 94 74
78 75 87 83 71 99 56 69 78 60
1197 39 104 86 67 79 98 102 82 91
46 120 73 125 132 86 48 55 112 28
42 24 130 100 46 57 31 129 137 59
102 51 135 53 105 110 107 46 108 117
1
A useful method for summarizing a set of data is the construction of a frequency table, or a
frequency distribution. That is, we divide the overall range of values into a number of classes and
count the number of observations that fall into each of these classes or intervals.
The general rules for constructing a frequency distribution are:
(i) There should not be too few or too many classes.
(ii) Insofar as possible, equal class intervals are preferred. But the first and last classes can
be open-ended to cater for extreme values.
In example 1, the sample size is 100 and the range for the data is 113 (137 - 24). A frequency
distribution with six classes is appropriate and it is shown below.
Frequency distribution of household town gas consumption
Town gas monthly
consumption
( in cubic metres)
Number of
households
20 - 39 5
40 - 59 15
60 - 79 25
80 - 99 30
100 - 119 18
120 - 139 7
Total 100
Class limits: are the numbers that typically serve to identify the classes in a listing of a frequency
distribution. Thus, in the above frequency distribution, for the class whose frequency is 30, its
lower class limit is 80 and upper class limit is 99.
As contrasted to a class limit, a class boundary is the precise point that separates one class from
another, rather than being a value indicated in one of the classes. A class boundary is typically
located midway between the upper limit of a class and the lower limit of the next higher class
adjoining it. Therefore the class boundary separating the class 60-79 and the class 80-99 is
halfway between 79 and 80, that is, at the point 79.5.
Class interval: is the width of a class. The class interval of a class is computed by subtracting the
lower limit (boundary) of the class from the lower limit (boundary) of the next class.

Class midpoint or class mark: is the point dividing the class into equal halves on the basis of
class interval. This point can be obtained by adding the lower and upper limits (boundaries) of a
class and dividing by 2.
Relative frequency of a class: is the frequency of the class divided by the total frequency of the
distribution.
Cumulative frequency distribution: shows the number of items of a series that are less than (or
more than) certain specified values.
2
Measure of Central Tendency
A value that would describe the 'centre' of a distribution would be visually located near the spot
where most of the data seem to be concentrated. Consequently, values that fulfil this role are
called measures of central tendency.
The most common measures of the central tendency of a data set are arithmetic mean or simply as
mean, median and mode.
The mean of a set of numerical data is the sum of the set divided by the number of observations,
that is, their average.
The median of a distribution is the value which divides the distribution so that an equal number of
values lie on either side of it, i.e., half of the items have values smaller or equal to it and half of
the items have values larger or equal to it.
The mode of a set of numerical data is the value which occurs most frequently.
Example 1 (calculating mean, median and mode for individual data)
The following table shows the hourly wage rates of eight sampled construction workers.
Worker i 1 2 3 4 5 6 7 8
Hourly wage
rate (
i
x
)
\$35 38 46 60 65 69 72 78
Mean
8
) (
8
1

·
·
i
i
x
x
)
8
(
8 7 6 5 4 3 2 1
x x x x x x x x + + + + + + +
·
875 . 57
8
463
· · (\$)
Location of the median: 5 . 4
2
9
2
1
· ·
+ n
th
Median = 5 . 62
2
65 60
2
5 4
·
+
·
+ x x
(\$)
Mode: the sample size is too small, mode cannot be identified.
Example 2 (calculating mean, median and mode for grouped data)
The following table shows the daily wages of a random sample of construction workers.
Calculate its mean, median and mode.
3
Daily Wages (\$) Number of Workers
200 - 399 5
400 - 599 15
600 - 799 25
800 - 999 30
1000 - 1199 18
1200 - 1399 7
Total 100
Solution
Daily Wages (\$)
Number of
Workers
i
f
Class Mark
i
x
i i
x f
200 - 399 5 299.5 1,497.5
400 - 599 15 499.5 7,492.5
600 - 799 25 699.5 17,489.5
800 - 999 30 899.5 26,985.5
1000 - 1199 18 1,099.5 19,791.0
1200 - 1399 7 1,299.5 9,096.5
Total 100 82,350.0
Mean 5 . 823
100
0 . 350 , 82
) (
6
1
6
1
· · ·

·
·
i
i
i
i i
f
x f
x (\$)
Daily Wages (\$) Number of Workers
i
f
Cumulative Frequency
i
F
200 - 399 5 5
400 - 599 15 20
600 - 799 25 45
800 - 999 30 75
1000 - 1199 18 93
1200 - 1399 7 100
4
Total 100
As 0.5n = 0.5(100) = 50, so the median lies in the 4th class.
Median =
) (
5 . 0
4
4
3
4
c
f
F n
L

+
where L is the lower class boundary,
c is the class interval.
8 . 832 ) 200 (
30
45 ) 100 ( 5 . 0
5 . 799 ·

+ · (\$)
Daily Wages (\$)
Number of
Workers
i
f
Class Interval
i
c
Relative Density
) 200 (
'
i
i
i
c
f
f ·
200 - 399 5 200 5
400 - 599 15 200 15
600 - 799 25 200 25
800 - 999 30 200 30
1000 - 1199 18 200 18
1200 - 1399 7 200 7
Total 100
As 30
'
4
· f is the largest relative density, so mode lies in the 4th class.
Mode ) (
) ( ) (
4
'
3
'
4
'
5
'
4
'
3
'
4
4
c
f f f f
f f
L
− + −

+ ·
3 . 858 ) 200 (
) 25 30 ( ) 18 30 (
25 30
5 . 799 ·
− + −

+ ·
(\$)
Mean
Advantages: (i) All values in the distribution are used in its calculation, so it can be
regarded as more representative than the other two measures.
(ii) Its method of calculation is simple and most people understand the
meaning of its result.
(iii) Its result can easily be used in further analysis.
5
Disadvantages: (i) Its result can be easily distorted by extreme values. As such, its result
may be rather lower or higher than the bulk of the values and
becomes unrepresentative.
(ii) In case of open end classes, mean can be calculated only if their class
marks are determined. If such classes contain a large proportion of
the values, then the mean may be subjected to substantial error.
Median
Advantage: Its result will not be affected by extreme values and open end
classes.
Disadvantage: It has to be supplemented by other statistics because it does not
reflect the distribution in the way that the mean does, that is,
including all values.
Mode
Advantages: (i) Its result will not be affected by extreme values and open end
classes.
(ii) If data are not grouped, it can be determined easily.
Disadvantages: (i) It has to be supplemented by other statistics.
(ii) It is difficult to obtain an accurate estimate of the mode if the values
are classified into a frequency distribution.
How to select a suitable measure
(i) Always select the mean whenever there is no special reason for choosing the other two
measures.
(ii) Select the median is the distribution consists of substantial amount of extreme large or
small values.
(iii) Select the mode if integral result is preferred as in cases the data are in ordinal scales.
6

Measure of data variation (variability)
A measure of central tendency is almost never, by itself, sufficient to provide an adequate
summary of the characteristics of a set of data. We will usually require, in addition, a measure of
the amount of variation in the data.
Example 1
Consider the following measurements, in grams, for two samples of strawberry jam bottled by
companies A and B:
Sample for
Company A
31 32 32 33 32
Sample for
Company B
28 29 32 35 36
Both samples have the same mean, 32 grams. It is obvious that company A, in comparison with
company B, bottles strawberry jam with a more consistent content. We say that the variability of
the observations is smaller for company A. Therefore in buying strawberry jam we would feel
more confident that the bottle we select will be closer to the advertised average content if we buy
from company A.
The most important measures of variability or dispersion are the range, mean deviation,
standard deviation and variance.
(There are some other measures like quartile deviation and percentiles. We shall not study these
measures. Read our textbook if interested)
The range of a set of numbers is the difference between the largest and the smallest number in
the set.
Example 2 (For individual data)
The following table shows the hourly wage rates of eight sampled construction workers.
Worker i 1 2 3 4 5 6 7 8
Hourly wage
rate (
i
x
)
\$35 38 46 60 65 69 72 78
The range is \$78 - \$35 = \$43.
Though range is simple and can be obtained easily, its result is unstable. This is particularly true
if the sample size is large. So whenever the sample size is over 10, we seldom choose to use
range to indicate variability of the data.
7
Mean deviation is the average of the absolute deviation of the numerical data from their mean.

Worker i 1 2 3 4 5 6 7 8
Hourly wage
rate (
i
x
)
\$35 38 46 60 65 69 72 78
875 . 57 − ·

i
i
x
x x
22.875 19.875 11.875 2.125 7.125 11.125 14.125 20.125
Mean deviation
656 . 13
8
25 . 109
8
875 . 57
8
1
· ·

· i
i
x
(\$)
The mean deviation is a good measure to show the extent of variation of the data in a distribution.
However, when this measurement is used in further analysis, it would give rise to some
unnecessary tedious mathematical problem as a result of its absolute value term. To avoid this
pitfall, we can use the standard deviation instead.
Standard deviation of a population (
) σ
is the square root of the average of the squared distances
of the observations from the mean.
N
x
N
i
i ∑
·

·
1
2
) ( µ
σ
, where
µ
is the population mean
To compute the sample standard deviation
) (s
we use the above formula, replacing
µ
by x
and N by 1 − n .
2
1
( )
1
n
i
i
x x
s
n
·

·

Worker i 1 2 3 4 5 6 7 8 Total
Hourly wage
rate (
i
x
)
\$35 38 46 60 65 69 72 78 463
2
( 35.875)
16.226(\$)
7
x
s

· ·
Variance is the square of the standard deviation.
8
2
2 1
( )
1
n
i
i
x x
s
n
·

·

Example 3 (for grouped data)
The following table shows the daily wages of a random sample of construction workers.
Calculate its mean deviation, variance, and standard deviation.

Daily Wages (\$) Number of Workers
200 - 399 5
400 - 599 15
600 - 799 25
800 - 999 30
1000 - 1199 18
1200 - 1399 7
Total 100
Solution
Daily Wages (\$)
Number of
Workers
i
f
Class Mark
i
x
5 . 823 − ·

i i
i i
x f
x x f
200 - 399 5 299.5 2,620
400 - 599 15 499.5 4,860
600 - 799 25 699.5 3,100
800 - 999 30 899.5 2,280
1000 - 1199 18 1,099.5 4,968
1200 - 1399 7 1,299.5 3,332
Total 100 21,160
Mean deviation 60 . 211
100
160 , 21
6
1
6
1
· ·

·

·
·
i
i
i
i i
f
x x f
(\$)
Daily Wages
(\$)
Number of
Workers Class Mark
2
( )
i i
f x x −
9
i
f
i
x
200 - 399 5 299.5 1, 372,880
400 - 599 15 499.5 1,574,640
600 - 799 25 699.5 384,400
800 - 999 30 899.5 173,280
1000 - 1199 18 1,099.5 1,371,168
1200 - 1399 7 1,299.5 1,586,032
Total 100 6,462,400
Variance
2
6462400
( ) 65, 276.77
99
s · ·
Standard deviation =
65276.77 255.49 ·
Comparison of the variation of two distributions
The values of the standard deviations cannot be used as the bases of the comparison because:
(a) units of measurements of the two distributions may be different, and
(b) average values of two distributions may be widely dissimilar.
The correct measure that should be used is the coefficient of variation
) (CV
.
% 100
x
s
CV ·
Example 4
The following table shows the summary statistics for the daily wages of two types of workers.
Worker's
Type
Daily Wages
Mean Standard deviation
I \$100 \$20
II \$150 \$24
Compare these two daily wages distributions.
Solution
In comparison Distribution Reason
Average magnitude
II > I 100 150 · > ·
I II
x x
Variation I > II
% 16 % 100
150
24
% 20 % 100
100
20
· · > · ·
II I
CV CV
Chapter Two - Probability
Introduction and concepts
“Perhaps it was man’s unquenchable thirst for gambling that led to the early development of
probability theory. In an effort to increase their winnings, gamblers called upon the
10
mathematicians to provide optimum strategies for various games of chance.” ---- from Walpole
R.E. Introduction to Statistics
Probability is the basis upon which the discipline of statistics has been developed and applied in
many fields associated with chance occurrences such as politics, business, weather forecasting,
and scientific research. Probability may be taken as a tool with which we may solve problems
involving uncertainties. In fact uncertainty is a basic element of human experiences. To cite some
examples: travelling time, number of customers, rainfall, temperature, share price movement,
length of our life, etc.
There are three approaches to understand probability. In the empirical approach, probability may
be taken as a relative frequency. As such the probability of an aeroplane arriving its destination
on time may be taken as the proportion of times the aeropline has been on time in the past, say,
one thousand times.
Suppose in a trial of an experiment, there are k possible outcomes which are equally likely. The
probability of the occurrence of an outcome is therefore 1/k. Thus in throwing a coin, the
probability of having a head is ½. In our course, we shall adopt this approach but the empirical
approach is always useful in giving us some intuition to understand the problem.
The third approach is very mathematical. A number of axioms have been set up and from these some
theorems of probability have been developed. This approach is too abstract and usually used by
mathematicians.
Some Basic Concepts
Sample space: is a set of all possible outcomes of an experiment.
Event: is a subset of a sample space.
To find the probability of an event we need to count the number of outcomes of the event and the
number of all possible outcomes of the experiment, and then to divide the former by the latter. Hence
the following counting rules may be helpful.
Some counting rules
Example 1
Three items are selected at random from a manufacturing process. Each item is inspected and
classified defective (D) or non-defective (N).
Its sample space is =
¹
;
¹
¹
'
¹
NNN NND NDN DNN
NDD DND DDN DDD
Example 2
11
The event that the number of defectives in above example is greater than 1.
Its sample space is = {DDD DDN DND NDD}
The probability of the event is 4/8 or ½.
Example 3
Suppose a licence plate containing two letters following by three digits with the first digit not
zero. How many different licence plates can be printed?
1st
Letter
2
nd
Letter
1st
Digit
2nd
Digit
3rd
Digit
Number of
Choices
A - Z
(26)
A - Z
(26)
1 - 9
(9)
0 - 9
(10)
0 - 9
(10)
Number of different licence plates that can be printed is
(26)(26)(9)(10)(10) = 608,400
Example 4
Find the possible permutations (the number of ways where sequence of the letters is counted)
from 3 letters A, B, C.
Consider the following tree diagram:
• The number of permutations of n distinct objects is

! ) 1 )( 2 ).....( 2 )( 1 )( ( n n n n P
n n
· − − ·
• The number of permutations of n distinct objects taken r at a time is
12

) 1 )( 2 )...( 1 )( ( + − + − − · r n r n n n P
r n

( )( )
)! (
!
) 1 )( 2 )...( 1 )( (
) 1 )( 2 )...( 1 )( ( ) 1 )( 2 )...( 1 )( (
r n
n
r n r n
r n r n r n r n n n

·
− − −
− − − + − + − −
·

e.g. The number of 3-letter words formed from 5 letters is
60
)! 3 5 (
! 5
3 5
·

· P
• The number of distinct permutations of n objects of which
1
n
are alike of the first
kind,
2
n are alike of the second kind, .....,
k
n
are alike of the kth kind and
n n n n
k
· + + + ...
2 1
is
) ! )...( ! )( ! (
!
2 1 k
n n n
n
Find the possible permutations of the following 5 letters: A A A B C
There are five objects of which three are alike.
! 3
! 5
! 3
5 5
· ·
P

Example 5
How many 7-letter words can be formed using the letters of the word 'BENZENE'?
(there are 1 B, 3 E, 2 N and 1 Z)
The number of 7-letter words that can be formed is
420
) ! 1 )( ! 2 )( ! 3 )( ! 1 (
! 7
·

• The number of combinations (number of ways where sequence is not counted) of n
distinct objects taken r at a time is
)! ( !
!
r n r
n
C
r n

·
Find the possible combinations of 5 distinct objects taken 3 at a time.

)! 3 5 ( ! 3
! 5

·
Example 6
13
The number of 3-person committees that can be formed from a group of 4 persons is
4
)! 3 4 ( ! 3
! 4
3 4
·

· C
Example 7
A box contains 8 eggs, 3 of which are rotten. Three eggs are picked at random. Find the
probabilities of the following events.
(a) Exactly two eggs are rotten.
(b) All eggs are rotten.
(c) No egg is rotten.
Solution:
(a) The 8 eggs can be divided into 2 groups, namely, 3 rotten eggs as the first group and 5
good eggs as the second group.
Getting 2 rotten eggs in 3 randomly selected eggs can occurred if we select randomly 2
eggs from the first group and 1 egg from the second group.
The number of this outcome is
( ) ( ) 15
1 5 2 3
· C C
Total number of possible outcomes of selecting 3 eggs randomly from the total 8 eggs is
56
3 8
· C
.
Thus the probability of having exactly two rotten among the 3 randomly selected eggs is
( ) ( )
56
15
3 8
1 5 2 3
·
C
C C

(b) Similarly, the probability of having all 3 rotten eggs is
( ) ( )
56
1
3 8
0 5 3 3
·
C
C C
(c) The probability of having no rotten egg is
( ) ( )
28
5
56
10
3 8
3 5 0 3
· ·
C
C C
14
Rules of probability
The following rules may help us to find the probability of an event.
Addition Rule: For any events that are not mutually exclusive
) ( ) ( ) ( ) ( B A P B P A P B A P ∩ − + · ∪
where B A∪ is the union of two sets A and B, it is the set of elements that belong to A
or to B or to both.
B A∩ is the intersection of two sets A and B, it is the set of elements that are common
to A and B.
Illustrative example
180 students took examinations in English and Mathematics. Their results were as follows:
Number of students passing English = 80
Number of students passing Mathematics = 120
Number of students passing at least one subject = 144
Then we can rewrite the above results as:
Probability that a randomly selected student passed English =
9
4
180
80
·
Probability that a randomly selected student passed Mathematics
3
2
180
120
· ·
Probability that a randomly selected student passed at least one subject
5
4
180
144
· ·
Find the probability that a randomly selected student passed both subject.
Solution
Let E be the event of passing English, and M be the event of passing Mathematics.
It is given that:
9
4
) ( · E P ;
3
2
) ( · M P ;
5
4
) ( · ∪E M P
As
) ( ) ( ) ( ) ( E M P M P E P E M P ∩ − + · ∪

31 . 0
45
14
5
4
3
2
9
4
) ( ) ( ) ( ) ( · · − + · ∪ − + · ∩ E M P M P E P E M P
Example 8
15
A card is drawn from a complete deck of playing cards. What is the probability that the card is a
heart or an ace?
Solution
Let A be the event of getting a heart, and B be the event of getting an ace.
The probability that the card is a heart or an ace is
) ( B A P ∪
.
) ( ) ( ) ( ) ( B A P B P A P B A P ∩ − + · ∪

13
4
52
16
52
1
52
4
52
13
· · − + ·
For mutually exclusive events,
) ( ) ( ) ( B P A P B A P + · ∪
What is the probability of getting a total of '7' or '11' when a pair of dice are tossed?
Solution
Total number of possible outcomes = (6)(6) = 36
Possible outcomes of getting a total of '7' :{1,6; 2,5; 3,4; 4,3; 5,2; 6,1}
Possible outcomes of getting a total of '11' : {5,6; 6,5}
Let A be the event of getting a total of '7', and B be the event of getting a total of '11'.
The probability of getting a total of '7' or '11' is
) ( B A P ∪
.
) ( ) ( ) ( ) ( ) ( ) ( B P A P B A P B P A P B A P + · ∩ − + · ∪
...A and B are mutually exclusive

9
2
36
2
36
6
· + ·
If A and A' are complementary events then
) ' ( 1 ) ( A P A P − ·
Example 9
A coin is tossed six times in succession. What is the probability that at least one head occurs?
Let A be the number of heads occurs in six successive tosses.
) 0 ( 1 ) 1 ( · − · ≥ A P A P
16
( )( )( )( )( )( )
64
63
2
1
2
1
2
1
2
1
2
1
2
1
1 · − ·
Conditional Probability
Let A and B be two events. The conditional probability of event A given that event B has
occurred, denoted by
) / ( B A P
is defined as
) (
) (
) / (
B P
B A P
B A P

·
provided that P(B) > 0.
Similarly, the conditional probability of B given that event A has occurred is defined as
) (
) (
) / (
A P
B A P
A B P

·
, provided P(A) > 0.
Example 10
A hamburger chain found that 75% of all customers use mustard, 80% use ketchup, and 65% use
both, when ordering a hamburger. What are the probabilities that:
(a) a ketchup-user uses mustard?
(b) a mustard-user uses ketchup?
Solution
Let A be the event of using mustard, and B be the event of using ketchup.
It is given that:
75 . 0 ) ( · A P
;
80 . 0 ) ( · B P
;
65 . 0 ) ( · ∩B A P

(a) P(a ketchup-user uses mustard)
8125 . 0
80 . 0
65 . 0
) (
) (
) / ( · ·

· ·
B P
B A P
B A P

(b) P(a mustard-user uses ketchup)
8667 . 0
75 . 0
65 . 0
) (
) (
) / ( · ·

· ·
A P
B A P
A B P

Multiplicative Rule
) / ( ) ( ) ( A B P A P B A P · ∩
or =
) / ( ) ( B A P B P
Statistically Independence: the occurrence or non-occurrence of one event has no effect on the
probability of occurrence of the other event.
17
Two events A and B are independent if and only if
) ( ) ( ) ( B P A P B A P · ∩
Example 11
A pair of fair dice are thrown twice. What is the probability of getting totals of 7 and 11?
Solution
Let
i
A
be the event of getting '7' in the i-th throw and j
B
be the event of getting '11' in the j-th
throw.
P(Getting totals of 7 and ll) ) ( ) ( ) (
2 1 2 1
A B P B A P B A P ∩ + ∩ · ∩ ·
) / ( ) ( ) / ( ) (
1 2 1 1 2 1
B A P B P A B P A P + ·
) ( ) ( ) ( ) (
2 1 2 1
A P B P B P A P + · .... j i
B A ,
are independent
54
1
36
6
36
2
36
2
36
6
·
,
_

¸
¸

,
_

¸
¸
+
,
_

¸
¸

,
_

¸
¸
·
Theorem of Total Probability
If the events
k
B B B ,..., ,
2 1
constitute a partition of the sample space S such that
) 0 ( ≠
i
B P

for i = 1, 2, ... , k, then for any event A of S
) ( ... ) ( ) ( ) (
2 1 k
B A P B A P B A P A P ∩ + + ∩ + ∩ ·

) / ( ) ( ... ) / ( ) ( ) / ( ) (
2 2 1 1 k k
B A P B P B A P B P B A P B P + + + ·
Example 12
Suppose 50% of the cars are manufactured in the United States and 15% of these are compact;
30% of the cars are manufactured in Europe and 40% of these are compact; and finally, 20% are
manufactured in Japan and 60% of these are compact. If a car is picked at random from the lot,
find the probability that it is a compact.
Let A be the event that the car is compact,

1
B be the event that the car is manufactured is United States,

2
B be the event that the car is manufactured in Europe, and

3
B
be the event that the car is manufactured in Japan.
) ( ) ( ) ( ) (
3 2 1
B A P B A P B A P A P ∩ + ∩ + ∩ ·

) / ( ) ( ) / ( ) ( ) / ( ) (
3 3 2 2 1 1
B A P B P B A P B P B A P B P + + ·

315 . 0 ) 60 . 0 )( 20 . 0 ( ) 40 . 0 )( 30 . 0 ( ) 15 . 0 )( 50 . 0 ( · + + ·
Baye's Theorem
If k
E E E ,...,
2 , 1 are mutually exclusive events such that
k
E E E ∪ ∪ ∪ ...
2 1
contains all
sample points of S, then for any event D of S with
0 ) ( ≠ D P
,
18

·

·

·
k
j
j
i i
i
D E P
D E P
D P
D E P
D E P
1
) (
) (
) (
) (
) / (

) / ( ) ( ... ) / ( ) ( ) / ( ) (
) / ( ) (
2 2 1 1 k k
i i
E D P E P E D P E P E D P E P
E D P E P
+ +
·
Example 13
Suppose a box contains 2 red balls and 1 white ball and a second box contains 2 red ball and 2
white balls. One of the boxes is selected by chance and a ball is drawn from it. If the drawn ball is
red, what is the probability that it came from the 1st box?
Solution

Let A be the event of drawing a red ball and B be the event of choosing the 1st box.
Given:
2
1
) ' ( ) ( · · B P B P ;
3
2
) / ( · B A P ;
4
2
) ' / ( · B A P
P(Coming from the 1st box/the drawn ball is red)
) / ( A B P ·

) ' ( ) (
) (
) (
) (
B A P B A P
B A P
A P
B A P
∩ + ∩

·

·

7
4
)
4
2
)(
2
1
( )
3
2
)(
2
1
(
)
3
2
)(
2
1
(
) ' / ( ) ' ( ) / ( ) (
) / ( ) (
·
+
·
+
·
B A P B P B A P B P
B A P B P
Chapter Three - Probability Distributions
To cope with uncertainties of outcome, a statistical model that describes the behavior of the outcome
is needed. These theoretical models which are very similar to relative frequency distributions, are
called probability distributions.
Random Variables - A random variable is a variable that takes on different numerical values
determined by the outcomes of a random experiment.
19
Example 1
An experiment of tossing a coin 3 times.
Let random variable, X be the number of heads achieved.
As S = {HHH HHT HTH THH TTH THT HTT TTT},
so X = {0, 1, 2, 3}
Discrete random variable - in a given interval, only a specified number of values can occur.
Continuous random variable - in a given interval, any value can occur.
Probability Distribution of a random variable - is a representation of the probabilities for all
the possible outcomes.
Example 2
The probability distribution of the number of heads occurred when a coin is tossed 4 times.
x 0 1 2 3 4
P(X=x)
16
1
16
4
16
6
16
4
16
1
That is,
16
) (
4 x
C
x X P · · ,
, 0 · x
1, 2, 4
Example 3
Consider an experiment of tossing two fair dice.
Let random variable, X be the sum of the two dice. Then the probability distribution of X is:
x 2 3 4 5 6 7 8 9 10 11 12
P(X=x)
36
1
36
2
36
3
36
4
36
5
36
6
36
5
36
4
36
3
36
2
36
1
The probability function
) (x f
, of a discrete random variable X expresses the probability that X
takes the value x, as a function of x. That is
) ( ) ( x X P x f · ·
20
where the function is evaluated at all possible values of x.
Properties of probability function
) ( x X P ·
:-
1.
0 ) ( ≥ ·x X P
for any value x.
2.

· ·
x
x X P . 1 ) (
Mathematical Expectations
The expected value, E(X), of a discrete random variable X is defined as

E X xP X x
x
( ) ( ) or
x
µ · ·

It is the mean of the probability distribution.
Let X be a random variable. The expectation of the squared discrepancy about the mean, (X −
µ x)
2
, is called the variance, denoted σ x
2
, and given by
Var X E X
x x
( ) [( ) ] or σ µ
2 2
· −
· − ·

( ) ( ) x P X x
x
x
µ
2
· · −

x P X x
x
x
2 2
( ) µ
Example 4
Calculate the mean and variance of the discrete probability distribution in example 2 and 3.
The Normal Distribution
Normal distribution is probability distribution of a continuous random variable. It is based on the
Law of Errors which states that
1. Errors are inevitable.
2. Large errors are less likely than small errors.
3. Positive and negative errors are equally likely.
Definition :
21
A continuous random variable X is defined to be a normal random variable if its probability
function is given by
f x
x
( )
( )
exp[ ( ) ] · −
− 1
2
1
2
2
σ π
µ
σ
for −∞ < x < +∞
where µ = the mean of X, σ = the standard deviation of X,
π = 3.14154
Notation : X ~ N(µ , σ
2
)
Properties of the normal distribution:-
1. It is a continuous distribution.
2. The curve is symmetric and bell-shaped about a vertical axis through the mean µ .
3. The total area under the curve and above the horizontal axis is equal to 1.
4. Area under the normal curve:
• Approximately 68% of the values in a normally distributed population are within
1 standard deviation from the mean.
• Approximately 95.5% of the values in a normally distributed population are
within 2 standard deviation from the mean.
• Approximately 99.7% of the values in a normally distributed population are
within 3 standard deviation from the mean.
The standard normal curve :
The distribution of a normal random variable with µ = 0 and σ =1 is called a standard normal
distribution. Usually a standard normal random variable is denoted by Z.
Notation : Z ~ N(0, 1)
Remark : Usually a table of Z is set up to find the probability P(Z ≥ z) for z ≥ 0.
Example 7
Given Z ~ N(0, 1)
(a) P(Z > 1.73) = 0.0418
(b) P(0 < Z < 1.73) = P(Z > 0) - P(Z > 1.73) = 0.5 - 0.0418 = 0.4582
22
(c) P(−2.42 < Z < 0.8) = 1 - P(Z < -2.42) - P(Z > 0.8)
= 1 - 0.00776 - .2119 = 0.78034
(d) P(1.8 < Z < 2.8) = P(Z > 1.8) - P(Z > 2.8) = 0.0359 - 0.00256 = 0.03334
(e) the value z that has
(i) 5% of the area below it
Let the corresponding z value be z1, then we have P(Z < z1) = 0.05.
From the standard normal distribution table we have P(Z < -1.64) = 0.05.
So z1 = -1.64
(ii) 39.44% of the area between 0 and z.
Let the corresponding z value be z1, then we have P(0 < Z < z1) = 0.3944.
From the standard normal distribution table we have P(0 < Z < 1.25) = 0.3944.
So z1 = 1.25
Theorem :
If X is a normal random variable with mean µ and standard deviation σ , then
Z
X
·
− µ
σ
is a standard normal random variable and hence
P x X x P
x
Z
x
( ) ( )
1 2
1 2
< < ·

< <
− µ
σ
µ
σ
Example 8
Given X ~ N(50, 10
2
), find P(45 < X < 62).
Solution:
,
_

¸
¸ −
<

<

· < <
σ
µ
σ
µ
σ
µ 62 45
) 62 45 (
X
P X P

,
_

¸
¸ −
< <

·
10
50 62
10
50 45
Z P
) 2 . 1 5 . 0 ( < < − · Z P
) 2 . 1 ( ) 5 . 0 ( 1 > − − < − · Z P Z P
= 1 - 0.3085 - .1151 = 0.5764
23
Example 9
The charge account at a certain department store is approximately normally distributed with an
average balance of \$80 and a standard deviation of \$30. What is the probability that a charge
account randomly selected has a balance
(a) over \$125;
(b) between \$65 and \$95.
Let X be the balance in the charge account. X ∼ N(80,
2
30 )
(a) )
125
( ) 125 (
σ
µ
σ
µ −
>

· >
X
P X P
0668 . 0 ) 5 . 1 ( )
30
80 125
( · > ·

> · Z P Z P
(b)
,
_

¸
¸ −
<

<

· < <
σ
µ
σ
µ
σ
µ 95 65
) 95 65 (
X
P X P

,
_

¸
¸ −
< <

·
30
80 95
30
80 65
Z P

) 5 . 0 ( ) 5 . 0 ( 1 ) 5 . 0 5 . 0 ( > − − < − · < < − · Z P Z P Z P
3830 . 0 3085 . 0 3085 . 0 1 · − − ·
Example 10
On an examination the average grade was 74 and the standard deviation was 7. If 12% of the
class are given A’s, and the grades are curved to follow a normal distribution, what is the lowest
possible A and the highest possible B?
Let X be the examination grade and x1 be the lowest grade for A.
12 . 0
7
74
12 . 0 ) (
1
1
·
,
_

¸
¸ −
> ⇒ · >
x
Z P x X P
From the standard normal distribution, we get

1210 . 0 ) 17 . 1 ( · > Z P
, and
1190 . 0 ) 18 . 1 ( · > Z P
so
12 . 0 ) 175 . 1 ( ≅ > Z P
Thus 175 . 1
7
74
1
·
− x
24
i.e. 83 2 . 82 ) 175 . 1 )( 7 ( 74
1
≅ · + · x
The highest possible B is 82.

The Binomial Distribution
A binomial experiment possesses the following properties :
1. There are n identical observations or trials.
2. Each trial has two possible outcomes, one called “success” and the other “failure”. The
outcomes are mutually exclusive and collectively exhaustive.
3. The probabilities of success p and of failure 1 −p remain the same for all trials.
4. The outcomes of trials are independent of each other.
Example 11
1. In testing 10 items as they come off an assembly line, where each test or trial may
indicate a defective or a non-defective item.
2. Five cards are drawn with replacement from an ordinary deck and each trial is labelled a
success or failure depending on whether the card is red or black.
Definition :
In a binomial experiment with a constant probability p of success at each trial, the probability
distribution of the binomial random variable X, the number of successes in n independent trials, is
called the binomial distribution.
Notation : X ~ b(n, p)
P(X = x) =
n
x
p q
x n x
¸
¸

_
,

x = 0, 1, …, n
p + q = 1
Example 12
Of a large number of mass-produced articles, one-tenth is defective. Find the probabilities that a
random sample of 20 will obtain
(a) exactly two defective articles;
(b) at least two defective articles.
Let X be the number of defective articles in a random sample of 20. X ∼ b(20,
10
1
)
(a) 28517 . 0
10
9
10
1
2
20
) 2 (
18 2
·
,
_

¸
¸

,
_

¸
¸

,
_

¸
¸
· · X P
25
(b)
60825 . 0 27017 . 0 12158 . 1
10
9
10
1
1
20
10
9
10
1
0
20
1
) 1 ( ) 0 ( 1 ) 2 (
19 20 0
· − − ·
,
_

¸
¸

,
_

¸
¸

,
_

¸
¸

,
_

¸
¸

,
_

¸
¸

,
_

¸
¸
− ·
· − · − · ≥ X P X P X P
Example 13
A test consists of 6 questions, and to pass the test a student has to answer at least 4 questions
correctly. Each question has three possible answers, of which only one is correct. If a student
guesses on each question, what is the probability that the student will pass the test?
Let X be the no. of correctly answered questions among 6 questions. X ∼ b(6,
3
1
)
( ) ( )
x x
x x
x
x X P X P

· ·
∑ ∑

,
_

¸
¸
· · · ≥
6
6
4
6
4
3
2
3
1
6
) ( ) 4 (
( ) ( ) ( ) ( ) ( ) ( ) 10014 . 0
3
2
3
1
6
6
3
2
3
1
5
6
3
2
3
1
4
6 0 6 1 5 2 4
·

,
_

¸
¸
+

,
_

¸
¸
+

,
_

¸
¸
·
Theorem
The mean and variance of the binomial distribution with parameters of n and p are
µ = np and σ
2
= npq respectively where p + q = 1.
Example 14
A packaging machine produces 20 percent defective packages. A random sample of ten packages
is selected, what are the mean and standard deviation of the binomial distribution of that process?
Let X be the no. of defective packages in a sample of 10 packages. X ∼ b(10, 0.2)
Its mean is µ = np = (10)(0.2) = 2
Its standard deviation is 265 . 1 ) 8 . 0 )( 2 . 0 )( 10 ( · · · npq σ
The Normal Approximation to the Binomial Distribution Theorem :
Given X is a random variable which follows the binomial distribution with parameters n and p,
then
P X x P
x np
npq
Z
x np
npq
( ) (
( . )
( )
( . )
( )
) · ·
− −
< <
+ − 05 05
if n is large and p is not close to 0 or 1.
Remark : If both np and nq are greater than 5, the approximation will be good.
26
Example 15
A process yields 10% defective items. If 100 items are randomly selected from the process, what
is the probability that the number of defective exceeds 13?
Let X be the no. of defective in a random sample of 100 items. X ∼ b(100, 0.1)
10 ) 1 . 0 )( 100 ( · · ·np µ
, 3 ) 9 . 0 )( 1 . 0 )( 100 ( · · · npq σ
) 5 . 13 ' ( ) 13 ( > ≅ > X P X P
by normal approximation
121 . 0 ) 167 . 1 (
3
10 5 . 13 5 . 13 '
· > ·
,
_

¸
¸ −
> ·
,
_

¸
¸ −
>

· Z P Z P
X
P
σ
µ
σ
µ
Example 17
A multiple-choice quiz has 200 questions each with four possible answers of which only one is
the correct answer. What is the probability that sheer guesswork yields from 25 to 30 correct
answers for 80 of the 200 problems about which the student has no knowledge?
Let X be the no. of correct answers for 80 with sheer guesswork. X ∼ b(80, 0.25)
20 ) 25 . 0 )( 80 ( · · ·np µ
, 15 ) 75 . 0 )( 25 . 0 )( 80 ( · · · npq σ
) 5 . 30 ' 5 . 24 ( ) 30 25 ( < < ≅ ≤ ≤ X P X P
by normal approximation
1196 . 0 00336 . 0 1230 . 0 ) 71 . 2 16 . 1 (
15
20 5 . 30
15
20 5 . 24
· − · < < ·

,
_

¸
¸ −
< <

· Z P Z P
The Poisson Distribution
Experiments yielding numerical values of a random variable X, the number of successes
(observations) occurring during a given time interval (or in a specified region) are often called
Poisson experiments.
A Poisson experiment has the following properties:
1. The number of successes in any interval is independent of the number of successes in
other interval.
2. The probability of a single success occurring during a short interval is proportional to the
length of the time interval and does not depend on the number of successes occurring
outside this time interval.
3. The probability of more than one success in a very small interval is negligible.
Examples of random variables following Poisson Distribution
27
1. The number of customers arrived during a time period of length t.
2. The number of telephone calls per hour received by an office.
3. The number of typing errors per page.
4. The number of accidents occurred at a junction per day.
Definition :
The probability distribution of the Poisson random variable X is called the Poisson distribution.
Notation : X ~ Po(λ )
where λ is the average number of successes occuring in the given time interval.
P(X = x) =
e
x
x −λ
λ
!
x = 0, 1, 2, …
e = 2.718283
Theorem: In a Poisson Distribution mean is equal to variance, i.e.,
2
σ µ· .
Example 17
The average number of radioactive particles passing through a counter during 1 millisecond in a
laboratory experiment is 4. What is the probability that 6 particles enter the counter in a given
millisecond?
Let X be the no. of particles entering the counter in a given millisecond. X ∼ Po(4)
1042 . 0
! 6
4
) 6 (
6 4
· · ·

e
X P
Example 19
Ships arrive in a harbour at a mean rate of two per hour. Suppose that this situation can be
described by a Poisson distribution. Find the probabilities for a 30-minute period that
(a) No ships arrive;
(b) Three ships arrive.
Let X be the no. of ship arriving in a harbour for a 30-minute period. X ∼ Po( 1
2
2
· )
28
(a) 3679 . 0
! 0
1
) 0 (
0 1
· · ·

e
X P
(b) 0613 . 0
! 3
1
) 3 (
3 1
· · ·

e
X P
Theorem :
The mean and variance of the Poisson distribution both have mean λ .
Poisson approximation to the binomial distribution
If n is large and p is near 0 or near 1.00 in the binomial distribution, then the binomial distribution
can be approximated by the Poisson distribution with parameter np.
Example 20
If the prob. that an individual suffers a bad reaction from a certain injection is 0.001, determine
the prob. that out of 2000 individuals, more than 2 individuals will suffer a bad reaction.
Sol
n
: According to binomial :
The required probability
= ( ) ( ) ( ) ( ) ( ) ( ) 1
2000
0
0001 0999
2000
1
0001 0999
2000
2
0001 0999
0 2000 1 1999 2 1998

¸
¸

_
,

+
¸
¸

_
,

+
¸
¸

_
,

¹
'
¹
¹
;
¹
. . . . . .
Using Poisson distribution:
P(0 suffers) =
2
0
1
0 2
2
e
e

·
!
 λ = np = 2
P(1 suffers) =
2
2 1
2
! 1
2
e
e
·

P(2 suffer) =
2
2!
2
2 2
2
e
e

·
Then the required probability = 1
5
0 323
2
− ·
e
.
General speaking, the Poisson distribution will provide a good approximation to binomial when
(i) n is at least 20 and p is at most 0.05; or
(ii) n is at least 100, the approximation will generally be excellent provided p< 0.1.
Example 21
29
Two percent of the output of a machine is defective. A lot of 300 pieces will be produced.
Determine the probability that exactly four pieces will be defective.
Let X be the no. of defective pieces among 300 pieces. X ∼ b(300, 0.02)
1338 . 0 ) 98 . 0 ( ) 02 . 0 ( ) 4 (
296 4
4 300
· · · C X P
By Poisson Approximation:
6 ) 02 . 0 )( 300 ( · · ·np λ
1338 . 0
! 4
6
) 4 (
4 6
· · ·

e
X P
CHAPTER 4 - SAMPLING DISTRIBUTIONS AND ESTIMATION
Definition
1. A sample statistic is a characteristic of a sample.
A population parameter is a characteristic of a population.
2. A statistic is a random variable that depends only on the observed random sample.
3. A sampling distribution is a probability distribution for a sample statistic. It indicates the
extent to which a sample statistic will tend to vary because of chance variation in random
sampling.
4. The standard deviation of the distribution of a sample statistic is known as the standard
error of the statistic.
An illustrating example
Suppose a population consists of four elements, {0,1,2,3}. A simple random sample of two
elements is to be drawn.
The population has two parameters: a mean
µ
of 1.5 and a variance
2
σ
of 1.6667.
Obviously there are six possible samples ( 6
4
2
· C ). They are
30
Sample Sample mean error Probability
0,1 0.5 -1.0 1/6
0,2 1.0 -0.5 1/6
0,3 1.5 0 1/6
1,2 1.5 0 1/6
1,3 2.0 0.5 1/6
2,3 2.5 1.0 1/6
From the above table, we can see that if we draw a sample and use the sample mean to estimate
the population mean, the accuracy of our estimate depends on which sample we have drawn,
which in turn depends on chance.
The probability distribution of sample mean is known as a sampling distribution of sample mean,
as compiled in the following table:
Sample mean 0.5 1.0 1.5 2.0 2.5
Probability 1/6 1/6 2/6 1/6 1/6
The expected value of sample mean is
Y y P y y E · · + + + + · ·

5 . 1 6 / 1 * 5 . 2 6 / 1 * 0 . 2 6 / 2 * 5 . 1 6 / 1 * 0 . 1 6 / 1 * 5 . 0 ) ( * ) (
.
Hence the average value of the sample mean is equal to the population mean. We call the sample
mean an unbiased estimator of the population mean.
The variance of the sample mean (i.e., the average square deviation of the sample mean from the
population mean) is: ∑
− · ) ( ) ( ) (
2
y P Y y y V
=
2 2 2 2 2
1 1 2 1 1
(0.5 1.5) * (1.0 1.5) * (1.5 1.5) * (2.0 1.5) * (2.5 1.5) *
6 6 6 6 6
− + − + − + − + −
=0.4167
Sampling Distribution of Mean
The Central Limit Theorem
If repeated samples of size n are drawn from any infinite population with mean
µ
and
variance σ
2
, and n is large (n ≥ 30), the distribution of x , the sample mean, is
approximately normal, with mean µ (i.e. ( ) E x µ · ) and variance σ
2
/n (i.e.
2
( ) V x
n
σ
· ), and this approximation becomes better as n becomes larger.
Notes: As in the previous illustrating example, we can see the following modifications:
31
(i) If the population is finite,
2
( ) (1 )
n
V x
N n
σ
· − ; where (1-n/N) is known as the finite
population correction factor. When N is very big, the factor is equal to 1.
(ii) If n is small, say less than 30, the sampling distribution is not so normal. A t-distribution
will be used (discussed later).
In the above example, N=4, n=2,
2
( ) (1 )
n
V x
N n
σ
· − =(1-2/4)(1.6667/2) = 0.4167. If the
population is big (or the sample is drawn with replacement), then
2
( ) V x
n
σ
· =1.6667/2=0.8333.
In this course we assume a big population or sampling with replacement.
Example 1
An electrical firm manufactures light bulbs that have a length of life that is approximately normal
distributed with mean equal to 800 hours and a standard deviation of 40 hours. Find the
probability that a random sample of 16 bulbs will have an average life of less than 775 hours.
Let X be the average life of the 16 bulbs. X ∼ N(
· · µ µ
x 800, · ·
n
x
2
2
σ
σ
16
40
2
)

,
_

¸
¸

< ·

,
_

¸
¸ −
<

· <
16
40
800 775 775
) 775 ( Z P
X
P X P
x
x
x
x
σ
µ
σ
µ

00621 . 0 ) 5 . 2 ( · − < · Z P
Example 2
The mean IQ scores of all students attending a college is 110 with a standard deviation of 10.
(a) If the IQ scores are normally distribution, what is the probability that the score of any one
student is greater than 112?
(b) What is the probability that the mean score in a random sample of 36 students is greater
than 112?
(c) What is the probability that the mean score in a random sample of 100 students is greater
than 112?
Solution
(a) Let X be the student's IQ score. X ∼ N(110,
2
10 )
32
4207 . 0 ) 2 . 0 (
10
110 112
) 112 ( · > ·
,
_

¸
¸ −
> · > Z P Z P X P
(b) Let
1
X be the mean score of a sample of 36 students.
1
X ∼ N(
, 110 · · µ µ
x
36
10
2 2
2
· ·
n
x
σ
σ )
1151 . 0 ) 2 . 1 (
36
10
110 112
) 112 (
1
· > ·

,
_

¸
¸

> · > Z P Z P X P
(c) Let
2
X be the mean score of a sample of 100 students.
2
X ∼ N(
, 110 · µ

100
10
2 2
·
n
σ
)
228 . 0 . 0 ) 2 (
100
10
110 112
) 112 (
2
· > ·

,
_

¸
¸

> · > Z P Z P X P
Estimation
Estimation is the process of using statistics from sample data to estimate the parameters of the
population. A statistic is a random variable which depends on which sample is drawn from a
population.
The followings are some examples
Estimator Population parameter
1.
x
µ
2. s
2
σ
2
3. P P
There are two important properties for an estimator, namely, unbiasedness and efficiency.
Unbiased estimator: An estimator, for example,
x
, is unbiased if and only if ( ) E x µ · .
33
Efficiency: The efficiency of an estimator, for example,
x
, is given by ( ) V x . The smaller the
( ) V x , the more accurate will be the
x
as an estimator.
There are two types of estimate
1. A point estimate is a single-value estimate of a population parameter, for example,
^ ^
; x P p µ · ·
.
2. An interval estimate of a population parameter gives an interval that may contain the true
value of the parameter with a certain probability (i.e. confidence); for example,
Pr( ) 0.99. a b µ < < ·
For a point estimate, both the accuracy and reliability of the estimation are unknown. For an
interval estimate, the width of the interval gives the accuracy and the probability gives the
reliability of the estimation.
Examples 3
(a) The mean and standard deviation for the quality point averages of a random sample of 36
college seniors are calculated to be 2.6 and 0.3, respectively. Find a 95% confidence
interval for the mean of the entire senior class.
(b) How large a sample is required in (a) if we want to be 95% confident of µ is off by less
than 0.05?
Solution
Let
µ
be the mean of the entire senior class.
Given: n = 36, 6 . 2 · x , s = 0.3, (1 - α ) = 0.95 05 . 0 · ⇒α
(a) A 95% confidence interval estimate for the
µ
is

,
_

¸
¸
+ < <

,
_

¸
¸
− ⇒ + < < −
36
3 . 0
96 . 1 6 . 2
36
3 . 0
96 . 1 6 . 2
ˆ ˆ
025 . 0 025 . 0
µ
σ
µ
σ
n
z x
n
z x
698 . 2 502 . 2 < < ⇒ µ
(b) Let
1
n be the required sample size.
To be 95% confident that
µ
if off by less than 0.05 would imply
34
05 . 0
3 . 0
96 . 1 05 . 0
ˆ
1 1
025 . 0

,
_

¸
¸
⇒ ≤

,
_

¸
¸
n n
z
σ
∴ 139 30 . 138
05 . 0
) 3 . 0 )( 96 . 1 (
2
1
≅ ·
1
]
1

¸

≥ n
A summary table for constructing (1 −α )% confidence interval for mean and proportion
Estimating Conditions Formula
Mean
Large samples (n ≥ 30) OR
σ is known
n
Z X
n
Z X
σ
µ
σ
α α
2 2
+ < < −
Mean *
Small samples and σ
unknown
n
s
t X
n
s
t X
2
,
2
,
α
ν
α
ν
µ + < < −
v=n-1
Proportion Large sample
n
q p
Z p
ˆ ˆ
ˆ
2
α
t
Difference of means
Large sample OR
1
σ and
2
σ are known
2
2
1
2
1
2
2 1
ˆ
) (
n n
Z X X
σ σ
α
+ t −
Difference of means
Small sample &
1
σ and
2
σ are unknown, assume
2 1
σ σ ·

2 1
2
,
2 1
1 1
) ( ) (
n n
s t X X
p
+ t −
α
ν
2
2 1
− + · n n ν ,
pooled estimate of sample standard
deviation:
2
) 1 ( ) 1 (
2 1
2
2 2
2
1 1
− +
− + −
·
n n
s n s n
s
p
Difference of means
Small sample &
1
σ and
2
σ are unknown,
assume
1
σ ≠
2
σ
2 2
1 2
1 2
,
2
1 2
( )
s s
X X t
n n
α
ν
− t +
35
1
) / (
1
) / (
) / / (
2
2
2
2
2
1
2
1
2
1
2
2
2
2 1
2
1

+

+
·
n
n s
n
n s
n s n s
υ
Difference of means Paired observations
, / 2
d
v
s
d t
n
α
t
;
1 2 d x x · −
and v=n-1
Difference of proportions Large samples
2
2 2
1
1 1
2
2 1
ˆ ˆ ˆ ˆ
) ˆ ˆ (
n
q p
n
q p
z p p + t −
α
Example 4
The contents of seven similar containers of sulfuric acid are 9.8, 10.2, 10.4, 9.8, 10.0, 10.2 and
9.6 liters. Find a 95% confidence interval for the mean of all such containers, assuming an
approximate normal distribution.
Solution
Let
µ
be the mean of all such containers.
Given: n = 7 ∑
·70 x

· 48 . 700
2
x
10
7
70
· · ·

n
x
x
08 . 0
6
7
70
48 . 700
1
) (
2
2
2
2
·

·

·

n
n
x
x
s
∴ s = 2828 . 0 08 . 0 · ; (1 -
α
) = 0.95 05 . 0 · ⇒α ,
447 . 2
025 . 0 , 6
· t
A 95% confidence interval estimate for the
µ
is

,
_

¸
¸
+ < <

,
_

¸
¸
− ⇒ + < < −
7
2828 . 0
447 . 2 10
7
2828 . 0
447 . 2 10
025 . 0 , 6 025 . 0 , 6
µ µ
n
s
t x
n
s
t x
262 . 10 738 . 9 < < ⇒ µ
Example 5
In a random sample of n = 500 families owning television sets in the city of Hamilton, Canada, it
was found that x = 340 owned color sets. Find a 95% confidence interval for the actual
proportion of families in this city with colour sets.
Let P be the actual proportion of families in this city with colour sets.
36
Given: n = 500, 68 . 0
500
340
ˆ · · ·
n
x
p ,
05 . 0 95 . 0 ) 1 ( · ⇒ · − α α
A 95% confidence interval for P is

,
_

¸
¸
+ < <
,
_

¸
¸

n
q p
z p P
n
q p
z p
ˆ ˆ
ˆ
ˆ ˆ
ˆ
025 . 0 025 . 0
72 . 0 64 . 0
500
) 32 )(. 68 (.
96 . 1 68 . 0
500
) 32 )(. 68 (.
96 . 1 68 . 0 < < ⇒ + < < − ⇒ P P
Examples 6
A standardized chemistry test was given to 50 girls and 75 boys. The girls made an average
grade of 76 with a standard deviation of 6, while the boys made an average grade of 82 with a
standard deviation of 8. Find a 96% confidence interval for the difference µ 1 and µ 2, where µ 1
is the mean score of all boys and µ 2 is the mean score of all girls who might take this test.
Given: 75
1
· n , 50
2
· n , 82
1
· x , 8
1
· s , 76
2
· x , 6
2
· s ,

04 . 0 96 . ) 1 ( · ⇒ · − α α

A 96% confidence interval for
2 1
µ µ − is:
2
2
2
1
2
1
02 . 0 2 1 2 1
2
2
2
1
2
1
02 . 0 2 1
ˆ ˆ
) (
ˆ ˆ
) (
n n
z x x
n n
z x x
σ σ
µ µ
σ σ
+ + − < − < + − −
50
6
75
8
05 . 2 ) 76 82 (
50
6
75
8
05 . 2 ) 76 82 (
2 2
2 1
2 2
+ + − < − < + − − ⇒ µ µ ,
30
1
> n & 30
2
> n , so
1 1
ˆ s · σ &
2 2
ˆ s · σ
57 . 8 43 . 3
2 1
< − < ⇒ µ µ
Example 7
In a batch chemical process, two catalysts are being compared for their effect on the output of the
process reaction. A sample of 12 batches is prepared using catalyst 1 and a sample of 10 batches
was obtained using catalyst 2. The 12 batches for which catalyst 1 was used gave an average
yield of 85 with a sample standard deviation of 4, while the average for the second sample gave
37
an average of 81 and a sample standard deviation of 5. Find a 90% confidence interval for the
difference between the population means, assuming the populations are approximately normally
distributed with equal variances.
Solution
Let
1
µ and
2
µ be the mean population yield using catalyst 1 and catalyst 2, respectively.
Given: 12
1
· n , 10
2
· n , 85
1
· x , 4
1
· s , 81
2
· x , 5
2
· s ,

10 . 0 90 . ) 1 ( · ⇒ · − α α
, 20 2 10 12 2
2 1
· − + · − + · n n ν ,
725 . 1
05 . 0 , 20
· t

pooled estimate of sample standard deviation
2
) 1 ( ) 1 (
2 1
2
2 2
2
1 1
− +
− + −
·
n n
s n s n
s
p
478 . 4
2 10 12
5 ) 1 10 ( 4 ) 1 12 (
2 2
·
− +
− + −
·
A 90% confidence interval for
2 1
µ µ − is:
2 1
05 . 0 , 20 2 1 2 1
2 1
05 . 0 , 20 2 1
1 1
) ( ) (
1 1
) ( ) (
n n
s t x x
n n
s t x x
p p
+ + − < − < + − − µ µ
10
1
12
1
) 478 . 4 )( 725 . 1 ( ) 81 85 (
10
1
12
1
) 478 . 4 )( 725 . 1 ( ) 81 85 (
2 1
+ + − < − < + − − ⇒ µ µ
31 . 7 69 . 0
2 1
< − < ⇒ µ µ
Example 8
The weight of 10 adults selected randomly before and after a certain new diet was introduced was
recorded as follows:
Before (
1
x ) After (
2
x )
Difference
1 76 81 -5
2 60 52 8
3 85 87 -2
4 58 70 -12
5 91 86 5
6 75 77 -2
7 82 90 -8
8 64 63 1
9 79 85 -6
10 88 83 5
38
Find a 98% confidence interval for the mean difference in weight.
Solution
i
d
d
n
·

= -1.6
2
2
( ( 1.6))
40.7
1
i
d
d
s
n
− −
· ·

For v = n-1 = 9;
0.01
2.821 t · .
A 98% confidence interval is
6.38
1.6 (2,821)
10
¸ _
− t

¸ ,
That is 7.29 4.09
d
µ − < <
Example 9
A certain change in a manufacturing procedure for component parts is being considered. Samples
are taken using both the existing and the new procedure in order to determine if the new
procedure results in an improvement. If 75 of 1500 items from the existing procedure were found
to be defective and 80 of 2000 items from the new procedure were found to be defective, find a
90% confidence interval for the true difference in the fraction of defectives between the existing
and the new process.
Solution
Let
1
P and
2
P be the true fraction of defectives of the existing and the new processes,
respectively.
Given: 1500
1
· n , 75
1
· x , 05 . 0
1500
75
ˆ
1
· · p
2000
2
· n , 80
2
· x , 04 . 0
2000
80
ˆ
2
· · p

10 . 0 90 . ) 1 ( · ⇒ · − α α
A 90% confidence interval for
1
P -
2
P is:
2
2 2
1
1 1
05 . 0 2 1 2 1
2
2 2
1
1 1
05 . 0 2 1
ˆ ˆ ˆ ˆ
) ˆ ˆ (
ˆ ˆ ˆ ˆ
) ˆ ˆ (
n
q p
n
q p
z p p P P
n
q p
n
q p
z p p + + − < − < + − −
39
2000
) 96 )(. 04 (.
1500
) 95 )(. 05 (.
64 . 1 ) 01 . 0 (
2000
) 96 )(. 04 (.
1500
) 95 )(. 05 (.
64 . 1 ) 01 . 0 (
2 1
+ + < − < + − ⇒ P P
021697 . 0 001697 . 0
2 1
< − < − ⇒ P P
Lecture 5 - Introduction to Test of Hypothesis
Statistical Hypothesis
Consider the following example:
A manufacturer of sports equipment has developed a new synthetic fishing line that he claims has a
mean breaking strength of 8 kilograms with a standard deviation of 0.5 kilogram. Test the
hypothesis that
µ
= 8 kilograms against the alternative that
µ
≠ 8 kilograms if a random
sample of 50 lines is tested and found to have a mean breaking strength of 7.8 kilograms.
Use a 0.01 level of significance.
When a random sample is drawn from a population (the 50 lines randomly selected), the sample
information can be used assess the validity of some conjecture, or hypothesis. Here
µ
= 8
kilograms is known as the null hypotheis and
µ
≠ 8 kilograms is the alternative hypothesis. They
are complementary to each other, and we need to decide which one to accept on the basis of the
sample result of 50 lines.
Now let us make a 95% confidence interval about the mean breaking strength of the population as
below:
0.5 0.5
7.8 1.96* 7.8 1.96* 0.95
50 50
P µ
¸ _
− < < + ·

¸ ,
;
i.e., (7.6614 7.9386) 0.95 P µ < < · .
As there is a probability of 0.95 that the mean breaking strength is between 7.66 kg and 7.94 kg,
it is highly unlikely that the null hypothesis
µ
= 8 kg is true and hence should be rejected.
There are four possible situations for the above decision making exercise:
0
H is correct
0
H is wrong
Accept
0
H
Correct decision Type 2 error
Reject
0
H
Type 1 error Correct decision
40
We still have a probability of 1-0.95, or 0.05 to reject a true
0
H . We call this probability ‘level of
significance’ or
α
, which is the probability of committing a type 1 error.
The rationale of hypothesis testing is simply outlined as above. There are however some formal
concepts and procedures to conduct the test. The details are put down below.
Some Hypothesis Testing Terminology
1. Null hypothesis, H0
A hypothesis that is held to be true until very strong evidence to the contrary is obtained.
H0 :
0
µ µ ·
2. Alternative hypothesis, H1
It is a hypothesis that is complement to the null hypothesis. Hence it will be accepted if
the null hypothesis is rejected.
1 0
: H µ µ ≠ (two-tail test)
1 0
: H µ µ > (One-tail test)
1 0
: H µ µ < (One-tail test)
In the one-tail test we have some expectation about the direction of the error when the
null hypothesis is wrong, while in the two-tail test we don’t have such expectation.
3. Test statistics
is the value, based on the sample, used to determine whether the null hypothesis should
be rejected or accepted.
4. Critical region
is a region in which if the test statistic falls the null hypothesis will be rejected.
5. Types of error
(a) Type I error: Reject H0 when H0 is true
(b) Type II error: Accept H0 when Ha is true
41
6. The significance level, α
is the probability of committing a type 1 error, i.e., P(Type I error) = α .
The probability of committing a type 2 error is β ; i.e., P(Type II error) = β .
Basic Steps in Testing Hypothesis
1. Formulate the null hypothesis.
2. Formulate the alternative hypothesis.
3. Specify the level of significance to be used.
4. Select the appropriate test statistic and establish the critical region.
5. Compute the value of the test statistic.
6. Conclusion: Reject H0 if the statistic has a value in the critical region, otherwise
accept H0.
42
Tests concerning means
The tests concerning means and proportions are summarized in the following table.
H0 Conditions Test statistic
0
µ µ ·
Large samples (n ≥ 30) OR σ is
known
n
x
z
σ
µ
0

·

0
µ µ · Small samples and σ unknown
n s
x
t
0
µ −
·
with 1 − · n υ
0 2 1
d · −µ µ
Large samples OR
1
σ and
2
σ are
known
2
2
2
1
2
1
0 2 1
) (
n n
d x x
z
σ σ
+
− −
·
0 2 1
d · −µ µ
Small sample &
1
σ and
2
σ are
unknown, assume
2 1
σ σ ·

,
_

¸
¸
+
− −
·
2 1
0 2 1
1 1
) (
n n
s
d x x
t
p
with 2
2 1
− + · n n υ and
2
) 1 ( ) 1 (
2 1
2
2 2
2
1 1 2
− +
− + −
·
n n
s n s n
s
p
if
1
σ =
2
σ but unknown
0 2 1
d · −µ µ
Small sample &
1
σ and
2
σ are
unknown,
assume
1
σ ≠
2
σ
2
2
2
1
2
1
0 2 1
) (
n
s
n
s
d x x
t
+
− −
·
with
1
) / (
1
) / (
) / / (
2
2
2
2
2
1
2
1
2
1
2
2
2
2 1
2
1

+

+
·
n
n s
n
n s
n s n s
υ
43
0 2 1
d · −µ µ Paired observations
n s
d d
t
d
0

·
with 1 − · n υ
p p ·
0
Large sample
0
0 0
ˆ

(1 )
p p
z
p p
n

·

p p
1 2
0 − · Large samples

,
_

¸
¸
+ −

·
2 1
2 1
1 1
) ˆ 1 ( ˆ
) ˆ ˆ (
n n
p p
p p
z
1 1 2 2
1 2
ˆˆ
ˆ
n p n p
p
n n
+
·
+
44
Example 1
A manufacturer of sports equipment has developed a new synthetic fishing line that he claims has a
mean breaking strength of 8 kilograms with a standard deviation of 0.5 kilogram. Test the
hypothesis that
µ
= 8 kilograms against the alternative that
µ
≠ 8 kilograms if a random
sample of 50 lines is tested and found to have a mean breaking strength of 7.8 kilograms.
Use a 0.01 level of significance.
Null hypothesis:
8 · µ
kilograms
Alternative hypothesis:
8 ≠ µ
kilograms
Level of significance: 0.01
Critical region: Z >
58 . 2
005 . 0
· z
or Z <
58 . 2
005 . 0
− · −z
Computation:
n = 50 8 . 7 · x 5 . 0 · σ
828 . 2
50
5 . 0
8 8 . 7
− ·

·

·
n
x
z
σ
µ
Conclusion: As the sample z (= -2.828) falls inside the critical region, so reject the null
hypothesis at 0.01 level of significance and conclude that
µ
is significantly smaller than 8
kilograms.
Example 2
The average length of time for students to register for fall classes at a certain college has been 50
minutes with a standard deviation of 10 minutes. A new registration procedure using modern
computing machines is being tried. If a random sample of 12 students had an average registration
time of 42 minutes with a standard deviation of 11.9 minutes under the new system, test the
hypothesis that the population mean is now less than 50, using a level of significance of (1) 0.05,
and (2) 0.01. Assume the population of times to be normal.
Let
µ
be the population mean time for students to register in the new registration procedure.
(1) Null hypothesis:
50 · µ
minutes
Alternative hypothesis:
50 < µ
minutes
Level of significance: 0.05
Critical region: (n = 12 < 30; and the new σ is unknown, so t-test should be used )
degree of freedom (ν ) = n -1 = 12 -1 =11
45
∴ t <
796 . 1
05 . 0 , 11
− · t
Computation:
n = 12 42 · x s = 11.9

329 . 2
12
9 . 11
50 42
− ·

·

·
n
s
x
t
µ
Conclusion: As sample t (= -2.329) falls inside the critical region, so reject the null
hypothesis at 0.05 level of significance and conclude that
µ
is significantly smaller than
50 minutes.
(2) Identical with those of (1) except the critical region would be replaced by:
718 . 2
01 . 0 , 11
− · <t t
and the corresponding conclusion would be changed as follows:
As sample t (= -2.329) falls outside the critical region, so reject the alternative hypothesis
at 0.01 level of significance and conclude that
µ
is not highly significantly smaller than
50 minutes.
Example 3
An experiment was performed to compare the abrasive wear of two different laminated materials.
Twelve pieces of material 1 were tested, by exposing each piece to a machine measuring wear.
Ten pieces of material 2 were similarly tested. In each case, the depth of wear was observed.
The samples of material 1 gave an average (coded) wear of 85 units with a standard deviation of
4, while the samples of material 2 gave an average of 81 and a standard deviation of 5. Test the
hypothesis that the two types of material exhibit the same mean abrasive wear at the 0.10 level of
significance. Assume the populations to be approximately normal with equal variances.
Let
1
µ and
2
µ be the mean abrasive wear of material 1 and 2 respectively.
Null hypothesis:
2 1
µ µ · , i.e. 0
2 1
· − µ µ
Alternative hypothesis:
2 1
µ µ ≠ , i.e. 0
2 1
≠ − µ µ
Level of significance: 0.10
Critical region: (As both
1
n and
2
n are smaller than 30 and their standard deviations are
unknown, so t-test has to be used.)
20 2 10 12 2
2 1
· − + · − + · n n ν
, ∴
725 . 1
05 . 0 , 20
· >t t
or
725 . 1
05 . 0 , 20
− · − < t t
46
Computation:
12
1
· n 85
1
· x 4
1
· s
10
2
· n 81
2
· x 5
2
· s
05 . 20
2 10 12
5 ) 9 ( 4 ) 11 (
2
) 1 ( ) 1 (
2 2
2 1
2
2 2
2
1 1 2
·
− +
+
·
− +
− + −
·
n n
s n s n
s
p
086 . 2
10
1
12
1
05 . 20
0 ) 81 85 (
1 1
) ( ) (
2 1
2 1 2 1
·
+
− −
·
+
− − −
·
n n
s
x x
t
p
µ µ
Conclusion: As the sample t (=2.086) falls inside the critical region, so reject the null hypothesis
at 0.10 level of significance and conclude that the mean abrasive wear of material 1 is
significantly higher than that of the material 2.
Example 4
Five samples of a ferrous-type substance are to be used to determine if there is a difference
between a laboratory chemical analysis and an X-ray fluorescence analysis of the iron content.
Each sample was split into two sub-samples and the two types of analysis were applied.
Following are the coded data showing the iron content analysis:
Sample
Analysis 1 2 3 4 5
x-ray 2.0 2.0 2.3 2.1 2.4
Chemical 2.2 1.9 2.5 2.3 2.4
Assuming the populations normal, test at the 0.05 level of significance whether the two methods
of analysis give, on the average, the same result.
Let
1
µ and
2
µ be the mean iron content determined by the laboratory chemical analysis and
X-ray fluorescence analysis respectively; and
D
µ be the mean of the population of differences of paired measurements.
Null hypothesis:
2 1
µ µ · or 0 ·
D
µ
Alternative hypothesis:
2 1
µ µ ≠ or 0 ≠
D
µ
Level of significance: 0.05
Critical region: (As n = 5 < 30, so t-test should be used.)
47

776 . 2
025 . 0 , 4
· >t t
or
776 . 2
025 . 0 , 4
− · − < t t
Computation:
Sample
Analysis 1 2 3 4 5
x-ray 2.0 2.0 2.3 2.1 2.4
Chemical 2.2 1.9 2.5 2.3 2.4
i
d -0.2 0.1 -0.2 -0.2 0
2
i
d
0.04 0.01 0.04 0.04 0
5 . 0
5
1
− ·

· i
i
d 13 . 0
5
1
2
·

· i
i
d
1 . 0
5
5 . 0
5
− ·

· ·

d
d
( )
02 . 0
) 4 )( 5 (
) 5 . 0 ( ) 13 . 0 )( 5 (
) 1 (
2
2
2
2
·
− −
·

·
∑ ∑
n n
d d n
s
d
5811 . 1
5
02 . 0
0 ) 1 . 0 (
− ·
− −
·

·
n
s
d
t
d
D
µ
Conclusion: As the sample t (=-1.5811) falls outside the critical region, so reject the alternative
hypothesis at 0.05 level of significance and conclude that there is no significant difference in the
mean iron content determined by the above two analyses.
Tests Concerning Proportions
Example 5
48
A manufacturing company has submitted a claim that 90% of items produced by a certain process
are non-defective. An improvement in the process is being considered that they feel will lower
the proportion of defective below the current 10%. In an experiment 100 items are produced with
the new process and 5 are defective. Is this evidence sufficient to conclude that the method has
been improved? Use a 0.05 level of significance.
Let P be the proportion of defective product in the new production process.
Null hypothesis: P = 0.1
Alternative hypothesis: P < 0.1
Level of significance: 0.05
Critical region:
64 . 1
05 . 0
− · − < z Z

Computation:
n = 100 x = 5 05 . 0
100
5
ˆ · · ·
n
x
p
667 . 1
100
) 9 . 0 )( 1 . 0 (
1 . 0 05 . 0
) 1 (
ˆ
− ·

·

·
n
P P
P p
z
Conclusion: As the sample z (=-1.667) falls inside the critical region, so reject the null hypothesis
at 0.05 level of significance and conclude that P is significantly smaller than 0.1. That is, the
production method has been improved in lowering the proportion of defective below the current
10%.
Example 6
A vote is to be taken among the residents of a town and the surrounding country to determine
whether a proposed chemical plant should be constructed. The construction site is within the
town limits and for this reason many voters in the country feel that the proposal will pass because
of the large proportion of town voters who favor the construction. To determine if there is a
significant difference in the proportion of town voters and county voters favoring the proposal, a
poll is taken. If 120 of 200 town voters favor the proposal and 240 of 500 county residents favor
it, would you agree that the proportion of town voters favoring the proposal is higher than the
proportion of county voters? Use a 0.025 level of significance.
Let
1
P and
2
P be the proportions of town voters and country voters, respectively, favouring
the proposal.
Null hypothesis:
2 1
P P · or 0
2 1
· −P P
Alternative hypothesis:
2 1
P P > or 0
2 1
> −P P
Level of significance: 0.025
49
Critical region:
96 . 1
025 . 0
· >z Z
Computation:
200
1
· n 120
1
· x 6 . 0
200
120
ˆ
1
1
1
· · ·
n
x
p
500
2
· n 240
2
· x 48 . 0
500
240
ˆ
2
2
2
· · ·
n
x
p
514 . 0
500 200
) 48 . 0 )( 500 ( ) 6 . 0 )( 200 ( ˆ ˆ
ˆ
2 1
2 2 1 1
·
+
+
·
+
+
·
n n
p n p n
p
870 . 2
500
1
200
1
) 486 . 0 )( 514 . 0 (
0 ) 48 . 0 6 . 0 (
1 1
) ˆ 1 ( ˆ
) ( ) ˆ ˆ (
2 1
2 1 2 1
·
1
]
1

¸

+
− −
·
1
]
1

¸

+ −
− − −
·
n n
p p
P P p p
z
Conclusion: As sample z (=2.870) falls inside the critical region, so reject the null hypothesis at
0.025 level of significance and conclude that the proportion of town voters favouring the proposal
is significantly larger than that of the country voters.
50
Chapter 7 - Chi-square Tests
There are two types of chi-square tests: goodness-of-fit test and tests for independence.
Goodness-of-fit Test
A test to determine if a population has a specified theoretical distribution. The test is based on
how good a fit we have between the frequency of occurrence of observations in an observed
sample and the expected frequencies obtained from the hypothesized distribution.
Theorem: A goodness-of-fit test between observed and expected frequencies is based on the
quantity
χ
test
2
2
·

( ) O E
E
i i
i
where χ
test
2
is a value of the random variable whose sampling distribution is
approximated very closed by the Chi-square distribution,
Oi is the observed frequency of cell i, and Ei is the expected frequency of cell i.
The number of degrees of freedom in a Chi-square goodness-of-fit test is equal to
the number of cells minus the number of quantities obtained from the observed
data that are used in the calculations of the expected frequencies.
For a level of significance equal to α χ χ
α
,
test
2 2
> constitutes the critical region. The decision
criterion described here should not be used unless each of the expected frequencies is at least
equal to 5.
Example 1
Consider the tossing of a die 120 times.
Faces
1 2 3 4 5 6
Observed 20 22 17 18 19 24
Expected
By comparing the observed frequencies with the expected frequencies, one has to decide whether
the die is fair die or not.
Null hypothesis: the die is a fair die, i.e.
6
1
) ( · · i X P for i = 1, 2, 3, 4, 5, and 6
Alternative hypothesis: the die is not a fair die
51
Level of significance: 0.05
Critical region: 5 1 6 1 · − · − · n ν ; ∴
07 . 11
2
05 . 0 , 5
2
· > χ χ
Computation:
Expected value = 20 )
6
1
( 120 ) ( · · ·i X nP
i 1 2 3 4 5 6
Observed
) (
i
O 20 22 17 18 19 24
Expected
) (
i
E

20 20 20 20 20 20
i i
E O − 0 2 -3 -2 -1 4

·

·
6
1
2
2
) (
i i
i i
E
E O
χ
7 . 1
20
4
20
) 1 (
20
) 2 (
20
) 3 (
20
2
20
0
2 2 2 2 2 2
· +

+

+

+ + ·
Conclusion: As the sample
2
χ (= 1.7) falls outside the critical region, so reject the alternative
hypothesis and conclude that the die is a fair die.
Example 2
The following distribution of battery lives may be approximated by the normal distribution.
Class boundaries Oi z-value p-value Ei
1.45 - 1.95 2
1.95 - 2.45 1
2.45 - 2.95 4
2.95 - 3.45 15
3.45 - 3.95 10
3.95 - 4.45 5
4.45 - 4.95 3
Chi-squared test can be applied to test whether the above frequency distribution can be
approximated by a normal distribution or not.
Null hypothesis: the distribution can be approximated by a normal distribution
Alternative hypothesis: the distribution cannot be approximated by a normal distribution
52
Level of significance: 0.05
Critical region:
2
05 . 0 ,
2
ν
χ χ >
where 3 − · n ν , and n is the number of cells.
Computation:
For finding the expected values, the mean and standard deviation of the frequency distribution
have to be found first.
Class boundaries
) (
i
f
Oi
Class mark
(
i
x
)
1.45 - 1.95 2 1.7
1.95 - 2.45 1 2.2
2.45 - 2.95 4 2.7
2.95 - 3.45 15 3.2
3.45 - 3.95 10 3.7
3.95 - 4.45 5 4.2
4.45 - 4.95 3 4.7
n = 40 ∑
· 5 . 136 fx 75 . 484
2
·

fx
4125 . 3
40
5 . 136
· · ·

f
fx
x
( )
( )
6969 . 0
39
40
5 . 136
75 . 484
1
/
2
2
2
·

·

·
∑ ∑
n
n fx fx
s
1
]
1

¸

< <

· −
6969 . 0
4125 . 3
6969 . 0
4125 . 3
i i
U
Z
L
value z
;

where
i
L
and
i
U
are the Lower and Upper Boundaries of the ith class.
1
]
1

¸

< <

· −
6969 . 0
4125 . 3
6969 . 0
4125 . 3
i i
U
Z
L
value p
53
1
]
1

¸

< <

·
6969 . 0
4125 . 3
6969 . 0
4125 . 3
) 40 (
i i
i
U
Z
L
P E
Class boundaries Oi z-value p-value Ei
1.45 - 1.95 2 Z<-2.10 .0179 0.716
1.95 - 2.45 1 -2.10<Z<-1.38 .0659 2.636
2.45 - 2.95 4 -1.38<Z<-0.66 .1708 6.832
2.95 - 3.45 15 -0.66<Z<0.05 .2653 10.612
3.45 - 3.95 10 0.05<Z<0.77 .2595 10.38
3.95 - 4.45 5 0.77<Z<1.49 .1525 6.1
4.45 - 4.95 3 1.49<Z .0681 2.724
In order to satisfy the rule that the expected value in each cell is larger than or equal to 5, we have
to combine the first three classes in to one cell and the last two classes into another cell. As such,
the number of cells (n) is 4.
824 . 8
) 824 . 8 8 (
38 . 10
) 38 . 10 10 (
612 . 10
) 612 . 10 15 (
184 . 10
) 184 . 10 7 ( ) (
2 2 2 2 2
2

+

+

+

·

·

E
E O
χ
901 . 2 ·
Conclusion: Since
841 . 3
2
05 . 0 , 1
· χ
, so the sample
2
χ (= 2.901) falls outside the critical region.
As such, reject the alternative hypothesis and conclude that the distribution of battery lives can be
approximated by a normal distribution.
Test for Independence
The Chi-square test procedure can also be used to test the hypothesis of independence of two
variables/attributes. The observed frequencies of two variables are entered in a two-way
classification table, or contingency table.
Remark: The expected frequency of the cell in the i
th
row and j
th
column in the contingency
table
E
ij
·
(total of row i) *(total of column j)
grand tota l
The degrees of freedom for the contingency table is equal to (r −1) (c −1) where r is the number of
rows and c is the number of columns in the table.
54
Example 3
Suppose that we wish to study the relationship between grade point average and appearance.
Appearance 1 2 3 4 Totals
attractive 14 ( ) 11 ( ) 10 ( ) 5 ( ) 40
ordinary 10 ( ) 16 ( ) 16 ( ) 14 ( ) 56
unattractive 3 ( ) 4 ( ) 7 ( ) 10 ( ) 24
Totals 27 31 33 29 120
Null hypothesis: There is no relationship between grade point average and appearance. That is,
the two characteristics are independent.
Alternative hypothesis: There is a relationship between grade point average and appearance. That
is, the two characteristics are not independent.
Level of significance: 0.05
Critical region:
2
05 . 0 ,
2
ν
χ χ >
, where ν = (r -1)(c - 1)
Computation:
Appearance 1 2 3 4 Totals
attractive 14
(9)
11
(10.33)
10
(11)
5
(9.67)
40
ordinary 10
( 12.6)
16
( 14.47)
16
(15.4)
14
(13.53)
56
unattractive 3
(5.4)
4
(6.2)
7
(6.6)
10
(5.8)
24
Totals 27 31 33 29 120
67 . 9
) 67 . 9 5 (
11
) 11 10 (
33 . 10
) 33 . 10 11 (
9
) 9 14 ( ) (
2 2 2 2 2
2

+

+

+

·

·

E
E O
χ

53 . 13
) 53 . 13 14 (
4 . 15
) 4 . 15 16 (
47 . 14
) 47 . 14 16 (
6 . 12
) 6 . 12 10 (
2 2 2 2

+

+

+

+
55
818 . 10
8 . 5
) 8 . 5 10 (
6 . 6
) 6 . 6 7 (
2 . 6
) 2 . 6 4 (
4 . 5
) 4 . 5 3 (
2 2 2 2
·

+

+

+

+
Conclusion: Since
596 . 12
2
05 . 0 , 6
· χ
, so sample
2
χ (=10.818) falls outside the critical region.
So reject the alternative hypothesis and conclude that there is no evidence to support there is
relationship between grade point average and appearance.
Test for Homogeneity
To test the hypothesis that several population proportions are equal.
Remark: The approach for the test of homogeneity is the same as for the test of
independence of variables/attributes.
Example 4
A study of the purchase decisions for 3 stock portfolio managers, A, B, and C was conducted to
compare the rates of stock purchases that resulted in profits over a time period that was less than
or equal to one year. One hundred randomly selected purchases obtained for each of the managers
showed the results given in the table. Do these data provide evidence of differences among the
rates of successful purchases for the three portfolio managers? Test with . 05 . 0 · α
Result
Manager
A B C
Purchases show profit 63 71 55
Purchase do no show profit 37 29 45
Total 100 100 100

Null hypothesis: the rates of stock purchases that resulted in profit were the same for the three
stock portfolio managers
Alternative: their rates were not all the same
Level of significance: 0.05
Critical region:
2 ) 1 3 )( 1 2 ( · − − · ν
; ∴
991 . 5
2
05 . 0 , 2
2
· > χ χ
Computation:
Result
Manager
A B C Total
Purchases show profit 63
(63)
71
(63)
55
(63)
189
56
Purchase do no show profit 37
(37)
29
(37)
45
(37)
111
Total 100 100 100 300
49 . 5
37
) 37 45 (
37
) 37 29 (
37
) 37 37 (
63
) 63 55 (
63
) 63 71 (
63
) 63 63 (
2 2 2 2 2 2
2
·

+

+

+

+

+

· χ
Conclusion: As the sample
2
χ (= 5.49) falls outside the critical region so reject the alternative
hypothesis and conclude that there is no sufficient evidence to support the rates of purchases
resulted in profit of the three portfolio managers were different.

57

A useful method for summarizing a set of data is the construction of a frequency table, or a frequency distribution. That is, we divide the overall range of values into a number of classes and count the number of observations that fall into each of these classes or intervals. The general rules for constructing a frequency distribution are: (i) (ii) There should not be too few or too many classes. Insofar as possible, equal class intervals are preferred. But the first and last classes can be open-ended to cater for extreme values.

In example 1, the sample size is 100 and the range for the data is 113 (137 - 24). A frequency distribution with six classes is appropriate and it is shown below. Frequency distribution of household town gas consumption Town gas monthly consumption ( in cubic metres) 20 - 39 40 - 59 60 - 79 80 - 99 100 - 119 120 - 139 Total Number of households 5 15 25 30 18 7 100

Class limits: are the numbers that typically serve to identify the classes in a listing of a frequency distribution. Thus, in the above frequency distribution, for the class whose frequency is 30, its lower class limit is 80 and upper class limit is 99. As contrasted to a class limit, a class boundary is the precise point that separates one class from another, rather than being a value indicated in one of the classes. A class boundary is typically located midway between the upper limit of a class and the lower limit of the next higher class adjoining it. Therefore the class boundary separating the class 60-79 and the class 80-99 is halfway between 79 and 80, that is, at the point 79.5. Class interval: is the width of a class. The class interval of a class is computed by subtracting the lower limit (boundary) of the class from the lower limit (boundary) of the next class. Class midpoint or class mark: is the point dividing the class into equal halves on the basis of class interval. This point can be obtained by adding the lower and upper limits (boundaries) of a class and dividing by 2. Relative frequency of a class: is the frequency of the class divided by the total frequency of the distribution. Cumulative frequency distribution: shows the number of items of a series that are less than (or more than) certain specified values.
2

Measure of Central Tendency A value that would describe the 'centre' of a distribution would be visually located near the spot where most of the data seem to be concentrated. Consequently, values that fulfil this role are called measures of central tendency. The most common measures of the central tendency of a data set are arithmetic mean or simply as mean, median and mode. The mean of a set of numerical data is the sum of the set divided by the number of observations, that is, their average. The median of a distribution is the value which divides the distribution so that an equal number of values lie on either side of it, i.e., half of the items have values smaller or equal to it and half of the items have values larger or equal to it. The mode of a set of numerical data is the value which occurs most frequently.

Example 1 (calculating mean, median and mode for individual data) The following table shows the hourly wage rates of eight sampled construction workers. Worker i Hourly wage rate ( x i ) 1 \$35 2 38 3 46 4 60 5 65 6 69 7 72 8 78

Mean

(x) =
=

∑x
i =1

8

i

(=

8

x1 + x 2 + x3 + x 4 + x5 + x6 + x7 + x8 ) 8

463 = 57 .875 (\$) 8 n +1 9 = = 4.5 th 2 2

Location of the median:

Median =

x 4 + x5 60 + 65 = = 62.5 (\$) 2 2

Mode: the sample size is too small, mode cannot be identified. Example 2 (calculating mean, median and mode for grouped data) The following table shows the daily wages of a random sample of construction workers. Calculate its mean, median and mode.
3

Daily Wages (\$) 200 - 399 400 - 599 600 - 799 800 - 999 1000 - 1199 1200 - 1399 Total

Number of Workers 5 15 25 30 18 7 100

Solution Number of Daily Wages (\$) 200 - 399 400 - 599 600 - 799 800 - 999 1000 - 1199 1200 - 1399 Total
6

Workers
fi

Class Mark
xi

f i xi

5 15 25 30 18 7 100

299.5 499.5 699.5 899.5 1,099.5 1,299.5

1,497.5 7,492.5 17,489.5 26,985.5 19,791.0 9,096.5 82,350.0

Mean ( x ) =

∑fx
i =1 6 i

i

∑f
i =1

=

82,350.0 = 823.5 (\$) 100

i

Daily Wages (\$) 200 - 399 400 - 599 600 - 799 800 - 999 1000 - 1199 1200 - 1399

Number of Workers
fi

Cumulative Frequency
Fi

5 15 25 30 18 7

5 20 45 75 93 100
4

Total

100

As 0.5n = 0.5(100) = 50, so the median lies in the 4th class. Median = L4 +

0.5n − F3 (c 4 ) f4

where

L is the lower class boundary,

c

is the class interval.

= 799 .5 +

0.5(100 ) − 45 ( 200 ) = 832 .8 (\$) 30
Number of Class Interval
ci

Relative Density

Daily Wages (\$)

Workers
fi

f i' =

fi (200 ) ci
5 15 25 30 18 7

200 - 399 400 - 599 600 - 799 800 - 999 1000 - 1199 1200 - 1399 Total

5 15 25 30 18 7 100

200 200 200 200 200 200

As f 4' = 30 is the largest relative density, so mode lies in the 4th class.

f 4' − f 3' (c 4 ) Mode = L4 + ( f 4' − f 5' ) + ( f 4' − f 3' )
= 799 .5 + 30 − 25 ( 200 ) = 858 .3 (\$) (30 −18 ) + (30 − 25 )

Advantages and disadvantages of each measure Mean Advantages: (i) (ii) (iii) All values in the distribution are used in its calculation, so it can be regarded as more representative than the other two measures. Its method of calculation is simple and most people understand the meaning of its result. Its result can easily be used in further analysis.

5

Its result will not be affected by extreme values and open end classes. Disadvantages: (i) (ii) How to select a suitable measure (i) (ii) (iii) Always select the mean whenever there is no special reason for choosing the other two measures. 6 . If data are not grouped. As such. Select the mode if integral result is preferred as in cases the data are in ordinal scales. (ii) Median Advantage: Disadvantage: Mode Advantages: (i) (ii) Its result will not be affected by extreme values and open end classes. It is difficult to obtain an accurate estimate of the mode if the values are classified into a frequency distribution. it can be determined easily. It has to be supplemented by other statistics because it does not reflect the distribution in the way that the mean does. including all values. In case of open end classes. mean can be calculated only if their class marks are determined. If such classes contain a large proportion of the values. Select the median is the distribution consists of substantial amount of extreme large or small values. It has to be supplemented by other statistics. then the mean may be subjected to substantial error. that is. its result may be rather lower or higher than the bulk of the values and becomes unrepresentative.Disadvantages: (i) Its result can be easily distorted by extreme values.

in grams.Measure of data variation (variability) A measure of central tendency is almost never. in comparison with company B. We say that the variability of the observations is smaller for company A. by itself. standard deviation and variance. Example 1 Consider the following measurements. for two samples of strawberry jam bottled by companies A and B: Sample for Company A Sample for Company B 31 28 32 29 32 32 33 35 32 36 Both samples have the same mean. a measure of the amount of variation in the data. its result is unstable. 32 grams. So whenever the sample size is over 10. This is particularly true if the sample size is large. Read our textbook if interested) The range of a set of numbers is the difference between the largest and the smallest number in the set. The most important measures of variability or dispersion are the range.\$35 = \$43. in addition. sufficient to provide an adequate summary of the characteristics of a set of data. we seldom choose to use range to indicate variability of the data. We will usually require. bottles strawberry jam with a more consistent content. Therefore in buying strawberry jam we would feel more confident that the bottle we select will be closer to the advertised average content if we buy from company A. Worker i Hourly wage rate ( x i ) 1 \$35 2 38 3 46 4 60 5 65 6 69 7 72 8 78 The range is \$78 . We shall not study these measures. 7 . It is obvious that company A. (There are some other measures like quartile deviation and percentiles. Example 2 (For individual data) The following table shows the hourly wage rates of eight sampled construction workers. Though range is simple and can be obtained easily. mean deviation.

125 6 69 11. we can use the standard deviation instead.226(\$) 7 Variance is the square of the standard deviation.125 7 72 14.875) 2 = 16.125 8 78 20.875 4 60 2. Worker i Hourly wage rate ( x i ) xi − x = x i −57 . 8 . Standard deviation of a population ( σ) is the square root of the average of the squared distances of the observations from the mean. To avoid this pitfall.875 8 = (\$) 109 .Mean deviation is the average of the absolute deviation of the numerical data from their mean. where µ is the population mean N To compute the sample standard deviation (s ) we use the above formula.656 8 The mean deviation is a good measure to show the extent of variation of the data in a distribution.875 1 \$35 22.25 = 13. s= ∑ (x − x ) i =1 i n 2 n −1 1 \$35 2 38 3 46 4 60 5 65 6 69 7 72 8 78 Total 463 Worker i Hourly wage rate ( x i ) s= ( x − 35. However. it would give rise to some unnecessary tedious mathematical problem as a result of its absolute value term.875 2 38 19.125 5 65 7. σ = ∑ (x i =1 N i − µ )2 . when this measurement is used in further analysis.125 Mean deviation ∑x i =1 8 i − 57.875 3 46 11. replacing µ by x and N by n −1 .

099.599 600 .399 400 .5 1. Calculate its mean deviation.5 499.5 699.999 1000 .799 800 .860 3. Daily Wages (\$) 200 .799 800 .s2 = ∑ (x − x ) i =1 i n 2 n −1 Example 3 (for grouped data) The following table shows the daily wages of a random sample of construction workers.999 1000 .160 = 211.60 (\$) 100 Number of Daily Wages (\$) 9 Workers Class Mark f i ( xi − x) 2 .968 3. and standard deviation.5 899. variance.1399 Total Workers fi Number of Workers 5 15 25 30 18 7 100 Class Mark xi f i xi − x = f i xi −823 .1399 Total Solution Number of Daily Wages (\$) 200 .100 2.299.620 4.599 600 .5 2.5 5 15 25 30 18 7 100 299.5 1.332 21.399 400 .1199 1200 .280 4.1199 1200 .160 Mean deviation = ∑f i =1 6 i 6 xi − x i ∑f i =1 = 21.

77 = 255.5 699. 276. and (b) average values of two distributions may be widely dissimilar. Solution In comparison Average magnitude Distribution II > I Variation Chapter Two .5 1.5 1.880 1. gamblers called upon the I > II Reason x II = 150 > x I = 100 CVI = 20 24 100% = 20% > CVII = 100% = 16% 100 150 10 .399 400 .299.599 600 .49 Comparison of the variation of two distributions The values of the standard deviations cannot be used as the bases of the comparison because: (a) units of measurements of the two distributions may be different.371.1399 Total 2 Variance ( s ) = 5 15 25 30 18 7 100 299.574.Probability Introduction and concepts “Perhaps it was man’s unquenchable thirst for gambling that led to the early development of probability theory. 372. V The correct measure that should be used is the coefficient of variation (C ) .400 6462400 = 65.1199 1200 .640 384. Worker's Type I II Mean \$100 \$150 Daily Wages Standard deviation \$20 \$24 Compare these two daily wages distributions.280 1.099.999 1000 .5 1. In an effort to increase their winnings.799 800 .168 1.77 99 Standard deviation = 65276.5 899.032 6.586.5 499.400 173.462.fi xi 200 . s CV = 100% x Example 4 The following table shows the summary statistics for the daily wages of two types of workers.

To find the probability of an event we need to count the number of outcomes of the event and the number of all possible outcomes of the experiment. Event: is a subset of a sample space. weather forecasting.E. and then to divide the former by the latter. There are three approaches to understand probability. there are k possible outcomes which are equally likely. Some Basic Concepts Sample space: is a set of all possible outcomes of an experiment. In our course. Each item is inspected and classified defective (D) or non-defective (N). Suppose in a trial of an experiment. This approach is too abstract and usually used by mathematicians.mathematicians to provide optimum strategies for various games of chance. DDD Its sample space is =  DNN DDN NDN DND NND NDD   NNN  Example 2 11 . one thousand times. temperature. business. Introduction to Statistics Probability is the basis upon which the discipline of statistics has been developed and applied in many fields associated with chance occurrences such as politics. rainfall. Probability may be taken as a tool with which we may solve problems involving uncertainties. Some counting rules Example 1 Three items are selected at random from a manufacturing process. To cite some examples: travelling time. The probability of the occurrence of an outcome is therefore 1/k. the probability of having a head is ½. Thus in throwing a coin. The third approach is very mathematical. probability may be taken as a relative frequency.from Walpole R.” ---. Hence the following counting rules may be helpful. As such the probability of an aeroplane arriving its destination on time may be taken as the proportion of times the aeropline has been on time in the past. In fact uncertainty is a basic element of human experiences. we shall adopt this approach but the empirical approach is always useful in giving us some intuition to understand the problem. etc. number of customers. say. A number of axioms have been set up and from these some theorems of probability have been developed. share price movement. length of our life. In the empirical approach. and scientific research.

400 Example 4 Find the possible permutations (the number of ways where sequence of the letters is counted) from 3 letters A... B.( 2)(1) = n! • The number of permutations of n distinct objects taken r at a time is 12 . How many different licence plates can be printed? 1st Letter Number of Choices A-Z (26) 2nd Letter A-Z (26) 1st Digit 1-9 (9) 2nd Digit 0-9 (10) 3rd Digit 0-9 (10) Number of different licence plates that can be printed is (26)(26)(9)(10)(10) = 608. Consider the following tree diagram: • The number of permutations of n distinct objects is n Pn = (n)( n −1)( n − 2).The event that the number of defectives in above example is greater than 1. Example 3 Suppose a licence plate containing two letters following by three digits with the first digit not zero.. C. Its sample space is = {DDD DDN DND NDD} The probability of the event is 4/8 or ½..

... The answer = 5! 3!(5 −3)! Example 6 13 ... n 2 are alike of the second kind.. The number of 3-letter words formed from 5 letters is 5 5! P3 = = 60 (5 −3)! • The number of distinct permutations of n objects of which n1 are alike of the first kind... ..( 2)(1) ) (n − r )( n − r −1).g.( 2)(1) = n! (n − r )! e... + n k = n is ( n1!)( n 2 !)..( n − r + 2)( n − r +1) = ( (n)( n −1).. 2 N and 1 Z) The number of 7-letter words that can be formed is 7! =4 0 2 (1 )( 3!)( 2!)(1 ) ! ! • The number of combinations (number of ways where sequence is not counted) of n distinct objects taken r at a time is n Cr = n! r!( n − r )! Find the possible combinations of 5 distinct objects taken 3 at a time.( n − r + 2)( n − r +1) )( (n − r )( n − r −1). 3 E.... n k are alike of the kth kind and n! n1 + n 2 + ..n Pr = (n)( n −1). ∴ The answer = 5 P5 5! = 3! 3! Example 5 How many 7-letter words can be formed using the letters of the word 'BENZENE'? (there are 1 B.( n k !) Find the possible permutations of the following 5 letters: A A A B C There are five objects of which three are alike.

3 of which are rotten. the probability of having all 3 rotten eggs is ( 3 C3 )( 5 C0 ) 8 C3 = 1 56 (c) The probability of having no rotten egg is ( 3 C 0 )( 5 C3 ) 8 C3 = 10 5 = 56 28 14 . (a) Exactly two eggs are rotten. (c) No egg is rotten. Solution: (a) The 8 eggs can be divided into 2 groups. Three eggs are picked at random. 3 rotten eggs as the first group and 5 good eggs as the second group. Getting 2 rotten eggs in 3 randomly selected eggs can occurred if we select randomly 2 eggs from the first group and 1 egg from the second group. The number of this outcome is ( 3 C 2 )( 5 C1 ) = 15 Total number of possible outcomes of selecting 3 eggs randomly from the total 8 eggs is 8 C 3 = 56 . Find the probabilities of the following events. namely. Thus the probability of having exactly two rotten among the 3 randomly selected eggs is ( 3 C 2 )( 5 C1 ) 8 C3 = 15 56 (b) Similarly. (b) All eggs are rotten.The number of 3-person committees that can be formed from a group of 4 persons is 4 C3 = 4! =4 3!( 4 − 3)! Example 7 A box contains 8 eggs.

31 9 3 5 45 Example 8 15 . Solution 144 4 = 180 5 Let E be the event of passing English. it is the set of elements that belong to A A ∩ B is the intersection of two sets A and B. A ∪ B is the union of two sets A and B. 9 3 P( M ∪ E ) = 4 5 As P( M ∪ E ) = P( E ) + P ( M ) − P( M ∩ E ) ∴ P ( M ∩ E ) = P ( E ) + P ( M ) − P( M ∪ E ) = 4 2 4 14 + − = = 0. Addition Rule: For any events that are not mutually exclusive P ( A ∪ B ) = P ( A) + P ( B ) − P ( A ∩ B ) where or to B or to both. to A and B. It is given that: P ( E ) = 4 2 . Their results were as follows: Number of students passing English = 80 Number of students passing Mathematics = 120 Number of students passing at least one subject = 144 Then we can rewrite the above results as: Probability that a randomly selected student passed English = 80 4 = 180 9 120 2 = 180 3 Probability that a randomly selected student passed Mathematics = Probability that a randomly selected student passed at least one subject = Find the probability that a randomly selected student passed both subject.Rules of probability The following rules may help us to find the probability of an event. P( M ) = . and M be the event of passing Mathematics. it is the set of elements that are common Illustrative example 180 students took examinations in English and Mathematics.

and B be the event of getting an ace..6.5.A and B are mutually exclusive = 6 2 2 + = 36 36 9 If A and A' are complementary events then P ( A) =1 − P ( A' ) Example 9 A coin is tossed six times in succession. What is the probability that at least one head occurs? Let A be the number of heads occurs in six successive tosses.5} Let A be the event of getting a total of '7'. The probability of getting a total of '7' or '11' is P ( A ∪B ) . P ( A ∪ B ) = P ( A) + P ( B ) What is the probability of getting a total of '7' or '11' when a pair of dice are tossed? Solution Total number of possible outcomes = (6)(6) = 36 Possible outcomes of getting a total of '7' :{1.A card is drawn from a complete deck of playing cards.6. What is the probability that the card is a heart or an ace? Solution Let A be the event of getting a heart. P ( A ∪ B ) = P ( A) + P ( B ) − P ( A ∩ B ) = 13 4 1 16 4 + − = = 52 52 52 52 13 For mutually exclusive events.3.1} Possible outcomes of getting a total of '11' : {5. 6.2.. and B be the event of getting a total of '11'. 2. 5.4. 4. P ( A ≥ 1) = 1 − P ( A = 0) 16 . The probability that the card is a heart or an ace is P ( A ∪B ) . 6. 3. P ( A ∪ B ) = P ( A) + P ( B ) − P ( A ∩ B ) = P ( A) + P ( B ) .

65 = = 0. and B be the event of using ketchup. provided P(A) > 0.8667 P ( A) 0.8125 P( B) 0. and 65% use both.80 P ( A ∩ B ) 0. the conditional probability of B given that event A has occurred is defined as P( B / A) = P( A ∩ B) . 17 . P ( B ) = 0.75 (b) P(a mustard-user uses ketchup) = P ( B / A) = Multiplicative Rule P ( A ∩ B ) = P ( A) P ( B / A) or = P ( B ) P ( A / B ) Statistically Independence: the occurrence or non-occurrence of one event has no effect on the probability of occurrence of the other event.80 . denoted by P ( A / B ) is defined as P( A ∩ B) P( B) P( A / B ) = provided that P(B) > 0. P ( A) Example 10 A hamburger chain found that 75% of all customers use mustard. when ordering a hamburger.=1 − 1 63 ( 2 )( 1 2 )( 1 2 )( 1 2 )( 1 2 )( 1 2 ) = 64 Conditional Probability Let A and B be two events. 80% use ketchup. What are the probabilities that: (a) a ketchup-user uses mustard? (b) a mustard-user uses ketchup? Solution Let A be the event of using mustard. P ( A ∩B ) = 0. The conditional probability of event A given that event B has occurred. It is given that: P ( A) = 0. Similarly.75 .65 = = 0.65 (a) P(a ketchup-user uses mustard) = P ( A / B ) = P ( A ∩ B ) 0.

. + P ( A ∩ Bk ) = P ( B1 ) P ( A / B1 ) + P ( B2 ) P ( A / B2 ) + .20 )( 0. 18 . 2... E k are mutually exclusive events such that E1 ∪ E 2 ∪ . 20% are manufactured in Japan and 60% of these are compact. Bk constitute a partition of the sample space S such that P ( Bi ≠ 0) for i = 1..50 )( 0.. . find the probability that it is a compact.Two events A and B are independent if and only if P ( A ∩ B ) = P ( A) P( B ) Example 11 A pair of fair dice are thrown twice. 30% of the cars are manufactured in Europe and 40% of these are compact.. k.15 ) + (0. then for any event A of S P ( A) = P ( A ∩ B1 ) + P ( A ∩ B2 ) + . ∪ E k contains all sample points of S. P(Getting totals of 7 and ll) = P ( A ∩ B ) = P ( A1 ∩ B2 ) + P ( B1 ∩ A2 ) = P ( A1 ) P ( B2 / A1 ) + P ( B1 ) P ( A2 / B1 ) = P ( A1 ) P ( B2 ) + P ( B1 ) P ( A2 ) . . B2 be the event that the car is manufactured in Europe.40 ) + (0... B1 be the event that the car is manufactured is United States.60 ) = 0. then for any event D of S with P ( D ) ≠ 0 .. E 2 .315 Baye's Theorem If E1. P ( A) = P ( A ∩ B1 ) + P ( A ∩ B2 ) + P ( A ∩ B3 ) = P ( B1 ) P ( A / B1 ) + P ( B2 ) P ( A / B2 ) + P ( B3 ) P ( A / B3 ) = (0.. B2 . If a car is picked at random from the lot.. Ai . and finally... What is the probability of getting totals of 7 and 11? Solution Let Ai be the event of getting '7' in the i-th throw and B j be the event of getting '11' in the j-th throw. + P ( Bk ) P ( A / Bk ) Example 12 Suppose 50% of the cars are manufactured in the United States and 15% of these are compact..30 )( 0... B j are independent  6  2   2  6  1 =    +    =  36  36   36  36  54 Theorem of Total Probability If the events B1 ... and B3 be the event that the car is manufactured in Japan. Let A be the event that the car is compact..

P ( E k ) P ( D / E k ) Example 13 Suppose a box contains 2 red balls and 1 white ball and a second box contains 2 red ball and 2 white balls..P( Ei / D) = P ( Ei ∩ D ) = P( D) P( Ei ∩ D) ∑ P( E j =1 k j ∩ D) = P( Ei ) P( D / Ei ) P ( E1 ) P ( D / E1 ) + P ( E 2 ) P ( D / E 2 ) + .Probability Distributions To cope with uncertainties of outcome. 19 . If the drawn ball is red.A random variable is a variable that takes on different numerical values determined by the outcomes of a random experiment. Random Variables . 2 P( A / B) = 2 . are called probability distributions. Given: P( B ) = P ( B ' ) = 1 . 3 P( A / B' ) = 2 4 P(Coming from the 1st box/the drawn ball is red) = P ( B / A) = P( A ∩ B) P( A ∩ B) = P ( A) P( A ∩ B) + P( A ∩ B' ) ( 1 )(2 ) P ( B) P( A / B ) 4 2 3 = = = P( B) P( A / B) + P( B' ) P( A / B' ) ( 1 )(2 ) + ( 1 )(2 ) 7 2 3 2 4 Chapter Three . One of the boxes is selected by chance and a ball is drawn from it.. what is the probability that it came from the 1st box? Solution Let A be the event of drawing a red ball and B be the event of choosing the 1st box. a statistical model that describes the behavior of the outcome is needed. These theoretical models which are very similar to relative frequency distributions.

of a discrete random variable X expresses the probability that X takes the value x. 3} Discrete random variable . That is f ( x) = P( X = x ) 20 . x = 0. Let random variable. X be the sum of the two dice. as a function of x. 1.Example 1 An experiment of tossing a coin 3 times. 1. any value can occur. 2. As so S = {HHH HHT HTH THH TTH THT HTT TTT}. Probability Distribution of a random variable . 4 16 Example 3 Consider an experiment of tossing two fair dice. Let random variable. only a specified number of values can occur. X be the number of heads achieved. x P(X=x) That is. Then the probability distribution of X is: x P(X=x) 2 1 36 3 2 36 4 3 36 5 4 36 6 5 3 6 7 6 36 8 5 36 9 4 36 10 3 36 11 2 36 12 1 3 6 The probability function f ( x ) . P ( X = x) = 4 0 1 1 6 1 4 1 6 2 6 1 6 3 4 1 6 4 1 1 6 Cx .is a representation of the probabilities for all the possible outcomes.in a given interval.in a given interval. X = {0. 2. Example 2 The probability distribution of the number of heads occurred when a coin is tossed 4 times. Continuous random variable .

∑P ( X x = x ) = 1. The Normal Distribution Normal distribution is probability distribution of a continuous random variable. Let X be a random variable. Properties of probability function P ( X = x) :1. is called the variance. denoted σ x2. Errors are inevitable. of a discrete random variable X is defined as E ( X ) or µ x = ∑ xP( X = x) x It is the mean of the probability distribution. and given by Var ( X ) or σ x = E [( X − µ x ) 2 ] 2 = ∑ ( x − µ x ) 2P ( X = x ) x = ∑ x2 P( X = x) − µ x x 2 Example 4 Calculate the mean and variance of the discrete probability distribution in example 2 and 3. 2. E(X). Positive and negative errors are equally likely.where the function is evaluated at all possible values of x. Definition : 21 . 3. Mathematical Expectations The expected value. (X − µ x)2. It is based on the Law of Errors which states that 1. 2. Large errors are less likely than small errors. The expectation of the squared discrepancy about the mean. P ( X = x ) ≥ 0 for any value x.

Approximately 99. It is a continuous distribution. π = 3.0.73) = P(Z > 0) .0418 P(0 < Z < 1.14154 Notation : X ~ N(µ . 3.0418 = 0.73) = 0.7% of the values in a normally distributed population are within 3 standard deviation from the mean. Area under the normal curve: • • • Approximately 68% of the values in a normally distributed population are within 1 standard deviation from the mean. σ 2) Properties of the normal distribution:1.5 . 4.P(Z > 1. The total area under the curve and above the horizontal axis is equal to 1. σ = the standard deviation of X. 2. Usually a standard normal random variable is denoted by Z. 1) Remark : Usually a table of Z is set up to find the probability P(Z ≥ z) for z ≥ 0.4582 22 .A continuous random variable X is defined to be a normal random variable if its probability function is given by f (x ) = 1 1 x−µ 2 exp[− ( ) ] 2 σ σ ( 2π ) for −∞ < x < +∞ where µ = the mean of X. The curve is symmetric and bell-shaped about a vertical axis through the mean µ . Example 7 Given Z ~ N(0.73) = 0.5% of the values in a normally distributed population are within 2 standard deviation from the mean. 1) (a) (b) P(Z > 1. Notation : Z ~ N(0. The standard normal curve : The distribution of a normal random variable with µ = 0 and σ =1 is called a standard normal distribution. Approximately 95.

(c) (d) (e) P(− 2.25) = 0.05.3085 .. Solution:  45 − µ X − µ 62 − µ  P (45 < X < 62) = P < <  σ σ   σ 62 − 50   45 − 50 = P <Z <  10   10 = P ( − . So z1 = -1.8) = P(Z > 1.8) = 1 .P(Z > 0.8) .00776 .2119 = 0.5 < Z <1.3944.8) = 0.25 Theorem : If X is a normal random variable with mean µ and standard deviation σ .42 < Z < 0.78034 P(1.03334 the value z that has (i) 5% of the area below it Let the corresponding z value be z1..0359 .8 < Z < 2.2) = 1 .3944.1151 = 0.64 (ii) 39.0.P(Z < -2. 102). find P(45 < X < 62).42) .5) − P ( Z >1.0. Let the corresponding z value be z1. So z1 = 1. then we have P(0 < Z < z1) = 0.P(Z > 2.5764 23 . From the standard normal distribution table we have P(0 < Z < 1.0.8) = 1 .64) = 0. then Z= X −µ σ is a standard normal random variable and hence P ( x1 < X < x2 ) = P ( x1 − µ x −µ <Z< 2 ) σ σ Example 8 Given X ~ N(50. then we have P(Z < z1) = 0.44% of the area between 0 and z.00256 = 0. From the standard normal distribution table we have P(Z < -1.05.2) 0 =1 − P ( Z < −0.

5) = 1 − 0.12 ⇒ P Z > 1  = 0.1210 . X ∼ N(80.17 ) = 0.5) =1 − P ( Z < −0. If 12% of the class are given A’s.175 7 24 .1190 so Thus P ( Z >1. What is the probability that a charge account randomly selected has a balance (a) (b) over \$125.12 x1 − 74 = 1.5) = 0.5) − P ( Z > 0. Let X be the balance in the charge account. between \$65 and \$95. and the grades are curved to follow a normal distribution. 30 2 ) (a) P ( X > 125 ) = P ( X −µ σ > 125 − µ σ ) = P( Z > 125 − 80 ) = P ( Z > 1. x − 74   P( X > x1 ) = 0.Example 9 The charge account at a certain department store is approximately normally distributed with an average balance of \$80 and a standard deviation of \$30.175 ) ≅ 0.5 < Z < 0.18 ) = 0. we get P ( Z >1.0668 30 (b)  65 − µ X − µ 95 − µ  P(65 < X < 95) = P < <  σ σ   σ 95 − 80   65 − 80 = P <Z <  30   30 = P ( −0.3830 Example 10 On an examination the average grade was 74 and the standard deviation was 7. what is the lowest possible A and the highest possible B? Let X be the examination grade and x1 be the lowest grade for A.3085 − 0.12 7   From the standard normal distribution. and P( Z >1.3085 = 0.

1 ) 1 0 Let X be the number of defective articles in a random sample of 20. …. one-tenth is defective.i. Each trial has two possible outcomes. Notation : P(X = x) = X ~ b(n. 4. one called “success” and the other “failure”.  20  1   9  P ( X = 2) =       2  10     10  2 18 (a) = 0. There are n identical observations or trials.28517 25 . The outcomes are mutually exclusive and collectively exhaustive. The probabilities of success p and of failure 1 − p remain the same for all trials. The outcomes of trials are independent of each other. x1 = 74 + (7)(1. The Binomial Distribution A binomial experiment possesses the following properties : 1.2 ≅ 83 The highest possible B is 82. at least two defective articles. 2. where each test or trial may indicate a defective or a non-defective item. Five cards are drawn with replacement from an ordinary deck and each trial is labelled a success or failure depending on whether the card is red or black. X ∼ b(20. n p+q=1 Example 12 Of a large number of mass-produced articles. Definition : In a binomial experiment with a constant probability p of success at each trial.e. 2. the number of successes in n independent trials. 1. p)  n x n − x  p q  x x = 0. is called the binomial distribution.175 ) = 82 . the probability distribution of the binomial random variable X. Example 11 1. Find the probabilities that a random sample of 20 will obtain (a) (b) exactly two defective articles. 3. In testing 10 items as they come off an assembly line.

5) − np <Z< ) ( npq ) ( npq ) if n is large and p is not close to 0 or 1. of defective packages in a sample of 10 packages. X ∼ b(10. 26 .2) Its mean is µ = np = (10)(0.10014 Theorem The mean and variance of the binomial distribution with parameters of n and p are µ = np and σ 2 = npq respectively where p + q = 1.12158 − 0. 6 6 6  P ( X ≥ 4) = ∑P( X = x) = ∑  1   3 x =4 x =4  x  ( ) (2 3 ) x 1 1 ) 3 6−x 6  =  1 4 3   ( ) (2 3 ) 4 2 6  +  1 5  3   ( ) (2 3 ) 5 6  +  1 6  3   ( ) (2 3 ) 6 0 = 0.2) = 2 Its standard deviation is σ = npq = (10 )( 0. 0. Each question has three possible answers.60825 Example 13 A test consists of 6 questions. the approximation will be good.8) =1. A random sample of ten packages is selected. and to pass the test a student has to answer at least 4 questions correctly.2)( 0. If a student guesses on each question.27017 = 0. Example 14 A packaging machine produces 20 percent defective packages. Remark : If both np and nq are greater than 5. what are the mean and standard deviation of the binomial distribution of that process? Let X be the no. of which only one is correct.5) − np ( x + 0.265 The Normal Approximation to the Binomial Distribution Theorem : Given X is a random variable which follows the binomial distribution with parameters n and p. what is the probability that the student will pass the test? Let X be the no. then P( X = x) = P( ( x − 0. X ∼ b(6. of correctly answered questions among 6 questions.P ( X ≥ 2) = 1 − P ( X = 0) − P ( X = 1) (b)  20  1   9  = 1 −       0  10    10  0 20  20  1  9  −      1  10 10     19 = 1 − .

25) µ = np = (80 )( 0. 0. X ∼ b(80.25 ) = 20 .Example 15 A process yields 10% defective items.5 − 20  = P <Z <  = P (1.1196   15 15   The Poisson Distribution Experiments yielding numerical values of a random variable X. The number of successes in any interval is independent of the number of successes in other interval. X ∼ b(100.75 ) = 15 P (25 ≤ X ≤ 30 ) ≅ P( 24 .00336 = 0. of correct answers for 80 with sheer guesswork.1) µ = np = (100 )( 0. what is the probability that the number of defective exceeds 13? Let X be the no.5) by normal approximation  24 .1)( 0.25 )( 0.5 −10   X '−µ 13 .1) = 10 .16 < Z < 2.5 − µ   = P >  = P Z >  = P ( Z > 1.5 < X ' < 30 .1230 − 0. The probability of a single success occurring during a short interval is proportional to the length of the time interval and does not depend on the number of successes occurring outside this time interval. What is the probability that sheer guesswork yields from 25 to 30 correct answers for 80 of the 200 problems about which the student has no knowledge? Let X be the no. σ = P ( X >13 ) ≅ P ( X ' >13 . A Poisson experiment has the following properties: 1.5) np = (1 )( 0. If 100 items are randomly selected from the process. The probability of more than one success in a very small interval is negligible.9) =3 q 00 by normal approximation 13 .71) = 0. 0. Examples of random variables following Poisson Distribution 27 . 2.5 − 20 30 . the number of successes (observations) occurring during a given time interval (or in a specified region) are often called Poisson experiments. 3.167 ) = 0. σ = npq = (80 )( 0.121 σ 3  σ    Example 17 A multiple-choice quiz has 200 questions each with four possible answers of which only one is the correct answer. of defective in a random sample of 100 items.

e. Suppose that this situation can be described by a Poisson distribution. P(X = x) = e− λ λx x! x = 0. i. µ = σ 2 . Example 17 The average number of radioactive particles passing through a counter during 1 millisecond in a laboratory experiment is 4. What is the probability that 6 particles enter the counter in a given millisecond? Let X be the no. Let X be the no. 4.718283 Theorem: In a Poisson Distribution mean is equal to variance. 2. 3.. … e = 2. The number of customers arrived during a time period of length t. 2.1. X ∼ Po( 2 2 = 1) 28 . Definition : The probability distribution of the Poisson random variable X is called the Poisson distribution. Notation : X ~ Po(λ ) where λ is the average number of successes occuring in the given time interval. Three ships arrive.1042 6! Example 19 Ships arrive in a harbour at a mean rate of two per hour. The number of accidents occurred at a junction per day. X ∼ Po(4) P ( X = 6) = e −4 4 6 = 0. The number of telephone calls per hour received by an office. of ship arriving in a harbour for a 30-minute period. The number of typing errors per page. of particles entering the counter in a given millisecond. 1. Find the probabilities for a 30-minute period that (a) (b) No ships arrive.

001) ( 0.001. the approximation will generally be excellent provided p< 0.001) ( 0.999)  ( 0. Example 20 If the prob.3679 0! e −113 = 0.323 e2 General speaking. more than 2 individuals will suffer a bad reaction.00 in the binomial distribution.0613 3! (b) P ( X = 3) = Theorem : The mean and variance of the Poisson distribution both have mean λ . Soln : According to binomial : The required probability  2000  2000  2000 0 2000 1 1999 2 1998  + + = 1 −   ( 0. determine the prob. then the binomial distribution can be approximated by the Poisson distribution with parameter np. the Poisson distribution will provide a good approximation to binomial when (i) n is at least 20 and p is at most 0.(a) P ( X = 0) = e −110 = 0. or (ii) n is at least 100. Then the required probability = 1 − Example 21 29 .1.999)  ( 0.05. that an individual suffers a bad reaction from a certain injection is 0.999)   1   2   0   Using Poisson distribution: P(0 suffers) = 20 e −2 1 = 2 0! e  λ = np = 2 P(1 suffers) = 21 e −2 2 = 2 1! e 2 2 e −2 2 = 2 2! e P(2 suffer) = 5 = 0. that out of 2000 individuals. Poisson approximation to the binomial distribution If n is large and p is near 0 or near 1.001) ( 0.

It indicates the extent to which a sample statistic will tend to vary because of chance variation in random sampling.1338 By Poisson Approximation: λ = np = (300 )( 0. X ∼ b(300. A simple random sample of two elements is to be drawn.3}. 0.02 ) = 6 P ( X = 4) = e −6 6 4 = 0.02) P ( X = 4) =300 C 4 (0. A lot of 300 pieces will be produced.2. A population parameter is a characteristic of a population. They are 30 . A sampling distribution is a probability distribution for a sample statistic.98 ) 296 = 0. Determine the probability that exactly four pieces will be defective. 4 Obviously there are six possible samples ( C 2 = 6 ). of defective pieces among 300 pieces. A statistic is a random variable that depends only on the observed random sample. Let X be the no. The standard deviation of the distribution of a sample statistic is known as the standard error of the statistic. 4.6667. {0. 2.02 ) 4 (0. The population has two parameters: a mean µ of 1. A sample statistic is a characteristic of a sample.5 and a variance σ 2 of 1.1338 4! CHAPTER 4 -SAMPLING DISTRIBUTIONS AND ESTIMATION Definition 1.Two percent of the output of a machine is defective.1. 3. An illustrating example Suppose a population consists of four elements.

e.0 1/6 1. the accuracy of our estimate depends on which sample we have drawn.e.0 Probability 1/6 1/6 1/6 1/6 1/6 1/6 From the above table. as compiled in the following table: Sample mean Probability 0.3 1. the sample mean. Hence the average value of the sample mean is equal to the population mean. σ2 ).5 * 1 / 6 =1.5 1/6 1.5 1.5 − 1.Sample 0.3 Sample mean 0.2 0.5 * 1 / 6 +1. is approximately normal.2 1. the average square deviation of the sample mean from the V ( y ) = ∑ y −Y ) 2 P ( y ) = ( population mean) is: 1 1 2 1 1 (0. which in turn depends on chance.5) 2 * + (1.1 0.5 − 1.5)2 * + (2.3 2.0 * 1 / 6 +1. with mean µ (i. we can see the following modifications: 31 . The probability distribution of sample mean is known as a sampling distribution of sample mean.e. and n is large (n ≥ 30).0 * 1 / 6 + 2. E ( x) = µ ) and variance σ 2/n (i. The variance of the sample mean (i. the distribution of x .0 − 1.0 2..5 − 1.5 * 2 / 6 + 2. we can see that if we draw a sample and use the sample mean to estimate the population mean.0 1/6 2.4167 Sampling Distribution of Mean The Central Limit Theorem If repeated samples of size n are drawn from any infinite population with mean µ and variance σ 2.5 2.5 0 0 0. We call the sample mean an unbiased estimator of the population mean.0 1.5)2 * + (1.5 1.5 error -1.5)2 * + (2.0 − 1.5 1. and this approximation becomes better as n becomes larger.5 1/6 The expected value of sample mean is E ( y ) = ∑y * P ( y ) = 0. V ( x) = n Notes: As in the previous illustrating example.5 = Y .5 2/6 2.5)2 * 6 6 6 6 6 =0.0 -0.

6667/2) = 0. When N is very big. Example 1 An electrical firm manufactures light bulbs that have a length of life that is approximately normal distributed with mean equal to 800 hours and a standard deviation of 40 hours. 10 2 ) 32 . X ∼ N( µ x = µ = 800. If the ) N n σ2 population is big (or the sample is drawn with replacement). the factor is equal to 1. V ( x ) = (1 − In this course we assume a big population or sampling with replacement. then V ( x ) = =1. Find the probability that a random sample of 16 bulbs will have an average life of less than 775 hours. say less than 30. n In the above example. n σ2 =(1-2/4)(1. A t-distribution will be used (discussed later). σ 2 40 2 2 = Let X be the average life of the 16 bulbs. σ x = ) n 16  X − µx 775 − µx P ( X < 775 ) = P <  σ σx x     775 −800  = P Z <   40   16        = P ( Z < −2. what is the probability that the score of any one student is greater than 112? What is the probability that the mean score in a random sample of 36 students is greater than 112? What is the probability that the mean score in a random sample of 100 students is greater than 112? Solution (a) Let X be the student's IQ score.00621 Example 2 The mean IQ scores of all students attending a college is 110 with a standard deviation of 10.4167. where (1-n/N) is known as the finite ) N n population correction factor. (a) (b) (c) If the IQ scores are normally distribution. the sampling distribution is not so normal.8333.(i) If the population is finite. n=2. (ii) If n is small. N=4. X ∼ N(110. V ( x ) = (1 − n σ2 .6667/2=0.5) = 0.

unbiasedness and efficiency.1 5 11    P( X 1   1 2 −1 0 1 1 >1 2 ) = PZ > 1  1 0  3 6  (c) Let X 2 be the mean score of a sample of 100 students.2) = 0.2) =0. = ) n 100   1 2 −1 1 10 >1 2 ) = PZ > 1  10  10 0     = P ( Z > 2) = 0.0. 3. for example. 33 .(b) 112 − 110   P ( X > 112 ) = P Z >  = P ( Z > 0. The followings are some examples Estimator 1.22 8    P( X 2 Estimation Estimation is the process of using statistics from sample data to estimate the parameters of the population. x . 2 X 1 ∼ N( µ x = µ = 110 . Unbiased estimator: An estimator. is unbiased if and only if E ( x ) = µ . σ x = σ 2 10 2 = ) n 36    = P ( Z >1. σ 2 10 2 X 2 ∼ N( µ =110 . Population parameter µ σ P 2 x s2 P There are two important properties for an estimator. 2. namely.4207 10   Let X 1 be the mean score of a sample of 36 students. A statistic is a random variable which depends on which sample is drawn from a population.

An interval estimate of a population parameter gives an interval that may contain the true value of the parameter with a certain probability (i.3.99. How large a sample is required in (a) if we want to be 95% confident of µ is off by less than 0.502 < µ < 2. for example. confidence).3. Examples 3 (a) The mean and standard deviation for the quality point averages of a random sample of 36 college seniors are calculated to be 2. both the accuracy and reliability of the estimation are unknown. Given: n = 36. respectively. A point estimate is a single-value estimate of a population parameter.025 ⇒ 2. x . For an interval estimate. The smaller the V ( x ) . Find a 95% confidence interval for the mean of the entire senior class.6 + 1. ^ ^ µ = x.Efficiency: The efficiency of an estimator. (1 (a) α ) = 0. P = p . For a point estimate.05? (b) Solution Let µ be the mean of the entire senior class. is given by V ( x ) . There are two types of estimate 1.3  < µ < 2. x = 2.6 − 1. s = 0.698 (b) Let n1 be the required sample size. To be 95% confident that µ if off by less than 0.96     36     ⇒ 2.025 ˆ ˆ  0.05 would imply 34 .3 σ σ < µ < x + z 0. for example. 2. for example.05 A 95% confidence interval estimate for the µ is x − z 0.95 ⇒ α = 0.6 . Pr(a < µ < b) = 0.6 and 0.96   n n  36   0.e. the width of the interval gives the accuracy and the probability gives the reliability of the estimation. the more accurate will be the x as an estimator.

05   ∴  (1.05 ⇒1. σ   0.025  ≤ 0.α ( s p ) 2 ν = n1 + n 2 − 2 . assume σ1 = σ 2 Formula X − Zα X − tν .96 )( 0. pooled estimate of sample standard deviation: sp = Difference of means Small sample & σ1 and σ2 are unknown. assume σ1 ≠ σ2 2 (n1 − 1) s12 + ( n2 − 1) s 2 n1 + n2 − 2 ( X 1 − X 2 ) ± tν .05  A summary table for constructing (1 − α )% confidence interval for mean and proportion Estimating Mean Mean * Conditions Large samples (n ≥ 30) OR σ is known Small samples and σ unknown Large sample Large sample OR σ1 and σ2 are known Small sample & σ1 and σ2 are unknown.α 2 2 s12 s2 + n1 n2 35 .α 2 2 σ σ < µ < X + Zα 2 n n s s < µ < X + tν .α 2 n n v=n-1 Proportion Difference of means ˆ p ± Zα 2 ˆˆ pq n 2 ( X 1 − X 2 ) ± Zα ˆ σ 12 σ 2 + n1 n2 1 1 + n1 n 2 Difference of means ( X 1 − X 2 ) ± tν .30 ≅ 139  0.3 ˆ  z 0.3)  n1 ≥   = 138 .96   n   n 1  1   2   ≤ 0.

0. ∑ x2 − (∑ x ) 2 n −1 n = 700.447 A 95% confidence interval estimate for the µ is x − t 6. Solution Let µ be the mean of all such containers.447   n 7    0. Find a 95% confidence interval for the mean of all such containers.025 = 2.48 ∑ x = 70 = 10 x= ∴ s= 0. 10.2 and 9. 10.08 α ) = 0.υ= 2 ( s12 / n1 + s 2 / n2 ) 2 2 ( s12 / n1 ) 2 ( s 2 / n2 ) 2 + n1 − 1 n2 − 1 Difference of means Paired observations d ± tv . 9.08 = 0. 36 .8. t 6.2828  < µ < 10 + 2.6 liters. d = x1 − x 2 n ˆ ˆ ˆ ˆ p1 q1 p 2 q 2 + n1 n2 Difference of proportions Large samples ˆ ˆ ( p1 − p 2 ) ± zα 2 Example 4 The contents of seven similar containers of sulfuric acid are 9. Let P be the actual proportion of families in this city with colour sets. assuming an approximate normal distribution.α / 2 and v=n-1 sd . Canada. Find a 95% confidence interval for the actual proportion of families in this city with colour sets. 0.025 s n < µ < x + t 6.447    7       ⇒9. 10. 0.738 < µ < 10 .2828 ⇒ 10 − 2.05 .4. 0.2828 .262 Example 5 In a random sample of n = 500 families owning television sets in the city of Hamilton.95 ⇒ α = 0.2.025 s  0.8. 10. Given: n = 7 ∑x = 70 n 7 ∑x s2 = (1 - 2 = 700 .48 − 70 6 2 7 = 0. it was found that x = 340 owned color sets.

while the average for the second sample gave 37 . The 12 batches for which catalyst 1 was used gave an average yield of 85 with a sample standard deviation of 4. n 500 (1 − α) = 0. 32 ) (. two catalysts are being compared for their effect on the output of the process reaction. The girls made an average grade of 76 with a standard deviation of 6.64 < P < 0.96 ⇒ 0.04 A 96% confidence interval for µ1 − µ2 is: ( x1 − x 2 ) − z 0.72 500 500 Examples 6 A standardized chemistry test was given to 50 girls and 75 boys. 68 )(. where µ 1 is the mean score of all boys and µ 2 is the mean score of all girls who might take this test.68 +1.43 < µ1 − µ 2 < 8. n 2 = 50 .05 ˆ ˆ n1 > 30 & n 2 > 30 . + < µ1 − µ 2 < (82 − 76 ) + 2. (1 − α) = .02 ˆ ˆ2 ˆ ˆ2 σ 12 σ 2 σ 12 σ 2 + < µ1 − µ 2 < ( x1 − x 2 ) + z 0.05 A 95% confidence interval for P is  ˆˆ   ˆˆ  ˆ ˆ p − z 0. x1 = 82 . 32 ) < P < 0. ˆ p = x = 340 = 0.68 −1.02 + n1 n 2 n1 n2 82 62 82 62 . 68 )(. x 2 = 76 .57 Example 7 In a batch chemical process.05 + 75 50 75 50 ⇒ (82 − 76 ) − 2.95 ⇒α = 0.68 .025  pq  < P < p + z 0.96 (. s 2 = 6 .96 ⇒ α = 0. Given: n1 = 75 . so σ1 = s1 & σ 2 = s 2 ⇒ 3. while the boys made an average grade of 82 with a standard deviation of 8.Given: n = 500. A sample of 12 batches is prepared using catalyst 1 and a sample of 10 batches was obtained using catalyst 2.025  pq  n n   ⇒ 0. s1 = 8 . Find a 96% confidence interval for the difference µ 1 and µ 2.

31 Example 8 The weight of 10 adults selected randomly before and after a certain new diet was introduced was recorded as follows: Adult 1 2 3 4 5 6 7 8 9 10 Before ( x1 ) 76 60 85 58 91 75 82 64 79 88 After ( x2 ) 81 52 87 70 86 77 90 63 85 83 Difference -5 8 -2 -12 5 -2 -8 1 -6 5 38 .05 ( s p ) (12 −1)4 2 + (10 −1)5 2 = 4. respectively. s 2 = 5 .0. s1 = 4 . 0.478 ) + 12 10 12 10 ⇒ (85 − 81) − (1.05 =1.an average of 81 and a sample standard deviation of 5.10 . x 2 = 81 . x1 = 85 . 0. assuming the populations are approximately normally distributed with equal variances. n 2 = 10 .69 < µ1 − µ 2 < 7.725 pooled estimate of sample standard deviation s p = 2 (n1 − 1) s12 + ( n2 − 1) s 2 n1 + n2 − 2 = A 90% confidence interval for µ1 − µ2 is: ( x1 − x 2 ) − t 20 .05 ( s p ) + n1 n 2 n1 n 2 1 1 1 1 + < µ1 − µ2 < (85 − 81) + (1.90 ⇒ α = 0. 1 Given: n1 = 12 .725 )( 4. t 20 .478 12 + 10 − 2 1 1 1 1 + < µ1 − µ2 < ( x1 − x 2 ) + t 20 . (1 − α) = . ν = n1 + n 2 − 2 = 12 + 10 − 2 = 20 .725 )( 4.478 ) ⇒ 0. Solution Let µ and µ2 be the mean population yield using catalyst 1 and catalyst 2. Find a 90% confidence interval for the difference between the population means.

Solution Let P and P2 be the true fraction of defectives of the existing and the new processes. 1 respectively.29 < µd < 4. If 75 of 1500 items from the existing procedure were found to be defective and 80 of 2000 items from the new procedure were found to be defective.821)    10  Example 9 A certain change in a manufacturing procedure for component parts is being considered.09  6. A 98% confidence interval is That is −7. x1 = 75 . x 2 = 80 .05 ˆ ˆ ˆ ˆ p1 q1 p 2 q 2 ˆ ˆ + < P1 − P2 < ( p1 − p 2 ) + z 0. find a 90% confidence interval for the true difference in the fraction of defectives between the existing and the new process.6)) 2 n −1 = 40.P2 is: 1 ˆ ˆ ( p1 − p 2 ) − z 0. Solution ∑d d= n i = -1.01 = 2. Given: n1 = 1500 . t0.05 n1 n2 ˆ ˆ ˆ ˆ p1 q1 p 2 q 2 + n1 n2 39 .6 ± (2.04 n 2 = 2000 .05 1500 ˆ p 2 = 80 2000 = 0. Samples are taken using both the existing and the new procedure in order to determine if the new procedure results in an improvement. ˆ p1 = 75 = 0.6 s 2 d ∑ (d = i − (−1. (1 − α) = .38  −1.10 A 90% confidence interval for P .90 ⇒ α = 0.Find a 98% confidence interval for the mean difference in weight.821 .7 For v = n-1 = 9.

8 kilograms.96* < µ < 7.94 kg.01) +1.96*  = 0.e.5 kilogram.5   P  7. 50 50   i.95 . the sample information can be used assess the validity of some conjecture. 05 )(. There are four possible situations for the above decision making exercise: H 0 is correct Accept H 0 Reject H 0 Correct decision Type 1 error H 0 is wrong Type 2 error Correct decision 40 . 05 )(.001697 < P1 − P2 < 0. Now let us make a 95% confidence interval about the mean breaking strength of the population as below: 0. P (7.01 level of significance. 96 ) + 1500 2000 ⇒ −0. or hypothesis. 95 ) (.021697 Lecture 5 .9386) = 0.8 + 1. 96 ) + < P − P2 < (0. 95 ) (. Here µ = 8 kilograms is known as the null hypotheis and µ ≠ 8 kilograms is the alternative hypothesis. Use a 0. They are complementary to each other.Introduction to Test of Hypothesis Statistical Hypothesis Consider the following example: A manufacturer of sports equipment has developed a new synthetic fishing line that he claims has a mean breaking strength of 8 kilograms with a standard deviation of 0. it is highly unlikely that the null hypothesis µ = 8 kg is true and hence should be rejected.64 1 1500 2000 (.95 that the mean breaking strength is between 7.5 0. Test the hypothesis that µ = 8 kilograms against the alternative that µ ≠ 8 kilograms if a random sample of 50 lines is tested and found to have a mean breaking strength of 7. and we need to decide which one to accept on the basis of the sample result of 50 lines. As there is a probability of 0.6614 < µ < 7.01) −1.95 . 04 )(.64 (.8 − 1.⇒(0. 04 )(.66 kg and 7.. When a random sample is drawn from a population (the 50 lines randomly selected).

The rationale of hypothesis testing is simply outlined as above. based on the sample. H1 : µ ≠ µ0 (two-tail test) H1 : µ > µ0 (One-tail test) H1 : µ < µ0 (One-tail test) In the one-tail test we have some expectation about the direction of the error when the null hypothesis is wrong.We still have a probability of 1-0. There are however some formal concepts and procedures to conduct the test. 4. The details are put down below. Critical region is a region in which if the test statistic falls the null hypothesis will be rejected. H0 : µ = µ0 2. Alternative hypothesis. Test statistics is the value. Types of error (a) (b) Type I error: Reject H0 when H0 is true Type II error: Accept H0 when Ha is true 41 .05 to reject a true H 0 . H0 A hypothesis that is held to be true until very strong evidence to the contrary is obtained. 5.95. Null hypothesis. used to determine whether the null hypothesis should be rejected or accepted. which is the probability of committing a type 1 error. 3. We call this probability ‘level of significance’ or α . Hence it will be accepted if the null hypothesis is rejected. H1 It is a hypothesis that is complement to the null hypothesis. Some Hypothesis Testing Terminology 1. while in the two-tail test we don’t have such expectation. or 0.

otherwise accept H0.. 4. 2. 3. 6. 42 . Conclusion: Reject H0 if the statistic has a value in the critical region.e. Specify the level of significance to be used.. Compute the value of the test statistic. 5. Formulate the null hypothesis. The probability of committing a type 2 error is β .6. The significance level. i. Basic Steps in Testing Hypothesis 1. α is the probability of committing a type 1 error. P(Type I error) = α . Formulate the alternative hypothesis. i.e. Select the appropriate test statistic and establish the critical region. P(Type II error) = β .

assume σ 1 = σ 2 1 1  sp  +  n n  2   1 with υ = n1 + n 2 − 2 and t= ( x1 − x 2 ) − d 0 s2 = p µ1 − µ2 = d 0 Small sample & σ1 and σ 2 are unknown.Tests concerning means The tests concerning means and proportions are summarized in the following table. assume σ1 ≠ σ 2 2 (n1 −1) s12 + (n2 −1) s2 n1 + n2 − 2 if σ1 = σ2 but unknown ( x − x2 ) − d 0 t= 1 2 s12 s 2 + n1 n2 with υ= 2 ( s12 / n1 + s 2 / n2 ) 2 2 ( s12 / n1 ) 2 ( s 2 / n2 ) 2 + n1 − 1 n2 − 1 43 . H0 Conditions Large samples (n ≥ 30) OR σ is known Small samples and σ unknown Test statistic x − µ0 z= σ n x − µ0 t= with υ = n − 1 s n µ = µ0 µ = µ0 µ1 − µ2 = d 0 Large samples OR σ1 and σ2 are known z= ( x1 − x 2 ) − d 0 2 σ 12 σ 2 + n1 n2 µ1 − µ2 = d 0 Small sample & σ1 and σ 2 are unknown.

µ1 − µ2 = d 0 Paired observations Large sample t= d − d0 sd n with υ = n − 1 p = p0 z= p1 − p2 = 0 Large samples ˆ p − p0 p0 (1 − p0 ) n ˆ ˆ ( p1 − p2 ) z= 1 1 ˆ ˆ  p(1 − p ) +    n1 n2  ˆˆ n p +n p ˆ p= 1 1 2 2 n1 + n2 44 .

5 kilogram.828 σ 0.5 z= x−µ 7. If a random sample of 12 students had an average registration time of 42 minutes with a standard deviation of 11.9 minutes under the new system. and the new degree of freedom (ν ) = n -1 = 12 -1 =11 σ is unknown. 8 − 8 = = −2.005 = −2. using a level of significance of (1) 0.05 Critical region: (n = 12 < 30.5 n 50 Conclusion: As the sample z (= -2.01 level of significance and conclude that µ is significantly smaller than 8 kilograms.58 or Z < − z 0. so t-test should be used ) 45 . (1) Null hypothesis: µ = 50 minutes Alternative hypothesis: µ < 50 minutes Level of significance: 0. Let µ be the population mean time for students to register in the new registration procedure. so reject the null hypothesis at 0.58 Computation: n = 50 x = 7.Example 1 A manufacturer of sports equipment has developed a new synthetic fishing line that he claims has a mean breaking strength of 8 kilograms with a standard deviation of 0.828) falls inside the critical region.05. Null hypothesis: µ = 8 kilograms Alternative hypothesis: µ ≠ 8 kilograms Level of significance: 0.8 kilograms.01 level of significance.01.005 = 2. Assume the population of times to be normal.01 Critical region: Z > z 0. and (2) 0. A new registration procedure using modern computing machines is being tried. Example 2 The average length of time for students to register for fall classes at a certain college has been 50 minutes with a standard deviation of 10 minutes. test the hypothesis that the population mean is now less than 50.8 σ = 0. Use a 0. Test the hypothesis that µ = 8 kilograms against the alternative that µ ≠ 8 kilograms if a random sample of 50 lines is tested and found to have a mean breaking strength of 7.

so reject the alternative hypothesis at 0.) ν = n1 + n2 − 2 = 12 + 10 − 2 = 20 .9 ∴ t= x−µ 42 − 50 = = −2.796 Computation: n = 12 x = 42 s = 11. 0. 0.e. µ1 − µ 2 = 0 Alternative hypothesis: µ1 ≠ µ 2 .725 or t < − 20 . Ten pieces of material 2 were similarly tested. 1 Null hypothesis: µ1 = µ 2 .10 level of significance. i. Twelve pieces of material 1 were tested.05 =1.e.05 level of significance and conclude that µ is significantly smaller than 50 minutes.10 Critical region: (As both n1 and n 2 are smaller than 30 and their standard deviations are unknown.329) falls outside the critical region.718 2 and the corresponding conclusion would be changed as follows: As sample t (= -2.05 = − . by exposing each piece to a machine measuring wear.01 = − . (2) Identical with those of (1) except the critical region would be replaced by: t <t11 .725 t 1 46 . The samples of material 1 gave an average (coded) wear of 85 units with a standard deviation of 4.05 = − . µ1 − µ 2 ≠ 0 Level of significance: 0. i. so reject the null hypothesis at 0. Assume the populations to be approximately normal with equal variances. Example 3 An experiment was performed to compare the abrasive wear of two different laminated materials. Let µ and µ2 be the mean abrasive wear of material 1 and 2 respectively. Test the hypothesis that the two types of material exhibit the same mean abrasive wear at the 0. so t-test has to be used. 0. the depth of wear was observed. 0. while the samples of material 2 gave an average of 81 and a standard deviation of 5.1 ∴ t < t11 .9 n 12 Conclusion: As sample t (= -2.329) falls inside the critical region.01 level of significance and conclude that µ is not highly significantly smaller than 50 minutes. In each case. ∴ t >t 20 .329 s 11.

5 4 2.05 Critical region: (As n = 5 < 30.Computation: n1 = 12 n2 = 10 x1 = 85 x 2 = 81 s1 = 4 s2 = 5 s2 = p 2 (n1 − 1) s12 + (n 2 − 1) s 2 (11)4 2 + (9)5 2 = = 20.05 level of significance whether the two methods of analysis give.05 + 12 10 = 2. so t-test should be used.2 2 2.) or µD = 0 or µD ≠ 0 47 .4 Assuming the populations normal. so reject the null hypothesis at 0.086 Conclusion: As the sample t (=2.4 2. and µD be the mean of the population of differences of paired measurements.1 2. Following are the coded data showing the iron content analysis: Sample Analysis x-ray Chemical 1 2.3 5 2.0 1.086) falls inside the critical region. test at the 0. the same result.10 level of significance and conclude that the mean abrasive wear of material 1 is significantly higher than that of the material 2.3 2. Let µ and µ2 be the mean iron content determined by the laboratory chemical analysis and 1 X-ray fluorescence analysis respectively.05 n1 + n 2 − 2 12 + 10 − 2 t= ( x1 − x 2 ) − ( µ1 − µ 2 ) sp 1 1 + n1 n2 = (85 − 81) − 0 1 1 20. Null hypothesis: µ1 = µ 2 Alternative hypothesis: µ1 ≠ µ 2 Level of significance: 0. on the average.9 3 2. Example 4 Five samples of a ferrous-type substance are to be used to determine if there is a difference between a laboratory chemical analysis and an X-ray fluorescence analysis of the iron content. Each sample was split into two sub-samples and the two types of analysis were applied.0 2.

9 0.1) − 0 = = −1.0 1.5811) falls outside the critical region. 0.1 0.04 4 2.5) 2 = 0.776 t 2 Sample Analysis x-ray Chemical di 1 2.3 -0.2 0.∴ t > t 4.13) − ( −0. 0.4 0 0 d i2 ∑d i =1 5 i = −0.5811 sd 0.2 -0.5 -0.025 = − .2 0.4 2.1 2.1 5 2 s = 2 d n∑d 2 − ( ∑d ) n( n −1) = (5)( 0.0 2.2 0.05 level of significance and conclude that there is no significant difference in the mean iron content determined by the above two analyses.02 (5)( 4) t= d − µ D ( −0.04 5 2.02 n 5 Conclusion: As the sample t (=-1.776 Computation: or t < − 4 .13 d = ∑d 5 = − 0. Tests Concerning Proportions Example 5 48 .04 2 2.5 ∑d i =1 5 2 i = 0.5 = −0.3 2.025 = 2. so reject the alternative hypothesis at 0.01 3 2.

667 Conclusion: As the sample z (=-1. respectively. That is.05 − 0.05 level of significance.1 Level of significance: 0.1 Alternative hypothesis: P < 0. To determine if there is a significant difference in the proportion of town voters and county voters favoring the proposal.9) 100 = −1.05 Critical region: Z < −z 0. Null hypothesis: P = P2 1 or P − P2 = 0 1 or Alternative hypothesis: P > P2 1 Level of significance: 0. Is this evidence sufficient to conclude that the method has been improved? Use a 0. An improvement in the process is being considered that they feel will lower the proportion of defective below the current 10%. favouring 1 the proposal. Null hypothesis: P = 0.05 = −1. Let P be the proportion of defective product in the new production process.025 P − P2 > 0 1 49 .05 n 100 z= ˆ p −P P (1 − P ) n = 0.1.1 (0. so reject the null hypothesis at 0. Example 6 A vote is to be taken among the residents of a town and the surrounding country to determine whether a proposed chemical plant should be constructed.025 level of significance. If 120 of 200 town voters favor the proposal and 240 of 500 county residents favor it. In an experiment 100 items are produced with the new process and 5 are defective. Let P and P2 be the proportions of town voters and country voters. would you agree that the proportion of town voters favoring the proposal is higher than the proportion of county voters? Use a 0.667) falls inside the critical region.05 level of significance and conclude that P is significantly smaller than 0. the production method has been improved in lowering the proportion of defective below the current 10%. a poll is taken.A manufacturing company has submitted a claim that 90% of items produced by a certain process are non-defective.1)( 0.64 Computation: n = 100 x=5 ˆ p= x 5 = = 0. The construction site is within the town limits and for this reason many voters in the country feel that the proposal will pass because of the large proportion of town voters who favor the construction.

514 )( 0.870 z= Conclusion: As sample z (=2.514 n1 + n 2 200 + 500 ˆ ˆ ( p1 − p 2 ) − ( P1 − P2 ) 1 1 ˆ ˆ p (1 − p )  +   n1 n2  = (0.6 n1 200 x 2 240 = = 0.6) + (500 )( 0.025 = 1.870) falls inside the critical region.48 ) = = 0.6 − 0.48) − 0 1   1 (0. 50 .486 )  +  200 500   = 2. so reject the null hypothesis at 0.Critical region: Z > z 0.48 n2 500 n 2 = 500 x 2 = 240 ˆ p2 = ˆ p= ˆ ˆ n1 p1 + n 2 p 2 (200 )( 0.96 Computation: n1 = 200 x1 = 120 ˆ p1 = x1 120 = = 0.025 level of significance and conclude that the proportion of town voters favouring the proposal is significantly larger than that of the country voters.

5. 3. Oi is the observed frequency of cell i. The number of degrees of freedom in a Chi-square goodness-of-fit test is equal to the number of cells minus the number of quantities obtained from the observed data that are used in the calculations of the expected frequencies. and Ei is the expected frequency of cell i. Faces 1 Observed Expected 20 2 22 3 17 4 18 5 19 6 24 By comparing the observed frequencies with the expected frequencies. 4.Chapter 7 . Example 1 Consider the tossing of a die 120 times. The test is based on how good a fit we have between the frequency of occurrence of observations in an observed sample and the expected frequencies obtained from the hypothesized distribution. Theorem: A goodness-of-fit test between observed and expected frequencies is based on the quantity χ2 = ∑ test (Oi − E i ) 2 Ei where χ 2 is a value of the random variable whose sampling distribution is test approximated very closed by the Chi-square distribution. one has to decide whether the die is fair die or not. The decision criterion described here should not be used unless each of the expected frequencies is at least equal to 5. i. Goodness-of-fit Test A test to determine if a population has a specified theoretical distribution.e. P( X = i) = 1 6 for i = 1. χ test > χ α constitutes the critical region.Chi-square Tests There are two types of chi-square tests: goodness-of-fit test and tests for independence. 2. 2 2 For a level of significance equal to α . Null hypothesis: the die is a fair die. and 6 Alternative hypothesis: the die is not a fair die 51 .

45 3.45 .95 3.45 .95 .4. Class boundaries 1.2.05 = 11 .95 1.Level of significance: 0.45 .0.4.45 4.1.95 .95 .95 2.2. Null hypothesis: the distribution can be approximated by a normal distribution Alternative hypothesis: the distribution cannot be approximated by a normal distribution 52 . ∴ χ > χ5.05 2 2 Critical region: ν = n − 1 = 6 − 1 = 5 .45 2.07 Computation: 1 Expected value = nP ( X = i ) = 120 ( ) = 20 6 i Observed (Oi ) Expected ( E i ) Oi − E i 1 20 20 0 6 2 22 20 2 3 17 20 -3 4 18 20 -2 5 19 20 -1 6 24 20 4 χ2 = ∑ i =1 (Oi − Ei ) 2 Ei = 0 2 2 2 ( −3) 2 ( −2) 2 ( −1) 2 4 2 + + + + + = 1.3.95 Oi 2 1 4 15 10 5 3 z-value p-value Ei Chi-squared test can be applied to test whether the above frequency distribution can be approximated by a normal distribution or not.7) falls outside the critical region. so reject the alternative hypothesis and conclude that the die is a fair die.45 .7 20 20 20 20 20 20 Conclusion: As the sample χ2 (= 1.3. Example 2 The following distribution of battery lives may be approximated by the normal distribution.

45 2. 0.1.45 .45 4.3.95 .6969   0.4125   L − 3.95 .95 .7 4.4.7 Class boundaries 1.45 .6969  where Li and U i are the Lower and Upper Boundaries of the ith class.4.6969  53 .05 2 2 Critical region: χ > χν .3. 0. and n is the number of cells.7 2. U − 3.95 1.45 .5 ∑ fx 2 = 484 .2 3.45 . ( fi ) Class mark ( xi ) 1.95 Oi 2 1 4 15 10 5 3 n = 40 x= ∑fx = =136 .2.4125 z − value =  i <Z < i .4125 40 2 2 484. the mean and standard deviation of the frequency distribution have to be found first.4125   L − 3.75 − 136.45 3.95 2.7 3.5 = 3.5 s= ∑ fx 2 − ( ∑ fx) / n n−1 = ( 39 40 = 0.2 4.05 where ν = n − 3 .2 2.Level of significance: 0.4125 p − value =  i <Z < i 0.6969   0.75 ∑ fx ∑f 136 .95 3. Computation: For finding the expected values.2.6969 ) U − 3.

4125   L − 3.832 10.2. reject the alternative hypothesis and conclude that the distribution of battery lives can be approximated by a normal distribution.636 6.05<Z<0.1 2. so the sample χ2 (= 2.612 ) 2 (10 − 10 .3.05 0.95 .0681 Ei 0.4.1. 54 .184 ) 2 (15 − 10 .716 2.4125 E i = (40 ) P  i <Z < i 0.45 2.45 .6969   0. or contingency table.77 0.05 = 3.U − 3.901 2 Conclusion: Since χ1. the number of cells (n) is 4.95 2. As such.1708 .824 = 2.841 .10<Z<-1.824 ) 2 = + + + E 10 . The observed frequencies of two variables are entered in a two-way classification table. As such.6969  Class boundaries 1.95 3. 0.2.612 10 .38) 2 (8 − 8.2653 .901) falls outside the critical region.724 In order to satisfy the rule that the expected value in each cell is larger than or equal to 5.49 1.0659 .38<Z<-0.45 . we have to combine the first three classes in to one cell and the last two classes into another cell.184 10 .95 Oi 2 1 4 15 10 5 3 z-value Z<-2.2595 .38 -1.95 .3.1525 .77<Z<1.4.66 -0. Test for Independence The Chi-square test procedure can also be used to test the hypothesis of independence of two variables/attributes.45 .45 4.95 .95 1.0179 . Remark: The expected frequency of the cell in the ith row and jth column in the contingency table E ij = (total of row i) * (total of column j) grand tota l The degrees of freedom for the contingency table is equal to (r − 1) (c − where r is the number of 1) rows and c is the number of columns in the table.612 10.38 8.45 . χ2 = ∑ (O − E ) 2 (7 − 10 .45 3.38 6.10 -2.66<Z<0.49<Z p-value .

Level of significance: 0.05 2 2 Critical region: χ > χν .1) Computation: Grade Point Average 2 3 11 (10.33) 16 ( 14. the two characteristics are independent.6) 2 (16 −14 . 0.05 .Example 3 Suppose that we wish to study the relationship between grade point average and appearance.8) 29 χ2 = ∑ (O − E ) 2 (14 − 9) 2 (11 − 10 .53 55 .53 ) 2 + + + 12 .33) 2 (10 − 11) 2 (5 − 9.4 13 .47) 4 (6.6 14 .4) 27 4 Totals 40 56 24 120 5 (9. Grade Point Average 2 3 4 11 ( ) 16 ( ) 4( ) 31 10 ( ) 16 ( ) 7( ) 33 5( ) 14 ( ) 10 ( ) 29 Appearance attractive ordinary unattractive Totals 1 14 ( ) 10 ( ) 3( ) 27 Totals 40 56 24 120 Null hypothesis: There is no relationship between grade point average and appearance.47 ) 2 (16 −15 .67 + (10 −12 . the two characteristics are not independent.53) 10 (5. Alternative hypothesis: There is a relationship between grade point average and appearance. That is.4) 7 (6.2) 31 10 (11) 16 (15. That is.33 11 9.67) 14 (13.6) 3 (5. where ν = (r -1)(c .6) 33 Appearance attractive ordinary unattractive Totals 1 14 (9) 10 ( 12.47 15 .4) 2 (14 −13 .67 ) 2 = + + + E 9 10 .

818) falls outside the critical region.2) 2 (7 − 6.991 Computation: Result Purchases show profit A 63 (63) Manager B 71 (63) C 55 (63) Total 189 56 . 0. One hundred randomly selected purchases obtained for each of the managers showed the results given in the table. so sample χ2 (=10. Do these data provide evidence of differences among the rates of successful purchases for the three portfolio managers? Test with α = 0.596 .05 =12 . Result Purchases show profit Purchase do no show profit Total A 63 37 100 Manager B 71 29 100 C 55 45 100 Null hypothesis: the rates of stock purchases that resulted in profit were the same for the three stock portfolio managers Alternative: their rates were not all the same Level of significance: 0.4 6.4) 2 ( 4 − 6.818 5. and C was conducted to compare the rates of stock purchases that resulted in profits over a time period that was less than or equal to one year. B.2 6.6) 2 (10 − 5.8 2 Conclusion: Since χ6. Remark: The approach for the test of homogeneity is the same as for the test of independence of variables/attributes. So reject the alternative hypothesis and conclude that there is no evidence to support there is relationship between grade point average and appearance.6 5.05 2 2 Critical region: ν = (2 −1)( 3 −1) = 2 . ∴ χ > χ2.05 = 5. A. Test for Homogeneity To test the hypothesis that several population proportions are equal.+ (3 − 5.8) 2 + + + = 10 . 0.05 . Example 4 A study of the purchase decisions for 3 stock portfolio managers.

49 63 63 63 37 37 37 Conclusion: As the sample χ2 (= 5. 57 .Purchase do no show profit Total 37 (37) 100 29 (37) 100 45 (37) 100 111 300 χ2 = (63 − 63) 2 (71 − 63) 2 (55 − 63) 2 (37 − 37 ) 2 (29 − 37 ) 2 (45 − 37 ) 2 + + + + + = 5.49) falls outside the critical region so reject the alternative hypothesis and conclude that there is no sufficient evidence to support the rates of purchases resulted in profit of the three portfolio managers were different.