See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/329772871

Introduction of Statistics

Book · December 2018

CITATIONS READS

0 62,751

1 author:

Z. A. Al-Hemyari
University of Nizwa
114 PUBLICATIONS   549 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

2. Assessing the performance of private HEIs: 2012-2014 View project

13. Quality and Performance Management Model for Higher Education Institutions: Statistical and Quantitative Approach View project

All content following this page was uploaded by Z. A. Al-Hemyari on 19 December 2018.

The user has requested enhancement of the downloaded file.

 Introduction
to
Statistics

Prof. Dr. Zuhair A. Al-Hemyari

 University of Nizwa

College of Arts and Sciences

Department of

Mathematical and Physical

Sciences

Section : Statistics
STAT101: Introduction to Statistics
Lecture/Tutorial (3:2) 4CR

Brief description of what statistics is all about. Common areas

of use. Types job opportunities. Sources of data : Internal data
and external data .Types of data: Nominal, ordinal, interval
and ratio. Statistical data terminology : Population,
population characteristics, sample ,population census ,
sampling and non-sampling errors,… and variables. Obtaining
data. Descriptive statistics: Summarizing data. One and two
way frequency tables and how to make them. Scatter plots.
Measures of central tendency, mean, median, mode (grouped
and non-grouped data) . Measures of variability, standard
deviation, range, skewness measure (grouped and non-grouped
data).Quartiles and percentiles. Definitions of probability.
Random variables and their distributions . Simple
uses/application of binomial and normal distributions. Simple
indices and rates . Scatter plots and simple linear regression.
STAT105: Statistics for Engineers
(Lecture/Tutorial (2:2) 3CR
This course is an introduction to the basic concepts of
probability and statistics. The examples and exercises strongly
emphasize engineering applications.
 Introduction to Statistics

STAT101

Instructor :

Prof. Dr. Zuhair A. Al-Hemyari

Contents

• Introduction
1.1 Examples of statistical problems
1.2 Sources of data
1.3 Statistical data terminology
1.4 The acquisition of data: Surveys and
experiments
1.5 Obtaining data
1.6 Constructing questionnaires and schedules
1.7 Variables and scales of measurement
Tutorial 1

• Frequency Distributions
2.1 Introduction
2.2 Frequency distributions
2.3 Graphical presentations
Tutorial 2

• Measures of Location
3.1 Introduction
3.2 The mean
3.3 The mean of a distribution
3.4 The coding method
3.5 The mode
3.6 The median
3.7 Other numerical measures
 3.7.1 Geometric mean
3.7.2 Quartiles and Percentiles

Tutorials 3.1 & 3.2

• Measures of Variation
4.1 Introduction
4.2 The range
4.3 The standard deviation
4.4 Measure of relative variation
4.5 Measure of skewness
Tutorial 4

• Introduction to Probability &

Random Variables
5.1 Introduction
5.2 The sample and event spaces
5.3 Computing probabilities from the sample space
5.4 Permutations, combinations, and other counting
rules
5.5 Random variable
5.6 Probability mass function
5.7 Probability density function
Tutorials 5.1 & 5.2
• Binomial & Normal distributions
6.1 Probability function
6.2 The binomial distribution
6.3 The normal distribution
Tutorial 6

• Regression Analysis
7.1 Introduction
7.2 Relationships between variables
7.3 Simple linear regression model
7.4 Fitting of a simple linear regression model
Tutorial 7

Probability Tables
1. Binomial cumulative distribution function.
2. Standard Normal distribution function

References
Introduction

to

Statistics

1
 1
Introduction

1.1 Examples of statistical problems

1.2 Sources of data
1.3 Statistical data terminology
1.4 The acquisition of data: Surveys and
Experiments
1.5 Obtaining data
1.6 Constructing questionnaires and schedules
1.7 Variables and scales of measurement
Tutorial: 1

2
1.1 Examples of Statistical Problems

The word "statistics" conveys a variety of meanings to people, many of

which are inaccurate or, at the very least, misleading. To some, the word
suggests only a plethora of mind-boggling tables, charts, and figures. Other
people consider statistics to be an imposing form of mathematics. The use of
the word certainly had an inauspicious beginning, as might be suspected
from a cursory study of the word, for it was originally a term used to denote
a collection of figures, graphs and the like which contained useful
information for the state (primarily budget information such as taxation
figures).
Used in the context of its original meaning, statistics generally refers to
information about an activity or a process that is expressed in numbers listed
in tables or illustrated in figures. But, since its early connotation, statistics
has grown to encompass a larger role than presenting us with charts, graphs,
and tables or figures. In a modern setting, statistics refers to the science of
collecting, presenting, and analyzing numerical data. A statistician is
a person who engages in one or more of the following tasks:
(1) the clerical activities of tabulating, summarizing, and displaying
statistical data.
(2) analyzing data by using statistical methods, usually for the purposes of
decision making.
(3) advancing the science of statistics by developing new and better
analysis methods.
The levels of expertise required by statisticians ranges from mastering
simple clerical operations with data to advanced training in applied
mathematics, and statisticians are needed at all levels.
The use of statistics has permeated almost every facet of our lives. The
daily newspapers and the televised news reports supply us with numerous
summaries of data such as stock market reports, financial summaries, and
crime statistics-and with the results of statistical analyses-weather forecasts,
political election outcome predictions, and so on.
Governments, businesses, and individuals collect statistical data required
to carry out their activities efficiently and effectively. The rate at which
statistical data are being collected is staggering and is primarily due to the
realization that better decisions are possible with more information and,
perhaps more importantly, to technological advances that have enabled the
efficient collection and analysis of large bodies of data.

3
 The most important technological advance in this area has, of course,
been the development of the electronic digital computer. Statistical concepts
and methods, and the use of computers in statistical analyses, have affected
virtually all disciplines biology, physics, engineering, economics, sociology,
psychology, business, and others. In business and economics, the
development and application of statistical methods have led to greater
production efficiency, to better forecasting techniques, and to better
management practices. It is becoming increasingly apparent that some
knowledge of statistics and computers is essential for careers in economics,
business, administration, and many other fields as well. To gain an
appreciation for the breadth of applications of statistics to business and
economic problems in particular, let us consider three examples.

Example 1.1
In operations management, a primary concern is controlling the quality of
the items being produced. If the product is a transistor radio battery, for
example, we may be concerned with the longevity of the batteries. Suppose
it is desired that at least 95 percent of the batteries last through at least 20
hours of continuous use. The actual percentage of batteries lasting m ore
than 20 hours could be determined by inserting each and every battery
produced into a transistor radio and recording its time to failure, but then
there would be no batteries to sell. Rather, a manager may wisely decide in a
day's production to pull every 100th battery off the production line, insert
the sampled batteries in electrical test circuits and record their times to
failure. The percentage of these batteries lasting through more than 20 hours
of continuous use could be used to estimate the percentage of all batteries
produced during that day which will last more than 20 hours. Moreover, if
this estimated percentage drops much below 95 percent (say to 80 percent),
the manager may wish to stop the production line until he can determine
why the percentage of bad batteries appears to be greater than the tolerated 5
percent. The manager is using a percentage statistic computed from a sample
of all batteries produced to arrive at a decision regarding the quality of the
set of all batteries produced on a given day.This example portrays a common
phenomenon in quality control: destructive sampling. It is impossible to test
the quality (longevity) of each battery produced because the test for
longevity will ordinarily involve its destruction.

4
 The manager has little recourse but to sacrifice a small number of
batteries (the sample) in order to gain information about the entire set of
batteries comprising the daily production (the population).

Example 1.2

Determining the saleability of a new product is a constant problem posed

to many marketing research groups. In order to determine whether a new
kitchen-ware product will sell, the marketers might conduct a house-to house
survey of 1,000 households selected randomly in the product target areas,
during which they present the product to the housewife for evaluation. The
percentage of the housewives willing to buy the product at its listed price,
together with other information obtained from the interviews could be used
to decide whether or not the new item should undergo full-scale production.

Example 1.3
Politicians and their supporters aelection as the campaign heads towards
final balloting. By sampling 1,000 regire immensely interested in knowing
their prospects of winning an stered voters prior to the election, the
percentage who claim they will vote for a given candidate may be used to
estimate the percentage of the votes the candidate will receive hi the
election. The estimated percentage could be used to decide, for example,
whether a greater campaign effort (more money) is required to assure the
candidate's election. There are many more examples in business and other
areas which might be cited, but the above five should indicate the many
ways in which statistics can be employed. In the first two examples,
statistics is used to describe large bodies of data. In this application, the
word "statistic" is being used to describe a specific numerical quantity such
as an average or a total, and the collection of statistics is used to summarize
or condense a large set of numbers. These compiled statistics may, in turn,
be used to assist in decision making. In the last three examples, statistics
may be interpreted in a much broader sense; namely, the process of drawing
conclusions about an entire population or collection of things based upon a
sample, a subset of the population or collection.

5
 Most students probably view statistics in the context of the first example
above; that is, as tables of figures, charts, and graphs (batting averages, pie
charts illustrating the sources of government revenue, and so on). This
concept is called descriptive statistics and was at one time the principal use
of statistics in business. Currently, there is an increasing interest in the
methods and uses of inferential statistics-the process of drawing inferences
about the whole (the population) from a subset of it (the sample), as
exemplified in Examples 1.1-1.3. Schematically, the process of drawing
inferences about an unknown population numerical quantity (the proportion
of defectives in a production lot, mean incomes of a class of laborers, etc..)
is illustrated in figure 1.1. Units are selected from the population to form the
sample which in turn is used to draw inferences about the population
characteristic of interest. Much of this text is devoted to the study of
statistical inference. In subsequent sections of this chapter, we will focus
attention on the sources of data, methods of obtaining data, and data
measurement considerations.

Figure 1.1 The Statistical Inference Process

1.2 Sources of Data

In most instances, businesses use internal data; that is:

data arising from bookkeeping practices, standard operating, business
procedures, or planned experiments by research divisions within the
company. Examples are profit and loss statements, employee salary
information, production data and economic forecasts. Occasionally, it may
be necessary or desirable to use sources of external data. By external,

6
We mean:
sources of data outside the firm. External data may be of two types:
primary data and secondary data. By primary data , we mean data obtained
from the organization which originally collected them. An example is the
population data collected by and available from the US. Bureau of the
Census. Secondary data come from a source other than the one which
originally collected them.

Ordinarily, if external data must be used, it is recommended that primary

data be sought out since it will not have undergone any "refining" by the
secondary source.

In the election survey (Example 1.3) in Section l. l, the Statistical Abstract

provide numerous tables of both primary and secondary data, such as past
voting records in districts and numbers of registered Democrats and
Republicans, which may supply important information in conjunction with
the internal sampled data on estimating a candidate's probability of being
elected.
There are many excellent sources of published (primary and secondary) data
which have been compiled by the state and government, by business and
economic associations, and by commercial sources (periodicals).
Some examples are: The Statistical Abstract of the United States (published
annually by the Bureau of the Census), Survey of Current Business
(published by the Department of Commerce), Monthly Labor Review
(published by the Bureau of Labor Statistics), Harvard Business Review
(periodical), Business Week (periodical), The Wall Street Journal
(periodical), Dun's Statistical Review (periodical) and The Journal of
Management Science (association journal). Additional sources of external
data, available in most reference libraries, are The Economic Almanac,
Federal Reserve Bulletin, Life Insurance Fact Book, International Financial
Statistics, and Business Conditions Digest.
Caution must always be exercised in using external sources of data,
particularly secondary sources, as they may contain errors in transcription
from the primary source. When external data are used, the conditions under
which the data were collected and summarized must be determined to assure
that they are relevant for the intended use. This determination usually re-
quires identifying and locating the primary source,

7
 which typically will discuss any restrictions placed on the data due to the
process of their collection. Thus, while secondary sources of data are
convenient, it usually is prudent to seek out and use primary sources of
external data.

1.3 Statistical Data Terminology

When statistical data are collected and analyzed, it is usually in the
context of populations and their characteristics.

Definition 1.1
Population and population characteristic
A population is the totality of units under study. A population
characteristic is an attribute of a population unit.
We may be interested, for example, in the salaries of workers in a particular
industry. If so, the population is the totality of these workers and the char-
acteristic of interest is each worker's salary. In collecting the salary data, we
may be interested in other population characteristics as well, including sex,
age, educational level, and other information. In general, a population unit
may have one or more characteristics of interest in a particular study.

As another illustration of a population, a firm may be interested in the

proportion of defective units of a certain product that it has produced in a
large lot stored in a warehouse. The population is the totality of units in the
warehouse and the characteristic is the acceptability of each unit of the
product-it is either defective or non-defective .

A population may be either entirely inspected or partially inspected. When

the data are produced by measuring the population characteristic for each
and every unit in the population, we say a census of the population has been
taken.

8
 Definition 1.2
A population census
A population census is the evaluation of each and every unit in the
population under study.
In some situations, it is possible to take a complete census of the popula-
tion. This rarely occurs in business unless the population size is very small,
due to cost and time considerations. A census of the US. population is under-
taken every ten years and it is truly a Herculean effort, subsidized, naturally,
by the taxpayers. The US. census produces a wealth of data of considerable
importance to the federal government and to firms and institutions, many of
whom view the census as an important source of external data.
In most instances, it is not possible to take a census of a population. It may
be too costly, too time consuming, or the evaluation process may destroy the
population unit as in Example 1.1.

Definition 1.3
A sample
A sample is a part of a population in which the population
characteristic is studied so that inferences may be made from the sample
study about the entire population.

A classic example of a situation in which samples must be used rather than

a census taken is destructive sampling in which the process of evaluating a
unit of the population destroys or irrevocably damages that unit.

Example 1.4

Suppose a tire manufacturer wishes to claim its new radial tire will last
40,000 miles or more. To support this claim, a sample of all tires produced
(the population) is selected for testing to determine how many miles the tires
will last. Since testing destroys the tires, a complete census of the population
is impossible.

The advantages of sampling to taking a census are rather obvious. A sample

is less expensive than a census, it can produce data more quickly, and the
data are often more reliable because more time can be spent studying each
sampled unit. But there clearly is a price to be paid as well. By looking at
only a portion of the population, we are subject to errors because the sample

9
may not be representative of the whole population.

Definition 1.4
Sampling error
Sampling error is the difference between the result of studying a
sample and infering a result about the population, and the result of a
census of the whole population.

Example 1.5

As an illustration of sampling error, suppose we are interested in the

average salary of unionized workers in a specific industry and we know
from union membership lists that there are presently 1,000 workers in this
industry. Had we taken a census, we may have found that the average salary
is, let us say, $25,000. Based upon a sample of 100 workers, we might find
that the average salary is $27,200. The difference between the two figures-
$2,200 is the sampling error.

Errors in acquiring and tabulating statistical data can arise in other ways as
well, and these errors are called nonsampling errors.
Definition 1.5
Nonsampling error
Nonsampling errors are errors that occur in acquiring, recording, or
tabulating statistical data that cannot be ascribed to sampling error.
They may arise in either a census or a sample.
Nonsampling errors are usually more difficult to control and detect than
sampling errors.
Example 1.6
Suppose we are acquiring data on the 1,000 unionized workers mentioned
above. If we approached a particular worker and asked for his or her income,
we could be lied to-a troublesome and frequent source of nonsampling error
when a sensitive question is asked directly of a person. (What is your grade
point average?) In some instances, a person may give a false response out of
ignorance rather than by design. Another source of nonsampling error is in
recording the data. A"7" may be written as a "9," the decimal point may be
incorrectly placed, and so on. Errors may also occur in tabulating the data-
keypunching errors in preparing computer cards and typing errors in

10
 transcribing data, for instance. It is always necessary to carefully edit data to
minimize the chance of nonsampling errors adversely affecting the statistical
analysis of the data.
The identification of the units in a population under study can often be a
surprisingly difficult task. We refer to a listing of population units as a
frame.

Definition 1.6
Population frame
The listing of all units in the population under study is called the
population frame.
Example 1.7
If the population is a production lot of units stored in a warehouse,
production records will give us a listing of the serial numbers of the units
from which each unit may be identified.
Example 1.8
If the population is the 1,000 unionized workers in a specific industry, union
membership records may serve as a frame.
But, what about a frame for all persons who will vote in a particular
election? A listing of registered voters is not appropriate, because in many
elections less than 50 percent of the registered voters actually vote. Some
classic errors have been made in identifying the frame.

1.4 The acquisition of Data: Surveys and Experiments

When internal or external data are not readily available or are incomplete
in a study attempting to answer questions about a population, a survey or an
experiment may be conducted to provide the required information.
Definition 1.7
Statistical survey
A survey is a process of collecting data from existing population
units, with no particular control over factors that may affect the
population characteristics of interest in the study.

11
 Most of us are very familiar with surveys. As students, we are asked about
our opinions regarding dining hall food, impending tuition hikes, teaching
effectiveness and so on. Filling out survey questionnaires or answering an
interviewer's questions has become a routine occurrence in most of our lives.
Example 1.9
suppose we are interested in acquiring data on the salaries of 1,000
unionized workers in a specific industry. The population characteristic
"salary" may be affected by a host of factors-age , race, sex, educational
level, etc.. As we elicit a particular worker's salary, we have no control over
educational level, age, and so on these are existing attributes of the worker.

In contrast to a survey is a statistical experiment in which we do exercise

control over factors that may affect the population characteristics of interest.

Definition 1.8
Statistical experiment
An experiment is a process of collecting data about population
characteristics when control is exercised over some or all factors that
may affect the characteristics of interest in the study.
Example 1.10
We may be interested, in the yield of a chemical process that is affected by
temperature and pressure. A variety of settings for temperature and pressure
could be selected, and the chemical process run for each setting to determine
the yield. In this way, the joint effect of temperature and pressure on yield is
studied in a controlled manner.
Example 1.11

In management, we may be interested in the effects of a training program

on the first year performance of new employees. A set of new employees
may be split into two groups such that both groups are approximately alike
in terms of age, sex, education, and other factors. The training program
could be administered to one group and not to the other (the control group).
At the end of the first year, performance characteristics could be measured to
assess the effects of the training program, accounting for factors other than
the training program that may affect performance.

12
 Experiments almost always provide better information than do surveys,
but both are extremely important and useful tools for acquiring data. Though
an experiment should be preferred to a survey, much of the data used in
statistical analyses in business and economics are survey data. There are a
number of reasons for this. First, most internal and external data are
collected by surveys. Second, it is not always possible to conduct an experi-
ment to acquire the needed information. An interesting example of this is the
effect of smoking on health. Virtually all data on the relationship between
smoking and health are survey data; other factors that may affect health,
such as age, race, sex, and physiological properties, are not in the control of
those collecting the data. To run an experiment in this case would involve
controlling persons' lives. Some people in the experiment would be required
to smoke while others would not. It is neither feasible nor desirable to
approach the acquisition of the data for the study of the relationship between
smoking and health in this way.

The planning of a survey or an experiment is essential to ensure that the

resulting information will be useful. A good plan usually involves the
following steps (these steps are applied to the quality control problem in
Example 1.1):
1. A clear and detailed statement of the problem.
The statement of the problem should clearly indicate that we are interested
in determining whether or not the percentage of good batteries (those lasting
through 20 hours) exceeds a specified percentage (95 percent). The
population is comprised of all batteries produced during a chosen period of
time (a day or a week) and the characteristic in the population of interest is
the number of hours the battery will continuously operate before failure.

2. A decision to survey or to experiment.

In this problem, it is possible to answer the question about the population by
experimentation. We may test each selected battery under the set of
conditions in which it was designed to operate.
3. A decision to take a census or a sample.
This is a case of destructive sampling. In determining the proportion of
batteries that will last 20 or more hours, the tested batteries are "spent." We
must, therefore, take a sample of batteries.

13
4. Designing the survey or experiment.
The experiment must be designed so that we isolate the characteristic of
interest-the lifetime of the battery. Test circuits must be constructed and
carefully monitored when the selected batteries are inserted for testing.
5. Collecting and analyzing the data.
For each battery, the time to failure is recorded and the proportion of bat-
teries lasting 20 hours or more is calculated.
6. Reaching conclusions about the population characteristics.
The sample proportion of batteries surviving 20 or more hours is used as an
estimate of the population proportion that survive 20 or more hours.
7. Reporting the results.
The report should include a thorough description of the problem, the
sampling design, the testing method and the inferences. Sufficient monies
should be allocated for a competent writing of the report. Indeed, many
companies employ technical writers to put into "laymen's" words the
experimental results.

The manner in which a sample is drawn, the methods of analyzing statistical

data , and the kinds of inferences that may be drawn from the
analysis (steps 4, 5 , and 6) are major topics in this text. It is important not to
minimize the other steps , particularly steps I and 7. A clear and detailed
statement of the problem is essential in planning a survey or an experiment.
And the best analysis of survey or experimental results is meaningless unless
the analysis can be accurately and understandably reported.

1.5 Obtaining Data

Once it has been determined that a survey or experiment is required, there
are a variety of methods that may be employed. The most difficult
problems arise when gathering information from people in surveys and the
methods most relevant to this situation will be emphasized.

1.5.1 Self -Enumeration

Self-enumeration is probably the most common method of acquiring data
from people in a survey or in an experiment. Questionnaires are usually
distributed to selected individuals by mail,

14
 although the distribution mechanism depends to a large extent on the
purpose and nature of the questionnaire. For example, if the purpose of the
questionnaire is to survey the attitudes of those using public transportation,
the questionnaire may be distributed to persons while commuting to and
from work on buses, subways, and trains.
The use of questionnaires suffers from two serious drawbacks. First, if the
respondent has difficulty in interpreting the questions, no one is available for
assistance. If this situation arises, the information received may contain a
high degree of nonsampling error or the respondent may become frustrated
and not bother completing or returning the questionnaire. Further, if a
questionnaire is mailed to a household, it is often not clear who in the house-
hold responded to it. Second, questionnaires have typically an extremely
poor response rate. It is not uncommon to have less than 30 percent returned
on the first mailing of a questionnaire. The principal advantage of a
questionnaire is the low cost relative to the other means of obtaining
information. Most mail questionnaires may be bulk mailed at a reasonable
rate. But it is almost always necessary to contact nonrespondents to the first
mailing by subsequent mailings, telephone calls, or personal interviews, and
these costs must be planned for in a well-designed self-enumeration survey
or experiment. In most instances, those who do respond to the first mailing
of a questionnaire are not representative of the entire population. To use only
their responses would tend to bias the analytical results. Some self-
enumeration questionnaires do enjoy high initial response rates. Examples
are questions asked on warranty cards that must be returned to the
manufacturer for warranty coverage of a new product .

1.5.2 Personal Interview

In most situations, the best method of eliciting information from indi-

viduals is by a personal interview. The interviewer personally contacts
individuals selected to participate in the survey or in the experiment.
Responses are recorded on a schedule (a questionnaire form filled out by the
interviewer).
The personal interview method produces a higher response rate, than self-
enumeration questionnaires and further allows the interviewer to clear up
any misunderstandings about any of the questions on the schedule. But
personal interviews are generally very expensive. Interviewers must be
carefully selected and trained,

15
 and sufficient remuneration must be provided to ensure that the interviewer
is competent and dedicated to the chore. It is always prudent in a personal
interview survey to call a selected set of respondents to ensure that they were
in fact contacted (as opposed to the interviewer filling in fake responses), to
ascertain if the interviewer's demeanor was appropriate, and to determine
whether the interviewer may have biased responses by making gestures
when stating the questions or recording the responses.

Overall, the personal interview method of conducting an experiment or

survey, where the population units are people, is the best way to acquire data
if the process is properly planned and executed, and if it can be afforded.

1.5.3 Telephone and Internet Interviews

Occasionally, it is possible to conduct an interview over the telephone or

internet with the interviewer working from a schedule as in a personal
interview. Polls to determine the most popular programs on television are
frequently conducted in this manner. Telephone and internet interviews are
usually less expensive than personal interviews, but the response rate is
lower and fewer questions may be asked before the respondent tires of the
proceedings. And, not everyone owns a phone or e-mail-even today.

1.6 Constructing Questionnaires and Schedules

There are three basic steps to constructing a questionnaire or schedule:

(1) designing the instrument, (2) the pretest, and (3) editing the results. The
construction of a questionnaire or schedule instrument is time consuming
and difficult. There is a natural tendency to rush through the construction of
the instrument so that the data collection process can commence. But time
spent in this stage of a well-planned survey or experiment is invariably
found to be extremely valuable in retrospect.
The proper construction of questionnaires is a skill which is generally
developed only by experience in the use of research methodology or by on -
the-job training. We will discuss only some of the basic concepts
concerning the construction of a questionnaire.

16
1.6.1 The Design

There are basically three kinds of questions that may be asked:

dichotomous, multiple choice, or free answer. In the dichotomous question,
the respondent is asked to select one of two responses, usually "yes" and
"no." For example, in a transportation study, a worker may be asked,

Did you drive a car to work this morning? YES ( ), NO ( ) .

The dichotomous question is simple and straightforward, and perhaps

comes closest to decisions that respondents are used to making.
In the multiple choice question, the respondent is asked to select one of a
number of responses:

What is the likelihood of your using the following services for preventive
health care purposes in the next two years? (a) Dental check-up, (b) Eye
exam, (c) General physical.

a b c
Extremely unlikely () () ()
Unlikely () () ()
Slightly unlikely () () ()
Not certain () () ()
Slightly likely () () ()
Likely () () ()
Extremely likely () () ()

The multiple choice question gives the respondent a greater range of

responses to choose from, but it may also request a more qualified response
than the respondent is prepared to make. For instance, a respondent may
answer "yes" to the dichotomous question, "will you have a physical this
year" when it is not a certain event-good intentions are not always realized.
Yet, the respondent may not be able to properly conceptualize the assign-
ment of a likelihood (slightly likely, likely, extremely likely) to the event, "I
will have a physical this year." All too often, responses in situations like this
lead to "end-loading"-selecting the response which most closely approaches
a simple "yes-no" response.

17
 In this case, the respondent would select the "extremely likely" response
in place of "yes" if he or she is given the multiple choice response format,
for instance.

In the free answer form, the respondent is asked to answer a question in

his or her own words in essay form:
What is your opinion of the dining hall food and service?
The difficulty with the free answer question is in classifying the responses.
This may not only be difficult and somewhat arbitrary, but it is also
extremely time consuming.
In most instruments, it usually is necessary to employ all three types of
questions to elicit the information required. ,
The order of the questions in the instrument can be extremely important.
The questionnaire or survey should begin slowly, with easily answered
questions to develop rapport with the respondent. Respondents tend to "tie"
questions together and a particular ordering of questions may produce a
different set of responses than another set for this reason.

The degree of directness of the questions is also important. If sensitive

questions are asked directly, respondents may distort their answers. This
invariably happens when a person is asked for his income. To elicit informa-
tion about sensitive questions, indirect questions may be employed. For
example, we may ask the respondent to indicate his salary range among a set
of ranges. Later, we may ask what proportion of his monthly income is spent
on food and much later, what his average monthly expenditure for food is.
We may be able to determine a person's salary indirectly in this way better
than by directly asking for his or her income. At the very least, we have a
consistency check to determine how reliable the responses are.
It is important that the questions are stated clearly and do not bias the results.
Ideally, the question should have the same meaning to every respondent in
the survey or experiment. And the questions should be relatively short. Bias
may arise when leading questions are used, such as:
The food in the dining hall is rotten.
Agree ( ) Uncertain ( ) Disagree ( ) .

18
 Given to the typical college student, the response will invariably be,
"agree." A less biased question might be, "The food in the dining hall is of
acceptable quality."

1.6.2 The Pretest

The pre-test is an essential step in constructing a questionnaire or schedule

instrument. The instrument is given to a small number of respondents to
determine whether there are any problems with it. Almost always there are.
There may be ambiguous questions, the ordering may require changing, and
some questions may have to be asked in different forms. The time to identify
difficulties with the instrument is before the full scale survey or experiment
is conducted-not after.
Further, the information gathered during the pretest phase may be used to
estimate statistics required for the proper planning of the statistical design of
the experiment.

1.6.3 Editing

The completed questionnaires or schedules must be carefully checked and

edited for errors. Often, it is possible to design in questions which represent
internal consistency checks for the respondent's answers. Finding recording,
transcription, or clerical errors can be very tedious work, but it is necessary
if the data are going to be of value in decision making. Today, the computer
is used extensively to edit data. Various computer assisted techniques have
been developed to identify "outliers"-responses which are greatly different
from the majority of the responses. Many outliers result from recording,
transcription, or clerical errors, or from false information provided by the
respondent.

1.7 Variables and Scales of Measurement

The characteristic of the population under study is called a variable if it
can take on two or more different values among the population units. For
instance, if we are interested in the incomes of workers in a particular
industry, we may also record other characteristics about the worker as well,
as for example age, race, level of education, and sex.

19
 In this instance, the five characteristics-income, age, race, level of
education, and sex-are variables in the survey or experiment.
Further, we would call income a dependent variable and the other four
independent variables if we are concerned with how sex, age, level of
education, and race affect income. Income is the basic variable of interest
and our interest in the other variables is in their influence on income.
If we are measuring a set of variables from a population, the determination
of which are dependent and independent variables is a function of the
purpose of the survey or experiment. An independent variable in one study
may be a dependent variable in another.
A quantitative variable is one that can be measured numerically, such as
income and age. A qualitative variable is one that is nonnumeric , such as
sex, race, and level of education (high school, college, graduate school,
etc ...
In preparing data for analysis, we must be familiar with the four numerical
scales of measurement: nominal, ordinal, interval and ratio. The nominal
scale applies whenever we have used numbers only to categorize outcomes
of a variable. For instance, we could let a "male" be 1 and a "female" be 0,
but this numerical assignment is clearly arbitrary-a female could be
assigned 100 and a male, 0. The ordinal scale differs from the nominal
scale in that the ordering of the numbers has meaning. An example is the
responses to a multiple-response question:

Strongly Agree Uncertain Disagree Strongly

agree disagree

-2 -1 0 +1 +2
The numerical assignments of -2, - 1, 0, 1, and 2 indicate the degree of
agreement, but they could just as easily have been 0, 10, 100, 200, and 500,
respectively. The key here is that while a -2 indicates stronger agreement
than a - 1, the difference between -2 and - 1 may not be the same as
between 0 and + 1. In the interval scale, the relative order of the numbers
is important, but so is the difference between them. This scale uses the
concept of unit distance such that the difference between any two numbers
may be expressed as some number of units. The interval scale requires a
zero point, but its location may be arbitrary.

20
 Good examples of interval scales are the Fahrenheit and Celsius
temperature scales. Both have different zero points and unit distances. The
principle of an interval scale is not violated by a change in scale or
location or both. The ratio scale is used when the interval size is important
and also the ratio between two numbers has meaning. By this, we mean it
is appropriate to speak of one number being, say, twice as big as another.
This is clearly not possible with an interval scale, where, for instance,
80°F is not twice as "hot" as 40°F-measured on the Celsius scale, these
two temperatures are 27°C and 4°C, respectively, and 27°C is not twice
4°C. Examples of instances when ratio scales are appropriate are
measurements of heights, weights, and age. Most of the statistical methods
we will develop in this book require that the variable be measured at least
on the interval scale.

21
Tutorial 1

l. Briefly describe each of the following terms:

a. A statistician.
b. Schedule.
c. Descriptive statistics.
d. Questionnaire.
e. Inferential statistics.
f. Survey.
g. Population.
h. Experiment.
i. Population characteristic.
j. Variable.
k. Census.
l. Quantitative variable.
m. Sample.
n. Qualitative variable.
q. Sampling error.
r. Dependent variable.
s. Nonsampling error.
t. Independent variable.
u. Frame.

2. Distinguish between a schedule and a questionnaire. What is each used

for?

3. Distinguish between a survey and an experiment. Which is preferred and

why?

4. Distinguish between primary and secondary data. Which is the most

reliable? Why?

5. There are three kinds of questions that may be used in a schedule or in a

questionnaire. Describe each, and discuss its advantages and disadvantages.

6. In constructing a schedule or questionnaire, there are three primary steps:

design, pretest and editing. Describe each step.

22
7. There are four measurement scales: nominal, ordinal, interval, and ratio.
Describe each, and give an example of a survey question that may use
measurements of each type.

8. For each of the following, indicate the scale of measurement: a. Red (1),
Blue (0), Yellow (- 1)
b. Extremely Likely (5), Likely (4), Indifferent (3), Unlikely (2) and Ex-
tremely Unlikely (1).
c. Pressure in pounds per square inch; from 0 to ∞.
d. Volume in cubic centimeters from 0 to ∞.
e. Age in years 0 to ?
f. Salary in dollars 0 to ?
g. Rank of a state in population 1 to 50.

9. For each of the following, indicate whether it is a quantitative or

qualitative variable.
a. Hair color.
b. Sales volume of an automotive firm.
c. Sex of an individual.
d. Number of persons unemployed.

10. Distinguish between sampling and nonsampling error. Which can occur
in a census? Which can occur in a sample?

11. A manufacturer buys electronic parts from a supplier with the

understanding that 1 percent or less of the parts are defective. In a particular
shipment of 5,000 parts, the supplier finds in a sample of 100 parts that none
are defective. The manufacturer decides to check the parts as well and, in
another sample of 100 parts, finds that four are defective. On this basis, the
manufacturer decides to reject the lot.
a. How is it possible that one sample produces 0 percent defectives while
another produced 4 percent defectives?
b. Is it possible that the manufacturer is making a mistake by not accepting
the shipment.

23
 2
Frequency Distributions

2.1 Introduction
2.2 Frequency distributions
2.3 Graphical presentations
Tutorial 2

24
2.1 Introduction

Grouping, classifying, and thus describing measurements and obser-

vations is as basic in statistics as it is in science and in many activities of
everyday life. To illustrate its importance in statistics, let us consider the
problem of an economist who wants to study the size of farms in the United
States. Not even giving a thought to the possibility of conducting a survey of
his own, since the expense would be staggering, he immediately turns to one
of the many organizations that specialize in the gathering of statistical data,
namely, the US. Department of Commerce. This department not only
provides government agencies with statistical data needed for over-all
planning and day-by-day operations, but it also makes this information
available to businessmen and research workers in various fields. Like other
organizations engaged in gathering statistical data, it thus faces the problem
of how to present the results of its surveys in the most effective and the most
usable form. With reference to the information needed by the above-
mentioned economist, the Department of Commerce could print sheets
containing millions of numbers, the actual sizes of all farms in the United
States; it is needless to say, however, that this would not be very effective
and, without some treatment, not very "usable."
When dealing with large sets of numbers, a good over-all picture and
sufficient information can often be conveyed by grouping the data into a
number of classes, and the Department of Commerce could, and in fact does,
publish its data on the size of farms in tables like the following:
Table 2.1

Size of Farms in 1964 Number of Farms

(acres) (thousands)
Under 10 183
10-49 637
50-99 542
100-179 633
180-259 355
260-499 451
500-999 210
1,000 and over 145
Total 3156

This kind of table is called a frequency distribution (or simply a

25
distribution) : It shows the frequencies with which the farm sizes are
distributed among the chosen classes. Tables of this sort, in which the data
are grouped according to numerical size, are called numerical or quantitative
distributions. In contrast, tables like the one given below, in which the data
are sorted according to certain categories, are called categorical or
qualitative distributions, as table 2.2 below:

Table 2.2
1967 Motor Vehicle
Registration
(thousands)
United States 96945
Other North and Central America 8900
South America 5490
Europe 65969
Africa 3822
Asia 13937
Oceania 5519

Although frequency distributions present data in a relatively compact

form, give a good over-all picture, and contain information which is ade-
quate for many purposes, there are evidently some things which can be
obtained from the original data that cannot be obtained from a distribution.
For instance, referring to the first of the above tables, we cannot find the
exact size of the smallest and largest farms, nor can we find the exact
average size of the 542000 farms in the 50-99 acre group. Nevertheless,
frequency distributions present raw (unprocessed) data in a more usable
form, and the price which we must pay, the loss of certain information, is
usually a fair exchange.
Data are sometimes grouped solely to facilitate the calculation of further
statistical descriptions.

26
2.2 Frequency Distributions

The construction of a numerical distribution consists essentially of three

steps:
(1) we must choose the classes into which the data are to be grouped,
(2) we must sort (or tally) the data into the appropriate classes, and
(3) we must count the number of items in each class.
Since the last two of these steps are purely mechanical, we shall concentrate
on the first, namely, the problem of choosing suitable classifications. Note
that if the data are recorded on punch-cards or tape, methods that are
nowadays widely used, the sorting and counting can be done automatically
in a single step.
The two things we shall have to consider in the first step are those of
determining the number of classes into which the data are to be grouped and
the range of values each class is to cover, that is, "from where to where"
each class is to go. Both of these choices are largely arbitrary, but they
depend to some extent on the nature of the data and on the ultimate purpose
the distribution is to serve. The following are some rules which are generally
observed:
(a) We seldom use fewer than 6 or more than 15 classes. This rule
reflects sound practice based on experience; in any given example,
the actual choice will have to depend on the number of
observations we want to group (we would hardly group 5
observations into 12 classes), and on their range.
(b) We always choose classes which will accommodate all the data. To
this end we must make sure that the smallest and largest values fall
within the classification, and that none of the values can fall into
possible gaps between successive classes.
(c) We always make sure that each item goes into only one class. In
other words, we must avoid successive classes which overlap, that
is, successive classes having one or more values in common.
(d) Whenever possible, we make the class intervals of equal length,
that is, we make them cover equal ranges of values. It is generally
desirable to make these ranges (intervals) multiples of 5, 10, 100,
etc.., or other numbers that are easy to work with, to facilitate the
tally (perhaps, mechanically) and the ultimate use of the table.

27
 Note that the first three, but not the fourth,of these rules were observed in
the construction of the farm-size distribution on page 23, assuming that the
figures were rounded to the nearest acre. (Had these figures been rounded to
the nearest tenth of an acre, a farm of, say, 49.6 acres could not have been
accommodated, as it would have fallen between the second class and the
third.) The fourth rule was violated in two ways: First, the intervals from 10
to 49 acres, 100 to 179 acres, and 260 to 499 acres, among others, cover
unequal ranges of values. Second, the first and last classes are open-for all
we know, the last class might include farms of a million acres or more, and
if we had grouped profits and losses instead of acreages, the first class might
even have included negative values. If a set of data contains a few values
that are much greater (or much smaller) than the rest, open classes can help
to simplify the over-all picture by reducing the number of required classes;
otherwise, open classes should be avoided as they can make it impossible (or
at least difficult) to give further descriptions of the data.
As we have pointed out in the preceding paragraph, the appropriateness of
a classification may depend on whether the data are rounded to the nearest
acre or to the nearest tenth of an acre. Similarly, it may depend on whether
data are rounded to the nearest dollar or the nearest cent, whether they are
given to the nearest inch, the nearest tenth of an inch, or the nearest
hundredth of an inch, and so on. Thus, if we wanted to group the amounts of
the sales made by a saleslady in a department store, we might use the
classification given in table 2.3 below

Table 2.3
Size of Sale
(dollars)

0.00 - 4.99

5.00 - 9.99

10.00 – 14.99

15.00 – 19.99

20.00 – 24.99

Etc.

28
And if we wanted to group the heights of children measured to the nearest
tenth of an inch, we might use the classification shown in table 2.4

Table 2.4

Height

(inches)

20.0 - 29.9

30.0 – 39.9

40.0 – 49.9

50.0 – 59.9

Etc.
Note that in each of these examples the nature of the data is such that a value
can fall into one and only one class.
To give a concrete illustration of the construction of a frequency dis-
tribution, let us consider the following data(table 2.5) representing the scores
which 150 applicants for secretarial positions in a large company obtained in
an achievement test:

Table 2.5

27 79 69 40 51 88 55 48 36 61
53 44 94 15 65 42 58 55 69 63
70 48 61 551 60 25 47 78 61 54
57 76 73 62 36 67 40 51 59 68
27 46 62 43 54 83 59 13 72 57
82 45 54 52 71 53 82 69 60 35
41 65 62 75 60 42 55 34 49 45
49 64 40 61 73 44 59 46 71 86
43 69 54 31 56 51 75 44 66 53
80 71 53 56 91 60 41 29 56 57
35 54 43 39 56 27 62 44 85 61
59 89 60 51 71 53 58 26 77 68

29
 62 57 48 69 76 52 49 45 54 41
33 61 80 57 42 45 59 44 68 73
55 70 39 58 69 51 85 46 55 67

Since the smallest of these scores is 17 and the largest is 94, it would
seem reasonable (for most practical purposes) to choose the nine classes
going from 10 to 19, from 20 to 29, . . ., and from 90 to 99. Performing the
actual tally and counting the number of values falling into each class, we
obtain the results shown in table 2.6 . The numbers shown in the right-hand
column of this table are called class frequencies; they give the number of
items falling into each class. Also, the smallest and the largest values that
can go into any given class are referred to as its class limits; thus, the class
limits of the above table are 10 and 19, 20 and 29, 30 and 39, and so on.
More specifically, 10, 20, 30, . . ., and 90 are referred to as the lower class
limits, while 19, 29, 39, ..., and 99 are referred to as the upper class limits of
the respective classes.

Table 2.6

Scores Tally Frequency

10 -19 1 1
20 – 29 111111 6
30 – 39 111111111 9
40 – 49 1111111111111111111111111111111 31
50 – 59 111111111111111111111111111111111111111111 42
60 – 69 11111111111111111111111111111111 32
70 – 79 11111111111111111 17
80 – 89 1111111111 10
90 – 99 11 2
150

If we are dealing with figures rounded to the nearest whole number, as in

the size-of-farms distribution on page 23, the class which has the limits 10
and 49 actually contains all values between 9.5 and 49.5. Similarly, if we are
dealing with measurements rounded to the nearest tenth of an inch, as in the
height distribution on page 27, the class which has the limits 30.0 and 39.9
actually contains all values between 29.95 and 39.95, and the class which
has the limits 40.0 and 49.9 actually contains all values between 39.95 and
49.95. It is customary to refer to these dividing lines between successive
classes as the class boundaries, although they are sometimes referred to

30
instead as the "real" class limits. In order to make this concept apply also to
the classes which are at the two extremes of a distribution, we simply act as
if the table were continued in both directions. Thus, the first class of the
above distribution of the 150 scores has the lower boundary 9.5, while the
last class has the upper boundary 99.5.

It is important to remember that class boundaries should always be

"impossible" values, namely, numbers which cannot occur among the values
we want to group. We make sure of this by accounting for the extent to
which the numbers are rounded when we choose appropriate classifications.
For instance, the class boundaries of the size-of-sales distribution on page 28
are -0.005, 4.995, 9.995, 14.995, and so on. Similarly, for the distribution of
the, scores, the class boundaries are 9.5, 19.5, 29.5, … , and 99.5, while the
figures themselves are, of course, whole numbers. Had there been scores less
than 10 in this example, we would have begun the table with the class 0-9,
whose boundaries are -0.5 and 9.5.
Two other terms used in connection with frequency distributions are
"class mark" and "class interval." A class mark is simply the mid-point' of a
class, and it is obtained by averaging the class limits (or boundaries), that is,
by dividing their sum by 2. Thus, the class marks of the distribution of the
scores are 14.5, 24.5, 34.5, ..., and 94.5, while those of the size-of-sales
distribution, table 2.3 on page 28 are 2.495, 7.495, 12.495, and so on. A
class interval is merely the length of a class (the range of values it can
contain), and it is given by the difference between its class boundaries. If the
classes of a distribution are all equal in length, their common class interval
(which we refer to as the class interval of the distribution) is also given by
the difference between any two successive class marks. Since 19.5 - 9.5 =
10, 29.5 - 19.5 = 10, ..., and 99.5 - 89.5 = 10, the distribution of the scores
has class intervals of length 10, and we say that this is the class interval of
the distribution. Note that the class interval is not given by the difference
between the respective upper and lower class limits, which in our example
would equal 9, and not 10.

Suppose now that in connection with the scores of the 150 applicants for
secretarial positions, it is of interest to know how many fell below various
levels. To provide this information, we have only to convert the distribution,
table 2.6on page 30 into what is called a cumulative frequency distribution
or simply a cumulative distribution. Successively adding the frequencies in
the table, we thus obtain the following "less than" cumulative distribution,
shown in table 2.7

31
 Table 2.7

Cumulative
Scores
Frequencies

Less than 10 0

Less than 20 1
Less than 30 7
Less than 40 16
Leas than 50 47
Less than 60 89
Less than 70 121
Less than 80 138
Less than 90 148
Less than 100 150

Note that in this table we could just as well have written "9 or less" instead
of "less than 10," "19 or less" instead of "less than 20," ..., and "99 or less"
instead of "less than 100."
If we successively add the frequencies starting at the other end of the
distribution, we similarly get a cumulative "or more" distribution (or a
cumulative "more than" distribution), which shows how many of the scores
are "10 or more" (or "more than 9"), how many are "20 or more" (or "more
than 19"), and so on.
Sometimes it is preferable to show what percentage of the items falls into
each class, or what percentage of the items falls above or below various
values. To convert a frequency distribution (or a cumulative distribution)
into a corresponding percentage distribution, we have only to divide each
class frequency (or each cumulative frequency) by the total number of items
grouped and multiply by 100. For instance, for the size-of-farm distribution
on page 25, it may be more informative to indicate that
(183/ 3,156)100 = 5 .8 per cent of the farms are under 10 acres, that
(637/3,156)100 =20.2 per cent of the farms are from 10 to 49 acres, and so
on. Generally speaking, percentage distributions are useful, especially when
we want to compare two or more sets of data.

32
For instance, it may well be more informative to say that the percentages of
farms under 10 acres in two counties are, respectively, 5 per cent and 6 per

cent, than to report that in one county 16 of 321 farms and in the other
county 43 of 717 farms are under 10 acres.
So far we have discussed only numerical distributions, but the general
problem of constructing categorical (or qualitative) distributions is very
much the same. Again we must decide how many classes (categories) to use
and what kind of items each category is to contain, making sure that all of
the items are accommodated and that there are no ambiguities. Since the
categories must often be selected before any data are actually obtained,
sound practice is to include a category labeled "others" or "miscellaneous."
When dealing with categorical distributions we do not have to worry about
such mathematical details as class limits, class boundaries, class marks, etc..;
on the other hand, we now have a more serious problem with ambiguities,
and we must be careful and explicit in defining what each category is to
contain. For instance, if we tried to classify items sold at a supermarket into
"meats," "frozen foods," "baked goods," and so oil, it would be difficult to
decide where to put, for example, frozen beef pies. Similarly, if we wanted
to classify occupations, it would be difficult to decide where to put a farm
manager, if our table contained (without qualification) the two categories
"farmers" and "managers." For this reason, it is often advisable to use
standard categories developed by the Bureau of the Census and other
government agencies.

2.3 Graphical Presentations

When frequency distributions are constructed primarily to condense large

sets of data and display them in an "easy to digest." form, it is usually
advisable to present them graphically, that is, in a form that appeals to the
human power of visualization.

Some of the common graphical presentations of statistical data are :

1)Histogram, 2)Bar chart, 3)Polygon, 4)Curve, 5)Pictogram,

5)Cumulative(less) distribution and 6)Cumulative(more) distribution.

The most common among all graphical presentations of statistical data is the
histogram, an example of which is shown in figure 2.1. A histogram is
constructed by representing

33
measurement or observations that are grouped (in figure 2.1 the scores)
on a horizontal scale, the class frequencies on a vertical scale, and
drawing rectangles whose bases equal the class interval and whose
heights are determined by the corresponding class frequencies. The
markings on the horizontal scale can be the class limits as in figure 2.1,
the class boundaries, the class marks, or arbitrary key values.

34
For easy readability it is generally preferable to indicate the class limits,
although the bases of the rectangles actually go from one class boundary to
the next. Similar to histograms are bar charts, like the one of figure 2.2,
where the lengths of the bars are proportional to the class frequencies, but
there is no pretense of having a continuous (horizontal) scale.
There are several points that must be watched in the construction of
histograms. First, it must be remembered that this kind of figure cannot
be used for distributions with open classes. Second, it should be noted

35
that the picture presented by a histogram can be very misleading if a
distribution hits unequal classes and no suitable adjustments are made. To
illustrate this point, let us regroup the distribution of the 150 scores by
combining all those from 60 to 79 into one class. Thus, the new distribution
is given by the following table

Table 2.8

Scores Frequency
10-19 1
20-29 6
30-39 9
40-49 31
50-59 42
60-79 49
80-89 10
90-99 2

and its histogram (with the class frequencies represented by the heights of
the rectangles) is shown in figure 2.3. This figure gives the impression that

just about half the scores fall on the interval from 60 to 79, where as

36
the correct proportion is close to 1/3,49/150 to be exact. This error is due
to the fact that when we compare the size of rectangles, triangles, and
other plane figures, we instinctively compare their areas and not their
sides. In order to correct for this, we simply draw the rectangles of the
histogram so that the class frequencies are represented by their areas, and
not by their heights. In figure 2.4 we accomplished this by reducing the
height of the rectangle representing the class 60-79 to half of what it was
in figure 2.3.

37
 The practice of representing class frequencies by means of areas is
especially important if histograms are to be approximated with smooth
curves. For instance, if we wanted to approximate the histogram of
figure 2.1 with a smooth curve, we could say that the number of scores
exceeding 69 is given by the shaded area of figure 2.5. Clearly, this area
is approximately equal to the sum of the areas of the corresponding three
rectangles.
An alternate, though less widely used, form of graphical presentation
is the frequency polygon (see figure 2.6). Here the class frequencies are
plotted at the class marks and the successive points are connected by
means of straight lines. Note that we added classes with zero frequencies

38
39
40
at both ends of the distribution in order to "tie down" the graph to the
horizontal scale.
If we apply the same technique to a cumulative distribution, we obtain
what is called an ogive. Note, however, that now the cumulative frequencies

41
are not plotted at the class marks-it stands to reason that the cumulative
frequency corresponding, say, to "less than 20" in our example should be
plotted at 20, or preferably at the class boundary of 19.5, since "less than 20"
actually includes everything up to 19.5. figure 2.7 shows an ogive
representing the cumulative "less than" distribution of the scores of the 150
applicants.

Although the visual appeal of histograms, frequency polygons, and ogives

exceeds that of frequency tables, there are ways in which distributions can
be presented even more dramatically and probably also more effectively. We
are referring here to the various kinds of pictograms ( pictorial pre-
sentations )(see, for example, figure 2.8 ) with which the reader must surely
be familiar through newspapers, magazines, advertising, and other sources.
The number of ways in which distributions (and other statistical data) can be
displayed pictorially is almost unlimited, depending only on the imagination
and artistic talent of the individual preparing the presentations.

42
Tutorial 2
1. Decide for each of the following quantities whether it can be deter-
mined on the basis of the distribution of the 150 scores on page 30; if
possible, give a numerical answer:
(a) The number of scores which were at least 50.
(b) The number of scores which were greater than 50.
(c) The number of scores which were 80 or less.
(d) The number of scores which were less than 80.
(e) The number of scores which were more than 90.
(f) The number of scores which were greater than 39 but at most 69.

2. If the amounts paid for the repairs of cars damaged in accidents are
grouped into a frequency table with the classes $0.00-$99.99,
$100.00$199.99, $200.00-$299.99, $300.00-$399.99, $400.00-
$499.99, and $500.00 or more, decide for each of the following
quantities whether it can be determined on the basis of this
distribution:
(a) How many of the amounts were less than $200.00.
(b) How many of the amounts were at least $200.00.
(c) How many of the amounts were more than $200.00.
(d) How many of the amounts were $200.00 or more.
3. The following is the distribution of the weekly earnings of 1, 216
secretaries in the Phoenix, Arizona, metropolitan area in March, 1969:

Number of
Weekly Earnings Secretaries
Under $80 21
$80- $99 296
$100-$119 494
$120-$139 247
$140-$159 119
$160 and over 39

Decide for each of the following quantities whether it can be determined

on the basis of this distribution; if possible give a numerical answer:

43
 (a) The number of secretaries with weekly earnings of at least $120.
(b) The number of secretaries with weekly earnings of more than $120.
(c) The number of secretaries with weekly earnings of more than $180.
(d) The number of secretaries with weekly earnings of less than $100.
(e) The number of secretaries with weekly earnings of at most $100.
(f) The number of secretaries with weekly earnings of at least $60.
4. The number of students absent from school each day are grouped into a
distribution having the classes 3-10, 11-18, 19-26, 27-34, and 35-42. Find
(a) the limits of each class, (b) the class boundaries, and (c) the class
marks.
5. The following is the distribution of the actual weight (in ounces) of 50
"one-pound" bags of coffee, which a grocery clerk filled from bulk stock:
Weight Number of bags
15.5 – 15.6 3
15.7 – 15.8 9
15.9 – 16.0 17
16.1 – 16.2 14
16.3 – 16.4 6
16.5 – 16.6 1
Find (a) the limits of each class, (b) the class marks, and (c) the class
boundaries.
6. The weights of certain laboratory animals, given to the nearest tenth of
an ounce, are grouped into a table having the class boundaries 11.45,
13.45, 15.45, 17.45, and 19.45 ounces. What are the limits of the four
classes of this distribution?
7. The class marks of a distribution of temperature readings, given to the
nearest degree Fahrenheit, are 113, 128, 143, 158, and 173. Find the
class boundaries of this distribution, and also the class limits.
8. Class limits and class boundaries have to be interpreted very carefully
when we are dealing with ages, for the age group from 5 through 9, for
example, includes all those who have passed their fifth birthday but not

44
yet reached their tenth. Taking this into account, what are the boundaries
and the class marks of the following age groups: 10-19, 20-29, 30-39, and
40-49.
9. A study of air pollution in a city yielded the following daily readings of
the concentration of sulfur dioxide (in parts per million):

.04 .11 .05 .01 .15 .12 .19 .06 .13 .03
.18 .01 .08 .11 .08 .14 .02 .14 .08 .10
.17 .09 .14 .07 .13 .11 .09 .05 .15 .08
.06 .05 .12 .10 .27 .12 .16 .10 .09 .I5
.07 .10 .17 .13 .20 .18 .11 .17 .14 .04
.22 .11 .09 .02 .12 .16 .15 .12 .13 .07
.05 .14 .04 .16 .19 .10 .06 .03 .16 .13
.18 .13 .11 .09 .06 .23 .11 .12 .07 .11

(a) Group these data into a table having the classes .00-.04, .05--.09,
.10-.14, .15-.19, .20-.24, and .25-.29.
(b) Convert the distribution obtained in (a) into a cumulative "less
than" distribution.
(c) Construct a histogram of the distribution obtained in (a).
(d) Draw an ogive of the cumulative "less than" distribution obtained
in (b) and use it to read off (roughly) the value below which we
should find the lowest, half of the data
10. The following are the number of customers a restaurant served for
lunch on 120 week days

50 64 55 51 60 41 71 53 63 64 46 59
66 45 61 57 65 62 58 65 55 61 50 55
53 57 58 66 53 56 64 46 59 49 64 60
58 64 42 47 59 62 56 63 61 68 57 51
61 51 60 59 67 52 52 58 64 43 60 62
48 62 56 63 55 73 60 69 53 66 54 52
56 59 65 60 61 59 63 56 62 56 62 57
57 52 63 48 58 64 59 43 67 52 58 47
63 53 54 67 57 61 65 78 60 66 63 58
60 55 61 59 74 62 49 63 65 55 61 54

45
(a) Group these data into a table having the classes 40-44, 45-49,
50-54, 55-59, 60-64, 65-69 ,70-74 and 75- 79.
(b) Convert the distribution obtained in (a) into a cumulative "less than"
distribution.
(c) Construct a histogram of the distribution obtained in (a).
(d) Draw an ogive of the cumulative "less than" distribution obtained in (b)
and use it to read off (roughly) the value below which we should find the
lowest, half of the data .

46
 3
Measures of Location

3.1 Introduction
3.2 The arithmetic mean
3.3 The mean of a distribution
3.4 The coding method
3.5 The mode
3.6 The median
3.7 Other numerical measures
3.7.1 Geometric mean
3.7.2 Quartiles and Percentiles

Tutorials: 3.1 & 3.2

47
3.1 Introduction
Descriptions of statistical data can be quite brief or quite elaborate,
depending partly on the nature of the data themselves, and partly on the
purpose for which they are to be used. Sometimes, we even describe the
same set of data in several different ways. To draw an analogy, a large motel
might describe itself to the public as having luxurious facilities, a heated
swimming pool, and TV in every room; on the other hand, it might describe
itself to the fire department by giving the floor space of each unit, the
number of sprinklers, and the number of employees. Both of these
descriptions may serve the purpose for which they are designed, but they
would hardly satisfy the State Corporation Commission in passing on the
owner's application for issuing stock. This would require detailed
information on the management of the motel, various kinds of financial
statements, and so on.
Whether we describe things statistically or whether we simply describe
them verbally, it is always desirable to say neither too little nor too much.
Thus, it may sometimes be satisfactory to present data simply as they are
and let them "speak for themselves"; in other instances it may be satisfactory
to group, classify, and present them using the methods of Chapter 2.
However, most of the time it is necessary to summarize them further by
means of one or more well-chosen descriptions. In this chapter and in
chapter 4 we shall concentrate mainly on two kinds of descriptions ,called
measures of location, and measures of variation.

The measures of location we shall study in this chapter are also referred
to at times as "measures of central tendencies," "measures of central values,"
and "measures of position." Except for some of the measures discussed in
Section 3.4, they may also be referred to crudely as "averages" in the sense
that they provide numbers that are indicative of the "center," "middle," or the
"most typical" of a set of data.
When we said that the choice of a statistical description depends partly
on the nature of the data themselves, we were referring among other things
to the following distinction: if a set of data consists of all conceivably possi-
ble (or hypothetically possible) observations of a certain phenomenon, We
refer to it as a population; if it contains only part of these observations, we
refer to it as a sample. The qualification "hypothetically possible" was
added to take care of such clearly hypothetical situations where, say, twelve
flips of a coin are looked upon as a sample from the population of all

48
possible flips of the coin, or where we shall want to look upon the weights of
eight 30-day-old calves as a sample of the weights of all (past, present, and
future) 30-day-old calves. In fact, we often look upon the results obtained in
an experiment as a sample of what we might obtain if the experiment were
repeated over and over again.
In this chapter and the next we shall limit ourselves to methods of
description without making generalizations, but it is important even here to
distinguish between samples and populations. As we have said before, the
kind of description we may want to use will depend on what we intend to do
later on, whether we merely want to present facts about populations or
whether we want to generalize from samples. We shall, thus, begin in this
chapter with the practice of using different symbols depending on whether
we are describing samples or populations; in Chapter 4 we shall carry this
distinction one step further by even using different formulas.

3.2 The Mean

There are many problems in which we have to represent data by means of
a single number which, in its way, is descriptive of the entire set. The most
popular measure used for this purpose is what the layman calls an "average"
and what, in statistics, is called an arithmetic mean., or simply a mean. We
gave the word "average," in quotes because it generally has a loose
connotation and different meanings-for example, when we speak of a batting
average, an average housewife, a person with average taste, and so on.

Definition 3.1
Arithmetic mean
The arithmetic mean of a set of n numbers is defined simply as their sum
divided by n.

Example 3.1
Given that the total attendance at major league baseball games in the years
1965, 1966, 1967, and 1968 was, respectively, 22.4, 25.2, 23.8, and 23.0
million, we find that the mean, namely, the "average" annual attendance for
these foul: years was (22.4+25.2+23.8+23.0)/4= 23.6 million

In order to develop a simple formula for the mean that is applicable to any
set of data, it will be necessary to represent the figures (measurements or
observations) to which the formula is to be applied with some general

49
symbols such as x, y, or z. In the above example, we could have represented

the annual attendance figures with the letter x and referred to the four values

as x1 (x sub-one), x2 (x sub-two), x3, and x4. More generally, if we have n

measurements which we designate x1, x2, x3, ..., and xn,, we can write

Mean x =(x1+x2+…..+xn)/n

This formula is perfectly general and it will take care of any set of data, but
it is still somewhat cumbersome. To make it more compact, we introduce the
symbol ∑ (capital sigma, the Greek letter for S), which is simply a
mathematical shorthand notation indicating the process of summation or
addition. If we write ∑x, this represents the "sum of the x's ," and we have

x‾=∑ xi/n

Using the sigma notation in this form, the number of terms to be added is not
stated explicitly; it is tacitly understood, however, to refer to all the X's with
which we happen to be concerned. . For a further discussion of the use of
subscripts and the ∑ notation, we shall finish simplifying our notation by
assigning a special symbol to the mean itself. If we look upon the x's as a
sample, we write their mean as x‾(x-bar); if we look upon them as a
population, we write their mean as μ .If we refer to sample data as y's or z's,
we correspondingly write their means as y or z. To further emphasize the
distinction between samples and populations, we denote the number of
values in a sample, the sample size, with the letter n and the number of
values in a population, the population size, with the letter N. We thus have
the formulas x =∑xi/n , µ=∑xi/N ,

x=∑xi/n , x=∑xi/n , µ=∑xi/N ,

50
depending on whether we are dealing with a sample or a population. In order
to distinguish between descriptions of samples and descriptions of
populations, statisticians not only use different symbols, but they refer to the
first as statistics and the second as parameters. Hence, we say that x is a
statistic and that µ is a parameter.
The popularity of the mean as a measure describing the "middle" or
"center" of a set of data is not just accidental. Anytime we use a single
number to describe a set of data, there are certain desirable properties we
must keep in mind.
Thus, some of the noteworthy properties of the mean are:
(1) it is familiar to most persons, although they may not call it by this name,
(2) it always exists, that is, it can be calculated for any kind of numerical
data,
(3) it is always unique, or in other words, a set of data has one and only one
mean,
(4) it takes into account each individual item ,
(5) it lends itself to further statistical manipulation , (it is possible to
combine the means of several sets of data into an over-all mean without
having to refer back to the original raw data), and
(6) it is relatively reliable in the sense that it does not vary too much when
repeated samples are taken from one and the same population, at least not as
much as some other kinds of statistical descriptions.
This question of reliability is of fundamental importance when it comes
to problems of estimation, hypothesis testing, and making predictions, and
we shall have a good deal more to say about it later in this book.
Since the computation of means is quite easy, involving only addition and
one division, there is usually no need to look for short-cuts or simplifi-
cations. However, if the numbers are unwieldy, that is, if each number has
many digits, or if the sample (or population) size is very large, it may be
advantageous to group the data first and then compute the mean from the
resulting distribution. Another reason why we shall investigate the problem
of obtaining means from grouped data is that published data are very often
available only in the form of distributions.

51
 3.3 The mean of a distribution

To obtain a formula for the mean of a distribution, let us write the

successive class marks as x1,, x2, ..., xk (assuming that there are k classed
and the corresponding class frequencies as fl, f2, . . ., fk . The total that goes
into the numerator of the formula for the mean is thus obtained by adding f1
times the value x1, f1 times the value x2, . . ., and fk. times the value xk; in
other words, it is equal to x1f1 + x2f2 +. . . + xkfk. Using the ∑ notation, we
can now write the formula for the mean of a distribution .

Definition 3.2
Mean of a distribution

X=∑xi . fi /∑ fi,

where n equals f1 + f2 +. ..+ fk, the sum of the class frequencies, or ∑ fi.
(When dealing with a population instead of a sample, we have only to
substitute µ for x in this formula and N for n.)
Example 3.2
To illustrate the calculation of the mean of a distribution, let us refer again
to the distribution of the scores of the 150 applicants on chapter 2. Writing
the class marks in the second column, we get
Class Marks Frequencies Products
X f x. f
10-19 14.5 1 14.5
20-29 24.5 6 147.0
30-39 34.5 9 310.5
40-49 44.5 31 1379.5
50-59 54.5 42 2289.0
60-69 64.5 32 2064.0
70-79 74.5 17 1266.5
80-89 84.5 10 845.0
90-99 94.5 2 189.0
Total 150 8505.0

and it follows that the mean of the distribution is

52
 x‾=8505.0/150=56.7

It is of interest to note that the mean of the original raw is 8500/150=56.67

so that the difference between the two means is extremely small.

3.4 The coding method

The calculation of the mean of the distribution of the 150 scores was fairly
easy because the frequencies were all small. Even so, the calculations can be
simplified by performing a change of scale; that is, we replace the class
marks with numbers that are easier to handle. This is also referred to as
"coding," and in our example, we might replace the class marks of the
distribution of the scores with the consecutive integers -4, -3, -2, -1, 0, 1, 2,
3, and 4. Of course, when we do something like this, we also have to account
for it in the formula we use to calculate the mean. Referring to the new
(coded) class marks as u's, it can easily be shown that the formula for the
mean of a distribution becomes

Definition 3.3
Coding (shortcut) Mean
The coding (shortcut) mean is given by
_
X = x0+(∑ui . fi/n) . c

where xo is the class mark (in the original scale) to which we assign 0 in the
new scale, c is the class interval, n is the number of items grouped, and
∑ ui*fi is the sum of the products obtained by multiplying each of the coded
class marks by the corresponding frequency.

Example 3.3

Illustrating this short-cut technique by recalculating the mean of the

distribution of the scores of the 150 applicants, we obtain

53
 Class
ui fi ui.f
Marks
14.5 -4 1 -4
24.5 -3 6 -18
34.5 -2 9 -18
44.5 -1 31 -31
54.5 0 42 0
64.5 1 32 32
74.5 2 17 34
84.5 3 10 30
94.5 4 2 8
Total 150 33

X = 54.5+(33/150)10=56.7,
should be noted that this agrees with the result obtained earlier; the short-
cut formula does not entail any further approximation, and it should always
yield the same result as the formula of definition 3.2.
Unless one can use an automatic computer, the short-cut method will
generally save a good deal of time; about the only time that the short-cut
method will not provide appreciable savings in time and energy is when the
original class marks are already easy-to-use numbers. In order to reduce the
work to a minimum, it is generally advisable to put the zero of the u-scale
near the middle of the distribution, preferably at a class mark having one of
the highest frequencies.

Remark 3.1:
A fact worth noting is that this short-cut method cannot be used for
distributions with unequal classes, although there exists a modification
which makes it applicable also in that case. Neither the short-cut formula nor
the formula on definition 3.2 is applicable to distributions with. open classes;
the means of such distributions cannot be found without going back to the
raw data or making special assumptions about the values which fall into an
open class.

54
Tutorial 3.1

1. Suppose we are given the high temperature recorded each day of

the year 1972 in Atlanta, Georgia. Give one illustration each of a
situation where these data would be looked upon (a) as a
population, and (b) as a sample.
2. Suppose that the final election returns from a given county show
that the two candidates for a certain office received, respectively,
16,283 and 13,559 votes. What office might these candidates be
running for so that we can look upon these figures (a) as a
sample and (b) as a population?
3. The dean of a college has in his files a complete record of how
many A's, B's, C's, etc.., each instructor gave to his students
during the academic year 1971-72. Give one illustration each of
a problem (situation) in which the dean would look upon this
information (a) as a sample and (b) as a population.
4. The following are the speeds (in miles per hour) at which 25 cars
were timed on the San Bernardino Freeway in early-morning
traffic: 52, 56, 54, 78, 71, 66, 69, 60, 70, 53, 55, 62, 67, 60, 56,
72, 73, 61, 68, 59, 67, 66, 67, 73, and 65. Find the mean of these
speeds and comment on the (misleading?) argument that "on the
average cars do not exceed the speed limit of 65 miles per hour
on this freeway in early-morning traffic."
5. The following are the monthly water bills which a resident of
Scottsdale, Arizona, received in 1971: $10.26, $9.29, $11.24,
$12.22, $19.07, $21.03, $22.50, $26.41, $18.09, $23.96, $16.18,
and $15.60. Find the mean, namely, the average water bill this
person paid per month in 1971.
6. The following are the number of seconds which 16 insects survived
after being sprayed with a certain insecticide: 121, 115, 79, 52, 102,
126, 81, 65, 109, 119, 115, 121, 103, 75, 59, and 110.
(a) Calculate the mean of these 16 measurements.
(b) Recalculate the mean of these 16 measurements by first
subtracting 100 from each value, finding the mean of the
numbers thus obtained, and then adding 100 to the result.

55
 (What general simplification does this suggest for the
calculation of means?)
7. Twenty-four cans of a floor wax, randomly selected from a large
production lot, have the following net weights (in ounces): 12.0, 11.9,
12.2, 12.0, 11.9, 12.0, 12.0, 12.1, 11.8, 12.0, 12.0, 12.1, 11.9, 11.9, 12.2,
12.1, 12.0, 11.9, 11.9, 12.1, 12.0, 12.0,11.9, and 12.0.
(a) Calculate the mean of these 24 weights.
(b) Recalculate the mean of these 24 weights by first
subtracting 12.0 from each value, finding the mean of the
numbers thus obtained, and then adding 12.0 to the result.
(What general simplification does this suggest for the
calculation of means?)
8. The following are the number of twists that were required to
break 20 forged alloy bars: 37, 29, 34, 21, 54, 38, 30, 26, 48, 37,
24, 33, 39, 51, 44, 38, 35, 29, 46, and 31. Find the mean of these
values.
9. In business and economics, there are many problems in which
we are interested in index numbers, that is, in measures of the
changes that have taken place in the prices (quantities, or values)
of various commodities. In general, the year or period we want to
compare by means of an index number is called the given year or
given period, while the year or period relative to which the
comparison is made is called the base year or base period.
Furthermore, given-year prices are denoted pn base-year prices
are denoted po, and the ratio pn / po for a given commodity is
called the corresponding price relative. A very simple kind of
index number is given by the mean of the price relatives of the
commodities with which we are concerned, multiplied by 100 to
express the index as a percentage.
(a) Find the mean of the price relatives comparing the 1969
prices of the given processed fruits and vegetables (in cents) with
those of 1965:

56
 1965 1969
Fruit cocktail, No. 303 can 26.1 27.9
Pears , No.21 can 47.0 50.9
Frozen orange juice, 6 oz, 23.7 24.3
Pears , No.303 can 23.7 24.6
Tomatoes, No. 303 can 16.1 19.6
Frozen broccoli, 10 oz, 26.4 27.6
(b) Find the mean of the price relatives comparing the following 1967
prices with those of 1960, where all prices are in cents per pound:
1960 1967
Copper 32.4 38.6
Lead 11.9 14.0
Zinc 12.9 13.8
10. If we substitute q's for p's in the index number of Exercise 9. where
given-year quantities (produced, sold, or consumed) are denoted qn and
base-year quantities are denoted q0 , we obtain a corresponding quantity
index. Given the following data in thousands of short tons, find the mean
of the quantity relatives comparing the 1967 production figures with
those of 1960:
1960 1967
Copper 1080 954
Lead 247 317
Zinc 435 549
11. Another way of obtaining an index comparing given-year prices with a
corresponding set of base-year prices (see Exercise 9) is to average the
two sets of prices separately, take the ratio of the two means, and then
multiply by 100 to express the index as a percentage. Canceling
denominators, the formula for such a simple aggregative index is thus

57
3.5 The Mode
Definition 3.4
Mode
The mode, denoted by mo, of a set n observations x1,x2,…,xn
(or of a frequency table)is the value of X which occurs with
greatest frequency.
Example 3.4

Find the mode for the following observations: 3, 7, 3, 5, 2, 8

Solution:
mo=3
Example 3.5

Compute the mode for the following 12 numbers:

2 , 3, 2, 5, -1, -2, -1, 2, -1, 5, 0, 8
Solution:
mo1=-1 , mo2=2
In this case, we will say that the set of numbers is bimodal.

Example 3.6

Determine the mode of the six measurements: 2, 3, -1, 4, 0, 1

Solutions:
In the case, we will say that the set of numbers does not have a
mode.

3.6 The Median

To avoid the difficulty met on section 3.2, where we showed that an

extreme value (perhaps, a gross error) can have a pronounced effect on the
mean, we sometimes describe the "middle" or "center" of a set of data with
other kinds of statistical descriptions.

58
One of these is the median, which is defined simply as:

Definition 3.5
Median
The median of a set of data ,is the value of the middle item (or the mean of
the values of the two middle items) when the data are arranged in an
increasing or decreasing order of magnitude.

If we have an odd number of items, there is always a middle item whose

value is the median. For example, the median of the five numbers 5, 10, 2, 7,
and 8 is 7, as can easily be verified by first arranging these numbers
according to size, and the median of the nine numbers 3, 5, 6, 9, 9, 10, 10,
12, and 13 is 9. Note that there are two 9's in this last example and that we
do not refer to either of them as the median. The median is a number and not
an item, namely, the value of the middle item. Generally speaking, if there
are n items and n is odd, the median is the value of the (n + 1)/2 th. largest
item. Thus, the median of 25 numbers is given by the value of the
(25+1)/2=13th largest, the median of 49 numbers is given by the value
(49+1)/2=25 th largest, and the median of 81 numbers is given by the value
of the (81+1)/2=41 st largest.
If we have an even number of items, there is never a middle item, and the
median is defined as the mean of the values of the two middle items. For
instance, the median of the six numbers 3, 6, 8, 10, 13, and 15 (which
are already ordered according to size) is (8+10)/ 2 = 9. It is halfway between
the two middle values (here the 3rd and the 4th) and, if we interpret it
correctly, the formula (n+1)/2 again gives the position of the median.
For the six given numbers the median is, thus, the value of the (6+1)/2=3.5th
largest, and we interpret this as "halfway between the values of the third and
the fourth." Similarly, the median of 100 numbers is given by
the value of the (100+1)/2= 50.5 th largest item, or halfway between the
values of the 50th and tile 51st.
It is important to remember that the formula (n+1)/2 is not a formula for the
median, itself; it merely tells us the position of the median, namely, the
number of items we have to count until we reach the item whose value is the
median (or the two items whose values have to be averaged to obtain the
median).

59
 To find the median of a distribution with a total frequency of n, we
must, so to speak, count n/2 items starting at either end and use def.3.6.

Definition 3.6
Median

The median of grouped data denoted by m is defined by

m=(Lower boundary of the class containing the median)+(((n/2)-cum..
frequency before the median class)/ frequency of median class) . c

Example 3.7
To illustrate this procedure , let us refer again to the distribution of the 150
scores since n=150 in this example , we will have to count n/2= 75 items
from either end. Beginning at the bottom of the distribution, we find that 47
of the values are less than 50 while 89 are less than 60, so that the median
must fall into the class whose limits are 50-59. Since 47 of the values fall
below this class, we must count another 75 - 47 = 28 of its 42 values, and we
accomplish this by adding 42 of the class interval of 10 to 49.5, the lower
boundary of the class. (We add 42 of the class interval because we want to
count 28 of the 42 values contained in this class.) We thus get

m= 49.5 +(28/42) . 10=56.2

rounded to one decimal.
Generally speaking, if L is the lower boundary of the class containing the
median, f is its frequency, c the class interval, and j the number of items we
still lack when reaching L, then the median of the distribution is given by
the formula
m=L+(j/f) . c
It is possible, of course, to arrive at the median of a distribution by
starting at the other end and subtracting an appropriate fraction of the class
interval from the upper boundary U of the class into which the median must
fall. For the distribution of the scores we thus obtain
m=59.5+(14/42) . 10=56.2
and the two answers are identical, as they should be.

60
3.7 Other Numerical Measures
There are numerous representative measures other than the mean, median,
and mode. The geometric mean is commonly used in business problems to
describe the "average" of ratios. Although the geometric mean is not as
important as the three principal representative measures (mean, median, and
mode).

3.7.1 Geometric mean

Definition 3.7
Geometric mean
The geometric mean of a set of n measurements, x1, x2,...,xn denoted by
GM is defined by

1/n
GM= [x1 . x2 . x3 …xn] .

The geometric mean is the nth root of the product of all the measurements.
It is not as easily computed as the arithmetic mean-the computation is eased
somewhat by taking logarithms of both sides of above equation

log(GM) =1/n [log x1 + log x2 +…+ log xn] ,

10

it is apparent that the geometric mean can be computed by taking the antilog
of the arithmetic mean of the logs of the measurements.

Example 3.8
Determine the geometric mean of the three measurements:
x1 =2 , x2=4 and x 3=8
Solution:
1/3
GM = ((2)(4)(8)) =4.

Example 3.9
Find the arithmetic and geometric mean of 100, 100, 100 and 1000.

61
Solution:

X = (100+100+100+1000)/4 =325

Geometric mean: log (GM) = 1/4[1og100 + 1og100 + 1og100 + Iog1000]

=1/4[2+2+2+3]=1/(4*9)=2.25
2.25

GM = (10) = 177.8.

The above example illustrates the fact that the geometric mean is less
affected by one (or two) extremely large (or small) values than is the
arithmetic mean. Unfortunately, the geometric mean is neither easy to
compute nor amenable to use for statistical inferences. It is very useful,
however, in averaging ratios-a process that frequently arises in computing
cost-of-living or other index numbers.

3.7.2 Quartiles and Percentiles

The mean, median, and mode can be thought of as measures of location -
they attempt to locate the most representative value. Other measures of
location are quartiles and percentiles.
Definition 3.8
Lower, middle, and upper quartiles
The lower quartile (q1) of a set of n measurements x1 , x2, ..., xn which
have been ordered from the smallest to the largest is the value of x that
exceeds 1/4 , of the measurements and is less than the remaining 3/4 .
The middle quartile (q2) is the median.

The upper quartile (q3) is the value of x that exceeds 3/4, of the
measurements is less than the remaining 1/4.

Example 3.10
Find the lower, middle, and upper quartiles for the data set:

62
20, 34, 17, 18, 28, 33, 12, 1S, 17, 12, 41,
4S, 18, 19, 16, 21, 26, 14, 26, 13, 29
Solution:
Ordered from the smallest to the largest, the 21 measurements are:

12, 12, 13, 14, 15, 16, 17, 17, 18, 18, 19, 20, 21, 26, 26, 28, 29,33,34,41, 45
↑ ↑ ↑
q1 = 15.25 q2 = 19 q3 = 28.75
To determine the first quartile, one fourth of the measurements is 21 /4 = 5.2
and three-fourths is 15.75. We wish to find, therefore, the measurement in
the data set such that 5.25 of the measurements are below it and 15.75 are
above it. Of course, no such measurement exists, so to find q1, we must
interpolate between the values of the fifth and sixth measurements, 15 and
16. This results in q1= 15.25.

Definition 3.9
The Pth percentile
,
The Pth percentile of a set of n measurements x1, x2, . . . , xn denoted by P,
is the value of x such that P percent of the values are less than P and (100 -
P) percent of the values are greater than P.

Example 3.11
Find the 85th percentile for the data set in Example 3.6.
Solution:
Since 85 percent of n = 21 is 17.85, we are looking for the measurement
such that 17.85 of the measurements are below it and 3.15 are above. This
value lies between 29 and 33. By interpolation
P = 32.4 .

By looking at the difference among the quartiles, we can get a feel for the
variability of the data. One measure of variability using the quartiles is the
interquartile range defined by
q3-q1 .
The larger the interquartile range, the more spread out the set of measure-
ments will be.

63
Tutorial 3.2

1. In the Complaints Department of a large department store on a given day.

the lengths (in seconds) of the first 100 telephone calls were recorded
(rounded to the nearest second), and the following frequency distribution
was constructed:
Class Class limit Class boundary Frequency
1 0-59 -0.5 to 59.5 5
2 60-119 59.5 to 119.5 20
3 120-179 119.5 to 179.5 40
4 180-239 179.5 to 239.5 25
5 240-299 239 5 to 299.5 10

a. Construct a histogram from this frequency distribution.

b. Construct a polygon from this frequency distribution.
c. Compute the approximate mean length of the 100 telephone calls.

2. The following 25 measurements represent the number of business trips

taken annually by 25 claim adjusters of the Acme Insurance Company:
33, 17, 2, 10, 12, 15, 22, 18, 20, 24,8, 27, 8,
12,17, 15, 21, 38, 16, 18,10,12, 9, 5, 28

Construct a frequency distribution with 5 classes for this data. Give the
relative frequencies and construct a histogram from the frequency
distribution. Compute the mean, median, and mode .

3. Find the median ,mean and the mode for the data given in Ex9 ,and Ex10
of tutorial 2.

64
 4
Measures of Variation

4.1 Introduction
4.2 The range
4.3 The standard deviation & the variance
4.4 The standard deviation & the variance of grouped data
4.5 Measure of relative variation
4.6 Measure of skewness
Tutorial: 4

65
4.1 Introduction

While representative measures provide certain information about a

distribution, more is needed before a clear picture of the shape of the
distribution can be formulated. In figure 4.1 for example, both distributions
have the same mean value, but obviously differ in another respect-the
amount of dispersion or variability of the values. The concept of variability
is very important in statistics. For example, in production management, a
major concern is the variability of the quality of a product being produced or
the variability of a crucial measurement of a product such as a bearing
diameter. More important, in statistical inference, we shall use the concept
of variability to determine how good our inferences are. For the moment, it
is sufficient to recognize the need for measuring the variability of a set of
values to get a better idea of the shape of the distribution of the measure-
ments.
Figure 4.1 Two dissimilar distributions with identical means

66
4.2 The range

The first and simplest measure of variability is the range.

Definition 4.1
Range
The range of a set of measurements x1, x2,..., xn is the algebraic
difference between the largest and smallest values.

Example 4.1
Given the following 6 numbers, determine their range.
x1=5 , x2=0 , x3=6 , x4=2 , x5=-2 , x6=9
Solution:
The largest number is 9 while the smallest is - 2. Thus the
range is 9 -(- 2) = 11.

While the range is very easy to compute, it is not a very satisfactory

measure of variability, as figure 4.2 illustrates. The distribution on the left
clearly is less variable than the distribution on the right, yet the ranges for
the two distributions are identical (10 - 0 = 10).
The mean of these six measurements is:

67
Figure 4.2 Two distributions with equal ranges

4.3 The Standard Deviation and the Variance.

Since the variation of a set of numbers is small if they are bunched
closely about their mean and it is large if they are spread over considerable
distances away from their mean, it would seem reasonable to define
variation in terms of the distances (deviations) by which numbers depart,
from their mean. If we have a set of numbers x1, x2,...,and xn whose mean is
x‾, we can write the amounts by which they differ from their mean as x‾1 -
x, x‾2-x,..., and x‾n-x. These quantities are called the deviations from the
mean and it suggests itself that we might use their average, namely, their
mean, as a measure of the variation of the numbers. This would not be a
bad idea, if it were not for the fact that we would always get 0 for an
answer, no matter how widely dispersed the data might be. As the quantity
∑ (xi – x¯ ) is always equal to zero-some of the deviations are positive,
some are negative, but they "average out," that is, their sum as well as their
mean are always equal to zero.
Since we are really interested in the magnitude of tile deviations and not
in their signs, we might simply "ignore" the signs and, thus, define a
measure of variation in terms of the absolute values of the deviations from
the mean. Indeed, if we added the values of the deviations from the mean as

68
if they were all positive and divided by n., we would obtain a measure of
variation called the mean deviation (see Exercises 5 and 6 on tutorial V).
Unfortunately, this measure of variation has the drawback that, owing to the
absolute values, it is difficult to subject it to any sort of theoretical
treatment; for instance, it is difficult to study mathematically how in
problems of sampling, mean deviations are affected by chance. However,
there exists another way of eliminating the signs of the deviations from the
mean, which is preferable on theoretical grounds: The squares of the
deviations from the mean cannot be negative; in fact, they are positive
unless a value happens to coincide with the, mean, in which case ∑(xi-x)is
equal to zero.
Definition 4.2
Variance
The variance of n measurements x1,x2,…,xn is denoted by s2 , is given by

s2 =∑( xi - x‾ )2/n,
and this is how, traditionally, the variance has been defined. Expressing
literally what we have done here mathematically, it has also been called the
mean-square deviation.
Nowadays, it has become the custom among most statisticians and
research workers to make a slight modification in this definition, which
consists of dividing the sum of the squared deviations from the mean by
n- 1 instead of n. Following this practice, which will be explained later,
let us thus formally define the sample variance , as

2
s* =∑( xi - x ) 2 /(n – 1) .

Definition 4.3
Standard deviation
The standard deviation is denoted by s(s*) is given by

s= (∑( xi – x‾ ) 2 / n) ½ ,

or
s*= (∑( xi – x‾ ) 2 /( n-1)) ½ .

69
The formulas we have given so far in this section are meant to apply to
samples, but if we substitute µ for x‾ and N(or N-1) for n(or n-1), we obtain
analogous formulas for the standard deviation and the variance of a
population. It has become fairly general practice to write population standard
deviations as S when dividing by N and S* when dividing by N - 1;
symbolically,

Definition 4.4
Population Standard Deviation
The Population standard deviation of N observation is denoted by S(S*) is
given by

2 1/2
S = (∑( xi- μ ) /N) ,
or

2 1/2
S* = (∑( x i- μ) /(N-1)).

To explain why we divide by n- 1 instead of n and N - I instead of N in the

formulas for s and S, let us point out that if we wanted to use sample
variances to estimate the respective variances of the populations from which
the samples were obtained, division by n instead of n - 1 would give us
values which on the average are too small. We cannot prove the following at
the level of this note, but it is shown in most textbooks on mathematical
statistics that the values would be too small on the average by the factor
(n-1)/n .

For instance, for n = 5 the estimates would on the average be (5-1)/5 = 0.80
or 80 per cent of what they should be,and hence 20 per cent too small. To
compensate for this we divide by n - 1 instead of n in the formulas for the
sample standard deviation and the sample variance. As the statisticians say,
this makes the sample variance
2 2
s unbiased; that is, if we calculate s* for several samples taken
from the same population, the values we get should average S2, the
variance of the population. Note, however, that this modification is of no
significance unless n is small; generally, its effect is negligible when it is
large, say 100 or more. The same applies to the difference between S2 and
S*2, which is negligible unless the size of the population is very small, and
in actual practice this is usually not the case.

70
Example 4.2

To illustrate the calculation of a sample standard deviation, let us find s for

the following data on the number of burglaries reported in a town during the
first six weeks of 1972: 12, 18, 7, 11, 15, and 9. First calculating x, we get

x‾=(12+18+7+11+15)/6=12.
and then the remainder of the calculations are as shown in the following
table
x (x-x‾) (x- x‾) 2

12 0 0
18 6 36
7 -5 25
11 -1 1
15 3 9
9 -3 9
0 80
and
s*= (∑( xi – x‾ ) 2 /( n-1)) ½

.
=(80/(6-1)) 1/2 =(16) 1/2 =4

Thus, x‾ = 12 provides us with an estimate of the average number of

burglaries in this town per week, and the value of the standard deviation,
s* = 4, tells us something about the variability of the figures from week to
week. How such a value of s is to be interpreted will be discussed
and how it can be used to judge how close x‾ = 12 might be to μ, the true
average number of burglaries in this town per week, will be discussed. (Note
that in the above table we totaled, as a check, the x‾ - x column, and as we
have indicated , the sum of its values must always equal zero.)
The calculation of s was very easy in this example, and this was due
largely to the fact that the x's, their mean, and hence also the deviations from

71
the mean were all whole numbers. Had this not been the case, it might have
been profitable to use the following short-cut formula for s:
.
s*=((n(∑ x2) – (∑ x) 2)/n(n-1)) ½

This formula does not involve any approximations and it can be derived
from the other formula for s by using the rules for summations. The
advantage of this short-cut formula is that we do not have to go through the
process of actually finding the deviations from the mean; instead we
calculate ∑x, the sum of the x's , ∑ x2, the sum of their squares, and
substitute directly into the formula. Referring again to the burglary data, we
now have

x x2
12 144
18 324
7 49
11 121
15 225
9 81
72 944

s*=((n(∑ x2) – (∑ x) 2)/n(n-1)) ½ ,

= (6(944)-(72)2/6.5) ½ =4.

It appears that in this particular example the "short-cut" method is actually

more involved; this may be the case, but in actual practice, when we are and
dealing with realistically complex data, the short-cut formula usually
provides considerable simplifications.
Example 4.3
To demonstrate the advantages of the short-cut formula, let us determine
the sample variance of the numbers 12, 7, 9, 5, 4, 8, 17, 2, 11, 14, 13, and 9,
using first the formula definition 4.3 and then the short-cut formula. Without
using the short-cut formula we get

72
 x (x- x‾ ) ( x- x‾ ) 2
12 2.75 7.5625
7 -2.25 5.0625
9 -0.25 0.0625
5 -4.25 18.0625
4 -5.25 27.5625
8 -1.25 1.5625
17 7.75 60.0625
2 -7.75 52.5625
11 1.75 3.0625
14 4.75 22.5625
13 3.75 14.0625
9 -0.25 0.0625
111 0 212.2500

and
x‾ = (111/12)=9.25 ,
s*2 =(212.2500/11)=19.3,

and working with the short-cut formula we get

x X2

12 144
7 49
9 81
5 25
4 16
8 64
17 289
2 4
11 121
14 196
13 169
9 81
111 1239

73
 s*2=(12(1239)-(111) 2) /12.11
=19.3.

Here the short-cut formula provided considerable simplifications.

A further simplification in the calculation of s* or s*2 consists of adding
all arbitrary positive or negative number to each measurement. It is easy to
prove that this would have no effect on the final result, and had we used this
trick in the last example, we might, have subtracted 10 (added -10) from
each number, getting 2, -3, -1, -5, -6, -2, 7, -8, 1, 4, 3, and – 1 instead of the
original numbers. The sum of these numbers is -9, the sum of their squares is
219, and substitution into the formula for
s*2 yields
s*2 =(12(219)-(-9) 2)/12.11
=19.13.

which is exactly what we had before. Since the purpose of this trick is to
reduce the size of the numbers with which we have to work, it is usually
desirable to subtract, a number that is close to the mean. In our example the
mean was 9.25, and the calculations might have been even simpler if we had
subtracted 9 instead of 10. Although the short-cut formula was given for use
with samples, we have only to substitute N2 for n throughout to make the
formula applicable to the calculation of s* or s*.

4.4 Standard deviation and Variance for grouped data.

If we want to calculate the standard deviation of data which have already

been grouped, we are faced with the same problem as on the mean.
Proceeding as we did in connection with the mean, and assigning the value
of the class mark to each value falling into a given class.

Definition 4.5
Standard deviation for Grouped Data
The standard deviation of grouped data is

s* = ( ( ∑( xi – x ) 2 . fi/(n-1)) 1/2 ,

74
 n=∑fi,

and, if we substitute µ for x and N or N - 1 for n - 1, we obtain analogous

formulas for s. Note that in this formula the x's are now the class marks and
the f's are the corresponding class frequencies.
The above formula serves to define s for grouped data, but it is seldom
used in actual practice. Either we use a computing formula analogous to the
short-cut formula ,

s*=((n(∑(xi2 fi)-(∑xifi) 2)/n(n-1)) ½ ,

where the x's are the class marks and the f's the corresponding class fre-
quencies, or we use the same kind of coding as in the calculation of the
mean of grouped data.

Following of the mean, we obtain

s*=c((n(∑(ui2 fi)-(∑uifi) 2)/n(n-1)) /12

This is the coding formula for computing the standard deviation of
grouped data. Note that this formula can be used only when the class
intervals are all equal.
Although this short-cut formula may look fairly complicated, it makes the
calculation of s very easy. Instead of having to work with the actual class
marks arid the deviations from the mean, we have only to find the sum of
the products obtained by multiplying each u by the corresponding f, the
sum of the products obtained by multiplying the square of each u by the
corresponding f, and substitute into the formula.

Example 4.4
To illustrate the use of this short-cut formula for the calculation of s for
grouped data, let us refer again to the distribution of the scores of the150
applicants. Using the same u-scale, we get

75
 Class Marks ui2.fi
ui fi ui .fi
xi
14.5 -4 1 -4 16
24.5 -3 6 -18 54
34.5 -2 9 -18 36
44.5 -1 31 -31 31
54.5 0 42 0 0
64.5 1 32 32 32
74.5 2 17 34 68
84.5 3 10 30 90
94.5 4 2 8 32
33 359
and
s*=c((n(∑(ui2 fi)-(∑uifi) 2)/n(n-1)) /12 ,

=10((150(359)-(33) 2 )150.149) ½
=15.4 .

The variation of the scores of the 150 applicants is, thus, measured by a
standard deviation of 15.4, and we shall indicate below how such a figure
might be interpreted.

4.5 Measures of Relative Variation

The standard deviation of a set of measurements is often used as an
indication of their inherent precision. If we repeatedly measure the same
quantity, for example, a person's temperature, the mileage a person gets with
a certain gasoline, or the weight of a piece of rock brought down from the
moon, we would hardly expect always to get the same result. Consequently,
the amount of variation we do find in repeated measurements of the same
kind provides us with information about their precision. To give an example,
suppose that 5 measurements of the length of a certain object have a
standard deviation of 0.20 in. Although this information may be important, it
does not allow us to judge the relative precision of these measurements; for
this purpose we would also have to know something about the actual size of
the quantity we are trying to measure. Clearly, a standard deviation of 0.20
in. would indicate that the measurements are very precise if we measured the
span of a bridge; on the other hand, they would be far from precise if we
measured the diameter of a small ball bearing.
This illustrates the need for measures of relative variation, that is, measures
which express the magnitude of the variation relative to the size of whatever

76
is being measured. The most widely used measure of relative variation is the
coefficient of variation.

Definition 4.6
Coefficient of Variation
The coefficient of variation of grouped (or non-grouped data) is denoted by
CV ,is given by the formula

CV = (s/ x ) . 100.
This simply expresses the standard deviation of a set of data (or distribution)
as a percentage of its mean. When dealing with populations, we ,analogously
define the coefficient of variation as

CV = (S/μ) . 100

If in the above example the standard deviation s = 15.4 and x‾=56.7 ,then

… CV=(15.4/56.7) . 100. …

By using the coefficient of variation, it is also possible to compare the

dispersions of two or more sets of data that are given in different units of
measurement. Instead of having to compare, say, the variability of weights
in pounds, lengths in inches, ages in years, and prices in dollars, we can
instead compare the respective coefficients of variation-they are all
percentages.

4.6 Skewness Measure

As suggested earlier in this chapter, one possible measure of the skewness
of a distribution of a set of measurements is the difference between its mean
and its median. We will use a function of this difference as a measure of
skewness.
Definition 4.7
Skewness measure( Sk)
The skewness measure of a set of n measurements x1,x2,..., xn (or
grouped data)denoted by Sk, is defined as three times the difference
between the mean and the median, divided by the standard deviation:

Sk=3( x - m)/s .

77
 If the distribution is skewed right, the mean will be larger than the median
and Sk will be positive. If the distribution is skewed left, the mean will be
smaller than the median and Sk will be negative.
The effect of dividing by s in Sk is to produce a statistic which is not depen-
dent on the unit of measurement. The mean, median and standard deviation
are all measured in the same units for a given data set.

The skewness measure Sk can be used in two ways. First, the sign of Sk
indicates the direction of skewness: +, skewed right and -, skewed left.
Second, if Sk is larger in magnitude in one data set than in another, the first
data set distribution is more skewed than the other. That is, Sk can be used
as a relative measure of the degree of skewness among data sets.

Example 4.5
Compute Sk for the following set of 5 measurements:
x1=10 , x2=4 , x3=4 , x4=6 and x5=1.

Solution:

The mean, median and standard deviation for this data set are x = 5, m = 4
and s= 2.97. Therefore,
Sk=3 . (5-4)/2.97=1.01
Since Sk is positive, the distribution of the 5 measurements is skewed to the
right.

78
Tutorial 4

1. Find the range , s2, s ,CV &Sk of the data of Ex. 6,7 & 8 of tutorial 3.
2. The following is the distribution of the percentage of students belonging
to a certain minority group in 40 schools:

Percentage Frequenc
y
0- 4 14
5- 9 11
10-14 7
15-19 6
20-24 2

a. Calculate the mean.

b. Calculate the mode.
c. Calculate the median.
d. Calculate s2 for this distribution
(i) without coding;
(ii) with coding.
e .Calculate C.V.
f. Calculate Sk.

3.Consider the following frequency distribution

Class Freq.
0-2 10
3-5 6
6-8 3
9-11 1
Calculate :
a.s2 & s.
b.median.
c.mode.
d. CV.
e. Sk.

79
4.Suppose that the random variable x has the following table

Class Freq.
0-7 2
8-15 10
16-23 8
24-31 3
32-39 2

a. Calculate the mean.

b. Calculate the mode.
c. Calculate the median.
d. s2 & s .
e. C.V .
f. range.
j. Sk.

80
 5
Introduction to
Probability & Random
Variables

5.1 Introduction
5.2 The sample and event spaces
5.3 Computing probabilities from the sample space
5.4 Permutations, combinations, and other counting rules
5.5 Random variable
5.6 Probability mass function
5.7 Probability density function
Tutorials 5.1 & 5.2

81
5.1 Introduction
Probability plays an integral role in inferential statistics by building a
"bridge" between the population and the sample taken from it. Our initial
applications of probability in this connection will be to make deductions
about a sample from a known population .The use of probability is as
indicated in figure 5.1: probability reasons from the population to the
sample, while statistical inferences are drawn about the population from the
sample.

Figure 5.1 Role of probability in the statistical inference process

Example 5.1
As an example of the use of probability in this context, consider the
national election for the office of president of the United States. Let us sup-
pose that only two candidates are listed on the ballot for the presidency, the
Democratic candidate (A), and the Republican candidate (B). Further,
suppose it is known in the population of registered persons who will vote on
election day that 60 percent will vote for A and 40 percent will vote for B. If
we now randomly sample one person from this population, what is the
probability that he or she will vote for A? Since we know that 60 percent of
the persons will vote for A, the probability that the one sampled person will
vote for A is 0.60. Knowledge of the probabilities of the two possible out-
comes of the experiment (voting for A or voting for B) enables us to deduce
the probability of the outcome in our sample of one.
Indeed, by using this knowledge, we could deduce the probabilities of
zero, one, or two persons voting for A in a sample of two, and so on for

82
larger sample sizes. Thus, if the population is known in the sense that the
probabilities associated with the values in it are known, then this knowledge
can be used to deduce the probabilities of the outcomes in the sample.
To illustrate the use of probability in making inferences from a sample to the
population, suppose candidate A conjectures before the election that 60
percent of the people in the voting population will vote for him. In order to
check this conjecture, his campaign manager randomly samples ten indi-
viduals from the population and finds that all ten intend to vote for candidate
B. If the probability of a randomly chosen person voting for A is really 0.60,
it is extremely unlikely that ten randomly chosen persons would all vote for
candidate B. It is more likely that the true percentage who will vote for A is
something considerably less than 60 percent. Hence, knowledge of this
experiment (sample outcome) indicates to A that more resources (cam-
paigning, etc..) may have to be employed if he is to have a chance of
winning the election. Candidate A is interested, of course, in testing whether
this sample is indicative of the population characteristics (voting patterns),
whether more sample information should be obtained, or whether the elec-
tion is likely to go to B (in which case A would be wasting his time by
campaigning further).
In practical situations, probability is used as a vehicle in drawing inferences
about unknown population characteristics. Additionally, as we shall see
later, probability concepts can be used to give us an indication of how good
these inferences are.
In this chapter, we will assume the population is known and compute the
probability of the occurrence of various sample outcomes. In effect, we will
be selecting a probability model depicting the outcomes in the population. In
practical applications of statistics, we shall see that the selection of this
model is an integral part of the statistical inference process.

5.2 The Sample and Event Spaces

In the presidential election example discussed in the previous section,

we defined a population which consists of registered persons who will vote
on election day. Suppose we assign a"1" to an individual if he or she intends
to vote for candidate A, and "0" if he or she intends to vote for candidate B.
The population can then be thought of as a collection of ones and zeroes.
How are these ones and zeroes generated? Each person in the population
must be contacted and represented by a"1" or a "0." The process of con-

83
tacting each person to determine the outcome ("1" or "0") is called an
experiment.

Definition 5.1
Experiment
An experiment is a process which results in one and only one outcome
of a set of disjoint outcomes, where the outcomes cannot be predicted
with certainty.
In the voting example, there are only two possible outcomes, and they are
disjoint (non-overlapping): a zero and a one. With our previous assumption
that only two candidates are listed on the ballot, each experiment results in
one and only one of the two possible experimental outcomes. And we cannot
predict with certainty the outcome before the experiment is conducted.
Repeated trials of this experiment will generate the population of zeroes and
ones. Other examples of experiments are:
Example 5.2
A professor at a large university is selected and his salary is recorded.
Example 5.3
A unit of a product is selected from an assembly line and is analyzed to
determine whether it is defective.
Example 5.4
A light bulb is randomly selected from the day's production and its time to
failure measured.
By repeating an experiment many times, a population of outcomes can be
generated. For example, if we repeated the experiment in 3. until each and
every light bulb in the day's production run had been tested to failure, the
population of all times to failure of this set of light bulbs would have been
generated. In the process of doing this, it should also be noted that the entire
day's production of light bulbs (the population) would have been destroyed.
We can also think of the sample being generated by repeated trials of an
experiment. For example, if we wanted to sample ten light bulbs, we could
repeat the experiment ten times.
The outcomes of an experiment are called simple events. Simple events shall
be denoted by the capital letter E subscripted to associate E; with a particular
outcome (ith) of an experiment.

84
 Definition 5.2
Simple event
A simple event is the outcome of an experiment.
Example 5.5
Suppose in our presidential election example that we randomly sample two
persons in the population of voters. A possible set of simple events
associated with this experiment is:
Vote By

Event First person Second person

E1 A A
E2 A B
E3 B A
E4 B B

These outcomes of the experiment can also be indicated by using an

outcome tree as shown in figure 5.2. The first "branch" of the tree is associ-
ated with the first person who votes either for A or for B. The second set of
branches is associated with the second person who votes either for A or for
B also.

The outcome tree represents a logical way to list the simple events of an
experiment. It is very practical if the number of events is not too large.

Figure 5.2 Outcome tree for Example 5.5

85
Example 5.6
Suppose three persons, A, B, and C, are interviewing for a job. Two will
be hired. The experiment is the selection of two of the three interviewed
individuals for the job. The simple events can be listed using an outcome
tree as illustrated in figure 5.3.

Figure 5.3 Outcome tree for Example 5.6

Notice in Example 5.6 that the six simple events listed specify not only the
two individuals selected, but also the order in which they are selected. That
is, E, and E, both result in the first two individuals, A and B, being selected.
If the order in which the two individuals are selected is not important, then
we need not distinguish between E1 and E2, E3 and E5. and E5, and E6. In this
case, a simpler set of outcomes would be:
E1*: A and B are selected,
E2*: A and C are selected,
E3*: B and C are selected.
As suggested above, it is often possible to define the outcomes and the
experimental simple events differently in the same experiment. To gain an

86
understanding of how to define the simple outcomes of an experiment,
consider the following example.

Example 5.7
Assume that Herman is to toss a "fair coin" twice. He informs you that
there are three possible outcomes (simple events) of this experiment:
E1*: No heads (two tails),
E2*: One head (one tail),
E3*: Two heads (no tails).
Herman tells you that the probability of any one of the three simple events
occurring is 1/3.He then wishes to wager with you on the outcome of one
trial of the experiment, say E2-one head occurring in two tosses of the coin.
Before deciding to accept a wager, you construct an outcome tree of a
single trial of the experiment.
From the outcome tree, it is clear that we may define another set of simple
events for this experiment:
E1: (H,H),
E2: (H.T),
E3: (T,H),
E4: (T,T).
If the coin is "fair," then the probability of each of the outcomes E1, E2, E3
and E4 in figure 5.4 occurring is 1/4.
In terms of the original three outcomes, E*1, E*2 and E*3 it is clear that
each does not have a 1/3, probability of occurring - the proper probabilities
are:
P( E1*)= 1/4 , [E1* = E1 (H,H)],
P(E2*)= 2/4 , [E2* = E2 or E3 (H,T) or (T,H)],
P(E3*)= 1/4 , [E3* = E4 (T,T)].

87
 Figure 5.4 Outcome tree for a coin-tossing experiment
1 st coin toss 2 nd coin toss

Definition 5.3
Event
An event is a subset of outcomes of an experiment.
Notice that any simple event of an experiment is an event because it is a
single outcome of the experiment.

Definition 5.4
Null event
A null event is an event containing no simple events in an experiment. It
is denoted by : ф.
In Example 5.5, an example of a null event is "no persons vote for A
or for B." It is impossible for this event to happen, because there are
only two candidates, A and B, and the population consists of
individuals who will vote in the election. In this instance, the event
set is empty, for it does not contain any of the simple events in the
experiment.
The simple events of an experiment and events defined to be
collections of these simple events can be portrayed graphically by a
Venn diagram. The Venn diagram associated with the simple events
in Example 5.5 and the event D defined above is shown in
figure 5.5. Each simple event in a

88
Figure 5.5 A Venn diagram for the simple events in Example 5.5

Venn diagram is shown as a "point" with its corresponding subscripted letter

E. The collection of all simple events in an experiment is the complete set of
sample points in the Venn diagram and is called the sample space. The event
D is illustrated in the Venn diagram by enclosing the sample points
belonging to it; the resulting closed region is called the event space, D.

Definition 5.5
Sample point
A sample point is a simple event in an experiment.

Definition 5.6
Sample space
A sample space is the set of all possible outcomes of an
experiment.

Definition 5.7
Event space
An event space is the collection of sample points
corresponding to an event defined over the sample space.

89
Figure 5.6 Venn diagram for the experiment shown in Figure 5.3

In Example 5.6 suppose that we define the following events:

F: A is chosen first
G: A is chosen without regard to selection order
H: A and B are chosen without regard to selection order.

Assume that we define the simple events as in Figure 5.3:

E1: A and B E3: B and A E5: C and A

E2:A andC E4:B andC E6: C and B.

The Venn diagram showing the sample space for this experiment, the sample
points, and the event spaces F, G, and H is illustrated in figure 5.6.

5.3 Computing Probabilities from the Sample Space

In Example 5.5, where two persons are randomly selected from the voting
population, there are four possible outcomes of the experiment. In the cor-
responding sample space for this experiment, illustrated in Figure 5.5, we
defined the event D to be, "at least one of the two persons votes for
candidate A." What is the probability that the event D occurs in this
experiment? This question can be answered directly from the sample space
associated with the experiment if the probabilities of the simple events
occurring are known. Thus, to answer a probability question about an event
in an experiment, we first must assign probabilities to the simple events

90
associated with the experiment. We will denote by P(Ei) the probability
assigned to the simple event Ei. The assigned probabilities P(E1 ), P(E2), . . .
, P(Em), where there are M simple events in the experiment sample space,
must satisfy three probability axioms for experiments that have a finite
number of outcomes.
Axiom 1

0≤ P(Ei)≤ 1 for i = 1, 2, ..., m.

The first axiom requires that every simple event be assigned as its proba-
bility a non-negative number between 0 and I inclusive.
Axiom 2
m
∑ P(Ei) = 1 .
i i =1

The second axiom requires that the probabilities assigned to all the simple
events in the experiment must total one.

Axiom 3

P(E1 or E2 or E3 or ...) = P(E1) + P(E2) + P(E3) +…

The third axiom requires that the probability of one or more members of a
set of simple events occurring in an experiment is the sum of their respective
probabilities.

Example 5.8
Suppose in Example 5.6 we assign the following probabilities to the simple
events in that experiment:

Simple events Probability

E1 (A,A) P(E1) = 1/4
E2 (A,B) P(E2)= 1/4
E3 (B,A) P(E3) = 1/4
E4 (B,B) P(E4) = 1/4

91
These probability assignments satisfy the above three axioms-each
probability assigned is a positive number between 0 and I inclusive, the
probabilities total one, and the probability that any member of a collection of
the simple events will occur is the sum of the probabilities of the members in
the collection.

These axioms are intuitive; most of us have an understanding of them before

taking any formal training in probability. If an event is certain to happen, its
probability of occurrence is 1 and if an event is certain not to happen, its
probability of occurrence is 0.
How do we, in fact, formally define probability? We will consider one way
of defining probability.

Definition 5.8
Relative frequency definition of probability
If an event E is defined in an experiment, the experiment is repeated a very
large number of times, say N, and the event E is observed to occur in n of
these N experimental trials, then
… P(E) =n/ N . …

The ratio n/N represents the proportion of the time that event E occurs in
repeated experiments.

5.4 Permutations Combinations and other Counting Rules

Numerous counting rules can be used to count the number of points in

sample and event spaces. We shall consider four of the most important
counting rules. Each will be presented without proof and followed by
examples.

Rule5.1

… m.n rule …

Suppose that there are m distinguishable objects in one group and n dis-

92
 tinguishable objects in another group. If one element is selected from each
group, it is possible to form m . n pairs of objects.

Example 5.9
Herman has decided to purchase a new hi-fi system with the money he
saved by buying a compact car instead of a large sedan. His hi-fi system will
be composed of a receiver, a pair of speakers, a record changer, and a tape
deck. In the store where he will make the purchase, there are 10 different
kinds of receivers, 5 kinds of speakers, 4 kinds of changers, and 8 kinds of
tape decks. How many systems can Herman choose from?

Solution:
Since he must select one element from each of the four groups, he can
choose from (10)(5)(4)(8) = 1600 possible systems.

The next two counting rules apply to a different type of experiment as

indicated in the following example.
Example 5.10
Suppose three persons, A, B, and C, are competing for two job
positions. How many ways can two people be selected for employment
from the three?
Solution:
In this problem, it is easy to list the possible different outcomes of the
experiment. There are three: AB, AC, and BC. However, we may be
interested in the order of the selection as well as the content of the
resulting pairs. If this is the case, there are six possible outcomes: AB,
BA, AC, CA, BC, and CB.
In Example 5.10, there are two different ways to view the pairs formed
by selecting two persons out of three. Suppose there are n distinguishable
objects from which we are selecting a subset of size r. If we are
concerned about the number of groups of size r that can be formed from
the n where one group of size r is different from another if its content is
different, we want to determine the number of combinations of r things
selected from n. If, on the other hand, we want to compute the number of
groups of size r that can be formed from n where one group of size r is
different from another in terms of both its content and the order in which
the r things were drawn, then we want to determine the number of
permutations of r things drawn from n.
In Example 5.10, the number of combinations of two persons drawn

93
from three is 3, while the number of permutations of two persons drawn
from three is 6.
In conjunction with the rules for computing the number of permutations and
combinations, the complete definitions follow.
Definition 5.9
Permutations
An ordered arrangement of r distinguishable objects is called a
permutation. The number of permutations of r objects taken from n
distinguishable objects will be denoted by Prn.

Definition 5.10
Combinations
A set of r distinguishable objects is called a combination. The number
of combinations of r objects taken from n distinguishable objects will be
denoted by Crn.

Rule 5.2
Permutations
Prn=n!/(n-r)!

Rule 5.3
Combinations
...Crn =n!/r!(n-r)!=(1/r!) Prn>>

The symbol n! is called "n-factorial"; n! = n(n - 1)(n - 2) ...(2)(1 ). Thus,

4! = 4(3)(2)(1) = 24 and 6! =(6)(5)(4)(3)(2)(l) = 720; 1! = 1 and, by
definition, 0! = 1.
Example 5.11
A committee of three is to monitor the activities of the local club. The
committee is to be formed by selecting three people from a group of five
persons. How many different committees could be formed?
Solution:
Nothing is mentioned about the order or arrangement of the three selected
individuals. Thus, one committee will be different from another if it has one

94
or more different people in it. We are only concerned about the content of
each group, and therefore we want to determine the number of combinations
of three things taken from five.

C35 = 5!/ 3!(5 - 3)!=5.4.3.2.1/(3.2.1)(2.1)

=10,

thus, it is possible to develop 10 different committees of three people

;elected from five. You can check this result by listing all possible groups of
three drawn from five, where the five people are labeled A. B, C. D, and E.

Example 5.12
A club committee of three is to be formed by selecting three people from a
group of five. One of the selected people will be chosen a chairman of the
committee, another the secretary, and the third person will simply be a
"member" of the committee. How many different committees can be
formed?
Solution:
Suppose the three people (denoted by A, B, and C, respectively) have been
chosen from among the five. Once we have this combination of three people,
we must then assign them to the three positions: chairman, secretary, and
member.

We can view this process as ordering the three persons. That is, let the first
position in the ordering (or permutation) be the chairman, the second be the
secretary, and the third be the member. The possible permutations are: ABC,
ACB, BAC, BCA, CAB, and CBA. Each of these combinations is a different
committee, although each contains the same three people, A, B, and C. Thus,
we want to compute the number of permutations of r = 3 people chosen from
n = 5.
P35=5!/(5-3)!=60
In this case, there are 60 different committees which can be formed. Notice
that this is six times as many committees that could be formed in Example
5.11 because for each combination in Example 5.11, there are 6 permuta-
tions in Example 5.12.

95
Tutorial 5.1

l. Give the sample space of each of the following experiments in the form of
a Venn diagram. Be certain to define the simple events corresponding to
the sample points in the sample space.
a. A fair coin is tossed three times.
b. A coin and a die are tossed together.
c. Two fair dice are tossed, and the sum of the dots of the two faces
turning up is recorded.
d . A student receives his score on a multiple choice exam containing 20
questions.
e. A student receives his grade on an exam.
f. The number of telephone calls received at a switchboard during a live
minute interval is recorded.
g. A child is selected in a first grade class, and his or her weight (to the
nearest pound) is recorded.
2. A committee is composed of two men and two women. One member of
the committee is selected to serve as chairman and another is selected to
serve as secretary.
a. Define the simple events comprising the sample space of this
experiment. Identify which sample points in the sample space belong to
the following event spaces:
b. The younger man is selected as chairman: Event A.
c. A man is selected as chairman: Event B.
d. A woman is selected as secretary: Event C.
e. Events A and C occur: Event D.
f. Events B or C or both occur: Event E.
g. Show the live event spaces in a Venn diagram.
3. Two college job recruiting officers. Herman and Bill, come to the
University of Truth campus to fill positions in their organizations. Each
officer is attempting to fill three positions. Three students qualify for the
positions described. and each will be interviewed by the two officers. If a
sample point is defined as a specific number of students hired by Herman
or Bill, define the following events as specific collections of sample
points: Note: there are six jobs and three students three jobs will not be
filled. (Hint: The sample space is two-dimensional.)
a. The sample space S which consists of all outcomes defining the
number of students hired by each officer.

96
 b. Event A: Herman hires at least two students.

c. Event B: All three students are hired by Bill.

d. Event C: Exactly one student is hired by Bill.
e. Event D: Bill hires two students and so does Herman.
4. When one card is drawn from a well-shuffled deck of 52 playing cards
what are the probabilities of getting:
a) a black king.
b) an ace.
c) a red card.
d) a king or a queen.
e) a black card.
Remark: Four groups (with two colors) 1 - 10 + King, queen, jack.

5. A bowel contains 17 red balls, 10 black, .10 white balls and 20 blue balls.
If one of these is drawn at random, what the probabilities that it will
be: a) a red b) a white c) a blue d) red or white e) white or
blue f) neither white nor red.

6. Bowel I contains 12 red and 13 blue balls. Bowel II contains 10 red, 15

blue balls and 15 black balls. One ball is drawn from bowel I and
placed in bowel II, then one ball is drawn from bowel II. Find
probabilities that it will be:
a) a red ball b) a blue ball
c) a black ball.

97
5.5 Random Variable

In a typical population, it is usually possible to identify more than one

characteristic of the units comprising it. For example, suppose the
population is the collection of all full-time students registered at your
university or college during the present academic term. In this instance, it is
possible to identify numerous characteristics of the population unit-earned
income, height, weight, sex, hair color, number of parking tickets
accumulated during the term, grade-point average and so on. In a statistical
study of the units in this population, we may be interested in just one
characteristic (e.g…, gradepoint average) or in a collection of such
characteristics (e.g…, sex and grade point average, or earned income and
grade point average).
In Chapter 1, we referred to a population characteristic as a variable if the
characteristic can assume one or more values in the population. We must
now more specifically define a "variable" when we use the word to mean a
measure of a population characteristic.

Definition 5.11
Random variable
A random variable is a numerically valued function whose value is
determined by a random experiment.

A more mathematically rigorous definition is:

A random variable is a numerically valued function defined over a sample
space.
Example 5.13
Suppose we consider the population of full-time students registered at
your university or college. If we are interested in the population character-
istic, "grade point average," and we select one student at random from this
population, then the characteristic may be viewed as a random variable; its
value is numeric and arises from a random experiment, and it is a function
because it defines a correspondence between members of one set (the stu-
dent population) and members of another set (the set of all possible grade
point averages, from 0.00 to 4.00). For each student, the random variable
defines one and only one grade point average, while more than one student

98
may have the same grade point average.

To better appreciate the concept of a random variable being a function. let us

consider an example.

Example 5.14
Suppose the random experiment is tossing a coin twice, and we define the
random variable, X = number of heads in the two tosses. figure 5.1
illustrates the correspondence between members of the experimental
outcomes and possible values of the random variable.

Figure 5.7 Random variable X: number of heads in two tosses of a coin

Notice that the random variable X is a function. To each member in the first
set (the simple events in the experimental sample space) there corresponds
one and only one member in the second set (the values of the random
variable). But each value of the random variable may correspond to one or
more simple events. Notice that we have written the random variable in
functional notation in figure 5.7 to emphasize its meaning.

Example 5.15
A production lot of 100 transistor radios contains 10 defectives. A retailer
decides to select two of the radios at random, and by extensive testing, to
determine whether they are defective. If neither radio is defective, he will
accept the lot. Define the random variable, X = number of defective radios

99
 (X = 0, 1, or 2). Determine the probabilities that X assumes each of its
three possible values.

Solution:
In the experiment of selecting two radios at random, define the events:
A1: First radio is defective
A2: Second radio is defective
A3: First radio is not defective
A4: Second radio is not defective
The four simple events of the experiment are given in Table 5.1. The prob-
ability of each simple event occurring is determined by using the multiplica-
tive law For example,
P(A1 ∩ A2) = P(A1)P(A2/A1 )= (10/100)(9/99) = 0.0091
Notice that the two events are not independent; the outcome of the first
selection affects the chance of the second radio being defective. Notice that
the probabilities in Table 5.1 sum to one-we have specified the four mutually
exclusive and collectively exhaustive simple events of the experiment.
Table 5.1 Outcomes and probabilities for Example 5.13
Simple events Probability Values of X
E1:A1∩A2 (10/100)(9/99)= 0.0091 2
E2:A1∩A4 (10/100)(90/99)= 0.0909 1
E3:A3∩A2 (90/100)(10/99)=0.0909 1
E4:A3∩A4 (90/100)(89/99)= 0.8091 0

Since the simple events are mutually exclusive and collectively exhaustive,
P(X = 2) = 0.009 1, P(X = 1) = 0.0909 + 0.0909 = 0. 1818, and P(X = 0) =
0.8091. The values of the random variable X and their probabilities of
occurrence are given in Table 5.2. This Table represents the probability
distribution of the random variable X-a list of each value and its probability
of occurrence.

100
 Table 5.2
Probability
distribution of X
X P(x)
0 0.8091
1 0.1818
2 0.0091

From the probability distribution the probability that the retailer will accept
the lot is 0.8091.
Table 5.3 Examples of discrete random variables

Definition 5.12
Discrete random variable
A random variable is discrete if its set of values is finite or countably
infinite in number.
A continuous random variable assumes values which occur on an interval or
intersection of intervals on the real line. The number of values that a
continuous random variable may assume is infinite. Examples of continuous
random variables include height, weight, and the diameter of ball bearings.
Although each of these variables is bounded (for example, the weight of
individuals is bounded between zero lbs. and, say 500 lbs.), the variable can
assume any of an infinite number of values between these bounds. Other
examples of continuous random variables are given in Table 5.4.

101
Table 5.4 Examples of continuous random variables
Definition of X Range of values of X
1. Diameter of 1/2 bolts produced in a machine shop X≥ 0
2. Weight of a student in your class X≥ 0
3. Amount of rainfall in inches recorded at a weather X≥0
station on a given day
4. Weight lost by a person weighing 300 lbs. on a X≤ 300
diet designed for weight loss (A negative value of X
indicates a weight gain.)

Definition 5.13
Continuous random variable
A random variable is continuous if it may assume all real number values in
an interval.
5.6 Probablity Mass Function

The probability distribution can be described by a function P(x), call a

probability mass function, which assigns probabilities to the values of
discrete random variable.
Definition 5.14
Probability mass function (finite case)
Let the random variable X assume a finite number of values, r in total, and
denote these values by x1, x2, ..., xr.
Let P(xi) be the probability that the random variable X assumes the value xi.
A probability mass function is a function which assigns probabilities to the
values of a discrete random variable such that the following two conditions
on the function P(x) are satisfied:

…1. 0 ≤ P(xi)≤1 , i = 1,2, . . ., r,…

r
2. ∑ P(xi) = 1.
i=1

102
 Figure 5.8 Three ways of presenting the distribution of the
discrete random variable in Example 5.15

Example 5.16
A committee of two persons is to be formed from four persons, three of
whom are female. The committee is formed anew at the end of each working
week. The members of the committee have the duty of arriving at the office
30 minutes early to prepare the morning coffee, turn on lights, plug in
machines and so on. In any given week, let X be the number of women
serving on the present committee. Form the probability distribution of the
random variable X.
Figure 5.9 Probability mass function for the random variable in Example
5.16

The average of the values of a random variable is called the expected value
of the random variable and is given for a discrete random variable:

103
 Definition 5.15
Expected value of a random variable (discrete case)
Let X be a discrete random variable with a finite number of values denoted
by x1, x2; ..., xn. The mean or expected value of the random variable,
denoted by E(X), is given by

r
E(X) = ∑ xi . p(xi) .
i=1

Example 5.17

Expected value of the random variable in Example 5.14 is given by the

following table :

Table 5.7

xi P(x) xP(x)
0 0.8091 0.0000
1 0.1818 0.1818
2 0.0091 0.0182
Total 0.2000

Definition 5.16
Variance of a random variable (discrete case)
Let X be a discrete random variable with a finite number of values denoted
b) x1, x2, ..., xn. The variance of the random variable, denoted by V(X), is
given by

r
V(X)=∑ (xi- E(x)) 2 p(xi) , or
i=1

V(X) =∑ xi2 p(xi) – [E(x)] 2 .

i=1

104
The first form of V(X) in definition 5.16 is a definitional form, and the
second is the "computing form." Giving two expressions for V(X) is similar
to giving two expressions for the population variance σ2 in chapter 4.

Example 5.18
Compute the variance of the random variable in Example 5.15.
Solution:
The easiest way to compute the variance of a discrete random variable is by
using a table similar to Table 5.8 below
Table 5.8 Partial computation of the
variance of the random variable in
Example 5.7
I xi P(xi) xi2 xi2P(xi)
1 0 0.8091 0 0.0000
2 1 0.1818 1 0.1818
3 2 0.0091 4 0.0364
0.2182
From Table 5.7, E(X) = 0.2.
Thus
3
V(x)=∑ xi2P(xi)- [E(x)] 2
i=1

= 0.2182 - (0.2)2 = 0.1782.

The variance of X is O.1782 and the standard deviation of X, denoted by s
is (0.1782) 1/2 = 0.422.

5.7 Probability Density Function

Definition 5.17
Probability density function (continuous case)
Let X be a continuous random variable defined over an interval of the real
line from a to b, as illustrated in Figure 5.13.
The probability density function of X, denoted by f(x) , must satisfy two
conditions:
1. f(x) ≥ 0 , a ≤ x ≤ b,

105
 2. The area under f(x) from X= a to X= b must be one.

Figure 5.10 Probability density function f(x) of a continuous

random variable

Notice the similarity between the conditions for P(x) to be a probability

mass function of a discrete random variable given in definition 5.12 and the
conditions placed on f(x) to be a probability density function for a con-
tinuous random variable given in definition 5.13. In the discrete case, prob-
ability is "massed" at the discrete values of the random variables while in the
continuous case, the probability is spread "densely" over the range of the
random variable. In the discrete case, the probability sticks must sum to one
while in the continuous case, the dense set of sticks (the area under the
function f(x) from a to b) must have an area of one unit.
Recall that if we desire to compute P(c ≤ X ≤ d) where d ≥ c for a discrete
random variable, we need only sum the probabilities of X taking on the
values c through d; that is:

106
 x=d
P (c≤ X ≤d)=∑ P(x) , d ≥ c.
x=c

Figure 5.11 Area under the curve corresponding to P(c ≤ X≤ d)

The analogy to a continuous random variable is straightforward. The

probability that a continuous random variable X takes on a value between c
and d is the area under f(x) between c and d as illustrated in Figure 5.11.
There is one significant difference in computing probabilities for a discrete
and a continuous random variable. If the discrete random variable X can
assume the value e, then the probability that X does assume this value is
P(e), the height of the stick over e on the stick diagram for X. However, if e
resides within the defined interval for a continuous random variable X, the
probability that X assumes the value of e is zero; that is, P(X = e) = 0. This
is true regardless of the numerical value of e over the defined interval of real
numbers for the random variable X. Thus, P(X = e) is not equal to f(e), the
height of the curve at the point X = e.
The reason for this may be argued as follows. Let us assume that the
continuous random variable is defined for values on the real line from X = a
to X = b. Assume that e is in the interval (a, b) and we let P(X = e) = f(e).
Since we have defined the probability that X assumes the value of e in (a, b)
in this manner, this definition must also hold for all other values, say e1, e2,

107
e3, . . . , in the interval (a, b). Since there is an infinite number of values (not
countable) between (a, b), we can see that the sum of the "probabilities" f(e)
+ f(e1 )+ f(e2) + f(e3) +••• will quickly exceed one, which means that any
particular number, say f(e2), can no longer be interpreted as a probability.
Indeed, a single f(ei) may exceed one by itself if the height of the curve at
the point X = e; is greater than one.

Another way to look at this is to write P(X = e) as P(e ≤ X≤ e) and use the
definition for the probability that X assumes a value between two points,
say c and d, as illustrated in figure 5.9. Since there is no area between e and
e, P(X = e) = P(e ≤ X≤ e) = 0.

Table 5.9 Comparison of properties of discrete and continuous random

variables

Example 5.18
Consider the probability density function
1 0 ≤x≤ 1 ,
f(x)=
0 x <0 or x> 1.

Find :
1. P(0 ≤ x ≤ 1)
2. P(0.25 ≤ x ≤ 0.60)
3. P(x > 0.75)

108
Solution:
1
1. P(0 ≤ x ≤ 1)= 0∫ f(x) dx
1
= 0∫ 1 dx = 1.
0.6
2. P(0.25 ≤ x ≤ 0.60) = 0.25 ∫ f(x) dx
1
3. P(x > 0.75) = 0.75∫ f(x) dx

Definition 5.18
Expectation of a random variable (continuous case)
Let X be a continuous random variable defined over the interval (a, b) with
probability density function f(x). The mean or expected value of the
random variable, denoted by E(X), is given by

b
E(X) = a∫ x f(x) dx.

Definition 5.19
Variance of a random variable (continuous case)
Let X be a continuous random variable defined over the interval (a, b)
with probability density function f(x). The variance of the random
variable, denoted by V(X), is given by

b
V(X) = a∫ [x -E(x)] 2 f(x)dx, or

b
V (X) =a∫ x2 f(x) dx-[E(x)] 2 .
It is interesting to note the similarity in the computation of the expected
value and the variance of a discrete and a continuous random variable as
illustrated in Table 5.10. The analogy to summing in the discrete case is
integrating in the continuous case .

109
Table 5.10 Comparison of the expected value and variance formulas
for discrete and continuous random variables

Example 5.19
Let

0.5, 0 ≤ x ≤ 2 ,

f(x)= {

0 , otherwise.

Find:
a. E(x)
b.Var (x)

Solution :
2
E(x)= 0∫ 1/2 x dx = x2 /4 ]02 = 1 ,
2
Var(x)= 0∫ 1/2 x2 dx – (1)2 =2/6.

110
Tutorial 5.2
I . In a certain population of voters, it is known that 60 percent are
Democrats and 40 percent are Republicans. If a sample of' three voters is
extracted from this population, find the probability distribution of the
random variable, X = Number of Democrats in the sample.

2. An automobile salesperson has a probability of 0.20 of selling a car to

each individual interested in buying a car with whom she speaks on the
showroom floor. On a certain day, the salesperson talks with four individuals
regarding the purchase of a car. Find the probability distribution of the
random variable, X = Number of cars sold. Show the distribution in the form
of a table, a stick diagram, and (if possible) a formula.

3. A multiple choice exam consists of four questions, each of which has four
possible answers. If a student is forced to guess on all four questions, what is
the probability distribution of the random variable, X = Number of correct
guesses?

4. In an organization consisting of 5 women and 10 men, a committee of

four individuals is to be selected at random from the 15 people. Find the
distribution of the random variable, X = Number of women on the
committee.

5. In a six-cylinder automobile engine. two spark plugs are defective. Three

spark plugs are removed at random and checked. Let X be the number of
defectives found (0, 1, or 2). Find the probability distribution of X.

6. Suppose two dice are tossed. Let X be the sum of the dots on the top faces
of the two dice. Find the probability distribution of X.

7. Let X be a random variable with probability distribution given by the

following table:

111
 X P(x) Find: a. The expected value
0 0.70 of X.
b. The variance of X.
c. The mode of X.
1 0.20 d. The range of X.
2 0.06
3 0.04

8.It is found that the probability distribution of X is:

X P(x)
0........0.95
1........0.03
2........0.015
3 .......0.003
4 .......0.0015
5 .......0.0005

Find:

a. The expected value

of X.
b. The variance of X.
c. The mode of X.
d. The range of X.

9. Consider the formula: P(x) = x2/21, x = 0, 1, 2, 4. Is P(x) a probability

mass function? If so, show the distribution of X in tabular form and compute
the expected value of X.

10. Simulate the experiment in Problem 5 by taking six scraps of paper

and marking on two of them, the letter "D." Place the six scraps of
paper in a box and randomly select three of the scraps. Record the
number marked D.Do this experiment 100 times. Construct a frequency
distribution for this sample and compare it with the theoretical
probability distribution determined in Problem 5.
11. Compute the expected value of the random variable X in Problem 5.
Find the sample mean number of defectives recorded per trial in the

112
Problem 10 simulation experiment. Does X, the sample mean, provide a
good estimate of E(X)" Should X provide a good estimate of E(X)?
Why?

12. From the most recent national census, it is found that the number of
children (X) in American families follows the following probability
distribution:
Number of Proportion of
children, X families. P(x)
0 ................. 0.48
1.................. 0.20
2.................. 0.15
3.................. 0.08
4 ................. 0.05
5 ................. 0.03
6 ………….. 0.01

It is assumed that the proportion of families with more than six children
is negligible.
a. Find the expected value and standard deviation of X.
b. Form a stick diagram of this distribution. Is the distribution skewed?
c. find var(X).

13. Consider the following functions

(2/3) x -1< x< 2 ,

f(x)=
0 otherwise

1/5 , -1 < x < 4 ,

f(x)=

0, otherwise.

113
 Calculate :

a.p(-1<x<0) ,

b.p(X>1) ,

c.p(X=2),

d.E(X) ,

e.var(X) .

114
 6
Binomial & Normal
distributions

6.1 Probability function

6.2 The binomial distribution
6.3 The normal distribution
Tutorial 6

115
6.1 Probability Function

The following tables, serve to illustrate what we mean by a probability

function, namely, a correspondence which assigns probabilities to the values
of a random variable.

Example 6.1
The first of the two tables which follow was easily obtained on the basis of
the assumption that each face of the die in question has a probability of 1/6
and the second was obtained by considering as equally likely the eight
possible outcomes HHH, HHT, HTH, THH, HTT, THT, TTH, and TTT of
three flips of a coin, where H stands for heads and T for tails:

Table 6.1 Table 6.2

Note that in each of these examples the sum of all the probabilities is 1.
Note also that since the values of probability functions are probabilities, they
must always be positive or zero, and cannot exceed 1.
Whenever possible, we try to express probability functions by means of
formulas which enable us to calculate the probabilities associated with the
various values of a random variable. With the usual functional notation.

116
 we can thus write
f(x)=1/6 for x=1,2,...,6,
for the first of the above examples, where f(1) represents the probability of
rolling a 1, f(2) represents the probability of rolling a 2, and so on.

6.2 The Binomial Distribution

There are many applied problems in which we are interested in the proba-
bility that an event will take place x times in n "trials," or in other words, x
times out of n, while the probability that it will take place in any one trial is
some fixed number p and the trials are independent. We may thus be
interested in the probability of getting 24 responses to 80 mail question-
naires, the probability that in a sample of 50 voters 32 will favor Candidate
A, the probability that 3 of 10 laboratory mice react positively to a new drug,
and so on. Referring to the occurrence of any one of the individual events as
a "success", we are thus interested in the probability of getting x successes in
n trials. To handle problems of this kind, which incidentally include, we use
a special probability function, that of the binomial distribution.
If p denotes the probability of a success on any given trial, the probability
of getting x successes in n trials (and hence, x successes and n - x failures) in
some specific order is px (1 - p)n-x. There is one factor p for each success, one
factor 1 - p for each failure, and the x factors p and n - x factors 1 - p are all
multiplied together by virtue of the assumption that the n trials are
independent. Since this probability is the same for each point of the sample
space where there are x successes and n- x failures (it does not depend on the
order in which the successes and failures are obtained), the desired
probability for x successes in n trials in any order is obtained by multiplying
px (1-p)n-x by the number of points of the sample space (that is, individual
outcomes) where there are x successes and n - x failures. In other words, px
(1-p)n-x is multiplied by the number of ways in which the x successes can be
distributed among the n trials, namely, by n!/x!(n-x)! we have thus arrived at
the following result:

117
 Definition 6.1
Binomial distribution
The probability of getting x successes in n independent trials is given
by
f(x)= (xn) px (1-p)n-x for n=0,1,2,...,n,
where p is the constant probability of a success for each individual
trial.
It is customary to say that the number of successes in n trials is a
random variable having the binomial probability distribution, or
simply the binomial distribution. The terms "probability distribution"
and "probability function" are often used interchangeably, although
some persons make the distinction that the term "probability
distribution" refers to all the probabilities associated with a random
variable, and not only those given directly by its probability function.
Incidentally, we refer to this distribution as the binomial distribution
because for x = 0, 1, 2, . . ., and n, the values of its probability function
are given by the successive terms of the binomial expansion of
((1 - p) + p)n.
Example 6.2
To illustrate the use of the above formula, let us first calculate the
probability of getting 5 heads and 7 tails in 12 flips of a balanced coin.
Substituting x = 5, n = 12, p=1/2, and (12!/5!.7!)=792
f(5)= 792 (1/2)5 (1-1/2)12-5= 99/512,

or approximately 0.19. Similarly, to find the probability that 7 of 10 mice

used in an experiment will react positively to a drug, when the probability
that any one of them will react positively is 4/5 we substitute x = 7, n = 10,
p =4/5 and (10!/7! 3!)= 120 , and we get

f(7)=120(4/5)7(1-4/5) 10-7
or approximately 0.20 .

118
Remark 6.1
Some of probability values of binomial distribution are given in Table
1(in appendix).

Example 6.3

To give an example in which we calculate all of the values of a bino-

mial distribution, suppose that a safety engineer claims that only 60
per cent of all drivers whose cars are equipped with seat belts use them
on short trips. Assuming that this figure is correct, what are the
probabilities that under such conditions 0, 1, 2, 3, 4, or 5 of 5 drivers
will be using their seat belts? Substituting n = 5, p = 0.60, and,
respectively, x= 0, 1, 2, 3, 4, and 5, we use

f(x)= (xn) px (1-p)n-x for n=0,1,2,3,4,5,

Then

where all the answers are rounded to three decimals. A histogram of this
distribution is shown in figure 6.1.

119
Figure 6.1 Histogram of binomial distribution with n=5 and p=0.60.

120
Figure 6.2 Some specific members of the binomial distributions where :
a- p = 0.2 , b- p = 0.5 , c- p = 0.8 ,

d- p = 0.2 , e- p = 0.5 , f- p = 0.8 .

Remark 6.2 If X is a binomial random variables then its mean E(x) and
variance , V(x) are given by:

… E(x) =n.p , var(x)= n.p(1-p) …

Example 6.4
Consider figure 6.2 find the mean and the variance for each p.m.f.
Solution:
i) E(x)=2(0.2) ,
Var(x)=2 (0.2)(0.8) ,

121
ii) E(x)=2(0.5) ,
Var(x)=2(0.5)(0.5) ,
iii) E(x)=2(0.8) ,
Var(x)=2(0.8)(0.2),
iv) E(x)=5(0.2) ,
Var(x)=5(0.2)(0.8) ,
v) E(x)=5(0.5) ,
Var(x)=5(0.5)(0.5) ,
vi) E(x)=5(0.8),
Var(x)=5(0.8)(0.2) .

6.3 The Normal Distribution

The normal distribution is "probably" the most important probability

distribution in statistics. It is a probability distribution of a continuous

Figure 6.3 Three forms of the normal distribution

random variable, yet it is often used to model the distribution of discrete

random variables as well as the distributions of other continuous random

variables.
The reason for the versatility in using the normal distribution as a probability
distribution model is indicated in figure 6.3. The basic form of the normal

distribution is that of a bell-it has a single mode and is symmetric about its
central value. The flexibility in using the normal distribution is due to the
fact that the curve may be centered over any number on the real line and that
it may be made flat or peaked to correspond to the amount of dispersion in
the values of a random variable. Many quantitative characteristics have
distributions similar in form to the normal distribution's bell shape.

122
Examples of random variables that have been modeled successfully by the
normal distribution are the height and the weight of people, the diameters of
bolts produced by a machine, the IQ of people, the life of batteries or light
bulbs, and so on. Typically, the type of experiment that produces a random
variable that can be successfully approximated by a normal random variable
is one in which the values of the random variable are produced by a measur-
ing process, where it is known that the measurements tend to cluster sym-
metrically about a central value. A random variable that is an average or a
sum of values of another random variable is, under very general conditions,
almost always distributed approximately as a normal random variable,
regardless of the form of the distribution of the random variable whose
values are summed or averaged. An example of such a random variable is
the mean grade point average of a randomly selected group of students. The
notion that a random variable that is an average is distributed as a normal
random variable is discussed in the next chapter with the central limit
theorem.
Unfortunately, if it is known that the distribution of a random variable is
symmetrically distributed with a single mode, the random variable may not
necessarily be a normal random variable. There are other distributions in
statistics that are unimodal and symmetric. However they also can often be
modeled successfully by the normal distribution. For a random variable to be
normally distributed, the mathematical expression delineating the form of
the bell must be of a specific type, as described in the following definition.

Definition 6.2

123
6.3.1 Mean and variance of the normal random variable

The mean and variance of the normal random variable may be determined
by performing the following integrations:

As might be suspected, these integrals are not easily evaluated. The results
of integration are rather simple, however, and are given in the following
theorem.
Remark 6.3

Notice that the mean depends only on the parameter µ, and that the variance
depends only on the parameter σ. Thus, the normal distribution may be
located over its central value on the real line independently of the amount of
dispersion σ2 specified for the distribution. Contrast this with the binomial
distribution (and others discussed thus far) in which the mean and the
variance both depend upon the parameters n and p, [E(X) = n.p ,
V(X) = n.p.(1-p) ] and hence are not independent of one another. This
property of the normal distribution adds immeasurably to its flexibility in
modeling the distributions of non-normal random variables.

124
 We will now return to the problem of computing probabilities associated
with a normal random variable.

6.3.2 Standardized normal distribution

Probabilities associated with any member of the normal distribution family

can be computed from a table of probabilities compiled for the standard
normal distribution.

Definition 6.3
Standard normal distribution

A normal distribution with µ= 0 and σ = 1 is called a standard normal

distribution. When a normal random variable X has a mean of zero and a
variance of one, it will be called a standardized normal random variable
and will be denoted by Z. The probability density function of the
standardized normal random variable Z is:

The form of the standard normal distribution is illustrated in figure 6.4. As

indicated in figure 6.4, and for any normal distribution, 68.27 percent of the
values of z lie within one standard deviation of the mean, 95.45 percent of
the values lie within two standard deviations of the mean, and 99.73 percent
of the values lie within three standard deviations of the mean.

Probabilities of a standardized normal random variable of the form P (0 ≤ Z≤

a) are provided in Table 2(appendix) . By using the fact that the normal
distribution is symmetric about its mean (zero in this case), and that the total
area under the curve is one (half to the left of zero, and half to the right), the
probability that Z resides in any interval on the real line may be determined
from this table, as the following example indicates.

125
 Figure 6.4 Standard normal distribution

Example 6.5

Find the area under the standard normal distribution curve for each of the
intervals listed below.
a. Between Z = 0 and Z = 2.0
b. Between Z=-1.28 and Z = 0.0
c. Between Z=-0.58 and Z= 2.54
d. Between Z = 1.20 and Z = 2.4
e. Greater than Z = 2.87

Solution:

a. In Table 1, proceed downward in the leftmost column until 2.0 is reached.

Select the first column marked .00 indicating that the second decimal place
is zero. The area read from the table is 0.4772.

126
b. Since the normal distribution is symmetric, the area between - 1.28 and
0.0 is equal to the area between 0.0 and 1.28. Thus, proceed down the
leftmost column until 1.2 is reached. Select the ninth column marked 0.08.
The resulting number in the table is 0.3997.

127
c. We may determine this area in two parts: total area = (area between 0.0
and 2.54) + (area between -0.58 and 0.0). The area between 0.0 and 2.54 is
0.4945 from Table 1. The area between -0.58 and 0.0 is the same as the area
between 0.0 and 0.58, which is 0.2190. The answer is 0.4945 + 0.2190 =
0.7135.

d. We may determine this area by differencing the area from 0 to 2.44 and
from 0 to 1.20. The area between 0 and 2.44 is 0.4927, and the area

128
 between 0 and 1.20 is 0.3849. Thus, the area between 1.20 and 2.44 is
0.4927 - 0.3849 = 0.1078.

e. Since the area between 0 and + ∞ is 0.5, we can determine the area from
2.87 to ∞ by subtracting the area from 0 to 2.87 (0.4979) from 0.5: 0.5000
-0.4979=0.0021.

In many problems, we will be given the area in a certain interval and be

asked to determine the value of Z that specifies the interval. This is the
reverse of the problems solved in Example 6.5 demonstrates the use of
Table 1 to solve the "reverse" problem.

Example6.6

Find the value of Z on the standard normal distribution axis for each of
the areas listed below.
a. The area between 0 and z is 0.3413 b. The area to the right of z is 0.8982

129
 Solution:
a. Table 1 must be used in reverse. We look in the body of the table for
the area 0.3413. It appears in the row marked 1.0 and the column marked
0.0. Thus, the value of Z is 1.00.

b. Since the area given is greater than 0.5, we know that z must be less
than zero. The area between z and 0 is 0.8982 - 0.5000 = 0.3982.
Now assume that z is positive and find z so that the area between 0
and z is 0.3982. The area of 0.3982 does not appear in the tables; the
closest numbers are 0.3980 and 0.3997. The exact value of z could
be determined by interpolation, but we will use z= 1.27 since 0.3980
is closer to 0.3982 than is 0.3997. We must remember that z must be
to the left of zero (a negative number). Thus, z = -1.27.

6.3.3 Areas under the normal distribution

Probabilities associated with a normal random variable X that is not

standardized can be determined from Table 1 by using the results of the
following theorem.

130
 Theorem 6.1
Standardization of a normal random variable
If X is a normal random variable, the mean of which is μ and the standard
deviation of which is σ, then
… Z=(x – μ )/σ ,…
is a standardized normal random variable with a mean of zero and a
standard deviation of one.
The following examples illustrate the use of Theorem 6.1, and more
generally, the applicability of the normal distribution model.

Example 6.7

The mean lifetime of 50-watt lightbulbs produced by the Stay-Bright

Lightbulb Company is 200 hours. It is known that the standard deviation is
20 hours. Assuming that the lifetimes of the lightbulbs are normally,
distributed, what are the probabilities that a single 50-watt lightbulb ex-
tracted from the production lot will
a. Burn out between 180 hours and 210 hours?
b. Burn out at a time greater than 250 hours?

Solution:

a. The solution on the X distribution with a mean of 200 and a standard

deviation of 20 is the area between X = x1 = 180 and X=x2= 210. This

131
 area can he determined by first standardizing x1 and x2:

Z1= (x1 – μ )/σ = (180 - 200)/20 = - 20/20 = - 1.00

The area between - 1.0 and 0.50 on the standard normal distribution will
equal the area between 180 and 210 on the X distribution. The area
between z1 = - 1.00 and z2 = 0.50 from Table 1 is:
Area(- 1.00 to 0.00) = 0.3413
+ Area(0.00 to 0.50) = 0.1915
0.5328
Thus , Prob (180 ≤ X ≤ 210) = P(-1.00≤Z≤0.50)= 0.5328. This answer
tells us that 53.28 percent of the 50-watt lightbulbs comprising the

132
133
Tutorial 6

2.

3. Check whether the following can be looked upon as probability

functions (defined in each case only for the given values of x) and
explain your answers:

(a) f(x) = 1/4 for x= 0, 1, 2, 3, or

(b) f(x) = (x+1)/10 for x = 0, 1, 2, or 3;

(c) f(x) = (x-2)/5 for x = 1, 2, 3, 4, or 5;

(d) f(x) =x2/30 for x = 0, 1, 2, 3, or 4.

134
4. In each case check whether the given values can be looked upon as the
values of the probability function of a random variable which can take on
only the values 1, 2, 3, and 4, and explain your answers:

(a) f(1) = 0.24, f(2) = 0.24, f(3) = 0.24 , f(4) = 0.24;

(b) f(1) = 1/6, f(2) = 2/6, f(3) = 2/6 , f(4) = 1/6;

(c) f(1) = 0. 13, f(2) = 0.38, f(3) = 0.04 , f(4) = 0.45;
(d) f( l ) = 1/2, f(2) = 1/4, f(3) = 1/8 , f(4) = 1/16.

5. Use the formula for the binomial distribution to find the probability of
getting :
(a) exactly 3 heads in 8 flips of a balanced coin; (b) at
most 3 heads in 8 flips of a balanced coin; (c) exactly 1 one
in 3 rolls of a balanced die; (d) at most 1 one in 3 rolls of a
balanced die,(e) calculate E(x) and Var(x) .

6. A multiple-choice test consists of 8 questions and 3 answers to each

question (of which only one is correct). If a student answers each question
by rolling a balanced die and checking the first answer if he gets a 1 or a
2, the second answer if he gets a 3 or a 4, and the third answer if he gets a
5 or a 6, find (by means of the formula for the binomial distribution) the
probability of getting
(a) exactly 3 correct answers;
(b) no correct answers;
(c) at least 6 correct answers.

135
 7
Regression Analysis

7.1 Introduction
7.2 Relationships between variables
7.3 Simple linear regression model
7.4 Fitting of a simple linear regression model
Tutorial 7

136
7.1 Introduction

Linear regression analysis is a technique used to predict the value of one

quantitative variable by using its relationship with one or more additional
quantitative variables. For example, if we know the relationship between
height and weight in adult. males, we can use regression analysis to predict
weight given a particular value for height.
The relationship between height and weight is familiar to us; generally,
the taller a person is, the more he weighs. Another example of a familiar
relationship is that of crop yield and the amount of fertilizer applied to the
land; the more fertilizer applied to the land, the greater the yield-to a point.
If too much fertilizer is applied, the crop will be killed off by the fertilizer
chemicals-the land will be "burned." An important relationship in business
is the relationship between the allocation of dollars to advertising effort and
the level of sales of a product; the more money expended in advertising, the
greater the level of sales (in general).
In this chapter, we will emphasize the development of regression analysis
when a single predictor variable is used to predict the variable of interest,
and where the relationship between the variables is linear. In this context,
the variable which is used to predict the variable of interest is called the
independent variable, and the variable we are trying to predict is called the
dependent variable. The analysis used is called simple linear regression
analysis-simple because there is only one predictor or independent variable,
and linear because of the assumed linear relationship between the
dependent and the independent variables.
Certainly, it is common to find that the variable of interest is related to
more than one predictor or independent variable, or that the relationship is
not linear. An example is the level of sales of a product and the advertising
expenditure. The sales level of a product generally "depends" upon more
predictor variables than advertising expenditure alone. However, we will
often find that quite good predictions are possible based upon a single
predictor variable.

137
7.2 Relationships between Variables

The concept of a functional relationship between two variables is

familiar to us. If a functional relationship exists between two variables, then
it is possible to represent the relationship by a formula Y = f'(X), where X is
the independent (or predictor) variable, and Y is the dependent (or pre-
dicted) variable.

Example 7.1
Suppose for every unit of a product sold, a company makes a profit of $3.
Let X = number of units sold, and Y =total profit. Then. Y = 3X. Illustrate
this linear relationship.

Figure 7.1 Graph of the functional relation Y= 3X

138
Solution:

This linear functional relationship is illustrated in Figure 7.1.

For example, if X = 10, Y= 3(10) =30; if X = 30, Y = 3(30) =90 and if X =
50; Y = 3(50) =$150, Notice that all three of the pairs (X, Y) of points fall
exactly on a straight line.

In Example 7.1, the functional relationship is linear. An example of a

nonlinear functional relationship is Y= Xz. If, for example, X= 2, then Y= 4.
A graph of the functional relationship Y = Xz is illustrated in figure 7.2.
In a statistical relationship, the variables are not perfectly related as they
are in a functional relationship. The pairs of points (X, Y) will not all lie
perfectly on the curve representing the relationship between the variables.

Example 7.2

An example of a statistical relationship is the relationship between heights

and weights of adult. males. Table 7.1 contains the heights and weights of
ten randomly selected males. Plot this data as a graph similar to Figure 7.1.

Solution:

These data are plotted in figure 7.2. Clearly, the taller a man is, the more
he weighs. But, the relationship is not a perfect one, as is evident in Figure
7.3. The line in figure 7.2 has been drawn to fit reasonably well through the
ten points, and the points are scattered about this line. The

139
Figure 7.2 Graph of the nonlinear functional relation Y= Xz

scattering of points suggests that some of the variation in weight is not

accounted for by height alone. For instance, two men are 70 inches tall, but
their weights differ- 185 pounds and 170 pounds. The variation in weight not
accounted for by height alone may be considered to be random in nature, but
may also be due to the failure to include other important independent
predictor variables. The randomness of the scattering of points about the
fitted line is an important element in assessing the validity of a regression
model. Before explaining how we fit a line to the data, we must first
describe the regression model and the assumptions necessary for its correct
application.

140
Table 7.1 Heights (X) and weights (Y) of ten randomly selected
adult males
Height (X) Weight (Y)
inches pounds
60 ...................... 110
65 ...................... 150
74 ...................... 200
70 ...................... 185
70 ..................... 170
66 ..................... 160
68 ...................... 180
72 ...................... 195
64 ...................... 135
71 ...................... 215

Figure 7.3 Plot of data in Table 7.1

141
7.3 Simple Linear Regression Model

The simple linear regression model is a mathematical way of stating the

statistical relationship that exists between two variables. The two principle
elements of a statistical relationship are:

(1) the tendency of the dependent variable Y to vary in a systematic way

with the independent variable X, and

(2) the scattering of points about the "curve" that represents the rela-
tionship between X and Y.

These two elements of a statistical relationship are represented in a simple

linear regression model by assuming that:

(i) there is a probability distribution of Y for each value of X, and

(ii) the means of these probability distributions fall perfectly on a line.

These two assumptions are illustrated in figure 7.4 for the Example 7.2
data. The systematic way in which Y varies as a function of X is identified
as a straight line, the regression line of Y on X. The regression line goes
perfectly through the means of the conditional probability distributions of Y,
given a value of X. The data are collected by taking random samples from
the conditional probability distribution of Y for values of X. For example,
from Table 7.1, when X = 60 inches, Y was observed to be 110 pounds. This
particular value of Y represents a random sample of size one drawn from the
conditional probability distribution of Y when X = 60 inches.

142
 Figure 7.4 Graphical form of the simple linear regression model

The formal statement of the simple linear regression model is:

Yi=β0 +βi.Xi+ei, i= 1,2,...,n,

where:
Yi = Value of the dependent variable in the ith trial ,
β0, β1= Parameters in the model,
Xi = Value of the independent variable in the ith trial,
ei = Random error term in the ith trial.
By trial, we mean an observed value of Y for a fixed value of X. For
example, in Table 7.1, the data are generated by making ten trials of the
simple experiment. In the first trial, we set X = 60 inches and from the
distribution of weights for all. adult males who are 60 inches tall, we sample
one whose weight is 110 pounds.
There are actually two ways we may acquire the needed sample information
as given in Table 7.1: by experimentation or by survey. To experimentally
generate the sample data, we would select a set of values of X, and for each
we would randomly sample one or more values of Y. For example, we may
be interested in the yield of a chemical compound Y measured in grams as a
function of pressure X in a chemical production process.

143
We could select a set of pressures (values of X) and then run the production
process at each pressure setting one or more times to produce observations
on Y. Alternatively, we could generate the sample data by taking a survey.
For example, we could randomly sample ten adult. males to determine their
heights and weights. But, the survey method has the disadvantage that we
must take whatever values of X(height) occur in the survey; the selection of
the set of values of X, the independent variable, is out of our control. We
might be so unfortunate, for instance, to find that all ten men in our survey
were 64 inches tall. We ideally want a spread of X values over the range of
interest and over which the regression line will be built.

It is always better to produce the sample data by experimentation, if

possible, for then we can control the independent variable X-the experiment
can be designed to suit our needs. When experimentation is not possible,
surveys must be used to generate the data.
The assumptions corresponding to the simple linear regression model are:

1. For the ith trial, the expected value of the error component ei is zero
[E(ei) = 0], and the variance of the error component [V(ei)] is σ2 and is
constant for all values of i, i= 1, 2, ..., n.
2. The error components in any pair of trials, say the ith and the jth, are
uncorrelated.
3. The terms βo and β1 in the model are parameters whose values are
typically unknown and must, therefore, be estimated from sample data.
Further, X; is considered to be a known constant in the model.

The consequences of these assumptions are:

1. The observed value of Y in the i th trial, Yi, is the sum of two
components; a constant and a random variable:
Yi = β0 + β1 Xi + ei
↑ ↑
constant random variable

144
2. By using the expectation operator rule given in Chapter 5, we get

E[Yi]=E[β0+ β1Xi+ei]= β0+ β1Xi+E[ei]. But,

E[ei]=0 , so E[Yi]= β0+ β1Xi.

Thus, the mean of the conditional probability distribution of Y given a value

of X, denoted by µ y/x, is equal to β0 + β1 Xi.
And. therefore, the regression function corresponding to the regression
model is

… E[ Y]= β0 + β1 X. …

3. By using the variance operator rule given in Chapter 5,

V[Yi]=V[β0+ βi Xi+ei]=V[β0+ β1 Xi]+V[ei]

=0+V[ei].

But, … V[ei] = σ2, so V[Yi] = σ2. …

Thus, the variance of the conditional probability distribution of Y given a

value of X, denoted by σ2y/x, is equal to σ2 and each conditional
probability distribution has the same variance, σ2.

4. The observed value of Y in the ith trial is larger or smaller than μy/x
by the amount ei, the value of the error component in the ith trial.
5. By the second assumption , the outcome in any trial is not affected by or
does not itself affect the error term in any other trial .

145
7.4 Fitting of a Simple Linear Regression Model
Since β0 and β1, are generally not known in a regression problem, they must
be estimated from sample data taken on the dependent variable Y for a
number of values of the independent variable X. These pairs of sample
values are obtained either by experimentation or by survey. The data given
in Table 7.1 were determined by survey- 10 adult. males were selected at
random and their heights and weights were recorded.
Returning to the data given in Table 7.1, we will first produce a scatter plot
of these data. The scatter plot is given in figure 7.5. In figure 7.6, we have
superimposed two "fitted" lines through this scatter of points,

Figure 7.5 Scatter plot of data given in table 7.1

146
Figure 7.6 Scatter plot and fitted lines for data in Table 7.1

Ŷ = 170 + 0 X and Ŷ = -310.76 + 7.07X, respectively. It is apparent in

Figure 7.6 that the line Ŷ = -310.76 + 7.07X fits the given data "better,"
but we must establish a criterion to evaluate when one line is "better" than
another so that we may find the best fitting line. The criterion we shall use
is called least squares. For each sample observation (Xi, Yi), the least
squares criterion considers the deviation of Yi from its expected value:

and requires that values of β0 and β1, be found which minimize:

The specific values of β0 and β1, that minimize LS are the regression coef-
ficient estimates, denoted by b0 and b1 respectively.

147
 Thus, the least squares criterion requires that we find a line, denoted by Ŷ =
b0+ b1 X, such that the sum of the squared vertical deviations between the
line and the scatter of points is minimized. In figure 7.6, the vertical
deviations corresponding to the line Ŷ = 170 + 0X, where b0 = 170 and
b1 = 0, are indicated. Obviously, the line Ŷ = -310.76 + 7.07X in figure
13.11, where b0 = -310.76 and b1 = 7.07, does much better in the least
squares sense because its vertical deviations from the scatter of points, when
squared and summed, will be less than the sum of squared deviations for the
line Ŷ = 170 + 0 X.

It turns out that the values of b0 and b1 , which minimize LS are solutions to
the following two simultaneous equations, which are referred to as the
normal equations:

Solving the normal equations for b0 and b1, produces the point estimators of
β0 and β1 respectively. The resulting formulas for b0 and b1,
are given below:

Example 7.3

Let us now fit a simple Linear regression model to the data on heights and

148
weights given in Table 7.1

Solution:

The computations for determining b0 and b1, are given in Table 7.2. The
format in this table provides a convenient worksheet for finding the neces-
sary components in the formulas for b0 and b1

Thus, the fitted regression line is Ŷ=- 310.76 + 7.07X and this is the best
fitting line based upon the least squares criterion.

TABLE 7.2 Computation worksheet for determining b0 and b1 in Example 7.3

149
Tutorial 7

I. Distinguish between dependent and independent variables in a regression

model.

2. Why is it important to plot a scatter diagram of the relationship between

variables in a simple linear regression model?

3. What is meant by "least squares" in a simple regression model?

4. Describe the normal equations and how they are derived.

5. Discuss the assumptions made in using simple linear regression about the
distributions of the conditional mean values.

6. What is meant by the coefficient of determination?

7. Plot each of the following sets of data as a scatter diagram. Which curves
seem to fit the data best?
Determine the regression equation for each set of data.

150
151
Table 1 Binomial Distribution Probability

152
Table 2 Standard normal distribution areas

153
 References

1.William G. Cochran.
Sampling Techniques.

2. Irwin Miller & John E. Freund

Probability and Statistics for Engineers.

3.John E. Freund
Modern Elementary Statistics .

4.Roger C. Pfaffenberger & James H. Patterson

Statistical Methods for Business and Economics .

5.Snedecor , G.W. and Cochran , W.G.

Statistical Methods .

154

View publication stats