Computing MAths

UNIT 1 PROBABILITY DISTRIBUTIONS
Structure
1.0
1.1
1.2
1.3
Introduction
Objectives
Random Variables
Discrete Random Variable
1.2.1
1.2.2
1.4
Page Nos.
5
Binomial Distribution
Poisson Distribution
Continuous Random Variable

1.4.1
1.4.2
1.4.3
1.4.4
Uniform Random Variable

Exponential Random Variable
Normal Distribution
Chi-square Distribution
1.0 INTRODUCTION
The discipline of statistics deals with the collection, analysis and interpretation of data.
Outcomes may vary even when measurements are taken under conditions that appear to
be the same. Variation is a fact of life. Proper statistical methods can help us understand
the inherent variability in the collected data, and facilitate the appropriate analysis of the
same.
Because of this variability, uncertainty is a part of many of our decisions. In medical
research, for example, interest may center on the effectiveness of a new vaccine for
AIDS; an agronomist may want to decide if an increase in yield can be attributed to a new
strain of wheat; a meteorologist may be interested in predicting whether it is going to rain
on a particular day; an environmentalist may be interested in testing whether new controls
can lead to a decrease in the pollution level; an economists interest may lie in estimating
the unemployment rate, etc. Statistics, and probabilistic foundations on which statistical
methods are based, can provide the models that may be used to make decisions in these
and many other situations involving uncertainties.
Any realistic model of a real world phenomenon must take into account the
possibilities of randomness. That is, more often that not, the quantities we are interested in
will not be predicted in advance, but rather will exhibit as inherent variation that should
be taken into account by the model. Such a model is, naturally enough, referred to as a
probability model.
In this unit we shall see what is a random variable and how it is defined for a
particular random experiment. We shall see that there are two major types of probability
distribution. We shall investigate their properties and study the different applications.
1.1 OBJECTIVES
After reading this unit, you should be able to
describe events and sample spaces associated with an experiment;

define a random variable associated with an experiment;
decide whether a random variable is discrete or continuous;
describe the following distributions
a) Binomial distribution
b) Poisson distribution
c) Uniform distribution
d) Exponential distribution
e) Normal distribution
f) Chi-square distribution.
1.2 RANDOM VARIABLES

Definition A Random experiment or a Statistical experiment is any act whose
outcome can not be predicted in advance. Any outcome of a random experiment is known
as event
We will start with the following illustrations :
1)
The number of telephone calls received by Monica, a telephone operator in a call
center in Delhi, between say 1:00 am and 3:00 am in the early morning.
2)
The amount of rainfall in Mumbai on August 1st.
3)
The number of misprints on a randomly chosen page of a particular book.
4)
The final results of the 5 one-day matches between India-Pakistan.
5)
The outcome of rolling die.
6)
The volume of sales of a certain item in a given year.
7)
Time to failure of a machine.
In all the above cases there is one common feature. These experiments describe the
process of associating a number to an outcome of the experiment (i.e. to an event). A
function which associates a number to each possible outcome of the experiment is called a
random variable. It is often the case that our primary interest is in the numerical value
of the random variable rather than the outcome itself. The following examples will help to
make this idea clear.
EXAMPLE 1: Suppose we are interested in the number of heads, say X, obtained in three
tosses of a coin.
SOLUTION
If we toss a coin three times, then the experiment has a total of eight possible outcomes,
and they are as follows;
a1={HHH}, a2={HHT}, a3={HTH}, a4={HTT}
a5={THH}, a6={THT}, a7={TTH}, a8={TTT}
Denoting the event corresponding to getting k heads, k=0,1,2,3,.. as{X=k}, observe that
{X=0} => {a8} ; {X=1} => {a4, a6,a7} ; {X=2} =>{a2, a3,a5} ; {X=1} =>{a1}
Note that each value in the support of X corresponds to some element (or set of elements)
in the sample space S. For example the , the value 0 corresponds to the element {a8},
while the value 1 corresponds to the set of elements { a4, a6,a7}
Therefore the sample space S, the set of all possible outcomes of this experiment
can be expressed as
S={a1, a2,,a8}
Since X is the characteristic, which denotes the number of heads out of the three tosses, it
is associated with each outcome of this experiment. Therefore, X is a function defined on
the elements of S and the possible values of X are {0,1,2,3} . The set of possible values
that X can take is called the support of X which may be denoted as . Observe, that X can
be explicitly expressed as follows;
X (a1)=3, X (a2)=X (a3)=X (a5)=2, X (a4)=X (a6)=X (a7)=1, X (a8)=0
It is interesting to observe that to each value there is always some element in sample
space or a set of element in sample spaces. For, example, the set of element in sample
spaces corresponding to the value 0 is the point {a8}; for 1, the set is {a4, a6, a7}, for 2,
the set is {a2, a3, a5} and for 3 the point is {a1}.
Therefore, we can easily make the following identification of events
corresponding to the values associated by X. Denoting the event corresponding to 0, as
{X=0}, similarly for other values, observe that
{X=0}={a8}; {X=1}={a4, a6, a7}; {X=2}={a2, a3, a5} {X=3}={a1}
If we assume that the coin is unbiased and the tosses have been performed independently,
the probabilities of all the outcomes of the sample space are equal, that is
1
P(a1)=P(a2)==P(a8) = . Therefore, using the probability law of disjoint events we can
8
easily obtain
P ({X=0}) = P ({a8}) =
1
8
3
8
3
P ({X=2}) = P ({a2, a3, a5}) = P ({a2})+P ({a3})+P ({a5}) =
8
1
P ({X=3}) = P ({a1}) =
8
P ({X=1}) = P ({a4, a6, a7}) = P ({a4})+P ({a6})+P ({a7}) =
Therefore, in this case the random variable X takes four values 0,1, 2,3 with the
probabilities 1/8,3/8,3/8,1/8 respectively. It is also important to observe that
P({X=0})+P({X=1})+P({X=2})+P({X=3})=1
It is not a coincidence, it is always true. If we add the probabilities of all possible values
of a random variable it is always one.
To sum up, we say that the random variable X, is a real valued function defined
on all the elements of a sample space of a random experiment. The random variable takes
different values, and for each value there is a probability associated with it. The sum of all
the probabilities of all the points in the support of the random variable X adds up to one.
The following figure will demonstrate the random variable X.
HHH HHT HTH HTT

THH THT TTH TTT
{0, 1, 2, 3}
Figure 1: Schematic representation of a random variable
Because of this representation , one can define probabilities for the set of
numbers(depending on the random variable) rather than working with arbitrary space and
this simplifies the problem considerably. Now we consider the following example.
EXAMPLE 2: You have purchased a new battery operated wristwatch and you have
inserted one new battery into it. Suppose you are interested in the following:
a) How long will it be before the first battery needs replacement ?
b) How many battery will have to be replaced during a one year period?
SOLUTION
Note that both (a) and (b) are random variables and let us discuss them one by one. In
case (a), we want to find the duration of time before the battery needs to be replaced. Note
that the variable takes values continuously along a line say from the time duration A to
time duration B. No values in between A and B are left out. In other words there is no
break in the values assumed by this random variable.
In case (b), the random variable is the number of batteries. This variable can take
values 0 or 1 or 2 etc. There is no continuity, since only non-negative integer values can
be assumed. So, the range of this variable is a discrete set of points. From this discussion
it is clear that the random variable X defined in Example 1 is also a discrete random
variable. The above examples shows that the random variables can be of two types. We
will distinguish between the following two types of random variables;
1. Discrete Random Variable , and
2. Continuous Random Variable.
Check Your Progress 1
E 1: Suppose you take a 50-question multiple-choice examination, guessing your answer,
and are interested in the number of correct answers obtained. Then
(a) What is the random variable X that you will consider for this situation?
(b) What is the set of possible values of X in this example?
(c) What does P(X=10) mean in this context?
Now in the next two sections we will describe the discrete and continuous random
variables in detail.
1.3 DISCRETE RANDOM VARIABLE

In this section, we define properly a discrete random variable and mention some of its
basic properties.
DEFINITION : A random variable X is said to be discrete, if the total number of values X
can take is finite or countably infinite(i.e the support of X is either finite or countable).
The support of X may be listed as {a0, a1, a2,}. Moreover, for each value of
ai, there is a probability associated with it. Suppose we denote them as {p0, p1, p2,},
therefore, we have P(X= ai)=pi for I=0,1,. From the properties of a random variable and
from the probability law, we have
(a) pi 0 for all i 0
(b)
= p0 + p1 + p2 + .... = 1
i =0
From the above discussions, it follows that there exists a function p: R as follows;
if
a = ai ;
i = 0,1, 2.....
p
p(a) = i
otherwise
0
This function p is called the probability mass function (p.m.f.) of the discrete random
variable X.
The collection of the pairs {(ai, pi;I=0,1,} is called the probability distribution of X.
Another function which plays a very important role for any random variable is
known as the cumulative distribution function (c.d.f.) or simply the distribution function
of the random variable. The c.d.f. F : R[0,1] of the random variable X is defined as
F(b) = P(X b), for - < b < .
In other words, F(b) denotes the probability that the random variable X takes on a value
which will be less than or equal to b. Some important properties of the c.d.f. F (.) are
(a) F (b) is a non-decreasing function of b.
(b) lim F (b) = 1
b
(c) lim F (b) = 0

b
Now we clarify these concepts with the same example discussed in the previous
section. Suppose X is the random variable denoting the number of heads obtained in three
independent tosses of a fair coin, then the probability mass function (p.m.f ) p is the
function, p: R, such that
1
3
1
p(0) = , p(1) = p(2) = , p(3) =
8
8
8
Therefore, p(ai) = pi 0, for all ai and
3
1 3 3 1
pi = + + + = 1
8 8 8 8
i =0
In this case the p.m.f of the random variable by the function p and the corresponding
1 3 3 1
probability distribution is the set 0, , 1, , 2, , 3, .This can also be
8 8 8 8
expressed in a tabular form as follows
TABLE 1
PROBABILITY DISTRIBUTION OF THE NUMBER OF HEADS
IN THREE INDEPENDENT TOSSES OF A FAIR COIN
The number of heads (X value)

0
1
2
Probability
1
8
3
8
3
8
1
8
Now let us see the graphical representation of the distribution.
0.50
0.40
Probability
0.30
0.20
0.10
0
Number of heads
Figure 2: Graphical representation of the distribution of .X
Graphically along the horizontal axis, plot the various possible values ai of a random
variable and on each value erect a vertical line with height proportional to the
corresponding probability pi.
Now let us consider the c.d.f of the random variable X. Note that if b<0, clearly
F(b)=P(X b)=0, because X takes values only {0,1,2,3}. If b=0, that is F(0)=P(X
0)=P(X=0)=1/8. If 0 b 1, then P( X b)=P(X=0)+P(0<X<b)=1/8+0=1/8. Similarly, if
b=1, F(1)= P(X 1) = P(X=0) + P(X=1)=1/8+3/8=4/8 and so on. Therefore, the c.d.f. F(.)
has the following form;
b<0
0 if
1
if 0 b < 1
8
4
F (b) =
if 1 b < 2
8
7
if 2 b < 3
b3
1 if
Note :Mathematical expectation or Expected values or Expectations forms

the fundamental idea in the study of probability distribution of any discrete
random variable X , the expected value (or mean) , denoted as E(X) is defined as
E(X) = x0p0+ x1p1+ x2p2+..=xipi
Where x0, x1, x2 etc are the values assumed by X and p0, p1, p2 etc are probabilities
of these values. Under special conditions (like all probabilities are equal )then
E(X) = mean of x0, x1 ,x2
Similarly for continuous variables X having density function p(x) where P[X=x] =
p(x) , the Expectation E(X) will be given by integral of xip(xi) w.r.t x
This concept of Expectation also contributes to the definition of Moment
Generating Function of X i.e Mx(t)= E(etx)
6
Example 3 A box contains twice as many red marbles as green marbles. One marble is
drawn at random from the box and is replaced ; then a second marble is drawn at random
from the box.If both marbles are green you win Rs. 50 ; if both marbles are red you loose
Rs. 10 ; and if the y are of different colour then neither you loose nor you win. Determine
the probability distribution for the amount you win or loose?
Solution Say X denote the amount you win (+) or loose (-) ; i.e X= +50 or -10
The probability that both marbles are green is 1/9 i.e. P[X=+50] =1/9
The probability that both marbles are red is 4/9 i.e. P[X=-10] =4/9
The probability that marbles are of different colours is 4/9 i.e. P[X=0] =4/9
Thus the probability distribution is given by following table
Amount(in Rs won(+) or lost(-))
+50
0
-10
Probability
1/9
4/9
4/9

E1: Which of the variables given below are discrete? Give reasons for your answer.
(a) The daily measurement of snowfall at Shimla.

(b) The number of industrial accidents in each month in West Bengal.
(c) The number of defective goods in a shipment of goods from a manufacturer.
1.3.1
Binomial Distribution
One very important discrete random variable (or discrete distribution) is the binomial
distribution. In this subsection, we shall discuss this random variable and its probability
distribution.
Quite often we have to deal with the experiments where there are only two
possible outcomes. For example, when a coin is tossed either a head or a tail will comes
up, a newborn is either a girl or a boy, a seed either germinates or fails to germinate. Let
us consider such an experiment. For example consider the same experiment of tossing a
coin independently three times. Note that the coin need not necessarily be a fair one, that
is P(Head) may not be equal at P(Tail)
This particular experiment has a certain characteristic. First of all it involves
repetition of three identical experiments (trials). Each trial has only two possible
outcomes: a Head or a Tail. We refer to the outcome Head as success and the outcome
Tail as failure. All trials are independent of each other. We also know that the
probability of getting a Head in a trial is p and probability of getting a tail in a trial is 1
p, that is
P(Head) = P(success) = p and P(Tail) = P(failure) = q = 1 - p
This shows that the probability of getting a success or a failure does not change from
one trial to another. If X denotes the total number of Heads, obtained in three trials, then
X is a random variable, which takes values {0,1,2,3}. Then regarding the above
experiment, we have observed the following;
(1)
(2)
(3)
(4)
It involves a repetition of n identical trials (Here n=3).

The trials are independent of each other.
Each trial has two possible outcomes.
The probability of success (p) and the probability of failure (q=1-p) remain
constant and do not change from trial to trial.
Now let us try to compute the probabilities P{X = 0}, P{X = 1}, P{X = 2} and P{X =
3} in this case. Note that
P(X=0) = P(getting tails in all three trials)
= P({TTT}) = (1-p)3 = q3.
Similarly,
P(X = 1) = P(getting one Tail and two Heads in three trials)
= P({THH,HTH,HHT}) = P({THH}) + P({HTH}) + ({HHT})
= (1-p)2p+(1-p)2p +(1-p)2p = 3(1-p)2p = 3q2p.
Similarly,
P(X = 2) = (getting two Tails and one Head in three trials)
= P({HTT,THT,TTH}) = P({HTT}) + P({THT}) + ({TTH})
= (1-p)p2 + (1-p)p2 + (1-p)p2 = 3(1-p)p2=3q2p
Finally
P(X=3) = P(getting Heads in three trials)
= P({HHH}) = p3
Now observe that instead of n=3, in the above example we can easily compute the
probability for any general n. Suppose we compute P(X = r), for 0 r n, then note that
P(X = r) = C(n,r)pr(1-p)n-r = C(n,r)prqn-r,
Where C(n,r) denotes the number of ways n places can be filled with r Heads and n-r
Tails. From your school mathematics, recall that it is the number of combination of n
objects taken r at a time and it can be calculated by the following formula:
n!
C ( n, r ) =
r !( n r )!
Therefore, for r = 0,1,,n,
n!
P( X = r ) =
p r q nr ,
r !(n r )!
where
n = the number of trial made
r = the number of success
p = the probability of success in a trial
q = 1-p = the probability of a failure.
Now we define the binomial distribution formally.
Let X represents the number of successes in the set of n independent identical trials. Then
X is a discrete random variable taking values 0,1,,n. The probability of the event
P(X= r) is given by
n!
p r q n r , r=0,1,2,..,n
r !(n r )!
where n, r, p, q are same as defined before. Such a random variable X is called a binomial
random variable and its probability distribution is called the binomial distribution. A
Binomial distribution has two parameters n and p
P( X = r ) =
E1 : A farmer buys a quantity of cabbage seeds from a company that claims that
approximately 90% of the seeds will germinate if planted properly. If four seeds are
planted, what is the probability that exactly two will germinate?
1.3.2
Poisson Distribution
In this subsection we will introduce another discrete distribution called Poisson

Distribution. First we shall describe the different situations where we can apply this
Poisson Distribution.
Suppose it is the first hour at a bank in a busy Monday morning , and we are
interested in the number of customers who might arrive during that hour, or during a 5minute or a 10-minute interval in that hour. In statistical terms , we want to find the
probabilities for the number of arrivals in a time interval.
To find this probability, we are making some assumptions similar to the binomial
distribution.
(a) The average arrival rate at any time, remains the same over the entire first hour.
(b) The number of arrivals in a time interval does not depend on what has happened
in the previous time intervals.
(c) It is extremely unlikely that more than one customer will arrive at the same time.
Under those above assumptions, we can find the required the probabilities. Suppose X
is the random variable denoting the number of customers that arrive in the first hour, then
P( X = i) =
e i
,
i!
i = 0,1,2,3,...
Where (the Greek letter Lambda) denotes the average arrival rate per hour. For
example, suppose we know that average number of customers that arrive in that bank
during the first hour is 60 and we want to find what is the chance there will be no more
than 3 customers in the first 10 minutes. Since we know that the average arrival rate per
hour is 60 , if we denote to be the average arrival rate per 10 minutes, then
60 10
=
=10. Therefore, we can use the above formula and get
60
e 1010i
P( X = i) =
,
i = 0,1, 2,3.....
i!
But we want to find the probability that no more than 3 customers will be there in the first
ten minutes and that is
P ( X 3) = P ( X = 0) + P( X = 1) + P( X = 2) + P( X = 3)
(1)
= e 10 + e 10 10 +
e 10 10 2 e 10 10 3
+
2!
3!
(2)
0.00005 +0.00050 + 0.00226 +0.00757 = 0.01038
(3)
What does this value 0.01038 indicate? It tells us that if the arrival rates are uniform then
there is only 1% chance that less than three customers will be there, or in other words,
there is a 99% chance that there will be more then 3 customers in the first 10 minutes.
Similarly if we want to find out the chance that there will be no more than 3
customers in the first 5 minutes, then similarly, as above we can see that in this case
60 5
= 5.
=
60
Therefore, if Y denotes the number of customers presents in the first 5 minutes, then
P (Y 3) = P(Y = 0) + P(Y = 1) + P(Y = 2) + P(Y = 3)
(4)
e 5 52 e 5 53
+
2!
3!
0.00674 +0.03369 +0.08422 +0.14037 = 0.26502
= e 5 + 5e5 +
(5)
(6)
From the above two examples it is clear that if we change the time unit (and hence the
value of ), the probabilities will change. The probability mass function (p.m.f ) given by
p (i ) = P ( X = i ) =
e i
, i = 0,1, 2,3, K
i!
represents the Poisson probability distribution. From the series expansion of e , it easily
follows that
e i
P( X = i) =
=1
i!
i =0
i =0
as it should be.
One point that should always be kept in mind is that a random variable denoting
the number of occurrences in an interval of time will follow a Poisson distribution, if the
occurrences have the following characteristics:
(a) The average occurrence rate per unit time is constant.
(b) Occurrence in an interval is independent of what has happened previously.
(c) The chance that more than one occurrence will happen at the same time is
negligible.
Now let us look at some situations where we can apply the Poisson distribution. Here is
an example
EXAMPLE 4: Calls at a particular call center occur at an average rate of 8 calls per 10
minutes. Suppose that the operator leaves his position for a 5 minute coffee break. What is
the chance that exactly one call comes in while the operator is away?
Solution: In this case the conditions (a), (b) and (c) are satisfied. Therefore if X denotes
the number of calls during a 5 minute interval, then X is a Poisson random variable with
8 5
= 4. Therefore,
=
10
P ( X = 1) =
e 4 41
= 4e 4 0.073
1!
10
That means the chance is 7.3% that the operator misses exactly one call.
Check your progress 4
E 1: If a bank receives on an average = 6 bad Cheques per day, what is the probability
that it will receive 4 bad checks on any given day
1.4 CONTINUOUS RANDOM VARIABLE

So far we have discussed about the discrete random variables in details and we have
provided two important discrete distributions namely binomial and Poisson distributions.
Now in this section we will be discussing another type of random variables namely
continuous random variables.
Let us look at the part (a) of Example 2. Note that we want to find the time of
occurrence rather than the number of occurrences. Therefore if the random variable X
denotes the time of occurrence of a particular event, then the random variable X can take
any value on the positive real line or may be any value on a fixed interval say (A,B).
Therefore, the random variable can take uncountably many values. This type of a
random variable which can take uncountably many values is called a continuous random
variable. For a continuous random variable X, the probability that X takes a particular
value is always zero, but we can always specify the probability of X of any interval
through a probability density function (p.d.f.). The exact details are given below.
DEFINITION: Let X be a continuous random variable which takes values in the interval
(A,B). A real valued function f(x): R R is called the p.d.f of X, if
(a) f (x) 0 and f (x) = 0, if x < A or x > B.
(b)
B
A
f ( x)dx = 1
(c) P (c < X < d) =
f ( x)dx
Now we shall see how we can use the graph of the p.d.f. of a continuous random variable
to study real life problems
Example 5: Suppose the Director of a training program wants to conduct a programme to
upgrade the supervisory skills of the production line supervisors. Because the programme
is self-administered, supervisors require different number of hours to complete the
programme. Based on a past study, it is known that the following p.d.f. shows the
distribution of time spent by a candidate to complete the program. From the graph it is
clear that the average time
11
250
Figure 3: The p.d.f. of the time spent by a candidate to complete the program
Spent by the candidate is 250 and it is symmetrically distributed around 250. How can the
Director use this graph to find the following. What is the chance that a participant selected
at random will require
(a) more than 250 hours to complete the program
(b) less than 250 hours to complete the program
SOLUTION:Since the graph is symmetric, therefore, it is clear that area under the curve
above 250 is half. Therefore, the probability that the random variable takes values higher
1
1
than 250 is . Similarly, the random variable takes value lower than 250 is also .
2
2
Please try the following exercise now:
Now in the following subsections we will consider different continuous distributions.
1.4.1
The Uniform Random Variable
The uniform distribution is the simplest of a few well-known continuous distributions,

which occur quite often. It can be explained intuitively very easily. Suppose X is a
continuous random variable such that if we take any subinterval of the sample space, then
the probability that X belongs to this subinterval is the same as the probability that X
belongs to any other subintervals of the same length. The distribution corresponding to
this random variable is called a uniform distribution and this random variable is called a
uniform random variable.
Formally, we define the uniform random variable as follows: The random
variable X is a uniform random variable between (A, B), if its p.d.f. is given by
f ( x) = B A
0
for
A< x< B
otherwise
From the p.d.f. it is clear that if A < a1 < b1 < B, A < a2 < b2 <B and b1-a1 = b2-a2, then
12
P(a1 < X < b1) = P(a2 < X < b2 ). Therefore, if the length of the intervals are same then the
corresponding probabilities will be also equal. Let us see some examples of such random
variables:
EXAMPLE 6: A train is likely to arrive at a station at any time uniformly between 6:15
am and 6:30 am. Suppose X denotes the time the train reaches, measured in minutes,
after 6:00 am.
SOLUTION In this case X is a uniform random variable takes value between (15,30).
Note that in this P(20 < X < 25) is same P(18 < x < 23) and that is equal to
25 30 23 18 1
=
=
30 15 30 15 3
E 1 An office fire drill is scheduled for a particular day, and the fire alarm is likely to ring
uniformly at any time between 10:00 am to 1:00 pm.
1.4.2
Exponential Random Variable
In making a mathematical model for a real world phenomenon, it is always necessary to

make certain simplifying assumptions so as to render the mathematical tractability. On the
other hand, however, we can not make too many simplifying assumptions, for then our
conclusions obtained from the mathematical model, would not be applicable to the real
world problem. Thus in short, we must take enough simplifying assumptions to enable us
to handle the mathematics but not so many that the mathematical model no longer
resembles the real world problem. One simplifying assumption that is often made is to
assume that certain random variables are exponentially distributed. The reason for this is
that the exponential distribution is both easy to work and is often a good approximation to
the actual distribution.
We use exponential distribution to model lifetime data that is the data, which are mainly
non-negative. Although, with proper modifications, it can be used to analyze any type of
data (not necessarily non-negative only). The property of the exponential distribution,
which makes it easy to analyze, is that it does not deteriorate with time. By this we mean
that if the lifetime of an item is exponentially distributed, then an item which has been in
use for say ten hours, is as good as a new item in regards to the amount of time remaining
until the item fails.
Now we define the exponential distribution formally: A continuous random
variable X is said to be an exponential random variable if the p.d.f of X is given for some
X>0, by
f(x) = e - x for x>0

0
otherwise
Here is known as the rate constant. It can be shown mathematically that the average
1
value or the mean values of X is .Shapes of f (x) for different values of are provided
in the figure below. From the figure,
13
=3
2.5
=2
2
1.5
=1
=0 5
0.5
0
0
Figure 4: The p.d.f. of the exponential distribution for different values of .
It is clear that f(x) is a decreasing function for all values of and f(x) tends to 0 as
x tends to . Now consider the following example.
EXAMPLE 7: suppose that the amount of time one spends in a bank to withdraw cash
from an evening counter is exponentially distributed with mean ten minutes, that is =
1/10. What is the probability that the customer will spend more than fifteen minutes in the
counter?
SOLUTION: If X represents the amount of time that the customer spend in the
counter than we need to find P(X>15). Therefore,
P (X > 15) =
=e
15
=e
3
2
0.223
15
P(X>15)= .223 represents that there is a 22.3 % chance that the customer has to
wait more than 15 minutes.
1.4.3
Normal Distribution
Normal distribution is undoubtedly the most used continuous distribution in different

areas like astronomy, biology, psychology and of course in probability and statistics also.
Because of its practical as well as theoretical importance it has received considerable
attentions in different fields. The normal distribution has a unique position in probability
theory, and it can be used as an approximation to other distributions. In practice normal
theory can frequently be applied with small risk of serious error, when substantially nonnormal distributions correspond more closely to the observed value. The work of Gauss in
1809 and 1816 established techniques based on normal distribution, which became
standard methods used during the nineteenth century. Because of Gausss enormous
contribution, it is popularly known as Gaussian distribution also.
14
We will now state the normal distribution formally: The random variable X is said
to be normally distributed with parameters and , if the p.d.f f(x) of X is given by
f ( x) =
( x )2
2 2
, where - < x <
Here is a real number lying between - and and is a real number lying
between 0 and .
The function f(x) may look a bit strange, but do not get bother. Just notice the
following important things. Note that it involves two parameters and , that means
corresponding to each and we get a distribution function. More over it can be seen
that for - < < and 0 < < , the function f(x) is symmetric about and is a bell
shaped one. Both and have nice interpretation. It can be easily checked that is
the average value or mean of the distribution and provides the measure of spread. The
p.d.f. of two different normal distributions are provided below.
0.8
0.7
(=1,=0.5)
0.6
0.5
0.4
0.3
(=0,=1)
0.2
0
0.1
-4
Figure 5: The p.d.f of the normal distribution for two different values of ( , ).
It is clear from the above figure that the p.d.f. is symmetric about and the shape
depends on . The spread of the distribution is more if is large.
Now let us find the P(a < X < b) for any a and b, when X follows normal
distribution with parameters and , note that,
P(a < X < b) =
1
2
( x )2
2 2
dx = a
1
2
z2
2
dz.
15
The last equality follows by simply making the transformation z =

follows
P(a < X < b) = P (
< Z<
. Therefore it
),
Where Z follows normal distribution with parameters 0 and 1. Although the probability
can not be calculated in a compact form, extensive tables are available for P (Z < z) for
different values of z. The table values can be used to compute P (a < X < b) for any
and .
Say we denote F(a) = P[Z<= a], the probability thet the standard normal variable Z takes
values less than or equal to a.The values of F for different values of a are calculated and
listed in table . One such table is given in the end of this unit
Note that the entries in the table are values of z for z=0.00 0.01,0.02 , ..0.09.To find the
probability that a random variable having standard normal distribution will take on a
value between a and b, we use the equation
P[a < Z < b] = F(b) F(a)
And if either a or b is negative then we can make use of the identity
F(-z) = 1 - F(z)
EXAMPLE 8 Use the table to find the following probabilities
(a) P[0.87 < Z < 1.28]
(b) P[-0.34 < Z < 0.62]
(c) P[Z >= 0.85]
(d) P[Z >= -0.65]
SOLUTION
a)P[0.87 < Z < 1.28] : Find F(1.28 from the table). In the table in the row for Z=1.2 find
the value under column 0.08 it will be 0.8997 . Similarly find F(0.87) =0.8078
so, P[0.87 < Z < 1.28] = 0.8997 0.8078 = 0.0919
b) Similarly P[-0.34 < Z < 0.62] = F(0.62) F(0.34) = F(0.62) [ 1- F(0.34)]
= 0.7324 (1 - 0.6331) = 0.3655
c) Similarly calculate P[Z > 0.85] = 1- P[Z <= 0.85] = 1-F(0.85) = 0.1977
d)P[Z > -0.65] = 1 P [ Z <= - 0.65]
= 1 F (-0.65)
= 1- F(1-F(0.65))
= 0.7422
Next we shall see that how to use the standard normal probability table to
calculate probability of any normal distribution
Standardising
Any normal random variable X, which has mean and variance 2 can be
standardized as follows.
Take a variable X, and
i)
subtract its mean (m or ) and then,
ii)
divide by its standard deviation(s or ).
16
We will call the result, Z, so

X
Z=
For example, suppose, as earlier, that X is an individuals IQ score and that it has a
normal distribution with mean = 100 and standard deviation = 15. To
standardize and individuals IQ score, X, we subtract = 100 and divide the result
by = 15 to give,
X 100
Z=
15
In this way every value of X, has a corresponding value of Z. For instance, when
130 100
90 100
X = 130, Z =
= 2 and when X = 90, Z =
= 0.67 .
15
15
The distribution of standardized normal random variables
The reason for standardizing a normal random variable in this way is that a
standardized normal random variable
X
Z=
has a standard normal distribution.

That is, Z is N(0,1). So, if we take any normal random variable, subtract its mean
and then divide by its standard deviation, the resulting random variable will have
standard normal distribution. We are going to use this fact to calculate (nonstandard) normal probabilities.
Calculating probabilities
With reference to the problem of IQ score, suppose we want tot find the
probability that an individuals IQ score is less than 85, i.e. P[X<85]. The
corresponding area under the pdf N(100,152) is shown in Figure 6 below.
2.5
2
f(x)
1.5
f(x)
P(X<85)
1
0.5
70
85
100
115
x F
Figure 6: area under the pdf N(100,152)
Figure 5: Density function f
(x)
We cannot use normal tables directly because these give N(0,1) probabilities.
Instead, we will convert the statement X<85 into an equivalent statement which
17
involves the standardized score, Z =
X 100
because we know it has a standard
15
normal distribution.
We start with X=85. To turn X into Z we must standardize the X, but to ensure
that we preserve the meaning of the statement we must treat the other side of the
inequality in exactly the same way. (Otherwise we will end up calculating the
probability of another statement, not X<85). Standardising both sides gives,
X 100 85 100
<
.
15
15
The left hand side is now a standard normal random variable and so we can call it
Z, and we have,
85 100
Z<
15
which is
Z < 1.
So, we have established that the statement we started with, X < 85 is equivalent to
Z < 1. This means that whenever an IQ score, X is less than 85 the
corresponding standardized score, Z will be less than 1 and so the probability we
are seeking, P[X<85] is the same P[Z < 1].
P[Z < 1] is just a standard normal probability and so we can look it up in Table
1 in the usual way, which gives 0.1587. We get that P[X < 85] = 0.1587.
This process of rewriting a probability statement about X, in terms of Z, is not
difficult if you are systematically writing down what you are doing at each stage.
We would lay out the working we have just done for P[X < 85] as follows.
X has a normal distribution with mean 100 and standard deviation 15. Let us find
the probability that X is less than 85.
X 100 85 100
P[ X < 85] = P
<
15
15
= P[Z 1] = 0.1587
Let us do some problems now.
Example 9: For each of these write down the equivalent standard normal
probability.
a) The number of people who visit a historical monument in a week is
normally distributed with a mean of 10,500 and a standard deviation of
600. Consider the probability that fewer than 9000 people visit in a week.
b) The number of cheques processed by a bank each day is normally
distributed with a mean of 30,100 and a standard deviation of 2450.
Consider the probability that the bank processes more than 32,000 cheques
in a day.
Solution: Here, we want to find the standard normal probability corresponding to
the probability P[X < 9000].
18
X 10500 9000 10500

<
a) We have P[ X < 9000] = P
= P[ Z < 2.5] .
600
600
b) Here, we want to find the standard normal probability corresponding to the
probability P[X > 32000].
X 30100 32000 30100
P[ X < 32000] = P
<
= P[ Z < 0.78]
2450
2450
Note: Probabilities like P[a < X < b] can be calculated in the same way. The only
difference is that when X is standardized, similar operations must be applied to
both a and b. that is, a < X < b becomes,
a X b
<
<
which is
a
b
<Z<
Example 10: An individuals IQ score has a N(100, 152) distribution. Find the
probability that an individuals IQ score is between 91 and 121.
Solution: We require P[91 < X < 121]. Standardising gives
91 100 X 100 121 100
P
<
<
15
15
15
The middle term is standardized normal random variable and so we have,
21
9
P < Z < = P[0.6 < Z < 1.4] = 0.9192 0.2743 = 0.6449 .
15
15
E1 If a random variable has the standard normal distribution , find the probability that it
will take on a value
a) Less than 1.50
b) Less than 1.20
c) Greater than 1.75
E2 A filling machine is set to pour 952 ml of oil into bottles. The amount to fill are
normally distributed with a mean of 952 ml and a standard deviation of 4 ml. Use the
standard normal table to find the probability that the bottle contains oil between 952 and
956 ml ?
1.4.4
Chi-Square Distribution
In the last subsection we have discussed normal distribution. The chi-square distribution
can be obtained from the normal distribution as follows. Suppose Z1,,Zn are n
independent identically distributed normal random variables with parameters 0 and 1,
then Z12 ++ Zn2 is said to have chi-square distribution with n degrees of freedom. The
degrees of freedom here basically indicates the number of independent components which
constitute the chi-square distribution. It has received several attention because of its
appearance in the constructing analysis of variable tables, contingency tables and for
obtaining the critical values of different testing procedure. Now we formally provide the
p.d.f of a chi-square random variable with n degrees of freedom.
19
If the random variable X has chi-square distribution with n-degrees of freedom,

then the p.d.f. of X is f(x)
f(x) =
n/2
1
e x / 2 x ( n / 2) 1
(n / 2)
0
if x >0
otherwise
here (.) is a gamma function and it is defined as
( a ) = x a 1e x dx
0
Although, the p.d.f of chi-square random variable is not a very nice looking one, do not
bother about that. Keep in mind that the shapes of density functions are always skewed. In
this case also if we want to compute P(a < X < b) for any a, b and n, explicitly it is not
possible to compute. Numerical integration is needed to compute this probability. Even
for chi-square distribution extensive tables of P(a < X < b) are available for different
values of a, b and n.
Note : We have a standard table corresponding to Chi- Square Distribution, many times
you may need to refer the values from the table. So the same is given at the end , method
of usage is similar to that discussed in Normal distribution.
EXAMPLE 9 Show that the moment generating function of a random variable X
which is chi-square distributed with v degrees of freedom is M(t) = (1 2t) v/2.
SOLUTION M (t ) = E (etx ) =
1
etx x ( v 2) / 2 e x / 2 dx
2v / 2 ( v/2) 0
1
= v/2
x ( v 2) / 2 e x (1 2t ) x / 2 dx
0
2 ( v/2)
Letting (1 2t ) x / 2 = u in the last integral we find
1
M (t ) = v / 2
2 ( v/2)
=
(1 2t ) v / 2
(v / 2)
2u
1 2t
( v 2) / 2
eu
2du
1 2t
u ( v / 2) 1e u du = (1 2t ) v / 2

E1 Let X1 and X2 be independent random variables, which are chi-square
distributed with v1 and v2 degrees of freedom respectively. Show that the moment
generating function of Z = X1 + X2 is (1 2t ) v ( v + v ) / 2
1
E2 Find the values of x2 for which the area of the right-hand tail of the x2
distribution is 0.05, if the number of degrees of freedom v is equal to (a) 15, (b)
21, (c) 50.
20
1.5 SUMMARY
In this unit we have covered following points:
a) A random variable is a variable that takes different values according to the chance
outcome
b) Types of random variables:
Discrete and Continuous
c) Probability distribution gives the probabilities with which the random variables
takes various values in their range
d) Discussed probability distributions :
a. Binomial Distribution : The probability of an event P[X=r] in this
distribution is given by
P(X = r) = C(n,r)pr(1-p)n-r = C(n,r)prqn-r,
b. Poisson Distribution : The probability of an event P[X=i] in this
distribution is given by
e i
,
i!
P( X = i) =
i = 0,1,2,3,...
c. Uniform Distribution : The probability density function is defined by

1
f ( x) = B A
0
A< x< B
for
otherwise
d. Normal Distribution : The probability for this distribution is determined

by calculating the area under the curve of probability density function
defined by
f ( x) =
1
2
( x )2
, where - < x <
2 2
e. Chi-Square Distribution : If the random variable X has chi-square

distribution with n-degrees of freedom, then the probability density
function of X is given by
f(x) =
n/2
1
e x / 2 x ( n / 2 ) 1
(n / 2)
0
if x >0
otherwise
here (.) is a gamma function and it is defined as
( a ) = x a 1e x dx
0
21
f.
Mathematical expectation or Expected values or Expectations

E(X) is defined as E(X) = x0p0+ x1p1+ x2p2+..=xipi
when all probabilities are equal then E(X) = mean of x0, x1 ,x2
Similarly for continuous variables X having density function p(x)
where P[X=x] = p(x) , the Expectation E(X) will be given by
integral of xip(xi) w.r.t x .
This concept of Expectation also contributes to the definition of
Moment Generating Function of X i.e Mx(t)= E(etx)
1.6 SOLUTIONS
E1 a) If X denotes the number of correct answers, then X is the random

variable for this situation
b) X can take the values 0,1,2,3..up to 50
c) P[X=10] means the probability that the number of correct answers is 10
E1 Case (a) is not discrete where as case (b) and (c) are discrete because in
case (a) we are taking values in an interval but in case(b) the number of
accident is finite , similarly you argue for case (c)
E1 This situation follows the binomial distribution with n=4 and p=90/100=9/10
The random variable X is the number of seeds that germinate. We have to
calculate the probability that exactly two of the four seedswill germinate. That is
P[X=2]. By applying Binomial formula , we get
P[X=2] = (4C2) * (9/10)2 *(1/10)2

= 6 * (81/100) * (1/100) = 486/10000 = 0.0486
So , the required probability is 0.0486
E1 Here we are dealing with the problem related to the receipt of bad Cheques,
which is an event with rare occurrence over an interval of time ( which is a day
In this case). So, we can apply Poisson distribution
Average bad Cheques received per day = 6

Thus by substituting = 6 and x=4 in Poisson formula we get
P[X=4] = (64 e-6)/4! = 0.135
E1 Suppose X denotes the time the fire alarm starts measured in minutes after 10:00 am.
Then clearly X is a uniform random variable between (0,180). If we want to find the
probability that fire alarm will ring before noon, then
(12 10) 60 2
= .
P (X 12 : 00 noon) =
180
3
22

E1
a) 0.9332
b) 0.1151
c) 0.9599
E2 The standard normal probability corresponding to this probability is given by
P[952 < Z < 956] = P[((952-952)/4) < ((X-952)/4) < ((952-956)/4)]

= P [0 < Z < 1]
= F(1) - F(0)
= 0.8413 0.5 = 0.343
E1 The moment generating function of Z = X1 + X2 is
M (t ) = E et ( X + X ) = E (etX ) E (etX ) = (1 2t ) v / 2 (1 2t ) v
1
/2
= (1 2t ) ( v + v ) / 2
1
using Example 9.
E2 Using the table in for Chi Square distribution we find in the column headed x2.95 the
values: (a) 25.0 corresponding to v = 15; (b) 32.7 corresponding to v = 21; (c) 67.5
corresponding to v = 50.
23
UNIT 2 PSEUDO RANDOM NUMBER GENERATION

2.0
2.1
2.2
2.3
2.4
2.5
2.6
2.7
Introduction
Objectives
Uniform random number generator
Generating random variates from
arbitrary distribution
Inverse Transform
Acceptance rejection method
Summary
Solutions
2.0 INTRODUCTION
A pseudo-random number generation is the methodology to develop algorithms and programs that can be used in, probability and
statistics applications when large quantities of random digits are needed. Most of these programs produce endless strings of singledigit numbers, usually in base 10, known as the decimal system. When large samples of pseudo-random numbers are taken, each of
the 10 digits in the set {0,1,2,3,4,5,6,7,8,9} occurs with equal frequency, even though they are not evenly distributed in the sequence.
Many algorithms have been developed in an attempt to produce truly random sequences of numbers, endless strings of digits in which
it is theoretically impossible to predict the next digit in the sequence based on the digits up to a given point. But the very existence of
the algorithm, no matter how sophisticated, means that the next digit can be predicted! This has given rise to the term pseudo-random
for such machine-generated strings of digits. They are equivalent to random-number sequences for most applications, but they are not
truly random according to the rigorous definition.
A simulation that has any random aspects at all, must involve sampling or generating random variables from different probability
distributions. These distributions are often specified, that is the form of the distribution functions is explicitly known, for example it
can be exponential, gamma, normal or Poisson as discussed in Unit 1.
Random number generation has intrigued scientists for several years and a lot of efforts has been spent on the creation of
randomness on a deterministic (non-random) machine, that is to design computer algorithms that are able to produce random
sequences of integers. This is not a trivial task. Such algorithms are called generators and all generators have flaws because all of
them construct the n-th number in the sequence as a function of the (n 1) th number, initialized with a non-random seed value.
Numerous techniques have been invented over the years that measure just how random a sequence is, and most well known generator
have subjected to rigorous testing. The mathematical tools that are required to design such an algorithm are largely number theoretic
and combinatorial in nature. These tools differ drastically from those needed when we want to generate sequences of integers with
certain non-uniform distributions given that a perfect uniform random number generator is available.
The methodology of generating random numbers has a long and interesting history. The earliest methods were essentially carried out
by hand, such as casting lots, throwing dice, dealing out cards or drawing numbered balls from a well-stirred urn. Many lotteries are
still operating in this way. In the early twentieth century statisticians joined gamblers in generating random numbers and mechanized
devices were built to generate random numbers more quickly. Some time later, electric circuits based on randomly pulsating vacuum
tubes were developed that delivered random digits at rates up to 50 per second. One such random number generator machine the
Electronic Random Number Indicator Equipment (ERNIE) was used by the British General Post Office to pick the winners in the
Premium Savings Bond lottery. Another electronic device was used by the Rand Corporation to generate a table of million random
digits. In India also Indian Statistical Institute has published a book just the collection of million random numbers in the mid twentieth
century which was used for sample survey planning.
As computers and simulation became more widely used, increasing attention was paid to methods of random number
generation compatible with the way computers work. A good uniform between 0 and 1, random generator should possess certain
properties as listed below:
Above all, the numbers produced should appear to be distributed uniformly on

[0, 1] and should not exhibit any correlation with each other; otherwise, the simulations results may be completely invalid.
From a practical point of view a generator should be fast and avoid the need for a lot of storage.
We should be able to produce a given stream of random numbers from a given initial (seed) value for at least two reasons. First
this can sometimes make debugging or verification of the computer program easier or we might want to use identical random
numbers in simulating different systems in order to obtain a more precise comparison.
In this unit we will describe how to generate U (0,1) (uniform between 0 and 1, see unit 1 for the actual definition) in a
1
computer. Once we have a U (0, 1) random number, we will use that to generate several other deviates from different discrete and
continuous distributions.
2. 1 OBJECTIVES
After reading this unit, you should know
how to generate U (0, 1) random number in a computer
how to generate random deviates from any discrete distribution
how to generate random numbers from many continuous distributions, like exponential, Weibull, gamma, normal, chi-square etc.
2.2 UNIFORM RANDOM NUMBER GENERATORS

As we had mentioned in the previous section that we need the uniform random number of generator for generating random numbers
from any other distributions. Therefore, it is very important to have a very good uniform random number generator. There are several
methods available to generate uniform random numbers. But currently the most popular one is the linear congruential generator
(LCG). Most of the existing softwares today use this LCGs proposed by Lehmer in the early 50s. The LCGs provides an algorithm
how to generate uniform random number between (0,1). It can be simply described as follows.
A sequence of integers Z1, Z2, is defined by the recursive formula
Zi = (aZi 1 + c) (mod m),
(1)
where m, the modulus, a, the multiplier, c, the increment and Z0, the seed or the initial value, are all non-negative integers. Those who
are not familiar with the definition of modules, note that for non-negative integers, x, y and z, x = y (mod z) means x is the remainder
when the integer y is divided the integer z. For example if y = 10 and z = 3, then x = 1, or if y = 10 and z = 2, then x = 0. Therefore,
from (1) it is clear that to obtain Zi, first divide aZi 1 + c by m and Zi is the corresponding remainder of this division. It is clear that 1
Zi m 1 and to obtain the desired random numbers Ui, for i = 1, 2. On [0, 1], we let Ui =
Zi
. The choice of the non-negative
m
integers, a, c and m are the most crucial steps, and they come from the theoretical considerations. In addition to non-negatively, they
also should satisfy 0 < m, a < m and c < m. Moreover, the initial value Z0 < m.
Immediately, two objections could be raised against LCGs. The first objection is one which is common to all random number
generators, namely the Zis defined by (1) are not really random. It can be easily seen that for i = 1, 2,.,
c(a i 1)
Zi = a i Z 0 +
(mod m),
1
a
so that every Zi is completely determined by m, a, c and Z0. However, by careful choice of these four parameters the aim is to induce a
behavior in the Zis that makes the corresponding Uis appear to be independent identically distributed U (0, 1) random variates when
subjected to variety to statistical tests.
The second objection to LCGs might be that the Uis can take on only the rational numbers 0,
1 2
( m 1)
,
, .,
; in fact
m m
m
the Uis might take only a fraction of these values, depending on the specifications of the four parameters. Therefore, there is no
possibility of getting a value of Ui between, say
0 .1
0 .9
0 .8
and
, whereas this should occur with probability
> 0. We will see later
m
m
m
that the modulus m is usually chosen to be very large, say 109 or more, so that the points in [0, 1] where Uis can fall are very dense.
This provides an accurate approximation to the true continuous U (0, 1) distribution, sufficient for most purposes.
Let us consider the following example:
Example 1 : Consider the LCG defined by m = 16, a = 5, c = 3 and Z0 = 7. The following table gives Zi and Ui (up to three decimal
places) for i = 1, ., 19. Note that Z17 = Z1 = 6, Z18 = Z2 = 1 and so on. Therefore, from i = 17, the sequence repeats itself. Naturally
we do not seriously suggest anybody to use this generator. The main reason here m is too small. This we are presenting just for
illustrative purpose.
Table 1
The LCG Zi = (5Z i 1 + 3) (mod 16) with Z0 = 7
2i
Zi
Ui
Zi
Ui
Zi
Ui
Zi
Ui
0
1
2
3
4
7
6
1
8
11
0.375
0.063
0.500
0.688
5
6
7
8
9
10
5
12
15
14
0.625
0.313
0.750
0.938
0.875
10
11
12
13
14
9
0
3
2
13
0.563
0.000
0.188
0.125
0.813
15
16
17
18
19
4
7
6
1
8
0.250
0.438
0.375
0.063
0.500
Note that the repeating behaviour of LCG is inevitable. By the definition of Zi, whenever it takes on a value it had taken previously,
from that point onward the sequence will repeat itself endlessly. The length of a cycle is called the period of a generator. For LCG, Zi
depends only on the previous value Zi 1 and since 0 Zi m 1, it is clear that the period is at most m. If it m, the LCG is said to
have full period. Clearly, if a generator is full period, any choice of the initial seed Z0 from {0,,m-1} will produce the entire cycle in
some order.
Since for large scale simulation projects may require hundreds of thousands of random numbers, it is desirable to have LCGs
with long periods. Moreover, it is desirable to have full period LCGs, because then it is assured to have every integer between 0 and m
1 exactly once in every cycle. Thus it is very important to know how to choose a, m and c so that the corresponding LCG will have
full period. We should also keep in mind that obtaining full period is just one desirable property for a good LCG. It should also have
good statistical properties, such as apparent independence, computational and storage efficiency and reproducibility. Reproducibility
is simple. For reproducibility, we must only remember that the initial seed Z0 initiates the generator with this value again to obtain the
same sequence of Ui exactly. Some of the standard LCGs with different values of a, m and c are presented below. These LCG have
been observed to perform very well in several machines and passed the standard statistical tests also.
Generator 1: a = 16807,
Generator 2: a = 1664525,
m = 231 1,
m = 232,
c = 0.
c = 1664525.
Fortunately today, most of the simulation packages and even simple scientific calculators have reliable U (0 1,) generator available.
E1 What do you mean by Pseudo random number generation? What is the practical advantage of the concept of random number
generation? Do you know any algorithm which works in designing the software for Random number generation?
2.3 GENERATING RANDOM VARIATES FROM ARBITRARY

DISTRIBUTIONS
A simulation that has any random aspects at all must involve generating random variates from different distributions. We usually use
the phrase generating a random variate to refer the activity of obtaining an observation or a realization on a random variable from the
desired distribution. These distributions are often specified as a result of fitting some appropriate distributional form. They are often
specified in advance, for example exponential, gamma or Poisson etc. In this section we assume that the distributional form is already
been specified including the values of the parameters and we address the issue of how we can generate random variate from this
specified distribution.
We will see in this section that the basic ingredient needed for every method of generating random variates from any
distribution is a source of independent identically distributed U (0, 1) random variates. Therefore, it is essential that a statistically
reliable U (0, 1) generator is available for generating random deviate correctly from any other distribution.Therefore, from now on
we assume that we have a reliable sequence of U(0,1) variates available to us.

There are several issues which should be kept in mind before using any particular generator. The most important issue is of
course the exactness. It is important that one should use an algorithm that results in random variates with exactly the desired
distribution, within the unavoidable external limitations of machine accuracy and the exactness of U (0,1) random number generator.
The second important issue is efficiency. Given that we have several choices, we should choose that algorithm which is efficient in
terms of both storage space and execution time. Now we provide some of the most popular and standard techniques to generator nonuniform random deviates, it may be both discrete or continuous and even mixed distribution also.
2.4 INVERSE TRANSFORM

Suppose we want to generate a random variate X that has a continuous and strictly increasing distribution function F, when 0 < F (x) <
1, i.e., whenever x1 < x2 then F (x1) < F (x2). We draw a curve of F in the Figure 1. Let F1 denote the inverse of the function F. Then
an algorithm for generating a random variate X having distribution function F is as follow .
ALGORITHM
Step 1: Generate U1 from U1 (0,1)

Step 2: Return X1 = F1 (U1).
1
0.9
U1
0.8
0.7
0.6
0.5
F(x)
0.4
0.3
0.2
0.1
-2
-1
X1
Figure 1: Distribution function of a continuous a random variable.
Note that when F is a strictly increasing function, F1 (U) will be always defined, since 0 U 1 and the range of F is [0, 1]. Figure 1
illustrates the algorithm graphically. According to the figure it is clear that the uniform random variable U1 results the random
variable X1 with the distribution function F. To show that the value X1 returned by the above algorithm has the desired distribution
function F, note that
P (X1 x) = P (F1 (U1 ) x) = P (U1 F (x)) = F (x).
The first equality follows from the definition of X1, the second equality follows because F is invertible and the third equality follows
because U1 follows U (0,1).
EXAMPLE 2: Let X have the Weibull distribution with the following probability density function:
e x x 1 if
f(x) =
0
if
x>0
-1
Find F
x<0
SOLUTION Here and both are known constants and both of them are strictly greater than 0.
Therefore, X has the distribution function
1 e x
0
F (x) =
if
if
x>0
x0
Therefore, to compute F1 (u), let us equate u = F (x) and we solve for x to obtain
F
(u) = [
{ In (1 u)} ]
Therefore to generate X from a Weibull distribution with = 2 and = 1, generate U from U (0, 1) and then set
X = [{ In (1 u)}]
1
2
In this case also as before, it is possible to replace U by 1 U, therefore we can use

X = [{ In u}]
1
2
to avoid one subtraction.

The inverse-transform method can be used also when the random variable X is discrete.
In this case the distribution function is
4
F (x) = P (X x) =
p( x ),
xi x
Where p(xi) is the probability mass function of X, i.e.,

p (xi) = P (X = xi).
We can assume that X can take values x1, x2,., where x1 < x2 < . Then the algorithm is as follows:
ALGORITHM
Step 1: Generate U from U (0,1)
Step 2: Determine the smallest positive integer I such that U F (x1) and return
X = xI . The following figure 2 illustrates the method. In that case we generate X=x4
Now to show that the discrete inverse transform method is valid, we need to show that P (X = xi) = p (xi) for all i = 1, we get X = x1, if
and only if U F (x1) = p (x1), since xis are in the increasing order and U follows U (0,1). For i 2, the algorithm sets X = xi if and
only if F (xi 1) < U F (xi), since the i chosen by the algorithm is the smallest positive integer such that U F (xi). Further, since U
follows U (0,1) and 0 F (xi 1) < F (xi) 1,
P (X = xi) = P {F (xi 1 < U F (xi)} = F (xi) F (xi 1) = p (xi).
Now consider the following example.
F(x)
x1
x2
x
x3
x4
X
x5
Figure 2: Distribution function of a discrete random variable

EXAMPLE 3: Suppose we want to generate a random sample from the following discrete probability distribution.
P (X = 1) =
1
1
1
, P (X = 2) = , P (X = 4) = . Generate a random sample from X?
2
4
4
SOLUTION
0
1
2
F (x) = 3
4
1
The distribution function of the random variable X is
if
x <1
if
1 x < 2
if
2 x<4
if
x4
The distribution function is X is presented in the Figure 3. If we want to generate a random sample from X, first generate a random
variable U from U (0, 1). If U
F(x)
1
1
3
, then assign X = 1. If < U , then assign X = 2, otherwise assign X = 4.
2
2
4
0.8
0.6
Figure 3: Distribution function of the random variable X of example 3.
E1 Let X have the exponential distribution with mean 1. The distribution function is
1 e x if x 0
Find F-1
if x < 0
0
(x)
E2 Consider another discrete random variable which may take infinitely many values. Suppose the random variable X has the
following probability mass function.
1
, i = 1, 2, 3, . Generate a random sample from X?
2i
P (X = i) = pi =
2. 5 ACCEPTANCE-REJECTION METHOD
In the last subsection we have discussed the inverse transformation method to generate random number from different non-uniform
distributions. Note that apparently, the inverse transformation method seems to be the most general method to generate random
deviate from any distribution function functions. In fact it can be used provided the distribution function can be written in an explicit
form, or more precisely the inverse of the distribution function can be computed analytically. For example, in case of exponential,
Weibull, Cauchy distributions the distribution function and also their inverses can be constructed analytically. Unfortunately that is
not the case in general. Suppose the random variable X follows gamma with the shape and scale parameters as and respectively.
Then the density function of X, say
fx (x | , ), is
1 x
x e
if
f (x | , ) =
0
if
The distribution function X, say F ( x | , ) =
x>0
x0
f ( y | , )dy can not be expressed in explicit form. Therefore, F1 (x | , ) also can
not be calculated explicitly. Exactly the same problem arises if X is a normal random variable. Suppose X is a normal random variable
with mean and variance 2, then the probability density function of X, say f (x | , ) is
f (x | , ) =
1
2
( x )2
2 2
, for < x < .
In this case also the distribution function can not be computed analytically and similarly its inverse also. Therefore in these cases we
can not apply the inverse transformation method to generate the corresponding random deviates. The acceptance-rejection method can
be used quite effectively to generate these random deviates. It can be described as follows.
Suppose we have a density function f (x) and we want to generate a random deviate from the density function f (x). The
distribution function of f (x) can not be expressed in explicit form. The acceptance-rejection method requires that we specify a
function g (x) that majorizes the function f (x), that is, f (x) g (x) for all x. Naturally, g (x) will not be a density function always, since
c=
g ( x)dx f ( x) = 1,
but the function h (x) =
1
g ( x) is clearly a density function provided c < . Now for any given f (x), we choose the function g (x),
c
such that c < and it is possible to generate random deviate from g (x) by a inverse transformation method. Then we can generate the
random deviate X from f (x) as follows:
ALGORITHM
Step 1: Generate Y having density function g (x).
Step 2: Generate U from U (0,1) which is independent of Y.
Step 3: If U
f (Y )
, X = Y, otherwise go back to Step 1 and try again.
g (Y )
Note that the algorithm is looping back to Step 1 until we generate a pair (Y, U) pairs in Steps 1 and 2 for which U
f (Y )
, when we
g (Y )
accept the value Y for X. Theoretically it can be shown that the random deviate X generated by the above algorithm has indeed the
probability density function f (x). Since it is not very easy to prove the result we do not provide it.
EXAMPLE 4: Consider the following density function
60 x 3 (1 x) 2 if
f (x) =
0
0 x 1
otherwise
In this case the distribution function of f (x) is
0
if
x<0
F (x) = 10 x 6 + 15 x 4 24 x 5 if 0 < x < 1
1
if
x >1
0.8
0.6
0.4
x)
0.2
0
0
0.2
0.4
0.6
0.8
Figure 4 : Distribution function of the density function f (x)
2.5
1.5
)
0.5
0
7
From the distribution function F (x) is provided in figure 4 .It is clear that the distribution function F (x) is a strictly increasing
function of x in [0, 1]. Therefore F1 (x) exists, but unfortunately to find F1 (x) we need to solve a six degree polynomial, which can
not be obtained analytically. We need to solve numerically only. Therefore, we can not generate random deviate from the density
function f (x) using the inversion method. But we will be able to generate random deviate from f (x) using the acceptance-rejection
method.
First let us look at the graph of the density function f (x). It is presented in the Figure 5. From the Figure 5 it is clear that f (x) is an
unimodal function with a unique maximum. The maximum can be easily obtained by the standard differential calculus, that is by
setting
df ( x)
= 0. We see that the maximum of f (x) occurs at x = 0.6 and the maximum value at 0.6, that is f (0.6) = 2.0736.
dx
Therefore, if we define
2.0736 if
0
g (x) =
0 < x <1
otherwise.
then clearly f (x) g (x) for all x. Now to calculate h (x), first we need to calculate c. Since,
1
c=
2.0735 = 2.0736, therefore,

0
2.0736
= 1 if 0 < x < 1
h (x) = c
0
otherwise.
It is immediate that h (x) is just the U(0,1) density function. Now the algorithm takes the following form.
- Step 1: Generate Y from U (0, 1)

- Step 2: Generate U from U (0,1) which is independent of Y.
- Step 3: If U
60Y 3 (1 Y ) 2
then return with X = Y,otherwise go back to Step 1.
2.0736
In this case X has the desired density function f (x).

E1 use acceptance-rejection method to generate a random deviate from gamma density function, . The gamma density function with
the shape parameter can be written as
1 1 x
x
e
f (x) = ( )
if
x>0
otherwise
2. 6 SUMMARY
In this unit we have discussed the meaning of pseudo random number generation and along with that we have described
uniform random number generator and arbitrary random number generator. Under Uniform random number generator case
we have emphasized on LCG (Linear Congruential Generator) and some objections related to LCG. Actually LCGs
provides an algorithm how to generate uniform random number between (0,1). It can be simply described as follows.
A sequence of integers Z1, Z2, is defined by the recursive formula
8
Zi = (aZi 1 + c) (mod m),
(1)
where m, the modulus, a, the multiplier, c, the increment and Z0, the seed or the initial value, are all non-negative integers
Then we have discussed the concept , algorithm and application of Inverse transforms for random number generation .
In brief Suppose we want to generate a random variate X that has a continuous and strictly increasing distribution function
F, when 0 < F (x) < 1, i.e., whenever x1 < x2 then F (x1) < F (x2). Let F1 denote the inverse of the function F. Then an
algorithm for generating a random variate of X having distribution function F is as follow .
ALGORITHM
Step 1: Generate U1 from U1 (0,1)
Step 2: Return X1 = F1 (U1).
The inverse-transform method can be used also when the random variable X is discrete.
In this case the distribution function is
p( x ),
F (x) = P (X x) =
xi x
Where p(xi) is the probability mass function of X, i.e.,

p (xi) = P (X = xi).
We can assume that X can take values x1, x2,., where x1 < x2 < . Then the algorithm is as follows:
ALGORITHM
Step 1: Generate U from U (0,1)
Step 2: Determine the smallest positive integer I such that U F (x1) and return
X = xI . The following figure 2 illustrates the method. In that case we generate X=x4
Note:
the inverse transformation method to generate random number from different non-uniform distributions. Note that apparently,
the inverse transformation method seems to be the most general method to generate random deviate from any distribution
function functions. In fact it can be used provided the distribution function can be written in an explicit form
Finally we had discussed the Acceptance Rejection method of random number generation, this method is quite important because
,when the distribution function can not be computed analytically and similarly its inverse also. Then in such cases we can not apply
the inverse transformation method to generate the corresponding random deviates. The acceptance-rejection method can be used quite
effectively to generate these random deviates.
In brief ,Suppose we have a density function f (x) and we want to generate a random deviate from the density function f (x). The
distribution function of f (x) can not be expressed in explicit form. The acceptance-rejection method requires that we specify a
function g (x) that majorizes the function f (x), that is, f (x) g (x) for all x. Naturally, g (x) will not be a density function always, since
c=
g ( x)dx f ( x) = 1,
but the function h (x) =
1
g ( x) is clearly a density function provided c < . Now for any given f (x), we choose the function g (x),
c
such that c < and it is possible to generate random deviate from g (x) by a inverse transformation method. Then we can generate the
random deviate X from f (x) as follows:
ALGORITHM
Step 1: Generate Y having density function g (x).
Step 2: Generate U from U (0,1) which is independent of Y.
Step 3: If U
f (Y )
, X = Y, otherwise go back to Step 1 and try again.
g (Y )
Note that the algorithm is looping back to Step 1 until we generate a pair (Y, U) pairs in Steps 1 and 2 for which U
f (Y )
, when we
g (Y )
accept the value Y for X. Theoretically it can be shown that the random deviate X generated by the above algorithm has indeed the
probability density function f (x). Since it is not very easy to prove the result we do not provide it.
9
2. 7 SOLUTIONS
E1 A pseudo-random number generation is the methodology to develop algorithms and programs that can be used in,
probability and statistics applications when large quantities of random digits are needed. Most of these programs produce
endless strings of single-digit numbers, usually in base 10, known as the decimal system. When large samples of pseudorandom numbers are taken, each of the 10 digits in the set {0,1,2,3,4,5,6,7,8,9} occurs with equal frequency, even though
they are not evenly distributed in the sequence.
Many algorithms have been developed in an attempt to produce truly random sequences of numbers, endless strings of
digits in which it is theoretically impossible to predict the next digit in the sequence based on the digits up to a given point.
But the very existence of the algorithm, no matter how sophisticated, means that the next digit can be predicted! This has
given rise to the term pseudo-random for such machine-generated strings of digits. They are equivalent to random-number
sequences for most applications, but they are not truly random according to the rigorous definition.
There are several methods available to generate uniform random numbers. But currently the most popular one is the linear
congruential generator (LCG). Most of the existing softwares today use this LCGs proposed by Lehmer in the early 50s.
E1 To find F-1, we set u = F(x) and solve for x to obtain

x = F 1(u ) = ln(1 u ).
Therefore, to generate random variate X from exponential distribution with mean 1, first generate U from a U (0,1) and
then let X = - In (1 U). Therefore X will have exponential distribution with mean 1. Since U and 1 U have the same U
(0, 1) distribution, we can also use X = In U. This saves a subtraction.
E2 Note that
2
i =1
= 1, therefore pi denotes the probability mass function of a discrete random variable. The
corresponding distribution function of the discrete random variable X can be written as
F (x) = m 1
i =1 2i
if
x <1
if
m x < m + 1,
where m is any positive integer. Now to generate a random deviate from the random variable X, first draw a random
m 1
m
0
1
1
1
sample U from U (0, 1), since 0 U 1, there exits a positive integer m such that i U < i , where i = 0,
i =1 2
i =1 2
i =1 2
then X = m,
E1 Our problem is to generate random deviate from f (x) for a given 0 < < 1. Note that we can not use the acceptancerejection method in this case. It is easily observed if we take
0
if
x0
x 1
if 0 < x < 1
g (x) =
( )
e x
if
x > 1,
( )
then f (x) g (x) for all x. In this case
c=
10
g (x) =
x 1
0 ( ) dx +
1
1 (e + a )
ex
dx =
.
( )
( ) ae
1
g (x) is
c
0
if
x0
1
x
h (x) =
if 0 x 1
b
e x
if
x > 1,
b
e+a
where b =
. The distribution function H (x) corresponds to the density function h (x) is
e
Therefore, h (x) =
x
b
H (x) = h (y) dy =
x
0
1 e
b
Which can be easily inverted as
if
0 x 1
if
x > 1,
(
bu
)
if u
b
H 1 (u) =
In b(1 u ) if u > 1
Therefore, it is possible to generate random deviate from the density function h (x) using the simple inversion method.
Generation of a random deviate Y from the density function h (x) can be performed as follows. First generate U1 from U (0,
1
1
b (1 U1 )
1), if U1 we set Y = (bU1) , in this case Y 1. Otherwise Y = In
and in this case Y > 1. Also note that
b
Y
if 0 Y 1
f (Y ) e
= 1
g (Y ) Y
if
Y >1
Now the algorithm to generate random deviate from a gamma density function with the shape parameter , for 0 < < 1
takes the following form:
Step 1: Generate U1 from U (0,1) and let P = bU1. If P > 1, go to Step 3 otherwise proceed to Step 2.
Step 2: Let Y = P and generate U2 from U (0, 1). If U2 e Y, return X = Y.

Otherwise go back to Step 1.
Step 3: Let Y = In
(b P )
and generate U2 from U (0, 1). If U2 Y 1, return X = Y, otherwise go back to Step 1.
11
Regression Analysis
UNIT 3 REGRESSION ANALYSIS

Structure
3.0
3.1
3.2
Introduction
Objectives
Simple Linear Regression
3.2.1
3.2.2
3.2.3
3.3
3.0
Least Squares Estimation

Goodness of Fit
Residual Analysis
Non-Linear Regression
3.3.1
3.4
3.5
Page Nos
Summary
Solutions
INTRODUCTION
In many problems there are two or more variables that are inherently related and it may be
necessary to explore the nature of their relationship. Regression analysis is a statistical
technique for modeling and investigating the relationship between two or more variables.
For example in a chemical process suppose that the yield of the product is related to the
process operating temperature. Regression analysis can be used to build a model that
expresses yield as a function of temperature. This model can be used to predict yield at a
given temperature level. It can also be used for process optimization or process control
purposes.
In general, suppose that there is a single dependent variable or response variable y and
that is related to k independent or regressor variables say x1,..,xk. The response
variable y is a random variable and the regressor variables x1,,xk are measured with
negligible error. The relationship between y and x1,..xk is characterized by a
mathematical model and it is known as the regression model. It is also known as the
regression of y on x1,..,xk. This regression model is fitted to a set of data. In many
situations the experimenter knows the exact form of the functional relationship
between y and x1, ..., xk, say (x1, ..., xk), except for a set of unknown parameters.
When the functional form is unknown, it has to be approximated on the basis of
past experience or from the existing information. Because of its tractability, a
polynomial function is popular in the literature.
In this unit we will be mainly discussing the linear regression model and when k = 1, that
is only one regressor variables. We will be discussing in details how to estimate the
regression line and how it can be used for prediction purposes from a given set of data.
We will also discuss briefly how we can estimate the function , if it is not linear.
3.1 OBJECTIVES
After reading this unit, you should be able to
Decide how two variables are related.
Measure the strength of the linear relationship between two variables.
Calculate a regression line that allows to predict the value of one of the variable if
1
the value of the other variable is known.

Analyze the data by the method of least squares to determine the estimated
regression line to be used for prediction.
Apply the least squares methods to fit different curves and use it for prediction
purposes.
3.2 SIMPLE LINEAR REGRESSION

We wish to determine the relationship between a single regressor variable x and a
response variable y(note: The linear regression with one independent variable is referred
to as simple linear regression ). We will refer to y as the dependent variable or response
and x as the independent variable or regressor. The regressor variable x is assumed to be a
continuous variable controlled by the experimenter. You know that it is often easy to
understand data through a graph. So, let us plot the data on Scatter diagram (a set of
points in a 2-D graph where horizontal axis is regressor and vertical axis is response)
Suppose that the true relationship between y and x is straight line. Therefore, each
observation y can be described by the following mathematical relation (model)
y = 0 + 1 x +
(1)
Yield (y)
80
100
120
140
160
180
200
220
Temperature (x)
Figure 1: Scatter diagram of yield versus temperature
where is a random variable with mean 0 and variance 2. The is known as the error
component and it is assumed to be small. If the error was absent then it was a perfect
relation between the variables y and x which may not be very practical. Let us look at the
following example.
Example 1: A chemical engineer is investigating the effect of process operating
temperature on product yield. The study results in the following data.
Temperature C (x)
Yield,%(y)
Regression Analysis
100 110 120 130 140 150 160 170 180 190
45
51
54
61 66 70
74
78
85
89
The scatter diagram between the temperature and the yield is presented in the Figure 1
above. From the Figure 1 it is clear that there is a linear relationship between yield and
temperature but clearly it is not perfect. For example we can not write the relationship
between y and x as follows
y = 0 + 1 x
Clearly the presence of the error is needed. Moreover the error is a random variable
because it is not fixed and it varies from one temperature to another. It may also vary
when two observations are taken at the same temperature. If there was a perfect
linear relationship between y and x we would have required just two points to find the
relationship. Since the relationship is not perfectly linear it is usually required much more
than two data points to find their relationship. Our main objective is to find the
relationship between them from the existing information (data points). Since it is assumed
that the relationship between x and y is linear therefore the relationship can be expressed
by the equation (1) and finding the relationship basically boils down finding the unknown
constants 0 and 1 from the observations.
Let us discuss this concept of linear regression by one more illustration/collection of data
described in the table 1 given below. This table encloses the data of 25 samples of
cement, for each sample we have a pair of observation (x,y) where x is percentage of SO3,
a chemical and y is the setting time in minutes. These two components are strongly
related; it is the percentage of SO3 which influences the setting time of any cement
sample, the recorded observations are given in table 1 below.
Table 1: Data on SO3 and Setting Time

S.No.
i
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
Percentage of SO3
x
1.84
1.91
1.90
1.66
1.48
1.26
1.21
1.32
2.11
0.94
2.25
0.96
1.71
2.35
1.64
1.19
1.56
1.53
Setting Time
Y (in minutes)
190
192
210
194
170
160
143
164
200
136
206
138
185
210
178
170
160
160
19
20
21
22
23
24
25
Total
Sum of
Squares
0.96
1.7
1.68
1.28
1.35
1.49
1.78
39.04
140
168
152
160
116
145
170
4217
64.446
726539
From the table 1, you see that setting time y increases as percentage of SO3
increases. Whenever you find this type of increasing (or decreasing) trend in a
table , same will be reflected in the scatter diagram ,and it indicates that there is a
linear relationship between x and y. By drawing the scatter diagram you can
observe that the relationship is not perfect in the sense that a straight line cannot
be drawn through all the points in the scatter diagram.
Nevertheless, we may approximate it with some linear equation. What formula
shall we use? Suppose, we use the formula y = 90 + 50x to predict y based on x.
To examine how good this formula is, we need to compare the actual values of y
with the corresponding predicted values. When x = 0.96, the predicted y is equal
to 138(= 90 + 50 x 0.96). Let (xi, yi) denote the values of (x, y) for the ith sample.
From Table-1, notice that x12=x19=0.96, whereas y12 = 138 and y19 = 140.
Let $y = 90 + 50 xi . That is, $yi is the predicted value of y (then using y = 90 + 50x
for the ith sample. Since, x12=x19=0.96, both $y12 and $y19 are equal to 138. Thus the
difference e$i = yi $yi , the error in prediction, also called residual is observed to
be e$ 12 = 0 and e$ 19 = 2. The formula we have considered above, y = 90 + 50x, is
called a simple linear regression equation , we will study these terms in detail in
our successive sections.
3.2.1
Least squares estimation
Suppose that we have n pairs of observations, say (x1 , y1),.(xn , yn). It is assumed that
the observed yi and xi satisfy a linear relation as given in the model (1). These data can be
used to estimate the unknown parameters 0 and 1 . The method we are going to use is
known as the method of least squares, that is, we will estimate 0 and 1 so that the sum
of squares of the deviations from the observations to the regression line is minimum. We
will try to explain it first using a graphical method in Figure 2. For illustrative purposes
we are just taking 5 data points (x, y) = (0.5, 57), (0.75, 64), (1.00, 59), (1.25, 68),
(1.50,74). The estimated regression line can be obtained as follows. For any line we have
calculated the sum of the differences (vertical distances) squares between the y value and
the value, which is obtained using that particular line. Now the estimated regression line
is that line for which the sum of these differences squares is minimum.
90
Regression Analysis
85
80
Estimated regression line

75
70
Observed point
Difference
65
60
Estimated point
55
0
0.5
1.5
Figure 2 Differences between y values and the estimated values using regression line
Matematically the sum of squares of the deviations of the observations from the
regression line is given by
n
L=
i2 =
i =1
(y
i
1 xi ) 2
i =1
The least squares estimators of 0 and 1 , are 0 and 1 which can be obtained by
solving the following two linear equations.
n
L
= 2 ( yi 0 1 xi ) = 0
0
i =1
n
L
= 2 ( y i 0 1 xi ) = 0
1
i =1
Simplifying these two equations yields

n 0 + 1
xi =
i =1
(2)
i =1
i =1
i =1
0 xi + 1 xi2
y x
(3)
i i
i =1
Solving (2) and (3) 0 and 1 can be obtained as follows:
0 = y 1 x
1 =
n
yx
i =1 i i
1
n
x2
i =1 i
(4)
( yi )( x )
1
( xi )
n
n
n
i =1 i
i =1
(5)
i =1
where y =
y and x = x
i
i =1
. Therefore, the fitted simple linear regression line
i =1
between these n points is

y = 0 + 1 x
(6)
Note : Linear Correlation and Regression

linear correlation and regression are very similar. One uses the correlation
coefficient to determine whether two variables are linearly related or not. The
correlation coefficient measures the strength of the linear relationship. Regression
on the other hand is used when we want to answer question about the relationship
between two variables.
Some Properties for Linear Correlation and Regression
(1) The line of regression of y on x always passes through( x, y ) where
x, and y are the mean of x and y values.

(2) There are always two line of regression one of y on x and other of x on y.
i.e., y = a 1 + b yx x or x = a 2 + b xy y
y
x
x
bxy = Regression coeff of y on x = r
y
where byx = Regression coeff of y on x = r
Correlation can be obtained by the following formula also,

r = bxy * byx (-1 r 1)
Angle between lines of regression is given by,
r 2 1 x * y
2
2
r x + y
= tan 1
Where r = correlation coeff between x and y

x = standard deviation of variable x
y = standard deviation of variable y
So, now, Regression equation of y on x can be rewritten as
( y y) = r
y
( x x)
x
And Regression equation of x on y as,

( x x) = r
x
( y y)
y
Example 1 (contd.) Now we will compute the estimates of 0 and 1 for the data points
given in Example 1. In this case it is observed
10
n = 10,
i =1
xi = 1450,
10
yi = 673,
i =1
x = 145,
y = 67.3
10
xi2 = 218,500,
i =1
10
10
x y
yi2 = 47, 225,
i i
i =1
Therefore,
1 =
Regression Analysis
= 101,570
i =1
101,570 10 1450 673

= 0.483,
218,500 10 14502
and
0 = 673 0.483 X 145 = 2.739.

The fitted simple linear regression line through those 10 points is
y = 2.739 + 0.483x
(7)
The best fitted line along with the data points are plotted in the Figure 3. Note that the
best fitted line can be used effectively for prediction purposes also. For example suppose
we want to know the expected yield when the temperature is 170C, for which the data is
not available. We can use the (7) for this purpose as follows.
y = 2.739 + 0.483 170 = 79.371 .
Therefore, the best fitted line shows that the expected yield at 170C is 79.371.
100
90
Best fitted line
80
70
Yield (y)
60
50
40
80
100
120
140
160
180
200
220
Temperature (x)
Figure 3: Data points and the best fitted regression line passing through these points
Soon in this section only, we will discuss the technique consisted of few steps; which can
be used to fit a line in best way, such that the error is minimum. In crux we will study the
technique to determine the best equation, that can fit a line in the data such that the error
is minimum. But before that lets see one more example.
7
Example 2: A survey was conducted to relate the time required to deliver a proper
presentation on a topic , to the performance of the student with the scores he/she receives.
The following Table shows the matched data:
Table 2
Hours (x)
Score (y)
0.5
57
0.75
64
1.00
59
1.25
68
1.50
74
1.75
76
2.00
79
2.25
83
2.50
85
2.75
86
3.00
88
3.25
89
3.50
90
3.75
94
4.00
96
(1) Find the regression equation that will predict a students score if we know how many
hours the student studied.
(2) If a student had studied 0.85 hours, what is the students predicted score?
Solution. We will arrange the data in the form of a chart to enable us to perform the
computations easily.
x
0.5
0.75
1.00
1.25
1.50
1.75
2.00
2.25
2.50
2.75
3.00
3.25
3.50
3.75
4.00
33.75
y
57
64
59
68
74
76
79
83
85
86
88
89
90
94
96
1188
Table 3
x2
0.25
0.56
1.00
1.56
2.25
3.06
4.00
5.06
6.25
7.56
9.00
10.56
12.25
14.06
16.00
93.44
xy
28.5
48.0
59.0
85.0
111.0
133.0
158.0
186.75
212.5
236.5
246.0
289.25
315.0
352.50
384.0
2863
In this case n = 15, therefore
15 2863 33.75 1188

1
= 10.857 , 0 = [1188 10.857 33.75] = 54.772
2
15
15 93.44 33.75
Therefore, the prediction line becomes:
Regression Analysis
y = 54.772 + 10.857 x
Now the predicted score when x = 0.85, is
y = 54.772 + 10.857 0.85 = 64.00

100
95
90
85
80
75
Score (y)
70
65
Predicted line
60
55
50
Hours (x)
Figure 4: Hours studied and the corresponding score with the best fitted regression line passing
through these points
Thus the predicted score of the student who had studied 0.85 hours is approximately
64.00.
We have plotted the individual scores and the hours studied with the best fitted prediction
line in the Figure 4. It shows the hours studied by the student and the corresponding score
follows a linear pattern and the predicted line can be used quite effectively to predict the
score of a student if we know how many hours the student had studied.
Now, its the time to discuss the technique for determining the best equation, i.e., the
equation which fits the line in a way that the overall error is minimized.
From above illustrations and examples you might have noticed that different
equations give us different residuals. What will be the best equation? Obviously,
$i are small.
the choice will be that equation for which es
$i zero
This means that whatever straight line we use, it is not possible to make all es
, where e$i = yi $yi (the difference). However, we would expect that the errors are
positive in some cases and negative in the other cases so that, on the whole, their
sum is close to zero. So, our job is to find out the best values of 0 and 1 in the
formula y = 0 + 1 x + e ( s.t. e 0) . Let us see how we do this.
$i are minimum. For

Now our aim is to find the values 0 and 1 so that the error es
that we state here four steps to be done.
1) Calculate a sum Sxx defined by
n
Sxx = x 2i n x
(8)
i =1
where xis are given value of the data and x =
xi
is the mean of the observed
n
values and n is the sample size.

The sum Sxx is called the corrected sum of squares.
2) Calculate a sum Sxy defined by
n
Sxy = xi yi n x y
(9)
i =1
where xis and yis are the x-values and y-values given by the data and x and
y are their means.
Sxy
3) Calculate
= 1 say. That is
Sxx
Sxy
1 =
(10)
Sxx
4) Find y 1 x = 0 , say.
Let us now compute these values of the data in Table 1: Data on SO3 and Setting
Time, we get
x = 1.5616, y = 168.68, Sxx = 3.4811, and Sxy = 191.2328.
Substituting these values in (10) and (11), we get
Sxy
1 =
= 54.943 and 0 = 168.68 54.943 x 1.5616 = 82.88
(11)
Sxx
Therefore, the best linear prediction formula is given by
y = 82.88 + 54.943x.
After drawing this line on the scatter diagram, you can find that this straight lines
is close to more points, and hence it is the best linear prediction.
Example 3: A hosiery mill wants to estimate how its monthly costs are related to its
monthly output rate. For that the firm collects a data regarding its costs and output for a
sample of nine months as given in the following table.
Table 4
10
Output (tons)
Production cost (thousands of dollars)
1
2
4
8
6
5
8
9
7
2
3
4
7
6
5
8
8
6
Regression Analysis
(a) Find the scatter diagram for the data given above.
(b) Find the regression equation when the monthly output is the dependent variable (x)
and monthly cost is the independent variable (y).
(c) Use this regression line to predict the firms monthly cost if they decide to produce 4
tons per month.
(d) Use this regression line to predict the firms monthly cost if they decide to produce 9
tons per month.
Solution
a) Suppose that xi denote the output for the ith month and yi denote the cost
for the ith month. Then we can plot the graph for the pair (xi, yi) of the
values given in Table . Then we get the scatter diagram as shown in Figure
below.
C
O
S
T
(T
H
O
U
S
A
N
D
S
OF
10
9
8
D 3
O
L 2
L
A 1
R)
0
1
4
5
6
7
OUTPUT (TONS)
Figure 5: Scatter Diagram
10
b) Now to find the least square regression line, we first calculate the sums Sxx
and Sxy from Eqn.(8) and (9).
n
Sxx = x 2i nx
i =1
Note that from Table(4) we get that

n
x
50
x = i =1 i =
n
9
11
i =1
y
i = 49
n
9
2
x
i
=
340
y=
y = 303
and x y = 319
2
i i
Therefore, we get that

n
xi yi nx y
1 = i =n1
x 2 nx
1
i =1
9 x 319 50 x 49
9 x 340 502
421
=
= 0.752
560
Correspondingly, we get
49
50
0 = (0.752) x
9
9
= 1.266
Therefore, the best linear regression line is
y = 1.266 + (0.752)x
c) If the firms decides to produce 4 tons per month, then one can predict that
its cost would be
1.266 + (0.752) x 4 = 4.274
Since the costs are measured in thousands of dollars, this means that the total
costs would be expected to be $4,274.
d) If the firms decides to produce 9 tons per month, then one can predict that
its cost would be 1.266 + (0.752) x 9=8.034
Since the costs are measured in thousands of dollars, this means that the
total costs would be expected to be $8034.
E 1: In partially destroyed laboratory record of an analysis of correlation data, the

following results only are legible.
Variance of x = 9
Regression equations : 8x 10y + 66 = 0
40x 18y 214 = 0
what were
12
(1) the mean values of x and y,

(2) the correlation coeff between x and y
(3) the standard deviation of y
E2 Since humidity influences evaporation, the solvent balance of water reducible paints
during sprayout is affected by humidity. A controlled study is conducted to examine the
relationship between humidity (X) and the extent of evaporation (Y) is given below in
table 5. Knowledge of this relationship will be useful in that it will allow the painter to
adjust his or her spray gun setting to account for humidity. Estimate the simple linear
regression line and predict the extent of solvent evaporation (i.e loss of solvent ,by
weight)when the relative humidity is 50%
Regression Analysis
Table 5
Observation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
(x)
Relative
humidity,
(%)
35.3
29.7
30.8
58.8
61.4
71.3
74.4
76.7
70.7
57.5
46.4
28.9
28.1
39.1
46.8
48.5
59.3
70.0
70.0
74.4
72.1
58.1
44.6
33.4
28.6
(y)
Solvent
Evaporation,
(%) wt
11.0
11.1
12.5
8.4
9.3
8.7
6.4
8.5
7.8
9.1
8.2
12.2
11.9
9.6
10.9
9.6
10.1
8.1
6.8
8.9
7.7
8.5
8.9
10.4
11.1
3.2.2 Goodness of Fit
We have seen in the previous subsection that the regression line provides estimates of the
dependent variable for a given value of the independent variable. The regression line is
called the best fitted line in the sense of minimizing the sum of squared errors. The best
fitted line shows the relationship between the independent (x) and dependent (y) variables
better than any other line. Naturally the question arises How good is our best fitted
line?. We want a measure of this goodness of fit. More precisely we want to have a
numerical value which measures this goodness of fit.
For developing a measure of goodness of fit, we first examine the variation in y. Let us
first try the variation in the response y. Since y depends on x, if we change x, then y also
changes. In other words, a part of variation in ys is accounted by the variation in xs.
13
Actually, we can mathematically show that the total variation in ys can be split up as
follows:
n
Syy =
S 2 xy
( yi y ) 2 =
S xx
i =1
( y y ) ,
i
(12)
i =1
where
Sxx =
( xi x) 2 ; Sxy =
i =1
( x x)( y y )
i
i =1
Now if we divide (12) by Syy on both sides, we get

1=
S 2 xy
S xx S yy
n
i =1
( yi yi ) 2
S yy
Since the quantities on the right hand side are both non-negative, none of them can exceed
one. Also if one of them is closer to zero the other one has to be closer to one. Thus if we
denote
S xy
R=
S xx S yy
then
S 2 xy
R =
S xx S yy
Since R2 must be between 0 and 1, R must be between 1 and 1. It is clear that if R2 = 1,

then
n
i =1
( yi yi ) 2
S yy
( y y )
= 0 or
=0
or
yi = yi
for all i.
i =1
Again when R2 is close to 1,
( y y )
i
is close to zero. When R is negative, it means
i =1
that y decreases as x increase and when R is positive y increases when x increases. Thus R
gives a measure of strength of the relationship between the variables x and y.
Now let us compute the value of R for Example 1. For calculating the numerical value of
R, the following formula can be used;
R=
S xy
S xx S yy
n
i =1
n
i =1
( xi x)( yi y )
( xi x ) 2
n
i =1
( yi y ) 2
n
x2
i =1 i
n
xy
i =1 i i
nx 2
nx y
n
y2
i =1 i
ny 2
Therefore, for Example 1, the value of R becomes;

101,570 10 145 67.3
R=
218,500 10 145
and
=
2
47225 10 67.3
3985
8250 1932.1
= 0.9981
R2 = 0.9963.
Therefore, it is clear from the value of R or from R2 that both of them are very close to
one. From the figure also it is clear that the predicted line fits the data very well.
14
Moreover R is positive means, there is a positive relation between the temperature and
yield. As the temperature increases the yield also increases.
Regression Analysis
Now the natural question is how large this R or R2 will be to say that the fit is very good.
There is a formal statistical test based on F-distribution which can be used to test whether
R2 is significantly large or not. We are not going into that details. But as a thumb rule we
can say that if R2 is greater that 0.9, the fit is very good, if it is between 0.6 to 0.8, the fit
is moderate and if it is less than 0.5 it is not good.
E1) For the data given in the table below compute R and R2
Table 6: $yi and e$i For Some Selected i
Sample
12
21
15
1
24
No. (i)
xi
0.96
1.28
1.65
1.84
2.35
yi
138
160
178
190
210
$yi
138
e$i
Note: $yi = 90 + 50x and e$i = yi $yi

3.2.3 Residual Analysis
Fitting a regression model to a set of data requires several assumptions. For example
estimation of the model parameters requires the assumptions that the errors are
uncorrelated random variables with mean zero and constant variance. If these assumptions
are not satisfied, then using the simple least squares method may not produce the efficient
regression line. In fitting a linear model, it is also assumed that the order of the model is
correct, that is if we fit a first order polynomial, then we are assuming that phenomenon
actually behave in a first order manner. Therefore, for a practitioner it is important to
verify these assumptions and the adequacy of the model.
We define the residuals as e i = y i y i , i = 1, 2, , n, where y i is an observation

and y i is the corresponding estimated value from the best fitting regression line.
Analysis of the residuals is frequently helpful in checking the assumption that errors are
independent and identically distributed with mean zero and finite variance and in
determining `whether the additional terms in the model would be required not. It is
advisable to plot the residuals
a) in time sequence (if known),
b) against the yi or
c) against the independent variable x. These graphs will usually look like one of the
four general patterns as shown in the Figures 6 9.
15
9
9
9
9
ei
9
9
9
9
9
9 9 9
9
9
9
9
9
9
yi
Figure 6 Pattern of the residual plot; satisfactory.
The Figure 6 indicates that the residuals are behaving in satisfactory manner and the
model assumptions can be assumed to be correct. The Figures 7 9 given below indicate
unsatisfactory behaviour of the residuals. The Figure 7 clearly indicates that the variances
are gradually increasing. Similarly the Figure 8 indicates that the variances are not
constant. If the residuals plot is like the Figure 9, then it seem the model order is not
correct, that means, the first order model is not the correct assumption. We should look
for higher order models.
*
*
*
ei
*
*
*
* *
*
* *
*
*
*
*
yi
*
*
16
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
ei
*
*
yi
Figure 7 Pattern of the residual plot; indicates the

Figure 8 Pattern of the residual plot; indicates the vari
variance is gradually increasing this case
not constant.
Regression Analysis
*
*
*
*
*
ei
*
*
*
*
*
*
*
yi
Figure 9 Patent of the residual plot; indicates the
model order is not correct.
Example 4: Now we provide the residual plots of the data given in Example 1. We have
plotted yi vs. ei. It is provided in the Figure 10. From the Figure 10, it is quite clear that
the residuals plot is quite satisfactory and apparently all the model assumptions are
satisfied in Figure 10 here.
1.5
1
0.5
ei
-0.5
-1
0
45
50
55
60
65
70
75
80
85
90
yi
Figure10 Pattern of the residual plot; satisfactory.
17
E1 What is the utility of residual plots? what is the disadvantage of residual plots?
3.3 NON-LINEAR REGRESSION

Linear regression is a widely used method for analyzing data described by models which
are linear in parameters. However, in many practical situations, people come across with
data where the relationship between the independent variable and the dependent variable
is no more linear. In that case definitely one should not try to use a linear regression
model to represent the relationship between the independent and dependent variable. Let
us consider the following example.
Example 5 Data on the amount of protein generated by a certain experiment were counted
and reported over time. They are presenting below in Table 7:
Time
(min)
0
10
20
30
40
50
60
70
Protein
(gm)
0.844
0.908
0.932
0.936
0.925
0.908
0.881
0.850
Time
(min)
80
90
100
110
120
130
140
150
Table 7
Protein
(gm)
0.818
0.784
0.751
0.718
0.685
0.685
0.628
0.603
Time
(min)
160
170
180
190
200
210
220
230
Protein
(gm)
0.580
0.558
0.538
0.522
0.506
0.490
0.478
0.467
Time
(min)
240
250
260
270
280
290
300
310
Protein
(gm)
0.457
0.448
0.438
0.431
0.424
0.420
0.414
0.411
We present the Time vs. Protein generated in the Figure 11.

From Figure 11 it is clear that the relationship between the time and protein generated is
not linear. Therefore, they can not be explained by a linear equation. In a situation like
this we may often go for a non-linear model to explain the relationship between the
independent and dependent variables and they are called the non-linear regression model.
A non-linear regression model can be formally written as
y = f ( x , ) + ,
(13)
where f ( x, ) is a known response function of k-dimensional vector of explanatory

variable x and p-dimensional vector of unknown parameter . Here also represents the
error component and it is assumed that it has mean zero and finite variance. Therefore, it
is clear that the non-linear regression model is a generalization of the linear regression
model. In case of linear regression model f (x, ) is a linear function, but there it can be
any non-linear function also. Similar to the linear regression model, here also our problem
is same, that is, if we observe a set of n, {(x1, y1),, (xn, yn)} how to estimate the
unknown parameters , when we know the functional form of f (x, ).
18
Regression Analysis
3.3.1 LEAST SQUARES ESTIMATION

Similar to the linear regression method here also to estimate the unknown parameters, we
adopt the same method. We find the estimate of by minimizing the residual sums of
squares, that is minimize
Q ( ) =
[ y f ( x , )] ,
2
(14)
i =1
with respect to the unknown parameters. The idea is same as before, that is we try to find
that particular value of for which the sum of squares of the distance between the points
yi and f ( xi , ) is minimum. Unfortunately in this case the minimum can not be performed
as easily as before. We need to adopt some numerical technique to minimize the function
Q( ) . This minimization can be performed iteratively and one technique that can be used
to accomplish this is the Gauss-Newton method.
1
0.9
0.8
Protein
0.7
0.6
0.5
0.4
0
50
100
150
200
250
300
350
Time
Figure 11 Time vs. Protein generated in an experiment.
You have already learned about the Gauss-Newton method in details before, we just give
a brief description for your convenience. We use the following notations below:
= (1 ,....., p ),
( k ) = (1( k ) ,........, p( k ) )
(15)
Expand the function f ( x, ) using a Taylor series expansion about the starting point (0)
and using the only the first order expansion, we get:
f ( xi , ) f ( xi , (0) ) + vi1 (1 1(0) ) + ..... + vip ( p p(0) )
where
vij =
f ( xi , )
j
for
= ( 0)
j = 1,.........., p.
19
Let ( ) = ( f ( x1 , ),....., f ( xn , ))' and y = y = ( y1 ,...... yn ) ' then in the matrix notation
we can write (15)
( ) ( (0) ) + V (0) ( (0) ),
where V (0) is the p derivative matrix with elements vij. Therefore, to compute the first
estimates beyond the starting value is to compute
b0 = [V (0)'V (0) ]1[ y ( (0) )]
and then solve for the new estimate (1) as
(1) = b0 + (0) .
This procedure is repeated then with ( 0 ) is replaced by (1) and V (0) by V (1) and this
produces a new set of estimates. This iterative procedure continues until convergence is
achieved. Naturally these computations can not performed by hands, we need calculators
or computers to perform these computations.
Example 5 (Contd). In this case it is observed (theoretically) that the following model (16)
can be used to explain the relationship the time and yield generated yi where
yt = 0 + 1e 1t + 2 e 2t + t .
(12)
Note that as we have mentioned for the general non-linear regression model, in this case
also the form of the non-linear function namely 0 + 1e 1t + e 2t is known, but the
parameters of the model, that is, = ( 0 ,1 , 2 , 1 , 2 ) is unknown. Given the data as
provided in Example 5, we want to estimate the unknown parameters.
We use the following initial guess 0 = 0.5,1 = 1.5, 2 = 1.0, 1 = 0.01, 2 = 0.02,
and finally using the Gauss-Newton algorithm we obtain the estimates of the parameters
as follows:
0 = 0.375,
1 = 1.936
2 = 1.465 ,
0 = 0.013
1 = 0.022
We have plotted the points and also the best fitted regression line, namely
y = 0.375 + 1.936e 0.013t 1.465e 0.013t
(17)
1
0.9
Fitted line
0.8
Protein
0.7
0.6
0.5
0.4
0
50
100
150
200
250
300
350
Figure 12 Time vs Protein generated in an experiment and the best fitted curve
20
Time
Regression Analysis
in the Figure 12. The Figure 12 indicates that the fitted regression curve provides a very
good relationship between the time and protein generated in that experiment. As before
the prediction curve, that is, the curve (17) can be used easily for prediction purposes also.
For example suppose we want to estimate the expected protein generation at the time 115
minutes after the experiment, then using (17), we obtain
y = 0.375 + 1.936e 0.013115 1.465e 0.013115 = 0.698.
Therefore, at the 115 minutes the expected protein generation is 0.698 gms.
Some points we should remember about the non-linear regression model is that we have
to know the functional form of the relationship, but the parameters involved are unknown.
Usually the functional forms are obtained from the physical nature of the process and they
are available. If they are completely unknown it is difficult to use the non-linear
regression model. In that case we need to try with some guess models but they are not
very easy and they are not pursued here. Another important issue is the choice of the
guess value of the iterative process. This is also a difficult problem. Usually from the
prior experimental results we may have to try some trial and error method to find the
initial guess values.
E1 Data on the amount of heat generated by friction were obtained by Count Rumford in
1798. A bore was fitted into a stationery cylinder and pressed against the bottom by
means of a screw. The bore was turned by a team of horses for 30 minutes and Rumford
measured the temperature at small in intervals of time. They are provided in the Table
below:
Time
(min)
4
5
7
12
14
16
20
Table 8
Temperature Time
(F)
(min)
126
24
125
28
123
31
120
34
119
37.5
118
41
116
Temperature
(F)
115
114
113
112
111
110
(1) Plot the time versus temperature and convince yourself that the linear regression
model is not the correct model in this case.
(2) A model based on Newtons law of cooling was proposed as
f (t , ) = 60 + 70e t
Using an initial guess of (0) = 0.01, find the least squares estimate of .
(3) Based on the fitted least squares regression line find the expected temperature at the
time 15th minute after starting the experiment.
21
3.4 SUMMARY
In this unit you have seen :
that regression analysis is an important technique, which can be used to verify the
results of any experiment.
How to determine the relationship between a dependent and an independent

variable by using the Scatter diagram
that by knowing the technique of regression you have an edge to analyse the
results in an organized way. Further this analysis is smoothened by application of
the concepts like least square estimation, goodness to fit and residual analysis.
that many times the data obtained by conducting an experiment does not follow
the linear relation. So, to handle such aspects we have also discussed the concept
of non linear regression, under we have emphasized least square estimation
technique.
Formulas and applications of following topics:
Simple Linear Regression

Goodness to Fit
Residual Analysis
Non-Linear Regression
3.5 SOLUTIONS
Check you r progress 1
E 1:
(1) Since both the regression lines pass through the point ( x , y ), we have
8 x 10 y + 66 = 0
40 x 18 y 214 = 0
Solving we get,
x = 13
y = 17.
Let 8x 10y + 66 = 0 and 40x 18y 214 = 0

Be the lines of regression of y and x and x on y respectively. Now, we put them in
the following form.
=
22
8
66
18
214
and x =
x +
y +
10
10
40
40
(4)
8 4
=
10 5
18 9
bxy = regression coeff of x on y =
=
40 20
4 9
9
Hence, r2 = bxy. byx =
=
5 20
25
3
= 0.6
So r =
5
byx = regression coeff of y on x =
Regression Analysis
Since, both the regression coeff are +ve, we take r = + 0.6

(3) We have, byx = r
y
x
4
3
y
=
x
5
5
3
y=4
Remarks (i) had we taken 8x 10y + 66 = 0 as regression equation of x on y and
40x 18y = 214, as regression equation of y on x.
10
40
Then
bxy =
and
byx =
8
18
10
40
or
r2
=
bxy byx =
x
= 2.78
8
18
so
r = 1.66
Which is wrong as r lies between 1.
E2
Observation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
(x)
Relative
humidity,
(%)
35.3
29.7
30.8
58.8
61.4
71.3
74.4
76.7
70.7
57.5
46.4
28.9
28.1
39.1
46.8
48.5
59.3
70.0
(y)
Solvent
Evaporation,
(%) wt
11.0
11.1
12.5
8.4
9.3
8.7
6.4
8.5
7.8
9.1
8.2
12.2
11.9
9.6
10.9
9.6
10.1
8.1
23
19
20
21
22
23
24
25
70.0
74.4
72.1
58.1
44.6
33.4
28.6
6.8
8.9
7.7
8.5
8.9
10.4
11.1
Summary statistics for these data are

x = 1314.90
y2 = 2286.07
n = 25
x2 = 76.308.53
y = 235.70
xy = 11824.44
To estimate the simple linear regression line, we estimate the slope 1 and
intercept 0. these estimates are
n xy [( x)( y )]
=1 => $ 1 = b1 =
n x 2 ( x)2
25(11,824.44) [(1314.90)(235.70)]
=
25(76,308.53) (1314.90) 2
= .08
0. => $ 0 = b0 = y b1 x
= 9.43 ( .08) (52.60) = 13.64
S
O
L
V
E
N
T
14
13
12
11
E 10
V
A 9
P
O 8
R
A 7
T
I 6
O
N 5
(% wt)
24
10
20
30
40
50
60
70
RELATIVE HUMIDITY (%)
80
90
100
Figure: A graph of the estimated line of regression of Y, the extent

of evaporation, on X, the relative humidity
Regression Analysis
Hence, the estimated regression equation is

Y | x = $y = 13.64 .08 x
The graph of this equation is shown in Figure above. To predict the extent of
solvent evaporation when the relative humidity is 50 %, we substitute the value 50
for x in the equation.
$y = 13.64 .08 x
to obtain $y = 13.64 .08(50) = 9.64 . That is, when there relative humidity is 50 %
we predict that 9.64% of the solvent, by weight, will be lost due to evaporation.
Recall from elementary calculus that the slope of a line gives the change in y for a
unit change in x. If the slope is positive, then as x increases so does y; as x
decreases, so does y. If the slope is negative, things operate in reverse. An
increase in x signals a decrease in y, whereas a decrease in x yields an increase in
y.
E1 R =
S xy
S xx S yy
= (138.076) / [(2.6929 * 7839.93)1/2] = 0.9504
So, R2 = 0.9033
E1 Residual plots are helpful in spotting potential problems. How ever theyare not always
easy to interpret. Residual patterns are hard to spot with small data set except in extreme
cases, residual plots are most useful with fairly large collection of data.

E1 Refer to example solved in the section 3.3.1
25
Table 1: The distribution function of standard normal random variable

z
0.00 0.01 0.02 0.03 0.04 0.05
0.06 0.07 0.08 0.09
0.0 0.5000 0.5040 0.5080 0.5120 0.5160 0.5199 0.5239 0.5279 0.5319 0.5359
0.1 0.5398 0.5438 0.5478 0.5517 0.5557 0.5596 0.5636 0.5675 0.5714 0.5753
0.2 0.5793 0.5832 0.5871 0.5910 0.5948 0.5987 0.6026 0.6064 0.6103 0.6141
0.3 0.6179 0.6217 0.6255 0.6293 0.6331 0.6368 0.6406 0.6443 0.6480 0.6517
0.4 0.6554 0.6591 0.6628 0.6664 0.6700 0.6736 0.6772 0.6808 0.6844 0.6879
0.5 0.6915 0.6950 0.6985 0.7019 0.7054 0.7088 0.7123 0.7157 0.7190 0.7224
0.6 0.7257 0.7291 0.7324 0.7357 0.7389 0.7422 0.7454 0.7486 0.7517 0.7549
0.7 0.7580 0.7611 0.7642 0.7673 0.7704 0.7734 0.7764 0.7794 0.7823 0.7852
0.8 0.7881 0.7910 0.7939 0.7967 0.7995 0.8023 0.8051 0.8078 0.8106 0.8133
0.9 0.8159 0.8186 0.8212 0.8238 0.8264 0.8289 0.8315 0.8340 0.8365 0.8389
1.0 0.8413 0.8438 0.8461 0.8485 0.8508 0.8531 0.8554 0.8577 0.8599 0.8621
1.1 0.8643 0.8665 0.8686 0.8708 0.8729 0.8749 0.8770 0.8790 0.8810 0.8830
1.2 0.8849 0.8869 0.8888 0.8907 0.8925 0.8944 0.8962 0.8980 0.8997 0.9015
1.3 0.9032 0.9049 0.9066 0.9082 0.9099 0.9115 0.9131 0.9147 0.9162 0.9177
1.4 0.9192 0.9207 0.9222 0.9236 0.9251 0.9265 0.9279 0.9292 0.9306 0.9319
1.5 0.9332 0.9345 0.9357 0.9370 0.9382 0.9394 0.9406 0.9418 0.9429 0.9441
1.6 0.9452 0.9463 0.9474 0.9484 0.9495 0.9505 0.9515 0.9525 0.9535 0.9545
1.7 0.9554 0.9564 0.9573 0.9582 0.9591 0.9599 0.9608 0.9616 0.9625 0.9633
1.8 0.9641 0.9649 0.9656 0.9664 0.9671 0.9678 0.9686 0.9693 0.9699 0.9706
1.9 0.9713 0.9719 0.9726 0.9732 0.9738 0.9744 0.9750 0.9756 0.9761 0.9767
2.0 0.9772 0.9778 0.9783 0.9788 0.9793 0.9798 0.9803 0.9808 0.9812 0.9817
2.1 0.9821 0.9826 0.9830 0.9834 0.9838 0.9842 0.9846 0.9850 0.9854 0.9857
2.2 0.9861 0.9864 0.9868 0.9871 0.9875 0.9878 0.9881 0.9884 0.9887 0.9890
2.3 0.9893 0.9896 0.9898 0.9901 0.9904 0.9906 0.9909 0.9911 0.9913 0.9916
2.4 0.9918 0.9920 0.9922 0.9925 0.9927 0.9929 0.9931 0.9932 0.9934 0.9936
2.5 0.9938 0.9940 0.9941 0.9943 0.9945 0.9946 0.9948 0.9949 0.9951 0.9952
2.6 0.9953 0.9955 0.9956 0.9957 0.9959 0.9960 0.9961 0.9962 0.9963 0.9964
2.7 0.9965 0.9966 0.9967 0.9968 0.9969 0.9970 0.9971 0.9972 0.9973 0.9974
2.8 0.9974 0.9975 0.9976 0.9977 0.9977 0.9978 0.9979 0.9979 0.9980 0.9981
2.9 0.9981 0.9982 0.9982 0.9983 0.9984 0.9984 0.9985 0.9985 0.9986 0.9986
3.0 0.9987 0.9987 0.9987 0.9988 0.9988 0.9989 0.9989 0.9989 0.9990 0.9990
Table 2: The critical values of chi-square distribution. The areas given across the top are
the areas to the right of the critical value. To look up an area on the left, subtract it from
one, and then look it up (i.e., 0.05 on the left is 0.95 on the right).
df
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
0.995 0.99
0.975 0.95
0.90
0.001
0.004
0.016
0.010
0.020
0.051 0.103
0.211
0.072
0.115
0.216 0.352
0.584
0.207
0.297
0.484 0.711
1.064
0.412
0.554
0.831 1.145
1.610
0.676
0.872
1.237 1.635
2.204
0.989
1.239
1.690 2.167
2.833
1.344
1.646
2.180 2.733
3.490
1.735
2.088
2.700 3.325
4.168
2.156
2.558
3.247 3.940
4.865
2.603
3.053
3.816 4.575
5.578
3.074
3.571
4.404 5.226
6.304
3.565
4.107
5.009 5.892
7.042
4.075
4.660
5.629 6.571
7.790
4.601
5.229
6.262 7.261
8.547
5.142
5.812
6.908 7.962
9.312
5.597
6.408
7.564 8.672
10.085
6.265
7.015
8.231 9.390
10.865
6.844
7.633
8.907 10.117 11.651
7.434
8.260
9.591 10.851 12.443
8.034
8.897
10.283 11.591 13.240
8.643
9.542
10.982 12.338 14.041
9.260
10.196 11.689 13.091 14.848
9.886
10.856 12.401 13.848 15.659
10.520 11.524 13.120 14.611 16.473
0.10
2.706
4.605
6.251
7.779
9.236
10.645
12.017
13.362
14.684
15.987
17.275
18.549
19.812
21.064
22.307
23.542
24.769
25.989
27.204
28.412
29.615
30.813
32.007
33.196
34.382
0.05
3.841
5.991
7.815
9.488
11.070
12.592
14.067
15.507
16.919
18.307
19.675
21.026
22.362
23.685
24.996
26.296
27.587
28.869
30.144
31.410
32.671
33.924
35.172
36.415
37.652
0.025
5.024
7.378
9.348
11.143
12.833
14.449
16.013
17.535
19.023
20.483
21.920
23.337
24.736
26.119
27.488
28.845
30.191
31.526
32.852
34.17
35.479
36.781
38.076
39.364
40.646
0.01
6.635
9.210
11.345
13.277
15.086
16.812
18.475
20.090
21.666
23.209
24.725
26.217
27.688
29.141
30.578
32.000
33.409
34.805
36.19
37.566
38.932
40.289
41.638
42.980
44.341
0.005
7.879
10.597
12.838
14.860
16.750
18.548
20.278
21.955
23.589
25.188
26.757
28.300
29.819
31.319
32.801
34.267
35.718
37.156
38.582
39.997
41.401
42.796
44.181
45.559
46.928

Computing MAths

Uploaded by

Copyright:

Available Formats

Computing MAths

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Computing MAths

Uploaded by

Copyright:

Available Formats

UNIT 1 PROBABILITY DISTRIBUTIONS

Continuous Random Variable

Uniform Random Variable

describe events and sample spaces associated with an experiment;

1.2 RANDOM VARIABLES

HHH HHT HTH HTT

Figure 1: Schematic representation of a random variable

1.3 DISCRETE RANDOM VARIABLE

(c) lim F (b) = 0

The number of heads (X value)

Now let us see the graphical representation of the distribution.

Note :Mathematical expectation or Expected values or Expectations forms

Check Your Progress 2

(a) The daily measurement of snowfall at Shimla.

It involves a repetition of n identical trials (Here n=3).

Check Your Progress 3

In this subsection we will introduce another discrete distribution called Poisson

0.00005 +0.00050 + 0.00226 +0.00757 = 0.01038

1.4 CONTINUOUS RANDOM VARIABLE

(c) P (c < X < d) =

The Uniform Random Variable

The uniform distribution is the simplest of a few well-known continuous distributions,

Exponential Random Variable

In making a mathematical model for a real world phenomenon, it is always necessary to

f(x) = e - x for x>0

in the figure below. From the figure,

Figure 4: The p.d.f. of the exponential distribution for different values of .

Normal distribution is undoubtedly the most used continuous distribution in different

, where - < x <

P(a < X < b) =

The last equality follows by simply making the transformation z =

P(a < X < b) = P (

We will call the result, Z, so

has a standard normal distribution.

involves the standardized score, Z =

X 10500 9000 10500

If the random variable X has chi-square distribution with n-degrees of freedom,

here (.) is a gamma function and it is defined as

EXAMPLE 9 Show that the moment generating function of a random variable X

which is chi-square distributed with v degrees of freedom is M(t) = (1 2t) v/2.

Letting (1 2t ) x / 2 = u in the last integral we find

Check your progress 7

Discrete and Continuous

c. Uniform Distribution : The probability density function is defined by

d. Normal Distribution : The probability for this distribution is determined

, where - < x <

e. Chi-Square Distribution : If the random variable X has chi-square

here (.) is a gamma function and it is defined as

Mathematical expectation or Expected values or Expectations

E1 a) If X denotes the number of correct answers, then X is the random

P[X=2] = (4C2) * (9/10)2 *(1/10)2

Average bad Cheques received per day = 6

Check your progress 6

P[952 < Z < 956] = P[((952-952)/4) < ((X-952)/4) < ((952-956)/4)]

UNIT 2 PSEUDO RANDOM NUMBER GENERATION

Above all, the numbers produced should appear to be distributed uniformly on

2.2 UNIFORM RANDOM NUMBER GENERATORS

2.3 GENERATING RANDOM VARIATES FROM ARBITRARY

we assume that we have a reliable sequence of U(0,1) variates available to us.