UNIT 3 PROBABILITY DISTRIBUTIONS

Structure

Page No.

3.1 Introduction
Objectives

3.2 Random Variable

Discrete Random Variable
Continuous Random Variable

3.3 Binomial Distribution

3.4 Poisson Distribution

3.5 Uniform Distribution

77

3.6 Normal Distribution

80

3.7 Summary

3.9 Appendix:Statistical Tables

-

3;l INTRODUCTION
In the first two units of this block you learnt a few methods for exploring and describing
numerical data. The examples you saw there showed that numerical observations on a
characteristic might need careful exploration for any useful and correct inference. In
general, when you decide to collect data on a characteristic, you have a specific
purpose; you want to either verify a hypothesis or want to estimate a quantity for
certain purpose. For example, you may want to know if the district of Jalandhar
produces more wheat in a year than the district of Faridkot in the state of Punjab. You
may have a hypothesis that in general, Jalandhar district produces more wheat per year
than Faridkot district. When you have such a specific question or hypothesis, you will
decide to collect the appropriate information or data, in this particular instance, on the
annual wheat production in the two districts of Punjab. You may find a secondary
source, such as a publication from an agency or a department of the government which
has already collected this information and has tabulated the data. Of course, you need
to be sure about the reliability of the source from where you have obtained the required
data. When you actually look at the data on the amount of annual wheat production in
the two districts, for say, the past ten years, what do you think you will And? Do you
think the amount produced in any district will be the same from year to year? This is
very unlikely as there are a very large number of factors that influence the amount of
wheat produced and these factors do not remain constant over the 10 years. Also it is
very unlikely that you will find Jalandhar district producing more wheat than the
district of Faridkot in each of the 10 years. In such a situation how does one compare
the two districts in respect of wheat production?
Firstly, it is clear from the above discussion that a realistic model will be to assume that
the annual production of wheat in a district is a random variable (result of a random
experiment introduced in the previous unit). The ten year observations may be thought
of as the outcome of ten repetitions of a random experiment for a district. Secondly, we
need to develop measures to compare the two districts with respect to the annual
production of wheat when this is thought of as outcome of a random experiment.

In certain situations or for certain problems, you may find that there is no reliable
secondary source where data on the appropriate characteristic are available, and that
you have to take measurements or make observations, if necessary after conducting a
controlled experiment. As in practice, it is never possible ta control or eliminate the
influence of all the factors, this experiment may also have to be modelled as a random
experiment. This need arises, for instance, when a scientist claims to have developed a
variety of wheat which yields more per acre than the existing varieties and you want to
verify this claim. To sum~xarise,you notice the following:
1) Given a specific purpose, data are collected on an appropriate characteristic.

2) In most real-life pr~blemsituations, the value of the characteristic of interest may

depend on a large number of factors which are not constant over space or time. It
is better, then, to model the situation as a random experiment and the value of the
chzracteristic of interest as an outcome of a random experiment.
3) Methods have to be developed to draw the correct conclusion or inference from the
data collekted on a characteristic, which is the outcome of a random experiment.
4)

If there is no secondary source of data, you may nwd to either conduct a controlled
experiment, or conduct a survey, to obtain data on the appropriate characteristic.

We start this unit by defining the notion of a random variable and its probability
distribution @ Section 3.2. The notions of discrete and continuous random variables are
introduced next, followed by the notions of expectation and variance. You will see that
to compare random variables or to draw inferences about them in a practical
application, their probability distiibutions are important. Also important are certain
measures of the distribution such as the mean and variance. These are discussed in
Sec.3.2. In the next four sections we have discussed the most commonly used four
probability distributions in detail.
Here is a list of what you should be able to do by the end of this unit.

Objectives
After reading this unit, you should be able to
0 specify when a variable is a random variable and classify it as discrete or continuous.
.
find the probability distribution of discrete and continuous random variables and cal*
culate the mean and variance of these distributions and use these measures to make
judgements about the real-life situation.
describe the following distributions
a) binomial distribution

b) Poisson distribution

c) uniform distribution
d) normal distribution

and calculate the m e a and variance associated with these distributions.

distinguish between the real-life situations which can be modelled (or studied) by
these distributions when to use binomial or when to use Poisson and likewise other
distributions.

3.2 RANDOM VARIABLE

Let us start with the following examples:
1) The number of telephone calls received by an operator in a specified interval of
time.

2) The amount of rainfall on a day.

Probability Distributions

Statistics and Probability

3) The outcome of throwing a die.

4)

The scores of students in an examination

5 ) The number of misprints on a randomly chosen page of a book.

6) The volume of sale of a certain manufactured item in a given year.

7) Time to failure of a machine.

All these examples have one common feature. They describe a characteristic which
associates a numerical value or number to each outcomes of a random experiment.
Recall from Unit 2 that random experiments are experiments the outcomes of which
cannot be predicted in advance. We call this characteristic a variable. JXs
characteristic depends on the outcome of the experiment and its value cannot be
predicted in advance. This variable is cdled a random varidble. "When a variable takes
di$erent values according to chance, (which can't be predicted before hand), it is called
a random variable." In order to make this idea clear, consider the follawing example.

Example 1: Suppose we are interested in the number X of heads obtained in three

tosses of a coin. Let us see what is the variable in this experiment.
For that we first find the set of possible outcomes of the experiments ie., the sample
space S. The outcomes are

Then

Let us denote by X(aj) the number of heads obtained when the outcome of our
experiment is aj, where j = 1,2, .. . ,8. You can easily cbeck that
X(al) = 3, X(az) = X(a3) = X(Q) = 2,
X(a5) = X(%) = X(a7) = 1, X(q) = 0
Then do you agree that the X maps elements of the sample s p a $ $ the values
0,1,2,3? i.e. X is a function from the space S to the set N * (0, I, $,-$),:

(1)

(2)

Also note that corresponding tq each value, there is always some sample P;oint or a set
of sample points. For example. the set of sample points corresponding la the value '0'
is the single point as, whereas for 1 the set is: {a5, as, a, 2. That rneans.correspondmg to
each value of X, there is a subset of the sample space S.
Now you again recall from Unit 2 that an event is a subset of a sample space: Thus we
note that that each value of X is associated with an event.
You can, therefore make the following identification of events corresponding to the
values associated by X. Denote the event corresponding to '0'-as [X= 01, likewise for
other values. Then
[ x = 0 ] = {as),

[x= 11 = (as,as,a7)

[ X = 2 l = {a2,a3,a4),[X=3l={al)
Assuming that all the sample points are equally likely, we assign probabilities of 118 to
each of the sample points.

From here, using the law of probability you can calculate the probabilities as follows:
P[X = 01 = P(ae) = 1/8,
P[X = 11 = P{as, %, a ~ =) Pias) P{&) P(87) = 3/8,
P[X = 21 = 3/8and PP[ = 31 = 1/8,
where we read PIX = j] as "probability that X equals j."

60

Probability Distributions

Then P has the following Property:

P[X = 01

+ P[X = 11 + P[X = 21 + P[X = 31 = 1.

***

Summing up our discussion above, we find that a random variable X is a function

defined on the set of outcomes of a random experiment and it takes on a numerid
value corresponding to each outcome of the experiment. Sometimes we use the
abbreviation r.v. for a random variable. Now these set of possible values is a numerical
representation of the original space. The following figure may help you to see this
representation.

Fig. 1 Representation of a r.v.

Because of this representation, instead of working with an abstract space, we can work
with a set of numbers,and this simplifies our problems considerably.
Now let us consider another example.

Example 2: You are sitting in a plane waiting for its take off. The pilot announces a
delay until some incoming planes land. Suppose you want to find the following:
i) How long will it be before take off.
ii) How many incoming planes are there.
Let us discuss the random variables for (i) and (ii).of the above example.

'

We first take (i) In this case we want to find the 'duration of time' before the plane takes
off. Note that the variable takes values continuously along a line as given below, say
from 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.

Fig.2

Now let us consider (ii) . Here the random variab1e.i~the number of planes. This
variable can only take the values 0 or 1 or 2 etc. as shown in Fig. 3. There is no
continuity, (see! Fig.3) since only non-negative integer values ean be assunied.

3tatistics and Probability

The above examples show that random vafiables can be of different types. There are
mainly two types of random vaiables:
1) discrete random v&.able
2) continuous random variable
The random variable shown in (ii) of Example 2 is discrete and that of (i) is conthuous.
In the next sibsectioil we shall iliscuss discrete random variables. Before that why
don't you try an exercise.
El) Suppose you take a 50-question multiple-choice exam., guessing every answer,
and are interested in the number of correct answers obtained. Then
a) What is the random variable you will consider for this situation?
b) What values might this random variable have?
c)

What would P[X = 401 means?

Discrete Random Variables

3.2.1

We first formally define a discrete random variable and familiarise you with some of the
properties1 aspects of a discrete r.v.
Definition 1: A random variable X is said to be discrete if the. number of values that X
can take is finite or soufitably infinite. These values may be listed as xo, xl, . . . , w h w
say, xo < XI < . . . ; these x's are called the jump points. They need not be equidistant,
Now let us consider the events associated with the values assigned by it.
Let the events be denoted by [X= xi],j = 0,1,2, . . .. Then, as stated earlier, we can
assign probability to these events. We denote P[X = xj], the probability of the event
[X = xj]. For further simplification we denote the probability for each j as
P[X=xj] = p j , i = 0 , 1 , 2 , 3 ,....
From the prope.*es of a random variable and by definition of probability it follows that

> 0 (for each i, i.e., each pi is a non-negative number),

ii) po + + . . . = 1 (The sum of the probabilities is 1).
i) pi

Now we have another definition.

Definition 2: Let p : X -t R be defined as
pi

ifx=xi,

i = 0 , 1 , 2 ,...

0,

otherwise
Then p is called the probability mass hnction(p.m.f) of the random variable X. The
collection of pairs (xi, pi), i = 0,1,2, . . . is called the probability dhtribution of X.
For example, suppose X is the r.v. denoting the number of heads obtained in three tosses
of a coin, then the probability mass function p is the funcuon p :X -t R such that

Note that p(xi) = pi

2 0 for all xi h d

Therefore, the probability distribution corresponding to ulls d

m vanable is the s&
((0,
(1, f ), (2, )), (3,i)). This is also expressed in a W a r form as follows:

i),

The number of heads Probability

(Values
of the r.v. X)
0
1

2
3

118
318
318
118

Suppose you have a random variable X assuming values XI,xa, . .. with probabilities
p1 , p2, . . . , respectively. You may also visualise this as an illustration of a frequency
distribution. In fact a frequency distribution tells you how the total probability "one" L
distributed over the possible values of the random variable.
Now let us see the graphical representation 01this dlstribation.
Graphically, along the horizontal axis, plot the various possible values xi of a random
variable and on each value erect a vertical line with height proportional to the
corresponding probability p,. (See Fig.4) Recall that in the bar diagram of a frequency
distribution, the observed frequencies are graphed along the vertical axis; the total
frequency (which is the same as the total number of re~titionsof the random
experiment) is thus distributed over the possible ouicomes.

Probability Distributions

Table 1: Probability distribution of number of heads in three tosses of a coin

,0.40

.-

1, 1 ,

;:;;

0.1 0

Fig. 4

It is therefore clear that associated with any random variable, a probability

distribution can be defined. Thus in the study of a ra?dom variable it is enough to
know the corres~ondingprobability distribution. This is illustrated in the following
example.

Example 3: Recall the problem of Sunil, the newspaper boy, which was presented in
in Section 1, Unit 2. When Sunil mentioned his dilemma about his 10 irregttl~r
customers to his sister Sunita, who is doicg a course in statistics at the local college. she
advised him, as a first step, to start maintaining a record for each of the 10 customers,
showiug for each day whether the customer has taken the newspaper from him or not.
Following her advise, Sunil had at the end of two months, 61 sets of observations, each
set corresponding to a day. F.ach set of observations was written as a sequence of two
numbers, 1's and O's, 1 at position i showing that customer i has bought the newspaper
on that day and a 0 in that position showing that customer i has not bought the
newspaper on that day. You will also find that the kth sequence (corresponding to the
kth day) is actually an observed sample point of the sample space corresponding to the
random experiment performed on day k with these 10 customers. Sunil has repeated
this random experiment with his 10 customers for 61 days! When Sunil reflected over
the mass of observations he hasmade it suddenly occurred to him that his record may be
sxcessive for solving his problem. Do you also get the same idea as Sunil did? If not,
think about it for a while now, and then read ahead.
Sunil reasoned as follows: After all, his daily gain will only depend on how many
newspapers he is able to sell on a particular day, irrespective of who among the 10 buys.
Therefore, it is enough for his purpose to note down the number of 1's that appear in the
kth sequence correspoilding to day k, to calculate his gain for that day. So why does he
have to maintain a sequence? Just the total number of customers buying on a day
should serve the purpose. Do you agree with Sunil? When Sunil showed his diary to
Sunita and mentioned to her his new idea, she appreciated his line of thought and told
him that the variable he wanted to consider, namely the total number of 1's in a
sequence, would be called a random variable by a statistician as the sequences were the
observed results of a randorn experiment. Sunil could forget about the sample space
containing the sequences, provided he knew the probability distribution of the random
variable chosen by him, namely the number of sales on a day to his irregular customers.
We shall consider the problem of finding the probability distribution of this random

Number Of Heads

63

Statistics and Probability

variable in the next section.

In the following sections you will see that there are some standard distributions and
most of the real-life problems can be solved by finding the distribution corresponding to
a probability model of the given situation. Let us try some exercises.
E2) Which of the random variables given below are discrete? Give reasons for your
answer.
1) The daily measurements of snowfall at Shimla
2) The number of industrial accidents in each month.
3) Thenumber of,defective goods in a shipment (lot) of goods from a
manufacturer.
E3) A box contains twice as many red marbles as green marbles. One marble is drawn
at random from the box and replaced; then a second marble is drawn at random
from the box. If both marbles are green, you win Rs. 50; if both are red you lose
Rs. 10; and if they are of different colours, you will win or lose nothing. Then
what is the probability distribution of the amount you win or lose.
Let us now return to the discussion of the three tosses of an unbiased coin. The r.v. X,
denoting the number of heads obtained, has the following probability distribution:

Suppose you want to calculate the probability of the event {X

< 2).

<

First note that the event {X 2) is the same as the event

{X=O)U{X= 1)u{X=2)]
Then, since the events are disjoint, we can write,

< 21

= P[X= 01 +PIX =-I] + P [ X = 21

- -1 + -3+ -3= 7
8 8 8 8 '
Therefore there are situations in which you are not only interested in the probability
P[X = xi] = pi, but are also interested in the probability of the X 5 xi, i.e., events of
the typz [X 5 xi] which we denote by P[X Ixi].
P[X

As we already pointed out, the probability distribution of a random variable is

analogous to a frequency distribution. It is therefofe not surprising that just as we found
it useful to calculate the mean of a frequency distribution, we now find it useful to
calculate the mean of a probability distribution. Another term often used to denote the
mean of a probability distribution is the expected value of the random variable which is
defined as follows:

Definition 3: For a discrete random variable X, its expected value (or mean), denoted
as E(X), is defined as:

where xo, XI,xz are the values assumed by X and po, pl ,pz are probabilities these
values.
Expected value is a fundamental idea in the study of probability distributions. For many
years, the concept has been put to considerable practical use in the insurance industry,
and in the last twenty years, it has been widely used by many others who must make
decisions under conditions of uncertainly.

I @

We shall illustrate this idea with a real-life problem given in Example 4.

Example 4: The Director of a breast cancer screening clinic wants to know how many
women will be screened on any one day. If past daily records of the clinic indicate that
the number of women screened daily ranges between 100 to 115. The following table
illustrates the number of times this level, between 100 to 115, has been reached during
the last 100 days.

Probability Distributions

Table3 : Number of women screened daily during 100 days

Number screened

Number of days
this level was observed

Probability that the random variable

will take on this value

100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115

1
2
3
5
6
7
9
10
12
11
9
8
6
5
4
2

.01
.02
.03
.05
.06
.07
.09
.10
.12
.ll
.09
.08
.06
0.05
.04
.02

Total - 100

Total - 1.OO

Let us see how the Director can use this past record to get information about the
long-run pattern of daily patient screenings.
We hope you have understood the real-life problem. We will try to model this in
statistical language. For that we shall first describe the 'random variable' of interest in
this problem.
The random variable here is the number of patients screened, on any given day. Note
that this is a discrete random variable, which can assume only non-negative integer
values with positive probability. The past record of the clinic indicates that the values of
@is random variable range between 100 and 115 patients daily. These values are given
in the 1st column of the table. The 2nd column contains the number of days each value
was observed. For example the value '103' occurred on 5 days. The last column gives
the probability/relative frequency for which a particular value is observed. How do you
calculate these probabilities? Note that the total number of days is 100 and that the
value '100' was observed only one day.
1
Probability that the value'100'is observed = - = .O1
100
In this manner you can calculate the other probabilities. (The thumb rule is 'divide each
value in the middle column by 100'). This is how the last column is obtained. Notice
that the sum of the values in the 1ast.column is one. The relative frequencies are taken
as probabilities. This is the statistician's empirical approach to assigning probabilities.
Now plot the 'observed values' (i.e. numbers in the 1st column) against the
probabilities in a graph. Then you get a graph as given in the next page.

65

Statistics and Probability

100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115
Daily number of women screened

Fig.5: Probability distribution for the discrete random variable 'number screened"

So, this is the graph of the probability distribution of the r.v.

Once we have a frequency distribution, we know that we can find its mean, variance
etc. Let us calculate the mean. Here is the mean

C xipi
Mean = -C pi

where xi's are the observed values in the first column and pi's are the frequencies
100x 1+101 x 2 +
Mean = 100

...+ 115 x 2

This we can rewrite as

1

Mean = 100x - + l o 1
100

-+...+
100
L

115x 100

So, essentially what we got is that 'if we multiply each number in the 1st column by the
corresponding number in the 3rd column' we will get the mean of the probability
distribution. This mean equals the expected value.
The above computation tells us that the expected value of the discrete random variable
"cumber screened is 108 women. What does this mean? It means that over a long
period of time, the number of daily screenings should average about 108.02. This does
not mean that 108 women will visit the clinic per day; this only says that over the long
run arid on an average 108 women can visit the clinic. This value is a long run average.
Now baaed on this expected value (or the mean) the director csn dedde on what
,
resourceslirmfiastru~tureis required to get ready for dealing with the expected number of
-people.

***

Why don't you try an exelcise now.

E4) A second-hand car dealer has sold as many as five cars in one day, and as few as
one. The dealar has tabulated sales records for a large number of days and found
that on 5 percent of the days co cars were sold. The dealer took 0.05 as the
probability of zero sales in a day, as shown in Table 3 below.Probabi1ities for sales
of 1,2,3,4 and 5 cars were assigned in the same manner (see Table below).
Table 3
Number of cars sold per day
0
1
2
3
4
5
Probability
0.05 0.15 0.35 0.25 0.12 0.18
He wants to find how many cars per day will be sold on the average over a long

run. How does he get this number?

3.2.2

Continuous Random Variable

The variables, we discussed in the last section such as 'number of women screened',
"number of heads obtained", etc. take on values 0 , 1 , 2 , 3 , . ... These are discrete
random variables. We saw that the values of a discrete random variable are graphed as
separated points and probabilities as lengths of vertical line segments (see Fig.4). We
also saw that a probability distribution of such a random variable contains all possible
values of the random variable, so the sum of all the probabilities must be 1.
On the contrary, a continuous random variable can take on any value in an interval.
For example, it can take all values x in the interval 0 5 x 5 1 of the form
C, 0.01,9.0002, etc. . . . ,0.98,0.99,1.00. As we have seen in the discrete case, we have
to assigrl probabilities to each value of the variable. How do we do this? Since the
possible values of X are uncountable, we cannot really speak of the ith value of X, hence
p(xi) becomes meaningless. Sc, what we shall do is to replace xi by any interval of the
type of (xi-l, xi) where 0 5 xi-1 < xi 5 1. -and define probability for such intervals.
To define this we consider a function defined on this interval 0 5 x 5 1, which assumes
non-negative values such that the total area under the graph cf this function and over
xi) as the area
the interval is 1. Then we define the probability of any interval
over this interval and under the graph off. (See Fig.6(b)) At present you don't have to
worry about these functions. Wherever the need arises we will specify these functions.
The above discussion may appear a little vague to you. You need not worry about this
at this stage. The ideas will be clear to you once we discuss particular cases of
continuous random variables and their probability distributions.
Now let us compare the graphs of probability distributions drawn for a discrete and
continuous random variable.

Pr&@llty dlstrlbuthn for a rllscrete random variable

Probabllltydistributiontor a contlnow random variable

ilhe probsblUty oS a horizontal intend la tlY .nr auw Ute intwvrri)

NMb)

Note that in Fig.6(b) we did not put vertical lines because a vertical distance tells us
what the height of the curve is, and that is not what we want. We want areas under the
portion of curves. You will see later that there are tables available for certain
continuous random variables where these areas are given for different segments. For

Probability Distributions

Statisticsand h b a b u t y

examples in Fig. 6@),the probability of the line segment 2.2

< x 5 4 is

P[2.2 5 x 1 4) = 0.32
and that of 0

< x < 2.2 is

It is important ta remember that a line segment has no area (zero area), because a line
has no width. Thus, the vertical line segment at 4 in the distribution of Figure 6 (b) has
an area of zero. This means the probability of a single value 4 is zero. In general, the
probability for an exact (single) value of a continuous random variable is zero.
Consequently, the probability of an interval is the same whether the endpoints are
included or not - because the endpoints have probability zero.
Now we formalise the above discussion and make the following definition.

Definition 4: Let X be a continuous random variable which takes on values in the

interval (ab). [i.e. all values between a and b, a < b)]. A function f(x) defined on X is
called the probability density function of X if
(i) f(x) is nonnegative for a 5 x < b i.e., f(x) > 0 f a all x lying between a and b.
(ii) the area under the graph and above the interval (a, b)is 1.
(iii) For any two real numbers c and d between a and b, the prabability that the random
variable adopt a value between c and d is equal to the area under the graph and
above the interval (c, d) i.e.
P[C

5 x 5 dl = area uider the graph over the interval(c,d)

Note: Those who are familiar with the mathematical concept of integration can easily
see that the above mentioned area is given by the integral of the function i.e.

and (ii) says that

We shall now see how we can use the graph of the distribution of a continuous random
variable to study real-life problems.

Problem 1: Suppose the Director of a training programme wants to conduct a

programme to upgrade the supervisory skills of production line supervisors. Because
the programme is self-administered, supervisors require different numbers of hours to
complete the programme. Based on a past study of participants, the following
distribution (see Fig.7) showing the time spent by candidates is available which shows
that average time spent is 500 hours.

Fig7

How can she use the graph of distribution, to find the following: What is the chance that
a participant selected at random will require

'

i)

more than 500 hours to aomplete the programme?

Probability Distribution$

ii) less than 500 hours to complete the p r o g r m e ?

Solution:
i) From the figure, we sg that half of the area under the curve is located on either
side of the mean of 500 hours. Thus, we get that the probability that the random
variable will take on value higher than 500 is one half, or 0.5.
ii) A similar argument shows that the chance is 0.5.
X

Why don't you try some exercise now.

E5) Suppose X is a continuous random variable defined on [2,4] and f is a function of
[2,4] such that
f(x) =
i)

{ 0,
3

for2<x<4
elsewhere

Draw a rough sketch of f(x),

ii) Does f define a probability density function of .f? if so why?

E6) Classify the r.v.'s given at the beginning of Sec. 3.1-as discrete or continuolls.
In the following sections we shall discuss some standard distribution.

3.3 BINOMIAL DISTRIBUTION

One of the important discrete random variables (or, discrete distributions) is the
binomial variable. In this section we shall discuss this random variable and its
probability distribution.
Many times we have to deal with experiments where there are only two possible
outcomes. For example, when a coin is tossed, either head or tail comes up, seed either
generates or fails to generate, a newborn is either a girl or boy.
Let us consider such an experiment. For example, consider the experiment of tossing a
fair coin 3 times. This experiment has certain characteristic. First of all, it involves
repetition of three identical experiments (trials). Each trial has only two possible
outcomes - a head or tail. We call outcome "head" success and outcome "tail" a failure.
All trials are independent of each other. We also know that probability of getting a head
in a trial and probability of gening a tail in a trial are bdth f .

and

do not change from

This shows that the probability of a "success" and of a "fail~re'~
one trial to another.
If X denotes the total number of heads, obtained in 3 trials, then X is a random
variables which takes values from (0, 1,2,3).
Suppose that p denote the probability of a success (i.e. getting a head) and q denote the
probability of failure (i.e. getting a tail).
Then regarding the above experiment, we have observed the following:

69

Statistics and Probability

1) It invokes a repetiticm of n idmica! hials (Here n = 3).

2) The trials are independent of each other.
3)

Each trials has two possible outcomes

3) The probabilities of a "success" (p) and of a "failure" (q) do not change.

If you go back for a moment to Sec. 3.1, you will see that we have already obtained the
probability distribution of this in Example 3. Let us look at the probabilities once again.
P[X = 01 = P[getting three tails]
= P[T,T,T] = q x q x q

Similarly,
P[X = 11 = ~ [ [ ~ l , P H T l , ~ l I
= P [TI'H] P[THT] P[HTT]

+ +

= q2p q2p q2p = 3q2p

Similarly,
P[X = 21 = P[THH] PlTHH]
= 3p2q

+ P[HHT]

and
P[X= 31 = p3
In fact the probability P[X = r],r =: 0,1,2,3 gives that if we toss a coin three times,
how maqy ways, or combinations, will yield r heads and n - r tails.
Now you recall from your school mathematics that the number of combinations of n
objects taken r at a time is calculated by the formula

In the case of tossing of three coins, ~ 3Therefore,

.
we rewrite the probabilities as
P[X = 01
P[X = 11
P[X = 21
P[X = 31

= c(3,0)~~=
q ~q3- ~
= ~ ( 3l)p1q3-1
,
= 3W2
= c(3, 2)p2q3-2 = 3p2q
= C(3, qp3q0 = p3

This suggests that the probability p[X = r] = p, for a given r can be calculated using
the formula

(3)

Pr = C(n, r)prqn-r
where r = number of successes
n = number of trials made
p =.probability of success in a trial
q = 1 - p =: probability of failwe in a trial.

Why don't you check this formula for n=5, i.e. tossing of a coin 5 times. For example,
try this exercise.
-

- -

E7) In the experiment of tossing a coin 5 times find the probability of getting 3 heads
and 2 tails. Verify that this probability is given by the formula given in Equation
(3).
Let us now sum up the points we have observed in the example, above.

An experiment consisting of n trials is performed such that

i)

each trial has two possible outcomes, viz., a 'success'@) and a "failure9'(q);

Probability Dbstdbutions

ii) the probability of success, p, is the same for any trial;

iii) the outcomes of different trials are statistically independent (i.e. the trials are
independent).

These trials are called Bernoulli trials.

......"

The sample space of this experiment consists of elements like "SSFSSF..

of
length n of 'S's and 'F's where S stands for the success and F stands for the failure.
Let X represent the number of successes (in any order whatsoever) in the set of n trials.
Then X is a discrete random variable taking integral values 0, 1, . . . ,n. The
probability of the event P[X = r] is given by

Binomial Distribution

P[X = r] = pr = C(n,r)prqn-'
where r = exact number of successes
n = number of trials made
p = probability of success on a trial
q = 1 - p = probability of failure on a trial and
n!
C(n,r) = (n - r)!r! (The earlier example illustrates how we got this formula for p,).
Such a random variable X is called a binomial random variable and its probability
distribution is called binomial distribution and is given by Eqn.(2).

Problem 2: A sales representative, calls on four potential clients. The probability that
she will obtain an order from each of them is and whether or not she obtains an order
from one of them is statistically independent of whether or not she obtains an order
from any of the others. What is the probability distribution of the number of orders she
will receive?
Solution: We note that there are two mutually exclusive events (obtaining an order or
no order) each time she makes a call and the probability of an order 112 each time.
Also the outcomes of the calls are statistically independent. Therefore this is a situation
where there are four Bernoulli trials and where the probability of a success (an order)
1

equals 112. Substituting n = 4 and p = - in Eqn.(4), we get that

2

Thus, the probability of no orders is 1/16, of one order is 114, of two order is 318, of
three orders is 114, and of four orders is 1/16.

Problem 3: It has been claimed that in 60% of all solar heat installations, the utility
bill is reduced by at least one-third. Accordingly, what are the probabilities that the
utility bill will be reduced by one-third in
i) four of five installations?
ii) at least four of the five installations?

James Bernoulli was a

seventeenth century swiss
mathematician who
performed some of the early
work on binomial

Statistics and hbabifity

Solution Here the random variable follows binomial distribution with p = 0.6, r = 4
and n=5.
To find (i), we have to calculate P[X = 41, which is given by
P [ X = 4 ] = ~(5,4)(0.6)~(0.4)
= 0.259
Now to find (ii), we have to find the probability that X is at least 4. This probability is
the sum of the probabilities that X = 4 and X = 5 because 'at least 4 means 4 or more'.
f l u s we have to find p[X = 41 f P[X = 51.
P[X = 51 = (5,5)(0.6)'
= 0.078

.; the required probability = 0.259 + 0.078 = 0.337.

Binomial distribution is very applicable in situations where we have to decide whether
to accept a lot of goods (items) coming out of of a manufacturing process. This decision
is based on how many defective items are in the lot. Companies (or firms) will generally
return the entire items if there is evidence that more than certain items is defective. To
make such decision, let us see how we can make use of the binomial distribution.
An item coming out of a manufacturing process can either be defective or non
defective. Consider a lot of N items produced by the manufacturing process. Let m of
these be defective. Suppose a quality control inspector draws a random sample of n
items from the lot, one by one, with replacement (i.e. an item drawn is put back in the
lot, after noting down whether it is defective or non defective, before the next item is
drawn at random). Let X be the number of defective items drawn by the inspector. Note
that there are n trials and in each trial the probability that a defektive item is picked
remains the same, namely ,; as the drawing is done with replacement and at random.
Also note that the trials are independent. Therefore, the random variable X defined
above is distributed as a Binomial (n,p) where p = g.

By now you must have got some idea for recognising those situations where we can
apply binomial formula. If we can apply binomial distribution to study a situation, then
we say that the situation can be modelled by binomial distribution.
I

Here are some exercises for you.

E8) A farmer buys a quantity of cabbage seeds from a company that claims that
approximately 80 % of the seeds will germinate if planted properly. If four seeds
are planted, what is the probability that exactly two will germinate?
E9) Consider again the data collected by Sunil, the newspaper boy. When Sunita, the
statistics student, saw the data, she started wondering if the number of customers
from among his ten irregular customers, who actually buy from him on a given
day, will follow a binomial distribution? What do you think? Under what
conditions will this random variable follows a binomial distribution.
E10) Sunita was still glancing through Sunil's diary wondering to herself if she could
think of the 10 customers as 'ten identical coins', when she noticed something
significant. She noticed that a lot more of the sequences had a 1 in the third
position than in the 8th position. Sunil remembered that customer 3 was the
management trslinee whom he called 'Alka Didi'. She was from a neighbouring
town and was undergoing training in a software company. She was interested in
news about software companies, science and environmental issues. She would
often buy from Sunil but not always. Customer 8, Sunil told his sister, was a
mysterious yoyng man by name Kapil, who was rumoured to be working for a
detective agency. One could rare$ find trim in the morning hours and if he was at

*
home in the morning hours, he would certainly buy fiom Sunil.

Probability Distributions

Given the situation above, do you stiM think that the nurilber of sales on a day can
be modelled as a binomially distributed random variable'?, Give reasons Fcr your
answer.
-

Once we have the probability distribution, we naturally ask what is the 'expected
value'. We shall see that now.

Expected Value of a Binomial Variable

We have already seen in Sec.3.2 that for a discrete random variable X, the 'Expected
Value' E(X) is

where xo, XZ,. . .are the values assumed by X and po, pl , . . . are the probabilities
associated with these values i.e.

If X is a binomid r.v., taking values 0, 1, . . . , ,n, then, we know that

I

We rewrite this expression in the sum notation C (called sigma) as

n

j= 1

Those who are familiar with binomial expansion can recognise that the second
expression on the R.H.S. is 1 - p p)"-'. Therefore we have

E(X) = np [ l = np.

This means that the expected number of successes is nu.

IE

Let us do a problem.

Problem 4: An oil exploration firm plans to drill six holes. It is believed that the
probability that each hole will yield oil is 0.1. Since the holes are in quite different
locations, the outcome of drilling one hole is statistically independent of that af drilling
any of the other holes.
(a) If the firm will be able to stay in business only if two or more holes produce oil,
what is the probability of its staying in business?
(b) Give the expected value of the number of holes that result in oil.

Solution: (a) If the firm can stay in business only if two or Inore holes produce oil, it
follows that the probability that it will stay in business cqunls i minus the probability
that the number of holes resulting in oil is 0 or 1. Eat11 Ilole drilled can be viewed as a
Bernoulli trial where the probability of success is . I . Thus. the probability that the
number of successes is 0 or 1 equals:

P(0 orl)

= P(0)
= .531

6!
O! 6!
+ P(1) = -(.96)
+
-(.1)(.9')
0!6!
1!5!

+ ,354= .885.

71

Statistfcs and Probability

Consequently, the probability that the firm will be able to stay in business is
1 -.885 = .115.
(b) The expected value of the number of holes yielding oil is 6 x 0.1
and p = .I.

= 0.6,

since n = 6

A problem with the binomial distribution is that if the number trials 'n' is very large
and probability 'p' is very small, computation of P[X = r] is cumbersome.
The distribution which we introduce in the next section may be useful in such a
situation.

3.4 POISSON DISTRIBUTION

Poisson
A nineteenth cectury swiss
mathematician.

In this section we introduce you to another discrete distribution called 'Poisson

distribution'. We will familiarise you with different situations where we can apply this
distribution. Let us try to understand this distribution through an example.
Suppose it is the the busy Friday noon hour at a bank, and we are interested in the
number of customers who might arrive during that hour, or during a 5-minute 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.
As in the case of binomial , here also we make some assumptions.
1) The average arrival rate at any unit time remains the same over the entire noon
hour.
2)

The number of arrivals in a time interval does not depend on what happened in
previous time intervals.

3)

It is extremely unlikely that there will be more than one arrival in a very short
interval of time. That means that it is impossible for more than one customer to get
through the revolving entrance door in a fraction of a second.

Under these assumption we find the required probability. For this we make use of the
following formula known as Poisson formula, given by

where X is the Greek letier lambda which denotes the average arrival rate per unit ~f
time and t is the number of units of time is the number of arrivals in t units of time
41so we know that X = 72 arrivals per hour is a constant for this situation. Since in the
question X is given in 'hour', to standardise the unit, we have to find 't' in hour.
i.e. 60 minutes = lhour

:. t = -hours
20
Then

To find P(4), we use the Table 2, given in the Appendix. This table shows p(x) for
selected values of A.

From-thetable, we get

What does this value 0.191 specify? This tells us that if the m v a l s are arrivals of
customers at a bank, there is 19.1% chance that exactly four customers will arrive in the
next 3 minutes.
If we vary the values of x and t, we can get different probabilities. This gives the
probability distribution which is called Poisson probability distribution.
In the above discussion we saw that the Poisson formula is applicable only if certain
conditions are specified. We re-state the formula now.

Poisson Formula
The Poisson Formula is given by

Poisson Distribution

where X is used to compute probabilities for the number of occurances in an interval of

time, if the occurrences have the following characteristic.

I)

the average occurrence rate per unit of time is constant

2)

occurrence in an interval is independent of what happened previously.

It rarely happens that there will be more than one occurrence in a very short
time interval
A distribution having probabilities given by Poisson Formula is called Poisson
distribution.
3)

I
I
I

Now let us see some situations where we can apply Poisson distribution . Here is an
example.

Problem 5: Calls at a telephone switch board occur at an average rate of six calls per
10 minutes. Suppose the operator leaves for a 5-minute coffee break, what is the
probability that exactly two calls come in (and so go unanswered)while the operator is
away?
Solution :Here you can check that the conditions 1.2 and 3 of the Poisson formula are
satisfied in this case. Therefore we can use the formula. Now that here X = $, In this
case t = 5 so that Xt = 3. Hence the required probability P(2) is given by

That means there is 0.2240 chance that two calls go unanswered.

I

Here are some exercises for you.

E l 1) If a bank receives on an average X = 6 bad checks per day, what is the probability
that it will receive 4 bad checks on any given day.
E12) A hospital has 20 kidney dialysis machines and that the chance of any one of them
malfunctioning during any day is .02. We want to find the probability that exactly
3 machines will be out of service on the same day. Then,
i) can we use the binomial farmula to find this probability? If yes, calculate the

Statistics and Probability

ii) can we use the Poisson formula to find this? If yes calculate the probability.
In the above exercise we have seen that the difference between the two calculations is
very small.
The Poisson formula can be used to approximate the binomial probability of r successes
in n binomial trials in the situations where n is large and probability of success 'p' is
small.
For instance, suppose we are interested in number of road accidents in a metropolitan
city or daily number of machine breakdown in a work shop etc., during a specified
interval of time. Each of these subintervals is so small that at best one and no more
occurrence happens within it. 'l7-m~we may look upon each subinterval as a trial. Each
trial leads to a "success" if the occurrence happens during that subinterval and to a
"failure" if the occurrence does not happen.Assume that the occurrences are
independent of each other. Hence, the total number of occurrences can be constructed
to be distributed binomially, the total number of trials being equal to the number of
subinten~alswhich we have ensured to be large; also, the length for each subinterval
being small, the probability of an arrival (success) ns likely to be small.
'Thus we have seen that there are situations where both binomial and Poisson are
applied. The rule of thump followed by most statisticians is that if n _> 20 and
p 0.05, then Poisson formula can be used to calculate binomial pobability.

<

It is clear that the Poisson calculation is simpler than the binomial calculation. An
advantage of the Poisson distribution. , if it is applicable, is that it has only one
parameter, A, whereas the binomial distribution has two parameters, n and p;
consequently, Poisson probabilities can be tabulated more compactly than binomial
probabilities. For example, the Poisson probability P(3) is the same for n = 200,
p = 0.01 as it is for n = 100, p = 0.02, and for any cther pair of n and p values whose
product is X = np = 2.
By now you must have got a fairly good idea where the Poisson formula can be used.
In all the situation we have considered so far, we have calculated the probability over an
interval of time. But there are situations where we need to calculate probability over a
region (or space) or something else as o m physical reference. In the following example
we given such a situation and illustrated how to use Poisson distribution to calculate the
probability.

Example 5: During second world war,a v-3 rocket hit in South London. Later a study
was conducted on'what are regions not affected by the rocket hit.Let us see how they
used Poisson distribution for this study.
They took X as the average number of hits per unit area (Note that earlier in the formula
X was average rate per unit time).Instead of the variable't' they replace the variable 'v',
and x denotes the number of hits per unit area. Then they assumed that all the
conditions to satisfy the Poisson formula holds in this case.With all this assumptions,
they calculated the probability using the formula

According the problem stated, they have to calculate the probability of 'no hit' per unit
area. That is, the x = 0 and v = 1, so that Xv = A. Now, to calculate A, what they did
was, they divided the area into 576 areas of equal size (the number 576 is chosen based
on some other study and they found that they were 537 hits).
537
:. the average number of hits per unit area X = - = 0.9323
576
Then the required Probability is

This means that if we take one region, then the probability that the region is not hit by
the rocket is 0.3936. Hence, out of 576 regions, the number of regions not hit by the
rocket is given by

Now, the actual number got from the record was that there are 229regions not hit by the
rocket. This number is quite close to 226. This shows that the values got using Poisson
formula are very close to the actual values.
Thus we saw that the Poisson distribution is very effective in studying various real-life
problems where the occurrence is very rare.
One of the main disadvantages of this distribution is that it is applicable only in
situation where the outcomes are independent i.e. each outcome is independent of what
happened previously.
In the next section we shall discuss another standard distribution.

3.5 UNIFORM DISTRIBUTION

The uniform distribution is the simplest of a few well-known continuous distributions
which occur often.
As we have seen in Sec.3.2 in the continuous case we are interested in behaviour of the
variable in the subintervals of the sample space; rather than at single points. If for
example, the sample space is what we call the unit interval [0, 11, and we set the random
variable X as a value selected from this interval, then we are no longer interested in the
outcome of the kind {X = a), but rather outcome of events of the kind {a < x < b) i.e.
values lying between the two numbers a and b, where 0 5 a 5 b 5 1.
Suppose X is a random variable such that if we take any subinterval of the sample
space, then the probability of this interval is the same as the probability of any other
subinterval of the same length. The distribution corresponding to this r;v. is called a
uniform distribution. As the name suggests the probability is uniform along
subintervals.
Let us see some examples of such sample spaces.

Example 6: A train is likely to arrive at a station at any time between 6.10 p.m. and
6.40 p.m. The time the train reaches, measured in minutes, after 6 p.m. is a random
variable X. Here X can take any value between 10 and 40 minutes. Therefore the
sample space is the interval (10,40). It is reasonable to assume that the likelihood for X
taking any value between 10 and 40 is equal. So if we take subintervals of equal
lengths, then the probability will be the same. The distribution corresponding to this r.v.
is uniform over the interval (10,40).

***
Example 7: An office fire drill is scheduled for a particular day, and the fire alarm is
likely to ring at any time between 9 a.m. and 5 p.m. The time the fire alarm starts,
measured in minutes, after 9 a.m. is therefore a random variable which takes any value
between 0 and 480 (= 8 hours = 8 x 60 = 480minutes) equally. The distribution
corresponding to this r.v. is uniform.

***
Now, why don't you look for such samplz spaces on your own. Try this exercise now.
E13) Verify whether the following situations can be described by uniform distribution
or not?

Probability Distributions

Statistics and Probability

a)

The average life span of a life bulb produced by a manufacturing company.

b)

The number of defective items produced by an assembly process.

Next we will see how we can define (calculate) the probabilities for this distribution. As
we have seen in Sec. 3.2 , in the case of a continuous distribution, the probabilities are
calculated using a function called 'probability density function' (p.d.f.). The p.d.f. for
uniform distribution is given as follows.

Definition 5: The pd.f. of a random variable X which is distributed uniformly in the

interval [a,b], where a < b is given by

I0,

otherwise

We can easily draw the graph of this distribution. It is given in Fig.8.

Fig. 8: draph of P.d.f. of a uniform distribution

Now let us see how we calculate different probabilities for this distribution. As stated in
Sec.3.2, for a continuous r.v., we calculate the probability of an interval rather than a
point. For example, what will be P[c < X < dl where a < c < d < b? We have seen
that it is given by the area above this interval and under the graph. The area is shown in
Fig.9.

Fig9 P [c < X < d] =Area of the rectangle shown

1
So, essentially it is the area of thc rectangle with length d - c and height = -i.e.
b -a
P[c < X

< d] = ( d - C )

1
x -

b -a
For example, if we take the situation in Example 4, let us find the probability that the

Pmbabiiity Diiributiollc

alarm sounds between 1 p.m. and 2 p.m. Here the pdf,

= 0, otherwise

To find the required probability, you have to find time elapsed in minutes between 9
a.m. and 1 p.m. and between 9 a.m. and 2 p.m.
For 1 p.m. this is 4 x 60 = 240 minutes.
Similarly, for 2 p.m., it is 5 x 60 = 300 minutes.
Therefore you have to calculate the probability P[240 < X < 3001. This is given by the

-10

This area is the rectangle with base 60(= 300 - 240) and height

&.

That is there is U.5% chance that the alarm sounds between 1 p.m. and 5 p.m. [Some
of you may think that this fact was rather obvious from the statement of the-problem
itself. But we have given this situation as an illustrative example. There are situations
which are complicated, where we can easily calculate the probability asing this
distribution.]
Next we state below the expected value of this distribution.

You can try this exercise now.

E14) Suppose that the weight of sugar obtained by processing a tank of sugar cane juice
is uniformly distributed with a mean of 10 kg. and range of 1.8 kg. Then
i) What are the largest and smallest weights of sugar obtained from a tank of
sugar can juice?
ii) What is the probability that a tank of juice will yield sugar weighing between
9 kg. and 10.5 kg.?
E15) A train is due to arrive at 5.30 p.m. but in practise is equally likely to amve at any
time between 2 minutes early and 30 minutes late. Let the time of.arrival
(expressed as minutesgrom due time) be X. Sketch the pdf f(x) of the r.v. X and
shade the areas given bellow
1) The probability that the train is less than 10 minutes late.

Statistics and Probability

2) The probability that the train is late, but less than 16 minutes late.
Next we shall discuss another continuous distribution which is widely used in statistical
problems.

3.6 NORMAL DISTRIBUTION

Normal distribution' is a class of distribution which can be used to study the probability
distribution occumng frequently in real-life situations, of biology, manufacturing
machines, psychology etc.
A particular form of this distribution was found by seventeenth - eighteenth century
mathematicians Abraham De Moivre and Pierre Laplace, while they were working on
various problems in probability. They found that the distribution corresponding to
certain random variables had got special property that when graphed, a bell-shaped
curve is obtained and came to be called the normal pattern. The graph of the pattern
became known as normal curve. Later this class of distribution was studied extensively
by another mathematician Karl Friedrich Gauss and therefore this became known as
'Gaussian distribution'.
We shall now state the distribution.

Definition 6: We say that a random variable X is normally distributed with parameters

p and a if the probability density function f(x) of X is given by,

where p is a real number lying between -oo and oo and a is a real number lying
between 0 and CQ.
The function f(x) may look rather formidable to you at first sight. At this stage we just
ask you to notice that it involves two parameters, a and p. Corresponding to each pair
(p, u), we get a distribution. Therefore there is a whole family of distributions, each
one specified by a particular pair of values for a and p.
The most important characteristic of this distribution is that the graph of pdf, f(x) for a
particular value of p and a is bell-shaped as shown in Fig.11.
The probability density function, pdf is also symmetrical about the mean p. The word
symmetrical means that the two halves of the curve are mirror images (see Fig.11). In
Fig. 11 you note that if we place a mirror on the dashed vertical line ( which occurs at
75 in Fig. 11) then the mirror image of the portion on the left is the same as the portion
on the right side.

Both p and a have a 'nice' interpretation. We have already said that the pdf is
symmetric about p, SO it is no surprise that p is the mean of the distribution. The
other constant, a2 dictates how spread out and flat the 'bell-shape' is and in fact u2is
the variance of the normal distribution.

As an illustration, the following figure shows that the normal pdfs for p and a are given
as follows:
A /L = 10,u = 1

p = 10,u=2

p=lO,u=3

p=15,u=l

10

15
Fig.12

Pdfs A, B and C all have the mean 10 and so they are all centred at x = 10. Of these
three curfes, C has the largest variance and so is the most 'spread out'. Curve I3 has a
smaller variance and so is less spread out, and curve A has the srriallest variance and so
is the most 'squeezed in'. Curves A and D have the same variance and so they have
exactly the same shape, but they have different means so they are centred at x = 10 and
x = 15 respectively.
Some notation

As a normal distribution is entirely specified by its parameters p and u we denote such

distribution by N ( p ,02)'where p is the mean and a2is the variance. So, for instance,
the curve shown in (A) above is the pdf N(10,l) the curve in (B) is the pdf.N(10,4)
and so on.

The standard normal distribution

The normal distribution with mean p = 0 and variance a2 = 1, is called the standard
nonnal distribution. Z is the notation usually used far a random variable which has this
distribution. A graph of the standard normal pdf, p(z) is shown in F1g.13.

-4

-3

-2

-1

Fig13

Notice that most of the area under the standard normal curve lies between -3 and +3.

Calculating Probabilities
The normal distribution is continuous and so the probability that the random variable X
lies between the interval (a, b) is are calculated by obtaining the area under the pdf
curve between a and b.

Probability Distributions

Statistics and Probability

For example, suppose an individual's IQ s

c
m X has a normal distribution with
p .= 100 and standard deviation a = 15. Fig. 14 shows the areas under the pdf which
correspond to P(X < 85) and P(115 < X < 120).

Rg.14

Unfortunately there are no 'nice' formulae for calculating such areas. But there are
tables available from which we can find out the area. Statistical software are also
available by which we can calculate the area.
Because the number of possible values for p and a is unlimited, the number of different
normal distributions is unlimited. However, probabilities for every normal distribution
can be obtained from a table of probability for standard normal distribution.
We shall first discuss how to use the table for calculating probabilities for a standard
normal distribution. Then we shall discuss how to use this to find the probability for
any normal distribution.

Using tables to calculate normal probability

We defiote by F(a) = P[Z 5 a], the probability that 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 a table. One such table is given in Sec. 3.9 Appendix.
Note that the entries in the table are the values of z for z=0.00,0.01,0.02,3.49. To find
the probability that a random variable having the 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, we also make use of the identity

In the following exampie we illustrate how we use the table to calculate different
probabilities.

Example 8: Suppose want to find the following probability

i) P[0.87 < Z < 1.281
ii), P[-0.34 < Z < 0.621
iii) P[Z 2 0.851
iv) P[Z 2 -0.651
We proceed as follows.
i) We know that
To find F(1.28), we find the row where 2k1.2, then move across that iow to the
column headed 0.08 and found the entry 0.8997. Similarly we catr find thatF(0.87)
= 0.8078. Then the requed Probability is

CI

Probability Distributions

ii) Similarly,
P[-0.34 < Z < 0.621 = F(0.62) - F(-0.34)
= F(0.62) - [l - F(0.34)]
by the identify F(z) = 1 - F(z).
= 0.7324 - (1 - 0.6331)

= 0.3655.

iii) From the previous unit (Unit 2), you have already learnt that

Hence we have
P[Z > 0.851 = 1 - P[Z 5 0.851
= 1 - F(0.85)
= 0.1977.
v)

As in (iii), we can write

P[Z > 0.65 = 1 - P[Z 5 -.0.65)]
= 1 - F(-0.65)

In the following exercise we ask you to find certain probabilities using the normal
distribution table.
E16) If a random variable has the standard normal distribution, find the probability that
it will take on a value
i) less than 1.50
ii) less than - 1.20
iii) greater than - 1.75
E17) A filling machine is set to pour 952 ml (rnillimetres) of oil into bottles. The
amounts of 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 a
bottle contains oil between 952 and 9$6 ml.
Next we shall see that how to use the standard normal probability table to calculate
probability of any normal distribution.
Standardising

i
L

Any normal random variable X, which has mean ,u and variance a2can be standardised
as follows.
Take the variable X,and
i) subtract its mean, p and then
ii) divide by its standard deviation, a.
We will call the result. 2. so
Z=- x - P
a

For example, suppose, as earlier, that X is an individual's IQ score and that it has a
normal distribution with mean p = 100 and standwd deviation a = 15. To standardise

Statistics and Probability

an individuals IQ score, X, we subtract p = 100 and divide the result by u = 15 to give,

In this way every value of X, has a corresponding value of Z. For instance, when
-2andwhen~=90,~=-=-0.67.
X = 1 3 0 , Z = T 130-100
-

The distribution of standardised normal random variables

The reason for standardising a normal random variable in this way is that a standardised
normal random variable

z=-x - P
u

has a standard normal distribution.

That is, Z is N(0,l). So if we take any normal random variable, subtract its mean and
then dividz by its standard deviation, the resulting random variable wiil have a standard
ncjrmal distribution. We are going to use this fact to calculate (non-standard) normal
probabilities.

Calculating probabilities
With reference to the prob!em of IQ score, suppose we want to find the probability that
an individual's IQ score is less than 85, i.e. P[X < 851. The corresponding area under
the pdf N(100,15~)is shown in Fig.15.

'fi-e cannat use r,omal tables directly because these give N ( 0 , l ) probabilities. Instead,
we will convert the statement X < 85 into an equivalent statement which involves the
because we know it has a standard nonnal distribution.
standardised score, Z =
We start with X =85. To turn X into Z we must standardise 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
stalement, not X < 85). 'Standadising' both sides gives, X-100 < 85-109
T.
The left hand side is now a standard normal randoin variable and so we can call it Z,
and we have

84

So we have established that the statement we started with, X < 85 is equivalent to

Z i-- 1. This means that whenever an IQ score, X, is less than 85 the corresponding
standardised score, Z will be less than - 1 2nd so the probability we are seeking,
P[X < 851 is the same as P[Z < - 11.

P[Z < -I]. 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 PIX < 853 = 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 < 851 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.
P[X < 851 = P

<

15
= P[Z < -11 = 0.1587

Let us do some problems now.

Problem 6: For each of these write down the equivalent standard normal probability.
a) The number of people who visit a historic 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.
The number of cheques processed by a bank each day is normalljr distributed with
a mean of 30,100 and a standard deviation of 2450. Consider tle probability that
the bank processes more than 32,000 cheqiles in a day.
Solution: Here we want to find the standard normal probability corresponding to the
probability P[X < 90001.
X - 10500 9000 - 105001
a) We have P[X < 90001 = P
<
= P[Z < -2.51.
600
b) Here we want to find the standard normal probability corresponding to the
probability P[X > 320001.
b)

X -- 30100 > 3 2-0-0 - 30100] =

P[X > 320001 = P -2450
2450

> 0.781

Note Probabilities like P[a < X < b] can be calculated in the same way. m e only
difference is that when X is standardised, similar operations must be applied to both a
and b. That is, a < X < b becomes
a - p X
< - p< - b - p
u

Cr

which is
a - p < ~ < b- - P
u

1
i

Problem 7: An individual's IQ score has a N(100, 1S2)distribution. Find the

probability that an individual's IQ score is between 91 and 121.

Solution: We require P[91 < X < 1211. Standardising gives

The middle term is a standardised normal random variable and so we have,

Probability Distributions

Statistics'and Probability

Try these exercises now.

E18) A flight is due at Palam airport at 1800 hours. Its arrival time has a normal
distribution with mean 1810 hours and standard deviation 10 minutes.
a) What is the probability that the flight arrives before its due time?
b) Passengers must check in for a connecting flight by 1830 at the latest. What
is the probability that passengers from the first flight arrive too late for the
connecting flight? (Assume no travelling time from aircraft to check-in.)
E19) The length of metallic strips produced by a machine has mean 100 cm. and
variance 2.25 cm. Only strips with a weight between 98 and 103 cm. are
acceptable. What proportion of strips will be acceptable? Yoq may assume that
the length of a strip has a normal distribution.
With this we come to an end of this unit.
Let us now summarise the points we have covered in this unit.

3.7 SUMMARY
In this unit we have covered the following points
1) A random variable is a variable that takes on different numerical values according
to chance outcomes
2)

There are two types of random variables - discrete and continuous;

3)

A probability distribution gives the probabilities with which the random variables
take an various values in their range.

4)

We have discussed three standard distributions:

a)

Binomial Distribution. The probabilities of an event P[X = T] in this

distribution is given by

P[X = r] = C(n, r)Pqn-'

b)

Poisson distribution: The probability of an event PIX = x] in this

distribution is given by

where X is a constant for a particular situation.

c)

Uniform Distribution The probability density function is defined by

1
-

b-a'
0,

ifa<xLb
elsewhere

The probability P[c < X

d)

< d] =

d-c
6-a

Normal distribution. The probability for this distribution is calculated by

finding the area, under the cuee of a fmction called probability density

function defined by

El) a) If X denote the number of correct answers, then X is the random variables for
this situation.

--

b)

X can take values 0,1,2, . . . up to 50

c)

P[X = 401 means the probability that the number of correct answers is 40.

Probability Distributions

E2) 1 and 2 is not discrete. 2 and 3 are discrete.

(1) is not discrete because it takes values in aii interval.
( 2 ) is discrete because the number of accidents is finite. Similarly, you argue for
the situation in (3).

E3) Let X denote the amount you win or lose. Then X takes values Rs. SO, 0 or - 10
(loss in Rs. 10). The probability that both the marbles are green is 119. The i.e.
P[X = 501 = 119. The probability that both the marbles are red is 419 i.e.
PIX = - 101 = 419.
The probability that the marbles are of different colou'r is 419 is i.e. P[X = 01 =
419.
Thus the probability distribution is as given in the following table.
Amount (in Rs. won (+) ot lost(-)
50
0

Probability
1/4
4/9

E4) He has to calculate the mean. It is given by

0 x 0.5+ 1 x 0.15+2 x 0.35+4 x 0.12 x 5 x 0.18
Mean =
0.05 0.15 + 0.35 + 0.25 0.12 + 0.18

This means that he can expect that on an 2 average cars will be sold per day over
long run (or more precisely 5 cars will be sold over (2 days)).

Fig.16

ii) The area under the graph and above the interval [2,4] is the area of the
rectangle shown in Fig.16 which is given by
Area = 2 x = 1.
:. f defines a probability density function of X.

E6) 1,3,4,5,6 are discrete. 2 and 7 are continuous.

E7) Here we have to calculate Probability of 3 heads and 2 tails. That is P[X = 31.
Tbe possibility of getting 3 heads and 2 tails in five tosses of a coin is given by
[ H H m I , [HHTHTI, [HHnHI, [HTHHTI, [HTTHHI, [THTHHI, [ m l ,
[TI-IHHT], [THHTH], [THTHH], [HTHTH].
Each of these events are having probability p3q2 and there are 10 such events.
Therefore we get

PIX= 31 = 1 0 ~ ~ ~ ~ .

87

If we apply the formula in Equation (3) to find P[X = 31, then we substitute
r = 3, n = 5 in the formula, and we get

PIX = 31 = ~ ( 5 . 3 ) ~ ~ ~ ~

E8) This situation follows binomial distribution with n=4 and p = #$ = $.The
random variable X is the ~unlberof seeds that germinate. We have to calculate the
probability that exactly two of the four seeds will germinate. That is P[X = 21.By
applying binomial formula, we get

Therefore the required probabiliw Is 0.154.

E9i If Xi denote the random variable that the ith customer buys the piiper on a given
dzy, then Xi's may not be identically distributed. Therefore Xi's may not be
binomially distributed. But if the customers are having the same business
activities or same kind of habits or working nature, then we can expect that Xi's
wiil be identically distributed. In such situatiorl we can expect that Xi's will
follow bi~omialdistribution.

+ + +

E10) The number of sales in a day is actually X1 X2 . . . Xio where each Xi Is

either 0 or 1 depending on whither customer i buys the paper or not on the given
day. Now since customer 8 is more likely to buy on a day, than customer 3, X3
and X8are not identically distributed. That is, PIXs = 11 > PIX3 = 11. Therefore
X1 X2 . . . Xlo cannot be thought of as binomially distributed random
variable.

+ + +

E l 1) Since the problem deals with the receipts of bad cheques which is an event with
rare occurrence over an interval of time (a day, in this case), we can apply Poisson
distribution.
Since on an average 6 bad cheques are received per day,
Substituting X = 6 ar,d x = 4 in the Poisson Formula, we get
64e-6
1296 x (0.0025)
P[X=4] = -4!
2.4

E12) Note that here the experiment or trial is 'checking the machine for its functioning.
There are 20 trials and each trial is identically distributed with probability 0.2.
i) The trials are independent also. Therefore we can apply binomial formula.
We are required to calculate P[X = 31. Then

ii) Here we have to check whether we can apply Poisson distribution.

Note that here the occurrences are 'function of the dialysis machines'. Then the
average rate of machines that go out of service in a day is a constant
X = 20 x 0.02 = 0.4.
88

Also note that we can make the subintervals so small that at best only one
machine go out of service. Thus conditio; (2) and (3) are satisfied. Therefore we
I

.,cab,apBIp Poissoa F d a tc calculat: the required probability P(3). Then

a(q

(0.4)~e-Q.~
=
,
31

fjM)j) The mean, 10 is the centre point of a line segment whose length is the range,
1.B kg. Hence, the line segment extends x (1.8)= 0.9kg. to the left and to
the right of' 10 i.e. 9.1 to 10.9 kg. Hence tbe smallest weight is 9.1 kg. and
the largest weight is 10.9 kg.
ii) We
requir6d.t~calculate p[9 < X < 10.51.
= 0.833.

That is the probability that the weight lies oetween 9.1kg. and 10.9 kg. is

0.833.

El51 Here the pdf f(x) is given by

5.28 < x'< 6
0,otherwise

The sketch of f(x) is as given belaw.

0.1

1/32 -

5.28 6
,

Fig.17

Tafind fie required probability we note that X can take values in the interval
(5,28,5.40). Hence the required probability is
1
0.003
P[S.28 < X < 5.401 = 0.2 X - =
= 0.0037
32
8
El@ a) 0.9332
b) Q,1151
C)
0.9599,
i)

E17)The standard normal probability c o v n d i n g to this probability is given by

P[952 < X < 9561 =
= P[O< z < I]
= F(l)- F(0)
= 0.8413 - 0.5
= 0.343.

Probability Distributions

Statisticsand Probability

E18) Let the time of arrival in minutes past 1800hrs be X. Then X follows normal
distribution N(10, lo2).
a) The required probability is P[X < 181. The standard probability
corresponding to this is

= F(O.O1)
= 0.5040.
b)

The required probability is P [X ) 301. Then

E19) We,- have

to find the probability
_ _ . P[98 < X

So, only 13 % will be acceptable.

< 1031. The standard probability Is

Probability Distributions

3.9 APPENDIX
Cumulative Standard Normal Probabilities P[Z < a] where Z -- N(0,l).

- POISSON PROBABILITY DISTRIBUTION

This table shows the value of

for selected values of x and for p = .005

to

8.0.

Continued

.....

Continued ....

Continued

.....