Chapter 3

Discrete Probability Distributions

 Motivation
 We are often interested in obtaining one of two
outcomes – “success or failure”, “satisfactory or
defective”, “heads or tails”, etc.
 We can represent these two outcomes by a
random variable X which takes on the values 1
(“success”) and 0 (“failure”).
 An experiment that has only two outcomes is
called a Bernoulli trial. This leads us to our
first special type of discrete random variable…
 The Bernoulli Distribution
 The simplest, non-trivial discrete distribution is
the Bernoulli distribution.
 If a RV X takes the values 1 and 0 with
probabilities p and 1 – p, respectively, then X is
said to be a Bernoulli random variable (or
have a Bernoulli distribution) with parameter
p.
 If X is a Bernoulli RV with parameter p, then
E(X) = p and V(X) = p(1 – p).
 Example 1
 Consider an experiment consisting of a sequence
of three independent Bernoulli RVs each with
parameter p; X1, X2, and X3.
 Since each variable has 2 possible values, there
are 23 = 8 possible overall outcomes; that is,
S=
{000,001,010,100,011,101,110,111}.
 What is the probability that we get 2 successes;
that is, 2 values of “1”?
 The Binomial Distribution
 The binomial distribution is characterized by:
1) a fixed number, n, of Bernoulli trials…
2) …which are independent of each other…
3) …and each have the same probability p of
“success”.
 If these 3 assumptions are satisfied, then the

total number of successes X has a binomial

distribution with parameters n and p. This is
written as X ~ B(n, p) .
 The Binomial Distribution
 If X ~ B(n, p) , then its PMF is given by
n x n x
 
P ( X  x)    p (1  p ) , x  0,1,..., n
 x
with E(X) = np and V(X) = np(1 – p).
 NOTE: If X ~ B(n, p), then we can think of X as the
sum of n Bernoulli RVs. More specifically,
X = X1 + X2 + … + Xn, where each Xi
is a Bernoulli RV with parameter p.
 Example 2
 A student is writing a multiple choice test with 10
questions. Each question has 4 possible answers.
If the student answers by pure guessing, what is
the probability that they will pass the test?
 Example 3
 A company sells packages of 10 widgets and has a
money back guarantee that none of the widgets in a
pack will be defective. Each widget has a probability
of 0.005 of being defective.
 a) What is the probability that a pack of widgets will
be returned?
 b) If you buy 3 packages of widgets, what is the
probability that you will return at least one of the
packages?
 Example 4
 Suppose that a lottery claims that each ticket has a
1 in 5 chance of winning a prize. How many tickets
would you need to purchase in order to have at
least an 80% chance of winning at least one prize?
 The Geometric Distribution
 Consider an experiment consisting of a series of
Bernoulli trials. If we define the RV X to be the
number of trials required to obtain the first success,
then X has a geometric distribution with
parameter p. The PMF is given by
x 1
P ( X  x )  (1  p ) p
and the CDF is given by
x
F ( x )  P ( X  x )  1  (1  pfor
) x = 1,2,…
 The Geometric Distribution

1
E( X ) 
p
 Example 5
 Some computer software has an error in it. To locate
the error, programmers conduct a series of
independent tests. Each test is based on a set of
random inputs for the software, and the probability that
a random test detects the error is 0.16
 a) What is the expected number of tests that will need
to be performed before the error is found?
 b) What’s the probability it will be found within 10
tries?
 c) How many tests should they plan to conduct if they
want to be at least 99% sure they’ll find the error.
 The Negative Binomial Distribution

 If we define the RV X to be the number of trials

required to obtain the r th success (r = 1, 2, 3,… ),
then X has a negative binomial distribution with
parameters p and r. The PMF is given by
 x  1 xr r
P ( X  x)   (1  p ) p , x  r , r  1,...
 r 1
 The Negative Binomial Distribution
 If X is a negative binomial RV with parameters p
and r, then
r r (1  p )
E( X )  and V ( X )  2
.
p p
 NOTE: If X is a negative binomial RV with parameters
p and r, then we can think of X as the sum of r
geometric RVs. More specifically, X
= X1 + X2 + … + Xr, where each Xi is a geometric RV
with parameter p.
 Example 6
 A married couple is going to keep having children
until they have 2 boys. Assume they are equally
likely to have a boy or a girl.
 a) What is the mean number of children they must
have before they have 2 boys?
 b) What is the standard deviation of the number of
children they must have before they have 2 boys?
 c) What is the probability they must have at least 3
children in order to have 2 boys?
 Example 7
 Suppose we are drawing cards from a standard 52 card
deck with replacement. What is the probability that…
 a) …the first heart drawn will be on the 5th draw?
 b) …the second club will be drawn on the 5th draw?
 c) …we will draw a king twice in 4 draws?
 d) What is the expected number of draws until we
draw a face card (a face card is a Jack, Queen, or
King)?
 The Hypergeometric Distribution
 Consider a group of N items where r of these items
having a certain characteristic (e.g. defective).
Suppose we select n of these N items without
replacement. Define the RV X to be the number of
“defective” items. Then X has a hypergeometric
distribution with PMF
 r  N  r   N 
P ( X  x )       ,
 x  n  x   n 
for max{0, n  r  N }  x  min{n, r}.
 The Hypergeometric Distribution
 If X is a hypergeometric RV with parameters N, n,
and r, then
nr nr  N  n  r 
E( X )  and V ( X )   1  .
N N  N  1  N
 Example 8
 In a receiving inspection department, lots of 100
pump shafts are received daily. Lots are accepted
if, in a random sample of 10 shafts, no more than
one shaft is found to be defective.
 a) What is the probability a lot of 100 shafts is
accepted if it has 5 defectives in it?
 b) What is the expected number of defective
shafts in the sample of 10 shafts?
 The Binomial Approximation to
the Hypergeometric Distribution
 When N is very large relative to n (n/N < 0.10, or
N/n > 10) sampling without replacement is very
much like sampling with replacement. This is due
to the fact that, while the probability of “success”
does not technically stay constant, it does not
change significantly from draw to draw.
 In such situations it is acceptable to use the
binomial distribution as an approximation to the
hypergeometric distribution with X ~ B(n, r/N) .
 Example 9
 Refer to Example 8. We had 100 shafts, 5 of which
were defective. We chose a sample of 10 shafts
and accepted the lot if no more than one of them
was defective.
 Using the hypergeometric distribution directly we
found the exact probability of accepting a lot was
0.92314.
 Using the binomial approximation, what is the
approximate probability that we accept the lot?
 The Poisson Distribution
 Next we consider a distribution which still counts the
number of “successes” or “events” to occur.
However, rather than counting how many successes
occur within a fixed number of trials, we count how
many successes occur within a certain amount of
time, distance, volume, etc.
 In such situations, the number of successes follows a
Poisson process or Poisson distribution.
 The Poisson Distribution
 Suppose some type of event occurs randomly at an
average rate of  per unit and X represents the
number of events to occur within t units. Then X is a
RV having a Poisson distribution with parameter
t or X ~ P( t ) and has PMF

P( X  x) 
e  t
t x
, x  0,1,2,...
x!
 If X is a Poisson random variable with parameter
t then E(X) = V(X)  =t .
 Example 10
 The number of surface flaws in plastic panels used
in a car’s interior has a Poisson distribution with a
mean of 0.05 flaws per square foot. Assume a
car’s interior contains 10 square feet of plastic
panel.
 a) What is the probability that there are no surface
flaws in a single car’s interior?
 b) If 10 cars are sold to a rental company, what is
the probability that at most one of them has at
least one interior surface flaw?
 Example 11
 Suppose the number of errors in a piece of
software follows a Poisson distribution with an
average of 3 errors.
 What is the probability of no errors in the software?
 What is the probability of at least 3 errors in the
software?
 The Poisson Approximation to
the Binomial Distribution
 If X ~ B(n,p) where n is very large and p is very
small, then it is suitable to approximate
probabilities from a binomial distribution using the
Poisson distribution with X ~ P( t  np ).
 Rule of Thumb: You should apply this
approximation when n is larger than 150 and p is
smaller than 0.01.
 Example 12
 Suppose a box contains 500 iPads, and each one
has a probability of 0.005 of being defective.
 What is the approximate probability that there are
at least 3 defective iPads in the box?

 Note: The exact probability is 0.4565082. How

good was this approximation?
 Summary
 The binomial, geometric, and negative binomial
distributions are all based on a series of
independent Bernoulli trials.
 For the binomial distribution, X represents the
total number of successes in a fixed number of trials.
 For the geometric distribution, X represents the
total number of trials needed to obtain the first
success.
 For the negative binomial distribution, X
represents the total number of trials needed to
obtain r successes.
 Summary
 The hypergeometric, Poisson, and multinomial
distributions – are NOT based on independent
Bernoulli trials.
 For the hypergeometric distribution, X
represents the total number of items of interest in a
sample of size n from a population of size N that has
r items of interest.
 If n/N < 0.10, we can use X ~ B(n, r/N) as a
reasonable approximation. (The smaller n/N gets,
the better the approximation.)
 Summary
 For the Poisson distribution, X represents
the total number of events of interest to occur
within a certain unit of time, distance, volume,
etc.
 If X ~ B(n, p) with p small and n large, then
we can use X ~ P(np) as a reasonable
approximation. The smaller p and the larger
n are, the better the approximation.