Chapter 4

Probability Distribution
 Contents
 4-1. Discrete Distribution
 4-2. Binomial Distribution
 4-3. Hypergeometric Distribution
 4-4. Poisson Distribution
 4-5. Continuous Uniform Distribution
 4-6. Normal Distribution
 4-7. Gamma Distribution
 4-8. Chi-Squared Distribution
 4-9. More Distributions
 4-1. Discrete Distribution

A discrete probability distribution can be represented by a

histogram , a tabular form or a formula.

x 1 2 3 4 5 6
f(x) 1/6 1/6 1/6 1/6 1/6 1/6

 Discrete uniform distribution k

1

k
 xi ,
1 i 1
f ( x; k )  , x  x1, x2 , x3  xk
k 1 k
2 
k
 i
( x   ) 2
.
i 1
 4-2. Binomial Distribution
 Bernoulli process
An experiment consists of repeated trials, each with 2 possible
outcomes. The probability of success, p, remains constant and
the repeated trials are independent.

The number X of success in n Bernoulli trials is called a

Binomial R.V. with the distribution:

n x
b( x; n, p )    p (1  p )n  x
 x
Binomial Distribution
 Binomial expansion

n  n  n  n  n 1  n  2 n 2 n n
( p  q)    q    pq    p q    p
 0 1  2 n
 b(0; n, p )  b(1; n, p )  b( 2; n, p )    b( n; n, p ).
n
Can show  b( x ; n , p )  1 ; [q  1  p]
x 0
  np ,  2  np(1  p )
Multinomial distribution
 n  x1 x2
f ( x1, x2 , xk ; p1, p2 , pk , n )    p1 p2  pkxk
 x1, x2 , xk 
 4-3. Hypergeometric Distribution
 Hypergeometric experiment
1. A random sample of size n is selected without
replacement from N items. (dependence among trials)
2. k of the N items may be classified as successes (red balls)
and N - k are classified as failures (white balls).
The number X of successes of a hypergeometric experiment
is a hypergeometric R.V. with the distribution:
 k  N  k 
  
 x  n  x 
h ( x; N , n , k )  , x  0,1,2, n.
N
 
n
 Relationship to Binomial Distribution

nk N n
2 k k
 ,    n  1  
N N 1 N N
If n is small compare to N, then the hypergeometric distribution
approaches the binomial distribution with p = k/N.
The hypergeometric distribution can be extended to the
multivariate hypergeometric distribution:

 a1  a2   ak 
  ... 
 x1  x2   xk 
h( x1, x2 ,..., xk ; a1, a2 ,..., ak , N , n )  ,  xi  n,  ai  N .
 
N
 
n
 4-4. Poisson Distribution
 Poisson experiment
Experiment yielding the number of outcomes (eg. phone calls)
occurring during a given time interval or in a specified region.
 Poisson Process
1. No memory: the no. of outcomes in disjoint
intervals are independent.
2. Uniform rate of occurrence: the average no. of
outcomes per unit time is a constant.
3. Discrete outcomes: one outcome at a time.
 Poisson Distribution

e  t ( t ) x
p ( x; t )  , x  0, 1, 2 ,
x!
where the Poisson R.V. X represents the number of outcomes
in a given time interval t, and  is the rate of occurrence.

  t ,  2  t
[Theorem]
Let X be a binomial R.V. with b(x;n,p) ,
When n  , p  0, and   np  const.,
b( x ; n , p )  p ( x ;  )
Poisson Distribution
4-5. Continuous Uniform Distribution

1
p( x; A, B )  , A X  B
B A
f (x)
 0 elsewhere
2
A B ( B  A)
 , 2 
2 12 A B
x


 Monte-Carlo Simulation

Example: Numerical integration

A R 2

 Hit-or-Miss method

Random number sphere ：

(x, y, z) or (r, ,)
uniform distribution check
Monte-Carlo Search
 4-6. Normal Distribution

The normal distribution is often referred to as the

Gaussian distribution, defined as:
1 (1 / 2 )[( x   ) /  ]2
n ( x;  ,  )  e 
2 

  E [ X ] ,  2  E [( X   ) 2 ] x1  x2

x2
P ( x1  X  x2 )   n( x;  , ) isdxrepresented by
x1
the area of the shaded region.
 Normal Distribution in Words (by Youden)

THE
NORMAL
LAW OF ERROR
STANDS OUT IN THE
EXPERIENCE OF MANKIND
AS ONE OF THE BROADEST
GENERALIZATIONS OF NATURAL
PHILOSOPHY. IT SERVES AS THE
GUIDING INSTRUMENT IN RESEARCHES
IN THE PHYSICAL AND SOCIAL SCIENCE AND
IN MEDICINE AGRICULTURE AND ENGINEERING
IT IS AN INDISPENSABLE TOOL FOR THE ANALYSIS AND THE
INTERPRETATION OF THE BASIC DATA FROM OBSERVATION AND EXPERIMENT
 Important features of normal distribution

 Normal distribution is a good approximation for modeling

a random phenomenon, in which many independent
random factors are involved. (Central limit theorem)
 For linear systems, Gaussian inputs result in Gaussian
outputs.
 The higher-order statistics of a Gaussian R.V. can be
derived from  and 2.
 Based on the central limit theorem, we can generate the
random numbers of normal distribution.
 Standard normal distribution
To evaluate the probability of the normal distribution, we require the tabulation of normal curve areas for quick reference.
Introduce the transformation

we obtain the standard normal distribution n(z;0,1). Thus ,

 1
X 
Z  0

The standard normal curve

P ( x1  X  x2 )  P ( z1  Z  z2 ), for any value of  and  .

Table of Cumulative Probability of Normal Distribution
 Applications
 A certain type of battery lasts, on average, 3 years, with = 0.5.
Assuming normal distribution, find the probability that a given
battery will last less than 2.3 years.
(Solution)
To find P ( X < 2.3) with n( x ; 3.0, 0.5), we use
X  X 3
Z 
 0.5
Thus x = 2.3 corresponds to z = -1.4. With the table of standard
normal distribution, we can find
P ( X < 2.3) = P ( Z < -1.4) =0.081.
 Normal Approximation to the Binomial

If X is a binomial R.V. with and

, then np  2  np(1  p)
X  np
Z  lim
n np (1  p )

has the standard normal distribution, as n  .

Can use the normal distr. to evaluate binomial probability

whenever p is not close to 0 or 1 . In general, if both np and
n(1-p) are greater than 5, the approximation will be good.
 Example

 A quiz has 80 questions each with 4 possible answers. What’s the

probability that pure guess-work yields 25~30 correct answers?
[Sol]:
The probability of getting a correct answer by guesswork is p = 1/4.
Let X represents the no. of correct answers, then
30 30.5
P(25  X  30)   b( x;80, 1
4) 
24.5
n( x;  ,  ) dx
x  25
Using the normal approximation with   np  (80)( 1 4 )  20 and
  npq  (80)( 1 4 )  3 4   3.873. The corresponding standard
normal z-values are
24.5  20 30.5  20
z1   1.16 and z2   2.71.
3.873 3.873
Thus, the probability is
30 30.5
P (25  X  30)   b( x;80, 1
4) 
24.5
n( x;  ,  ) dx
x  25
 P (1.16  Z  2.71)
 P ( Z  2.71)  P( Z  1.16)
 0.9966  0.8770 (from Table A.3)
=0.1196.

Compare with the exact probability!

4-7. Gamma Distribution
Gamma function
  1  x
( )  0 x e dx , (  0)

Integrating by parts can be show, for   1

( )  (  1) (  1)

When is a positive integer n,

It can be shown that

(n)  (n  1)!

Γ(1)  1 , Γ(1 / 2)  
 Gamma Distribution

It can be show that, if X has a normal density n( x;0, ) , then

1
Y  X2  ( y; , 2 2 ).
2
has the Gamma distribution
 1
  x 1e  x /  , x  0
( x; ,  )    ( )
 0 , elsewhere

  
 2   2

If   1, it is called the exponential distribution .

 Relationship to the Poisson Process

 Using the Poisson distribution, the probability of no event

occurring up to time t is given by
p(0;t) = e –t.
Now let X be the time to the first Poisson event (eg. failure),
then P ( X  x )  e-x .

Thus, the cumulative distribution function for X is given by

P (0  X  x )  1 - e-x ,
and the p.d.f. is just the exponential distribution with = 1/:
f ( x )   e-x . (   mean time to failure)
 Application to reliability analysis
the ability of a system or component to perform its required
functions for a specified period of time.

A system contains a type of component whose time to failure is

T years, modeled by exponential distribution with mean time to
failure   5. If 5 of the components are installed, what is the
probability that at least 2 are still functioning at the end of 8 years?
(Solution)
Let X be the number of components functioning after 8 years, then X follows a binomial distribution with

 t / 5
p  P (T  8)  1 / 5 8e dt  0.2
8
5
P ( X  2)   b( x;5,0.2)  0.263.
x 2
 4-8. Chi-Squared Distribution
Another special case of gamma distribution is obtained by letting
and    / 2   2:
1
p( x; )   / 2 x / 21e  x / 2 , x  0
2 ( / 2)

where  is positive integer called the degrees of freedom.

   ,  2  2 .
Chi-squared distribution is important in variance analysis and statistical testing.
 4-9. More Distributions
Lognormal distribution
The R.V. X has a lognormal distribution if Y = ln(X) has a normal distribution.

 1 e [ln( x )  ]2 /( 2 2 ) , x  0
f ( x )   2  x
 0, x0
Weibull distribution (strength analysis)
Student’s t – distribution (unknown  )


F – distribution (variance analysis)