Selected Probability Models: Bernoulli Trials - Most Common Type of Situation Modeled

Selected Probability Models
Bernoulli trials - most common type of situation modeled:

independently repeated, two possible outcomes (S or F),
probability of success (p) is the same from trial to trial.
Bernoulli distribution P(X = 1) = p = 1 P(X = 0).
E(X ) = p, Var(X ) = p(1 p), mX (t) = E(etX )
= e0t (1 p) + et1 p = 1 p + pet .
Binomial distribution, BIN(n, p) n Bernoulli trials:
Sn = X1 + X2 + + Xn , Xi is a Bernoulli trial, i = 1, 2, . . . , n.
Sn is a number of successes in n Bernoulli trials.
P
E(Sn ) = E(Xi ) = np,
P
Var(Sn ) = Var(X1 + + Xn ) = Var(Xi ) = np(1 p).

P(Sn = k ) = kn pk (1 p)nk , k = 0, 1 . . . , n.

P
mSn (t) = n0 etSn Snn pSn (1 p)nSn

P
= n0 (et p)Sn (1 p)nSn Snn = (pet + 1 p)n .
1/1
Geometric distribution, GEO(p) X is the number of failures

before the first success. X = 0, 1, 2, . . .
X
P
x
E(X ) = x=0 p(1 p) x = p
xq x = p
|x=1{z }
= q + 2q 2 + 3q 3 + 4q 4 + = (q + q 2 + q 3 + q 4 + )
+(q 2 + q 3 + q 4 + )
+(q 3 + q 4 + )
+(q 4 + )
q
q2
q3
q
2
3
1
= 1q + 1q + 1q + = 1q (q + q + q + ) = (1q)
2 =
|
{z
}
q
.
p2
q
1q
Finally, E(X ) = p pq2 = qp .
2/1
Var (X ) = E(X 2 ) (E(X ))2 , E(X 2 ) =
x=1 p(1
p)x x 2 =?
mX (t) = E(etX ) = e0t p + e1t pq + e2t pq 2 + e3t pq 3 +

= 1p + et pq + e2t pq 2 + e3t pq 3 + = p(1 + et q + (et q)2 + )
1
t 1
= p 1e
t q = p(1 qe )
0
m (t) = (1)p(1 qet )2 (qet ) = pqet (1 qet )2 ,

0
m1 = m (0) = pq(1 q)2 = pqp2 =
pq
p2
q
p
00
d
t
t 2 = pq[et (1 qet )2
dt pqe (1 qe )
+et (2)(qet )(1 qet )3 ] = pqet (1 qet )3 [1
m (t) =
qet + 2qet ]
= pqet (1 + qet )(1 qet )3 .

00
m2 = m (0) = pq(1 + q)(1 q)3 =

Now Var(X ) = m2 m12 =
q(1+q)
p2
pq(1+q)
p3
( qp )2 =
q(1+q)
.
p2
q+q 2 q 2
p2
q
.
p2
3/1
S(x) = P(X > x) = P(X x + 1) = P(X = x + 1) + P(X =

x + 2) + = pq x+1 + pq x+2 + pq x+3 +
1
= pq x+1 (1 + q + q 2 + ) = pq x+1 1q
=
pq x+1
p
= q x+1 .
Therefore F (x) = 1 S(x) = 1 q x+1 .

Problem 9.1.9 p. 269 X number of failures, U number of
trials including the first success. X + 1 = U.
P(S) = ( 16 )6 6 =
1
65
= p and q = 1 p = 1
E(U) = 1 + E(X ) = 1 +
q
p
= 1 + 65 (1
Var(U) = Var(X +1) = Var(X ) =
q
p2
1
)
65
1
.
65
= 1 + 65 1 = 65 .
= (65 )2 (1 615 ) = 65 (65 1).
HW 9.1.12, p. 270
4/1
Negative Binomial distribution generalization of a

geometric distribution waiting for r successes:
n
}|
{
z
S
| F F S F F S{z. . . S S S F S} S last (nth) result is a success.
n1 trials, r 1 successes
Let k be the number of failures, k = 0, 1, . . ..

1 r 1 k
k +r 1 r k
P(X = k ; p, r ) = k +r
p
q
p
=
r 1
r 1 p q ,
X = X1 + X2 + + Xr , Xi GEO(p).
E(X ) = E(X1 + + Xr ) = rE(Xi ) =
rq
p.
Xi s are independent,
so that Var(X ) = Var(X1 + + Xr ) = r Var(Xi ) =
rq
.
p2
Also, mX (t) = [mXi (t)]r = pr (1 qet )r
5/1
Example
Suppose that an ordinary die is rolled repeatedly.
(i) Find the probability that the third 6 occurs on the 7th roll.
(ii) Find the probability that the number of rolls until the first 6
occurs is at most nine.
(i) r = 3, p = 16 , X NB(3, 16 ), X is the number of failures.

4
P(X + 3 = 7) = P(X = 4) = 62 ( 61 )3 ( 56 )4 = 15 657 = 0.0335.
(ii) X GEO(p), X is the number of failures.
P(X + 1 9) = P(X 8) = FX (8) = 1 q 9 = 1 ( 56 )9 = 0.806.
6/1
Hypergeometric distribution N is a population size (a

successes and N a failures). Sample of size n is selected
and X is the number of successes in a sample of size n.
Range of X : X a and X n so that X min(a, n).
Also, n X N a so that X n + a N and X 0,
therefore X max(0, n + a N).
Finally, max(0, n (N a)) X min(a, n).
N n
a
a
E(X ) = an
,
Var(X
)
=
n
(1
)
(
).
N
N
N
N
1
| {z }
fpc
fpc - finite population correction factor.
7/1
Example Nine chips are in a bin five red and four blue. Six
chips are to be selected at random.
Obtain the distribution of X the number of red chips selected.
N = 5 + 4 = 9, a = 5, N a = 4, n = 6,
max(0, n (N a)) X min(a, n)
max(0, 6 (9 5)) X min(5, 6)
2X 5
(52)(44)
P(X = 2) = 9 =
(6)
5 4
( )( )
P(X = 4) = 4 9 2 =
(6)
X
p
(53)(43)
P(X = 3) = 9 =
(6)
5 4
(5)(1)
30
,
P(X
=
5)
=
=
84
(96)
10
84 ,
5
42
20
42
15
42
2
42
20
42
5
E(X ) = 2 42
+3
10
E(X ) = na
N = 3 .
Var(X ) = n Na (1
+4
a Nn
N )( N1 )
15
42
+5
2
42
40
84
4
84
10
3 ,
or
5
= 6 59 (1 59 )( 96
91 ) = 9 .
8/1
Poisson distribution Random variable X has a Poisson

n
distribution, POI(), if P(X = n) = n! e , > 0, n = 0, 1, 2, . . .
mX (t) =
E(etX )
tk k e
e
k =0
k!
t
e ee
X
(et )k
k =0
k!
{z
=1
ee
t
e(e 1) .
=
0
t
t
0
mX (t) = e(e 1) et = et+(e 1) , m1 = m (0) = e0 e0 = .
00
mX = [et+(e 1) ] = et+(e 1) (t + (et 1))

t
= et+(e 1) (1 + et ).
00
m2 = m (0) = e0+(e
0 1)
(1 + e0 ) = (1 + )
Var(X ) = m2 m12 = (1 + ) 2 = .
9/1
Poisson distribution is the only discrete distribution that has the

mean and the variance equal.
E(X ) = - interpretation of parameter is the average count
per unit (intensity).
Theorem If random variables X POI(1 ) and Y POI(2 )
are independent then variable X + Y has POI(1 + 2 )
distribution.
t
Proof: mX (t) = e1 (e 1) , mY (t) = e2 (e 1) , and

t
t
t
mX +Y (t) = mX (t)mY (t) = e1 (e 1) e2 (e 1) = e(1 +2 )(e 1) .
mX +Y (t) is a moment generating function of POI(1 + 2 )
distribution.
10 / 1
Problem 9.3.3 p. 288
= 1.5.
(i) Let X5 be the total number of accidents in five consecutive

months, X5 POI(5 1.5 = 7.5).
P(X5 = 5) =
e7.5 (7.5)5
5!
= 0.1094
(ii) Let X1 be the total number of accidents in one month,

[P(X1 = 1)]5 = ( e
1.5 1.5
1!
)5 = 0.0042.
11 / 1
Poisson approximation to binomial distribution

Theorem If p 0 and n in
such
a way that
k
n k
nk
=
e
limn np = > 0 then limn
p (1 p)
k
!
k
| {z }
|
{z
}
POI(=np)
BIN(n,p)
for k = 0, 1, . . . , n large, p small.
If p 1, successes and failures need to be relabeled.
Problem 9.3.2 p. 288 n = 500, = 0.3 (per page), Y is the
number of misprints per page,
S page contains at least three misprints. Y POI(0.3).
P
j
0.3 =
P(S) = 1 P(Y < 3) = 1 2j=0 0.3
j! e
0
0.3
0.3
0.3 (1+0.3+0.045) = 0.0036.
1e0.3 ( 0.3
0! + 1! + 2! ) = 1e
X500 BIN(500, 0.0036), = 500 0.0036 = 1.8, V POI(1.8).

P(X500 > 1) = 1 P(X500 1) 1 [P(V = 0) + P(V = 1)]
= 1 e1.8 (1 + 1.8) = 1 e1.8 2.8 = 0.5371.
12 / 1

Selected Probability Models: Bernoulli Trials - Most Common Type of Situation Modeled

Uploaded by

Copyright:

Available Formats

Selected Probability Models: Bernoulli Trials - Most Common Type of Situation Modeled

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Selected Probability Models: Bernoulli Trials - Most Common Type of Situation Modeled

Uploaded by

Copyright:

Available Formats

Selected Probability Models

Bernoulli trials - most common type of situation modeled:

Geometric distribution, GEO(p) X is the number of failures

Finally, E(X ) = p pq2 = qp .

Var (X ) = E(X 2 ) (E(X ))2 , E(X 2 ) =

mX (t) = E(etX ) = e0t p + e1t pq + e2t pq 2 + e3t pq 3 +

m (t) = (1)p(1 qet )2 (qet ) = pqet (1 qet )2 ,

m1 = m (0) = pq(1 q)2 = pqp2 =

= pqet (1 + qet )(1 qet )3 .

m2 = m (0) = pq(1 + q)(1 q)3 =

S(x) = P(X > x) = P(X x + 1) = P(X = x + 1) + P(X =

Therefore F (x) = 1 S(x) = 1 q x+1 .

Var(U) = Var(X +1) = Var(X ) =

= (65 )2 (1 615 ) = 65 (65 1).

Negative Binomial distribution generalization of a

so that Var(X ) = Var(X1 + + Xr ) = r Var(Xi ) =

Also, mX (t) = [mXi (t)]r = pr (1 qet )r

Suppose that an ordinary die is rolled repeatedly.

Hypergeometric distribution N is a population size (a

fpc - finite population correction factor.

Poisson distribution Random variable X has a Poisson

mX = [et+(e 1) ] = et+(e 1) (t + (et 1))

Poisson distribution is the only discrete distribution that has the

Proof: mX (t) = e1 (e 1) , mY (t) = e2 (e 1) , and

Problem 9.3.3 p. 288

(i) Let X5 be the total number of accidents in five consecutive

(ii) Let X1 be the total number of accidents in one month,

Poisson approximation to binomial distribution

X500 BIN(500, 0.0036), = 500 0.0036 = 1.8, V POI(1.8).

You might also like