Families of Random Variables

Homework 7

Problems
1. Let X ∼ N egbin(4, 1/3).

(a) Find P {X > 12}

∞ 12
X
x 1 1 X 1 1
P {X > 12} = Cx+4−1 ( )4 (1 − )x = 1 − x
Cx+4−1 ( )4 (1 − )x = 0.8341 (1)
x=13
3 3 x=0
3 3

(b) Find the smallest value for x so that P {X > x} = 0.05, 0.01, 0.001
∞ x
X
i 1 1 X 1 1
P {X > x} = Ci+4−1 ( )4 (1 − )i = 1 − i
Ci+4−1 ( )4 (1 − )i (2)
i=x+1
3 3 i=0
3 3

So minimal x such that P {X > x} = .05 is 17, minimal x such that P {X > x} = .05 is 23,

x 17 23 29 30 40
P {X > x} 0.9538 0.9918 0.9987 0.9991 1.0000

minimal x such that P {X > x} = .05 is 30.

(c) Simulate 10,000 independent observations of N egbin(4, 1/3) random variable and check your an-
swers to (a) and (b) from the simulation.
X¡-rnbinom(10000,4,1/3)
X¡-sort(X)
X[10000*(1-.05)]
X[10000*(1-.01)]
X[10000*(1-.001)]
S ¡- X¿12
table(S)
From table(S) we know P {X > 12} = 0.8303, and the three quantiles are 17, 23 and 30. They
are quite similar to answers in last two parts.

2. For some natural numbered-valued random variables, the case {X = 0} cannot be observed. Thus, we
can define and second random variable Y so that

P {Y = y} = P {X = y|X 6= 0}.

1
 (a) Find the mass function for Y .
P (X = y, y 6= 0) P (X = y, y 6= 0)
P {Y = y, y 6= 0} = P {X = y|X 6= 0} = = (3)
P (X 6= 0) 1 − P (X = 0)

P {Y = 0} = P {X = 0|X 6= 0} = 0 (4)

(b) Give this mass function for X ∼ Bin(n, p)

P {X = 0} = Cn0 p0 (1 − p)n = (1 − p)n (5)

P (X = y, y 6= 0) C y py (1 − p)( n − y)
P {Y = y, y 6= 0} = = n (6)
1 − P (X = 0) 1 − (1 − p)n

(c) Give this mass function for X ∼ Geom(p)

P {X = 0} = p(1 − p)0 = p (7)

P (X = y, y 6= 0) p(1 − p)y
P {Y = y, y 6= 0} = = = p(1 − p)y−1 (8)
1 − P (X = 0) 1−p

(d) Find the mean for these random variables,

P
X yP (X = y, y 6= 0) EX
EY = yP {Y = y, y 6= 0} = = (9)
1 − P (X = 0) 1 − P (X = 0)
EX np
EYBin = = (10)
1 − P (X = 0) 1 − (1 − p)n
p
EX 1−p p
EYGeom = = = (11)
1 − P (X = 0) 1−p (1 − p)2

3. On the other hand, the mass function for a zero inflated random variable Y is a mixture of a degenerate
random variable taking only the value 0. Thus, the mass function

fY (y) = αI0 (y) + (1 − α)fX (y)

2
for some natural numbered-valued random variable X (mean µX and variance σX ) and some α ∈ (0, 1)
(a) Verify that fY is a mass function.
Note I0 (y) ≥ 0 and fX (y) ≥ 0, so fY (y) ≥ 0
X X X
fY (y) = (αI0 (y)) + ((1 − α)fX (y))
X X
=α I0 (y) + (1 − α) fX (y) (12)
= α + (1 − α) = 1

So fY (y) is mass function.

2
 2
(b) Find the mean of Y in terms of µX and σX .
X X X
EY = yfY (y) = αyI0 (y) + (1 − α)yfX (y)
(13)
= αy + (1 − α)EX

2
(c) Find the variance of Y in terms of µX and σX .
X X X
EY 2 = y 2 fY (y) = αy 2 I0 (y) + (1 − α)y 2 fX (y)
(14)
= αy 2 + (1 − α)µ2X

V ar(Y ) = EY 2 − (EY )2 = αy 2 + (1 − α)EX 2 − [αy + (1 − α)EX]2

= (1 − α)(µ2X + σX
2
) − 2α(1 − α)yµX − (1 − α)2 µ2X (15)
= α(1 − α)µ2X + (1 − 2
α)σX − 2α(1 − α)yµX

Challenging Problems
1. Let X ∼ Beta(α, β)

(a) Write the density of X as an exponential family with natural parameters η1 , η2 .

Γ(α + β) α−1
f (x) = x (1 − x)β−1
Γ(α)Γ(β)
Γ(α + β)
= exp[(α − 1)lnx + (β − 1)ln(1 − x)]
Γ(α)Γ(β)
= exp{(α − 1)lnx + (β − 1)ln(1 − x) − [ln(Γ(η1 + 1)) + ln(Γ(η2 + 1)) − ln(Γ(η1 + η2 + 2))]}
(16)
η1 = α − 1, η2 = β − 1 (17)

(b) Find E ln X
A(η1 , η2 ) = ln(Γ(η1 + 1)) + ln(Γ(η2 + 1)) − ln(Γ(η1 + η2 + 2)) (18)
∂A(η1 , η2 ) Γ0 (η1 + 1) Γ0 (η1 + η2 + 2)
ElnX = = − (19)
∂η1 Γ(η1 + 1) Γ(η1 + η2 + 2)
Z ∞
Γ0 (η1 + 1) = η1 tη1 −1 e−t dt = η1 Γ(η1 ) (20)
0
Z ∞
Γ0 (η1 + η2 + 2) = η1 tη1 −1 e−t dt = (η1 + η2 + 2)Γ(η1 + η2 + 1) (21)
0

η1 Γ(η1 ) (η1 + η2 + 1)Γ(η1 + η2 + 1)

ElnX = − =0 (22)
Γ(η1 + 1) Γ(η1 + η2 + 2)

(c) Find EX ln X

3
 (d) What does these values tell you in the case of X ∼ Beta(0, 0) = U nif (0, 1)?
2. Let X1 , . . . , Xn be independent and identically distributed with kurtosis α4 .
(a) Find the kurtosis α4,n of X̄n , the average of X1 , . . . , Xn .
For i = 1, . . . , n,
Xi − µX 4 E{Xi − µX }4
E{( ) } − 3 = α4 = −3 (23)
σX σX
n Pn
4 1X 4 E{ i=1 (Xi − µX )}4
E{X̄ − µX } = E{ (Xi − µX )} = (24)
n i=1 n4
Consider
E{(Xi1 − µX )(Xi2 − µX )(Xi3 − µX )(Xi4 − µX )} = 8µ4X − 8µ4X = 0 (25)
E{(Xi1 − µX )2 (Xi2 − µX )(Xi3 − µX )} (26)
n
X̄ − µX 4 1 X Xi − µ 4
α4,n = E{( ) }−3= 4 E{ } −3
σX n i=1 α
(27)
n(α4 + 3)
= −3
n4

(b) Find
lim α4,n .
n→∞

n(α4 + 3)
lim α4,n = lim − 3 = −3 (28)
n→∞ n→∞ n4