0% found this document useful (0 votes)
38 views9 pages

Chapter 5 - Creating New Univariate Distributions

1. Transformations involve creating a new random variable Y as a function of an original random variable X to produce a new distribution for Y. For discrete X, the probability mass function of Y is defined in terms of the probability mass function of X. For continuous X, the cumulative distribution function of Y is defined using the cumulative distribution function of X and the transformation function. 2. Common transformations include scaling, inverting, transforming to a power, and shifting. These transformations preserve the shape of the original distribution but change parameters like location and scale. 3. Theorems and examples are provided to derive the distributions resulting from various transformations applied to uniform, normal, Pareto, exponential and Gamma distributions.

Uploaded by

Samuel Salacup
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
38 views9 pages

Chapter 5 - Creating New Univariate Distributions

1. Transformations involve creating a new random variable Y as a function of an original random variable X to produce a new distribution for Y. For discrete X, the probability mass function of Y is defined in terms of the probability mass function of X. For continuous X, the cumulative distribution function of Y is defined using the cumulative distribution function of X and the transformation function. 2. Common transformations include scaling, inverting, transforming to a power, and shifting. These transformations preserve the shape of the original distribution but change parameters like location and scale. 3. Theorems and examples are provided to derive the distributions resulting from various transformations applied to uniform, normal, Pareto, exponential and Gamma distributions.

Uploaded by

Samuel Salacup
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 9

Mathematical Statistics I

Module for Math 150.1

an
lim
Di
UP
s,
ic
at
em
h
at
M
of
te
itu
st
In
Chapter 5: Creating New Univariate
Distributions

5.1 Transformations
Transformation is a technique of constructing other distributions. We create a new ran-

an
lim
dom variable, hence creating another distribution, by considering functions of the original
random variable.

Di
UP
For a discrete random variable X with pmf fX (x), then the pmf of the random variable
s,

Y = g(X) is
ic

X
fY (y) = fX (x).
at

x∈g −1 (y)
em

Example 1. If X ∼ Unif{1, 2, . . . , n} and Y = 2X + 1, then Y ∼ Unif{3, 5, 7, . . . , 2n + 1}.


h
at

Example 2. If X ∼ Unif{−10, −9, . . . , 9, 10} and Y = |X|, then Y has pmf


M

(
of

1
21
, y=0
fY (y) = .
te

2
21
, y ∈ {1, 2, . . . , 10}
itu

For continuous random variables, we use the cdf instead of the pdf.
st
In

Theorem 5.1 (CDF Technique) Let X is be a continuous random variable with pdf
fX (x) and cdf FX (x). Also, let g(x) be an invertible function and Y = g(X).

1. If g(x) is monotonically increasing, then

FY (y) = P[Y ≤ y] = P[g(X) ≤ y] = P[X ≤ g −1 (y)] = FX (g −1 (y)).

2. If g(x) is monotonically decreasing, then

FY (y) = P[Y ≤ y] = P[g(X) ≤ y] = P[X ≥ g −1 (y)] = 1 − FX (g −1 (y)).

2
Chapter 5. Creating New Univariate Distributions 3

Example 3. Find the pdf of Y = X 2 , where X ∼ Unif[−1, 1].


Solution: Since Y = X 2 , then Y can only take on values on [0, 1]. Let y ∈ [0, 1], then
√ √
FY (y) = P[Y ≤ y] = P[X 2 ≤ y] = P[− y ≤ X ≤ y]
√ √
√ √ 1+ y 1− y √
= FX ( y) − FX (− y) = − = y
2 2
1
Thus, fY (y) = FY0 (y) = √ 1(0,1) (y).
2 y
Remark. Common Transformations (and how the new distribution is named)
• Y = θX, θ 6= 0 : scaled
• Y = X 1/τ , τ > 0, τ 6= 1 : transformed
• Y = X −1 : inverse

an
• Y = X 1/τ , τ < 0, τ 6= −1 : inverse transformed

lim
• Y = eX : log

Di
• Y = X ± c, c 6= 0: translated / shifted UP
Theorem 5.2 Let X be a continuous random variable with pdf fX (x) and cdf FX (x). Let
s,

Y = θX, where θ > 0. Then


ic
at

y  1 y 
em

FY (y) = FX and fY (y) = fX


θ θ θ
h

Remark. The parameter θ is a scale parameter for the random variable Y .


at
M

Example 4. If X ∼ N (µ, σ 2 ) and Y = θX, θ > 0. What is the distribution of Y ?


of

Solution: Based on the previous theorem,


te

(  2 ) (  2 )
itu

1 y  1 1 1 yθ − µ 1 1 y − µθ
fY (y) = fX = √ exp = √ exp .
st

θ θ θ σ 2π 2 σ σθ 2π 2 σθ
In

Thus, Y ∼ N (µθ, (σθ)2 ).


Example 5. If X has the Pareto Type II (also called Lomax) distribution P a(α, θ), where
α, θ > 0. Then its pdf and cdf are as follows

αθα

θ
fX (x) = and FX (x) = 1 − , x>0
(x + θ)α+1 x+θ
Suppose Y = (1 + r)X where r > −1. What is the distribution of Y ?
Solution: Since 1 + r > 0, then
  !α  α
y θ (1 + r)θ
FY (y) = FX =1− x =1−
1+r 1+r
+θ x + (1 + r)θ

Therefore, Y ∼ P a(α, (1 + r)θ).


Chapter 5. Creating New Univariate Distributions 4

Next, we look at how raising to a power affects the distribution of a random variable.

Theorem 5.3 Let X be a continuous random variable with pdf fX (x) and cdf FX (x) with
FX (0) = 0. Let τ > 0.

• if Y = X 1/τ , then FY (y) = FX (y τ ) and fY (y) = τ y τ −1 fX (y τ )

• if Y = X −1/τ , then FY (y) = 1 − FX (y −τ ) and fY (y) = τ y −τ −1 fX (y −τ )

Example 6. Suppose X has the exponential distribution with parameter 1. Determine


the cdf of the inverse, transformed and inverse transformed exponential distribution with
scale parameter θ > 0.
Solution: X ∼ Exp(1), thus FX (x) = 1 − e−x . Let θ > 0.

an
h  i  

lim
If Y = θX −1 , then FY (y) = 1 − 1 − exp − yθ = exp − yθ (Inverse Exponential)

Di
τ
If Y = θX 1/τ , τ > 0, then FY (y) = 1 − exp − yθ
  
(Weibull Distribution)
h τ i UP
If Y = θX −1/τ , τ > 0, then FY (y) = exp − yθ (Inverse Weibull Distribution)
s,

Definition 5.4 The incomplete gamma function with parameter α > 0 is given by
ic

ˆ x
at

1
em

Γ(α; x) = tα−1 e−t dt


Γ(α) 0
h
at

while the gamma function is defined by


M

ˆ ∞
of

Γ(α) = tα−1 e−t dt


0
te
itu

Example 7. Find the pdf’s of the scaled, inverse, transformed and inverse transformed
st

Gamma distribution with pdf


In

xα−1 e−x
fX (x) =
Γ(α)
Solution: Note that X ∼ Γ(α, 1).  α
y α−1 e−y/θ 1 1
• Scaled Gamma [Y = θX]: fY (y) = α = y α−1 e−y/θ
θ Γ(α) Γ(α) θ
Thus, Y ∼ Γ(α, 1/θ).
θα e−θ/y
• Inverse Gamma: fY (y) = α+1
y Γ(α)
τ uα e−u  y τ
• Transformed Gamma: fY (y) = where u =
yΓ(α) θ  τ
τ uα e−u θ
• Inverse Transformed Gamma: fY (y) = where u =
yΓ(α) y
Chapter 5. Creating New Univariate Distributions 5

Lastly, we look at how exponentiation affects the distribution of certain random variables.

Theorem 5.5 Let X be a continuous random variable with pdf fX (x) and cdf FX (x) with
fX (x) > 0 for all real x. Let Y = exp X. Then, for y > 0,
1
FY (y) = FX (ln y) fY (y) = fX (ln y)
y

Example 8. Lognormal distribution


Let X ∼ N (µ, σ 2 ) and Y = exp(X). Then
"  2 #
1 1 ln y − µ
fY (y) = √ exp − .
yσ 2π 2 σ

an
5.2 Mixed Random Variables

lim
Di
There exists certain random variables that are neither discrete nor continuous, but are a
UP
mixture of both. These random variables are called mixed random variables. In particular,
a mixed random variable has a continuous part and a discrete part. We shall illustrate such
ics,

random variables via some examples.


at
em

Example 9. Consider the random variable X having cdf


h

(
x2 , 0 ≤ x < 21
at

FX (x) = .
M

1, x ≥ 12
of

Below is the graph of the cdf


te
itu
st
In

Note that even though X(Ω) = [0, 21 ] (an interval), FX is not continuous. Hence X is not
a continuous random variable.
Chapter 5. Creating New Univariate Distributions 6

Definition 5.6 Let X be a random variable with cdf F (x). X is called a mixed random
variable, and its distribution a mixed distribution, if F (x) has the form
k
X
F (x) = αi Fi (x),
i=1

k
X
with 0 < αi < 1, αi = 1, k ≥ 2, and Fi (x) is cdf of either a continuous or discrete
i=1
random variable. The αi ’s are called the weights of the Fi ’s.

In general, the cdf FX of a mixed random variable X can be written as the convex combi-
nation of the cdf’s of a continuous rv and a discrete rv., i.e.,

an
FX (x) = αC(x) + βD(x)

lim
where α, β ≥ 0 and α + β = 1, and C is a cdf of some continuous rv while D is the cdf of

Di
some discrete rv. UP
s,

In our previous example, we see that


ic
at

(
1
x2 , 0 ≤ x < 2
em

αC(x) = and
1 1
4
, x≥ 2
h
at

(
1
0, 0 ≤ x <
M

2
βD(x) = .
3 1
, x≥
of

4 2
1
te

Since C and D are cdf’s then lim C(x) = lim D(x) = 1. Hence we take α = 4
and
x→+∞ x→+∞
itu

β = 34 . Finally,
st

(
1
4x2 , 0 ≤ x <
In

2
C(x) = and
1
1, x≥ 2
(
1
0, 0 ≤ x < 2
D(x) = .
1
1, x ≥ 2

Here, we can clearly see that C is a cdf of a continuous rv, while D is a cdf of a discrete rv.

Remark. The pdf of a mixed random variable exists, but it uses what we call the “Dirac
Delta Function” and is beyond the coverage of our course.

Theorem 5.7 Let X be a mixed random variable with cdf F (x). Then there exists α, β > 0
with α + β = 1, a continuous cdf C(x) and a discrete (step-function) cdf D(x) such that

F (x) = αC(x) + βD(x) for all x.


Chapter 5. Creating New Univariate Distributions 7

Moreover, if f (x) is the pdf of C(x) and p(x) is the pmf of D(x) then for any n ∈ N :
ˆ +∞ X
n
E[X ] = α xn f (x) dx + β xnk p(xk )
−∞ xk

Example 10. Find the mean of the mixed random variable from Example 9.
Solution: Take f (x) as the pdf of C(x), i.e., f (x) = C 0 (x). Thus
(
1
8x, 0 ≤ x < 2
f (x) = .
0, otherwise

Similarly, let p(x) be the pmf of D(x), then p(x) = 1{ 1 } (x).


2

Hence,

an
ˆ

lim
1  
2 1
E[X] = α x f (x) dx + β p(1/2)
2

Di
0
ˆ 1  
1 2 2 3 1 UP 11
= 8x dx + (1) = .
4 0 4 2 24
ic s,

5.3 Truncation
at
em

A truncated distribution is a conditional distribution that is obtained by restricting the


h
at

domain of some other known distribution. Truncation is usually used on a normal distri-
M

bution.
of
te

Definition 5.8 Let X be a random variable with cdf Fx (x). Let a, b be in the domain of
itu

FX such that a < b and FX (a) 6= FX (b). The cdf of the truncated random variable X given
st
In

that a < X ≤ B is given by


FX (x) − FX (a) FX (x) − FX (a)
F (x|a < X ≤ b) = =
P(a < X ≤ b) FX (b) − FX (a)

Theorem 5.9 Let X be a continuous random with cdf F (x) and pdf f (x). Suppose a, b is
in the range of X such that a < b and F (a) 6= F (b). Then the function

f (x)
g(x) = 1[a,b] (x)
F (b) − F (a)
is a pdf. Furthermore, the above is the pdf of the truncated random variable X given that
a ≤ X ≤ b.
Chapter 5. Creating New Univariate Distributions 8

Example 10. Let X ∼ N (µ, σ 2 ). Now suppose that we condition on X ∈ [a, b]. Find the
mean and variance of the truncated distribution.
b−µ a−µ
 
Solution: The probability that X ∈ [a, b] is given by Φ σ
−Φ σ
, where Φ is the cdf
of a standard normal rv. Then the pdf of the truncated rv is given by
n 2 o
√1 exp −1 x−µ
f (x) σ 2π 2 σ
f (x|a ≤ X ≤ b) = b−µ
 a−µ
= b−µ
 a−µ
 , a ≤ x ≤ b.
Φ σ −Φ σ Φ σ −Φ σ
To obtain the raw moments of the truncated distribution, we first solve its mgf.
ˆ b
 tX etx φ(x) dx
a

m(t) = E e |a ≤ X ≤ b =
Φ b−µ a−µ
 
− Φ

an
σ σ
Note that
ˆ b ˆ b ˆ b

lim
1 1 2
n o
x−µ 2 −1
e 2σ2 [
tx tx −1 ( ) x−(σ 2 t+µ)] −(σ 2 t+µ)2 +µ2
e φ(x) dx = √ e e2 σ dx = √ dx

Di
a σ 2π a σ 2π a
ˆ b −1  x−(σ2 t+µ)2 2
UP
1 −1
[ µ2 −(σ 2 t+µ)2
]
= √ e 2σ2 e2 σ
dx
σ 2π a
ˆ b
s,

2 2
 
2
−1 x−(σ t+µ)
1
ic

σ 2 t2
= eµt+ 2 √ e 2 σ
dx
at

a σ 2π
em

b − σ2t − µ a − σ2t − µ
    
2 2
µt+ σ 2t
= e Φ −Φ
h

σ σ
at

Thus,
M

2 2
h  2   2 i
µt+ σ 2t b−σ t−µ
e Φ σ
− Φ a−σσ t−µ
of

m(t) =
Φ b−µ − Φ a−µ
 
te

σ σ
itu

b−µ
 a−µ
!
φ − φ
So E [X|a ≤ X ≤ b] = m0 (0) = µ − σ σ σ
st

b−µ
 a−µ
 .
Φ σ −Φ σ
In

Likewise,

E X 2 |a ≤ X ≤ b = m00 (0)
 
" # " #
b−µ a−µ b−µ a−µ
φ0 − φ0
 
φ −φ
= σ 2 + µ2 + σ 2 σ
b−µ
 σ
a−µ
 − 2µσ σ
b−µ
 σ
a−µ

Φ σ
−Φ σ
Φ σ
−Φ σ
Note that:
φ0 (x) = xφ(x).
∴ V ar(X|a ≤ X ≤ b) = E X 2 |a ≤ X ≤ b − (E [X|a ≤ X ≤ b])2
 
 
b−µ
 b−µ  a−µ
 a−µ  " b−µ
 a−µ
 #2 
 φ − φ σ φ −φ
= σ2 1 + σ σ σ
− σ σ
.
Φ b−µ − Φ a−µ b−µ a−µ
   
 σ σ
Φ σ
−Φ σ

Chapter 5. Creating New Univariate Distributions 9

Exercises.
X
1. Let X ∼ Exp(1) and Y = with θ > 0. Find the distribution of Y .
θ
2. Let X ∼ P oi(λ) and Y = cX, c > 0. Find the distribution of Y .

3. Let X be a standard Cauchy random variable, i.e., its pdf is given by


1
f (x) = , x ∈ R.
π(1 + x2 )

Let T = X −1 . Find the distribution of T .

4. Suppose X is the equally weighted mixture Y ∼ U ({0, 1, 2, 5}) and Z ∼ U [0, 4].
Find the mean and variance of X.

an
lim
5. Let X ∼ N (µ, σ 2 ). Let Y = X|X < b and Z = X|X > a. Find the mean and

Di
variance of Y and Z.
UP
6. Let X ∼ U (0, b), b > 0. Define Y = X|X < a with 0 < a < b. Show that
Y ∼ U (0, a).
ics,
at
h em
at
M
of
te
itu
st
In

You might also like