186

Chapter 6

Functions of Random Variables

6.1 Introduction
We now look at determining probability distributions of functions of random variables
(of statistics); for example, statistic
n
1X
Ȳ = Yi .
n i=1

Knowing probability distribution of a statistic, we are able to determine measures of

“goodness” of an estimation; for example, probability of difference between statistic
Ȳ and parameter µ. Distribution of a function of random variables is derived from
a joint distribution of all random variables. Assuming simple random sampling with
large sample size (which approximates independence), joint distributions of discrete
and continuous, respectively, random variables is given by,

p(y1 , y2 , . . . , yn ) = p(y1 )p(y2) · · · p(yn )

f (y1, y2 , . . . , yn ) = f (y1)f (y2 ) · · · f (yn )

6.2 Finding the Probability Distribution of a

Function of Random Variables
Methods which determine probability distributions of functions of random variables
include:

• method of distributions functions

given distribution of Y , determines distribution of U = f (Y )
could integrate density of Y to get distribution of Y , then use method to get distribution of U = f (Y ), then
differentiate to get density of U

187
188 Chapter 6. Functions of Random Variables (ATTENDANCE 11)

• method of (univariate) transformations,

given density of Y , determines density of U = f (Y ) directly without going through the distributions
requires function U = f (Y ) to be continuous one-to-one monotonic

• method of moment–generating functions.

given moment-generating function of Y , determines moment-generating function of U = f (Y )
useful if distributions/densities have a recognizable moment-generating functions

A method of multivariate transformations is also discussed.

6.3 The Method of Distribution Functions

To determine probability distribution of function, U, of random variables
Y1 , Y2 , . . . , Yn , using method of distribution functions, identify both curve U = u
and region U ≤ u in (y1 , y2, . . . , yn ) space. Then determine F (u) = P (U ≤ u) by
integrating f (y1 , y2, . . . , yn ) over region U ≤ u. Lastly, find f (u) = F ′ (u) = dF
du
.
This method is demonstrated in one–dimensional U(Y ), two–dimensional
U(Y1 , Y2) and nonlinear U(Y ) cases. In addition, invertible distribution F (y) and
uniform distribution are related to one another for sampling purposes.

Exercise 6.3 (The Method of Distribution Functions)

1. One dimensional U(Y ). Density
1 − y2 ,

0≤y≤2
f (y) =
0 elsewhere
(a) Distribution F (y).
y t=y
t2
Z   
t
FY (y) = 1− dt = t −
0 2 4 t=0
y2 y y2 y3
(i) 4
(ii) y − 4
(iii) y − 4
(iv) y − 4

(b) Density for U = 2Y .

u u 
FU (u) = P (U ≤ u) = P (2Y ≤ u) = P Y ≤ = FY
2 2
u 2


2
u u u
= − 2 = − ,
2 4 2 16
dF
fU (u) = FU′ (u) = =
du
3 2u 1 2u 1 u 1 2u
(i) 2
− 32
(ii) 2
− 32
(iii) 2
− 32
(iv) 2
− 16
,
Section 3. The Method of Distribution Functions (ATTENDANCE 11) 189

where 0 ≤ u2 ≤ 2, or
(i) −1 ≤ u ≤ 2 (ii) −2 ≤ u ≤ 1 (iii) 0 ≤ u ≤ 3 (iv) 0 ≤ u ≤ 4

(c) Density for U = 2 − 2Y .

   
2−u 2−u
FU (u) = P (U ≤ u) = P (2 − 2Y ≤ u) = P Y ≥ = 1 − FY
2 2
2
!
2−u


2−u 4 + 4u + u2
= 1− − 2 = ,
2 4 16
dF
fU (u) = FU′ (u) = =
du
3 2u 1 2u 1 u 1 2u
(i) 2
− 32
(ii) 2
+ 32
(iii) 2
+ 32
(iv) 4
+ 16
,

where 0 ≤ 2−u
2
≤ 2, or
(i) −1 ≤ u ≤ 2 (ii) −2 ≤ u ≤ 1 (iii) −1 ≤ u ≤ 1 (iv) −2 ≤ u ≤ 2

(d) Density for U = Y 2 .

√
√

FU (u) = P (U ≤ u) = P Y 2 ≤ u = P Y ≤ u = FY


u
√ 2
√ ( u) √ u
= u− = u− ,
4 4
dF
fU (u) = FU′ (u) = =
du
3 2u 1 1 √1 1 1
(i) 2
− 32
(ii) 2
√
u
− 4
(iii) u
− 4
(iv) 2
√
u
+ 41 ,
√
where 0 ≤ u ≤ 2, or
(i) 0 ≤ u ≤ 1 (ii) 0 ≤ u ≤ 2 (iii) 0 ≤ u ≤ 3 (iv) 0 ≤ u ≤ 4

(e) Comment Notice if the main goal was to determine the density of U =
f (Y ), this method requires first integrating the density of Y to get the
distribution of Y , then use this method to get the distribution of U = f (Y ),
then differentiate to finally get the density of U; it does not determine the
density of U = f (Y ) directly from the density of Y .
(i) True / (ii) False

2. Two dimensional U(Y1 , Y2). Bivariate density function is

 1
12
, 0 ≤ y1 ≤ 2, 0 ≤ y2 ≤ 6
f (y1 , y2 ) =
0 elsewhere
190 Chapter 6. Functions of Random Variables (ATTENDANCE 11)

f(y1,y2)
y
2

1/12 6

0 2 y1

Figure 6.1: Continuous bivariate function

y
2
u = y1
6
y1 < u

0 2 y1

Figure 6.2: Continuous bivariate: U = Y1

(a) Density for U = Y1 .

Function u = y1 ranges 0 to 2, y2 ranges 0 to 6 and y1 ranges 0 to u.

FU (u) = P (U ≤ u) = P (Y1 ≤ u)
Z 6Z u Z 6  y =u Z 6
1 y1 1 u
= dy1dy2 = dy2 = dy2 =
0 0 12 0 12 y1 =0 0 12

6 5 4 3
(i) 12
(u) (ii) 12
(u) (iii) 12
(u) (iv) 12
(u)

dF 1 1 1 1
So fU (u) = FU′ (u) = du
= (i) 2
(ii) 3
(iii) 4
(iv) 5
(b) Density for U = 3Y1 + Y2 .
In figure (a), u = 3y1 + y2 ranges 3(0) + 0 = 0 to 3(2) + 0 = 6, y2 ranges 0
Section 3. The Method of Distribution Functions (ATTENDANCE 11) 191

y
2
u=3 y + y
u=3 y + y 1 2
1 2

6 6
3 y1 + y < u
2
3 y1 + y > u
2

0 2 y1 0 2 y1

(a) (b)

Figure 6.3: Continuous bivariate: U = 3Y1 + Y2

u−y2
to u, y1 ranges 0 to 3
.
 
u − y2
FU (u) = P (U ≤ u) = P (3Y1 + Y2 ≤ u) = P Y1 ≤ =
3
Z u Z u−y2 Z u  y = u−y2
3 1 y1 1 3
= dy1 dy2 = dy2 =
0 0 12 0 12 y1 =0
Z u y =u
1 2uy2 − y22 2
  
1 u − y2
= dy2 = =
0 12 3 12 6 y2 =0

1
(i) 72 (u2 + 3)
1
(ii) 72 (u2 + 2)
1
(iii) 72 (u2 + 1)
1
(iv) 72 (u2 )

dF u+3 u+2 u+1 u

So fU (u) = du
= (i) 36
(ii) 36
(iii) 36
(iv) 36

In figure (b), easier to first determine non–shaded triangular region then

subtract from one (1) to find shaded region. For non–shaded region, u =
3y1 + y2 ranges u = 3(2) + 0 = 6 to u = 3(2) + 6 = 12, y2 ranges u − 6 to
6 and y1 ranges u−y3
2
to 2.
FU (u) = P (U ≤ u) = P (3Y1 + Y2 ≤ u) = 1 − P (3Y1 + Y2 > u)
Z 6 Z 2 Z 6  y =2
1 y1 1
= 1− dy1dy2 = 1 − dy2 =
u−y2 12 12 u−y
y1 = 3 2
u−6 3
u−6
Z 6 y =6
1 −2(u − 6)y2 + y22 2
  
1 u − y2
= 1− 2− dy2 = 1 −
u−6 12 3 12 6 y2 =u−6

1
(i) 72
(35 − 12(u − 6) + (u − 6)2 )
192 Chapter 6. Functions of Random Variables (ATTENDANCE 11)

1
(ii) 72 (36 − 16(u − 6) + (u − 6)2 )
1
(iii) 1 − 72 (36 − 16(u − 6) + (u − 6)2 )
1
(iv) 1 − 72 (36 − 12(u − 6) + (u − 6)2 )

dF 12−u 13−u 14−u 15−u

So fU (u) = du
= (i) 36
(ii) 36
(iii) 36
(iv) 36

(c) Density and Expected Value for U = Y1 Y2 .

y2 y1 y2 > u
6
y2 = u_
2
y1 y2 < u y1 y2 = u
0 2 y1

Figure 6.4: Continuous bivariate: U = Y1 Y2

Easier to first determine non–shaded region then subtract from one (1) to
find shaded region. For non–shaded region, u = y1 y2 ranges u = (0)(6) = 0
to u = (2)(6) = 12, y2 ranges u2 to 6, y1 ranges yu2 to 2.
 
u
FU (u) = P (U ≤ u) = P (Y1 Y2 ≤ u) = 1 − P Y1 > =
y2
Z 6Z 2 Z 6  y =2
1 y1 1
= 1− dy1 dy2 = 1 − dy2 =
u u 12 u 12 y1 = yu2
2 y2 2
Z 6
1 1
[2y2 − u ln y2 ]yy22 =6


= 1− 2 − uy2−1 dy2 = 1 − =u =
u 12 12 2
2

1 u


(i) 1 − 12 12 + u − u ln 12
1 u


(ii) 1 − 12 12 − u + u ln 12
1 u


(iii) 12 12 + u − u ln 12
1 u


(iv) 12 12 − u + u ln 12

So fU (u) = dF du
= (choose one)
11 1 u
(i) − 12 − 12 ln 12 
2
(ii) 1 − 12 1
1 − ln u2 − u12
 2

1
(iii) 12 ln u2 − u12
 
1 u u2
(iv) 12 1 + ln 2 − 12
Section 3. The Method of Distribution Functions (ATTENDANCE 11) 193

And
12 u=12
u2 u2
  
11 1 1
Z
u u
E(u) = u· − − ln du = − ln − =
0 12 12 12 12 2 12 4 u=0

(choose one) (i) −8 (ii) −3 (iii) 5 (iv) 6

3. More two dimensional U(Y1 , Y2 ): Potato Chips.

f(y1,y2)
y
2

8
1/12
2

49 51 y1

Figure 6.5: Continuous bivariate functions: potato chips

Bivariate density function for weight of bag and amount of salt is

 1
12
, 49 ≤ y1 ≤ 51, 2 ≤ y2 ≤ 8
f (y1, y2 ) =
0 elsewhere
(a) Density and Expected Value for U = Y1 .

y u= y
1
2

8
y1 < u

2
0 y1
49 51

Figure 6.6: Continuous bivariate, potato chips: U = Y1

Function u = y1 ranges 49 to 51, y2 ranges 2 to 8 and y1 ranges 49 to u.

FU (u) = P (U ≤ u) = P (Y1 ≤ u)
Z 8Z u Z 8  y =u Z 8
1 y1 1 1
= dy1 dy2 = dy2 = (u − 49) dy2 =
2 49 12 2 12 y1 =49 2 12
194 Chapter 6. Functions of Random Variables (ATTENDANCE 11)

6 5 4 3
(i) 12
(u − 49) (ii) 12
(u − 49) (iii) 12
(u − 49) (iv) 12
(u − 49)

dF 1 1 1 1
So fU (u) = FU′ (u) = du
= (i) 2
(ii) 3
(iii) 4
(iv) 5

And
51  
1
Z
E(u) = u· du =
49 2
(choose one) (i) 49 (ii) 49.5 (iii) 50 (iv) 50.5
(b) Density and Expected Value for U = 3Y1 + Y2 .

y u=3 y + y
u=3 y + y 2 1 2
1 2

8 8
3 y1 + y < u
2
3 y1 + y > u
2
2 2
0 y1 0 y1
49 51 49 51
(a) (b)

Figure 6.7: Continuous bivariate, potato chips: U = 3Y1 + Y2

In figure (a), U = 3Y1 + Y2 ranges 3(49) + 2 = 149 to 3(51) + 2 = 155. So

y2 ranges 2 to u − 147 and y1 ranges 49 to u−y
3
2
.
 
u − y2
FU (u) = P (U ≤ u) = P (3Y1 + Y2 ≤ u) = P Y1 ≤ =
3
Z u−147 Z u−y2 Z u−147  y = u−y2
3 1 y1 1 3
= dy1 dy2 = dy2 =
2 49 12 2 12 y1 =49
Z u−147 y =u−147
1 (2u − 6(49))y2 − y22 2
  
1 u − y2
= − 49 dy2 = =
2 12 3 12 6 y2 =2

1
(i) 72 (−(u − 147)2 + 2(u − 147)(u − 149) + 1)
1
(ii) 72 (−(u − 147)2 + 2(u − 147)(u − 149) + 2)
1
(iii) 72 (−(u − 147)2 + 2(u − 147)(u − 149) + 3)
1
(iv) 72 (−(u − 147)2 + 2(u − 147)(u − 149) + 4)

dF u−147 u−148 u−149 u−150

So fU (u) = dv
= (i) 36
(ii) 36
(iii) 36
(iv) 36
Section 3. The Method of Distribution Functions (ATTENDANCE 11) 195

In figure (b), easier to first determine non–shaded triangular region then

subtract from one (1) to find shaded region. For non–shaded region, u =
3y1 + y2 ranges u = 3(51) + 2 = 155 to u = 3(51) + 8 = 161, y2 ranges
u − 153 to 8 and y1 ranges u−y 3
2
to 51.

FU (u) = P (U ≤ u) = P (3Y1 + Y2 ≤ u) = 1 − P (3Y1 + Y2 > u)

Z 8 Z 51 Z 8  y =51
1 y1 1
= 1− dy1 dy2 = 1 − dy2 =
u−y2 12
u−153 12 y1 = 3
u−y2
u−153 3
Z 8 y =8
1 [306 − 2u]y2 + y22 2
  
1 u − y2
= 1− 51 − dy2 = 1 − =
u−153 12 3 12 6 y2 =u−153

1
(i) 1 − 72 ((u − 153)2 − 16(u − 153) + 64)
1
(ii) 1 − 72 ((u − 153)2 − 16(u − 153) + 65)
1
(iii) 72 ((u − 153)2 − 16(u − 153) + 66)
1
(iv) 72 ((u − 153)2 − 16(u − 153) + 67)

dF 158−u 159−u 160−u 161−u

So fU (u) = du
= (i) 36
(ii) 36
(iii) 36
(iv) 36

And
155
 Z 161 
 
u − 149 161 − u
Z
E(u) = u· du + u· du
149 36 155 36
 3 u=155  u=161
u 149u2 161u2 u3
= − + − =
108 72 u=149 72 108 u=155

(choose one) (i) 154 (ii) 155 (iii) 156 (iv) 157
(c) Density for U = 2Y1 + Y2 .

u=2 y + y
1 2
y u=2 y + y
2 1 2
u=2 y + y 8
1 2
4
8 6 8
2 y1 + y < u
2
4 2 y1 + y > u
1 2
2 2
2
0 y1 0 y1
49 51 49 51
(a) 49 50 51 (c)
(b)

Figure 6.8: Continuous bivariate, potato chips: U = 2Y1 + Y2

196 Chapter 6. Functions of Random Variables (ATTENDANCE 11)

In figure (a), U = 2Y1 + Y2 ranges u = 2(49) + 2 = 100 to u = 2(51) + 2 =

104, y2 ranges 2 to u − 98 and y1 ranges 49 to u−y
2
2
. In other words (choose
one)
(i)
u−y2
u−98
1
Z Z
2
FU (u) = dy1 dy2
2 49 12
(ii)
49−u−y2
2
1
Z Z
3
FU (u) = dy1 dy2
u−98 49 12
(iii)
u−y2
u−147
1
Z Z
2
FU (u) = dy1dy2
2 49 12
In figure (b), U = 2Y1 + Y2 ranges u = 2(51) + 2 = 104 to u = 2(49) + 8 =
1
106. In triangular regions 1, 2, 3 and 4, each with total probability of 12 ,
function y2 ranges u − 102 to 4, 4 to u − 100, u − 100 to 6 and 6 to u − 98,
respectively. and so, accounting for y1 ,
u−y2
4 51 u−100
1 1 1
Z Z Z Z
2
FU (u) = − dy1dy2 + dy1dy2
12 u−102
u−y2
2
12 4 50 12
u−y2
6 50 u−98
1 1 1
Z Z Z Z
2
+ − dy1 dy2 + dy1 dy2
12 u−100
u−y2
2
12 6 49 12

In figure (c), easier to first determine non–shaded triangular region then

subtract from one (1) to find shaded region. For non–shaded region, u =
2y1 + y2 ranges u = 2(49) + 8 = 106 to u = 2(51) + 8 = 110, y2 ranges
u − 104 to 8 and y1 ranges u−y 2
2
to 51. So (choose one)
(i)
u−y2
u−98
1
Z Z
2
FU (u) = dy1 dy2
2 49 12
(ii)
8 51
1
Z Z
FU (u) = dy1 dy2
u−104
u−y2
2
12
(iii)
8 51
1
Z Z
FU (u) = 1 − dy1dy2
u−104
u−y2
2
12
Section 3. The Method of Distribution Functions (ATTENDANCE 11) 197

4. Invertible distribution F (y) and uniform.

Draw a random sample of size 5 from Pareto distribution,
(
0,   y<β
F (y) = β
α
1− y , y ≥β

(a) Find Inverse. Since  α

β
F (y) = 1 − = u,
y
then
 y= F −1 (y) =G(u)
α= (choose one)
α 1 1
(i) βy (ii) 1 − βαy
(iii) 1 − β(1 − u)− α (iv) β(1 − u)− α

(b) Sample Pareto (α = 3, β = 3) using uniform distribution. One possible

random sample of size 5 from uniform on (0,1) gives

0.2189, 0.9913, 0.5775, 0.2846, 0.8068.

Use your calculator: 7 STO MATH PRB ENTER ENTER (again!) for “seed”,
then MATH PRB ENTER ENTER (again, five times, for sample).

Complete corresponding random sample of size 5 from Pareto dis-

tribution (α = 3, β = 3)
(
0, y<β=3
F (y) =  α  3
1 − βy = 1 − y3 , y ≥ β = 3

u 0.2189 0.9913 0.5775 0.2846 0.8068

1
g(u) = 3(1 − u)− 3 3.2575 14.586
(c) Sample Pareto (α = 1, β = 5) using uniform distribution.
Complete random sample of size 5 from Pareto distribution (α = 1, β = 5)
(
0,   y<5
F (y) = β
α
5
1− y = 1 − y, y ≥ 5

u 0.2189 0.9913 0.5775 0.2846 0.8068

g(u) = 5(1 − u)−1 6.4012 574.71
(d) Understanding this idea. Draw a random sample of size 5 from distribution

0, y<5
F (y) = 5
1 − y, y ≥ 5
198 Chapter 6. Functions of Random Variables (ATTENDANCE 11)

F(y)

G(u) = 5/(1 - u) = y

F(y) = 1 - 5/y = u

1
0 1 5 y

Figure 6.9: Invertible distribution F (y) and uniform.

Distribution F (y) is invertible, has inverse G(u) = F −1 (y),

5 5
F (y) = 1 − = u if and only if G(u) = y = F −1 (y) =
y 1−u
and so
   
5 5 5
P (G(U) ≤ y) = P ≤y =P U ≤1− =1− .
1−u y y
In other words, since distribution 0 ≤ F (y) ≤ 1 and u = F (y)
(choose one) (i) 0 ≤ u ≤ 1 (ii) u ≤ 0 (iii) u ≥ 1 (iv) u ≥ ∞;
that is, u, like a uniform random variable, has a value between 0 and 1.

And since distribution 0 ≤ F (y) ≤ 1 and F (y) = 1 − y5 ,

(i) 0 ≤ 1 − y5 ≤ 1 (ii) 1 − y5 ≤ 0 (iii) 1 − y5 ≥ 1 (iv) 1 − y5 ≥ ∞,
probability 1 − y5 has (of course) a value between 0 and 1.
 
So U = 1 − Y5 must be uniform on (0, 1) where P U ≤ 1 − y5 = 1 − y5
and since F (y) is invertible, y = G(u) = F −1 (y),
5
F (y) = P (G(U) ≤ y) = 1 − .
y
(e) Invertible distribution F (y) and uniform does not always work.
Distribution,
y − 14 y 2, 0 ≤ y ≤ 2

F (y) =
0, elsewhere
cannot be inverted because
1
F (y) = y − y 2 = u if and only if G(u) = y = F −1 (y) = ?
4
(choose one) (i) True (ii) False
Section 4. The Method of Transformations (ATTENDANCE 11) 199

6.4 The Method of Transformations

If Y is a continuous random variable with density f (y) > 0 and u = h(y) is a strictly
monotone differentiable function of y, then U has probability density

d −1
fU (u) = fY h−1 (u) h (u)
du

Method of transformations is a special case of method of distribution functions which

because h(y) is a strictly monotone differentiable function of y allows the use of the
chain rule of differentiation.

Exercise 6.4 (The Method of Transformations)

1. One dimensional U(Y ). Density

1 − y2 ,

0≤y≤2
f (y) =
0 elsewhere

(a) Density for U = 2Y .

u = h(y) = 2y, strictly monotonically increases in y and

y = h−1 (u) = (choose one) (i) 21 (ii) 12 − u (iii) u2 (iv) u
2
−1

dh−1 1 1 u u
du
= (choose one) (i) 2
(ii) 2
− u (iii) 2
(iv) 2
−1

dh−1 1
u
So fU (u) = fY [h−1 (u)] du
= fY 2 2
= (choose one)
1 1 u u u
(i) 2
(ii) 2
− 8
(iii) 2
(iv) 2
−1

where 0 ≤ u2 ≤ 2 or (choose one)

(i) −1 ≤ u ≤ 2 (ii) −2 ≤ u ≤ 1 (iii) −1 ≤ u ≤ 1 (iv) 0 ≤ u ≤ 4

This matches method of distribution result above.

(choose one) (i) True (ii) False
(b) Density for U = 2 − 2Y .

u = h(y) = 2 − 2y decreases in y and

u
y = h−1 (u) = (choose one) (i) 21 (ii) 12 − u (iii) 2
(iv) 2−u
2

1 u u
dh−1
du
= (choose one) (i) 2
(ii) − 12 (iii) 2
(iv) 2
−1

dh−1
= fY 2−u − 21 = (choose one)
 
So fU (u) = fY [h−1 (u)] du 2
1 1 u 2+u u
(i) 2
(ii) 2
− 8
(iii) 8
(iv) 2 − 1
200 Chapter 6. Functions of Random Variables (ATTENDANCE 11)

where 0 ≤ 2−u
2
≤ 2 or (choose one)
(i) −1 ≤ u ≤ 2 (ii) −2 ≤ u ≤ 1 (iii) −1 ≤ u ≤ 1 (iv) −2 ≤ u ≤ 2

(c) Density for U = Y 2 .

u = h(y) = y 2 increases in y and

q √ u
y = h−1 (u) = (choose one) (i) 12 (ii) u (iii) 2
(iv) 2−u
2

dh−1 1 1 u u
du
= (choose one) (i) 2
(ii) 2
√
u
(iii) 2
√
u
(iv) 2
−1

dh−1 √ 1
So fU (u) = fY [h−1 (u)] du
= fY [ u] √
2 u
=
3 2u 1 1 √1 1 1
(choose one) (i) 2
− 32
(ii) √
2 u
− 4
(iii) u
− 4
(iv) 2
√
u
+ 41 ,
√
where 0 ≤ u ≤ 2, or
(i) 0 ≤ u ≤ 1 (ii) 0 ≤ u ≤ 2 (iii) 0 ≤ u ≤ 3 (iv) 0 ≤ u ≤ 4

2. Another one dimensional U(Y ).

Given exponential density with mean β = 3,
( y y
1 −β
e = 13 e− 3 , y>0
f (y) = β
0 elsewhere

determine density of function U = Y + 3.

u = h(y) = y + 3 increases in y and

u
y = h−1 (u) = (choose one) (i) 12 (ii) u − 3 (iii) 2
(iv) 2−u
2

1 u
dh−1
du
= (choose one) (i) 2
(ii) − 12 (iii) 2
(iv) 1

dh−1
So fU (u) = fY [h−1 (u)] du
= fY [u − 3] |1| = (choose one)
u−3 u−3
1 1
(i) 2
(ii) 2
− u − 3 (iii) 31 e− 3 (iv) e− 3 −1

where y = u − 3 ≥ 0 or (choose one)

(i) u ≥ 0 (ii) u ≥ 1 (iii) u ≥ 2 (iv) u ≥ 3

3. Two dimensional U(Y1 , Y2). Bivariate density function is

 1
12
, 0 ≤ y1 ≤ 2, 0 ≤ y2 ≤ 6
f (y1 , y2 ) =
0 elsewhere
Section 4. The Method of Transformations (ATTENDANCE 11) 201

(a) Density for U = 3Y1 + Y2 .

For fixed y1 , u = h(y2 ) = 3y1 + y2 increases in y2 and

y2 = h−1 (y1 , u) =
√
(choose one) (i) 21 + y1 (ii) uy1 (iii) u − 3y1 (iv) 2−u
2

dh−1 1 1 u
du
= (choose one) (i) 2
(ii) 2
√
u
(iii) 2
√
u
(iv) 1

dh−1 1
So gU (y1 , u) = fY [y1 , h−1 (y1 , u)] du
= fY [y1 , u − 3y1 ] |1| = 12
|1| =
3 2u 1 1 √1 1 1
(choose one) (i) 2
− 32
(ii) √
2 u
− 4
(iii) u
− 4
(iv) 12

y
2
u=3 y + y
u=3 y + y 1 2
1 2

6 6
3 y1 + y < u
2
3 y1 + y > u
2

0 2 y1 0 2 y1

(a) (b)

Figure 6.10: Continuous bivariate: U = 3Y1 + Y2

Since 0 ≤ y2 = u − 3y1 ≤ 6 (choose one)

(i) u−6
3
≤ y1 ≤ u3
(ii) u+6
3
≤ y1 ≤ u3
(iii) −2 ≤ y1 ≤ u3
(iv) u−63
≤ y1 ≤ 0

In figure (a), u = 3y1 + y2 ranges 3(0) + 0 = 0 to 3(2) + 0 = 6,

so either −2 ≤ u−6 3
≤ y1 ≤ 0 or 0 ≤ y1 ≤ u3 ≤ 2. Since only second
inequality is possible,
u  y1 = u3
1 1
Z
3
fU (u) = dy1 = y1 =
0 12 12 y1 =0

u+3 u+2 u+1 u

(choose one) (i) 36
(ii) 36
(iii) 36
(iv) 36

In figure (b), u = 3(2) + 0 = 6 to u = 3(2) + 6 = 12,

so either 0 ≤ u−6
3
≤ y1 ≤ 2 or 2 ≤ y1 ≤ u3 ≤ 4. Since only first inequality
202 Chapter 6. Functions of Random Variables (ATTENDANCE 11)

is possible, y1 =2
2 
1 1
Z
fU (u) = dy1 = y1 =
u−6
3
12 12 y1 = u−6
3

12−u 13−u 14−u 15−u

(choose one) (i) 36
(ii) 36
(iii) 36
(iv) 36

This matches method of distribution result above.

(choose one) (i) True (ii) False
(b) Density for U = Y1 Y2 .

For fixed y1 , u = h(y2 ) = y1 y2 increases in y2 and

y2 = h−1 (y1 , u) =
√
(choose one) (i) y1u+1 (ii) uy1 (iii) u − 3y1 (iv) u
y1

dh−1 1 1 √1
du
= (choose one) (i) 2
(ii) y1
(iii) u
(iv) 1 + y1
h i
So gU (y1 , u) = fY [y1 , h−1 (y1 , u)] dh−1
du
= fY y1 , yu1 y11 = 1
12
1
y1
=
3 2u 1
√ − 1 √1 1 1
(choose one) (i) 2
− 32y1
(ii) 2 u 4
(iii) u
− 4
(iv) 12y1

y2 y1 y2 > u
6
y2 = u_
2
y1 y2 < u y1 y2 = u
0 2 y1

Figure 6.11: Continuous bivariate: U = Y1 Y2

Since 0 ≤ y2 = yu1 ≤ 6 (choose one)

(i) u−6
3
≤ y1 ≤ u3
(ii) u+6
3
≤ y1 ≤ u3
(iii) 0 ≤ y1 ≤ u6
(iv) u6 ≤ y1 ≤ 2

So
2
1 1
Z
fU (u) = dy1 = [ln y1 ]yy11 =2
=u =
u
6
12y1 12 6

1 u 1 u 1 u u
(choose one) (i) 1 − 12
ln 12 (ii) 12
ln 12 (iii) − 12 ln 12 (iv) 36
Section 4. The Method of Transformations (ATTENDANCE 11) 203

4. Relationship between method of transformations and distribution functions.

True / False. Method of transformations is a special case of method of distri-
bution functions which, because u = h(y) is a strictly monotone differentiable
function of y, means y = h−1 (u) exists and so allows use of the chain rule of
differentiation: derivative of distribution FU (u) has density

d

fU (u) = FY h−1 (u)
dy

d −1
= FY′ h−1 (u) h (u)
du

d −1
= fY h−1 (u) h (u)
du