Module 19
Joint Distribution of Random Variables
() Module 19 Joint Distribution of Random Variables 1 / 19
(Ω, P(Ω), P): a given probability space;
In many situations we may be interested in two or more numerical
characteristics of the sample space simultaneously;
Example 1:
E: Casting a red and a white die;
Ω = {(i, j) : i, j ∈ {1, 2, . . . , 6}} ,
where, in (i, j) ∈ Ω, i denotes the number of spots on upper face of
red die and j denotes the number of spots on upper face of white die.
Define r.v.s X1 : Ω → R and X2 : Ω → R as:
X1 ((i, j)) = i + j = sum of numbers on upper faces of two dice;
X2 ((i, j)) = |i − j| = absolute difference of numbers on upper faces
of two dice.
() Module 19 Joint Distribution of Random Variables 1 / 19
We may be interested in studying X1 and X2 simultaneously, i.e., we
may be interested in studying the function
X : Ω → R2 ,
where X = (X1 , X2 )0 , X ((i, j)) = (X1 (i, j), X2 (i, j))0 , (i, j) ∈ Ω.
Definition 1: A function
X = (X1 , . . . , Xp )0 : Ω → Rp
is called a p-dimensional random vector (r.v.); here
Rp = {x = (x1 , . . . , xp ) : −∞ < xi < ∞, i = 1, . . . , p}
denotes the p-dimensional Euclidean space.
For A ⊆ Rp (i.e., A ∈ P(Rp )), define
X −1 (A) = {ω ∈ Ω : X (ω) ∈ A} .
() Module 19 Joint Distribution of Random Variables 2 / 19
Further define the set function PX : P(Rp ) → [0, 1] as
PX (A) = P X −1 (A)
= P ({ω ∈ Ω : X (ω) ∈ A}) , A ∈ P(Rp ).
() Module 19 Joint Distribution of Random Variables 3 / 19
Result 1:
(Rp , P (Rp ) , PX ) is a probability space, i.e., PX (·) is a probability function
on P(Rp ).
Proof: Same as for p = 1
Definition 2 : The probability space (Rp , P(Rp ), PX ) is called the
probability space induced by r.v. X (or simply the induced probability
space).
X
Rp , P (Rp ) , PX
(Ω, P(Ω), P) −→
() Module 19 Joint Distribution of Random Variables 4 / 19
Notations:
For A ∈ P (Rp ) , {X ∈ A} ≡ {ω ∈ Ω : X (ω) ∈ A};
For a = (a1 , . . . , ap ), b = (b1 , . . . , bp ) ∈ Rp
(a, b] = (a1 , b1 ] × (a2 , b2 ] × · · · × (ap , bp ]
= {x = (x1 , . . . , Xp ) : ai < xi ≤ bi , i = 1, . . . , p}
denotes the p-dimensional semi-closed rectangle with vertices
{z = (z1 , . . . , zp ) ∈ Rp : zi ∈ {ai , bi } , i = 1, . . . , p} ,
(in all 2p vertices).
() Module 19 Joint Distribution of Random Variables 5 / 19
∆p ≡ ∆p ((a, b])
= {z = (z1 , . . . , zp ) ∈ Rp : zi ∈ {ai , bi } , i = 1, . . . , p}
= set of all vertices of p-dimensional rectangle (a, b]
[p
= ∆k,p ((a, b]),
k=0
where
∆k,p ≡ ∆k,p ((a, b])
= {z = (z 1 , . . . , z p ) ∈ ∆p : k of zi s are aj s}, k = 0, 1, . . . , p.
() Module 19 Joint Distribution of Random Variables 6 / 19
Example 2:
p=2
∆2 = {(b1 , b2 ), (a1 , b2 ), (b1 , a2 ), (a1 , a2 )}
∆0,2 = {(b1 , b2 )}
∆1,2 = {(a1 , b2 ), (b1 , a2 )}
∆2,2 = {(a1 , a2 )}
() Module 19 Joint Distribution of Random Variables 7 / 19
Notation :
For a = (a1 , . . . , ap ), b = (b1 , . . . , bp ) ∈ Rp
(−∞, a] = {z ∈ Rp : zi ≤ ai , i = 1, . . . , p}
(a, ∞) = {z ∈ Rp : zi > ai , i = 1, . . . , p}
(a, b) = {z ∈ Rp : ai < zi < bi , i = 1, . . . , p}
[a, b], [a, b), (−∞, a) and [a, ∞) are defined similarly.
a ≤ b means ai ≤ bi , i = 1, . . . , p.
a < b means ai ≤ bi , i = 1, . . . , p, with aj < bj , for some
j ∈ {1, . . . , p}.
() Module 19 Joint Distribution of Random Variables 8 / 19
Definition 3:
Let X = (X1 , . . . , Xp )0 be a p-dimensional r.v.
(a) The joint distribution function (d.f.) of X is the function
FX : Rp → R defined by
FX (x) = PX ((−∞, x])
= P(X −1 ((−∞, x]))
= P({ω ∈ Ω : X (ω) ≤ x})
= P({ω ∈ Ω : Xi (ω) ≤ xi , i = 1, . . . , p}), x ∈ Rp .
(b) The joint d.f. of any subset of r.v.s {X1 , . . . , Xp } is called a marginal
d.f. of FX (·).
() Module 19 Joint Distribution of Random Variables 9 / 19
Example 3:
Let X = (X1 , X2 , X3 , X4 )0 be a r.v. with d.f. FX (·). Then
FX1 (.) : d.f. of X1 ;
Fx2 ,x4 (.) : joint d.f. of (X2 , X4 )0
Fx1 ,x3 ,x4 (·) : joint d.f. of (X1 , X3 , X4 )0
are marginal d.f.s of X = (X1 , X2 , X3 , X4 )0 .
() Module 19 Joint Distribution of Random Variables 10 / 19
Result 2:
Let X = (X1 , . . . , Xp )0 be a r.v. with joint d.f. FX1 ,...,Xp (·). Then the
marginal d.f. of (X1 , . . . , Xp−1 ) is
FX1 ,...,Xp−1 (x1 , . . . , xp−1 ) = lim FX1 ,...,Xp−1 ,Xp (x1 , . . . , xp−1 , t),
t→∞
(x1 , . . . , xp−1 ) ∈ Rp−1 .
Proof: Let (x1 , . . . , xp−1 ) ∈ Rp−1 be fixed. Then
FX1 ,...,Xp−1 (x1 , . . . , xp−1 ) = P ({X1 ≤ x1 , . . . , Xp−1 ≤ xp−1 })
= P ({X1 ≤ x1 , . . . , Xp−1 ≤ xp−1 , Xp < ∞})
∞
!
[
= P {X1 ≤ x1 , . . . , Xp−1 ≤ xp−1 , Xp ≤ t}
t=1
= lim P ({X1 ≤ x1 , . . . , Xp−1 ≤ xp−1 , Xp ≤ t})
t→∞
= lim FX1 ,...,Xp−1 ,Xp (x1 , . . . , xp−1 , t).
t→∞
() Module 19 Joint Distribution of Random Variables 11 / 19
Remark 1:
The above result suggests that to get a marginal d.f. take (in limit) the
arguments of unwanted variables in the joint d.f. to ∞.
Result 3: Let X = (X1 , . . . , Xp )’ be a p-dimensional r.v. with joint d.f.
FX (·). Then, for any p-dimensional rectangle (a, b] (a, b ∈ Rp , a < b),
P ({X ∈ (a, b]}) = P ({ai < Xi ≤ bi , i = 1, . . . , p})
Xp X
= (−1)k FX (z).
k=0 z∈∆k,p ((a,b])
Proof: For p = 2
P ({a1 < X1 ≤ b1 , a2 < X2 ≤ b2 }) = P ({X1 ≤ b1 , a2 < X2 ≤ b2 })
−P ({X1 ≤ a1 , a2 < X2 ≤ b2 })
P ({X1 ≤ b1 , a2 < X2 ≤ b2 }) = P ({X1 ≤ b1 , X2 ≤ b2 , })
−P ({X1 ≤ b1 , X2 ≤ a2 , })
() Module 19 Joint Distribution of Random Variables 12 / 19
= FX1 ,X2 (b1 , b2 ) − FX1 ,X2 (b1 , a2 )
P ({X1 ≤ a1 , a2 < X2 ≤ b2 }) = P ({X1 ≤ a1 , X2 ≤ b2 , })
−P ({X1 ≤ a1 , X2 ≤ a2 , })
= FX1 ,X2 (a1 , b2 ) − FX1 ,X2 (a1 , a2 ).
Therefore
P ({X ∈ (a, b]}) = F X1 ,X2 (b1 , b2 ) − [FX1 ,X2 (a1 , b2 ) + FX1 ,X2 (b1 , a2 )]
+FX1 ,X2 (a1 , a2 )
2
X X
= (−1)k FX1 ,X2 (z1 , z2 ).
k=0 z∈∆k,2 ((a,b])
Thus the result holds for p = 2. Suppose, for any probability function,
result holds for some p = m.
() Module 19 Joint Distribution of Random Variables 13 / 19
Let
Y = (X1 , . . . , Xm , Xm+1 ) = (X , Xm+1 ) , a0 = (a1 , . . . , am ), b0 =
(b1 , . . . , bm ), a = (a1 , . . . , am , am+1 ) and b = (b1 , . . . , bm , bm+1 ). Then
P ({Y ∈ (a, b]}) = P X ∈ (a0 , b0 ], Xm+1 ∈ (am+1 , bm+1 ]
= P X ∈ a0 , b0 | {Xm+1 ∈ (am+1 , bm+1 ]} ×
P ({Xm+1 ∈ (am+1 , bm+1 ]}) .
Using the result for p = m (for conditional probability function) we have
P X ∈ a0 , b0 | {Xm+1 ∈ (am+1 , bm+1 ]}
Xm X
= (−1)k P ({X ≤ z}| {Xm+1 ∈ (am+1 , bm+1 ]}) .
k=0 z∈∆k,m ((a0 ,b0 ])
Thus
P ({Y ∈ (a, b]})
m
X X
= (−1)k P({X ≤ z, am+1 < Xm+1 ≤ bm+1 })
k=0 z∈∆k,m ((a0 ,b0 ])
() Module 19 Joint Distribution of Random Variables 14 / 19
m
X X
= (−1)k [P({X ≤ z, Xm+1 ≤ bm+1 })
k=0 z∈∆k,m ((a0 ,b0 ])
−P(X ≤ z, Xm+1 ≤ am+1 )]
m+1
X X
= (−1)k P({Xi ≤ zi , i = 1, . . . , m + 1})
k=0 z∈∆k,m+1 ((a,b])
m+1
X X
= (−1)k (−1)k FX (z).
k=0 z∈∆k,m+1 ((a,b])
() Module 19 Joint Distribution of Random Variables 15 / 19
Take Home Problems
1. Let X = (X1 , X2 ) be a r.v. having the joint d.f.
xy 2 ,
if 0 ≤ x < 1, 0 ≤ y < 1
x, if 0 ≤ x < 1, y ≥ 1
(x,y )
F x1 ,x2 = y 2, if x ≥ 1, 0 ≤ y < 1 .
1, if x ≥ 1, y ≥ 1
0, otherwise
Find the marginal d.f.s of X1 and X2 .
() Module 19 Joint Distribution of Random Variables 16 / 19
2. Consider the function G : R2 → R defined by
x, if 0 ≤ x < 1, y ≥ 1
2
y , if x ≥ 1, 0 ≤ y < 1
G (x, y ) = .
1, if x ≥ 1, y ≥ 1
0, otherwise
Verify whether or not
2
X X
(−1)k FX (z1 , z2 ) ≥ 0,
k=0 z∈∆k,2 (a,b]
holds for all rectangles (a, b], where
a1 ∈ [0, 1), a2 ∈ [0, 1), b1 ∈ [1, ∞) and b2 ∈ [1, ∞).
() Module 19 Joint Distribution of Random Variables 17 / 19
Abstract of Next Module
We will derive various properties of joint distribution function
() Module 19 Joint Distribution of Random Variables 18 / 19
Thank you for your patience
() Module 19 Joint Distribution of Random Variables 19 / 19