Multivariate Probability Distributions

CHAPTER 5
Bivariate and Marginal Probability
Distributions

 In the previous chapters we dealt with experiments that produced a

single numerical response, or random variable. We discusses, for
example, the height of a person or the weight of a person.

 For an experiment, one often can make several different measurements

Y1, . . . , Yn simultaneously,

 – the height and weight of a person;

 – the temperature and rainfall of a day;
 – the two coordinates of a needle randomly dropped on a table;
 – the number of 1s and the number of 6s in 10 rolls of a die.
 If only two random variables are involved, these joint distributions are
called bivariate distributions.

 Example. We are interested in the effect of seat belt use on saving

lives. If we consider the following random variables X1 and X2 defined
as follows:
 X1 =0 if child survived
 X1 =1 if child did not survive

 And X2 = 0 if no belt
 X2 = 1 if adult belt used
 X2 = 2 if child seat used
 The following table represents the joint probability distribution of X1
and X2 . In general we write P(X1 = x1 , X2 = x2 ) = p(x1 , x2) and call
p(x1 , x2) the joint probability function of (X1 , X2).

 X1
 0 1
 --------------------------------------
 0 | 0.38 0.17 | 0.55
 X2 1 | 0.14 0.02 | 0.16
 2 | 0.24 0.05 | 0.29
 ------------------------------------------
 0.76 0.24
 Probability that a child will both survive and be in a child seta when
involved in an accident is:

 P(X1 = 0, X2 = 2) = 0.24

 Probability that a child will be in a child seat:

 P(X2 = 2) = P(X1 = 0, X2 = 2) + P(X1 =1, X2 = 2) = 0.24+0.05= 0.29

 Definition. For discrete random variables X1 and X2 , the joint
probability distribution is p(x1 , x2) = P(X1 = x1 , X2 = x2 ) .

 Marginal probability function of X1 is given by

 p1 (x1) =∑x2 p(x1 , x2). This is the probability function of X1 .


 The conditional probability distribution for X1 given X2 fixes a value of X2
and looks at the relative frequencies for values of X1. For example,
conditioning on X2 = 0 produces

 P(X1 =0 | X2 = 0) = [P(X1 =0 , X2 = 0) ] / [P( X2 = 0) ] = 0.38/0.55=0.69

 and similarly

 P(X1 =1 | X2 = 0) = 0.31.

 These two values show how survivorship relates to the situation in which no
seat belt is in use, 69% of children in such situation survived fatal accidents.
Diagram
Probabilities

Independent Random Variables
Exercise
 Let X and Y denote the proportion of two different chemicals in a sample
mixture of chemicals used as an insecticide. Suppose X and Y have joint
probability density given by:

 2, 0 ≤ x ≤ 1,0 ≤ y ≤ 1,0 ≤ x + y ≤ 1
f ( x, y) = 
 0, elsewhere

 (Note that X + Y must be at most unity since the random variables denote
proportions within the same sample).
 1) Find the marginal density functions for X and Y.
 2) Are X and Y independent?
 3) Find P(X > 1/2 | Y =1/4).

 1) 1− x
f1 ( x ) =  ∫
 2dy = 2(1 − x ), 0 ≤ x ≤ 1
0
 0 otherwise


1− y
f 2 ( y) =  ∫

 2dx = 2(1 − y ), 0 ≤ y ≤ 1
0
 0 otherwise

 2) f1(x) f2(y)=2(1-x)* 2(1-y) ≠ 2 = f(x,y), for 0 ≤ x ≤ 1-y.
 Therefore X and Y are not independent.

 3)

1
1 f ( x, y = )
1 1
 1 1 1 4 dx = 2 2
P  X > | Y =  = ∫ f ( x | y = )dx = ∫ ∫/ 2 1 3 =
 2 4  1/ 2 4 1/ 2 f ( y =
1
) 1 2(1 − )
4 4
 If X and Y are independent, then it follows that
E[G(X)H(Y)]=E[G(X)]E[H(Y)].

 The covariance is a measure of how strongly related are two variables

X and Y.

 If Y tends to be large when X is large, and small when X is small,

then X and Y will have a positive covariance.

 If, on the other hand Y tends to be small when X is large, and large
when X is small, then X and Y will have a negative covariance.
While covariance measures the direction of the association between two
random variables, correlation measures the strength of the association.
 Clearly, the correlation between the two variables should not be dependent
on the actual units used, but simply how the variables are related to each
other. This can be achieved simply by normalizing or scaling the
covariance, which yields the correlation coefficient.
>Plot(x,y)
Example
 The proportions X and Y of two chemicals found in samples of an
insecticide have the joint probability density function

2, 0 ≤ x ≤ 1,0 ≤ y ≤ 1,0 ≤ x + y ≤ 1

f ( x, y ) = 
0, elsewhere

 The random variable Z=X + Y denotes the proportion of the insecticide

due to both chemicals combined.
 1) Find E(Z) and V(Z)
 2) Find an interval in which values of Z should lie at least 50% of the
samples of insecticide.
 3) Find the correlation between X and Y and interpret its meaning.
 Solutions
1 1− x 1
2
E( X +Y ) = ∫ ∫ ( x + y )2dydx = ∫ (1 − x )dx =
2

0 0 0
3

1 1− x 1 4
2 2 x 23 1
E[( X + Y ) 2 ] = ∫ ∫0 ( x + y ) 2
2dydx = ∫0 3 (1 − x ) 3
dx = ( x − ) |1
0=  =
0
3 4 34 2

2
1 2 1
V ( Z ) = E ( Z 2 ) − E ( Z )2 = −   =
2 3 1
 Solution
 Using Tchebysheff’s theorem with k= 2 we have,

2 2 2 2
P  − < X +Y < +  ≥ 0 .5

 3 18 3 18 
 The desired interval is (1/3, 1).
 Solution
1− X 1− X
f (X ) = ∫
0
f ( X , Y )dy = ∫ 2dY = 2(1 − X ),0 ≤ X ≤1
0
1− Y 1− Y
f (Y ) = ∫
0
f ( X , Y )dx = ∫ 2dX = 2(1 − Y ),0 ≤ Y ≤1
0
1
1 1
z z 2 3
 1 1  1 1
E ( X ) = E (Y ) = ∫ z 2(1 − z )dz = 2∫ z − z 2dz = 2 −  = 2 −  = 2  =
0 0  2 3  0  2 3  6 3
1
1 1
z z 
3 4
1 1  1  1
E ( X 2 ) = E (Y 2 ) = ∫ z 2 (1 − z )dz = 2∫ z 2 − z 3dz = 2 −  = 2 −  = 2  =
0 0  3 4  0  3 4   12  6
 Solution
2
1 1 1 1 1
Var ( X ) = Var (Y ) = E ( X ) − ( E ( X )) = −   = − =
2 2

6 3 6 9 18
1 1− x 1 1− x 1 1− x

E ( XY ) = ∫ ∫ xy 2dxdy = ∫ x ∫ 2 ydydx = ∫ x ( y 2 ) dx =
0 0 0 0 0 0
1
1 1
x 2x x  2
1 2 1 1 3 4
= ∫ x (1 − x ) dx = ∫ x − 2 x + x dx = 
2 2
 − 3
+ 
 = − + =
0 0  2 3 4  0 2 3 4 12
2
1 1 1 1
−  −
12  3  12 9 1
ρ= = =−
2 1 2
1
  18
 18 
HOMEWORK 3 – Due January 25,
2007
 4.57
 4.63
 5.3
 5.9
 5.19
 6.15
 6.24
 6.35