You are on page 1of 2

Statistics 351 (Fall 2015

)
Prof. Michael Kozdron

November 13, 2015

Lecture #23: The Bivariate Normal Density Function
Reference. §5.5 pages 125–126
Last lecture we derived the density function for the multivariate normal by starting with
independent and identically distributed N (0, 1) random variables, performing a linear transformation, and using the change-of-variables formula from Chapter 1.
⇤] > 0 is
Definition III. A random vector X with E(X) = µ and Cov(X) = ⇤ where det[⇤
µ, ⇤ ) if and only if its density function is given by
N (µ
✓ ◆ n2

1
1
1
p
x µ )0⇤ 1 (x
x µ) .
x) = fX (x1 , . . . , xn ) =
exp
fX (x
(x
2⇡
2
⇤]
det[⇤
µ, ⇤ )
The case of the bivariate normal is of particular importance. Let X = (X1 , X2 )0 2 N (µ
with

 2
µ1
⇢ 1 2
1
µ=
and ⇤ =
.
2
µ2
⇢ 1 2
2
2
j ),

In particular, Xj 2 N (µj ,

j = 1, 2, with Cov(X1 , X2 ) = ⇢

Cov(X1 , X2 )

Corr(X1 , X2 ) = p
=
Var(X1 ) Var(X2 )

Furthermore,

⇤] =
det[⇤

Provided

1

1

⇢2

2 2
1 2

=

2 2
1 2 (1

1 2

Notice that

= ⇢.

1 2

⇢2 ).

⇤] > 0 so that X has a density. Since
> 0, and 1 < ⇢ < 1, then det[⇤


2
2
1
1
⇢ 1 2
⇢ 1 2
2
2
=
=
2
2
2
2
2
⇢ 1 2
⇢ 1 2
⇤]
det[⇤
⇢)
1
1
1 2 (1

1
1/ 12
⇢/ 1 2
=
,
2
⇢/ 1 2
1/ 22
1 ⇢

> 0,

2 2
1 2

1 2.

2

we conclude that
fX (x1 , x2 )
=

2⇡

1 2

1
p

1

⇢2

exp

(

1
⇢2 )

2(1

In the special case that µ = 0 ,
fX (x1 , x2 ) =

2⇡

p

2
1

1
1

=

x1

µ1
1

◆2

2⇢

(x1

µ1 )(x2

µ2 )

+

1 2

x2

µ2
2

◆2 !)
(⇤)

2
2

= 1, we find

1
exp
x21
2)
2
2(1

2⇢x1 x2 + x22

.

Remark. The density is nicely typed on page 126. At the top of page 126, a general formula
for ⇤ 1 is used (and given in equation (1.6) on page 119). We will not require this fact, and
will only be concerned with the bivariate density as derived above.
23–1

.

fX (x) fY |X=x (y) = Since X 2 N (µx . ⇤ ) with Suppose that X = (X.Y (x. y) is given by fX.Y (x. X 2 N (µx . and Corr(X. To compute the conditional density function for Y |X = x. x2 ). we see that Y |X = x 2 N ✓ µy + ⇢ y (x 2 y (1 µx ). y2 ). Suppose further that ⇤] > 0 so that X has a density given by Definition III.6 pages 127–130 µ. Y 2 N (µy . By definition.Y (x.Y (x. det[⇤ Goal. In particular. y) by fX (x) and simplifying gives ( ✓ 1 1 fY |X=x (y) = p p exp y 2 y2 (1 ⇢2 ) 1 ⇢2 y 2⇡ µy ⇢ y x In other words.Conditional Distributions for the Bivariate Normal Reference. E(Y |X) = µy + ⇢ y (X µx ) x and Var(Y |X) = 23–2 2 y (1 ⇢2 ). (x We also know from (⇤) that the joint density fX. 2 ◆ ⇢) . (x + ✓ y µx ) µy y ◆2 ) ◆2 !) . y) . . x In particular. 2 x) we know that fX (x) = x 1 p 2⇡ ⇢ exp 1 2 2 x µx ) 2 . . Y ) = ⇢. fX. §5. y) = 2⇡ x 1 p 1 y ⇢2 exp ( 1 2(1 ⇢2 ) ✓ x µx x ◆2 2⇢ (x µx )(y µy ) x y Dividing fX. Y )0 2 N (µ   µx µ= and ⇤ = ⇢ µy 2 x ⇢ x y x y 2 y .