The Density of the F Distribution

Stat 305 Spring Semester 2006

The purpose of this document is to determine the pdf of the Fm;n distribution. Recall
that the Fm;n distribution is the ratio of two (scaled) independent 2 random variables, the
…rst having m degrees of freedom and the second having n degrees of freedom.

Proposition 1 If X is Fm;n , then

m+n
2
mm=2 nn=2 x(m=2) 1
fX (x) = m n (m+n)=2
, x > 0.
2 2
(n + mx)

Proof. Suppose that X = (U=m)=(V =n) where U and V are independent 2 random vari-
ables, U having m degrees of freedom and V having n degrees of freedom. For convenience,
let c = m=2 and d = m=2. To compute the distribution of X, we need the joint pdf of U
and V . First note that
1
fU (u) = uc 1 e u=2
, u > 0 and
(c)2c
1
fV (v) = vd 1e v=2
, v > 0.
(d)2d
Therefore, since U and V are independent by assumption,
1 1
fU;V (u; v) = c
uc 1 e u=2 v d 1 e v=2
(c)2 (d)2d
1
= uc 1 v d 1 e u=2 e v=2 , u; v > 0.
(c) (d)2c+d
We will …rst …nd the distribution of the random variable U=V by using the cdf method.
Speci…cally,

FU=V (x) = P (U=V x) = P (U xV ).

Putting U on the horizontal axis and V on the vertical axis, we can graph the inequality
U xV . See Figure 1. Integrating …rst with respect to u and then with respect to v, we
obtain
Z 1 Z xv
1
P (U xV ) = c+d
uc 1 e u=2 du v d 1 e v=2 dv.
(c) (d)2 0 0

We will solve this integral without actually integrating anything! Di¤erentiate the integral
with respect to x moving the d=dx past the …rst integral. We get
Z 1
0 1
fU=V (x) = FU=V (x) = c+d
(vx)c 1 e vx=2 v v d 1 e v=2 dv
(c) (d)2
c 1 Z0 1
x
= v c+d 1 e [(x+1)=2]v dv.
(c) (d)2c+d 0

1
(Note that the extra v in the …rst line comes from the application of the Chain Rule to the
inner integral.)

Figure 1: The inequality U xV

Observe that
x+1 c+d
2 c+d 1 [(x+1)=2]v
f (v) = v e
(c + d)
is
R 1the pdf of a Gamma random variable with parameters = c + d and = (x + 1)=2. Since
0
f (v)dv = 1,
xc 1 (c + d)
fU=V (x) =
(c) (d)2c+d x+1 c+d
2
c 1
(c + d) x
= .
(c) (d) (x + 1)c+d
We now have that
fX (x) = f(U=m)=(V =n) (x) = f(n=m)(U=V ) (x)
m m
= fU=V x
n n
m+n m (m=2) 1
m 2 n
x
= m n (m+n)=2
n 2 2
m
x+1
n
m+n m=2 n=2 (m=2) 1
2
m n x
= .
m
2
n
2
(n + mx)(m+n)=2

2
This completes the proof.

Using Maple, we can plot the graphs of Fm:n using the following commands.

f := (m; n; x) > (GAMMA((m + n)=2) m^(m=2) n^(n=2) x^((m=2) 1))=

(GAMMA(m=2) GAMMA(n=2) (n + m x)^((n + m)=2));

plot(f(2; 5; x); x = 0::6);

m = 2, n = 5 m = 10, n = 5

3
m = 10, n = 20 m = 40, n = 20

m = 10, n = 10 m = 80, n = 80