.
f ( y | ) g (T ( y ) | )h( y ) h( y )
Since this ratio does not depend on , the assumptions of
the theorem imply that S ( x ) S ( y ) . Thus, S ( X ) is at
least as coarse a partition of the sample space as T ( X ) , and
consequently S ( X ) is minimal sufficient.
Example 1: Consider the ratio
x
n
x
i1 i (1 ) i1 i
n
f (x | )
.
n
n
f ( y | ) i1 yi (1 ) n i1 yi
If this ratio is constant as a function of , then we must
n
n
have i 1 xi i 1 yi . Since we have shown that
n
n
p( x | ) x (1 ) n x
x
n
exp[ x log
n log(1 )]
x
1
This is a one-parameter exponential family with
n
( ) log
,
B
(
n
log(1
),
T
(
x
)
x
,
h
(
x
)
1
x
The family of distributions obtained by taking iid samples
from one-parameter exponential families are themselves
one-parameter exponential families.
Specifically, suppose X ~ P and {P : } is an
exponential family, then for X1 , K , X n iid with common
distribution P ,
h
(
x
)
exp
)
T ( xi ) nB( )
i
i
i 1
p( x1 , K , xn | )
A sufficient statistic is
n
i 1
n
i 1
p( x | ) h( x ) exp{ ( )T ( x ) B ( )}
Then the family of the distributions of the statistic T ( X ) is
a one-parameter exponential family of discrete distributions
whose pdf may be written
h *(t ) exp{ ( )t B ( )}
for suitable h*.
Proof: By definition,
P [T ( x) t ] p ( x | )
{ x:T ( x ) t }
h( x) exp[ ( )T ( x) B( )]
{ x:T ( x ) t }
exp[ ( )t B( )]{
*
h
If we let (t )
{ x:T ( x ) t }
h( x)}
{ x:T ( x ) t }
x e 1
p( x | )
exp{x log }
x!
x!
Letting log , we have
1
p ( x | ) exp{ x exp( )}, x={0,1,2,...} .
x!
We have
1
A( ) log e x
x 0 x !
(e ) x
log
x!
x 0
log exp(e ) e
Thus, { :| A( ) | } .
Note that if 1 , then would still be a one-parameter
exponential family but it would be a strict subset of the
canonical one-parameter exponential family generated by T
and h with natural parameter space { :| A( ) | }
.
h( x) exp[(s )T ( x) A( )]dx
{exp[ A( s ) A( )]} L h( x) exp[( s )T ( x) A( s )]dx
exp[ A( s ) A( )]
E [ X ]
d
e
e
log
d log
d2
Var [ X ] 2 e
e
log
d
log
Example 2: Suppose X 1 ,K , X n is a sample from a
population with pdf
x
x2
p ( x | ) 2 exp( 2 ), x 0, 0
2
This is known as the Rayleigh distribution. It is used to
model the density of time until failure for certain types of
equipment. The data comes from an exponential family:
n
xi2
n xi
p ( x1 ,K , xn | ) 2 exp( 2 )
i 1 2
i 1
1
xi exp( 2
2
i 1
n
x
i 1
2
i
n log 2 )
Here
1
1
2
,
, B( ) n log 2 , A( ) n log(2 ) .
2
2
2
n
X
i 1
2
i
has mean
g ( x3 , ) g ( x4 , ) T ( x3 ) T ( x4 )
is constant as a function of .
1
exp log
2
(1
(
x
)
exp{ log log[1 ( x ) 2 ]}, , x
Thus, for the Cauchy family,
g ( x , ) log log[1 ( x ) 2 ] .
For any four sample points x1 , x2 , x3 , x4 ,
g ( x1 , ) g ( x2 , ) log[1 ( x1 ) 2 ] log[1 ( x2 ) 2 ]
g ( x3 , ) g ( x4 , ) log[1 ( x3 ) 2 ] log[1 ( x4 ) 2 ]
This is not constant as a function of so the Cauchy
family is not an exponential family.
II. Multiparameter exponential families
One-parameter exponential families have a natural onedimensional sufficient statistic regardless of the sample
10
p ( x | ) h( x ) exp{ j ( )T j ( x ) B ( )}
j 1
2
i 1 i
exp 2 2 i 1 xi 2 2 2
2
11
n!
p1 y1 ( x ) L pk yk ( x )
y1 ( x ) L yk ( x )
y1 ( x )
yk 1 ( x )
p1
p
n!
L k 1
pk n
y1 ( x ) L yk ( x ) pk
p k
n!
n!
y1 ( x ) L yk ( x )
k 1
Moments of Sufficient Statistics: As with the oneparameter exponential family, it is convenient to index the
family by (1 ,K ,k ) . The analogue of Theorem 1.6.2
that calculates the moments of the sufficient statistics is
Corollary 1.6.1:
A
A
E0 T ( X )
(0 ), K ,
(0 )
k
1
2 A
Var0 T ( X )
(0 )
a b
12
pi
))]
pk
E[ y j ( x )]
n log 1 e i
j
i 1
nei
k 1
1 e
i 1
pi
pk
k 1
1
i 1
pi
pk
npi
1
pi
pk
pk
n
k 1
nei e j
i
Cov0 [ yi ( x ), y j ( x )]
n log 1 e
npi p j , i j
k 1
j k
i 2
i 1
(1 e )
i 1
Var0 [ yi ( x )]
i
n
log
1
npi (1 pi ) .
j 2
i 1
k 1
p ( x | ) h( x ) exp{ j ( )T j ( x ) B( )}
j 1
13
( xi ) 2
1
exp{
}
2
2
2
p( x | )
i 1
1
1
exp
2
2
2
i
i 1
i 1 i
2
n 2
2
2
1
1
n
1 n
n
exp
2 i 1 i
i 1 i
2
2
2
14