Statistical Models
(Chapter 6.) A statistical model P is a collection of probability distributions (or a collection of densities). An example of a nonparametric model is
(
)
Z
P=
p:
p(x; ) :
2 /2
}.
Statistics
MX = E[etx ] = E[e
Xi t/n
]=
E[eXi (t/n) ]
= [MX (t/n)] =
1
1 t/n
n
1
=
1 /nt
n
.
1
(1 + x2 )
j=k
n
X
j=k
n
[FX (x)]j [1 FX (x)]nj
j
n!
[FX (x)]k1 p(x) [1 FX (x)]nk .
(k 1)!(n k)!
2
Sufficiency
3.1
Sufficient Statistics
Pn
i=1
Xi . Then,
P (X n = xn and T = t)
.
P (T = t)
But
n
P (X = x and T = t) =
0
T (x1 , . . . , xn ) 6= t
P (X1 = x1 , . . . , Xn = xn ) T (x1 , . . . , xn ) = t.
Hence,
n
P (X = x ) =
n
Y
e xi
i=1
Now, T (xn ) =
xi !
en xi
en t
Q
Q
=
=
.
(xi !)
(xi !)
xi = t and so
P (T = t) =
en (n)t
t!
since T Poisson(n).
Thus,
t!
P (X n = xn )
= Q
P (T = t)
( xi )!nt
P
which does not dependP
on . So T =
P i Xi is a sufficient statistic for . Other sufficient
statistics are: T = 3.7 i Xi , T = ( i Xi , X4 ), and T (X1 , . . . , Xn ) = (X1 , . . . , Xn ).
3.2
Sufficient Partitions
0,
0,
1,
0,
1,
0,
1,
1,
0)
1)
0)
0)
1)
1)
0)
1)
8 elements
t
t=0
t=1
t=1
t=1
t=2
t=2
t=2
t=3
Xi .
p(x|t)
1
1/3
1/3
1/3
1/3
1/3
1/3
1
4 elements
t
p(x|t)
t=0
(1 )2
t=0
(1 )
t=0
(1 )
t=0
2
t=1
(1 )2
t=1
(1 )
t=1
(1 )
t=1
2
2 elements
4
3.3
P
en Xi
1
=Q
en i Xi .
p(xn ; ) = Q
(xi !)
(xi !)
Example 11 X1 , , Xn N (, 2 ). Then
n
p(x ; , ) =
1
2 2
n2
P
(xi x)2 + n(x )2
exp
.
2 2
(a) If known:
n
p(x ; ) =
1
2 2
n2
P
(xi x)2
exp
2 2
{z
}
exp
|
h(xn )
n(x )2
.
2 2
{z
}
3.4
T (X1 , , Xn ) =
Xi2 .
i=m+1
0,
0,
1,
0,
1,
0,
1,
1,
0)
1)
0)
0)
1)
1)
0)
1)
t
t=0
t=1
t=1
t=1
t=2
t=2
t=2
t=3
p(x|t)
1
1/3
1/3
1/3
1/3
1/3
1/3
1
Xi .
u
u=0
u=1
u=1
u=1
u = 73
u = 73
u = 91
u = 103
p(x|u)
1
1/3
1/3
1/3
1/2
1/2
1
1
3.5
Theorem 15 Define
p(y n ; )
R(x , y ; ) =
.
p(xn ; )
Suppose that T has the following property:
n
en yi p(y n ; )
yi xi
Q
Q
p(y n ; ) = Q
=
,
p(xn ; )
yi
yi !/ xi !
P
P
P
which is independent of iff
yi =
xi . This implies that T (Y n ) =
Yi is a minimal
sufficient statistic for .
The minimal sufficient statistic is not unique. But, the minimal sufficient partition is unique.
Example 17 Cauchy.
p(x; ) =
Then
1
.
(1 + (x )2 )
n
Q
{1 + (xi )2 }
p(y ; )
= i=1
.
n
Q
p(xn ; )
2
{1 + (yj ) }
n
j=1
If T is sufficient, then T contains all the information you need from the data to compute the
likelihood function. It does not contain all the information in the data. We will define
the likelihood function in the next set of notes.