8.1.

8 Then likelihood function is o given by L x1 ; : : : ; xn j 2 =

2n 1
P n 2
exp 2 2 i=1 (xi 0) : By factorization (Theorem 6.1.1),
Pn 2 Pn 2
i=1 (xi 0 ) is su¢ cient. Further, E 2 i=1 (xi 0) = n 2 ; so
P n 2
n 1 i=1 (xi 0 ) is unbiased for
2
: Since this su¢ cient statistic is complete,
1
P n 2 2
we have that n i=1 (xi 0 ) is UMVU for :
n
8.1.12 The likelihood function is given by L (x1 ; : : : ; xn j ) = whenever
> x(n) and 0 otherwise. Therefore, when we know x(n) we know the likelihood
function and so x(n) is su¢ cient. Then x(n) has density given by n nxn 1 for
R n
0 < x < and E x(n) = 0 n
nxn dx = n+1n xn+1 0 = n+1 n
: So n+1
n x(n) is
UMVU for :
8.1.16 We have that
!
2
2 2 2 2 4 c
MSE( ; 2) cs = E( ; 2) cs = E( ; 2) X 1
(n 1)
( )
2
4 c 2 2c
= E( ; 2) X E( ; 2) (X) + 1
n 1 n 1

where X = (n 1) s2 = 2 2
(n 1). So E( ; 2 ) (X) = n 1 and Var( ; 2 ) (X) =
2
2(n 1); which implies E( ; 2 ) X 2 = 2(n 1) + (n 1) : Di¤erentiating the
above expression with respect to c; and setting the derivative equal to 0, gives
that the optimal value satis…es
c 1
2 E( ;
2) X2 E( ; 2) (X) = 0
(n 1) n 1
or
E( ; 2 ) (X) (n 1) n 1
c = (n 1) = (n 1) 2 = :
E( ; 2 ) (X 2 ) 2(n 1) + (n 1) n+1
We have that the bias equals

n 1 2 2
E( ; 2) cs2 2
= (c 1) 2
= 1 2
= 2
:
n+1 n+1

8.1.22 The log-likelihood function is given by l ( j x1 ; : : : ; xn ) = n 0 ln nx,

so S ( j x1 ; : : : ; xn ) = n 0 = nx; S 0 ( j x1 ; : : : ; xn ) = n 0 = 2 ; which implies
I ( ) = n 0 = 2 : Since ( ) = 1
; 0( )= 2
the information lower bound
4 2
for unbiased estimators is given by 1= =n 0 = 1=n 0 2 : Note that
1
by Exercise 8.1.7 x= 0 is UMVU for and this has variance 0 =n 02 2 =
2
1=n 0 ; which is the Cramer-Rao lower bound.
8.2.8 What we care in optimal hypothesis testing theory is type I and II errors,
i.e., signi…cance level and power function. Hence, we must ignore the di¤erence

1
of two test procedures whenever two tests have the same signi…cance level and
the same power function.
8.2.13 We have that E (') = for every ; so it is of exact size : For this
test, no matter what data is obtained, we randomly decide to reject H0 with
probability :
8.2.14 Suppose that '0 is a size UMP test for a speci…c problem and let '
be the test function of Problem 8.2.13. Then for such that the alternative is
true we have that E ('0 ) E (') = , so '0 is unbiased.
8.2.20 The UMP size test for H0 : = 0 versus H0 : = 1 is of the form
nx
L( 1 j x1 ; : : : ; xn ) ( 1) e 1
= nx > c0
L( 0 j x1 ; : : : ; xn ) ( 0) e 0

or, equivalently, whenever nx (ln 1 ln 0 ) > ( 1 0 ) ln c0 ; and since 1 >

0 ; this is equivalent to rejecting whenever nx > ( 1 0 ) ln c0 = (ln 1 ln 0 ) :
Now recall that nx Poisson(n 0 ) under H0 so we must determine the smallest
k such that P 0 (nx > k) and then put = ( P 0 (nx > k)) =P 0 (nx = k) :
Since this test does not involve 1 ; it is UMP size for H0 : = R0 versus H0 :
1 1 x
> 0 . From Problem 8.2.19 we have that P (nx > k) 1 x! y e y dy;
and we see that this is increasing in : Therefore, this test is UMP size for
H0 : 0 versus H0 : > 0.
8.2.21 When H0 : = 0 holds, the log-likelihood Pn and score functions are given
2
by l x1 ; : : : ; xn j 2 = n ln 2 2 1 2 i=1 (x 0 ) ; S x1 ; : : : ; xn j
2
=
n 1
P n 2 2
2 + 2 4 i=1 (x 0 ) : Then S x1 ; : : : ; xn j = 0 leads to the MLE
2 1
Pn 2
^H0 = 0 ; ^H = n i=1 (x 0 ) ; and the maximized log-likelihood equals
0
2
Pn 2 n
l x1 ; : : : ; xn j ^H0 = n ln n n ln i=1 (x 0) 2:
When Ha : 6= 0 holds, the log-likelihood Pn function is given by
2
l x1 ; : : : ; xn j ; 2 = n ln 2 2 1 2 i=1 (x ) :
Pn 2
By Example 6.2.6 the MLE is given by ^ = x; ^ 2 = n1 i=1 (x x) and the
2
maximized log-likelihood is given by l x1 ; : : : ; xn j ^; ^ =
Pn 2 n
n ln n n ln i=1 (x x) 2 : Then the likelihood ratio test rejects whenever

2 l x1 ; : : : ; xn j ^; ^ 2 2
l x1 ; : : : ; xn j ^H 0
n n
! Pn
X 2
X 2 (x 0)
2
=2 n ln (x x) + n ln (x 0) = 2n ln Pi=1
n 2 > x1
i=1 i=1 i=1 (x x)
2 2
where x1 is the (1 )th quantile of the (2 1) = (1) distribution.