Math152F09Assign13Solutions PDF

Math 152.
Rumbos
Fall 2009
Solutions to Assignment #13

1. Consider a test of the simple hypotheses
H : =
versus H1 : = 1
based on a random sample from a distribution with pmf ( ), for =

1, 2, . . . , 7. The values of the likelihood function at and 1 are given in the
table below.
1
( ) 0.01
(1 ) 0.06
2
0.01
0.05
3
0.01
0.04
4
0.01
0.03
5
0.01
0.02
6
0.01
0.01
7
0.94
0.79
Use the NeymanPearson Lemma to nd the most powerful test for H versus
H1 with signicance level = 0.04. Compute the probability of Type II error
for this test.
Solution: Table 1 shows the values of the likelihood ratio statistic
in the third row. Observe that if we let = 1/3 and the region
dened by , then
= P( 1/3 = ) = 0.04.
Thus, by the NeymanPearson Lemma, the test that rejects H if
1
( )
0.01
0.06
(1 )
( )/(1 ) 1/6
2
0.01
0.05
1/5
3
0.01
0.04
1/4
4
0.01
0.03
1/3
5
0.01
0.02
1/2
6
7
0.01 0.94
0.01 0.79
1
1.19
Table 1: Likelihood ratios

( )
1
(1 )
3
is the most powerful test at signicance level = 0.04.
The power of the test is
(1 ) = P( 1/3 = 1 ) = 0.18.
Thus, the probability of a Type II error is 1 (1 ) = 82%.
Math 152. Rumbos
Fall 2009
2. Let 1 , 2 , . . . , be a random sample from a Poisson() distribution.

(a) Find the most powerful test for testing
H : =
versus H1 : = 1 ,
for 1 > .
Solution: According to the NeymanPearson Lemma, the most
powerful test is the LRT. To nd the LRT rejection region, we
rst compute the likelihood ratio statistic
(1 , 2 , . . . , ) =
where
for =
( 1 , 2 , . . . , )
,
(1 1 , 2 , . . . , )
( 1 , 2 , . . . , ) =
,
1 !2 ! !
(1)
=1
We then have that

1
(1 , 2 , . . . , ) = 1 =
1
= ,
where we have set = 1 / and = /1 .

Since, 1 > , > 1 and < 1.
The rejection region of the LRT is
:
(1 , 2 , . . . , ) ,
for some (0, 1) determined by the signicance level of the test,

or
,
(2)
where =
.
=1
Taking the natural logarithm on both sides of the inequality in

(2), we obtain
ln + ln ln ,
from which we get that
ln ln
> 0.
ln
Math 152. Rumbos
Fall 2009
Thus, the LRT rejects H if

,
for some > 0, where =
(3)
=1
(b) Show that the test found in part (a) is uniformly most powerful for testing
H : =
versus H1 : > .
Solution: Fix in (3) so that P( > ) = , where

Poisson( ). Note that this value of depends only on and
. Furthermore, by the result of part (a), the test that reject
H : = versus H : = 1 , if
,
is the most powerful test at level for every 1 > . It then
follows that the test that rejects H if
,
is the uniformly most powerful test of H versus H1 : > .
3. Given a random sample, 1 , 2 , . . . , , from a distribution with distribution
function ( ). We say that a statistic = (1 , 2 , . . . , ) is sucient
for is the joint distribution (1 , 2 , . . . , ) can be written in the form
(1 , 2 , . . . , ) = (, )(1 , 2 , . . . , ),
for some functions : 2 and : .
Let 1 , 2 , . . . , be a random sample from a Poisson() distribution. Find
a sucient statistic for . Justify your answer based on the denition given
above.
Solution: According to (1) int he solution to part (a) of Problem 2
in this assignment, the likelihood function in this case is
( 1 , 2 , . . . , ) = (, )(1 , 2 , . . . , ),
where
(, ) = ,
Math 152. Rumbos
Fall 2009
(1 , 2 , . . . , ) =
and
=
1
,
1 !2 ! !
.
=1
It then follows that =
is a sucient statistic for . Observe
=1
that =
1
is also a sucient statistic for .
4. Suppose that 1 , 2 , . . . , forms a random sample from distribution with

distribution function ( ).
(a) Show that if is a sucient statistic for , then the likelihood ratio statistic
for the test of
H : = versus H1 : = 1
is a function of .
Solution: In this case, the likelihood function is
( 1 , 2 , . . . , ) = (, )(1 , 2 , . . . , )
for all possible values of the parameter . It then follows that the
likelihood ratio statistic is
(1 , 2 , . . . , ) =
(, )
,
(, 1 )
which is a function of .
(b) Explain how knowledge of the distribution of under H may be used to

choose a rejection region that yields the most powerful test at level .
Solution: Knowing the distribution of , assuming that the null
hypothesis is true, it is possible to nd a value, , for , such that
(
)
(, )
P
= .
(, 1 )
The LRT rejection region is then given by
:
(, ) (, 1 );
that is, if the value of given by the sample falls in the region ,
the null hypothesis is rejected.
Math 152. Rumbos
Fall 2009
5. Derive a likelihood ratio test for

versus H1 : 2 = 2
H : 2 = 2
based on a sample from a normal(, 2 ) distribution.

Solution: The likelihood ratio statistic is
sup (, 1 , 2 , . . . , )
(1 , 2 , . . . , ) =
(
,
1 , 2 , . . . , )
(4)
where
and
2 are the MLEs for and 2 , respectively; That is,
1
,
==
=1
and
=
( )2 .
=1
2
When we maximize (, 1 , 2 , . . . , ) over we obtain

that = . It then follows from (4) that
(1 , 2 , . . . , ) =
(
, 1 , 2 , . . . , )
,
(
,
1 , 2 , . . . , )
(5)
where
(, 1 , 2 , . . . , ) =
1
(2)/2
=1 ( )
2 /2 2
for and > 0. We then have from (5) that

( ) ( )2 /22
=1
(1 , 2 , . . . , ) =
2
2
=1 ( ) /2
(
=
= /2 /2 /2 ,
where we have set =
2
.
2
/2
/2
Math 152. Rumbos
Fall 2009
We then have that

(1 , 2 , . . . , ) = ( )
where is the statistic
1
( )2 ,
=
2 =1
(6)
0.0
0.2
0.4
g(t)
0.6
0.8
1.0
and () = /2 /2 /2 . Observe that has a 2 ( 1) distribution when the null hypothesis is true. Observe also that () has a
graph like the one sketched in Figure 1. It then follows that for any
Figure 1: Sketch of graph of () for = 10 and 0 4

(0, 1), there exists 1 and 2 such that 0 < 1 < 1 < 2 and
(1 ) = (1 ) = .
Furthermore,
() for 1 or 2 .
Math 152. Rumbos
Fall 2009
We then have that the LRT rejection region,

:
(, 2 , . . . , ) ,
can be expressed in terms of the statistic in (6) as

:
1 or 2 .
The LRT rejection region can also be expressed in terms of the sample
variance, 2 , as follows
2
1 2
1 2
1 or 2
2 ,
for 0 < 1 < 1 < 2 , or, equivalently,

:
2 2 1 or 2 2 + 2 ,
for some positive values of 1 and 2 determined by the signicance

level of the test.

Math152F09Assign13Solutions PDF

Uploaded by

Copyright:

Available Formats

Math152F09Assign13Solutions PDF

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Math152F09Assign13Solutions PDF

Uploaded by

Copyright:

Available Formats

Math 152.

Solutions to Assignment #13

based on a random sample from a distribution with pmf ( ), for =

Table 1: Likelihood ratios

Math 152. Rumbos

2. Let 1 , 2 , . . . , be a random sample from a Poisson() distribution.

We then have that

where we have set = 1 / and = /1 .

for some (0, 1) determined by the signicance level of the test,

Taking the natural logarithm on both sides of the inequality in

Math 152. Rumbos

Thus, the LRT rejects H if

Solution: Fix in (3) so that P( > ) = , where

Math 152. Rumbos

It then follows that =

is a sucient statistic for . Observe

4. Suppose that 1 , 2 , . . . , forms a random sample from distribution with

(b) Explain how knowledge of the distribution of under H may be used to

Math 152. Rumbos

5. Derive a likelihood ratio test for

based on a sample from a normal(, 2 ) distribution.

When we maximize (, 1 , 2 , . . . , ) over we obtain

for and > 0. We then have from (5) that

Math 152. Rumbos

We then have that

Figure 1: Sketch of graph of () for = 10 and 0 4

Math 152. Rumbos

We then have that the LRT rejection region,

can be expressed in terms of the statistic in (6) as

for 0 < 1 < 1 < 2 , or, equivalently,

for some positive values of 1 and 2 determined by the signicance

You might also like