Dr. Ceylan Yozgatlıgil 02.04.

2007
METU-SPRING 2008
STAT 272
MIDTERM EXAM I

SOLUTIONS

1. Let be a random sample of random variables having p.d.f.

where a is known and is unknown.

a) (5 pts.) Is this p.d.f. belongs to a family of location, scale or location-scale type? Why? Is the
unknown parameter location or scale parameter? Explain.

Scale family of distributions. .

Let .

not depend on .

is a scale parameter.
b) (5 pts.) Find a Pivotal Quantity for using a sufficient statistic of which is
.

Since is scale parameter and X(n) is a s.s. for , a pivotal quantity is .

c) (10 pts.) Construct a 100 % confidence interval for with equal-tail probabilities using the

Pivotal Quantity that you found in part b)? (Hint: ).

1
100 % confidence interval for : .
d) (10 pts.) Using the distribution function technique, find a pivotal quantity for and obtain a upper
90% confidence limit for .

Since is a strictly increasing function of , we can take the exponent of the limits.

Hence, 90% upper confidence limit for is .

2
2. Suppose and are independent and identically distributed from the p.d.f.

a) (5 pts.) Show that is a location parameter.

Let .

Hence, is a location parameter.

b) (10 pts.) Show that and , which is a complete and sufficient statistic

for , are independent

Let . Then,

Since the distribution of Z is independent of , it is an ancillary statistic and since Y is a c.s.s. for ,
by Basu Therorem, they are independent.

3. Let be a random sample of random variables having p.d.f.

a) (15 pts.) Using the general method, construct a 100 % confidence interval for . (Hint:
).

3
100 % confidence interval for :

b) (5 pts.) Find an approximate 100 % confidence interval for using the central limit
theorem. (Hint: ).

By CLT,

An approximate 100 % for :

4
4. Consider a random sample and . It is known that is a sufficient

statistic for and . Suppose that we want to test the hypothesis

a) (10 pts.) Define the rejection region of the test in terms of and find the critical value, c for
a given significance level . (You can use the p.d.f. of Y, to be able to define c).

Reject H0 if

Reject H0 if .

b) (5 pts.) Determine the power function, .

c) (5 pts.) Find the p-value for . For what significance level, is the null hypothesis
rejected?

p-value=

If p-value= < , reject H0.

d) (10 pts.) Find a 100 % one-sided confidence limit for , using the rejection region of
the above test. How can we decide to reject or accept the null hypothesis by using this
confidence limit? Discuss.

5
 Hence, 100 % upper confidence limit for is .

If is in the confidence interval, then we reject the null hypothesis with significance level
of .
e) (5 pts.) Suppose that 95% confidence interval for is given by . What would
be your decision about the null hypothesis? Why?

We cannot use this confidence interval to decide on the above hypothesis because above test is
one-sided test and this interval is two sided. The acceptance regions will be different and we can
come up with wrong decisions. Also, this confidence coefficient of the interval is 95%. Even if it
is an upper one-sided confidence limit, the significance level of the test and confidence coefficient
of the interval should use the same value to be able to get consistent results.

STAT 272 - MIDTERM EXAM II

SOLUTIONS

1. Let X be one observation from the density

.
For testing
vs.
We used the test that rejects if and only if .
a) (10 pts.) Find the size of this test.

b) (10 pts.) Find the power function of this test.

6
2. Let X = (X1, ...,Xn) be a random sample from the density

where θ  Θ = (0,∞).
a) (15 pts.) Find the Monotone Likelihood Ratio Statistics for

L depends on only through and L is an increasing function of .

Therefore, MLRS= .

b) (20 pts.) Find a Uniformly Most Powerful test of size α for testing
H0 : θ  θ0 versus H1 : θ < θ0.
Let θ0 be a positive constant.

Reject H0 if <c ~ Reject H0 if ln <ln c.  Reject H0 if .

.
We need to find the probability distribution of Y.

Consider Z= - ln X.  X=e-Z .

Therefore, . .

Uniformly Most Powerful test of size α: Reject H0 if

c) (10 pts.) Construct a 100(1-)% Uniformly Most Accurate Confidence Interval for .

Rejection region:

Acceptance region:

7
 So, Replace by .

Hence, the UMA 100 CI for :

3. (35 pts.) Let be i.i.d. as where  and 2 are unknown. Derive the
Generalized Likelihood Ratio Test of size for testing
vs.
based on . Write down the power function of this test.

MLE of and MLE of .

8
Reject H0 if  Reject H0 if .

9
Hence, reject H0 is .

It is known that . So,

Hence,

GLRT of size : Reject H0 if .

10
Dr. Ceylan Yozgatlıgil 09.06.2007
METU-SPRING 2008
STAT 272
FINAL EXAM

QUESTIONS

1. (15 pts.) Let be a random sample from a Beta( ,1) with pdf
,
and zero elsewhere. For testing
vs. ,
find the most powerful test of size .

Reject H0 if .

So, reject H0 if . Now, we need to find the distribution of . To be able to find

this, lets look at the distribution on .

This means that . Thus, and .

The MPT of size : Reject H0 if or .

2. (15 pts.) Consider a random sample of size n from a distribution with pdf

and zero elsewhere. Derive a uniformly most powerful test of size of

vs. .

11
L depends on sample observations only through and L is an increasing

function of . Thus, MLRS= .

Reject H0 if >c. We need to find the distribution of .

Let .

That is .

The UMPT of size : Reject H0 if .

3. Let be a random sample from a Uniform( , ) with pdf

,
and zero elsewhere.
a) (15 pts.) Find the generalized likelihood ratio test of size for testing
vs. .
and the MLE of is .

and

12
Hence, the generalized likelihood ratio is

Reject H0 if where . Hence,

Reject H0 if and .

and .

Thus, and ,

The GLRT of size : Reject H0 if and . Since , .

Thus, . This means that, the GLRT of size : Reject H0 if .
(10 pts.) Find the power of the test.

4. (15 pts.) Let where is unknown. Consider the test vs.

where . For a random sample of size n, define the rejection, acceptance

and continuation regions in terms of for a sequential probability ratio test with

, approximately.

and .

The simple likelihood ratio:

Reject H0 as soon as . Hence,

13
Accept H0 as soon as . Hence,

Continue if .

5. Let be a random sample from an exponential distribution with mean ,

, and assume that the prior density of also is exponential with mean
where is known.

a) (15 pts.) Find the posterior distribution of .

b) (5 pts.) Find the Bayesian point estimate of for a squared error loss function.

The Bayesian point estimate of .

6. (10 pts.) In the class, we did an experiment with Bonibon’s. Explain the aim of that
experiment.

14
Dr. Ceylan Yozgatlıgil 9.4.2009
METU-SPRING 2009
STAT 272
MIDTERM EXAM I

SOLUTION

1. Let be i.i.d. r.v.s from N(,2) distribution.

a) (5 pts.) Are  and  location or scale parameters? Explain.

Hence,  and  are location and scale parameters, respectively.

b) (15 pts.) Show that and , which is a complete and sufficient statistic for , are

independent ( : a r.v., : the sample mean, : the sample standard deviation).

and .

So, .

Hence,

So, is an ancillary statistics. By Basu theorem, T and css are independent.

15
2. Let be a random sample of random variables having p.d.f.

a) (5 pts.) Is this p.d.f. belongs to a family of location, scale or location-scale type? Why? Is the
unknown parameter location or scale parameter? Explain.

Scale family of distributions. .

Let .

not depend on .

is a scale parameter.

b) (5 pts.) Find a Pivotal Quantity for using a sufficient statistic for .

 is a ss for .

Hence, . But we know that which is also a pivotal quantity. Since it has a

statistical table, we prefer to use as a P.Q.

c) (10 pts.) Construct a 100 % confidence interval for with equal-tail probabilities using the
Pivotal Quantity that you found in part b.

A 100(1-)% CI for :

16
3. Let be a random sample having pdf

a) (15 pts.) Using the general method and an MLE of , construct a 100 % confidence interval
for .
MLE of  =

Hence, and

100(1-)% CI for :

b) (10 pts.) Using the distribution function technique, find a pivotal quantity for and obtain a lower
95% confidence limit for .

This P.Q is a decreasing function of .

17
95% LCL for :

4. (20 pts.) Let be a random sample of random variables having p.d.f.

Find an approximate 100 % confidence interval for by using the central limit theorem.
(Hint: If ).

CLT:

100(1-)% approximate CI for :

18
5. Let be a random sample having pdf N(, 2=).

a) (10 pts.) Suggest two methods to construct 100(1−)% (approximate) confidence intervals for
. Explain how you would get a confidence interval for  using your methods. If you propose
any transformation, explain the type of the transformation (e.g. log transformation or square
root transformation and so on). Do not attempt to find the confidence intervals.

As an estimate of , let’s consider which is the MVUE of . Then,

We cannot obtain a proper confidence interval for  using the above random interval.

Method 1: We need to get rid of term in the denominator. That is, we need to use a transformation
so that the variance of the function of is independent of . We can use the delta method.
Find a function of , say . Then, the approximate variance of will be . This

means that . for some constants a c

and a. Hence, using transformation we can get rid of the  term in the variance. Then, we can

obtain a CI for first , then for . Because, .

Method 2: We can use the limiting distribution of the MLE of , which is .

Since the variance part is also a function of , we need to apply delta method again to be able to
obtain a proper interval for .

b) (5 pts.) If the resulting confidence intervals for  are different than each other, which one do
you prefer as an interval estimate for ? Why?

We will prefer an interval with the smallest expected length. Because it will give us an interval which
is closer to the unknown parameter for the same confidence coefficient.

Dr. Ceylan Yozgatlıgil 9.4.2009

METU-SPRING 2009
STAT 272
MIDTERM EXAM I

19
 SOLUTION

2. Let be i.i.d. r.v.s from N(,2) distribution.

a) (5 pts.) Are  and  location or scale parameters? Explain.

Hence,  and  are location and scale parameters, respectively.

b) (15 pts.) Show that and , which is a complete and sufficient statistic for , are

independent ( : a r.v., : the sample mean, : the sample standard deviation).

and .

So, .

Hence,

So, is an ancillary statistics. By Basu theorem, T and css are independent.

20
2. Let be a random sample of random variables having p.d.f.

a) (5 pts.) Is this p.d.f. belongs to a family of location, scale or location-scale type? Why? Is the
unknown parameter location or scale parameter? Explain.

Scale family of distributions. .

Let .

not depend on .

is a scale parameter.

b) (5 pts.) Find a Pivotal Quantity for using a sufficient statistic for .

 is a ss for .

Hence, . But we know that which is also a pivotal quantity. Since it has a

statistical table, we prefer to use as a P.Q.

c) (10 pts.) Construct a 100 % confidence interval for with equal-tail probabilities using the
Pivotal Quantity that you found in part b.

A 100(1-)% CI for :

21
3. Let be a random sample having pdf

a) (15 pts.) Using the general method and an MLE of , construct a 100 % confidence interval
for .
MLE of  =

Hence, and

100(1-)% CI for :

b) (10 pts.) Using the distribution function technique, find a pivotal quantity for and obtain a lower
95% confidence limit for .

This P.Q is a decreasing function of .

22
95% LCL for :

4. (20 pts.) Let be a random sample of random variables having p.d.f.

Find an approximate 100 % confidence interval for by using the central limit theorem.
(Hint: If ).

CLT:

100(1-)% approximate CI for :

23
5. Let be a random sample having pdf N(, 2=).

c) (10 pts.) Suggest two methods to construct 100(1−)% (approximate) confidence intervals for
. Explain how you would get a confidence interval for  using your methods. If you propose
any transformation, explain the type of the transformation (e.g. log transformation or square
root transformation and so on). Do not attempt to find the confidence intervals.

As an estimate of , let’s consider which is the MVUE of . Then,

We cannot obtain a proper confidence interval for  using the above random interval.

Method 1: We need to get rid of term in the denominator. That is, we need to use a transformation
so that the variance of the function of is independent of . We can use the delta method.
Find a function of , say . Then, the approximate variance of will be . This

means that . for some constants a c

and a. Hence, using transformation we can get rid of the  term in the variance. Then, we can

obtain a CI for first , then for . Because, .

Method 2: We can use the limiting distribution of the MLE of , which is .

Since the variance part is also a function of , we need to apply delta method again to be able to
obtain a proper interval for .

d) (5 pts.) If the resulting confidence intervals for  are different than each other, which one do
you prefer as an interval estimate for ? Why?

We will prefer an interval with the smallest expected length. Because it will give us an interval which
is closer to the unknown parameter for the same confidence coefficient.

Dr. Ceylan Yozgatlıgil 12.5.2009

METU-SPRING 2009
STAT 272
MIDTERM EXAM II
(Time Duration: 90 minutes)

24
 QUESTIONS

1. (25 pts.) Let be a random sample from a Poisson( ) with pdf

To test vs , derive the most powerful test of size  using the randomized test.

2. Consider a random sample of size n with pdf

for .
b) (25 pts.) Construct the uniformly most powerful test of size for testing
vs. .
c) (5 pts.) Give the p-value formula for this test.
d) (5 pts.) Obtain the uniformly most accurate confidence interval for .

3. Let be i.i.d. r.v.s with pdf

where a is known.
a) (35 pts.) Derive the generalized likelihood ratio test of against
based on a random sample of size n.
b) (5 pts.) Find the power function of the test.

Dr. Ceylan Yozgatlıgil 8.6.2009

METU-SPRING 2009
STAT 272
MIDTERM EXAM II
(Time Duration: 120 minutes)

QUESTIONS

1. If X have the pdf

,
consider testing

Using a single observation x,

f) (5 pts.) Define the rejection region of the test and find the critical value, c such that
=0.008.
g) (5 pts.) Determine the power function, .
h) (5 pts.) Find the p-value for . For what significance level, is the null
hypothesis rejected?
i) (5 pts.) Find a 100 % one-sided confidence limit for , using the rejection
region of the above test. How can we decide to reject or accept the null hypothesis by
using this confidence limit? Discuss.

25
2. Consider a random sample of size n from a distribution with pdf
,  Exp(1/)
and zero elsewhere.
a) (15 pts.) Derive a uniformly most powerful test of size of
vs. .

b) (10 pts.) For a given value of write the p-value formula.

3. (20 pts.) Let be a random sample from a Beta( ,1) with pdf
,
and zero elsewhere. Consider the test vs. where . Define the
rejection, acceptance and continuation regions for a sequential probability ratio test with
≈0.01 and ≈0.02 approximately.

4. Let be a random sample from a normal distribution with mean  and

variance 1, and assume that the prior density of also is normal with mean  and
variance 1 where  is known.
c) (10 pts.) Find the posterior distribution of .
d) (5 pts.) Find the Bayesian point estimate of for a squared error loss function.
e) (5 pts.) Obtain 95% Bayesian confidence interval for .

5. (15 pts.) Explain the differences between classical approach and Bayesian approach in
statistical inference.
BONUS QUESTION: (20 pts.)
 Explain why we performed a Topitop experiment.
 Explain which steps that we followed in Topitop experiment.
 What was our hypothesis at the beginning of the experiment?
 What is your conclusion on the hypothesis at the end of the experiment?

Dr. Ceylan Yozgatlıgil 6.5.2010

METU-SPRING 2009
STAT 272
MIDTERM EXAM II
(Time Duration: 90 minutess)

QUESTIONS
1. Let where  is unknown.
a. (5 pts.) Is the unknown parameters  location and scale parameter? Explain.
b. (10 pts.) Is , where ,
an ancillary statistics? Explain.
c. (10 pts.) Find a Pivotal Quantity for  using the MLE of .
d. (15 pts.) Construct a 100 % confidence interval for with equal-tail
probabilities using the Pivotal Quantity that you found in part c.

2. Let X = (X1, ...,Xn) be a random sample from the density

26
 where θ  Θ = (0,∞).
a. (10 pts.) Construct a 100 % confidence interval for with equal-tail
probabilities using the distribution function method.
b. (10 pts.) Construct a 100 % confidence interval for with equal-tail
probabilities using the general method.
c. (10 pts.) Let L be the length of the confidence interval that you obtained in part b.
Find the expected value of L.
3. Let X have the p.d.f.

Let be a random sample of size n.

a. (15 pts.) Construct a lower 90% confidence limit for using an appropriate
Pivotal Quantity with a sufficient statistics for .
b. (15 pts.) Construct a 100(1)% approximate confidence interval for  using the
central limit theorem.

Dr. Ceylan Yozgatlıgil 6.5.2010

METU-SPRING 2009
STAT 272
MIDTERM EXAM II
ANSWERS
1. Suppose that we have a random sample of size n from the density function

a) (15 pts.) Derive the most powerful test of size  for testing against

Reject H0 is L k ln L ln k

27
 The most powerful test of size  for testing against is :

Reject H0 if

b) (5 pts.) Find the Type II error for the test given in a).

c) (5 pts.) Write the p-value formula for a given value of

2. Let be a r.s. from Normal where  is known.

a) (20 pts.) Derive a uniformly most powerful test of size  for testing the following
hypothesis

28
 2 2
1 n

L( ) depends on (x ,...,x ) only through and L( ) is a nondecreasing

function of . Hence, MLRS= .

0

UMPT: Reject H if MLRS>c.

The uniformly most powerful test of size : Reject H0 if .

b) (10 pts.) Construct a 100(1)% uniformly most accurate confidence interval for
.

Rejection region:

Acceptance region

A 100(1)% uniformly most accurate confidence interval for :

3. A random sample is drawn from Pareto population with pdf

where  is known.
c) (25 pts.) Derive the generalized likelihood ratio test of size  for testing
against based on a random sample of size n.

29
 0 0

GLRT: Reject H if < .

GLRT of size : Reject H if .

d) (5 pts.) Find the power function of the test.

4. Give brief answers to the following questions.

a) (5 pts.) Explain why we cannot minimize both Type I and Type II errors at the same
time.

b) (5 pts.) Explain why we don’t say “accept H0” when we cannot reject the null
hypothesis.
c) (5 pts.) What is the p-value? Explain.

Dr. Ceylan Yozgatlıgil 1.6.2010

METU-SPRING 2010
STAT 272
FINAL EXAM

SOLUTIONS

1. Consider a random sample from a distribution with

.
a) (10 pts.) Derive a uniformly most powerful test of size  for testing

30
 L depends on only through and L is a decreasing function of

. MLRS= . Since H1: >1, the rejection region will be MLRS>c.

Reject H0 if

To be able to find the value k, we have to find the distribution of or

Let Then, .
Hence,

Then,

UMPT of size : Reject H0 if

b) (5 pts.) Find a 100(1)% uniformly most accurate confidence interval for .

Rejection region is

Acceptanceregion is

Then,

A 100(1)% uniformly most accurate confidence interval for :

31
 .

2. Consider a random sample from a distribution with

where  is known.
a) (15 pts.) Derive a generalized likelihood ratio test of size  for testing

Since  cannot be equal to 1, we need to ignore the first one. So,

Now, we need to find MLE of .

Hence,

GLRT: Reject H0 if <0. Reject H0 if ln<ln0.

32
 Let . Then,

z
k1 n k2 y

Rejection region

Hence, Reject H0 if

So, the rejection criteria will be reject H0 if .

33
 Let T=XX=T+dx/dt=1.

GLRT of size : Reject H0 if

b) (10 pts.) Find the power function for this test.

c) (10 pts.) Find the asymptotic test results for the test given in part a) using the
asymptotic distribution of

Reject H0 if .

Reject H0 if

34
 Reject H0 if

3. (15 pts.) Let be a random sample from a distribution with pdf

If , find the Sequential Probability Ratio Test of

Reject H0 if

Reject H0 if

Reject H0 if

Reject H0 if

Accept H0 if

Continue if

4. (20 pts.) Let X(n) be the nth order statistic of a random sample of size n from a
distribution with pdf
.
Let  be an observed value of the random variable which has pdf
.
Using the squared error loss function, find the Bayesian point estimate of .
Note: cdf of X(n):

35
 Using the squared error loss function, the Bayesian point estimate of :

5. (15 pts.) Let be a random sample from a Normal distribution with

mean  and variance  . Consider a test
2

Draw pdf(s) of normal distribution and on this (or these) plot(s) show the region of the
power for =20. (Do not attempt to derive a test.)

(=20)

=10 =20

Dr. Ceylan Yozgatlıgil

11.4.2011
METU-SPRING 2011
STAT 272
MIDTERM EXAM I

36
 QUESTIONS

1. Let X = (X1, ...,Xn) be a random sample from the density

a. (5 pts.) Is the unknown parameters  location or scale parameter? Explain.

b. (10 pts.) Show that , where ,

is an ancillary statistics?

c. (10 pts.) Show that by assuming that the sufficient statistic you

found is also complete.

d. (5 pts.) Find a Pivotal Quantity for  using the MLE of .
e. (10 pts.) Construct an upper 95% confidence limit for using the Pivotal
Quantity that you found in part d.

2. Let X = (X1, ...,Xn) be a random sample from the density

where θ  Θ = (0,∞).

a. (10 pts.) Construct a 90% lower confidence interval for using the distribution
function method.
b. (10 pts.) Construct a 100 % confidence interval for with equal-tail
probabilities using the general method.
3. Let X have the p.d.f.

Let be a random sample of size n.

a. (10 pts.) Construct a 100(1)% approximate confidence interval for  using the
central limit theorem.
b. (10 pts.) Construct a 100(1)% approximate confidence interval for ln using the
asymptotic distribution of the MLE of .

37
 4. Please answer all the following steps.
a) (5 pts.) Create a research question.
b) (5 pts.) State the null and the alternative hypothesis based on your research question.
c) (5 pts.) Discuss the type of errors that may occur and when they can occur.
d) (5 pts.) Which type error is more serious in your study? Please discuss it by giving
strong justification.
e) (5 pts.) What is the p-value related to your hypothesis?

Dr. Ceylan Yozgatlıgil

9.5.2011
METU-SPRING 2011
STAT 272
MIDTERM EXAM II

QUESTIONS

1. Let X be a random variable whose pmf under H0 and H1 is given by

x 1 2 3 4 5 6 7
0.02 0.02 0.02 0.01 0.02 0.01 0.90
0.06 0.06 0.05 0.02 0.03 0.01 0.77

a) (10 pts.) Find the Most Powerful Test of size =0.05 to test H0 versus H1.
b) (10 pts.) Compute the probability of Type II error for this test.
2. Consider a random sample of size n from Normal(0, 2).
a) (15 pts.) Derive the most powerful test of size  for testing

b) (5 pts.) Find the power of the test given in a).

3. A random sample is drawn from the following pdf

 cdf of X(n):
 cdf of X(1):
a) (15 pts.) Derive the Uniformly Most Powerful Test of size  for testing
against based on a random sample of size n.
b) (5 pts.) Derive a 100(1)% confidence interval for  using the UMP test result that
you found in part a).

38
4. Let X = (X1, ...,Xn) be a random sample from the density

a) (5 pts.) Find the Monotone Likelihood Ratio Statistics for .

b) (20 pts.) Derive a Uniformly Most Powerful test of size α for testing

H0 : θ = θ0 versus H1 : θ < θ0.

5. In each of the following cases, write the p-value.

a) (5 pts.) For testing H0:=1/2 versus H1<1/2, 4 successes are observed out of 10
Bernoulli trials.
b) (5 pts.) For testing H0:=1 versus H1:>1, X=3 is observed, where X~Poisson().
c) (5 pts.) For testing H0:=1 versus H1:>1, X1=3, X2=5 and X3=1 is observed, where
Xi~Poisson(), independent.

Dr. Ceylan Yozgatlıgil 7.6.2011

METU-SPRING 2011
STAT 272
MIDTERM EXAM II

QUESTIONS

1. Consider a random sample of size n from a distribution with pdf

and zero elsewhere.

a) (10 pts.) Derive a uniformly most powerful test of size for testing

vs. .

b) (10 pts.) Find the power function of the test.

2. Consider a random sample from a distribution where 
is known.
a) (10 pts.) Derive a generalized likelihood ratio test of size  for testing

b) (5 pts.) Find the power function for this test.

c) (10 pts.) Find the asymptotic test results for the test given in part a) using the
asymptotic distribution of

39
 3. (10 pts.) Suppose a population is Poisson distributed with mean . For a random
sample of size n, derive a sequential probability ratio test for testing versus
with and , approximately.

4. (15 pts.)Let be a sample variance based on a random sample of size

n from a Normal population with mean  and variance 2. Consider the prior
distribution for 2 to be the inverted gamma distribution, IG(,), with pdf

Find the posterior distribution of 2.

5. Let be a random sample from an exponential distribution with mean .

For testing

a likelihood ratio test is obtained as or

a) (7 pts.) Obtain the confidence interval from this test and interpret it.
b) (8 pts.) Find the p-value of the test.
c) (7 pts.) Draw the type II probability error region on the graph for where
.
d) (8 pts.) Discuss how we can use Bayesian approach to test the same hypothesis.

Dr. Ceylan Yozgatlıgil 03.04.2018

METU-SPRING 2018
STAT 304
MIDTERM EXAM I
(Time Duration: 100 minutes)

QUESTIONS

5. Let X = (X1, ...,Xn) be a random sample from the density

where α is known.

 cdf of X(n):
 cdf of X(1):


40
 a) (5 pts.) Is the unknown parameters β location or scale parameter? Justify your
answer with calculation.
b) (5 pts.) Find a Pivotal Quantity for using the MLE of .
c) (10 pts.) Construct a lower 95% confidence limit for using the Pivotal
Quantity that you found in part b.
d) (10 pts.) Construct a 90% confidence interval for using the distribution
function method.
e) (5 pts.) Compare the intervals that you found in part c) and d), if possible.

6. Let X = (X1, ...,Xn) be a random sample from the density

for .

a) (10 pts.) Find a sufficient statistic for .

b) (10 pts.) Construct a 100(1)% confidence interval for using the general
method and equal tail probabilities.
7. Let X have the p.d.f.
.

Let be a random sample of size n.

a) (10 pts.) Construct a 100(1)% exact confidence interval for  using any
method that you want.
b) (10 pts.) Construct a 100(1)% approximate confidence interval for  using
the central limit theorem.
c) (5 pts.) Explain how you can compare these two intervals. Which
mathematical measure you want to choose to decide on the more precise one?
8. Please answer all the following steps.
f) (5 pts.) Create a research question related to any subject that you can imagine.
g) (5 pts.) State the null and the alternative hypothesis based on your research question.
h) (5 pts.) Discuss the type of errors that may occur and when they can occur.
i) (5pts.) Which type error is more serious in your study? Please discuss it by giving
strong justification.

41
Dr. Ceylan Yozgatlıgil 8.5.2018
METU-SPRING 2018
STAT 304
MIDTERM EXAM II
(Time Duration: 120 minutes)

QUESTIONS

1. Consider a random sample (X1, ...,Xn) and two simple hypotheses and
.

a) (10 pts.) Find the Most Powerful Test of size =0.05.

b) (10 pts.) Find the power function of this test.

2. X is a single r.v. with pdf

a) (10 pts.) Find the MPT of size  for . Find the p-value of the test if
x is observed as 0.6. What is your conclusion on the test.
b) (10 pts.) To test , find the UMPT of size , if exits. Find the power of
the test.
c) (10 pts.) To test , the following procedure was used: Reject H 0 if
X>1/2. Find the power and the size of the test.

3. Consider a r.s. (X1, ...,Xn) from a distribution with pdf

zero elsewhere, where .

a) (10 pts.) Find the uniformly most powerful test of size  for testing
.
b) (10 pts.) Using the test that you construct in part b), construct 100(1)% Uniformly
most Accurate Confidence Interval. Discuss how you can use this interval to decide
on the hypothesis.

42
 c) (10 pts.) Find the p-value formula for this test and explain the meaning of the p-
value.

Hint: cdf of X(n):

cdf of X(1):

4. Consider a random sample from a distribution where both 

and  are unknown.
2

d) (10 pts.) Derive a generalized likelihood ratio test of size  for testing

e) (10 pts.) Find the power function for this test.

Dr. Ceylan Yozgatlıgil 29.5.2018

METU-SPRING 2018
STAT 304
FINAL EXAM
(Time Duration: 120 minutes)

QUESTIONS

3. (20 pts.) Let be a random sample from a distribution with pdf

If , find the Sequential Probability Ratio Test of

4. Suppose that we have a random sample of size n from the Pareto density function

where  is known.
 cdf of X(n):
 cdf of X(1):

c) (10 pts.) Derive the uniformy most powerful test of size  for testing
against
d) (5 pts.) Derive the 100(1)% uniformly most accurate confidence interval for .
e) (10 pts.) Derive a 100(1)% lower confidence bound for  using any exact method
that you want.

43
 f) (5 pts.) Compare the confidence intervals that you derived in part b) and part c). If
they are different than each other, which one do you prefer? Explain the reason.
g) (5 pts.) Find the p-value of the test.
3. A random sample follows the following pdf

a) (20 pts.) Derive the generalized likelihood ratio test of size  for testing
against based on a random sample of size n.
b) (5 pts.) Find the power function of the test.
c) (5 pts.) Find the asymptotic test results for the test given in part a) using the
asymptotic distribution of

4. Let be a sample variance based on a random sample of size n from a

Normal population with mean  and variance 2. Consider the prior distribution for 2
to be the gamma distribution, Gamma(,), with pdf
a) (15 pts.) Find the posterior distribution of 2.
b) (5 pts.) Find the Bayesian estimator of 2 under the squared error loss function.

Dr. Ceylan Yozgatlıgil 29.03.2019

METU-SPRING 2019
STAT 304
MIDTERM EXAM I
(Time Duration: 100 minutes)

QUESTIONS

1. (10 pts.) Consider a random sample of size n, . Show that the statistic

where

is location-scale invariant.

2. Consider a random sample of size n from a distribution with pdf

and zero elsewhere.

44
 a) (5 pts.) Is this p.d.f. belongs to a family of location, scale or location-scale type? Why? Is the
unknown parameter location or scale parameter? Explain.

b) (10 pts.) Find a Pivotal Quantity for considering the type of the parameter.

c) (10 pts.) Construct a 100 % confidence interval for with equal-tail probabilities using
the Pivotal Quantity that you found in part b.

3. Consider a random sample of size n from a distribution with pdf

 cdf of X(n):

 cdf of X(1):
a. (10 pts.) Construct a 100(1-α)% upper confidence limit for  using the general
method.
b. (10 pts.) Construct an approximate 100(1-α)% confidence interval for  using the central
limit theorem.

4. Consider the test vs , where

x 2 4 6

f0(x) 0.10 0.15 0.75

f1(x) 0.60 0.30 0.10

a) (10 pts.) For a single x, find the rejection region for a randomized test with =0.15.
b) (5 pts.) Find also .
5. The random variable X has pdf

In order to test hypothesis about , a single observation of X, denoted x, is taken.

a) (10 pts.) Find the most powerful test of the hypothesis versus of
size α.
b) (10 pts.) Find the power of the test.
c) (10 pts.) Calculate the p-value of the test and discuss its meaning.

Dr. Ceylan Yozgatlıgil 29.4.2019

METU-SPRING 2019

45
 STAT 304
MIDTERM EXAM II
(Time Duration: 100 minutes)

QUESTIONS

1. Let X be a single random variable with probability density function given below:

a) (13 pts.) Find the most powerful level 0.1 test of H0:  = 0 versus H1:  = 1.

b) (12 pts.) Obtain the power of this test.

2. Suppose that where ’s are fixed constants that are not all
the same, ’s are independent and identically distributed as , and is an
unknown parameter.
a) (20 pts.) Derive a uniformly most powerful test of size  for testing
versus .
b) (10 pts.) Obtain the 100(1α)% Uniformly Most Accurate (UMA) confidence
region for Discuss how you can use this interval to decide on the hypothesis.
c) (10 pts.) Find the p-value of the test.

3. Consider two random samples from a distribution and

from a distribution which are independent, and both and
are unknown.
f) (20 pts.) Derive a generalized likelihood ratio test of size  for testing

g) (15 pts.) Derive the asymptotic test results for the test given in part a) using the
asymptotic distribution of

BONUS:(15 pts.) Find the power function for this test.

46
Dr. Ceylan Yozgatlıgil 29.05.2019
METU-SPRING 2019
STAT 304
FINAL EXAM
(Time Duration: 120 minutes)

QUESTIONS

1. Let X have the p.d.f.

.

Let be a random sample of size n.

a)(10 pts.) Construct a 100(1)% exact confidence interval for  using any
method that you want.
b)(10 pts.) Construct a 100(1)% approximate confidence interval for  using
the central limit theorem.
c)(5 pts.) Explain how you can compare these two intervals. Which mathematical
measure you want to choose to decide on the more precise one?

2. A random sample is drawn from the following pdf

 cdf of X(n):
 cdf of X(1):
c) (10 pts.) Derive the Uniformly Most Powerful Test of size  for testing
against based on a random sample of size n.
d) (5 pts.) Find the p-value for this test.
e) (10 pts.) Derive a 100(1)% confidence interval for  using the UMP test result that
you found in part a).
3.Suppose that X is a continuous random variable with p.d.f.

c. (15 pts.) Derive the generalized likelihood ratio test of against

based on a random sample of size n.

d. (5 pts.) Find the power function of the test.

47
 4.(10 pts.) Suppose a population is Poisson distributed with mean . For a random sample
of size n, derive a sequential probability ratio test for testing versus
with and , approximately.

4. (20 pts.) Let X(n) be the nth order statistic of a random sample of size n from a
distribution with pdf
.
Let  be an observed value of the random variable which has pdf

Using the squared error loss function, find the Bayesian point estimate of .
Note: cdf of X(n):

Dr. Ceylan Yozgatlıgil 01.06.2020

METU-SPRING 2020
STAT 304
FINAL EXAM – PART I
(Time Duration: 180 minutes)

QUESTIONS

1. Let be a random sample with the following pdf

where  > 0 is an unknown parameter.

a) Does this family of distributions belong to location or scale family? Justify your
answer.
b) Does this family of distributions have a monotone likelihood ratio (MLR) in the MLE
of ? Justify your answer.
c) Use the MLE of  to construct the shortest 95% confidence interval for .

2. Let be a random sample from N(µ, 1) with an unknown µ. After sample is

taken, an intern forgets to record the values of in a study and instead only

records for where I(.) is an indicator function. Use the

48
 observed data Y to construct a uniformly most powerful (UMP) test of size α for
testing
.
Hint: Consider the Bernoulli experiment with success probabilities coming from the given
distribution.

3. Suppose that form a random sample from a distribution with the following
p.d.f.

Let and we would like to test H0 : 2    4 vs. H1 :  < 2 or  > 4.

The critical region is chosen to be C = {X(n) < 3 or X(n) > 4}.

a) Determine the power function of the test.
b) Determine the size (maximum type I error) of the test.

4. Let be independent samples (not i.i.d.) from the probability distribution

with pdf

where i > 0, i = 1, 2, respectively.

a) Derive the likelihood ratio test of size α for testing H 0 : 1 = 2 vesus H1 : 1 < 2 when
{i, i = 1, 2} are unknown.
b) Derive the power function of your test.

5. Let X = ( ) be a random sample from the uniform distribution on the interval

(0, θ), where θ > 0 is unknown. Let the prior distribution of θ be the log-normal
distribution with parameter (μ, σ2), where μ ∈ R and σ2 > 0 are known constants.
a) Find the posterior density of log θ.
b) Find the rth posterior moment of θ.
c) Find a value that maximizes the posterior density of θ.

49