P a g e | 83

CHAPTER 6

Hypothesis Testing

6.1 Introduction
6.2 The Neyman-Pearson Lemma
6.3 Likelihood Ratio Tests
6.4 Hypotheses for a Single Parameter
6.5 Testing of Hypotheses for Two Samples
6.6 Chapter Summary
6.7 Computer Examples
 P a g e | 84

Exercise 6.1

6.1.1

a.

b.

6.1.2

a. Support for a presidential candidate is changed after the start of the use of TV commercial.
b. Proportion of viewers is greater than or equal to 30%.
c. The median grade point average is not equal to 2.9.

6.1.3

a. = Probability of type I error = P(reject | is true)

=

= 1-0.8491

= 0.1509

b. = P(accept | is false)
=

= 0.2784

c. =
 P a g e | 85

= 0.5968

d. For

From the binomial table we can observe that, for , and

for and

falls between and . However, for , which exceeds 0.01.

If we want to limit to be no more than 0.01, we will take . That is we reject if .

For

Similar to above method, we can see that

falls between and . However, for , which exceeds 0.03.

If we want to limit to be no more than 0.03, we will take . That is we reject if .

e. When . From part (d), rejection region is of the form .

For

6.1.4

a. = p(reject | is true)
=

=1-0.39
 P a g e | 86

=0.61

=p(not reject | is false)

=0.213

The type I error is increased whereas type II error is decreased.

(b) Similar to ‘a’ except n=20 and n=25

6.1.5 ,

a. = Probability of type I error = P(Reject | is true)

b. = P(Accept | is false)
 P a g e | 87

c.

=160.78 rounded up to 161

6.1.6

a. =probability of type I error = p(reject | is true)

b. =P(do not reject | is false)

 P a g e | 88

6.1.7

Under Null Hypothesis, there are 10 green and 10 red balls.

Under Alternative hypothesis there are 15 green and 5 red balls.

All 3 balls are green

All 3 balls are not green

= All 3 balls are green

=1-

6.1.9

=173.19 rounded up to 174

Exercise 6.2

6.2.1
 P a g e | 89

a. Therefore, if ,
the most powerful test is to reject if
 P a g e | 90

b. If , the most powerful test is to reject if

Where

6.2.2

Following the proof in 7.2.1 we can write: The rejection region of the test does not depend on the

value of when vs. . Hence the test is uniformly most powerful. On the

other hand, the test vs. is not uniformly most powerful.

6.2.3

The Neyman-Pearson Lemma gives

And a critical region . We need

has a c.d.f .

6.2.4
 P a g e | 91

The most powerful test is: reject if

Taking natural logarithm, we have

Hence is the most powerful test where

6.2.5

For any
 P a g e | 92

Therefore the most powerful test is, Reject H0 if

Where

6.2.6

So

implies
 P a g e | 93

Thus the rejection region is

Which is the most powerful test at level of .

6.2.7

where

Taking natural logarithm, we have

To find the rejection region for a fixed value of , write the region as

, where

6.2.8
 P a g e | 94

This implies which is the most powerful test of the hypothesis.

6.2.10

Mean= ,variance =25,sample size(n)=20, =0.05.

To test,

Now,
 P a g e | 95

Simplifying, we get,

That is the Neyman Pearson Lemma tells us that the rejection region for the most powerful test for
testing.

is of the form , where is selected so that the size of the critical region

=0.05. Under the null hypothesis, the sample mean is normally distributed with mean=5 and standard

deviation= .

Therefore, the critical value is deemed to be .

Thus the Neyman Pearson lemma tells us that the rejection region for the most powerful test for

, under the normal probability model is:

The power of such a test when is

Exercises 6.3

6.3.1
 P a g e | 96

Here and

Hence

σ^
Similarly for the alternative hypothesis, mle 2

if

Since

The likelihood ratio test has the rejection region:

 P a g e | 97

Reject if , which is equivalent to

which is equal to

6.3.3

Here and

The joint pdf

The MLE’s are and

Under H0, . MLE’s on are

and

After some algebra we can show that , Reject if

where
 P a g e | 98

6.3.4

Let

We reject if or

If , then

where

Similarly we reject if where is a constant.

6.3.5
 P a g e | 99

for

MLE of is ,

We reject the null hypothesis if

, or

6.3.6

To test,

vs.

This is one sided z test:

Using sample data, we calculate the pooled sample proportion (p) and the standard error (SE). Using
those measures, we compute the z-score test statistic z.
 P a g e | 100

Since we have a one-tailed test, the P-value is the probability that the z-score is less than .
We use the Normal Distribution Table to find . Thus, the .

Since the P-value 0.80785 is greater than the significance level 0.01, we cannot reject the null hypothesis.

Exercises 6.4

6.4.1

a. The observed test statistic ( ) is

b. p-value= = =0.0384

c. Smallest =0.0384

p-value

Since P value is greater than 0.02, we fail to reject the null hypothesis.

6.4.2

Here and both are greater than 5, we can use normal approximation to the binomial test
hypothesis

^p =0.2571
 P a g e | 101

Test statistic =

For =0.02,

Hence the rejection region is

Since 5.59 is greater than 2.55, we reject . Hence the evidence supports the claim.

6.4.3

a.

b. Here

c. Assumptions: even though population standard deviation is unknown, because of the large sample
size, normal distribution is assumed.
 P a g e | 102

6.4.4

a. , , ,

Yes, the null hypothesis is rejected.

b.

Where calculated t-value<

So we fail to reject the null hypothesis.

c. Due to the variability of the samples, even if we keep the sample size and level of significance
fixed, we might end up in making completely different decisions. That is why, many times, we
may have to do many samples before making decisions.

6.4.5

, ,

a. P(reject | is true)=0

Since the probability of rejecting null hypothesis equals to zero. Therefore, the null hypothesis is
accepted.

b.

vs
 P a g e | 103

Rejection region is

Since t=2.642 is greater than 1.47, we fail to reject the null hypothesis

6.4.6

and

We can approximate normal

Rejection region is =

We fail to reject the null hypothesis

p-value=

So we fail to reject the null hypothesis even at level

6.4.7.

Since , using normal approximation to binomial

95% confidence interval for

 P a g e | 104

vs ,Test Statistic

Since we have enough evidence to reject H0.

6.4.8

Rejection region is

i.e.

Yes, customer has preference over ivory color.

6.4.9

, , ,

Since n<30 we use t distribution

 P a g e | 105

Rejection region is

. This is sufficient evidence to confirm the belief

6.4.10

a. , ,

The data is normally distributed.

Rejection region for z is where

We fail to reject the null hypothesis

b. 90% confidence interval for is

c. From a, we can see that it is reasonable to take , 44 falls in the confidence interval. The
argument is supported by the confidence interval in (b).

6.4.11
 P a g e | 106

130
120
Normal Q-Q Plot
Sample Quantiles

110
100
90

-2 -1 0 1 2

Theoretical Quantiles

The QQ plot does not support the normality assumption. Better to use a non-parametric test.

, ,

vs

Rejection region is

Hence we reject the null hypothesis

6.4.12

, ,

Rejection region is

We do not reject the null hypothesis.

 P a g e | 107

6.4.13

, ,

For , and a degree of freedom of 9, the rejection region is

Test statistic

Since value of test statistic does not fall in the rejection region, we cannot reject .

Hence the company’s claim can be substantiated under 0.05 level of confidence.

6.4.14

, , ,

Rejection region is

Hence we reject the null hypothesis that the expenditure per consumer is increased from 1994 to 1995

6.4.15 ,

252
^p= =0. 63
400
 P a g e | 108

>5 and >5, we can use normal approximation of the binomial distribution

Rejection region is

Since , we fail to reject the null hypothesis.

6.4.16

Where

Thus we reject the null hypothesis. That is price of gas is changed recently.

6.4.17 , , ,

vs

Rejection region is
 P a g e | 109

Thus we fail to reject the null hypothesis

Exercises 6.5

6.5.1

Rejection region for z is

Since , we fail to reject the null hypothesis. To see the significant difference we need to
have (p value) level of significance.

6.5.2

, , , , , ,

Rejection region for

 P a g e | 110

Since -2.508<-0.412364, we fail to reject the null hypothesis

6.5.3 , , , , , ,

Rejection region is i.e.

Since 4.6124>1.6787, we reject the null hypothesis. Thus implies that there exists significant
evidence to show that the male’s salaries are higher than that of female.

6.5.4

Here , and

, and
 P a g e | 111

Rejection region is

Reject . Yes, African Americans are more likely to get cancer.

6.5.5 , , , , , ,

Since and , we use normal distribution

Use

Rejection region is

Since 80.2758>1.96, we reject the null hypothesis. Therefore we can conclude that there exist a
difference between mean weights of sources A and B.

6.5.6

, , , , ,

Rejection region for

Since 8.235>2.325, we reject the null hypothesis.

6.5.7 Let be the upstream BOD and be the downstream BOD values

, , , , , ,
 P a g e | 112

a.

Rejection region is i.e.

Since -5.259<-1.7341, we have enough evidence to reject the null hypothesis. Thus the mean
BOD for the downstream is more than for the upstream.

b.

Test statistic

From the F-table

Rejection region is and

Since the observed value of the statistic 1.1923<4.026, we fail to reject the null hypothesis

c. Let Upstream - Downstream

 P a g e | 113

From t-table

We reject the null hypothesis. This implies that the mean of BOD for downstream is more than
the upstream.

6.5.8

, , , , , ,

Assuming the data follow normal distribution

The degrees of freedom for the t-distribution are given by:

Use

Rejection region is

Since 5.0007>2.131, we reject the null hypothesis. Therefore, we can conclude that there exist a
difference in the individuals’ driving ability under the two conditions.
 P a g e | 114

6.5.9 Let Post IVCP – Pre IVCP

a. vs , , ,

From t-table , Since we have enough evidence to reject

the null hypothesis. , , vs

Test statistic , From the F-table ,

Rejection region is and , we fail to reject the null hypothesis.

6.5.10

Now,

We know that,

So we have,

Hence we get,
 P a g e | 115

Therefore, is an unbiased estimator of .

6.5.11 , vs

Test statistic , From the F-table ,

Rejection region is and , Since the observed value of the statistic 1.5<2.90,
we fail to reject the null hypothesis.

6.5.12

, , , , ,

a.

Since the variances are equal and unknown

Rejection region is

Since , thus does not fall in the rejection region. We fail to reject , in
other words, the mean SAT score in math is same for 1999 and 1989.
 P a g e | 116

b. Test statistic

From the F-table

Rejection region is and

Since observed value of test statistic = 0.6622<0.3493, there is evidence that the variances are
equal.

6.5.13 , , , , ,
,

a. vs , Since the variances are equal and unknown

, ,

Rejection region is Since , thus does not fall in the rejection

region. We fail to reject , in other words, the mean SAT score in math is same for 1999 and 1989.

b. Test statistic , From the F-table

Rejection region is and

Since observed value of test statistic (1.5124)<2.862, there is evidence that the variances are
equal.