1 STATISTICAL HYPOTHESES

Required Readings
Anderson, David, Dennis J. Sweeney & Thomas A. Williams (2011): Statistics for Business
and Economics, 9th edition, CENGAGE Learning.

Venkatesh, Seshamani & Ndhlovu Obrien (2016): Statistical Methods for Economic Analysis,
Printech Ltd.

Spiegel, Murray, John Schiller & R. Alu Srinivasan (2012): Probability and Statistics, 4th
edition, Schaum’s outline series, McGraw-Hill.

Spiegel, Murray, & Stephens J. Larry (2008): Theory and Problems of Statistics, 4th edition,
Schaum’s outline series, McGraw-Hill.

1.1 Unit introduction

Welcome to Unit 5 Hypothesis Testing.

This unit, takes you further from the previous chapter on statistical inference. Recall that a
sample was used to develop point and interval estimates of population parameters.
1.2 Unit Aim
This unit aims to familiarize you with hypotheses testing both for small samples and large
samples.

1.3 Unit learning outcomes

By the time you finish working through this unit, you should be able to;

1. Distinguish between Type I error and Type II error

2. Should be able to conduct hypothesis tests for small sample as well as large samples

1
 1.4 Contents list

1. Tests of hypotheses and significance; Type I and Type II errors; level of significance;
one-tailed and two-tailed tests; P value;
2. Large sample tests for variables and attributes-tests based on normal distribution; small
sample tests for variables and attributes-tests based on t-distribution and chi-square
distribution

1.5 Time required

You must be done with this unit in three weeks.
1.6 Testing of hypotheses and significance
Under hypothesis testing, we seek to establish a claim or conjecture about the value of the
population parameter on the basis of information contained in the sample. Thus we can define
a hypothesis as a statement that something is true. Under hypothesis testing, there are two
competing assumptions or hypotheses that are tested using information contained in the sample.
These two hypotheses are the null and alternative hypotheses.
Note that the claim or hypothesis is what is called the alternative hypothesis, and its denoted
by either 𝐻1 or 𝐻𝑎 . On the other hand, the null hypothesis denoted by 𝐻0 is the opposite
hypothesis that nullifies the original claim or alternative hypothesis. We can therefore
conclude that the alternative hypothesis is what the test attempts to establish or we look for
evidence to support the claim.
1.6.1 Type I and type II errors
As previously noted, the null and alternative hypotheses are competing statements about the
value of the population parameter. Thus either one of them can be true. To this end, we
expect a researcher to accept 𝐻0 when 𝐻0 is true and reject 𝐻0 when 𝐻𝑎 is true. However,
there is a possibility of errors in the decision making process. In other words, sometimes
correct decisions may not be reached. There are two types of errors.
1) Type I error
We commit a type I error when we reject 𝐻0 when 𝐻0 is true. Note that type I error is
also called the level of significance (i.e 𝛼 = 0.1, 0.05, 0.01 etc). Thus if the level of
significance or type I error is small, we have a high degree of confidence that the
researcher’s conclusion to reject 𝐻0 is correct.
2) Type II error

2
 This is the failure to reject 𝐻0 when 𝐻1 is true. Or accepting 𝐻0 when 𝐻0 is false.
The table below shades more light on the two types of errors.

Population Condition
𝐻0 True 𝐻𝑎 True
Accept Correct
𝐻0 Conclusion Type II error
Conclusion
Reject Correct
𝐻0 Type I error Conclusion
Type I error, also called the level of significance can be expressed as:
𝛼 = 𝑃(𝑇𝑦𝑝𝑒 𝐼)
= 𝑃(𝑟𝑒𝑗𝑒𝑐𝑡 𝐻0 𝑔𝑖𝑣𝑒𝑛 𝑡ℎ𝑎𝑡 𝐻0 𝑖𝑠 𝑡𝑟𝑢𝑒)
= 𝑃(reject 𝐻0 |𝐻0 is true)
On the other hand, type II error can be expressed as:
𝛽 = 𝑃(𝑇𝑦𝑝𝑒 𝐼𝐼)
= 𝑃(reject 𝐻0 |𝐻1 is true)
Under hypothesis testing, the two possible conclusions are:
1. If you reject 𝐻0 , conclude that 𝐻1 is probably true
2. If you fail to reject 𝐻0 , conclude that there is no sufficient evidence from the sample
data to support the claim (𝐻1 ).
Note that hypotheses are always statements about the population distribution of interest and
not statements about the sample.
Note further that the probability of correctly rejecting 𝐻0 when 𝐻0 is false is called the power
of the test, expressed as 1 − 𝛽. In other words, 1 − 𝛽 is the probability of correctly rejecting
𝐻0 is one minus the probability of making type II error.

1.6.2 Conducting a hypothesis test

The following steps will assist us in carrying out the test
1. Formulate the hypothesis. That is, specify 𝐻0 and 𝐻1 as well as 𝛼.

3
 2. Select an appropriate test statistic. The test statistic is used as basis for deciding
𝜎 2
𝑋 −𝜇 ̅
whether 𝐻0 should be rejected. For example, for 𝑋̅~𝑁(𝜇, 𝑛 ), 𝑍 = 𝜎⁄ 𝑛 is the test
√
𝑋−𝜇
statistic assuming 𝜎 is known. Or for 𝑋~𝑁(𝜇, 𝜎 2 ), 𝑍 = is the test statistic.
𝜎

3. Determine the rejection region. This consist of a set of values for the test statistic that
lead to the rejection of 𝐻0 .
a) For a two-tailed test, at the level of significance 𝛼, the rejection regions are as
shown below.

Rejection Region Rejection Region

Acceptance Region

i.e do not reject 𝐻0

−𝑍𝛼⁄2 𝑍𝛼⁄2

Note that −𝑍𝛼⁄2 and 𝑍𝛼⁄2 are called the critical values because they separate the
rejection and non-rejection regions.

b) For a one-tailed test at 𝛼 level of significance, rejection regions are shown below:
(i) Left-tailed. That is, 𝐻0 : 𝜇 = 𝜇0 vs 𝐻1 : 𝜇 < 𝜇0

Rejection region

Acceptance region
−𝑍𝛼

(ii) Right-tailed. That is, 𝐻0 : 𝜇 = 𝜇0 vs 𝐻1 : 𝜇 > 𝜇0

Rejection region

Acceptance region

𝑍𝛼

2. Calculate the value of the test statistic. For example

𝑋̅ −𝜇
𝑍 = 𝜎⁄ large sample with 𝜎 known
√𝑛
𝑋̅ −𝜇
𝑡 = 𝑠⁄ small sample with 𝜎 unknown
√𝑛

4
3. Draw conclusion
1.6.3 One-tailed and two tailed tests
One tailed tests
With one-tailed tests, the rejection region is either on the lower or upper tail of the sampling
distribution. By rejection region, we mean the area corresponding to the level of significance
of , for example, 𝛼 = 0.1,0.05,0.01, etc. Recall that we use information contained in the
sample to establish a claim or conjecture. For example, to test the claim that the mean height
of all UNZA students (𝜇) is greater than 1.5m, you can assess whether the observed sample
mean (𝑋̅) falls in the upper tail of the sampling distribution by computing a test statistic
which is given by
𝑋̅−𝜇0
𝑍= 𝜎⁄√𝑛
𝑋̅−𝜇0
} for large sample case.
𝑍= 𝑠⁄√𝑛

Note that in the above example, the direction of the claim is one side because it specifies that
the value of 𝜇 is above or on one side of 1.5.
The null hypothesis (𝐻0 ) in this case or opposite is 𝜇 is equal or less than 1.5. note that it is
customary to have 𝐻0 specify that there is no difference and thus an equal(=) sign will be
used. In the above example, the two competing hypotheses are:
𝐻0 : 𝜇 = 𝜇0
⇒ 𝐻0 : 𝜇 = 1.5 vs
𝐻1 : 𝜇 > 1.5.
Two-tailed tests
Unlike under a one-tailed hypothesis test, the two sided tailed tests have rejection regions in
both the lower and upper tails of the sampling distribution. In this case, the direction of the
claim is unknown. Thus the two competing hypotheses are:
𝐻0 : 𝜇 = 𝜇0 vs 𝐻1 : 𝜇 ≠ 𝜇0
For a statement where 𝐻1 or 𝐻𝑎 is 𝜇 ≠ 𝜇0
In conclusion, the rules are:
1. One-tailed test: left tailed
𝐻0 : 𝜇 ≥ 𝜇0
𝐻1 : 𝜇 < 𝜇0
𝑋̅ −𝜇0
𝑍= assuming 𝜎 is known
𝜎⁄√𝑛
𝑋̅ −𝜇0
𝑍= 𝜎 estimated by 𝑠.
𝑠⁄√𝑛

5
 Rejection rule
Using test statistic: reject 𝐻0 if 𝑍 < −𝑍𝛼 and using p-value: reject 𝐻0 if p-value< 𝛼
2. One-tailed test: right tailed
𝐻0 : 𝜇 ≤ 𝜇0
𝐻1 : 𝜇 > 𝜇0
𝑋̅ −𝜇0
𝑍= assuming 𝜎 is known
𝜎⁄√𝑛
𝑋̅ −𝜇0
𝑍= 𝜎 estimated by 𝑠.
𝑠⁄√𝑛

Rejection rule
Using test statistic: reject 𝐻0 if 𝑍 > 𝑍𝛼 and using p-value: reject 𝐻0 if p-value> 𝛼

3. Two-tailed test:
𝐻0 : 𝜇 = 𝜇0
𝐻1 : 𝜇 ≠ 𝜇0
𝑋̅ −𝜇0
𝑍= assuming 𝜎 is known
𝜎⁄√𝑛
𝑋̅ −𝜇0
𝑍= 𝜎 estimated by 𝑠.
𝑠⁄√𝑛

Rejection rule
Using test statistic: reject 𝐻0 if 𝑍 < 𝑍−𝛼/2 or if 𝑍 > 𝑍𝛼/2 and using p-value: reject 𝐻0
if p-value< 𝛼.

Confidence interval approach

For a two-tailed test, the confidence interval can be used to determine whether the
hypothesized population value falls in the rejection region or acceptance region. Recall that
for the mean, the 1 − 𝛼 confidence interval was given as:
𝜎
𝑋̅ ± 𝑍𝛼⁄2
√𝑛
For a large sample case and assuming 𝜎 is known. Note that we do not reject 𝐻0 if region
contain all values of the sample mean 𝑋̅ that are within -𝑍𝛼⁄2 and 𝑍𝛼⁄2 standard errors of the
hypothesized value 𝜇0 .

6
Decision rule: if the confidence interval contains the hypothesized value, do not reject 𝐻0
otherwise reject 𝐻0 .
Example: Large sample statistical test
A quality control manager believes that 12-gram box of breakfast cereal should contain an
average of 6 gram of bran. He measures a random sample of 60 booles of a popular cereal for
bran content. Suppose the data yield a sample mean of 5.9 gram of bran and a sample
standard deviation of 0.37 gram of bran. Does the data suggest that the bran content differs
from 6 grams?
Solution
1. Hypothesis
This is a two-tailed test
𝐻0 : 𝜇 = 6 𝑣𝑠 𝐻1 : 𝜇 ≠ 6

2. Test statistic
𝜎2
𝑋̅~𝑁(𝜇, )
𝑛
𝑋̅ −𝜇
Now since 𝑛 = 60 > 30 we use 𝑍 = 𝜎⁄ if 𝜎 is given. But since 𝜎 is unknown, we
√𝑛

approximate the test statistic using

𝑋̅ − 𝜇
𝑍=
𝑠⁄√𝑛
We are given that 𝑋̅ = 5.9, 𝑛 = 60, 𝑠 = 0.37, 𝜇 = 6. Therefore
5.9 − 6 −1.1
𝑍= = = −2.09
0.37⁄√60 0.37⁄√60
3. Rejection region

−𝑍𝛼⁄2 𝑍𝛼⁄2

𝛼 = 0.05 and since it is two-tailed 𝑍𝛼⁄2 = 𝑍0.025 = 1.96

4. Decision rule
Reject 𝐻0 if 𝑍 < −𝑍𝛼⁄2 or 𝑍 > 𝑍𝛼⁄2

7
 5. Conclusion
Since 𝑍 < −𝑍𝛼⁄2 that is -2.09<-1.96 we reject 𝐻0 and conclude that there is some
evidence that the mean bran content differs from 6 grams.
Small-sample test with unknown 𝝈
One-tailed Two-tailed
𝐻0 : 𝜇 = 𝜇0 𝐻0 : 𝜇 = 𝜇0
𝐻1 : 𝜇 < 𝜇0 (𝑜𝑟 𝜇 > 𝜇0 ) 𝐻1 : 𝜇 ≠ 𝜇0
Test statistic
𝑋̅ −𝜇 𝑋̅ −𝜇0
𝑡 = 𝑠⁄ 𝑡=
√𝑛 𝑠⁄√𝑛

Rejection region
𝑡 < −𝑡𝛼,𝑛−1 𝑡 < −𝑡𝛼/2,𝑛−1
Or 𝑡 > −𝑡𝛼,𝑛−1 𝑡 > 𝑡𝛼/2,𝑛−1
Note that the variability of the sampling distribution for a t-distribution depends on the
sample size n-1, and recall that the n-1 was what we referred to as the degrees of freedom.
Example
TOYOTA wants to test a new engine to see whether it meets new air pollution standards. The
mean emissions 𝜇 of all engines of this type must be less than 20 parts per million carbon.
Ten engines are manufactured for testing purposes, and the mean and standard deviation of
the emissions of the sample engines are determined to be 17.1 and 3 parts per million carbon
respectively. Do the data supply sufficient evidence to allow the manufacturer(TOYOTA)
conclude that this type of engine meets pollution standards? Assume the manufacturer is
willing to risk a type I error of 1% (i.e with probability 0.01).

Solution
1. Hypothesis
This is a one-tailed test
𝐻0 : 𝜇 = 20 𝑣𝑠 𝐻1 : 𝜇 < 20
2. Test statistic

8
 𝜎2
𝑋̅~𝑁(𝜇, )
𝑛
𝑋̅ −𝜇
Now since 𝑛 = 10 < 30 we use 𝑡 = 𝜎⁄ if 𝜎 is given. But since 𝜎 is unknown, we
√𝑛

approximate the test statistic using

𝑋̅ − 𝜇
𝑡=
𝑠⁄√𝑛
We are given that 𝑋̅ = 17.1, 𝑛 = 10, 𝑠 = 3, 𝜇 = 20. Therefore
17.1 − 20
𝑡= = −3.06
3⁄√10

3. Rejection region
𝛼 = 0.01 and since it is one-tailed −𝑡𝛼,𝑛−1 = −𝑡0.01 = −2.821.

Rejection region

−𝑡𝛼,𝑛−1

4. Decision rule
Reject 𝐻0 if 𝑡 < −𝑡𝛼,𝑛−1

−2.821

5. Conclusion
Since 𝑡 < −𝑡𝛼,𝑛−1 that is -3.06<-2.821 we reject 𝐻0 and conclude that there is some
evidence that the mean emission of all engines are less than 20 parts per million
carbon. That is, the manufacturer concludes that 𝜇 < 20 meets the pollution
standards.

HYPOTHESIS TESTS ABOUT A POPULATION PROPORTION

𝐻0 : 𝑝 ≥ 𝑝0 𝐻0 : 𝑝 ≤ 𝑝0 𝐻0 : 𝑝 = 𝑝0

9
𝐻𝑎 : 𝑝 < 𝑝0 𝐻𝑎 : 𝑝 > 𝑝0 𝐻𝑎 : 𝑝 ≠ 𝑝0
𝑝0 − 𝑝
𝑍=
𝜎𝑝̅0

𝑝(1 − 𝑃)
𝜎𝑝̅0 = √
𝑛

Reject 𝐻0 is 𝑍 < 𝑍𝛼 for 𝐻𝑎 : 𝑝 < 𝑝0 and 𝑍 > 𝑍𝛼 for 𝐻𝑎 : 𝑝 > 𝑝0 and for 𝐻𝑎 : 𝑝 ≠ 𝑝0 , reject 𝐻0
if 𝑍 < −𝑍𝛼⁄2 or 𝑍 > 𝑍𝛼⁄2 .
P-VALUE APPROACH
This is another approach that is used to draw conclusions when conducting a hypothesis test.
In other words, conclusions can be drawn based on a probability called a p-value. The p-
value is also known as the observed level of significance. We can therefore define the p-value
as the observed area to the right of the computed/observed test statistic.
DECISION RULES WHEN USING THE P-VALUE APPROACH
1. For a one-tailed test
(a) Reject 𝐻0 if the p-value< 𝛼, meaning the test statistic is in the rejection
region.
(b) Fail to reject 𝐻0 if the p-value> 𝛼, meaning that the test statistic is not in
the rejection region.
2. For a two-tailed test
(a) Reject 𝐻0 if the p-value< 𝛼/2, meaning the test statistic is in the rejection
region.
(b) Fail to reject 𝐻0 if the p-value> 𝛼/2, meaning that the test statistic is not
in the rejection region.

For a two-tailed test, one can also use an alternative method that involves doubling the p-
value and then comparing it to the given level of significance 𝛼. That is, if 2(p-value) > 𝛼,
one fails to reject 𝐻0 whereas if 2(p-value) < 𝛼, one rejects 𝐻0 .
HYPOTHESIS TESTS ABOUT THE DIFFERENCE BETWEEN THE MEANS OF
TWO POPULATIONS: INDEPENDENT SAMPLES
We can use independent simple random samples to estimate the difference between two
population means (𝜇1 − 𝜇2 ). Recall that 𝑋̅1 − 𝑋̅2 represents the point estimator of the
difference between the means of the two populations. Thus to estimate 𝜇1 − 𝜇2 we select a
simple random sampling of size 𝑛1 observations from population 1 and a simple random

10
sampling of size 𝑛2 observations from population 2 to compute 𝑋̅1 and 𝑋̅2 respectively.
Recall that the standard deviation of the difference between two population means is:

𝜎12 𝜎22
𝜎𝑋̅1 −𝑋̅2 = √ +
𝑛1 𝑛2

Assuming 𝜎12 and 𝜎22 are known. If 𝜎12 and 𝜎22 are unknown, we use the sample standard
deviations as estimates of the population standard deviations and estimate 𝜎𝑋̅1 −𝑋̅2 using

𝑠12 𝑠22
𝑆𝑋̅1 −𝑋̅2 = √ +
𝑛1 𝑛2

HYPOTHESIS TEST ABOUT 𝝁𝟏 − 𝝁𝟐

a) Large-sample case with unknown variance
𝐻0 : 𝜇1 − 𝜇2 = 𝐷0 or 𝐻0 : 𝜇1 = 𝜇2
One-tailed test
Right tailed Left tailed
𝐻1 : 𝜇1 − 𝜇2 > 𝐷0 𝐻1 : 𝜇1 − 𝜇2 < 𝐷0
Two tailed
𝐻1 : 𝜇1 − 𝜇2 ≠ 𝐷0
Test statistic
𝜎12 𝜎22
𝑋̅1 − 𝑋̅2 ~𝑁 (𝜇1 − 𝜇2 , + )
𝑛1 𝑛2
(𝑋̅1 − 𝑋̅2 ) − (𝜇1 − 𝜇2 )
𝑍=
𝜎12 𝜎22
√
𝑛1 + 𝑛2

But since 𝜎12 and 𝜎22 are unknown, we estimate them with 𝑠12 and 𝑠22 . Therefore,
(𝑋̅1 − 𝑋̅2 ) − (𝜇1 − 𝜇2 )
𝑍=
𝑠12 𝑠22
√
𝑛1 + 𝑛2

Rejection region
One tailed test
Reject 𝐻0 if
𝑍 > 𝑍𝛼 (right tailed)
𝑍 < −𝑍𝛼 (left tailed)
Two-tailed test

11
Reject 𝐻0 if 𝑍 < −𝑍𝛼⁄2 or 𝑍 > 𝑍𝛼⁄2 .
Example
A university investigation, conducted to determine whether car ownership affects
academic achievement, was based on two random samples of 100 male students, each
drawn from the student body. The grade point average for 𝑛1 = 100 nonowners of
cars had an average and variance equal to 𝑋̅1 = 2.7 and 𝑠12 = 0.36 as opposed to
𝑋̅2 = 2.54 and 𝑠12 = 0.4 for 𝑛2 = 100 cars owners. Do the data provide sufficient
evidence to indicate that a difference in the mean achievement between car-owners
and non-car owners. Test using 𝛼 = 0.1.
Solution
We wish to detect a difference if it exists between the mean academic achievement for
non-car owners 𝜇1 and car owners 𝜇2 .
𝜇1 𝜇2
𝑛1 = 100 𝑛2 = 100
𝑋̅1 = 2.7 𝑋̅2 = 2.54
𝑠12 = 0.36 𝑠12 = 0.4
1. Hypotheses
𝐻0 : 𝜇1 − 𝜇2 = 0 vs 𝐻1 : 𝜇1 − 𝜇2 ≠ 0
This is a two tailed test.
2. Test statistic
𝑛1 and 𝑛2 are large i.e (𝑛1 and 𝑛2 ≥ 30)
𝜎12 𝜎22
∴ 𝑋̅1 − 𝑋̅2 ~𝑁 (𝜇1 − 𝜇2 , + )
𝑛1 𝑛2
(𝑋̅1 − 𝑋̅2 ) − (𝜇1 − 𝜇2 )
⇒𝑍=
𝜎12 𝜎22
√
𝑛1 + 𝑛2

(𝑋̅1 − 𝑋̅2 ) − (𝜇1 − 𝜇2 )

≃𝑍=
𝑠12 𝑠22
√
𝑛1 + 𝑛2

(2.7 − 2.54) − 0
=
√0.36 + 0.4
100 100
𝑍 ≈ 1.34

12
 3. Rejection region
It is two tailed therefore 𝛼 = 0.1, 𝛼/2 = 0.05. and this implies that
𝑍𝛼⁄2 = 𝑍0.05 = 1.645
Reject 𝐻0 if 𝑍 < −𝑍𝛼⁄2 or 𝑍 > 𝑍𝛼⁄2

-1.645 1.645

4. Conclusion
Since 𝑍 > 𝑍0.05 that is 1.84 > 1.645 we reject 𝐻0 and conclude that there is some
evidence of difference in the academic performance between the two groups.
b) Small sample case
𝐻0 : 𝜇1 − 𝜇2 = 𝐷0 or 𝐻0 : 𝜇1 = 𝜇2
One-tailed test
Right tailed Left tailed
𝐻1 : 𝜇1 − 𝜇2 > 𝐷0 𝐻1 : 𝜇1 − 𝜇2 < 𝐷0
Two tailed
𝐻1 : 𝜇1 − 𝜇2 ≠ 𝐷0
Test statistic
(𝑋̅1 − 𝑋̅2 ) − (𝜇1 − 𝜇2 )
𝑡=
1 1
√𝑆 2 ( + )
𝑛1 𝑛2
(𝑛1 −1)𝑠12 +(𝑛2 −1)𝑠22
Where 𝑠 2 = 𝑛1 +𝑛2 −2

Rejection region
One tailed test
Reject 𝐻0 if
𝑡 > 𝑡𝛼,𝑛1 +𝑛2 −2 (right tailed)

13
 𝑡 < −𝑡𝛼,𝑛1 +𝑛2 −2 (left tailed)
Two-tailed test
Reject 𝐻0 if 𝑡 > 𝑡𝛼,𝑛1 +𝑛2 −2 or 𝑡 < −𝑡𝛼,𝑛1 +𝑛2 −2 .

1.7 INFERENCES ABOUT POPULATION VARIANCES

We extend the discussion to inferences about population variances. Recall that the variance
plays a vital role in determining the variability of the observations from the desired mean.
Recall also that the sample variance

2
∑𝑛𝑖=1(𝑋𝑖 − 𝑋̅)2
𝑠 =
𝑛−1
Was used as the point estimator of the population variance 𝜎 2 . To make inferences about the
population variance using the sample variance, we make use of the sampling distribution of
the quantity
(𝑛 − 1)𝑠 2
χ2 =
𝜎2
therefore, whenever a sample random sample of size n is selected from a normally distributed
(𝑛−1)𝑠2
population, the sampling distribution of χ2 = follows a chi-square distribution with n-
𝜎2
(𝑛−1)𝑠2
1 degrees of freedom. Since the sampling distribution of χ2 = follows a chi-square
𝜎2

distribution, we can employ the chi-square distribution to develop interval estimates and
carry out hypothesis tests about a population.

With 2 df

With 5 df

With 10 df

0 (𝑛 − 1)𝑠 2
𝜎2

INTERVAL ESTIMATION OF 𝝈𝟐
Recall that each simple random sample generates different sample estimates. We cannot
therefore expect the sample variance to provide the exact value of the population variance.

14
Instead, we can expect the interval estimate to contain all possible values of the population
variance. Note that the notation χ2𝛼,𝑛−1𝑑𝑓 denotes the chi-square critical value that provides an
area or probability of 𝛼 to the right of the stated χ2𝛼,𝑛−1𝑑𝑓 .

Area or probability

0 χ2𝛼

2
The area to the left of χ2𝛼 is therefore χ1−𝛼 . Combining the above two critical chi-square
values, we can say that the 1 − 𝛼 confidence interval containing χ2 can be written as
2 2 2
χ1−𝛼 ⁄2,𝑛−1 ≤ χ ≤ χ𝛼⁄2,𝑛−1

(𝑛−1)𝑠2 .
But we know that χ2 = 𝜎2

Therefore, we can re-write the interval estimate of χ2 as:

2
(𝑛 − 1)𝑠 2
χ1−𝛼 ⁄2,𝑛−1 ≤ ≤ χ2𝛼⁄2,𝑛−1
𝜎2
To find the interval estimate for 𝜎 2 , we perform some algebraic manipulation on the
inequality. Performing algebraic manipulation on the leftmost inequality, we get:

2
(𝑛 − 1)𝑠 2
χ1−𝛼 ⁄2,𝑛−1 ≤
𝜎2
2
𝜎 2 χ1−𝛼 ⁄2,𝑛−1 ≤ (𝑛 − 1)𝑠
2

And this implies that:

(𝑛 − 1)𝑠 2
𝜎2 ≤ 2 … … … … … . . (𝑎)
χ1−𝛼 ⁄2,𝑛−1

Likewise, working with the rightmost inequality results into:

(𝑛 − 1)𝑠 2
≤ χ2𝛼⁄2,𝑛−1
𝜎2
(𝑛 − 1)𝑠 2 ≤ χ2𝛼⁄2,𝑛−1 𝜎 2
(𝑛 − 1)𝑠 2
⇒ ≤ 𝜎 2 … … … … … (𝑏)
χ2𝛼⁄2,𝑛−1
Combining equations (a) and (b) above, we get the interval estimate for 𝜎 2 as:

15
 (𝑛 − 1)𝑠 2 2
(𝑛 − 1)𝑠 2
≤𝜎 ≤
χ2𝛼⁄2,𝑛−1 2
χ1−𝛼 ⁄2,𝑛−1

We can therefore conclude that the respective (1 − 𝛼)100% confidence interval for the
population variance (𝜎 2 ) and population standard deviation(𝜎) are:

(𝑛 − 1)𝑠 2 2
(𝑛 − 1)𝑠 2
≤𝜎 ≤
χ2𝛼⁄2,𝑛−1 2
χ1−𝛼 ⁄2,𝑛−1

(𝑛 − 1)𝑠 2 (𝑛 − 1)𝑠 2
√ ≤𝜎≤√
χ2𝛼⁄2,𝑛−1 2
χ1−𝛼 ⁄2,𝑛−1

TEST OF HYPOTHESIS ABOUT A POPULATION VARIANCE

In some situations, it may be necessary to conduct a hypothesis test about the variance of the
population. Suppose 𝑋~𝑁(𝜇, 𝜎 2 )
Null hypothesis
𝐻0 : 𝜎 2 = 𝜎02
One tailed test about 𝜎 2
Right tailed Left tailed
𝐻𝑎 : 𝜎 2 > 𝜎02 𝐻𝑎 : 𝜎 2 < 𝜎02
Two tailed
𝐻𝑎 : 𝜎 2 ≠ 𝜎02
Test statistic
(𝑛 − 1)𝑠 2 2
χ2 = ~χ𝑛−1
𝜎02
Rejection region
One tailed test
Reject 𝐻0 if

16
 1) χ2𝑜𝑏𝑠 > χ2𝛼,𝑛−1 (right tailed)

χ2𝛼,𝑛−1

2) χ2𝑜𝑏𝑠 < χ2𝛼,𝑛−1 (left tailed)

χ2𝛼,𝑛−1
Two tailed test
Reject 𝐻0 if
2
χ2𝑜𝑏𝑠 < χ1−𝛼/2 or χ2𝑜𝑏𝑠 > χ2𝛼/2 or reject 𝐻0 if p-value< 𝛼.

INFERENCES ABOUT THE VARIANCES OF TWO POPULATIONS: TWO

INDEPENDENT RANDOM SAMPLES
In some situations, the researcher may wish to compare the variances of two populations. To
make inferences about two population variances 𝜎12 and 𝜎22 , two respective sample variances
𝑠12 and 𝑠22 can be used.
Note that whenever the variances of the two normal populations from which the samples are
𝑠2
selected are equal (i.e 𝜎12 = 𝜎22 ), the test statistic 𝐹 = 𝑠12 can be used to make inferences about
2

the difference in the two population variances (𝜎12 − 𝜎22 ). We can therefore conclude that
whenever independent random samples of size 𝑛1 and 𝑛2 are chosen from two normal
𝑠2
populations with equal variances, the sampling distribution of 𝐹 = 𝑠12 has an F-distribution
2

with 𝑛1 − 1 degrees of freedom in the numerator and 𝑛2 − 1 degrees of freedom for the
denominator, with 𝑠12 from a random sample of size 𝑛1 from population 1 and with 𝑠22 from a
random sample of size 𝑛2 from population 2.

17
Note that 𝐹𝛼,𝑛1 −1,𝑛2 −1 is the critical value for the F-distribution that gives an area or
probability of 𝛼 to the right of the stated F-value.

0 𝐹𝛼,𝑛1 −1,𝑛2 −1

Below is the summary of test procedures on differences between two population variances.
Note that in hypothesis tests with 𝐻0 : 𝜎12 = 𝜎22 , the population with a larger variance is
denoted as population 1 so that we are guaranteed that the rejection of 𝐻0 will only occur in
the upper tail of the F-distribution. Although deciding to denote the population with the larger
variance as population 1 is not common in practice, deciding which population to denote as
population 1 or population 2 is necessary.
Summary
We can compute the lower tail.
F-value (𝐹1−𝛼,𝑑𝑓1 ,𝑑𝑓2 ) using the upper tail
F-value (𝐹𝛼,𝑑𝑓1 ,𝑑𝑓2 ) using the following relationship:
1
𝐹(1−𝛼),𝑑𝑓1 ,𝑑𝑓2 =
𝐹𝛼,𝑑𝑓1 ,𝑑𝑓2
That is, 𝐹(1−𝛼),𝑑𝑓1 ,𝑑𝑓2 is the reciprocal of the 𝐹𝛼 value with 𝑑𝑓1 in the numerator and 𝑑𝑓2 in the
denominator.

Differences Between Two Variances

Hypothesis Test statistic Critical value or rejection
Two tailed
𝑠2
𝐻0 : 𝜎12 = 𝜎22 𝐹 = 𝑠12 𝐹𝑜𝑏𝑠 > 𝐹𝛼/2,𝑛1 −1,𝑛2 −1 or
2

𝐻1 : 𝜎12 ≠ 𝜎22 𝐹𝑜𝑏𝑠 < 𝐹1−𝛼/2,𝑛1 −1,𝑛2 −1

Or if p-value< 𝛼
Right tailed
𝑠2
𝐻0 : 𝜎12 = 𝜎22 𝐹 = 𝑠12 𝐹𝑜𝑏𝑠 > 𝐹𝛼/2,𝑛1 −1,𝑛2 −1 or
2

18
𝐻1 : 𝜎12 > 𝜎22 Or if p-value< 𝛼
Left tailed
𝑠2
𝐻0 : 𝜎12 = 𝜎22 𝐹 = 𝑠12 𝐹𝑜𝑏𝑠 < 𝐹1−𝛼/2,𝑛1 −1,𝑛2 −1 or
2

𝐻1 : 𝜎12 < 𝜎22 Or if p-value< 𝛼

1.8 Formative Evaluation

1. With proper statistical intuition distinguish between Type I and Type II errors in
hypothesis testing.
2. Define significance level of the test and power of the test. Clearly, explain how these
two concepts relate to Type I and Type II errors in hypothesis testing.
3. A random sample of 16 values, 𝑋1 , 𝑋2 , 𝑋3 , … , 𝑋16 , was drawn from a normal
distribution and gave the following summary statistics:
∑16 16 2
𝑖=1 𝑋𝑖 = 51.2, ∑𝑖=1 𝑋𝑖 = 243.19

Calculate a 95% confidence interval for the population mean.

4. In 1989, the Johannesburg Stock Exchange (JSE) boomed, and the Allshare Index
showed an annual return of 55.5%. A sample of industrial shares yielded the
following results:
54.5 47.8 42.3 59.8 33.1 49.7 16.0 50.3
52.8 56.1 23.2 52.5 65.7 47.5 32.5 46.7
Test at the 5% significance level, whether the performance of industrial shares lagged
behind the market as a whole.
5. Two training procedures are compared by measuring the time that it takes trainees
to assemble a device. A different group of trainees are taught using each method. Is
there a difference in the two methods? Use 𝛼 = 0.01.

Time to Assemble Method I Method II

Sample size 10 12

Sample mean 35 31

19
 Sample Std 4.9 4.5
Deviation

1.9 Unit Summary

This unit defined and discussed type I and type II error, and thereafter conducted hypothesis
tests for small sample as well as large samples. The unit concluded by considering inferences
about a population variance.

20