Chapter 15

Inference about population variances

Chapter outline

15.1 Inference about σ2

15.2 Inference about σ12/σ22
Learning objectives

LO1 Understand the chi-squared distribution and the

F-distribution
LO2 Identify the sampling distribution of the sample
variance and derive an interval estimate for the
population variance
LO3 Test hypotheses regarding population variance
LO4 Identify the sampling distribution of the ratio of
two sample variances and derive an interval
estimate for the ratio of two population variances
LO5 Test hypotheses comparing two population
variances.
 15.5

Introduction
In this chapter we make inferences about population
variance (s) by utilising the approach developed previously
for making statistical inference about population
parameters such as population mean μ and population
proportion p.

As before, this can be achieved in 3 steps:

Step 1: Identify the parameter to be estimated or tested.
Step 2: Specify the parameter’s estimator and its sampling
distribution.
Step 3: Construct an interval estimator or perform a test.
 15.6

Inference about population variances

For a single population, to draw inferences about
variability, the parameter of interest is the population
variance σ2.
To compare the variability of two populations, the
parameter of interest is the ratio of the two variances
σ12/σ22.
Inference about σ12/σ22 is important as, noted earlier,
the inference about the population mean difference
μ1 - μ2 depends on whether the variances are equal
(σ12 = σ22) or unequal (σ12 ≠ σ22).
 15.7

15.1 Inference about 2

If we are interested in drawing inferences about a normal
population’s variability, the parameter we need to
investigate is the population variance σ2.
The sample variance s2 is an unbiased, consistent and
efficient point estimator for σ2.
Moreover, the statistic
(n − 1) s 2
2 =
2
has a distribution called Chi-squared with d.f. = n – 1, if the
population is normally distributed.
 15.8

15.1 Inference about 2…

(n − 1) s 2
2 = d. f . = = n −1
2
 15.9
The 2 table
A =0.01

d.f. = 10
A =0.01

1-A 
.990 .010

01,10 = 23.2093
Degrees of
freedom      

1 0.0000393 0.0001571 0.0009821 . . 6.6349 7.87944
.
.
10 2.15585 2.55821 3.24697 . . 23.2093 25.1882
. . . . . .
. . . . . . . .
 15.10

Estimating the population variance σ2

 15.11

Estimating the population variance σ2

From the following probability statement

P(21-/2 < 2 < 2/2) = 1 – 

we have (by substituting 2 = [(n – 1)s2]/σ2)

(n -1)s 2
(n -1)s 2
<s <
2

c a /2,n-1
2
c 2
1-a /2,n-1
 15.12

Estimating the population variance 2

The confidence interval for σ2 is

(n − 1) s 2
Lower confidence limit = LCL =
2 /2,n−1
(n − 1) s 2
Upper confidence limit = UCL =
12− /2,n−1

where (1 – α) is the confidence level.

 15.13

Testing the population variance 2

Our hypotheses:
Null hypothesis H0: 2 = 02
Alternative hypothesis is then
HA: 2  02 or H A :  2 > 0 2 or H A : 2 <  0 2
Test statistic is
(n − 1)s 2
2 = .
 2

The test statistic has a 2 distribution with d.f. = n – 1.

 15.14

Identifying factors
 15.15

Example 1 – Consistency of fills from a

container filling machine
(Example 15.2, p575)

A container-filling machine is believed to fill 1-litre

containers so consistently that the variance of the filling
will be less than 1 cc (0.001 litre). A random sample of 25
1-litre fills was taken and the results recorded. The data
are provided in file XM15-02.
a. Construct a 95% confidence interval for the population
variance 2.
b. Do these data support the belief that the variance is
less than 1cc at a 5% significance level?
 15.16

Example 1(a) – Solution

Identifying the technique
Problem objective: To describe the population of 1-litre fills
from a filling machine.
Data type: The data are numerical
Parameter of interest: Variability of the fills 2.
Assume that the population is normally distributed.
The (1-)100% confidence interval for σ2 is

(n − 1) s 2 (n − 1) s 2
LCL = UCL =
 /2,n−1
2
12− /2,n−1
 15.17

Example 1(a) – Solution

Solving manually
Note that (n – 1)s2 = xi2 – (xi)2/n

From the sample (data is presented in units of cc-1000 to

avoid rounding) we can calculate

xi = 24,996.4 and xi2 = 24,992,821.3, n=25.

Then (n – 1)s2 = 24 992 821.3 – ((24996.4)2/25) = 20.8.

95% confidence interval for 2

(n − 1)s 2 20.8 (n − 1)s 2 20.8

LCL = = = 0.528 and UCL = = = 1.677.
0.025
2
,24 39.3641 0.975
2
,24 12.4011
 15.18

Example 1(b) – Solution

Solving manually
1. The hypotheses:
H0: 2 = 1
HA: 2 < 1
(n − 1) s 2
2. The test statistic is  =
2
.
 2

3. Level of significance: α = 0.05

4. Decision rule: Reject H0 if   1− , n−1 = 0.95, 25−1 = 13.8484
2 2 2

(or Reject H0 if p-value <  = 0.05)

5. Value of the test statistic:

(n − 1)s 2 20.8
 =
2
= = 20.8
 2
1 2
 15.19

Example 1(b) – Solution

 = 0.05 1 -  = 0.95

Rejection
region
 2  13.8484
13.8484 20.8 2
0.95,
2
25−1

6. Conclusion: Since 2 = 20.8 > 13.8484, do not reject the

null hypothesis.
There is insufficient evidence to support the hypothesis that the
variance is less than 1cc.
 15.20

Example 1(b) – Solution

Using Excel (Data Analysis Plus)
 15.21

Example 1(b) – Solution

Using Excel (Data Analysis Plus)
In the Data Analysis Plus dialogue box (shown below),
enter the input and the output is presented in the next
slide.
 15.22

Example 1(b) – Solution

Using Excel (Data Analysis Plus)
The Excel output is presented below.

Conclusion: As p-value =
0.3484 > 0.05 = , we do
not reject Ho (2 = 1).

There is not enough evidence to support the hypothesis that the

variance is less than 1cc.
 15.23

Checking the required condition

The fills do not appear to be extremely non-normal,

which supports the validity of the conclusions drawn in
Example 1.
 15.24

15.2 Inference about 12/22

In this section we discuss how to compare the variability
of two normal populations.
In particular, we draw inference about the ratio of two
population variances.
This question is interesting because:
• Variances can be used to evaluate the consistency of
processes.
• The relationships between variances determine the
technique to be used to test the difference between
two population means (μ1-μ2).
 15.25

Inference about 12/22

Point estimator of 12/22
• Recall that s2 is an unbiased estimator of 2.
• Therefore, it is not surprising that we estimate
12/22 by s12/s22.
Sampling distribution of s12/s22
• We have [s12/12] is distributed as 2(1) and [s22/22]
is distributed as 2(2). Therefore the ratio F =
[s12/12]/[s22/22] is distributed as an F(1,2)
distribution, where 1=n1-1, 2=n2-1.
• The test statistic for 12/22, F = [s12/12]/[s22/22],
follows an F- distribution.
 15.26

F-distribution
 15.27

F-distribution…
 15.28

Reading F-values from the F-table

Example:
F0.05, 4, 8 = 3.84
F0.05, 8, 4 = 6.04

Note: The table

gives only right-
tail values.
 15.29

Reading F-values from the F-table

F-tables provide only right-tail values. If we need left-
tail values, we use the transformation:
1
F1-A,v1,v2 =
FA,v2 ,v1

For example,

1 1
F0.95,4,8 = = = 0.166
F0.05,8,4 6.04
 15.30

Estimating the ratio of two population

variances
From the following probability statement:
P(F1-(/2) < F < F /2) = 1 – 
 15.31

Estimating the ratio of two population

variances
By substituting the statistic F = [s12/12]/[s22/22] we can
isolate σ12/σ22 and build the following interval estimator:

 s12  1  2  s 2 
   1
  1 F / 2, ,
 s2  F  2  s2  2 1
 2  / 2 ,  ,
1 2 2  2
where 1 = n1 − 1 and  2 = n2 − 1

Here we have also used that: 1

= F /2,v2 ,v1
F1−( /2),v1,v2
 15.32

Estimating the ratio of two population

variances
 15.33

Factors that identify…

 15.34

Example 2

Two independent samples drawn from two normal

populations provide the following information:
n1 = 10 s12= 20376.2
n2 = 20 s12= 214004
Determine the 90% confidence interval estimate of the
ratio of the two population variances.
 15.35

Example 2 – Solution
We find F/2,v1,v2 = F0.05,9,19 = 2.42 and F/2,v2,v1 = F0.05,19,9 = 2.94

LCL = (s12/s22)[1/F/2,v1,v2 ]
= (s12/s22)[1/F0.05,9,19 ]
= (20376.2/214004)[1/2.42] = 0.04
UCL = (s12/s22)[ F/2,v2,v1 ]
= (s12/s22)[ F0.05,19,9 ]
= (20376.2/214004)[2.94] = 0.28
95% confidence interval estimate for 12/22 is [0.04,0.28].
 15.36

Testing the equality of two population

variances
Our null hypothesis is always
H0: 12/22 = 1
Under this null hypothesis, the F-test statistic
s12/12
F=
s22/22
becomes
s12
F= d.f.: v1 = n1-1, v2 = n2-1,
s22
 15.37

Testing the equality of two population

variances…
In comparing two population variances, we will almost always
test the null hypothesis specifying that the population
variances are equal, σ12 = σ22. That is,
H0: σ12/σ22 = 1

As was the case in all other tests, we can formulate any of the
three possible alternative hypotheses and the corresponding
decision rule based on an F-test.
1 HA: σ12/σ22 ≠ 1, where the rejection region is F > F/2
or F < F1-(/2)
2 HA: σ12/σ22 > 1, where the rejection region is F > F
3 HA: σ12/σ22 < 1, where the rejection region is F < F1- 
 15.38

Example 3
(Example 15.5, p589)
XM15-05 In Example 14.2, we applied the unequal-
variances t-test of μ1 – μ2. We chose that test statistic
after calculating the standard deviation of the sample of
consumers of high-fibre cereal to be 142.75 and the
standard deviation of the sample of non-consumers of
high-fibre cereal to be 462.61. The difference between
the two sample standard deviations appears to indicate
that the population standard deviations (and, of course,
variances) differ. We can make this process more formal
by conducting an F-test of σ12/σ22.
 15.39

Example 3 – Solution
Solving manually
1. The null and alternative hypotheses:
H0: σ12/σ22 = 1
HA: σ12/σ22  1 s12
2. The test statistic under Ho is F = 2
s2
3. Level of significance: α = 0.05
4. Decision rule:
Reject H0 if F  F /2,1, 2 = F0.025,9,19 = 2.88
or F  F1− /2,1, 2 = F0.975,9,19 = 1/ F0.025,19,9 = 1/ 3.67 = 0.272
OR if p-value <  = 0.05
Otherwise, do not reject H0.
 15.40

Example 3 - Solution…
Solving manually
5. Value of the test statistic: F = s12/s22 = 0.0952
6. Conclusion: As F = 0.0952 < 0.272, we reject Ho in
favour of the alternative (σ12/σ22  1).

There is sufficient evidence in the data to conclude at

the 5 percent level of significance that the two
variances differ.
 15.41

Example 3 - Solution…
Using Excel (Data Analysis)
In the F-Test Two Sample for Variances dialogue box (shown
below), in Data Analysis, enter the input and the output is
presented in the next slide.
 15.42

Example 3 - Solution…
Using Excel (Data Analysis)
The Excel output is presented below.

Conclusion: As p-value =
2 0.0005 = 0.001 < 0.05
= , we reject Ho in
favour of the alternative
(σ12/σ22  1).

There is sufficient evidence in the data to conclude

at the 5 percent level of significance that the two
variances differ.