Lecture 9

SAMPLING DISTRIBUTIONS CHI -SQ AND F-DIST

INFERENCES CONCERNING THE VARIANCE

Spring 2021 DR. MAHA A. HASSANEIN 1

Study Outline
Sampling Distribution of the Variance
◦ Chi-squared
◦ F-distribution

Inferences Concerning Variances

◦ Estimation of Variances
◦ Hypothesis Concerning one Variance
◦ Hypothesis concerning two variances

Spring 2021 DR. MAHA A. HASSANEIN 2

Sampling Distribution of the
Variance
Theorem 6.5
If 𝑆 2 is the variance of a random sample of size n taken from
a normal population having variance 𝜎 2 , then
2 σ 𝑛 ത 2
(𝑛 − 1)𝑆 𝑖=1 𝑋𝑖 − 𝑋
𝜒2 = 2
=
𝜎 𝜎2
Is a random variable having the Chi-squared distribution
with the degrees of freedom (df) 𝜈 = 𝑛 − 1.
The sample variance (𝑆 2 ) is an estimate of the population
variance (𝜎 2 ).

Spring 2021 DR. MAHA A. HASSANEIN 3

 Chi-Squared Distribution

Spring 2021 DR. MAHA A. HASSANEIN 4

Table: Chi-squared Distribution
The 𝜒 2 distribution is already normalized in its definition through
including s in its definition.
Therefore no z-transforms are needed and we can work directly with a
table that gives right tail areas under the 𝜒 2 distribution.
That table is the Chi-squared Distribution Table, in the Appendix, and it
gives values of 𝜒 2 for given values of area to the right of 𝜒 2
We’ll need \chi^2_{\rm left} and \chi^2_{\rm right} such that the tail
areas are equal and such that the area between them is \cal{C}

Spring 2021 DR. MAHA A. HASSANEIN 5

Table for Chi-squared

Spring 2021 DR. MAHA A. HASSANEIN 6

Example 1
Optical frim purchase glass with variance of refractive index
1.26 × 10−4 . The firm rejects shipment if sample variance of 20
pieces exceeds 2.00 × 10−4 . Assume sample values from normal
population.What is the probability that a shipment will be rejected?
Solution
𝜎 2 = 1.26 × 10−4 𝑎𝑛𝑑 𝑠 2 = 2.00 × 10−4 . 𝑤𝑖𝑡ℎ 𝜈 = 𝑛 − 1 = 19
−4
19 . 2.00 × 10
𝜒2 = −4
= 30.2
1.26 × 10
2
From Table chi-sq: 𝜒0.05 = 30.1 .
Thus, the probability that the sample variance exceeds
2.00 × 10−4 𝑖𝑠 𝑙𝑒𝑠𝑠 𝑡ℎ𝑎𝑛 0.05.

Spring 2021 DR. MAHA A. HASSANEIN 7

Spring 2021 DR. MAHA A. HASSANEIN 8
F-Distribution
The F-distribution results when you build up a distribution of the ratio
of the two 𝜒 2 sample values.
𝑆12
𝐹 = 2 , 𝑆1 > 𝑆2
𝑆2
The ratio follows an F-distribution if 𝜎1 = 𝜎2 .
That F distribution has two degrees of freedom:
one for the numerator 𝜈1 = 𝑛1 − 1
one for the denominator 𝜈2 = 𝑛2 − 1.

So we denote the distribution more specifically as 𝐹𝜈1 ,𝜈2

Spring 2021 DR. MAHA A. HASSANEIN 9

The exact shape of the 𝐹𝜈1,𝜈2 distribution depends on the choice of
𝜈1 𝑎𝑛𝑑 𝜈2 , But it roughly looks like a 𝜒 2 distribution

F and t are related such that 𝐹1,𝜈2 = 𝑡 2

Spring 2021 DR. MAHA A. HASSANEIN 10

Difference between Two
Variances – the F Distributions
Assume that the two populations have a distribution that is almost
normal

Spring 2021 DR. MAHA A. HASSANEIN 11

Two normal populations lead to two 𝜒 2 distributions that represent
distributions of sample variances.

Spring 2021 DR. MAHA A. HASSANEIN 12

Sampling Distribution of the
ratio of Variances

Spring 2021 DR. MAHA A. HASSANEIN 13

Tables for F-Distribution

Spring 2021 DR. MAHA A. HASSANEIN 14

Spring 2021 DR. MAHA A. HASSANEIN 15
Example 2
If two independent random samples of size 𝑛1 = 7 𝑎𝑛𝑑 𝑛2
= 13 taken from normal population. What is the probability that
the variance of the first sample will be at least three times as large
as that of the second sample.
Solution
𝑆12 = 3𝑆22 𝑤𝑖𝑡ℎ 𝜈1 = 𝑛1 − 1 = 6 and 𝜈2 = 𝑛2 − 1 = 12
𝑆12
𝐹= 2=3
𝑆2
From Table F-dist: 𝐹0.05 = 3 .
Thus, the probability that the variance of the first sample will be at
least three times as large as that of the second sample is 0.05.

Spring 2021 DR. MAHA A. HASSANEIN 16

Tables for F-Distribution

Spring 2021 DR. MAHA A. HASSANEIN 17

Spring 2021 DR. MAHA A. HASSANEIN 18
Important Property of F-Distribution
1
𝐹1−𝛼 𝜈1 , 𝜈2 =
𝐹𝛼 𝜈2 , 𝜈1

Ex. Find 𝐹0.95 12,6

Since 𝐹0.05 6,12 = 3 , then
1 1
𝐹0.95 12,6 = =
𝐹0.05 6,12 3

Spring 2021 DR. MAHA A. HASSANEIN 19

The Estimation of Variances
The sample variance
σ 𝑛 ത 2
𝑖=1 𝑋 𝑖 − 𝑋
𝑆2 =
𝑛−1
Is an unbiased estimator of 𝜎 2 . ( point Estimate)

Spring 2021 DR. MAHA A. HASSANEIN 20

 The 𝛼 in the Chi-squared
Distribution Table

𝛼/2
2 2 1-𝛼
𝜒right = 𝜒𝛼/2 𝛼/2
2
𝜒𝑙𝑒𝑓𝑡 = 𝜒2 𝛼
(1− )
2

let 𝜒𝛼2 be the table value that corresponds to 𝛼. In other words 𝜒𝛼2
is the 𝜒 2 value that corresponds to a right tail area of 𝛼.
Spring 2021 DR. MAHA A. HASSANEIN 21
 2
Confidence Interval for 𝜎
For Confidence level (1-𝛼), Finally, the confidence
interval for 𝜎 2 is given by

𝑛 − 1 𝑠2 2 <
𝑛 − 1 𝑠 2

2 < 𝜎 2
𝜒α/2 𝜒1−𝛼/2

where the 𝜒 2 distribution with 𝜈 = 𝑛 − 1 degrees

of freedom.

Spring 2021 DR. MAHA A. HASSANEIN 22

Example 3
Find the 90% confidence interval on 𝜎 2 for the following data
𝟓𝟗, 𝟓𝟒, 𝟓𝟑, 𝟓𝟐, 𝟓𝟏, 𝟑𝟗, 𝟒𝟗, 𝟒𝟔, 𝟒𝟗, 𝟒𝟖
Solution
𝑠 2 = 28.2 𝑎𝑛𝑑 𝜈 = 𝑛 − 1 = 9
2 2 2
𝜒right = 𝜒𝛼/2 = 𝜒0.05 = 16.919
2 2 2
𝜒𝑙𝑒𝑓𝑡 = 𝜒(1−𝛼 = 𝜒0.95 = 3.325
)2
9 . 28.2 2
9 28.2
< 𝜎 <
16.919 3.325
The confidence interval for the variance
15.0 < 𝜎 2 < 76.3

Spring 2021 DR. MAHA A. HASSANEIN 23

 1 - 𝜶/𝟐 𝜶/𝟐

Spring 2021 DR. MAHA A. HASSANEIN 24

Hypothesis Concerning One
Variance
Statistic for test concerning variance ( normal population)
𝑛 − 1 𝑆 2
𝜒2 =
𝜎2

Alternative Hypothesis Critical Region

𝜎 2 < 𝜎02 2
𝜒 2 < 𝜒1−𝛼

𝜎 2 > 𝜎02 𝜒 2 > 𝜒𝛼2

𝜎 2 ≠ 𝜎02 2
𝜒 2 < 𝜒1−𝛼/2
2
𝜒 2 > 𝜒𝛼/2

Spring 2021 DR. MAHA A. HASSANEIN 25

 Example 4
Thickness of silicon wafers is acceptable if 𝜎, the population standard deviation of
thickness is at most 0.5 mil. Use the 0.05 level of significance to test the null
hypothesis 𝜎 = 0.5 against the alternative hypothesis𝜎 > 0.5, if thickness of 15
wafers cut have a standard deviation of 0.64 mil
Solution.
𝐻0 : 𝜎 = 0.5
𝐻1 : 𝜎 > 0.5
2
𝛼 = 0.05, 𝜒0.05 = 23.685
Statistic for test concerning variance ( normal population)
2
𝑛 − 1 𝑆 2 14. 0.642
𝜒 = = = 22.94 ≯ 23.685
𝜎2 0.52
Fail to reject the null hypothesis, there is not sufficient evidence to conclude 𝜎 > 0.5

Spring 2021 DR. MAHA A. HASSANEIN 26

Spring 2021 DR. MAHA A. HASSANEIN 27
Hypothesis Concerning two
variances
Critical region for testing 𝜎12 = 𝜎22 ( normal populations)

Alternative Test Statistic Critical Region

Hypothesis
𝜎12 < 𝜎22 𝐹 = 𝑆2 2 /𝑆12 𝐹 > 𝐹𝛼 (𝑛2 − 1, 𝑛1 − 1)

𝜎12 > 𝜎22 𝐹 = 𝑆1 2 /𝑆22 𝐹 > 𝐹𝛼 (𝑛1 − 1, 𝑛2 − 1)

𝜎12 ≠ 𝜎22 𝐹 = 𝑆𝑀 2 /𝑆𝑚

2
𝐹 > 𝐹𝛼/2 (𝑛𝑀 − 1, 𝑛𝑚 − 1)

Spring 2021 DR. MAHA A. HASSANEIN 28

 Example
To determine if there is less variability in the silver plating by company 1 than in
that of company 2. if independent random samples of size 12 of the two
companies yield 𝑠1 = 0.035 𝑚𝑖𝑙 𝑎𝑛𝑑 𝑠2 = 0.062 𝑚𝑖𝑙, test the null hypothesis 𝜎12
= 𝜎22 ( normal populations) against the alternative hypothesis 𝜎12 < 𝜎22 at the 0.05
level of significance.
Solution.
𝐻0 : 𝜎12 = 𝜎22
𝐻1 : 𝜎12 < 𝜎22
𝛼 = 0.05, 𝐹0.05 = 2.82 ( 𝑏𝑦 𝑖𝑛𝑡𝑒𝑟𝑝𝑜𝑙𝑎𝑡𝑖𝑜𝑛)
𝑆22 (0.062)2
Statistic for test ∶ 𝐹 = = = 3.14 > 2.82
𝑆12 (0.035)2

Reject the null hypothesis, the data support that company 1 is less variable than
company 2.

Spring 2021 DR. MAHA A. HASSANEIN 29

Spring 2021 DR. MAHA A. HASSANEIN 30
 Text book
Chapter 9. sec 9.1-9.3

Reference

Spring 2021 DR. MAHA A. HASSANEIN 31