Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variance

Examples: Measure of Location Vs Dispersion

Marks scored by a student

Investment Options

Profit of a company

Productivity/Efficiency: Time taken to complete a task

 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variance

 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variance

Relevance of variance: Measure of consistency, risk

1- Population variance

Sample variance, s2 is a point estimate of population variance, σ2

When a sample is selected from a population that follows a normal distribution the following
ratio follows a chi-square (ꭓ2 ) distribution with n-1 degrees of freedom.

(n-1) s2 / σ2

Chi-square (ꭓ2 ) distribution: Skewed distribution

 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variance

1- Population variance

Chi-square (ꭓ2 ) distribution: Skewed distribution

 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variance

1- Population variance: Chi-square (ꭓ2 ) distribution table
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variance

Interval Estimate:
χ2α : Denotes that chi-square value for which the area to the right is α
At 1- α confidence level and n-1 degrees of freedom

(𝐧 − 𝟏)𝐬 𝟐 𝟐
(𝐧 − 𝟏)𝐬 𝟐

𝟐
≤ 𝛔 ≤ 𝟐
𝛘𝛂/𝟐 𝛘(𝟏−𝛂/𝟐)

(𝐧−𝟏)𝐬 𝟐
Lower confidence limit:
𝛘𝟐𝛂/𝟐

(𝐧−𝟏)𝐬 𝟐
Upper confidence limit:
𝛘𝟐(𝟏−𝛂/𝟐)
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variances

Interval Estimate: e.g. 95% confidence interval

(n−1)s2
Lower confidence limit: 2
χα/2

(n−1)s2
Upper confidence limit: 2
χ(1−α/2)
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variance

Problem: Create a confidence interval for population variance of soft drink discharged from a
vending machine, when a sample of 20 fills were taken and it was found that mean discharge
volume was 2 litres with a variance of 0.05 litre squared. Use 95% C.L.

df = n-1 = 19; α = 5%

χ2α/2 = ?

χ2(1−α/2) = ?
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variance

Problem: Create a confidence interval for population variance of soft drink discharged from a
vending machine. A sample of 20 fills were taken and it was found that mean discharge volume
was 2 litre with a variance of 0.05 litre squared. Use 95% C.L.

df = n-1 = 19; χ2α/2 = 32.852 ; χ2(1−α/2) = 8.907

(n−1)s2
Upper confidence limit: 2 = 19* 0.05 / 8.907 = 0.1067
χ(1−α/2)

(n−1)s2
Lower confidence limit: 2 = 19* 0.05 / 32.852 = 0.0289
χα/2
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variance

Types of Hypotheses:

1. H0 : σ2 >= k
H1 : σ2 < k

2. H0 : σ2 <= k
H1 : σ2 > k

3. H0 : σ2 = k
H1 : σ2 ≠ k
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variance

(n−1)s2 (n−1)s2
Test statistic, χ2 − calc = =
σ2H0 k

This is compared with χ2 - critical (χ2 df, α OR χ2 df, 1-α) and then inference is made about the
hypothesis.
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variance

Problem1: Production manager of a firm, that manufactures polyester films, thinks that in
recent years due to supply chain uncertainty, the variance of weekly supply time of one of its
major suppliers has increased from 4 days squared in the past. To test this hunch, she randomly
takes a sample of supply data of 30 weeks from past and finds that the variance in this sample
is 5 days squared. What can she infer at 5% significance level?
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variance

Problem1: Production manager of a firm that manufactures polyester films thinks that in recent
years due to supply chain uncertainty, the variance of weekly supply time of one of its major
suppliers has increased from 4 days squared in the past. To test this hunch, she randomly takes
a sample of supply data of 30 weeks from past data and finds that the variance from this
sample is 5 days squared. What can she infer at 5% significance level?

. H0 : σ2 <= 4
H1 : σ2 > 4

n = 30; s2 = 5

χ2 - critical = χ2 df, α = χ2 29, 0.05

 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variance

Problem1: Production manager of a firm that manufactures polyester films thinks that in recent
years due to supply chain uncertainty, the variance of weekly supply time of one of its major
suppliers has increased from 4 days squared in the past. To test this hunch, she randomly takes
a sample of supply data of 30 weeks from past data and finds that the variance form this
sample is 5 days squared. What can she infer at 5% significance level?

. H0 : σ2 <= 4
H1 : σ2 > 4

n = 30; s2 = 5
χ2 - critical = χ2 df, α = χ2 29, 0.05 = 42.557
χ2 - calc = 29*5/4 = 36.25
Since χ2 - calc < χ2 - critical , we accept H0
And infer that …
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variance

Problem2: Machine operator at a ball bearings manufacturing company suspects that due to
changes in settings by new workers, the variability of bearings (measured using the diameter)
manufactured from a given machine has changed from 2 mm squared. To verify this opinion he
collects a sample of 20 bearings, measures their diameter and notes that the standard
deviation of the diameter is now 1.8 mm. What inference can he draw at 90 % confidence
level?
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variance

Problem2: Machine operator at a ball bearings manufacturing company suspects that due to
changes in settings by new workers, the variability of bearings (measured using the diameter)
manufactured from a given machine has changed from 2 mm squared. To verify this opinion he
collects a sample of 20 bearings, measures their diameter and notes that the standard
deviation of the diameter is now 1.8 mm. What inference can he draw at 90 % confidence
level?

H0 : σ2 = 2
H1 : σ2 ≠ 2

n = 20; s2 = 3.24
χ2 - critical values:
χ2 − critical lower = χ2 df, 1−α/2 ; χ2 − critical upper = χ2 df, α/2
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variance

H0 : σ2 = 2
H1 : σ2 ≠ 2

n = 20; s2 = 3.24
χ2 − critical lower = χ2 df, 1−α/2 = 10.117 ; χ2 − critical upper = χ2 df, α/2 = 30.144

χ2 - calc = 19*3.24/2 = 30.78

Since χ2 - calc > χ2 - critical upper ,

we reject H0 and infer that…
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variances

2- Population variances
Types of Hypotheses:

1. H0 : σ12 − σ22 >= 0

H1 : σ12 − σ22 < 0

2. H0 : σ12 − σ22 <= 0

H1 : σ12 − σ22 > 0

3. H0 : σ12 −σ22 = 0
H1 : σ12 −σ22 ≠ 0
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variances

2- Population variances
When samples are selected from two populations that follow the normal distribution, the ratio
of sample variances s12 / s22 follows a skewed distribution called as F-distribution.

Test statistic, F – calc = s12 / s22

This is compared with F-critical (Fndf, ddf, α) to accept/reject the null hypothesis,
where ndf is numerator degrees of freedom (n1-1) and ddf is denominator degrees of freedom
(n2-1) and α is significance level.

Note 1: The sample variance which is higher is treated as sample taken from population one (in
which case even the lower tail test can be converted to upper tail test).
Note 2: For a two tailed test, F-calc is compared with F-critical upper to accept/reject the null
hypothesis
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variances

2- Population variances

F-distribution: Skewed Distribution

 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variances

2- Population variances: F-distribution table
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variances

2- Population variances: F-distribution table
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variances

2- Population variances: F-distribution table
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variances

2- Population variances: F-distribution table
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variances

2- Population variances

Problem1: A manufacturer believes that one of it two suppliers has been inconsistent in recent
times and therefore this supplier’s weekly supply time variability has increased when compared
to the other supplier. He studies a data of supply time variations from a random sample of 21
weeks for each of these suppliers and finds that the standard deviation of supply time of
supplier 1 is 3 days as compared to 2 days for supplier 2. What can be inferred at 5%
significance level?
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variances

2- Population variances

Problem1: A manufacturer believes that one of it two suppliers has been inconsistent in recent
times and therefore this supplier’s weekly supply time variability has increased when compared
to the other suppliers. He studies a data of supply time variations from a random sample of 21
weeks for each of these suppliers and finds that the standard deviation of supply time of
supplier 1 is 3 days as compared to 2 days for supplier 2. What can be inferred at 5%
significance level?

s12 = 9 ; s22 = 4; n1 = n2 = 21

H0 : σ12 − σ22 <= 0

H1 : σ12 − σ22 > 0
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variances

2- Population variances

s12 = 9 ; s22 = 4; n1 = n2 = 21

H0 : σ12 − σ22 <= 0

H1 : σ12 − σ22 > 0

Test statistics, F – calc = s12 / s22 = 9/4 = 2.25

This is compared with F-critical (F20, 20, 0.05) = 2.12

Since F – calc > F-critical we reject H0 and infer that….

 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variances

2- Population variances

Problem2: A manufacturer believes that weekly supply time variability of it two suppliers is
different. He studies a data of supply time variations from a random sample of 21 weeks for
each of these suppliers and finds that the supply time variability of supplier 1 is 3 days squared
as compared to 2 days squared for supplier 2. What can be inferred at 5% significance level?
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variances

2- Population variances

Problem2: A manufacturer believes that weekly supply time variability of it two suppliers is
different. He studies a data of supply time variations from a random sample of 21 weeks for
each of these suppliers and finds that the supply time variability of supplier 1 is 3 days squared
as compared to 2 days squared for supplier 2. What can be inferred at 5% significance level?

H0 : σ12 −σ22 = 0
H1 : σ12 −σ22 ≠ 0

s12 = 3 ; s22 = 2 ; n1 = n2 = 21
 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variances

2- Population variances

H0 : σ12 −σ22 = 0
H1 : σ12 −σ22 ≠ 0

s12 = 3 ; s22 = 2 ; n1 = n2 = 21

Test statistics, F – calc = s12 / s22 = 1.5

F-Critical Lower = F20, 20, 0.975 = ?

F-Critical Upper = F20, 20, 0.025 = 2.46

Since F – calc < F-critical Upper, we accept H0 and infer that ….

 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variances

2- Population variances

Problem3: A bottling plant manager believes that the variability of one of its machine (m1) is
less than that of second machine (m2). He randomly collects a sample of 26 bottles from
machine1 and 21 bottles from machine 2 and realizes that the variance of the liquid filled in a 2
litre bottle in m1 is 1.5 ml squared and that in a 2 litre bottle in m2 is 2 ml squared. What can
the manager infer at 1% significance level?

s12 = 1.5 ; s22 = 2; n1 = 26, n2 = 21

H0 : σ12 − σ22 >= 0

H1 : σ12 − σ22 < 0

Since, s12 < s22 here, we make population 1 as population 2 so that

 Vishal Mishra (IBS, Hyderabad)

Hypothesis Testing – Variances

2- Population variances

Problem3: A bottling plant manager believes that the variability of one of its machine (m1) is
less than that of second machine (m2). He randomly collects a sample of 26 bottles from
machine1 and 21 bottles from machine 2 and realizes that the variance of the liquid filled in a 2
litre bottle in m1 is 1.5 ml squared and that in a 2 litre bottle in m2 is 2 ml squared. What can
the manager infer at 1% significance level?

s12 = 2 ; s22 = 1.5; n1 = 21, n2 = 26

H0 : σ12 − σ22 <= 0

H1 : σ12 − σ22 > 0

Thereby converting a lower tail test to upper tail test. Now we can proceed as usual.