Kathmandu University

Course: MATH 208, ENVE II/II

Prepared by: Dr. Samir Shrestha

Point Estimation and Confidence Interval Estimation

Statistical Inference
Statistical inference is a process by which information from a sample data is used to draw a
conclusion about the population.

Two measure areas of Statistical Inferences

1. Parameter estimation (i) Point estimation (ii) Interval estimation

2. Hypothesis testing

Point Estimator: A point estimator of a population parameter is a single numerical value of a

statistics calculated from a sample taken from given population.
Let a population is described by a random variable which follows a probability distribution
function 𝑓𝑋 and 𝑓𝑋 is characterized by the parameter(s) 𝜃 (e.g. 𝜃 = 𝑝, 𝜇, 𝜎 2 etc. ). If 𝑋1 , 𝑋2 , … , 𝑋𝑛
are random sample of size n taken from a population, then the statistics 𝜃 = 𝑔(𝑋1 , 𝑋2 , … , 𝑋𝑛 )
corresponding to the parameter 𝜃 is called the point estimator of 𝜃. Note that the statistic 𝜃 is also
a random variable.

Examples:
𝑋1 +𝑋2 +⋯+𝑋𝑛
 Sample mean, 𝑋 = is a point estimator of the population mean 𝜇, that
𝑛
means, 𝜇 = 𝑋 .
1
 Sample variance, 𝑆 2 = 𝑛−1 𝑛
𝑖=1 𝑋𝑖 − 𝑋 2
is a point estimator of population variance 𝜎 2 ,
that means, 𝜎 2 = 𝑆 2 .
 Mean difference of two population 𝜇1 − 𝜇2 has point estimator 𝜇1 − 𝜇2 = 𝑋1 − 𝑋2

Good Estimator: (Unbiased Estimator)

An estimator 𝜃 of a population parameter 𝜃 is called the unbiased estimator if 𝐸 𝜃 = 𝜃.

Examples: Then sample mean 𝑋 and Sample variance 𝑆 2 are the unbiased estimators of
population mean 𝜇 and population variance 𝜎 2 , that means,

 𝐸 𝑋 =𝜇
 𝐸(𝑆 2 ) = 𝜎 2
1. Proof of 𝐸 𝑋 = 𝜇
Let 𝑋1 , 𝑋2 , … , 𝑋𝑛 be the sample taken from the population 𝑋 with population mean
𝜇 = 𝐸(𝑋). Then, 𝐸 𝑋𝑖 = 𝜇 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑖 = 1,2, … 𝑛. Now,
1
 1
𝐸 𝑋 = 𝐸 𝑋1 + 𝑋2 + ⋯ + 𝑋𝑛
𝑛
1
𝑜𝑟, 𝐸 𝑋 = (𝐸 𝑋1 ) + 𝐸(𝑋2 ) + ⋯ + 𝐸(𝑋𝑛 )
𝑛
1
𝑜𝑟, 𝐸 𝑋 = 𝜇 + 𝜇 + ⋯ + 𝜇 = 𝜇
𝑛
2. Proof of 𝐸(𝑆 2 ) = 𝜎 2 (Left for Exercise)

Methods to Estimate the Population Parameters

 The method of maximum likelihood (Mostly used)
 The method of moments
 Bayesian inference
Standard Error of the Estimator 𝜽 of 𝜽 (Precision of the Estimation)
Standard error of the estimator 𝜽 is defined as the standard deviation 𝜃 , that means,

𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝑒𝑟𝑟𝑜𝑟 𝑜𝑓 𝜃, σ𝜃 = Var(𝜃)

𝜎
Example: Show that the standard error of sample mean 𝑋, σX = √𝑛 .
1
Proof: 𝑉𝑎𝑟 𝑋 = 𝑛 2 𝑉𝑎𝑟(𝑋1 + 𝑋2 + ⋯ + 𝑥𝑛 )
1
𝑉𝑎𝑟 𝑋 = 2 𝑉𝑎𝑟(𝑋1 + 𝑋2 + ⋯ + 𝑥𝑛 )
𝑛
1
𝑉𝑎𝑟 𝑋 = 2 𝑉𝑎𝑟 𝑋1 + 𝑉𝑎𝑟(𝑋2 + ⋯ + 𝑉𝑎𝑟(𝑋𝑛 )]
𝑛
1
𝑉𝑎𝑟 𝑋 = 2 (𝜎 2 + 𝜎 2 + ⋯ + 𝜎 2 )
𝑛
𝑛𝜎 2
𝑉𝑎𝑟 𝑋 = 2
𝑛
𝜎2
𝑉𝑎𝑟 𝑋 = , taking square root, we get,
𝑛
𝜎
σX = √𝑛 .

Confidence Interval Estimation

Single sample Confidence Interval Estimations
Point estimation does not provide enough information about the parameter of interest. For example
from a given sample 𝑋1 , 𝑋2 , … , 𝑋𝑛 of size n, we do not expect sample mean 𝑋 exactly equal to the
population mean 𝜇 but close to 𝜇. Rather than estimating the population parameter 𝜃, by the point
estimator 𝜃, it is sometime more valuable to specify an interval 𝐿 ≤ 𝜃 ≤ 𝑈 that contains the
population parameter 𝜃 with certain degree of confidence, that means, we find 𝐿 ≤ 𝜃 ≤ 𝑈 such
that 𝑃 𝐿 ≤ 𝜃 ≤ 𝑈 = 1 − 𝛼.

 The interval 𝐿 ≤ 𝜃 ≤ 𝑈 is called the 1 − 𝛼 100% confidence interval (two sided).

 L: lower confidence limit and U: upper confidence limit.
2
  (1 − 𝛼): Confidence coefficient or confidence level.
 𝛼 : Significance level.

Mostly we choose 𝛼 = 0.1 𝑜𝑟 0.05 𝑜𝑟 0.001

One-sided confidence interval

I. Lower confidence Interval

𝐿 ≤ 𝜃 such that 𝐿 ≤ 𝜃 = 1 − 𝛼 , called one-sided 1 − 𝛼 100% lower confidence
interval.
II. Upper confidence Interval
𝜃 ≤ 𝑈 such that 𝜃 ≤ 𝑈 = 1 − 𝛼, called one-sided 1 − 𝛼 100% upper confidence
interval.

Remark: Half-interval 𝜃 − 𝐿 or 𝑈 − 𝜃 is called the accuracy of the estimator. Longer the

confidence interval, the more confident we are that interval actually contains the true
parameter 𝜃. On the other hand, longer the interval, the less information we have about the true
value of 𝜃.
1. The Confidence Interval on the Mean 𝝁 of a Normal Distribution (Variance 𝝈𝟐 Known)
Two-Sided Confidence Interval
Let 𝑋1 , 𝑋2 , … , 𝑋𝑛 be n-samples taken from a normally distributed population 𝑋~𝑁(𝜇, 𝜎 2 ) with
unknown mean 𝜇 but known variance 𝜎 2 , then two-sided 𝟏 − 𝜶 𝟏𝟎𝟎% confidence interval
𝝈 𝝈 1 𝑛
of 𝝁 is 𝑿 − 𝒛𝜶/𝟐 ≤ 𝝁 ≤ 𝑿 + 𝒛𝜶/𝟐 , where 𝑠𝑎𝑚𝑝𝑙𝑒 𝑚𝑒𝑎𝑛, 𝑋 = 𝑛 𝑖=1 𝑋𝑖 ,
√𝒏 √𝒏
𝜎 𝜎
That means, 𝑃 𝑋 − 𝑧𝛼/2 ≤ 𝜇 ≤ 𝑋 + 𝑧𝛼/2 = 1−𝛼
√𝑛 √𝑛
𝑋1 +𝑋2 +⋯+𝑋𝑛 𝜎
Proof: We have 𝑋 = . So, 𝑋 = 𝜇 and 𝑉 𝑋 = .
𝑛 √𝑛
𝑋 −𝜇
Then, define the statistics 𝑍 = 𝜎/ ~𝑁(0,1)
√𝑛

Now, 𝑃 −𝑧𝛼/2 ≤ 𝑍 ≤ 𝑧𝛼/2 = 1 − 𝛼

𝑋−𝜇
⇒ 𝑃 −𝑧𝛼/2 ≤ ≤ 𝑧𝛼/2 = 1 − 𝛼
𝜎/√𝑛
⇒ 𝑃 −𝑧𝛼/2 𝜎/√𝑛 ≤ 𝑋 − 𝜇 ≤ 𝑧𝛼 /2 𝜎/√𝑛 = 1 − 𝛼
⇒ 𝑃 𝑋 − 𝑧𝛼/2 𝜎/√𝑛 ≤ 𝜇 ≤ 𝑋 + 𝑧𝛼/2 𝜎/√𝑛 = 1 − 𝛼
⇒ 𝑋 − 𝑧𝛼 /2 𝜎/√𝑛 ≤ 𝜇 ≤ 𝑋 + 𝑧𝛼/2 𝜎/√𝑛 is two sided 1 − 𝛼 100%
confidence interval of 𝜇.

3
 Figure: Two-sided 1 − 𝛼 100% confidence interval with 𝐿 = 𝑋 − 𝑧𝛼/2 𝜎/√𝑛 and 𝑈 = 𝑋 + 𝑧𝛼 /2 𝜎/√𝑛

Remarks:
1. Accuracy of confidence interval is 𝑧𝛼/2 𝜎/√𝑛
2. Length of the confidence interval is 2𝑧𝛼/2 𝜎/√𝑛
3. Error in the estimation is 𝑬 = 𝑿 − 𝝁 . The minimum number of sample size n
𝑍𝛼 /2 𝜎 2
required to keep the error E is given by 𝒏 = (take n as a ceiling valued
𝐸
integer).

One-Sided Lower Confidence Interval

Let 𝑋1 , 𝑋2 , … , 𝑋𝑛 be n-samples taken from a normally distributed population 𝑋~𝑁(𝜇, 𝜎 2 ) with

unknown mean 𝜇 but known variance 𝜎 2 , then one-sided lower 𝟏 − 𝜶 𝟏𝟎𝟎% confidence
𝝈 1 𝑛
interval of 𝝁 is 𝑿 − 𝒛𝜶 ≤ 𝝁, where 𝑋 = 𝑛 𝑖=1 𝑋𝑖 ,
√𝒏
𝜎
That means, 𝑃 𝑋 − 𝑧𝛼 ≤ 𝜇 = 1−𝛼
√𝑛

Figure: one-sided lower 1 − 𝛼 100% confidence interval with 𝐿 = 𝑋 − 𝑧𝛼 /2 𝜎/√𝑛

One-Sided Upper Confidence Interval

Let 𝑋1 , 𝑋2 , … , 𝑋𝑛 be n-samples taken from a normally distributed population 𝑋~𝑁(𝜇, 𝜎 2 ) with

unknown mean 𝜇 but known variance 𝜎 2 , then one-sided upper 𝟏 − 𝜶 𝟏𝟎𝟎% confidence
𝝈 1 𝑛
interval of 𝝁 is 𝝁 ≤ 𝑿 + 𝒛𝜶 , where 𝑋 = 𝑛 𝑖=1 𝑋𝑖 ,
√𝒏
𝜎
That means, 𝑃 𝜇 ≤ 𝑋 + 𝑍𝛼 =1−𝛼
√𝑛

4
 Figure: one-sided lower 1 − 𝛼 100% confidence interval with 𝑈 = 𝑋 + 𝑧𝛼 /2 𝜎/√𝑛

Values of 𝒁𝜶/𝟐 and 𝒁𝜶 for different values of 𝜶

𝜶 𝒛𝜶/𝟐 𝜶 𝒛𝜶
0.1 𝒛𝟎.𝟎𝟓 = 𝟏. 𝟔𝟒 0.1 𝒛𝟎.𝟏 = 𝟏. 𝟐𝟗
0.05 𝒛𝟎.𝟎𝟐𝟓 = 𝟏. 𝟗𝟔 0.05 𝒛𝟎.𝟎𝟓 = 𝟏. 𝟔𝟒
0.01 𝒛𝟎.𝟎𝟎𝟓 = 𝟐. 𝟓𝟖 0.01 𝒛𝟎.𝟎𝟏 = 𝟐. 𝟑𝟑
Example1:
The life in hours of 75 Watt light is known to be approximately normally distributed with standard
deviation 𝜎 = 25 𝑕𝑜𝑢𝑟𝑠 . A random sample of 20 bulbs has a mean life of 𝑥 = 1014 𝑕𝑜𝑢𝑟𝑠.
a. Construct a 95% two-sided confidence interval on the mean life.
b. Construct 95% lower confidence interval on the mean life.
c. We want 95% confidence that error in estimating mean life is less than 5 hours. What sample
size should be used?
d. We wanted with 95% confident that the total width of the confidence interval on the mean life
to be 8 hours. What sample size should be used?

Answer: 𝜎 = 25, 𝑥 = 1014, 𝑛 = 20

(a) For 95% two-sided confidence interval, 𝛼 = 0.05. So 95% confidence interval is given by
𝜎 𝜎
𝑥 − 𝑧𝛼/2 ≤ 𝜇 ≤ 𝑥 + 𝑧𝛼/2
√𝑛 √𝑛
25 25
⇒ 1014 − 1.96 × ≤ 𝜇 ≤ 1014 + 1.96 × (𝑼𝒔𝒆: 𝒛𝜶/𝟐 = 𝒛𝟎.𝟎𝟐𝟓 = 𝟏. 𝟗𝟔)
√20 √20
⇒ 1003.04 ≤ 𝜇 ≤ 1024.96

So, 1003.04 ≤ 𝜇 ≤ 1024.96 is the required 95% two-sided confidence interval.

(b) For 95% lower confidence interval , 𝛼 = 0.05. So 95% confidence interval is given by
𝜎
𝑥 − 𝑧𝛼 ≤ 𝜇
√𝑛
25
⇒ 1014 − 1.64 × ≤ 𝜇 (𝑼𝒔𝒆: 𝒛𝜶 = 𝒛𝟎.𝟎𝟓 = 𝟏. 𝟔𝟒)
√20
⇒ 1004.83 ≤ 𝜇

So, 1004.83 ≤ 𝜇 < ∞ is the required 95% one-sided lower confidence interval.
(c) For 95% confidence interval (𝛼 = 0.05), to keep the error of the amounts E=5, so sample size
needed is

5
 𝑧𝛼/2 𝜎 2
𝑛≈
𝐸
2
1.96 × 25
⇒ 𝑛≈
5
⇒ 𝑛 ≈ 96.04
⇒ 𝑛 = 97 (minimum sample sized needed
)

(d) With 95% confident (𝛼 = 0.05) that the total width of the confidence interval is 8, the sample size
needed is
𝜎
2𝑧𝛼 /2 =8
√𝑛
2𝑧𝛼/2 𝜎 2
⇒ 𝑛=
8
2 × 1.96 × 25 2
⇒ 𝑛=
8
2 × 1.96 × 25 2
⇒ 𝑛=
8
⇒ 𝑛 = 150.6 (rounding up)
𝒏 = 𝟏𝟓𝟏

2. The Confidence Interval on the Mean 𝝁 of a Normal Distribution (Variance

𝝈𝟐 Unknown)

Two-Sided Confidence Interval

Let 𝑋1 , 𝑋2 , … , 𝑋𝑛 be n-samples taken from a normally distributed population 𝑋~𝑁(𝜇, 𝜎 2 ) with

unknown mean 𝜇 and unknown variance 𝜎 2 , then two-sided 𝟏 − 𝜶 𝟏𝟎𝟎% confidence
𝑺 𝑺 1 𝑛
interval of 𝝁 is 𝑿 − 𝒕𝜶/𝟐,𝒏−𝟏 ≤ 𝝁 ≤ 𝑿 + 𝒕𝜶/𝟐,𝒏−𝟏 , where 𝑋 = 𝑛 𝑖=1 𝑋𝑖 and 𝑆 2 =
√𝒏 √𝒏
1 𝑛 2
𝑖=1 𝑋𝑖 − 𝑋
𝑛−1
𝑆 𝑆
That means, 𝑋 − 𝑡𝛼 /2,𝑛−1 ≤ 𝜇 ≤ 𝑋 + 𝑡𝛼/2,𝑛−1 = 1 − 𝛼 , where 𝑡𝛼/2,𝑛−1 is t-value of
√𝑛 √𝑛
t-distribution with (n-1) degrees of freedom.
𝑋1 +𝑋2 +⋯+𝑋𝑛
Proof: We have 𝑋 = 𝑛

𝑋 −𝜇
Then, define the statistics 𝑡 = 𝑆/ which is t-distributed with n-1 degrees of freedom.
√𝑛

6
 Now, 𝑃 −𝑡𝛼/2,𝑛−1 ≤ 𝑍 ≤ 𝑡𝛼 /2,𝑛−1 = 1 − 𝛼
𝑋−𝜇
⇒ 𝑃 −𝑡𝛼/2,𝑛−1 ≤ ≤ 𝑡𝛼/2,𝑛−1 = 1 − 𝛼
𝑆/√𝑛
⇒ 𝑃 −𝑡𝛼/2,𝑛−1 𝑆/√𝑛 ≤ 𝑋 − 𝜇 ≤ 𝑡𝛼 /2,𝑛−1 𝑆/√𝑛 = 1 − 𝛼
⇒ 𝑃 𝑋 − 𝑡𝛼 /2,𝑛−1 𝑆/√𝑛 ≤ 𝜇 ≤ 𝑋 + 𝑡𝛼/2,𝑛−1 𝑆/√𝑛 = 1 − 𝛼
⇒ 𝑋 − 𝑡𝛼 /2,𝑛−1 𝑆/√𝑛 ≤ 𝜇 ≤ 𝑋 + 𝑡𝛼/2,𝑛−1 𝑆/√𝑛 is two sided 1 − 𝛼 100%
confidence interval of 𝜇.

One-Sided Lower Confidence Interval

Let 𝑋1 , 𝑋2 , … , 𝑋𝑛 be n-samples taken from a normally distributed population 𝑋~𝑁(𝜇, 𝜎 2 ) with

unknown mean 𝜇 and unknown variance 𝜎 2 , then one-sided 𝟏 − 𝜶 𝟏𝟎𝟎% lower
𝑺 1 𝑛 1 𝑛
confidence interval of 𝝁 is 𝑿 − 𝒕𝜶,𝒏−𝟏 ≤ 𝝁, where 𝑋 = 𝑛 𝑖=1 𝑋𝑖 and 𝑆 2 = 𝑛−1 𝑖=1 𝑋𝑖 −
√𝒏
𝑋2
𝑆
That means, 𝑋 − 𝑡𝛼 ,𝑛−1 ≤ 𝜇 = 1 − 𝛼 , where 𝑡𝛼 ,𝑛−1 is t-value of t-distribution with (n-1)
√𝑛
degrees of freedom.

One-Sided Upper Confidence Interval

Let 𝑋1 , 𝑋2 , … , 𝑋𝑛 be n-samples taken from a normally distributed population 𝑋~𝑁(𝜇, 𝜎 2 ) with

unknown mean 𝜇 and unknown variance 𝜎 2 , then one-sided 𝟏 − 𝜶 𝟏𝟎𝟎% upper
𝑺 1 𝑛 1 𝑛
confidence interval of 𝝁 is 𝝁 ≤ 𝑿 + 𝒕𝜶,𝒏−𝟏 , where 𝑋 = 𝑛 𝑖=1 𝑋𝑖 and 𝑆 2 = 𝑛−1 𝑖=1 𝑋𝑖 −
√𝒏
𝑋2
𝑆
That means, 𝜇 ≤ 𝑋 + 𝑡𝛼 ,𝑛−1 , = 1 − 𝛼 , where 𝑡𝛼 ,𝑛−1 is t-value of t-distribution with (n-
√𝑛
1) degrees of freedom.

Example 1:
The wall thickness of 25 glass 2-liter bottle was measured by a quality control engineer. The sample
mean 𝑥 = 4.05 𝑚𝑚 and sample standard deviation𝑠 = 0.08 𝑚𝑚.
a. Find 90% two-sided confidence interval on the mean wall thickness.
b. Find 90% lower confidence interval on the mean wall thickness.

Answer: Given 𝑥 = 4.05, 𝑠 = 0.08, 𝑛 = 25

(a) For 90% two-sided confidence interval, 𝛼 = 0.1. The confidence interval is given by
𝑥 − 𝑡𝛼/2,𝑛−1 𝑠/√𝑛 ≤ 𝜇 ≤ 𝑥 + 𝑡𝛼/2,𝑛−1 𝑠/√𝑛
0.08 0.08
⇒ 4.05 − 1.711 × √25 ≤ 𝜇 ≤ 4.05 + 1.711 × √25 (Use: 𝒕𝜶/𝟐,𝒏−𝟏 = 𝒕𝟎.𝟎𝟓,𝟐𝟒 = 𝟏. 𝟕𝟏𝟏 )
⇒ 4.023 ≤ 𝜇 ≤ 4.077

7
 So, 4.023 ≤ 𝜇 ≤ 4.077 is the required 95% two-sided confidence interval.
(b) For 90% lower confidence interval, 𝛼 = 0.1. The confidence interval is given by
𝑥 − 𝑡𝛼 ,𝑛−1 𝑠/√𝑛 ≤ 𝜇
0.08
⇒ 4.05 − 1.318 × √25 ≤ 𝜇 (Use: 𝒕𝜶 ,𝒏−𝟏 = 𝒕𝟎.𝟏,𝟐𝟒 = 𝟏. 𝟑𝟏𝟖 )
⇒ 4.029 ≤ 𝜇
So, 4.029 ≤ 𝜇 < ∞ is the required 95% one-sided lower confidence interval.

Example 2:

The compressive strength of concrete is being tested by a civil engineer. He tested 16 specimens
and obtained the following data

2216 2237 2249 2204 2225 2301 2281 2263

2318 2255 2275 2295 2250 2238 2300 2217

a. Find 95% two-sided confidence interval on the mean strength.

b. Find 95% lower confidence interval on the mean strength.
c. Find 95% two-sided confidence interval on the on the mean strength assuming that 𝜎 2 = 36 .
Compare the interval with the one from part (a).

3. The Confidence Interval on the variance 𝝈𝟐 of a Normal Distribution

Two-Sided Confidence Interval

Let 𝑋1 , 𝑋2 , … , 𝑋𝑛 be n-samples taken from a normally distributed population 𝑋~𝑁(𝜇, 𝜎 2 )

with unknown mean 𝜇 and unknown variance 𝜎 2 , then two-sided 𝟏 − 𝜶 𝟏𝟎𝟎% confidence
(𝒏−𝟏)𝑺𝟐 (𝒏−𝟏)𝑺𝟐 1 𝑛
interval of 𝝈𝟐 is 𝝌𝟐 ≤ 𝝈𝟐 ≤ 𝝌𝟐 , where 𝑆 2 = 𝑛−1 𝑖=1 𝑋𝑖 − 𝑋 2
𝜶/𝟐,𝒏−𝟏 𝟏−𝜶/𝟐,𝒏−𝟏
(𝑛−1)𝑆 2 (𝑛−1)𝑆 2
That means, ≤ 𝜎2 ≤ 𝜒2 = 1 − 𝛼 , where 𝜒 2 𝛼/2,𝑛−1 and 𝜒 2 1−𝛼/2,𝑛−1 are
𝜒 2 𝛼 /2,𝑛 −1 1−𝛼 /2,𝑛 −1

𝜒 2 -value of 𝜒 2 -distribution with (n-1) degrees of freedom.

(𝑛 −1)𝑆 2
Proof : We define the statistics 𝜒 2 = which is 𝜒 2 -distributed with n-1 degrees of
𝜎2
freedom.

Now, 𝑃 𝜒 2 1−𝛼/2,𝑛−1 ≤ 𝜒 2 ≤ 𝜒 2 𝛼 /2,𝑛−1 = 1 − 𝛼

8
 (𝑛 − 1)𝑆 2
⇒ 𝑃 𝜒 2 1−𝛼/2,𝑛−1 ≤ ≤ 𝜒 2 𝛼/2,𝑛−1 = 1 − 𝛼
𝜎2
(𝑛 − 1)𝑆 2 2
(𝑛 − 1)𝑆 2
⇒ 𝑃 ≤𝜎 ≤ 2 =1−𝛼
𝜒 2 𝛼/2,𝑛−1 𝜒 1−𝛼/2,𝑛−1
(𝑛−1)𝑆 2 (𝑛−1)𝑆 2
⇒ 𝜒2 ≤ 𝜎2 ≤ 𝜒2 is two sided 1 − 𝛼 100% confidence
𝛼 /2,𝑛 −1 1−𝛼 /2,𝑛 −1

interval of 𝜎 2 .

One-Sided Lower Confidence Interval

Let 𝑋1 , 𝑋2 , … , 𝑋𝑛 be n-samples taken from a normally distributed population 𝑋~𝑁(𝜇, 𝜎 2 )

with unknown mean 𝜇 and unknown variance 𝜎 2 , then one-sided 𝟏 − 𝜶 𝟏𝟎𝟎% lower
(𝒏−𝟏)𝑺𝟐 1 𝑛
confidence interval of 𝝈𝟐 is ≤ 𝝈𝟐 where 𝑆 2 = 𝑛−1 𝑖=1 𝑋𝑖 − 𝑋 2
𝝌𝟐 𝜶,𝒏−𝟏
(𝑛−1)𝑆 2
That means, ≤ 𝜎 2 = 1 − 𝛼 , where 𝜒 2 𝛼,𝑛−1 is 𝜒 2 -value of 𝜒 2 -distribution with (n-
𝜒 2 𝛼 ,𝑛 −1

1) degrees of freedom.

One -Sided Upper Confidence Interval

Let 𝑋1 , 𝑋2 , … , 𝑋𝑛 be n-samples taken from a normally distributed population 𝑋~𝑁(𝜇, 𝜎 2 )

with unknown mean 𝜇 and unknown variance 𝜎 2 , then one-sided 𝟏 − 𝜶 𝟏𝟎𝟎% upper
(𝒏−𝟏)𝑺𝟐 1 𝑛
confidence interval of 𝝈𝟐 is 𝝈𝟐 ≤ , where 𝑆 2 = 𝑖=1 𝑋𝑖 − 𝑋 2
𝝌𝟐 𝟏−𝜶,𝒏−𝟏 𝑛−1
(𝑛−1)𝑆 2
That means, 𝜎 2 = 𝜒 2 = 1 − 𝛼 , where 𝜒 2 1−𝛼,𝑛−1 is 𝜒 2 -value of 𝜒 2 -distribution with
1−𝛼 ,𝑛 −1

(n-1) degrees of freedom.

Example:
The wall thickness of 25 glass 2-liter bottle was measured by a quality control engineer. The sample
mean 𝑥 = 4.05 𝑚𝑚 and sample standard deviation𝑠 𝑠 = 0.08 𝑚𝑚.
d. Find 95% two-sided confidence interval on the variance of the wall thickness data.
e. Find 95% lower confidence interval on the variance of the wall thickness data.
f. Find 95% upper confidence interval on the variance of the wall thickness data.

Answer: Given that 𝑠 = 0.08, 𝑠 2 = 0.0064, 𝑛 = 25.

(a) The 95% two-sided confidence interval, 𝛼 = 0.05. The class interval is given by
(𝑛 − 1)𝑠 2 2
(𝑛 − 1)𝑠 2
≤ 𝜎 ≤
𝜒 2 𝛼 /2,𝑛−1 𝜒 2 1−𝛼/2,𝑛−1
Using Chi-square 𝜒 2 − 𝑇𝑎𝑏𝑙𝑒, 𝑤𝑒 𝑔𝑒𝑡 𝜒 2 𝛼 ,𝑛−1 = 𝜒 2 0.025,24 = 39.364 and
2
2 2
𝜒 1−𝛼/2,𝑛−1
=𝜒 0.975,24
= 12.401

9
 24 × 0.0064 24 × 0.0064
⇒ ≤ 𝜎2 ≤
39.364 12.401
⇒ 0.0039 ≤ 𝜎 2 ≤ 0.0124
⇒ √0.0039 ≤ 𝜎 ≤ √0.0124
⇒ 0.062 ≤ 𝜎 ≤ 0.111

The required 95% two-sided confidence interval is 0.062 ≤ 𝜎 ≤ 0.111

(b) The 95% lower confidence interval, 𝛼 = 0.05. The class interval is given by
(𝑛 − 1)𝑠 2
≤ 𝜎2
𝜒 2 𝛼 ,𝑛−1
Using Chi-square 𝜒 2 − 𝑇𝑎𝑏𝑙𝑒, 𝑤𝑒 𝑔𝑒𝑡 𝜒 2 𝛼,𝑛−1 = 𝜒 2 0.05,24 = 36.42
24 × 0.0064
⇒ ≤ 𝜎2
36.42
⇒ 0.0042 ≤ 𝜎 2
⇒ √0.0042 ≤ 𝜎
⇒ 0.0649 ≤ 𝜎
The required 95% lower confidence interval is 0.0649 ≤ 𝜎 ≤ 0.

(c) The 95% upper confidence interval, 𝛼 = 0.05. The class interval is given by
2
(𝑛 − 1)𝑠 2
𝜎 ≤ 2
𝜒 1−𝛼,𝑛−1
Using Chi-square 𝜒 2 − 𝑇𝑎𝑏𝑙𝑒, 𝑤𝑒 𝑔𝑒𝑡 𝜒 2 1−𝛼,𝑛−1 = 𝜒 2 0.95,24 = 13.85
24 × 0.0064
⇒ 𝜎2 ≤
13.85
2
⇒ 𝜎 ≤ 0.0111
⇒ 𝜎 ≤ √0.0111
⇒ 0 ≤ 𝜎 ≤ 0.1053
The required 95% lower confidence interval is 0 ≤ 𝜎 ≤ 0.1053

Two-samples Confidence Interval Estimations

A. Confidence Interval on the Difference (𝝁𝟏 − 𝝁𝟐 ) Between the Means of Two Normal
Distributions (Variances Known)

Let 𝑋11 , 𝑋12 , … , 𝑋1𝑛 1 be 𝑛1 - samples taken from a 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛1~ 𝑁(𝜇1 , 𝜎12 ) and Let
𝑋21 , 𝑋22 , … , 𝑋2𝑛 2 be 𝑛2 - samples taken from a 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛2~ 𝑁(𝜇2 , 𝜎22 ) where the
population means 𝜇1 𝑎𝑛𝑑 𝜇2 are known but the population variances 𝜎12 𝑎𝑛𝑑 𝜎22 unknown.
1 𝑛1 1 𝑛2
If 𝑋1 = 𝑛 𝑖=1 𝑋1𝑖 and 𝑋2 = 𝑛 𝑖=1 𝑋2𝑖 are the sample means of population1 and
1 2
population2 respectively. Then,
(i) two-sided 𝟏 − 𝜶 𝟏𝟎𝟎% confidence interval of( 𝝁𝟏 − 𝝁𝟐 ) is
10
 𝜎12 𝜎22 𝜎12 𝜎22
𝑿𝟏 − 𝑿𝟐 − 𝒛𝜶/𝟐 + ≤ (𝝁𝟏 − 𝝁𝟐 ) ≤ 𝑿𝟏 − 𝑿𝟐 + 𝒛𝜶/𝟐 +
𝑛1 𝑛2 𝑛1 𝑛2
(ii) one-sided 𝟏 − 𝜶 𝟏𝟎𝟎% lower confidence interval of 𝝁𝟏 − 𝝁𝟐 is
𝜎12 𝜎22
𝑿𝟏 − 𝑿𝟐 − 𝒛𝜶 + ≤ 𝝁𝟏 − 𝝁𝟐
𝑛1 𝑛2
(iii) one-sided 𝟏 − 𝜶 𝟏𝟎𝟎% upper confidence interval of 𝝁𝟏 − 𝝁𝟐 is
𝜎12 𝜎22
−𝝁𝟐 ≤ 𝑿𝟏 − 𝑿𝟐 + 𝒛𝜶 +
𝑛1 𝑛2

Example 1:
Two machines are used to fill plastic bottles with dishwashing detergent. The standard deviation of
fill volume are known to be 𝜎1 = 0.15 𝑓𝑙𝑢𝑖𝑑 𝑜𝑢𝑛𝑐𝑒𝑠 and 𝜎2 = 0.18 𝑓𝑙𝑢𝑖𝑑 𝑜𝑢𝑛𝑐𝑒𝑠 for the two
machines, respectively. Two samples of 𝑛1 = 12 bottles from machine-1 and 𝑛1 = 10 bottles from
machine-2 are selected and sample mean fill volumes are 𝑥1 = 30.87 𝑓𝑙𝑢𝑖𝑑 𝑜𝑢𝑛𝑐𝑒𝑠 and 𝑥2 =
30.68 𝑓𝑙𝑢𝑖𝑑 𝑜𝑢𝑛𝑐𝑒𝑠.
a. Construct a 95% two-sided confidence interval on the mean difference in fill volume.
b. Construct 95% upper confidence interval on the mean difference in fill volume.

Answer: Given information is listed in the flowing table: (Known variances 𝝈𝟐𝟏 𝒂𝒏𝒅 𝝈𝟐𝟐 )

Population-1 Population-2
𝜎1 = 0.15 𝜎2 = 0.18
𝑛1 = 12 𝑛2 = 10
𝑥1 = 30.87 𝑥2 = 30.68
(a) For 95% two-sided confidence interval ( 𝛼 = 0.05 ) on the mean difference 𝜇1 − 𝜇2 , the
confidence interval is given by

𝜎12 𝜎22 𝜎12 𝜎22

𝑥1 − 𝑥 2 − 𝑧𝛼/2 + ≤ (𝜇1 − 𝜇2 ) ≤ 𝑋1 − 𝑋2 + 𝑧𝛼/2 +
𝑛1 𝑛2 𝑛1 𝑛2
(Using z-table 𝒛𝜶/𝟐 = 𝒛𝟎.𝟎𝟐𝟓 = 𝟏. 𝟗𝟔)

0.15 2 0.18 2
⇒ 30.87 − 30.68 − 1.96 × + ≤ (𝜇1 − 𝜇2
12 10

0.15 2 0.18 2
≤ 30.87 − 30.68 + 1.96 × +
12 10
⇒ 0.0498 ≤ (𝜇1 − 𝜇2) ≤ 0.3302
The required 95% two-sided confidence interval is 0.0498 ≤ (𝜇1 − 𝜇2 ) ≤ 0.3302
(b) For 95% upper confidence interval (𝛼 = 0.05) on the mean difference 𝜇1 − 𝜇2 , the confidence
interval is given by

𝜎12 𝜎22
(𝜇1 − 𝜇2 ) ≤ 𝑋1 − 𝑋2 + 𝑧𝛼 +
𝑛1 𝑛2
11
 (Using z-table 𝒛𝜶 = 𝒛𝟎.𝟎𝟓 = 𝟏. 𝟔𝟒)

0.15 2 0.18 2
⇒ (𝜇1 − 𝜇2 ≤ 30.87 − 30.68 + 1.64 × +
12 10
⇒ (𝜇1 − 𝜇2) ≤ 0.3073
The required 95% upper confidence interval is (𝜇1 − 𝜇2 ) ≤ 0.3073

B. Confidence Interval on the Difference (𝝁𝟏 − 𝝁𝟐 ) Between the Means of Two Normal
Distributions (Variances unknown)

Case I: If the sample sizes 𝑛1 ≥ 30 𝑎𝑛𝑑 𝑛2 ≥ 30 , we use confidence interval of the result
A by replacing 𝜎12 𝑏𝑦 𝑆12 and 𝜎22 𝑏𝑦 𝑆22 .

Case II: (𝜎12 = 𝜎22 )

Let 𝑋11 , 𝑋12 , … , 𝑋1𝑛 1 be 𝑛1 - samples taken from a 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛1~ 𝑁(𝜇1 , 𝜎12 ) and Let
𝑋21 , 𝑋22 , … , 𝑋2𝑛 2 be 𝑛2 - samples taken from a 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛2~ 𝑁(𝜇2 , 𝜎22 ) where the
population means 𝜇1 , 𝜇2 and population variances 𝜎12 , 𝜎22 unknown but 𝜎12 = 𝜎22 assumed.
1 𝑛1 1 𝑛1
If 𝑋1 = 𝑛 𝑖=1 𝑋1𝑖 and 𝑆12 = 𝑛 𝑖=1 𝑋1𝑖 − 𝑋1 2
are of population1 and 𝑋2 =
1 1 −1
1 𝑛2 1 𝑛2
𝑖=1 𝑋2𝑖 and 𝑆22 = 𝑛 𝑖=1 𝑋2𝑖 − 𝑋2 2
are of population2. Then,
𝑛2 2 −1
(i) two-sided 𝟏 − 𝜶 𝟏𝟎𝟎% confidence interval of( 𝝁𝟏 − 𝝁𝟐 ) is

1 1 1 1
𝑿𝟏 − 𝑿𝟐 − 𝒕𝜶/𝟐,𝒏𝟏+𝒏𝟐 −𝟐 𝑆𝑃 + ≤ (𝝁𝟏 − 𝝁𝟐 ) ≤ 𝑿𝟏 − 𝑿𝟐 + 𝒕𝜶/𝟐,𝒏𝟏 +𝒏𝟐 −𝟐 𝑆𝑃 +
𝑛1 𝑛2 𝑛1 𝑛2

(ii) one-sided 𝟏 − 𝜶 𝟏𝟎𝟎% lower confidence interval of( 𝝁𝟏 − 𝝁𝟐 ) is

1 1
𝑿𝟏 − 𝑿𝟐 − 𝒕𝜶,𝒏𝟏 +𝒏𝟐 −𝟐 𝑆𝑃 + ≤ (𝝁𝟏 − 𝝁𝟐 )
𝑛1 𝑛2
(iii) one-sided 𝟏 − 𝜶 𝟏𝟎𝟎% upper confidence interval of ( 𝝁𝟏 − 𝝁𝟐 ) is

1 1
(𝝁𝟏 − 𝝁𝟐 ) ≤ 𝑿𝟏 − 𝑿𝟐 + 𝒕𝜶,𝒏𝟏+𝒏𝟐 −𝟐 𝑆𝑃 +
𝑛1 𝑛2

𝑛 1 −1 𝑆12 + 𝑛 2 −1 𝑆22
where 𝑆𝑃2 = is combined (pooled) sample variance.
𝑛 1 +𝑛 2 −2

Example 2:

The diameter of steel rods manufactured on two different extrusion machines is being investigated.
Two random samples of sizes 𝑛1 = 15 and 𝑛2 = 18 are selected and sample mean and variance are
12
𝑥1 = 8.73, 𝑠1 2 = 0.30, 𝑥2 = 8.68, 𝑠2 2 = 0.34 respectively. Assume 𝜎1 2 = 𝜎2 2 . Construct a 95%
two-sided confidence interval on the difference in mean rod diameter.

Answer: Given information is listed in the flowing table: (Unknown variances 𝒃𝒖𝒕 𝝈𝟐𝟏 =
𝝈𝟐𝟐 𝒂𝒔𝒔𝒖𝒎𝒆𝒅)

Population-1 Population-2
𝑛1 = 15 𝑛2 = 18
𝑥1 = 8.73 𝑥2 = 8.68
𝑠12 = 0.30 𝑠22 = 0.34
(a) For 95% two-sided confidence interval ( 𝛼 = 0.05 ) on the mean difference 𝜇1 − 𝜇2 , the
confidence interval is given by
1 1
𝒙𝟏 − 𝒙𝟐 − 𝒕𝜶/𝟐,𝒏𝟏 +𝒏𝟐 −𝟐 𝑠𝑃 + ≤ (𝝁𝟏 − 𝝁𝟐 )
𝑛1 𝑛2

1 1
≤ 𝒙𝟏 − 𝒙𝟐 + 𝒕𝜶/𝟐,𝒏𝟏 +𝒏𝟐 −𝟐 𝑠𝑃 +
𝑛1 𝑛2
𝑛 1 −1 𝑠12 + 𝑛 2 −1 𝑠22 14×0.30+17×0.34
Where, 𝑠𝑃2 = = =0.3219 ⇒ 𝑠𝑝 = √0.3219 = 0.5674
𝑛 1 +𝑛 2 −2 15+18−2
(Using t-table 𝑡𝛼/2,𝑛 1 +𝑛 2 −2 = 𝑡0.025,31 = 𝑧0.025 = 1.96)

1 1
⇒ 8.73 − 8.68 − 1.96 × 0.5674 + ≤ (𝜇1 − 𝜇2 )
15 18

1 1
≤ 8.73 − 8.68 + 1.96 × 0.5674 +
15 18
⇒ −0.3388 ≤ (𝜇1 − 𝜇2 ) ≤ 0.4388

The required 95% two-sided confidence interval is −0.3388 ≤ (𝜇1 − 𝜇2 ) ≤ 0.4388

Case III: (𝜎12 ≠ 𝜎22 )

Let 𝑋11 , 𝑋12 , … , 𝑋1𝑛 1 be 𝑛1 - samples taken from a 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛1~ 𝑁(𝜇1 , 𝜎12 ) and Let
𝑋21 , 𝑋22 , … , 𝑋2𝑛 2 be 𝑛2 - samples taken from a 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛2~ 𝑁(𝜇2 , 𝜎22 ) where the
population means 𝜇1 , 𝜇2 and population variances 𝜎12 , 𝜎22 unknown but 𝜎12 ≠ 𝜎22 assumed.
1 𝑛1 1 𝑛1
If 𝑋1 = 𝑛 𝑖=1 𝑋1𝑖 and 𝑆12 = 𝑛 𝑖=1 𝑋1𝑖 − 𝑋1 2
are of population1 and 𝑋2 =
1 1 −1
1 𝑛2 1 𝑛2
𝑖=1 𝑋2𝑖 and 𝑆22 = 𝑛 𝑖=1 𝑋2𝑖 − 𝑋2 2
are of population2. Then,
𝑛2 2 −1
(iv) two-sided 𝟏 − 𝜶 𝟏𝟎𝟎% confidence interval of( 𝝁𝟏 − 𝝁𝟐 ) is

13
 𝑆12 𝑆22 𝑆12 𝑆22
𝑿𝟏 − 𝑿𝟐 − 𝒕𝜶/𝟐,𝝂 + ≤ (𝝁𝟏 − 𝝁𝟐 ) ≤ 𝑿𝟏 − 𝑿𝟐 + 𝒕𝜶/𝟐,𝝂 +
𝑛1 𝑛2 𝑛1 𝑛2

(v) one-sided 𝟏 − 𝜶 𝟏𝟎𝟎% lower confidence interval of( 𝝁𝟏 − 𝝁𝟐 ) is

𝑆12 𝑆22
𝑿𝟏 − 𝑿𝟐 − 𝒕𝜶,𝝂 + ≤ (𝝁𝟏 − 𝝁𝟐 )
𝑛1 𝑛2
(vi) one-sided 𝟏 − 𝜶 𝟏𝟎𝟎% upper confidence interval of ( 𝝁𝟏 − 𝝁𝟐 ) is

𝑆12 𝑆22
(𝝁𝟏 − 𝝁𝟐 ) ≤ 𝑿𝟏 − 𝑿𝟐 + 𝒕𝒕𝜶,𝝂 +
𝑛1 𝑛2

𝑆2 2 2
1 +𝑆 2
𝑛1 𝑛2
where 𝜈 = 2 2 −2
𝑆2
1 𝑆2
2
𝑛1 𝑛1
+
𝑛 1 +1 𝑛 2 +1

Example 3:

Random samples of sizes 𝑛1 = 15 and 𝑛2 = 10 are drawn from two independent normal
populations. The sample means and variances are 𝑥1 = 300, 𝑠1 2 = 16, 𝑥2 = 325, 𝑠2 2 = 48
Assume 𝜎1 2 ≠ 𝜎2 2 . . Construct a 95% two-sided confidence interval on 𝜇1 − 𝜇2 . (Student
Exercise)
C. Confidence Interval on the Difference (𝝁𝟏 − 𝝁𝟐 ) for paired Observations

When two samples of the two populations are collected in a pair under homogeneous
conditions but conditions may change from one pair to another.

Let (𝑋11 , 𝑋21 ), (𝑋12 , 𝑋22 ), … ..,(𝑋1𝑛 , 𝑋2𝑛 ) be set of n- paired observations, where
𝑋1𝑖 ~ 𝑁 𝜇1 , 𝜎12 and 𝑋2𝑖 ~ 𝑁 𝜇2 , 𝜎22 , 𝑖 = 1,2, … , 𝑛. We want to find the confidence interval
for the mean difference 𝜇1 − 𝜇2 .

Let us denote by the notations: 𝜇𝐷 = 𝜇1 − 𝜇2 , 𝐷𝑖 = 𝑋1𝑖 − 𝑋2𝑖 , 𝑖 = 1, … , 𝑛.

1 𝑛 1 𝑛
Let 𝐷 = 𝑛 𝑖=1 𝐷𝑖 and 𝑆𝐷2 = 𝑛−1 𝑖=1 𝐷𝑖 − 𝐷 2 . Then,

(i) two-sided 𝟏 − 𝜶 𝟏𝟎𝟎% confidence interval of (𝝁𝟏 − 𝝁𝟐 ) is

𝑺𝑫 𝑺𝑫
𝑫 − 𝒕𝜶/𝟐,𝒏−𝟏 ≤ 𝝁𝑫 ≤ 𝑫 + 𝒕𝜶/𝟐,𝒏−𝟏
√𝒏 √𝒏
(ii) one-sided 𝟏 − 𝜶 𝟏𝟎𝟎% lower confidence interval of (𝝁𝟏 − 𝝁𝟐 ) is
𝑺𝑫
𝑫 − 𝒕𝜶,𝒏−𝟏 ≤ 𝝁𝑫
√𝒏
(iii) two-sided 𝟏 − 𝜶 𝟏𝟎𝟎% upper confidence interval of (𝝁𝟏 − 𝝁𝟐 ) is
𝑺𝑫
𝑫 − 𝒕𝜶,𝒏−𝟏 ≤ 𝝁𝑫
√𝒏
14
 Example:
Memory capacity of 10 students was tested before and after the training and scores are tabulated as
follows:
Roll 1 2 3 4 5 6 7 8 9 10
Before Training 12 14 11 8 7 10 6 0 5 6
After Training 15 16 10 7 5 12 10 2 3 8

Find the 95% two-sided confidence interval for mean difference on ( 𝜇1 − 𝜇2 ).

Answer: Difference of memory capacity before and after the training is tabulated below:
𝐷𝑖 -3 -2 1 1 2 -2 -4 -2 2 -2
1 1
The sample mean of difference 𝐷 = 𝑛 𝑛𝑖=1 𝐷𝑖 ⇒ 𝐷 = 10 10 𝑖=1 𝐷𝑖 ⇒ 𝐷 = −0.9 and
1 𝑛 1 10
sample variance of the difference is 𝑠𝐷2 = 𝑖=1 𝐷𝑖 − 𝐷 2
⇒ 𝑠𝐷2 = 𝑖=1 𝐷𝑖 − 𝐷 2
⇒
𝑛−1 9
𝑠𝐷2=4.7667⇒𝑠𝐷=2.1833.
For 95% two-sided confidence interval ( 𝛼 = 0.05 ) for the mean difference 𝜇𝐷 , the
confidence interval is given by
𝑠𝐷 𝑠𝐷
𝐷 − 𝑡𝛼/2,𝑛−1 ≤ 𝜇𝐷 ≤ 𝐷 + 𝑡𝛼/2,𝑛−1
√𝑛 √𝑛
(Using t-table 𝑡𝛼/2,𝑛−1 = 𝑡0.025,9 = 2.262)
2.1833 2.1833
⇒ −0.9 − 2.262 × ≤ 𝜇𝐷 ≤ −0.9 + 2.262 ×
√10 √10
⇒ −2.4617 ≤ 𝜇𝐷 ≤ 0.6617
From this confidence interval we conclude that there is not significance difference in
memory capacity because the confidence interval contains zero.

𝜎 2
D. Confidence Interval on the ratio of Variances 𝜎1 2 of Two Normal Distribution
2
Populations
Let 𝑋11 , 𝑋12 , … , 𝑋1𝑛 1 be 𝑛1 - samples taken from a 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛1~ 𝑁(𝜇1 , 𝜎12 ) and Let
𝑋21 , 𝑋22 , … , 𝑋2𝑛 2 be 𝑛2 - samples taken from a 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛2~ 𝑁(𝜇2 , 𝜎22 ) where the
population means 𝜇1 , 𝜇2 and population variances 𝜎12 , 𝜎22 are all unknown. Let 𝑆12 =
1 𝑛1 1 𝑛2
𝑖=1 𝑋1𝑖 − 𝑋1 2
be sample variance of population1 and 𝑆22 = 𝑛 𝑖=1 𝑋2𝑖 − 𝑋2 2
𝑛 1 −1 2 −1
be of population2. Then,
𝜎1 2
(i) two-sided 1 − 𝛼 100% confidence interval of is
𝜎2 2

𝑆1 2 𝜎1 2 𝑆1 2
𝐹1−𝛼/2,𝑛 2 −1,𝑛 1 −1 ≤ ≤ 𝐹𝛼/2,𝑛 2 −1,𝑛 1 −1
𝑆2 2 𝜎2 2 𝑆2 2
𝜎1 2
(ii) one -sided 1 − 𝛼 100% lower confidence interval of is
𝜎2 2

15
 𝑆1 2 𝜎1 2
𝐹1−𝛼,𝑛 2 −1,𝑛 1 −1 ≤
𝑆2 2 𝜎2 2

𝜎1 2
(iii)one -sided 1 − 𝛼 100% upper confidence interval of is
𝜎2 2

𝜎1 2 𝑆1 2
≤ 𝐹𝛼 ,𝑛 2 −1,𝑛 1 −1
𝜎2 2 𝑆2 2

Example:
Random sample of size 20 were drawn from two independent normal populations. The sample
mean and standard deviation were 𝑥1 = 22.0, 𝑠1 = 1.8, 𝑥2 = 21.5, 𝑠2 = 1.5. Construct a 95%
𝜎 2
two-sided confidence interval on the ratio of population variances 𝜎1 2 .
2
Answer: The given information:
𝑥1 = 22.0, 𝑠1 = 1.8, 𝑥2 = 21.5, 𝑠2 = 1.5,

𝜎12
For 95% two-sided confidence interval, 𝛼 = 0.05 and the confidence interval for is
𝜎22
given by

𝑠1 2 𝜎1 2 𝑠1 2
𝐹 ≤ ≤ 𝐹
𝑠2 2 1−𝛼/2,𝑛 2 −1,𝑛 1 −1 𝜎2 2 𝑠2 2 𝛼/2,𝑛 2 −1,𝑛 1 −1

(Using F-table, we find 𝐹𝛼 /2,𝑛 2 −1,𝑛 1 −1 = 𝐹0.025,19,19 = 2.5 and 𝐹1−𝛼/2,𝑛 2 −1,𝑛 1 −1 =
1 1
=𝐹 = 0.40)
𝐹𝛼 /2,𝑛 1 −1,𝑛 2 −1 0.025 2,19,19

1.8 2 𝜎1 2 1.8 2
⇒ × 0.4 ≤ ≤ × 2.5
1.5 2 𝜎2 2 1.5 2

𝜎1 2
⇒ 0.5760 ≤ ≤ 3.6
𝜎2 2

𝜎2 𝜎1 2
The required two-sided 95% confidence interval for 𝜎12 is 0.5760 ≤ ≤ 3.6
2 𝜎2 2

16