MA2540/MA4240: Applied Statistics

Dr. Sameen Naqvi

Department of Mathematics, IIT Hyderabad
Email id: sameen@math.iith.ac.in
 Example 6

Suppose we have conducted a study on the acidity levels of 61 soil

samples collected from a particular agricultural region. The sample
mean acidity level is 5.8 pH units, and the sample standard
deviation is 0.6 pH units.

(a) Calculate the 95% confidence interval for the mean acidity
level in the soil of this agricultural region.

(b) Calculate the 99% confidence interval for the mean acidity
level in the soil of this agricultural region.
 Solution

(a) Given: n = 61, df = 60, X̄ = 5.8, S = 0.6, t0.025,n−1 = 2.000.

0.6
The interval is: 5.8 ± 2 √ = 5.8 ± 0.154, i.e., (5.646, 5.954).
61

Thus, we are 95% confident that the mean acidity level in the
soil is 5.646 pH to 5.954 pH.

Width of interval = 0.308.

 Solution

(b) Given: n = 61, df = 60, X̄ = 5.8, S = 0.6, t0.005,n−1 = 2.660.

0.6
The interval is: 5.8 ± 2.660 √ = 5.8 ± 0.204, i.e.,
61
(5.596, 6.004).

Thus, we are 99% confident that the mean acidity level in the
soil is 5.596 pH to 6.004 pH.

Width of interval = 0.408.

 Confidence and Precision

I Wider intervals have poorer precision.

I Higher the confidence level, the wider is the width of the

interval and thus less precision.
 Determining Sample Size : t-interval
 
I Since E = tα/2,n−1 √S , we determine the sample size by
n
solving the equation for n:
(tα/2,n−1 )2 S 2
n= .
E2

I Here, approximate value of S = range

4 .
 Determining Sample Size : t-interval

I Note that the t-value on the right depends upon n.

I Crude Method: Simply replace the t-value that depends on

n with a Z -value that doesn’t because as n increases, the
t-distribution approaches the standard normal distribution.

I Thus,
(zα/2 )2 S 2
n≈ .
E2

I Iterative method: Start with an initial guess for n, plug in

the formula, and iteratively solve for n.
CI for population variance
 (C). CI for population variance

Theorem 3

If X1 , X2 , . . . , Xn are normally distributed and a = χ21−α/2,n−1 and

b = χ2α/2,n−1 , then a (1 − α)100% CI for the population variance
σ 2 is: !
(n − 1)S 2 (n − 1)S 2
,
b a
and a (1 − α)100% CI for the population standard deviation σ is:
p p !
(n − 1)S (n − 1)S
√ , √
b a
 (C). CI for population variance contd.
Proof
It is known that if X1 , X2 , . . . , Xn are normally distributed with
mean µ and population variance σ 2 , then
(n − 1)S 2
∼ χ2n−1 .
σ2

With a = χ21−α/2 and b = χ2α/2 , and using

 (C). CI for population variance contd.

we have
h (n − 1)S 2 i
P a≤ ≤ b =1−α
σ2

Considering
(n − 1)S 2
a≤ ≤ b,
σ2

and simplifying, we get

(n − 1)S 2 (n − 1)S 2
≤ σ2 ≤ .
b a
 Example 7

I A pharmaceutical company produces pills with an intended

active ingredient concentration of 20 milligrams per tablet. A
quality control analyst at the company is concerned about the
variation in the actual active ingredient concentrations and
wants to estimate the population standard deviation (σ) of
the concentrations.
I To do this, the analyst randomly selects a sample of n = 15
pills from a production batch and measures their active
ingredient concentrations. The sample yields a sample
variance of 3.6.
I Use this random sample data to calculate a 95% confidence
interval for σ of the active ingredient concentrations in these
pills.
 Solution
Here,
a = χ21−α/2,n−1 = χ20.975,14 = 5.629
and
b = χ2α/2,n−1 = χ20.025,14 = 26.119.
Substituting in the formula,
 
14 × 3.6 2 14 × 3.6
≤σ ≤
26.119 5.629
and simplifying, we get 95% confidence interval for σ 2

1.93 ≤ σ 2 ≤ 8.95 .

This leads to 95% confidence interval for σ

(1.39 ≤ σ ≤ 2.99).
CI for population proportion
 (D). CI for Population Proportion
Theorem 4
For large random samples, a 100(1 − α)% CI for population
proportion p is: r
p̂(1 − p̂)
p̂ ± zα/2 .
n

I Proof. We know that, for large n,

p̂ − p
Z=q ∼ N(0, 1).
p(1−p)
n

Now,
h p̂ − p i
P − zα/2 ≤ q ≤ zα/2 ≈ 1 − α.
p(1−p)
n
 (D). CI for Population Proportion contd.
Now, consider the inequality inside the brackets:
−zα/2 ≤ qp̂−p ≤ zα/2
p(1−p)
n
r r
p(1 − p) p(1 − p)
−zα/2 ≤ p̂ − p ≤ +zα/2
r n rn
p(1 − p) p(1 − p)
−p̂ − zα/2 ≤ −p ≤ −p̂ + zα/2
r n r n
p(1 − p) p(1 − p)
p̂ − zα/2 ≤ p ≤ p̂ + zα/2
n n
Replace population proportions (p) that appear at endpoints of the
interval with sample proportion (p̂) to get an (approximate)
100(1 − α)% CI for p
r r
p̂(1 − p̂) p̂(1 − p̂)
p̂ − zα/2 ≤ p ≤ p̂ + zα/2 .
n n
 Example 8

A marketing agency conducted a survey to investigate the

preference for eco-friendly packaging among consumers in a
city.

Out of 600 respondents, 420 expressed a preference for

eco-friendly packaging.

Using this sample proportion, the marketing agency wants to

estimate, with 95% confidence, the parameter p, which is the
proportion of all consumers in the city who prefer eco-friendly
packaging. What is the confidence interval for p based on this
sample proportion?
 Solution

Given: n = 600, sample proportion p̂ = 420

600 = 0.70, and
z0.025 = 1.96. Substituting in the formula for Cl for p, we get:
r
0.70(1 − 0.70)
0.70 ± 1.96
600
i.e.,
0.70 ± 0.037 = (0.663, 0.737)
Thus, we can be 95% confident that between 66.3% and 73.7% of
the population in the city prefer eco-friendly packaging.
CIs for difference of two population means
 CIs for µ1 − µ2

(A.) when the populations are independent and normally

distributed with unknown common variance σ 2 - Two sample
Pooled t-interval.

(B.) when the populations are independent and normally

distributed with unknown and unequal variances - Welch’s
t-interval.

(C.) when the populations are dependent and normally distributed

- Paired t-interval.
 (A). Two-Sample Pooled t-interval
Theorem 1
If X1 , X2 , . . . , Xn ∼ N(µ1 , σ 2 ) and Y1 , Y2 , . . . , Ym ∼ N(µ2 , σ 2 ) are
independent random samples, then a (1 − α)100% CI for the
difference in the population means, µ1 − µ2 is:
r
1 1
(X − Y ) ± tα/2,n+m−2 Sp + ,
n m

where Sp2 , the “pooled sample variance”

(n − 1)SX2 + (m − 1)SY2
Sp2 =
n+m−2

is an UE of the common variance σ 2 .

I Note: See Theorem 4, Week 5.

 (A). Two-Sample Pooled t-interval contd.
Proof:
It is known that
(X − Y ) − (µ1 − µ2 )
T = q ∼ tn+m−2 .
1 1
Sp n + m

Also,
 
(X − Y) − (µ1 − µ2 )
P −tα/2,n+m−2 ≤ q ≤ tα/2,n+m−2  = 1−α.
1 1
Sp n + m

Consider the inequality within the bracket

r
1 1
−tα/2,n+m−2 Sp + ≤ (X − Y ) − (µ1 − µ2 )
n m
r
1 1
≤ tα/2,n+m−2 Sp +
n m
 (A). Two-Sample Pooled t-interval contd.

On simplification, we get
r
1 1
(X − Y ) − tα/2,n+m−2 Sp + ≤ µ1 − µ 2
n m
r
1 1
≤ (X − Y ) + tα/2,n+m−2 Sp +
n m

Thus, (1 − α)100% CI for the difference in the population

means is r
1 1
(X − Y ) ± tα/2,n+m−2 Sp + .
n m
 Example 1

Suppose the number of products sold by the two sales team, A and
B, weekly is as follows:

Team A Team B
28, 35, 30, 32, 29, 34, 31, 33, 24, 29, 26, 31, 27, 30, 28, 32,
27, 36, 30, 32, 28, 35, 31, 33, 25, 33, 29, 31, 24, 29, 28, 32,
29, 34, 30, 32, 31 26, 30, 27, 31, 28

Is there statistically significant evidence to conclude that there

is a difference in the average number of products sold between
two sales teams?
 Solution

I Let Xi and Yi be the number of products sold by Team A and

Team B in the i th week, respectively.

I Since sample variances SX2 = 6.05 and SY2 = 6.63 are not that
different, we can assume the population variances are similar.

I The pooled sample variance

(21 − 1)6.05 + (21 − 1)6.63

Sp2 = = 6.68.
21 + 21 − 2
which implies Sp = 2.58.
 Solution contd.

I For m = n = 21, if we calculate a 95%Cl, we have

t0.025,21+21−2 = t0.025,40 = 2.021.

Also, x̄ = 31.43 and ȳ = 28.57. Thus, the 95%Cl for the

difference in population means are
r
1 1
(31.43 − 28.57) ± 2.021(2.58) + = (1.250, 4.470).
21 21
I Since the interval does not contain the value 0, we can
conclude that the population means differ.
 (B). Welch’s t-interval (if σX2 6= σY2 )
Theorem 2
If data is normally distributed and the population variances σX2 and
σY2 can’t be assumed to be equal, then a (1 − α)100% CI for the
difference in the population means, µX − µY is:
s
SX2 S2
(X − Y ) ± tα/2,r + Y,
n m

where the r d.f. are approximated by:

2
SX2 SY2

n + m
r= .
(SX2 /n)2 (SY2 /m)2
n−1 + m−1

I Note: See Theorem 3, Week 5.

 Example 2

I In Example 1, the following statistics were given:

n = 21, x̄ = 31.43, SX2 = 6.05

m = 21, ȳ = 28.57, SY2 = 6.63.

What is the difference, if any, in the mean number of products

sold by sales teams (Team A and Team B)?
 Solution

I Here,
2
SX2 SY2

+ 6.05 6.63 2


n m 21 + 21
r= 2 2 = (6.05/21)2 2 ≈ 40.07.
(SX2 /n)
+
(SY2 /m)
20 + (6.63/21)
20
n−1 m−1

So, dr e = 40.

Using a t-table, we get t0.025,40 = 2.021.

Thus, Welch’s interval is

r
6.05 6.63
(31.43 − 28.57) ± 2.021 +
21 21
 Solution contd.

Thus, 95%CI for µX − µY is (1.360, 4.360).

I Recall from Example 1 that the two-sample pooled t-interval

was (1.250, 4.470).

I Comparing the two intervals, we note that they aren’t that

different. The reason is that sample variances aren’t really all
that different.

I Rule of thumb: Use Welch’s interval if

SX2 SY2
> 4 or >4
SY2 SX2
 (C.) Paired t-interval

Theorem 3

When dealing with pairs of dependent measurements, sample mean

difference, D should be used to estimate population mean
difference, µD . The (1 − α)100% t-interval is
S 
D
D ± tα/2,n−1 √
n

I Note: See Theorem 6, Week 5.

 Example 3

I Suppose you want to investigate if the installation of a new

air filtration system in a factory has had an impact on the
level of a specific air pollutant (e.g., particulate matter) in the
factory environment.

I The collected data on the concentration of particulate matter

(in micrograms per cubic meter) before and after the
installation of the filtration system for ten different days, is as
follows:
 Example 3

Day Before installation After installation

1 45 38
2 50 42
3 48 40
4 55 48
5 42 35
6 47 41
7 53 45
8 52 40
9 49 39
10 46 37
 Solution

I Xi : concentration of particulate matter before installation of

the filtration system.

I Yi : concentration of particulate matter after installation of

the filtration system.

I Calculating Di = Xi − Yi removes the effect of the air

filtration system, and therefore, Di ’s are independent.
 Solution contd.

Day Xi Yi Di = Xi − Yi
1 45 38 7
2 50 42 8
3 48 40 8
4 55 48 7
5 42 35 7
6 47 41 6
7 53 45 8
8 52 40 12
9 49 39 10
10 46 37 9
 Solution contd.

I Thus, the 95% CI for µD is

SD
D ± t0.025,9 √ .
n
I From the given data, we get

1.62
7.4 ± 2.262 √ = (6.241, 8.559).
10
I Since 95% confidence interval does not include 0, we can
conclude that installation of the filtration system has a
significant effect in reducing the particulate matter.
CIs for ratio of two population variances
 CIs for ratio of two population variances

Theorem 4
If X1 , X2 , . . . , Xn ∼ N(µX , σX2 ) and
Y1 , Y2 , . . . , Ym ∼ N(µY , σY2 ) are independent samples, then a
(1 − α)100% CI for σX2 /σY2 is:
!
1 SX2 SX2
, F (m − 1, n − 1) 2 .
Fα/2 (n − 1, m − 1) SY2 α/2 SY
CIs for ratio of two population variances contd.

Proof
(n−1)SX2 (m−1)SY2
We know that σX2
∼ χ2n−1 and σY2
∼ χ2m−1 .
Also, by the independence of the two samples,
(m−1)SY2
σY2
/(m − 1) σX2 SY2
F = = · ∼ F (m − 1, n − 1).
(n−1)SX2
/(n − 1) σY2 SX2
σX2

Therefore,
" #
σX2 SY2
P F1−α/2 (m−1, n−1) ≤ 2 · 2 ≤ Fα/2 (m−1, n−1) = 1−α
σY SX
CIs for ratio of two population variances contd.

Simplifying the quantity within the bracket and using the fact
that
1
F1−α/2 (m − 1, n − 1) = ,
Fα/2 (n − 1, m − 1)

the (1 − α)100% CI for σX2 /σY2 is:

1 SX2 σX2 SX2

≤ ≤ F α/2 (m − 1, n − 1) .
Fα/2 (n − 1, m − 1) SY2 σY2 SY2
 Example 4

I In Example 1, the following statistics were given:

n = 21, x̄ = 31.43, SX2 = 6.05

m = 21, ȳ = 28.57, SY2 = 6.63.

Estimate, with 95% confidence, the ratio of the two

population variances.
 Solution

I From the F -table

1 1
F0.025 (20, 20) = 2.47 and F0.975 (20, 20) = = .
F0.025 (20, 20) 2.47

σ2
Then, the 95% CI for X2 is
σY
σX2
   
1 6.05 6.05
≤ 2 ≤ 2.47 .
2.47 6.63 σY 6.63

Simplifying, we get the 95% CI as (0.367, 2.237)

CIs for difference of two population proportions
 CIs for difference of two population proportions
I Therem 5
For large random samples, an approximate 100(1 − α)% CI for the
difference in two population proportions p1 − p2 is:
s
p̂1 (1 − p̂1 ) p̂2 (1 − p̂2 )
(p̂1 − p̂2 ) ± zα/2 + .
n1 n2

Proof. We know that

!
Y1 p1 (1 − p1 )
p̂1 = ∼ N p1 ,
n1 n1
and !
Y2 p2 (1 − p2 )
p̂2 = ∼ N p2 ,
n2 n2
CIs for difference of two population proportions contd.
By independence,
!
p1 (1 − p1 ) p2 (1 − p2 )
(p̂1 − p̂2 ) ∼ N p1 − p2 , + .
n1 n2

Now,
" #
(p̂1 − p̂2 ) − (p1 − p2 )
P − zα/2 ≤ q ≤ zα/2 ≈ 1 − α
p1 (1−p1 ) p2 (1−p2 )
n1 + n2

Simplifying the quantity within the bracket, we get the

approximate 100(1 − α)% CI for p1 − p2 :
s
p̂1 (1 − p̂1 ) p̂2 (1 − p̂2 )
(p̂1 − p̂2 ) ± zα/2 + .
n1 n2
 Example 5

A marketing research company conducted a study to compare

the effectiveness of two advertising campaigns, Campaign X
and Campaign Y, in attracting new customers to a retail store.

It was found that in a sample of 400 people who were exposed

to Campaign X, 200 of them visited the store. For Campaign
Y, in a sample of 250 people, 100 of them visited the store.

Calculate a 95% confidence interval for the difference in the

proportions of people who visited the store as a result of the
two advertising campaigns (Campaign X and Campaign Y).
 Solution
Data Campaign X Campaign Y
Sample size 400 250
# of people visited the store 200 100
Sample proportion 0.50 0.40
I Substituting in the formula, we get
r
0.50 × 0.50 0.40 × 0.60
(0.50 − 0.40) ± 1.96 +
400 250
which simplifies to

0.10 ± 0.078 = (0.022, 0.178)

I We can be 95% confident that there are between 2.2% and
17.8% more visitors to store due to campaign X than
campaign Y.
Thank you for listening!