Statistics I

22nd Class
Influences Based on Two Samples:
Confidence Intervals and Tests of
Hypothesis
19 June 2023
Makoto Shimizu
 Chapter 8
Inferences Based
on a Two Samples
Confidence
Intervals and
Tests of Hypothesis

2
 Content
1. Identifying the Target Parameter
2. Comparing Two Population Means: Independent
Sampling
3. Comparing Two Population Means: Paired
Difference Experiments
4. Comparing Two Population Proportions:
Independent Sampling
5. Determining the Required Sample Size
6. Comparing Two Population Variances:
Independent Sampling

3
 Where We’re Going
1. Learn how to identify the target parameter
for comparing two populations.
2. Learn how to compare two population
means using confidence intervals and tests
of hypotheses
3. Apply these inferential methods to
problems where we want to compare
• Two population proportions
• Two population variances

4
 Where We’re Going
4. Determine the sizes of the samples
necessary to estimate the difference
between two population parameters with a
specified margin of error

5
 8.1

Identifying the Target Parameter

6
 Determining the Target Parameter

Parameter Key Words or Phrases Type of Data

Mean difference; differences
1 – 2 in averages Quantitative

Differences between
proportions, percentages,
p1 – p2 fractions, or rates; compare Qualitative
proportions

( 1 ) 2 Ratio of variances;
differences in variability or Quantitative
( 2 ) 2 spread; compare variation

7
 8.2

Comparing Two Population Means:

Independent Sampling

8
 Properties of the Sampling
Distribution of ( x1  x2 )
1. The mean of the sampling distribution ( x1  x2 ) is ( 1   2 ).
2. If the two samples are independent, the standard deviation of
the sampling distribution is
 12  22
 x x   
1 2
n1 n2

where 1 and 2 are the variances of the two populations being

 2 2

sampled, and n1 and n2 are the respective sample sizes. We also refer
to   x1  x2  as the standard error of the statistic ( x1  x2 ).

3. The sampling distribution of ( x1  x2 ) is approximately normal for large

samples the Central Limit Theorem.

9
Large-Sample Confidence Interval for (μ1
– μ2): Normal (z) Statistic
 12 ,  22 known:
 12  22
x1  x2  z 2 x  x   x1  x2  z 2
1 2
n1

n2

 12 ,  22 unknown:
s12 s22
 x1  x2   z 2  x  x    x1  x2   z 2 
1 2
n1 n2

10
 Large-Sample Test of Hypothesis for (µ1 –
µ2): Normal (z) Statistic
One-Tailed Test
H0: (µ1 – µ2) = D0
Ha: (µ1 – µ2) < D0 [or Ha: (µ1 – µ2) > D0 ]
where D0 = Hypothesized difference between the means (the
difference is often hypothesized to be equal to 0)
z
x1  x2  D0
 12  22
Test statistic:  x  x   x  x   
1 2
1 2
n1 n2
Rejection region: z < –z s12 s22
 
[or z > z when Ha: (µ1 – µ2) > D0 ] n1 n2
11
 Large-Sample Test of Hypothesis for (µ1 –
µ2): Normal (z) Statistic
Two-Tailed Test
H0: (µ1 – µ2) = D0
Ha: (µ1 – µ2) ≠ D0
where D0 = Hypothesized difference between the means (the
difference is often hypothesized to be equal to 0)

z
x1  x2  D0  12  22
Test statistic:  x  x   x  x   
1 2
1 2
n1 n2
2 2
Rejection region: |z | > z s s
 1
 2
n1 n2
12
 Conditions Required for Valid Large-Sample
Inferences about
(μ1 – μ2)
1.The two samples are randomly selected in an independent
manner from the two target populations.
2.The sample sizes, n1 and n2, are both large (i.e., n1 ≥ 30 and
n2 ≥ 30). [Due to the Central Limit Theorem, this condition
guarantees that the sampling distribution of  1 2 will be
x x
approximately normal regardless of the shapes of the
underlying probability distributions of the populations.
s 2
s 2

Also, 1 and 2 will provide good approximations to 1 and 2
2
 2

when the samples are both large.]

13
 Example: Comparing Mean Car Prices
• An independent study of retail prices of 50 random
retail automobile sales in the United States and Japan
provide the following data:
Country Samples Mean Std. Dev
USA 50 26596 1981.4404
Japan 50 27236 1974.0934

• A claim is made that higher retail prices are being

charged for Japanese automobiles in Japan than in the
United States. Conduct a test of hypothesis to compare
the mean retail prices for the two countries. Use  =
0.05 for the test.

14
Example: Comparing Mean Car Prices (Cont.)
• Let1 and  2represent the population mean retail sales
prices in the United States and Japan, respectively. If the
claim is true, then the mean retail price in Japan will exceed
the mean in the United States, i.e.
1  2 or  1  2   0.
• The elements of the test:
• H0:  1  2   0, i.e. 1  2 ; note that D0 = 0 for this test.
• Ha:  1  2   0, i.e. 1  2
z
 x1  x2   D0

( x1  x2 )  0
• Test statistic:  x x   x x 
1 2 1 2

• Rejection region: z < –z0.05 = – 1.645

15
Example: Comparing Mean Car Prices (Cont.)

• Use the given summary statistics and obtain:

z
 x1  x2   D0

(25,596  27,236)  0
 x x   12  22
1 2

n1 n2
640 640 640
    1.62
s2
s 2
(1,981) 2
(1,974) 2 396
1
 2

n1 n2 50 50

Rejection region: z < –z0.05 but z   z0.05 :  1.62  1.635.

Thus, samples do not provide sufficient evidence at  = 0.05 to
conclude that the mean retail price in Japan exceeds that in the
United States.
16
 Small Independent Samples Confidence Interval
for (μ1 – μ2): Student’s t-Statistic

 1 1
x 1 
 x2  t 2
s   
2
p
 n1 n2 

n
s2  1
 1 s1
2
 n2  1 s 2
2
where p
n1  n2  2

and t/2 is based on (n1 + n2 – 2) degrees of freedom.

17
Small Independent Samples Test of Hypothesis for
(µ1 – µ2): Student’s t-Statistic
One-Tailed Test
H0: (µ1 – µ2) = D0
Ha: (µ1 – µ2) < D0 [or Ha: (µ1 – µ2) > D0 ]

t
x1  x2  D0
Test statistic:
1 1
s   
2
p
 n1 n2 
Rejection region: t < –t
[or t > t when Ha: (µ1 – µ2) > D0 ]
p-value = P(t < tc) [or P(t > tc)]
where t is based on (n1 + n2 – 2) degrees of freedom.
18
Small Independent Samples Test of Hypothesis for
(µ1 – µ2): Student’s t-Statistic
Two-Tailed Test
H0: (µ1 – µ2) = D0
Ha: (µ1 – µ2) ≠ D0
t
x1  x2  D0
Test statistic:
1 1
s   
2
p
 n1 n2 
Rejection region: |t| > t /2
• p-value = 2P(t < |tc|)
where t/2 is based on (n1 + n2 – 2) degrees of freedom.

19
Conditions Required for Valid Small-Sample Inferences about
(μ1 – μ2)

1. The two samples are randomly selected in an

independent manner from the two target
populations.
2. Both sampled populations have distributions that are
approximately equal.
3. The populations variances are equal
(i.e.,  12   22).

20
 Example: Managerial Success
• A researcher is hired to compare the average success index for
two groups of managers by a manufacturer. Group 1 interact
with people outside a manager’s group while Group 2 managers
have little outside interactions.
• The following data was collected:
Success Std
Samples Mean
Index Deriv
Group 1 12 65.33 6.61
Group 2 15 49.47 9.33

a. Use the data table to estimate the true mean difference

between the success indexes of managers in the two groups.
Use a 95% confidence interval.

21
Example: Managerial Success (Cont.)
df = n1 + n2 – 2 = 12 + 15 – 2 = 25
t/2 = t0.025 = 2.064 (t-Static Table)
Calculate the pooled estimate of variance:

1  1 2  2
n  1 s 2
 n  1 s 2

s 2p 
n1  n2  2
12  16.61  15  19.33
2 2


12  15  2
 67.97

22
Example: Managerial Success (Cont.)
Find confidence interval for  1  2 , the difference between
mean managerial success indexes for the two groups:
1 1   1 1 
 x1  x2   t 2 s 2p      (65.33  49.47)  t0.025 67.97   
 n1 n2   12 15 

 15.86  (2.06)(3.19)

 15.86  6.58

The confidence interval is (9.28, 22.44).

23
Example: Managerial Success (Cont.)

b. Interpret the interval.

The confidence interval includes positive differences
only. So, we are 95% confident that  1  2  exceeds
0. In fact, we estimate the mean success index, 1 ,
for managers with a high volume of outsider
interaction (group 1) to be anywhere between 9.28
and 22.44 points higher than the mean success
index, 2, of managers with few interactions (group
2).

24
 Approximate Small-Sample
Procedures when  1   2
2 2

1. Equal sample sizes (n1 = n2 = n)

Confidence interval: x1  x2  t 2 s12  s22  n

Test statistic for H0: 1  2   0
t  x1  x2   
s12  s22 n

where t is based on v = n1 + n2 – 2 = 2(n – 1) degrees of

freedom.
25
Approximate Small-Sample Procedures
when (Cont.)  1   2
2 2

2. Unequal sample sizes (n1 ≠ n2)

Confidence interval: x1  x2  t 2 s12  

n1  s22 n2 
Test statistic for H0: 1  2   0
t  x1  x2    
s12 n1  s22 n2 
where t is based on degrees of freedom equal to (next
slide)
26
 Approximate Small-Sample
Procedures when  1   2(Cont.)
2 2

v
s 2
1 n1  s n2
2
2 2

s
2
1 n1   s
2 2
2 n2 
2

n1  1 n2  1

Note: The value of v will generally not be an

integer. Round v down to the nearest integer to
use the t-table.

27
 8.3

Comparing Two
Population Means:
Paired Difference Experiments

28
 Paired Differences
• A valid method of analyzing data is by pairing the
data and then taking their differences. Paired
differences can provide more information about the
difference between population means than an
independent samples experiment.
• Let xd = sample mean difference
sd = sample standard deviation of differences
nd = number of differences = number of pairs

29
Paired-Difference Confidence Interval
for µd = (µ1 – µ2)
Large Sample, Normal (z) Statistic
d sd
xd  z 2  xd  z 2
nd nd

Small Sample, Student’s t-Statistic

sd
xd  t 2
nd
where t/2 is based on (nd – 1) degrees of freedom.

30
 Paired-Difference Test of Hypothesis
for µd = (µ1 – µ2)
One-Tailed Test
H0: µd = D0
Ha: µd < D0 [or Ha: µd > D0 ]
Large Sample, Normal (z) Statistic
xd  D0 xd  D0
Test statistic: z  
 d nd sd nd
Rejection region: z < –z
[or z > z when Ha: µd > D0 ]
• p-value = P(z < zc) [or P(z > zc)]
• where zc is the calculated of the test statistic.
31
 Paired-Difference Test of Hypothesis
for µd = (µ1 – µ2) (Cont.)
One-Tailed Test
H0: µd = D0
Ha: µd < D0 [or Ha: µd > D0 ]
• Small Sample, Student’s t-Statistic
xd  D0
Test statistic: t 
sd nd
Rejection region: t < –t
[or t > t when Ha: µd > D0 ]
• p-value = P(t < tc) [or P(t > tc)]
where t is based on (nd – 1) degrees of freedom.
32
 Paired-Difference Test of Hypothesis
for µd = (µ1 – µ2) (Cont.)
Two-Tailed Test
H0: µd = D0
Ha: µd ≠ D0
Large Sample
xd  D0 xd  D0
Test statistic: z 
 d nd sd nd
Rejection region: |z| > z
• p-value = 2P(z > |zc|)
• where zc is the calculated of the test statistic.

33
Paired-Difference Test of Hypothesis for
µd = (µ1 – µ2) (Cont.)
Two-Tailed Test
H0: µd = D0
Ha: µd ≠ D0
Small Sample
xd  D0
Test statistic: t
sd nd
Rejection region: |t| > t
• p-value = 2P(t > |tc|)
• where t is based on (nd – 1) degrees of freedom and
tc is the calculated of the test statistic
34
 Conditions Required for Valid Large-Sample
Inferences about µd
1. A random sample of differences is selected
from the target population of differences.
2. The sample size nd is large (i.e., nd ≥ 30); due
to the Central Limit Theorem, this condition
guarantees that the test statistic will be
approximately normal regardless of the shape
of the underlying probability distribution of
the population.

35
 Conditions Required for Valid Small-Sample
Inferences about µd

1. A random sample of differences is selected

from the target population of differences.
2. The population of differences has a
distribution that is approximately normal.

36
 Paired-Difference Experiment
Data Collection Table

Observation Group 1 Group 2 Difference

1 x11 x21 d1 = x11 – x21
2 x12 x22 d2 = x12 – x22
   
i x1i x2i di = x1i – x2i
   
n x1n x2n dn = x1n – x2n

37
 Example: Comparing Mean
Salaries of Males and Females
• An experiment is conducted to compare the starting
salaries of male and female college graduates who find jobs.
Pairs are formed by choosing a male and a female with the
same major and similar grade point averages (GPAs).
• Suppose a random sample of 10 pairs is formed in this
manner and the starting annual salary of each person is
recorded. The results are shown in the table (next slide).
• Compare the mean starting salary, 1, for males with the
mean starting salary, 2, for females using a 95% confidence
interval. Interpret the results.

38
 Example: Comparing Mean
Salaries of Males and Females (cont)
• Data on Annual Salaries for Matched Pairs of College Graduates

Paired difference statistics:

xd  400
sd  435

39
 Example: Comparing Mean
Salaries of Males and Females (cont)
Note that t0.025 = 2.262 (t-Static Table) for df = nd – 1 = 9
The 95% confidence interval for d  ( 1  2 )
of this small sample is
sd sd
xd  t 2  xd  t0.025
nd nd
 435 
 400  2.262  
 10 
 400  311
 ($89, $711)
As the interval falls above 0, we infer that 1  2  0 , i.e.
the mean salary for males exceeds the mean salary for females.
40
 8.4

Comparing Two
Population Proportions:
Independent Sampling

41
Properties of the Sampling Distribution of
(p1 – p2)
1. The mean of the sampling distribution of p̂1  p̂2 is
(p1 – p2); that is,
E p̂1  p̂2   p1  p2
2. The standard deviation of the sampling distribution
of p̂1  p̂2 is
p1q1 p2 q2
 p̂1  p̂2   
n1 n2
3. If the sample sizes n1 and n2 are large, the sampling
distribution of p̂1  p̂2  is approximately normal.

42
Large-Sample (1 – )% Confidence Interval for (p1 – p2)

p1q1 p2 q2
p̂1  p̂2  z 2 p̂  p̂   p̂1  p̂2  z 2
1 2
n1

n2
p̂1q̂1 p̂2 q̂2
 p̂1  p̂2   z 2 
n1 n2

43
 Conditions Required for Valid Large-Sample
Inferences about
(p1 – p2)

1. The two samples are randomly selected in an

independent manner from the two target
populations.
2. The sample sizes, n1 and n2, are both large so that
the sampling distribution of p̂1  p̂2 will be

approximately normal. (This condition will be

satisfied if both n1 pˆ1  15, n1qˆ1  15, and
n2 pˆ 2  15, n2 qˆ2  15.)

44
 Large-Sample Test of Hypothesis
about (p1 – p2)
One-Tailed Test
H0: (p1 – p2) = 0
Ha: (p1 – p2) ≠ 0 [or Ha: (p1 – p2) > 0 ]
z
 p̂1  p̂2 
Test statistic:  p̂  p̂ 
1 2

Rejection region: z < –z

[or z > z when Ha: (p1 – p2) > 0 ]
• p-value = P(z < zc) [or P(z > zc)]
p1q1 p2 q2 1 1 x x
Note:  p̂  p̂     p̂q̂    , p̂  1 2 .
1 2
n1 n2  n1 n2  n1  n2
45
 Large-Sample Test of Hypothesis
about (p1 – p2)
Two-Tailed Test
H0: (p1 – p2) = 0
Ha: (p1 – p2) ≠ 0
z
 p̂1  p̂2 
Test statistic:  p̂  p̂ 
1 2

Rejection region: | z | > z

• p-value = 2P(z > |zc|)
p1q1 p2 q2 1 1 x1  x2
Note:  p̂  p̂     p̂q̂    , p̂  .
1 2
n1 n2  n1 n2  n1  n2

46
Example: Comparing Car Repair Rates
• A consumer advocacy group wants to determine if there is a
difference between the proportions of the two leading
automobile models that need major repairs (more than $500)
within 2 years of their purchase. A sample of 400 two-year
owners of model 1 is contacted, and a sample of 500 two-year
owners of model 2 is contacted. The numbers x1 and x2 of
owners who report that their cars needed major repairs within
the first 2 years are 53 and 78, respectively.
• Test the null hypothesis that no difference exists between the
proportions in populations 1 and 2 needing major repairs against
the alternative that a difference does exist. Use  = 0.10.

47
 Example: Comparing Car Repair Rates
(Cont.)
• Define p1 and p2 as the true proportions of model 1 and
model 2 owners, respectively, whose cars needed major
repairs within 2 years, the elements of the test are
• H0 : (p1 - p2) = 0
• Ha : (p1 - p2) ≠ 0

• Test statistic:  pˆ1  pˆ 2   0

z
 ( pˆ  pˆ )
1 2

• Rejection region: ( = 0.10):

| z | z /2  z0.05  1.645

48
 Example: Comparing Car Repair Rates
(Cont.)
• Calculate the sample proportions of owners who
needed major car repairs:
x1 53
pˆ1    0.1325
n1 400
x2 78
ˆp2    0.1560
n2 500
 pˆ1  pˆ 2   0  pˆ1  pˆ 2 
Then z  
 ( pˆ  pˆ
2) 1 1 
ˆp qˆ   
1

 n1 n2 

where

and ^ =1 − 𝑝
𝑞 ^ =1 − 0.1456=0.8544 49
 Example: Comparing Car Repair Rates
(Cont.)
• p̂ is a weighted average of p̂1 and p̂2, with more
weight given to the larger sample of model 2 owners.
• The computed value of the test statistic is
0.1325  0.1560   0.0235
z    0.99
 1 1  0.0237
(0.1456)(0.8544)   
 400 500 

Note that | z |  0.99  z /2  1.645

The samples provide insufficient evidence at  = 0.10 to
detect a difference between the proportions of the two
models that needed repairs within 2 years. 50
 Example: Comparing Car Repair Rates
(Cont.)
• The rejection region for this example:

51
 Exercise
In an exit poll, 42 of 75 men sampled supported a ballot initiative
to raise the local sales tax to build a new football stadium. In the
same poll, 41 of 85 women sampled supported the initiative.
Find and interpret the p-value for the test of hypothesis that the
proportions of men and women who support the initiative are
different.

52
 Solution for the Exercise
42 41
^ 1=
𝑝 ^
=0.56∧𝑝 2= =0. 48
75 85

The test statistic is 1.01

The p-value is
The probability of observing a value of z more contradictory to the
null hypothesis is 0.3124.

53
 8.5

Determining the Required Sample

Size

54
 Determination of Sample Size for Estimating
µ1 – µ2
To estimate (μ1 – μ2) with a given margin of error ME
and with confidence level (1 – ), use the following
formula to solve for equal sample sizes that will achieve
the desired reliability:
  2  1 2
2
z  2
  2

n1  n2 
(SE) 2
You will need to substitute estimates for the values of  12
and  22 before solving for the sample size. These
2 2
s
estimates might be sample variances 1 and 2 froms
prior sampling (e.g., a pilot sample) or from an
educated (and conservatively large) guess based on the
range–that is, s ≈ R/4.
55
 Example: μ1 – μ2 Sample Size
What sample size is needed to estimate
μ1 – μ2 with 95% confidence and a margin of error of
5.8? Assume prior experience tells us σ1 =12 and σ2
=18.

n1  n2 
1.96 
2
12 2
 182 
 53.44  54
2
(5.8)

56
 Determination of Sample Size for Estimating
p1 – p2
To estimate (p1 – p2) with a given margin of error ME
and with confidence level (1 – ), use the following
formula to solve for equal sample sizes that will achieve
the desired reliability:
 z  pq  p q 
2
 2 1 1 2 2
n1  n2 
(SE) 2
You will need to substitute estimates for the values of
p1 and p2 before solving for the sample size. These
estimates might be based on prior samples, obtained
from educated guesses, or, most conservatively,
specified as p1 = p2 = 0.5.
57
 Example: p1 – p2 Sample Size
What sample size is needed to estimate
p1 – p2 with 90% confidence and a width of 0.05?
width 0.05
ME    0.025
2 2

1.645 0.5  0.5  0.5  0.5 

2

n1  n2  2
 2164.82  2165
(0.025)

58
 8.6

Comparing Two Population

Variances: Independent Sampling

59
 F-Test for Equal
Population Variances
One-Tailed Test
H0:  2
1   2
2

Ha:  2
1   2
2 
(or Ha: 1
2
  2
2 )

2
s  s12 2
 or F  s 2 when H a :  1   2 
2
Test statistic: F 
2
2
s
1
2

Rejection region: F > F

• p-value = P(F > Fc)
where F is based on v1 = numerator degrees of
freedom and v2 = denominator degrees of freedom; v1
and v2 are the degrees of freedom for the numerator
and denominator sample variances, respectively. 60
 F-Test for Equal
Population Variances
Two-Tailed Test
H0:  2
1   2
2

Ha:  2
1   2
2
s12  s 2
2
Test statistic: F  2 when s1  s2  or 2 when s2  s1 
2 2 2 2
s2  s1 
Rejection region: F > F
• p-value = P(F* < 1/Fc) + P(F > Fc)
where F/2 is based on v1 = numerator degrees of freedom
and v2 = denominator degrees of freedom; v1 and v2 are
the degrees of freedom for the numerator and
denominator sample variances, respectively.
61
 Conditions Required for a Valid F-Test for
Equal Variances

1. Both sampled populations are normally distributed.

2. The samples are random and independent.

62
 Key Ideas
Key Words for Identifying the Target Parameter

 –  Difference in means or averages

d Paired difference in means or averages
p1 – p2 Difference in proportions, fractions,
percentages, rates
12

2
2 Ratio (or difference) in variances, spreads

63
 Key Ideas
Determining the Sample Size

  2  1 2 
2
z  2
  2
Estimating  – : n1  n2  2
(SE)

 z  pq  p q 
2
Estimating p – p: n1  n2 
 2 1 1 2 2
2
(SE)

 z   
2 2
• Estimating  2 d
d : nd 
(SE) 2
64
 Key Ideas
Conditions Required for Inferences about µ1 – µ2
Large Samples:
1. Independent random samples
2. n1 ≥ 30, n2 ≥ 30
Small Samples:
1. Independent random samples
2. Both populations normal
3.  12   22

65
 Key Ideas

Conditions Required for Inferences about 1 2
/  2
2

Large or small Samples:

1. Independent random samples
2. Both populations normal

66
 Key Ideas
Conditions Required for Inferences about µd
Large Samples:
1. Random sample of paired differences
2. nd ≥ 30

Small Samples:
1. Random sample of paired differences
2. Population of differences is normal

67
 Key Ideas
Conditions Required for Inferences about p1 – p2

Large Samples:
1. Independent random samples
2. n1p1 ≥ 15, n1q1 ≥ 15
3. n2p2 ≥ 15, n2q2 ≥ 15

68
 Key Ideas
Using a Confidence Interval for (µ1 – µ2) or (p1 – p2)
to Determine whether a Difference Exists
1. If the confidence interval includes all positive numbers
(+, +): Infer µ1> µ2 or p1 > p2
2. If the confidence interval includes all negative numbers
(–, –): Infer µ1 < µ2 or p1 < p2
3. If the confidence interval includes 0 (–, +): Infer no
evidence of a difference

69