Chapter 9: Inferences Based on Two Samples

CHAPTER 9
Section 9.1
1.
a.

E X Y E X E Y 4.1 4.5 .4 , irrespective of sample sizes.

b.

V X Y V X V Y
X Y

c.

2.

12 22 1.8 2 2.0 2

.0724 , and the s.d. of

m
n
100
100

.0724 .2691 .

A normal curve with mean and s.d. as given in a and b (because m = n = 100, the CLT
implies that both X and Y have approximately normal distributions, so X Y does
also). The shape is not necessarily that of a normal curve when m = n = 10, because the
CLT cannot be invoked. So if the two lifetime population distributions are not normal,
the distribution of X Y will typically be quite complicated.

The test statistic value is

xy

z 1.96 . We compute

s12 s 22 , and H0 will be rejected if either z 1.96 or

m n
z

42,500 40,400
1900
2200

45
45
2

2100
4.85
433.33
. Since 4.85 >

1.96, reject H0 and conclude that the two brands differ with respect to true average tread lives.

3.

The test statistic value is

z 2.33 . We compute

x y 5000
s12 s 22

m n

, and H0 will be rejected at level .01 if

42,500 36,800 5000

2200 2 1500

45
45

700
1.76
396.93
, which is

not > 2.33, so we dont reject H0 and conclude that the true average life for radials does not
exceed that for economy brand by significantly more than 500.

23

Chapter 9: Inferences Based on Two Samples

4.
a.

From Exercise 2, the C.I. is

x y 1.96

s12 s 22

2100 1.96 433.33 2100 849.33

m n

1250.67,2949.33 . In the context of this problem situation, the interval is

moderately wide (a consequence of the standard deviations being large), so the
information about 1 and 2 is not as precise as might be desirable.
b. From Exercise 3, the upper bound is
5700 1.645 396.93 5700 652.95 6352.95 .
5.
a.

Ha says that the average calorie output for sufferers is more than 1 cal/cm 2/min below that

12 22
.2 2 .4 2 .1414 , so

m
n
10
10

.64 2.05 1
z
2.90 . At level .01, H0 is rejected if z 2.33 ; since
.1414

for non-sufferers.

2.90 < -2.33, reject H0.

b.

P 2.90 .0019

c.

1.2 1

1 2.33
1 .92 .8212
.1414

d.

mn

.2 2.33 1.28

.2 2

65.15 , so use 66.

24

Chapter 9: Inferences Based on Two Samples

6.
a.

H0 should be rejected if z 2.33 . Since

18.12 16.87
2.56 1.96

40
32

3.53 2.33

, H0

should be rejected at level .01.

b.

1 0

1 2.33
.50 .3085
.3539

c.

2.56 1.96
1
1.96

.1169
.0529 n 37.06 , so use
2
40
n
n
1.645 1.28
n = 38.

d. Since n = 32 is not a large sample, it would no longer be appropriate to use the large
sample z test of Section 9.1. A small sample t procedure should be used (Section 9.2),
and the appropriate conclusion would follow. Note, however, that the test statistic of 3.53
would not change, and thus it shouldnt come as a surprise that we would still reject H 0 at
the .01 significance level.
7.
1

Parameter of interest: 1 2 the true difference of means for males and

females on the Boredom Proneness Rating. Let 1 mens average and 2

womens average.
H0: 1 2 0

H a: 1 2 0

4
5
6
7

x y o x y 0

s12 s 22

m n

s12 s 22

m n

RR: z 1.645

10.40 9.26 0 1.83

4.83 2 4.68 2

97
148

Reject H0. The data indicates the average Boredom Proneness Rating is higher for
males than for females.

25

Chapter 9: Inferences Based on Two Samples

8.
a.
1

2
3
4
5
6
7

Parameter of interest: 1 2 the true difference of mean tensile strength of the

1064 grade and the 1078 grade wire rod. Let 1 1064 grade average and 2
1078 grade average.
H0: 1 2 10
Ha: 1 2 10

x y o x y 10

s12 s 22

m n

s12 s 22

m n

RR: p value

107.6 123.6 10
2

1.3
2.0

129 129

6
28.57
.210

For a lower-tailed test, the p-value = 28.57 0 , which is less than any ,
so reject H0. There is very compelling evidence that the mean tensile strength of the
1078 grade exceeds that of the 1064 grade by more than 10.

b. The requested information can be provided by a 95% confidence interval for 1 2 :

x y 1.96
9.
a.

s12 s 22

16 1.96 .210 16.412,15.588 .

m
n

point estimate x y 19.9 13.7 6.2 . It appears that there could be a

difference.

b.
H0: 1 2 0 ,Ha: 1 2 0 ,

19.9 13.7
2

39.1
15.8

60
60

6.2
1.14
5.44
, and the

p-value = 2[P(z > 1.14)] = 2( .1271) = .2542. The p value is larger than any reasonable
, so we do not reject H0. There is no significant difference.
c.

No. With a normal distribution, we would expect most of the data to be within 2 standard
deviations of the mean, and the distribution should be symmetric. 2 sds above the mean
is 98.1, but the distribution stops at zero on the left. The distribution is positively
skewed.

d. We will calculate a 95% confidence interval for , the true average length of stays for
patients given the treatment. 19.9 1.96

10.
26

39.1
19.9 9.9 10.0,21.8
60

Chapter 9: Inferences Based on Two Samples

a.

The hypotheses are H0: 1 2 5 and Ha: 1 2 5 . At level .001, H0 should be

rejected if z 3.08 . Since z

65.6 59.8 5 2.89 3.08 , H

.2272

cannot be

rejected in favor of Ha at this level, so the use of the high purity steel cannot be justified.
b.

11.

1 2 o 1 , so 3.08
.53 .2891
.
2272

X Y z

/2

s12 s 22 . Standard error = s . Substitution yields

n
m n

x y z / 2 SE1 2 SE 2 2
5.5 3.8 1.96 0.3 2

. Using .05, z / 2 1.96 , so

0.2

0.99,2.41 . We are 95% confident that the

true average blood lead level for male workers is between 0.99 and 2.41 higher than the
corresponding average for female workers.
12.

The C.I. is

x y 2.58

s12 s 22

8.77 2.58 .9104 8.77 2.46

m
n

11.23,6.31 . With 99% confidence we may say that the true difference between
the average 7-day and 28-day strengths is between -11.23 and -6.31 N/mm2.
13.

1 2 .05 , d = .04, .01, .05 , and the test is one-tailed, so

2

.0025 .0025 2.33 1.645

n

.0016

14.

49.38 , so use n = 50.

The appropriate hypotheses are H0: 0 vs. Ha: 0 , where 2 1 2 . ( 0 is

equivalent to 2 1 2 , so normal is more than twice schizophrenic) The estimator of

is

4

, is the square root of

2 X Y , with Var 4Var X Var Y

2
1

2
2

2
2
Var , and is obtained by replacing each i with S i . The test statistic is then

(since o 0 ), and H0 is rejected if z 2.33. With

2 2.69 6.35 .97 and

4.03
4 2.3

43
45
2

.9236 ,

.97
1.05 ; Because 1.05 > -2.33, H0 is not rejected.
.9236

15.

27

Chapter 9: Inferences Based on Two Samples

a.

1 2 o
increases (the numerator is

1 2 o
2 o

decreases, so z 1

positive), so z

As either m or n increases,

decreases, so

decreases.
b. As decreases, z increases, and since z is the numerator of n , n increases also.

16.

xy
s12 s 22

n n

.2

2 . For n = 100, z = 1.41 and p-value = 21 1.41 .1586 .

n

For n = 400, z = 2.83 and p-value = .0046. From a practical point of view, the closeness of
x and y suggests that there is essentially no difference between true average fracture
toughness for type I and type I steels. The very small difference in sample averages has been
magnified by the large sample sizes statistical rather than practical significance. The pvalue by itself would not have conveyed this message.

Section 9.2
17.

a.

b.

52
10
2 2

106
2

c.

2
2

37.21
17.43 17
.694 1.44

24.01
21.7 21
.694 .411

7.84
18.27 18
.018 .411

5
10

62
10

52
10
2 2

156
2

5
10

62
15

22
10
2 2

14

156
2

2
10

62
15

14
28

Chapter 9: Inferences Based on Two Samples

d.

18.

52
12
2 2

624
2

5
12

62
24

11

23

12.84
26.05 26
.395 .098

With H0: 1 2 0 vs. Ha: 1 2 0 , we will reject H0 if

.164 2
6
2 2
.164
6

.2405

.240 2
5

6.8 6

, and the test statistic

22.73 21.95
.164 2
6

p value .

.2405

.78
6.17 leads to a p-value of 2[ P(t > 6.17)] < 2(.0005)
.1265

=.001, which is less than most reasonable ' s , so we reject H0 and conclude that there is a
difference in the densities of the two brick types.

19.

For the given hypotheses, the test statistic

and the d.f. is

4.2168 4.8241 2
4.2168 2 4.8241 2
5

115 .7 129.3 10
5.032
6

9.96

5.386

3.6
1.20 ,
3.007

, so use d.f. = 9. We will reject H0 if

t t .01,9 2.764; since 1.20 > -2.764, we dont reject H0.

20.

We want a 95% confidence interval for 1 2 . t .025,9 2.262 , so the interval is

13.6 2.262 3.007 20.40,6.80 . Because the interval is so wide, it does
not appear that precise information is available.

21.

Let 1 the true average gap detection threshold for normal subjects, and 2 the

corresponding value for CTS subjects. The relevant hypotheses are H0: 1 2 0 vs. Ha:

1 2 0 , and the test statistic t

d.f.

.0351125 .07569 2
.0351125 2 .07569 2
7

1.71 2.53
.82

2.46 . Using
.0351125 .07569 .3329

15.1

29

, or 15, the rejection region is

Chapter 9: Inferences Based on Two Samples

t t .01,15 2.602 . Since 2.46 is not 2.602 , we fail to reject H0. We have
insufficient evidence to claim that the true average gap detection threshold for CTS subjects
exceeds that for normal subjects.

30

Chapter 9: Inferences Based on Two Samples

Let 1 the true average strength for wire-brushing preparation and let 2 the average
strength for hand-chisel preparation. Since we are concerned about any possible difference
between the two means, a two-sided test is appropriate. We test H 0 : 1 2 0 vs.

22.

H a : 1 2 0 . We need the degrees of freedom to find the rejection region:

1.58 2
12
2
1.58 2
12

4.1201

4.012
5

11

2.3964
14.33
, which we round down to 14, so
.0039 .1632

11

we reject H0 if t t .025,14 2.145 . The test statistic is

19.20 23.13

1.582
12

4.012
12

3.93
3.159 , which is 2.145 , so we reject H
1.2442

and

conclude that there does appear to be a difference between the two population average
strengths.
23.
a.
Normal Probability Plot for Poor Quality Fabric

.999

.999

.99
.95

.99
.95

Probability

Probability

Normal Probability Plot for High Quality Fabric

.80
.50
.20

.80
.50
.20
.05

.05
.01

.01

.001

.001

0.8

1.3

1.8

1.0

2.3

1.5

2.0

2.5

P:

H:
Av erage: 1.50833
StDev : 0.444206
N: 24

Av erage: 1.58750
StDev : 0.530330
N: 24

Anderson-Darling Normality Test

A-Squared: 0.396
P-Value: 0.344

Anderson-Darling Normality Test

A-Squared: -10.670
P-Value:
1.000

Using Minitab to generate normal probability plots, we see that both plots illustrate
sufficient linearity. Therefore, it is plausible that both samples have been selected from
normal population distributions.

31

Chapter 9: Inferences Based on Two Samples

b.
Comparative Box Plot for High Quality and Poor Quality Fabric

Poor
Quality

High
Quality

0.5

1.5

2.5

extensibility (%)

The comparative boxplot does not suggest a difference between average extensibility for
the two types of fabrics.

c.

We test H 0 : 1 2 0 vs. H a : 1 2 0 . With degrees of freedom

.0433265 2
.00017906

10.5 , which we round down to 10, and using significance level .

05 (not specified in the problem), we reject H0 if t t .025,10 2.228 . The test

.08
.38 , which is not 2.228 in absolute value, so we
statistic is t
.0433265
cannot reject H0. There is insufficient evidence to claim that the true average
extensibility differs for the two types of fabrics.
24.
a.

95% upper confidence bound: x + t.05,65-1SE = 13.4 + 1.671(2.05) = 16.83 seconds

b. Let 1 and 2 represent the true average time spent by blackbirds at the experimental and
natural locations, respectively. We wish to test H0: 1 2 = 0 v. Ha: 1 2 > 0. The
13.4 9.7
relevant test statistic is t
= 1.37, with estimated df =
2.05 2 1.76 2

( 2.05 2 1.76 2 ) 2
2.05 4 1.76 4

64
49

112.9. Rounding to t = 1.4 and df = 120, the tabulated P-value is

very roughly .082. Hence, at the 5% significance level, we fail to reject the null
hypothesis. The true average time spent by blackbirds at the experimental location is not
statistically significantly higher than at the natural location.

32

Chapter 9: Inferences Based on Two Samples

c.

25.

95% CI for silvereyes average time blackbirds average time at the natural location:
(38.4 9.7) (2.00) 1.76 2 5.06 2 = (17.96 sec, 39.44 sec). The t-value 2.00 is
based on estimated df = 55.

We calculate the degrees of freedom

5 .5 2
28
2 2
5 .5
28

731.8

27

7 .8 2
31

53.95

, or about 54

30

(normally we would round down to 53, but this number is very close to 54 of course for this
large number of df, using either 53 or 54 wont make much difference in the critical t value)
so the desired confidence interval is

91.5 88.3 1.68

5.5 2
28

731.8

3.2 2.931 .269,6.131 . Because 0 does not lie inside this interval, we can be
reasonably certain that the true difference 1 2 is not 0 and, therefore, that the two
population means are not equal. For a 95% interval, the t value increases to about 2.01 or so,
which results in the interval 3.2 3.506 . Since this interval does contain 0, we can no
longer conclude that the means are different if we use a 95% confidence interval.

26.

Let 1 the true average potential drop for alloy connections and let 2 the true
average potential drop for EC connections. Since we are interested in whether the potential
drop is higher for alloy connections, an upper tailed test is appropriate. We test
H 0 : 1 2 0 vs. H a : 1 2 0 . Using the SAS output provided, the test statistic,
when assuming unequal variances, is t = 3.6362, the corresponding df is 37.5, and the p-value
for our upper tailed test would be (two-tailed p-value) =

1
2

.0008 .0004 . Our p-

value of .0004 is less than the significance level of .01, so we reject H0. We have sufficient
evidence to claim that the true average potential drop for alloy connections is higher than that
for EC connections.
27.

The approximate degrees of freedom for this estimate are

2
11.32
8.32
6
8
2
2 2
, which we round down to 8, so
11.3
8.32
6
8

893.59
8.83
101.175

t .025,8 2.306 and the desired interval is

40.3 21.4 2.306

11.32
6

8.83 18.9 2.306 5.4674

2

18.9 12.607 6.3,31.5 . Because 0 is not contained in this interval, there is

strong evidence that 1 2 is not 0; i.e., we can conclude that the population means are not
33

Chapter 9: Inferences Based on Two Samples

equal. Calculating a confidence interval for 2 1 would change only the order of
subtraction of the sample means, but the standard error calculation would give the same result
as before. Therefore, the 95% interval estimate of 2 1 would be ( -31.5, -6.3), just the
negatives of the endpoints of the original interval. Since 0 is not in this interval, we reach
exactly the same conclusion as before; the population means are not equal.
28.

We will test the hypotheses: H 0 : 1 2 10 vs. H a : 1 2 10 . The test statistic is

x y 10

2.752
10

2.75 2
10
2
2.75 2
10

4.44 2
5

4.445

4.44 2
5

4.5
2.08
2.17

The degrees of freedom

22.08
5.59 5
, and the p-value from table A.8 is
3.95

approx .045, which is < .10 so we reject H0 and conclude that the true average lean angle for
older females is more than 10 degrees smaller than that of younger females.

29.

Let 1 the true average compression strength for strawberry drink and let 2 the true
average compression strength for cola. A lower tailed test is appropriate. We test

H 0 : 1 2 0 vs. H a : 1 2 0 . The test statistic is t

44.4 2
29.4 2 15 2
14

14

29.4 15

2.10 .

1971.36
25.3
, so use df=25. The p-value
77.8114

14

P(t 2.10) .023 . This p-value indicates strong support for the alternative
hypothesis. The data does suggest that the extra carbonation of cola results in a higher
average compression strength.
30.
a.

We desire a 99% confidence interval. First we calculate the degrees of freedom:

2
2 .2 2
4 .3 2
26
26
2
2
, which we would round down to 37, except that
2 .2 2
4 .3 2
26
26

26

37.24

26

there is no df = 37 row in Table A.5. Using 36 degrees of freedom (a more conservative

choice), t .005, 36 2.719 , and the 99% C.I. is

34

Chapter 9: Inferences Based on Two Samples

33.4 42.8 2.719

9.4 2.576 11.98,6.83

2.2 2
4.32
. We
26
26
are 99% confident that the true average load for carbon beams exceeds that for fiberglass
beams by between 6.83 and 11.98 kN.

b. The upper limit of the interval in part a does not give a 99% upper confidence bound.
The 99% upper bound would be 9.4 2.434 .9473 7.09 , meaning that the
true average load for carbon beams exceeds that for fiberglass beams by at least 7.09 kN.

35

Chapter 9: Inferences Based on Two Samples

31.
a.
Comparative Box Plot for High Range and Mid Range
470

mid range

460
450
440
430
420
mid range

high range

The most notable feature of these boxplots is the larger amount of variation present in the
mid-range data compared to the high-range data. Otherwise, both look reasonably
symmetric with no outliers present.
b. Using df = 23, a 95% confidence interval for mid range high range is

438.3 437.45 2.069

6.1183 .85 8.69 7.84,9.54 .

15.12
17

Since plausible values for mid range high range are both positive and negative (i.e.,
the interval spans zero) we would conclude that there is not sufficient evidence to suggest
that the average value for mid-range and the average value for high-range differ.

32.

Let 1 the true average proportional stress limit for red oak and let 2 the true
average proportional stress limit for Douglas fir. We test H 0 : 1 2 1 vs.

H a : 1 2 1 . The test statistic is

degrees of freedom

8.48 6.65 1

.79 2
14

.2084 2

.79 2
14

13

1.28 2
10

1.28 2
10

13.85 13

1.83
.2084

1.818 .

, the p-value = P(t > 1.8) = .

048. We would reject H0 at significance levels greater than .046 (e.g., the standard 5%
significance level). At = .05, there is sufficient evidence to claim that true average
proportional stress limit for red oak exceeds that of Douglas fir by more than 1 MPa.

36

With

Chapter 9: Inferences Based on Two Samples

33.

Let 1 and 2 represent the true mean body mass decrease for the vegan diet and the control
diet, respectively. We wish to test the hypotheses H0: 1 2 1 v. Ha: 1 2 > 1. The
(5.8 3.8) 1
t
relevant test statistic is
3.2 2 2.8 2 = 1.33, with estimated df = 60 using the

32
32
formula. Rounding to t = 1.3, Table A.8 gives a one-sided P-value of .098 (a computer will
give the more accurate P-value of .094). Since our P-value > = .05, we fail to reject H0 at the
5% level. We do not have statistically significant evidence that the true average weight loss
for the vegan diet exceeds that for the control diet by more than 1 kg.

34.
a.

Following the usual format for most confidence intervals: statistic (critical value)
(standard error), a pooled variance confidence interval for the difference between two
means is

x y t / 2 , m n 2 s p

1
m

1n .

b. The sample means and standard deviations of the two samples are x 13.90 ,
s1 1.225 , y 12.20 , s 2 1.010 . The pooled variance estimate is s 2p

m 1 2
n 1 2 4 1
4 1
2
2
s1
s2
1.225
1.010
m n 2
m n 2
4 4 2
4 4 2
1.260 , so s p 1.1227 . With df = m+n-1 = 6 for this interval, t .025, 6 2.447

and the desired interval is

13.90 12.20 2.447 1.1227

1
4

1
4

1.7 1.943 .24,3.64 . This interval contains 0, so it does not support the
conclusion that the two population means are different.
c.

Using the two-sample t interval discussed earlier, we use the CI as follows: First, we need
to calculate the degrees of freedom.
2
1.2252
1.012
4
4
2
2
so t .025 , 5 2.571 . Then the
1.2252
1.012
4
4

.3971
5.78 5
.0686

interval is

13.9 12.2 2.571

1.70 2.571 .7938 .34,3.74

1.225 2
1.012
4
4
. This interval is slightly wider, but it still supports the same conclusion.

37

Chapter 9: Inferences Based on Two Samples

35.

There are two changes that must be made to the procedure we currently use. First, the
equation used to compute the value of the t test statistic is:

x y
sp

1 1 where sp is

m n

defined as in Exercise 34 above. Second, the degrees of freedom = m + n 2. Assuming

equal variances in the situation from Exercise 33, we calculate sp as follows:

7
9
2
2
2.6
2.5 2.544 . The value of the test statistic is, then,
16
16

sp

32.8 40.5 5
1 1
2.544

8 10

2.24 2.2

. The degrees of freedom = 16, and the p-

value is P ( t < -2.2) = .021. Since .021 > .01, we fail to reject H0.

Section 9.3
36.

d 7.25 , s D 11.8628

Parameter of Interest: D true average difference of breaking load for fabric in

unabraded or abraded condition.

H0 : D 0
Ha : D 0

RR: t t .01, 7 2.998

Fail to reject H0. The data does not indicate a significant mean difference in breaking
load for the two fabric load conditions.

d D
sD / n

d 0
sD / n

7.25 0
11 .8628 / 8

1.73

38

Chapter 9: Inferences Based on Two Samples

37.
a.

This exercise calls for paired analysis. First, compute the difference between indoor and
outdoor concentrations of hexavalent chromium for each of the 33 houses. These 33
differences are summarized as follows: n = 33, d .4239 , s d .3868 , where d =
(indoor value outdoor value). Then t .025,32 2.037 , and a 95% confidence interval
for the population mean difference between indoor and outdoor concentration is

.3868

.4239 2.037

.4239 .13715 .5611,.2868 . We can

33

be highly confident, at the 95% confidence level, that the true average concentration of
hexavalent chromium outdoors exceeds the true average concentration indoors by
between .2868 and .5611 nanograms/m3.
b. A 95% prediction interval for the difference in concentration for the 34th house is

d t.025,32 s d 1 1n .4239 2.037 .3868 1 331 1.224,.3758 .

This prediction interval means that the indoor concentration may exceed the outdoor
concentration by as much as .3758 nanograms/m3 and that the outdoor concentration may
exceed the indoor concentration by a much as 1.224 nanograms/m3, for the 34th house.
Clearly, this is a wide prediction interval, largely because of the amount of variation in
the differences.
38.
a.

The median of the Normal data is 46.80 and the upper and lower quartiles are 45.55
and 49.55, which yields an IQR of 49.55 45.55 = 4.00. The median of the High data
is 90.1 and the upper and lower quartiles are 88.55 and 90.95, which yields an IQR of
90.95 88.55 = 2.40. The most significant feature of these boxplots is the fact that their
locations (medians) are far apart.

Comparative Boxplots
for Normal and High Strength Concrete Mix
90
80
70
60
50
40
High:

Normal:

39

Chapter 9: Inferences Based on Two Samples

b. This data is paired because the two measurements are taken for each of 15 test conditions.
Therefore, we have to work with the differences of the two samples. A normal
probability plot of the 15 differences shows that the data follows (approximately) a
straight line, indicating that it is reasonable to assume that the differences follow a normal
distribution. Taking differences in the order Normal High, we find d 42.23 ,
and s d 4.34 . With t .025,14 2.145 , a 95% confidence interval for the difference
between the population means is
4.34
42.23 2.404 44.63,39.83 . Because
42.23 2.145
15

0 is not contained in this interval, we can conclude that the difference between the
population means is not 0; i.e., we conclude that the two population means are not equal.
39.
a.

A normal probability plot shows that the data could easily follow a normal distribution.

b. We test H 0 : d 0 vs. H a : d 0 , with test statistic

t

d 0

167.2 0

2.74 2.7 . The two-tailed p-value is 2[ P( t > 2.7)] =

sD / n
228 / 14
2[.009] = .018. Since .018 < .05, we can reject H0 . There is strong evidence to support
the claim that the true average difference between intake values measured by the two
methods is not 0. There is a difference between them.
40.

From the data, n = 10, d = 105.7, sd = 103.845.

a. Let d = true mean difference in TBBMC, postweaning minus lactation. We wish to test
105.7 25
the hypotheses H0: d 25 v. Ha: d > 25. The test statistic is t
=
103.845 / 10
2.46; at 9df, the corresponding P-value is around .018. Hence, at the 5% significance
level, we reject H0 and conclude that true average TBBMC during postweaning does
exceed the average during lactation by more than 25 grams.
b. A 95% upper confidence bound for d = d + t.05,9sd/
10 = 165.89 grams.
c.

41.

n = 105.7 + 1.833(103.845)/

No. If we pretend the two samples are independent, the new standard error is is roughly
235, far greater than 103.845/ 10 . In turn, the resulting t statistic is just t = 0.45, with
estimated df = 17 and P-value = .329 (all using a computer).

We test H 0 : d 5 vs. H a : d 5 . With d 7.600 , and s d 4.178 ,

7.600 5
2.6
t

1.87 1.9 . With degrees of freedom n 1 = 8, the

1.39
4.178 / 9
corresponding p-value is P( t > 1.9 ) = .047. We would reject H0 at any alpha level greater
than .047. So, at the typical significance level of .05, we would reject H0, and conclude that
the data indicates that the higher level of illumination yields a decrease of more than 5
seconds in true average task completion time.
40

Chapter 9: Inferences Based on Two Samples

42.
1

Parameter of interest: d denotes the true average difference of spatial ability in

brothers exposed to DES and brothers not exposed to DES. Let
d exp osed un exp osed .

2
3

H0 : D 0
Ha : D 0

RR: P-value < .05, df = 9

A.8)
Reject H0. The data supports the idea that exposure to DES reduces spatial ability.

d D

sD / n

d 0
sD / n

12.6 13.7 0 2.2 , with corresponding p-value .028 (from Table

0 .5

43.
a.

Although there is a jump in the middle of the Normal Probability plot, the data follow a
reasonably straight path, so there is no strong reason for doubting the normality of the
population of differences.

b. A 95% lower confidence bound for the population mean difference is:

sd

23.18

38.60 1.761

d t .05,14

38.60 10.54 49.14 .

15

We are 95% confident that the true mean difference between age at onset of Cushings
disease symptoms and age at diagnosis is greater than -49.14.
c.
44.

A 95% upper confidence bound for the corresponding population mean difference is
38.60 + 10.54 = 49.14.

We need to check the differences to see if the assumption of normality is plausible. A normal
probability plot validates our use of the t distribution. A 95% upper confidence bound for D

sd

is d t .05,15

508.645
2635.63 222.91 = 2858.54.
16

2635.63 1.753

We are 95% confident that the true mean difference between modulus of elasticity after 1
minute and after 4 weeks is at most 2858.54.
45.

From the data, n = 12, d = 0.73, sd = 2.81.

a. Let d = the true mean difference in strength between curing under moist conditions and
laboratory drying conditions. A 95% CI for d is d t.025,11sd/ n = 0.73 2.201(2.81)/
10 = (2.52 MPa, 1.05 MPa). In particular, this interval estimate includes the value
zero, suggesting that true mean strength is not significantly different under these two
conditions.
b. Since n = 12, we must check that the differences are plausibly from a normal population.
The normal probability plot below strongly substantiates that condition.
41

Chapter 9: Inferences Based on Two Samples

Normal Probability Plot of Differences

Normal
99

95
90

Percent

80
70
60
50
40
30
20
10
5

46.

-7.5

-5.0

-2.5

0.0
Differences

2.5

5.0

x1 , y1 6,5 , x 2 , y 2 15,14 , x3 , y 3 1,0 , and x 4 , y 4 21,20 ,

With

d 1 and s d 0 (the dIs are 1, 1, 1, and 1), while s1 = s2 = 8.96, so sp = 8.96 and t = .16.

Section 9.4
47.

1 .150 , and p 2 .300 ,

H0 will be rejected if z z .01 2.33 . With p
p

30 80
210

.263 , and q .737 . The calculated test statistic is

200 600 800

.150 .300

.263 .737

1
200

1
600

.150
4.18 . Because 4.18 2.33 , H0 is
.0359

rejected; the proportion of those who repeat after inducement appears lower than those who
repeat after no inducement.
48.
a.

1
H0 will be rejected if z 1.96 . With p
p 2

63
.2100 , and
300

75
63 75
.4167 , p
.2875 ,
180
300 180

.2100 .4167

.2875 .7125 3001 1801

.2067
4.84 .
.0427

Since 4.84 1.96 ,

H0 is rejected.
b.

p .275 and

.0432 , so power =

1.96 .0421 .2
1.96 .0421 .2
1


.0432
.0432

1 6.54 2.72 .9967 .

42

Chapter 9: Inferences Based on Two Samples

49.
1

H 0 : p1 p 2 0
H a : p1 p 2 0

Reject H0 if p-value < .10

6
7

50.

Parameter of interest: p1 p2 = true difference in proportions of those responding to

two different survey covers. Let p1 = Plain, p2 = Picture.

p 1 p 2

p q m1 1n
104
207

109
213

213 207
420
420 2071 2131

.1910 ; p-value = .4247

Fail to Reject H0. The data does not indicate that plain cover surveys have a lower
response rate.

Let .05 . A 95% confidence interval is

224

395
126
266 1.96

p 1 p 2 z / 2

p1q1
m

p2nq2

224 171
395
395 126
140
266 266

395

.0934 .0774 .0160,.1708 .

266

51.
a.

Let p1 and p2 denote the true incidence rates of GI problems for the olestra and control
groups, respectively. We wish to test H0: p1 2 = 0 v. Ha: p1 p2 0. The pooled

proportion is p
statistic is z =

529(.176) 563(.158)
= .1667, from which the relevant test
529 563
.176 .158

(.1667)(.8333)[529 1 563 1 ]

= 0.78. The two-sided P-value is

2P(Z 0.78) = .433 > = .05, hence we fail to reject the null hypothesis. The data do not
suggest a statistically significant difference between the incidence rates of GI problems
between the two groups.
b.

52.

1.96
n

(.35)(1.65) / 2 1.28 (.15)(.85) (.2)(.8)

(.05) 2
common sample size of m = n = 1211 would be required.

1210.39 , so a

Let p1 = true proportion of irradiated bulbs that are marketable; p2 = true proportion of
untreated bulbs that are marketable; The hypotheses are H 0 : p1 p 2 0 vs.

H 0 : p1 p 2 0 . The test statistic is

43

p 1 p 2

p q
1
m

1
n

1
. With p

153
.850 , and
180

Chapter 9: Inferences Based on Two Samples

p 2

119
272
.661 , p
.756 , z
180
360

.850 .661

.756 .244 1801 1801

.189
4.2 .
.045

The p-value = 1 4.2 0 , so reject H0 at any reasonable level. Radiation appears to

be beneficial.
53.
a.

A 95% large sample confidence interval formula for ln is

mx n y

. Taking the antilogs of the upper and lower bounds

mx
ny
gives the confidence interval for itself.
ln z / 2

b.

189
11, 034
104
11, 037

1.818 , ln .598 , and the standard deviation is

10,845
10,933

.1213 , so the CI for ln is

11,034 189 11,037 104
.598 1.96 .1213 .360,.836 . Then taking the antilogs of the two bounds
gives the CI for to be 1.43,2.31 . We are 95% confident that people who do not
take the aspirin treatment are between 1.43 and 2.31 times more likely to suffer a heart
attack than those who do. This suggests aspirin therapy may be effective in reducing the
risk of a heart attack.
54.
a.

The after success probability is p1 + p3 while the before probability is p1 + p2, so p1 +

p3 > p1 + p2 becomes p3 > p2; thus we wish to test H 0 : p 3 p 2 versus H a : p3 p 2 .

b. The estimator of (p1 + p3) (p1 + p2) is

c.

X1 X 3 X1 X 2
n

X3 X2
.
n

p p3
X3 X2
2
, which is estimated by
n
n

When H0 is true, p2 = p3, so Var

X3 X2
X X2
n
p 2 p 3
3
. The Z statistic is then
.
p 2 p 3
X2 X3
n
n
d. The computed value of Z is

200 150
200 150

2.68 , so P 1 2.68 .0037 .

At level .01, H0 can be rejected but at level .001 H0 would not be rejected.

44

Chapter 9: Inferences Based on Two Samples

55.

p 1

15 7
29
.550 , p 2
.690 , and the 95% C.I. is
40
42

.550 .690 1.96 .106 .14 .21 .35,.07 .

56.

2.7719
.25 .25
, so L=.1 requires

n
n
n

Using p1 = q1 = p2 = q2 = .5, L 21.96

n=769.

Section 9.5
57.
a.

From Table A.9, column 5, row 8, F.01,5,8 3.69 .

b. From column 8, row 5, F.01,8,5 4.82 .

c.

F.95,5,8

d.

F.95,8,5

e.

F.01,10,12 4.30

f.

F.99,10,12

g.

F.05,6, 4 6.16 , so P F 6.16 .95 .

F.05,8,5

.207 .

1
.271
F.05,5,8

1
F.01,12,10

h. Since F.99,10,5

1
.212 .
4.71

1
.177 ,
5.64

P .177 F 4.74 P F 4.74 P F .177 .95 .01 .94 .

58.
a.

Since the given f value of 4.75 falls between F.05,5,10 3.33 and F.01,5,10 5.64
, we can say that the upper-tailed p-value is between .01 and .05.

b. Since the given f of 2.00 is less than F.10,5,10 2.52 , the p-value > .10.
c.

The two tailed p-value = 2 P F 5.64 2(.01) .02 .

45

Chapter 9: Inferences Based on Two Samples

d. For a lower tailed test, we must first use formula 9.9 to find the critical values:
1
1
F.90,5,10
.3030 , F.95,5,10
.2110 ,
F.10,10,5
F.05,10,5

F.99,5,10

1
.0995 . Since .0995 < f = .200 < .2110, .01 < p-value < .05
F.01,10,5

(but obviously closer to .05).

e.

59.

There is no column for numerator d.f. of 35 in Table A.9, however looking at both df = 30
and df = 40 columns, we see that for denominator df = 20, our f value is between F.01 and
F.001. So we can say .001< p-value < .01.

We test

H 0 : 12 22 vs. H a : 12 22 . The calculated test statistic is

2.75 2
4.44 2

.384 . With numerator d.f. = m 1 = 10 1 = 9, and denominator d.f. = n

1 = 5 1 = 4, we reject H0 if f F.05,9, 4 6.00 or

f F.95,9, 4 1

F.05, 4,9

3.63

.275 .

Since .384 is in neither rejection region, we do

not reject H0 and conclude that there is no significant difference between the two standard
deviations.

60.

With 1 true standard deviation for not-fused specimens and 2 true standard

deviation for fused specimens, we test H 0 : 1 2 vs. H a : 1 2 . The calculated test

statistic is f

277.3 2
205.9 2

1.814 . With numerator d.f. = m 1 = 10 1 = 9, and

denominator d.f. = n 1 = 8 1 = 7, f 1.814 2.72 F.10,9,7 . We can say that the pvalue > .10, which is obviously > .01, so we cannot reject H0. There is not sufficient
evidence that the standard deviation of the strength distribution for fused specimens is smaller
than that of not-fused specimens.

61.

Let

12 variance in weight gain for low-dose treatment, and 22 variance in weight

gain for control condition. We wish to test H 0 : 1 2 vs. H a : 1 2 . The test statistic
2

is

s12
, and we reject H0 at level .05 if f F.05,19, 22 2.08 .
s 22

54 2
32 2

2.85 2.08 , so reject H0 at level .05. The data does suggest that there is

more variability in the low-dose weight gains.

62.

For the hypotheses H 0 : 1 2 versus H a : 1 2 , we find a test statistic of f = 1.22.

At df = (47,44) (40,40), 1.22 < 1.51 indicates the P-value is greater than 2(.10) = .20.
46

Chapter 9: Inferences Based on Two Samples

Hence, H0 is not rejected. The data does not suggest a significant difference in the two
population variances.

47

Chapter 9: Inferences Based on Two Samples

63.

S 2 / 2
P F1 / 2,m 1,n 1 12 12 F / 2,m 1,n 1 1 . The set of inequalities inside the
S2 / 2

2
S 2 F1 / 2,m 1, n 1 22 S 22 F / 2,m 1, n 1
parentheses is clearly equivalent to
2
.
S12
1
S12
22
2
2
Substituting the sample values s1 and s 2 yields the confidence interval for 2 , and
1
taking the square root of each endpoint yields the confidence interval for
we need F.05, 3, 3 9.28 and F.95,3,3
074, the C. I. for

64.

22
12

1
.108 . Then with s1 = .160 and s2 = .
9.28

is (.023, 1.99), and for

A 95% upper bound for

2
is
1

2
. m = n = 4, so
1

s 22 F.05,9,9
s12

2
is (.15, 1.41).
1

3.59 2 3.18
.79 2

8.10 . We are

confident that the ratio of the standard deviation of triacetate porosity distribution to that of
the cotton porosity distribution is at most 8.10.

Supplementary Exercises
65.

We test H 0 : 1 2 0 vs. H a : 1 2 0 . The test statistic is

x y
s12 s 22

m
n

807 757

27 2 412

10
10

241 2
72.9 2 168.1 2
9

15.6

50
241

50
3.22
15.524
. The approximate d.f. is

, which we round down to 15. The p-value for a two-

tailed test is approximately 2P(T > 3.22) = 2( .003) = .006. This small of a p-value gives
strong support for the alternative hypothesis. The data indicates a significant difference. Due
to the small sample sizes (10 each), we are assuming here that compression strengths for both
fixed and floating test platens are normally distributed. And, as always, we are assuming the
data were randomly sampled from their respective populations.

48

Chapter 9: Inferences Based on Two Samples

66.
a.
Comparative Boxplot of Tree Density Between
Fertilizer Plots and Control Plots
1400

Fertiliz

1300

1200

1100

1000
Fertiliz

Control

Although the median of the fertilizer plot is higher than that of the control plots, the
fertilizer plot data appears negatively skewed, while the opposite is true for the control
plot data.
b. A test of H 0 : 1 2 0 vs. H a : 1 2 0 yields a t value of -.20, and a twotailed p-value of .85. (d.f. = 13). We would fail to reject H0; the data does not indicate a
significant difference in the means.
c. With 95% confidence we can say that the true average difference between the tree density
of the fertilizer plots and that of the control plots is somewhere between 144 and 120.
Since this interval contains 0, 0 is a plausible value for the difference, which further
supports the conclusion based on the p-value.
67.

Let p1 = true proportion of returned questionnaires that included no incentive; p2 = true

proportion of returned questionnaires that included an incentive. The hypotheses are

H 0 : p1 p 2 0 vs. H 0 : p1 p 2 0 . The test statistic is

p 1

p 1 p 2

p q m1 1n

75
66
.682 , and p 2
.673 . At this point we notice that since p 1 p 2 , the
110
98

numerator of the z statistic will be > 0, and since we have a lower tailed test, the p-value will
be > .5. We fail to reject H0. This data does not suggest that including an incentive increases
the likelihood of a response.

49

Chapter 9: Inferences Based on Two Samples

68.

Summary quantities are m = 24, x 103.66 , s1 = 3.74, n = 11, y 101.11 , s2 = 3.60.

We use the pooled t interval based on 24 + 11 2 = 33 d.f.; 95% confidence requires
t.025,33 2.03 . With s 2p 13.68 and s p 3.70 , the confidence interval is

2.55 2.03 3.70

1
24

111 2.55 2.73 .18,5.28 . We are confident that

the difference between true average dry densities for the two sampling methods is between
-.18 and 5.28. Because the interval contains 0, we cannot say that there is a significant
difference between them.

69.

The center of any confidence interval for 1 2 is always x1 x 2 , so

473.3 1691.9
609.3 . Furthermore, half of the width of this interval is
2
1691.9 473.3
1082.6 . Equating this value to the expression on the right of the
2
x1 x 2

95% confidence interval formula, we find

s12 s 22
1082.6

552.35 . For a 90%

n1 n 2
1.96

interval, the associated z value is 1.645, so the 90% confidence interval is then
609.3 1.645 552.35 609.3 908.6 299.3,1517.9 .

70.
a.

A 95% lower confidence bound for the true average strength of joints with a side coating

5.96

is x t .025,9

63.23 1.833

63.23 3.45 59.78 . That is,

10

with a confidence level of 95%, the average strength of joints with a side coating is at
least 59.78 (Note: this bound is valid only if the distribution of joint strength is normal.)
b. A 95% lower prediction bound for the strength of a single joint with a side coating is

x t .025,9 s 1 1n 63.23 1.833 5.96 1 101

63.23 11.46 51.77 . That is, with a confidence level of 95%, the strength of a
single joint with a side coating would be at least 51.77.
c.

For a confidence level of 95%, a two-sided tolerance interval for capturing at least 95%
of the strength values of joints with side coating is x (tolerance critical value)s. The
tolerance critical value is obtained from Table A.6 with 95% confidence, k = 95%, and n
= 10. Thus, the interval is
63.23 3.379 5.96 63.23 20.14 43.09,83.37 . That is, we can be
highly confident that at least 95% of all joints with side coatings have strength values
between 43.09 and 83.37.

d. A 95% confidence interval for the difference between the true average strengths for the
two types of joints is 80.95 63.23 t .025,
50

9.59 2
10

5.96 2
10

. The

Chapter 9: Inferences Based on Two Samples

91.109681 35.105216 2

approximate degrees of freedom is

91.109681 2 35.105216 2
9

15.05

, so we use 15

d.f., and t .025,15 2.131 . The interval is , then,

17.72 2.131 3.57 17.72 7.61 10.11,25.33 . With 95%
confidence, we can say that the true average strength for joints without side coating
exceeds that of joints with side coating by between 10.11 and 25.33 lb-in./in.

71.

m = n = 40, x 3975.0 , s1 = 245.1, y 2795.0 , s2 = 293.7. The large sample 99%

confidence interval for 1 2 is 3975.0 2795.0 2.58

1180 .0 1560 1020,1340 .

245.12 293.7 2

40
40

The value 0 is not contained in this interval so we can

state that, with very high confidence, the value of 1 2 is not 0, which is equivalent to
concluding that the population means are not equal.
72.

This exercise calls for a paired analysis. First compute the difference between the amount of
cone penetration for commutator and pinion bearings for each of the 17 motors. These 17
differences are summarized as follows: n = 17, d 4.18 , s d 35.85 , where d =
(commutator value pinion value). Then t .025 ,16 2.120 , and the 95% confidence
interval for the population mean difference between penetration for the commutator armature
bearing and penetration for the pinion bearing is:

35.85
4.18 18.43 22.61,14.25 . We would have
17

4.18 2.120

to say that the population mean difference has not been precisely estimated. The bound on the
error of estimation is quite large. Also, the confidence interval spans zero. Because of this,
we have insufficient evidence to claim that the population mean penetration differs for the two
types of bearings.

51

Chapter 9: Inferences Based on Two Samples

73.

Since we can assume that the distributions from which the samples were taken are normal, we
use the two-sample t test. Let 1 denote the true mean headability rating for aluminum
killed steel specimens and 2 denote the true mean headability rating for silicon killed steel.
Then the hypotheses are H 0 : 1 2 0 vs. H a : 1 2 0 . The test statistic is

.66

.03888 .047203

.086083 2
.03888 2 .047203 2
29

.66
2.25 . The approximate degrees of freedom
.086083

57.5

, so we use 57. The two-tailed p-value

29

2 .014 .028 , which is less than the specified significance level, so we would reject
H0. The data supports the articles authors claim.

74.

Let 1 denote the true average tear length for Brand A and let 2 denote the true average
tear length for Brand B. The relevant hypotheses are H 0 : 1 2 0 vs.

H a : 1 2 0 . Assuming both populations have normal distributions, the two-sample t

test is appropriate. m = 16, x 74.0 , s1 = 14.8, n = 14, y 61.0 , s2 = 12.5, so the
2
14.8 2
12.5 2
16
14
2
2
approximate d.f. is
, which we round down to 27.
14.8 2
12.5 2
16
14

15

The test statistic is

74.0 61.0
14.8 2
16

1214.5

27.97

13

2.6 .

From Table A.7, the p-value = P( t > 2.6)

= .007. At a significance level of .05, H0 is rejected and we conclude that the average tear
length for Brand A is larger than that of Brand B.
75.
a.

The relevant hypotheses are H 0 : 1 2 0 vs. H a : 1 2 0 . Assuming both

populations have normal distributions, the two-sample t test is appropriate. m = 11,
x 98.1 , s1 = 14.2, n = 15, y 129.2 , s2 = 39.1. The test statistic is

31.1

18.3309 101.9207

freedom

31.1

120.252

120.252 2
18.3309 2 101.9207 2
10

14

2.84 . The approximate degrees of

18.64

, so we use 18. From Table A.8,

the two-tailed p-value 2 .006 .012 . No, obviously, the results are different.

52

Chapter 9: Inferences Based on Two Samples

b. For the hypotheses H 0 : 1 2 25 vs. H a : 1 2 25 , the test statistic
changes to t

31.1 25
120.252

.556 . With degrees of freedom 18, the p-value

P t .6 .278 . Since the p-value is greater than any sensible choice of , we

fail to reject H0. There is insufficient evidence that the true average strength for males
exceeds that for females by more than 25N.
76.
a.

The relevant hypotheses are H 0 : 1 2 0 (which is equivalent to saying

1 2 0 ) versus H a : 1 2 0 (which is the same as saying 1 2 0 ).

The pooled t test is based on d.f. = m + n 2 = 8 + 9 2 = 15. The pooled variance is

m 1 2
n 1 2 8 1
9 1
2
2
s1
s2
4.9
4.6
m n 2
m n 2
8 9 2
89 2
22.49 , so s p 4.742 . The test statistic is

s 2p

x *y *
sp

1
m

1n

18.0 11.0
4.742

1
8

19

3.04 3.0 .

From Table A.7, the p-value

associated with t = 3.0 is 2P( t > 3.0 ) = 2(.004) = .008. At significance level .05, H0 is
rejected and we conclude that there is a difference between 1 and 2 , which is
equivalent to saying that there is a difference between 1 and 2 .

b. No. The mean of a lognormal distribution is e / 2 , where and are the

parameters of the lognormal distribution (i.e., the mean and standard deviation of ln(x)).
So when 1 2 , then 1 2 would imply that 1 2 . However, when

1 2 , then even if 1 2 , the two means 1 and 2 (given by the formula

above) would not be equal.

77.

This is paired data, so the paired t test is employed. The relevant hypotheses are
H 0 : d 0 vs. H a : d 0 , where d denotes the difference between the population
average control strength minus the population average heated strength. The observed
differences (control heated) are: -.06, .01, -.02, 0, and -.05. The sample mean and standard
deviation of the differences are d .024 and s d .0305 . The test statistic is

.024
.0305

1.76 1.8 . From Table A.8, with d.f. = 5 1 = 4, the lower tailed p-

value associated with t = -1.8 is P( t < -1.8) = P( t > 1.8 ) = .073. At significance level .05, H0
should not be rejected. Therefore, this data does not show that the heated average strength
exceeds the average strength for the control population.

78.

Let 1 denote the true average ratio for young men and 2 denote the true average ratio for
elderly men. Assuming both populations from which these samples were taken are normally
53

Chapter 9: Inferences Based on Two Samples

distributed, the relevant hypotheses are H 0 : 1 2 0 vs. H a : 1 2 0 . The value
of the test statistic is

7.47 6.71
.22 2 .28 2

7.5

. The d.f. = 20 and the p-value is P(t >

13
12
7.5) 0. Since the p-value is .05 , we reject H0. We have sufficient evidence to claim
that the true average ratio for young men exceeds that for elderly men.
79.

The normal probability plot below indicates the data for good visibility does not follow a
normal distribution, thus a t-test is not appropriate for this small a sample size. (The plot for
poor visibility isnt as bad.)
99

95
90
80

Percent

70
60
50
40
30
20
10
5

-1

1
Good

80.
a.

A 95% CI for 37,dry = 325.73 t.025,5(34.97)/ 6 = 325.73 2.571(14.276) = (289.03,

362.43). We are 95% confident that the true average breaking force in a dry medium at
37 is between 289.03 N and 362.43 N.

b. The relevant estimated df = 9. A 95% CI for 37,dry 37,wet = (325.73 306.09) t.025,9
34.97 2 41.97 2
= (30.81,70.09). We are 95% confident that the true average

6
6
breaking force in a dry medium at 37 is between 30.81 N less and 70.09 N more than the
true average breaking force in a wet medium at 37.

c.

We wish to test H0: 37,dry 22,dry = 0 v. Ha: 37,dry 22,dry > 0. The relevant test statistic is t
(325.73 170.60) 100
= 2.58. The estimated df = 9 again, and the
34.97 2
39.08 2

6
6
approximate P-value is .015. Hence, we reject H0 and conclude that true average force in
a dry medium at 37 is indeed more than 100 N greater than the average at 22.
=

54

Chapter 9: Inferences Based on Two Samples

81.

We wish to test H0: 1 2 versus Ha: 1 2

Unpooled:
With H0: 1 2 0 vs. Ha: 1 2 0 , we will reject H0 if p value .

2
.79 2 1.522

14
12
2
2
.79 2
1.522
14
12

13

, and the test statistic

11

8.48 9.36
.79 2
14

15.95 15

52
1.12

.88
1.81 leads to a p-value of about 2[P(t15 > 1.8)] =2(.046)
.4869

= .092.
Pooled:
The degrees of freedom are m n 2 14 12 2 24 and the pooled variance is

13
11
2
2

.79
1.52 1.3970 , so s p 1.181 . The test statistic is
24
24
.88
.88
t

1.89 . The p-value = 2[P( t > 1.9 )] = 2( .035) = .070.

24
1
1
.465
1.181 14 12
With the pooled method, there are more degrees of freedom, and the p-value is smaller than
with the unpooled method. That is, if we are willing to assume equal variances (which might
or might not be valid here), the pooled test is more capable of detecting a significant
difference between the sample means.
82.

Because of the nature of the data, we will use a paired t test. We obtain the differences by
subtracting intake value from expenditure value. We are testing the hypotheses H0: d = 0 vs
Ha: d 0. Test statistic t

1.757
1.197

3.88 with df = n 1 = 6 leads to a p-value of

2[P(t>3.88)] .008. Using either significance level .05 or .01, we would reject the null
hypothesis and conclude that there is a difference between average intake and expenditure.
However, at significance level .001, we would not reject.
83.
a.

With n denoting the second sample size, the first is m = 3n. We then wish

20 2 2.58

900 400
, which yields n = 47, m = 141.

3n
n

b. We wish to find the n which minimizes 2 z / 2

n which minimizes

900
400
, or equivalently, the

400 n
n

900
400

. Taking the derivative with respect to n and

400 n
n
55

Chapter 9: Inferences Based on Two Samples

equating to 0 yields 900 400 n

84.

2
400 n 2 0 , whence 9n 2 4 400 n , or
5n 2 3200n 640,000 0 . This yields n = 160, m = 400 n = 240.
Let p1 = true survival rate at 11 C ; p2 = true survival rate at 30 C ; The hypotheses are
2

H 0 : p1 p 2 0 vs. H a : p1 p 2 0 . The test statistic is

p 1

p 1 p 2

p q m1 1n

. With

73
102
175
.802 , and p 2
.927 , p
.871 , q .129 .
91
110
201

.802 .927

.871 .129 911 1101

.125
2.63 .
.0475

The p-value =

2 2.63 2(.0043) .0086 , so reject H0 at most reasonable levels (.10, .05, .01).
The two survival rates appear to differ.
85.

We want to test the hypothesis H0: 1 1.52 v. Ha: 1 > 1.52 or, using the hint, H0: 0 v.
Ha: > 0. Our point estimate of is X 1 1.5 X 2 , whose estimated standard error
equals s ()

12
2
s12
s2
(1.5) 2 2 . Plug in
(1.5) 2 2 , using the fact that V ()
n1
n2
n1
n2

the values provided to get a test statistic t =

22.63 1.5(14.15) 0

0.83. A
2.8975
conservative df estimate here is = 50 1 = 49, and t.05,49 1.676. Since 0.83 < 1.676, we fail
to reject H0 at the 5% significance level. The data does not suggest that the average tip after an
introduction is more than 50% greater than the average tip without introduction.
86.
a.

For the paired data on pitchers, n = 17, d = 4.066, and sd = 3.955. t.025,16 = 2.120, and the
resulting 95% CI is (2.03, 6.10). We are 95% confident that the true mean difference
between dominant and nondominant arm translation for pitchers is between 2.03 and
6.10.
b. For the paired data on position players, n = 19, d = 0.233, and sd = 1.603. t.025,18 = 2.101,
and the resulting 95% CI is (0.54, 1.01). We are 95% confident that the true mean
difference between dominant and nondominant arm translation for position players is
between 2.03 and 6.10.
c.

Let 1 and 2 represent the true mean differences in side-to-side AP translation for
pitchers and position players, respectively. We wish to test the hypotheses H0: 1 2 = 0
v. Ha: 1 2 > 0. The data for this analysis are precisely the differences utilized in parts a
4.066 0.233

3.955 2 1.603 2 = 3.73. The estimated df = 20

17
19
(using software), and the corresponding P-value is P(t > 3.73) = .001. Hence, even at the
1% level, we concur with the authors assessment that this difference is greater, on
average, in pitchers than in position players.
and b. Hence, the test statistic is t =

56

Chapter 9: Inferences Based on Two Samples

87.

n
200 14.142
,

, so 1.645
14.142
n
n

0 0 , 1 2 10 , d = 1,

giving .9015, .8264, .0294, and .0000 for n = 25, 100, 2500, and 10,000 respectively. If
the i ' s referred to true average IQs resulting from two different conditions, 1 2 1
would have little practical significance, yet very large sample sizes would yield statistical
significance in this situation.

88.

H 0 : 1 2 0 is tested against H a : 1 2 0 using the two-sample t test, rejecting

H0 at level .05 if either t t .025,15 2.131 or if t 2.131 . With x 11 .20 ,

s1 2.68 , y 9.79 , s 2 3.21 , and m = n = 8, sp = 2.96, and t = .95, so H0 is not

rejected. In the situation described, the effect of carpeting would be mixed up with any
effects due to the different types of hospitals, so no separate assessment could be made. The
experiment should have been designed so that a separate assessment could be obtained (e.g., a
randomized block design).

89.

H 0 : p1 p 2 will be rejected at level

z z.05 1.645 . With

p 1

250
2500

in favor of H a : p1 p 2 if either

.10 , p 2

167
2500

.0668 , and

.0834 ,
p

.0332
4.2 , so H0 is rejected . It appears that a response is more likely for a white
.0079

name than for a black name.

90.

The computed value of Z is z

34 46

1.34 . A lower tailed test would be

34 46
appropriate, so the p-value 1.34 .0901 .05 , so we would not judge the drug
to be effective.

57

Chapter 9: Inferences Based on Two Samples

91.
a.

Let

1 and 2 denote the true average weights for operations 1 and 2, respectively.

The relevant hypotheses are H 0 : 1 2 0 vs. H a : 1 2 0 . The value of the

test statistic is

1402.24 1419.63
10.97 2 9.96 2
30

The d.f.

17.39

4.011363 3.30672

17.39

7.318083

6.43

30

7.318083 2
4.011363 2 3.30672 2
29

57.5

, so use df = 57. t .025, 57 2.000

29

, so we can reject H0 at level .05. The data indicates that there is a significant difference
between the true mean weights of the packages for the two operations.

b.

H 0 : 1 1400 will be tested against H a : 1 1400 using a one-sample t test with

test statistic

x 1400
s1

. With degrees of freedom = 29, we reject H0 if

t t .05, 29 1.699 . The test statistic value is t

1402.24 1400
10.97

30

2.24
1.1 .
2.00

Because 1.1 < 1.699, H0 is not rejected. True average weight does not appear to exceed
1400.

92.

First, Var X Y

1 2
1 1

under H0, where can be estimated for the
m n
m n

mX nY
variance by the pooled estimate
. With the obvious point estimates 1 X ,
mn
(X Y ) 0
X Y
Z

2 Y , we have a large-sample test statistic of

X Y . With
1 1

n m
m n

x 1.616 and y 2.557 , z = -5.3 and p-value = 2 5.3 .0006 , so we

would certainly reject H 0 : 1 2 in favor of H a : 1 2 .

93.

1 2 , or
A large-sample confidence interval for 1 2 is ( ) z

1
2
/2
m

x
y
. With x 1.62 and y 2.56 , the 95% confidence interval
m n
for 1 2 is -.94 1.96(.177) = -.94 .35 = (-1.29,-.59).
( x y ) z / 2

58