MAT 540 Statistical Concepts For Research

287
Chapter 10
COMPARING TWO TREATMENTS

10.1 First group using first letter:

{B, C} {B, E} {B, H} {B, P}
{C, E} {C, H} {C, P}
{E, H} {E, P}
{H, P}

10.2 For convenience, we list the second group using Greek letters:

{ , } { , } { , }
{ , } { , } { , }
{ , } { , } { , }
{ , } { , } { , }
{ , } { , } { , }

10.3 (a) We use the first letter of the first names:

{ (S, G), (T, E) } { (S, G), (T, R) } { (S, G), (E, R) }
{ (S, J), (T, E) } { (S, J), (T, R) } { (S, J), (E, R) }
{ (J, G), (T, E) } { (J, G), (T, R) } { (J, G), (E, R) }

(b) There are three sets each consisting of three pairs.

288 CHAPTER 10. COMPARING TWO TREATMENTS
10.4

Unit Treatments Response Design
(a) Person
vitamin C capsule
no vitamin C

Number of colds in 5
months
Independent
samples
(b) Plant
self-fertilization
cross-fertilization

Height after 3 months Matched pair
(c) Person
husband
wife

Number of children Matched pair
(d) Dog
reward training
reward punishment

Learning times Independent
samples

10.5 (a) A point estimate of
1 2
is given by 73 66 7 x y = = .
Estimated S.E. =
2 2
1 2
1 2
151 142
2.48
52 44
s s
n n
+ = + =
(b) A large sample 95% confidence interval for
1 2
is given by
( ) 0.05
2
2 2
1 2
1 2
7 1.96(2.48) 7 4.9
s s
x y z
n n
+ = = or (2.1, 11.9).

10.6 A large sample 98% confidence interval for
1 2
is given by
( ) 0.02
2
2 2 2 2
1 2
1 2
(8.2) (7.6)
(76.4 81.2) 2.33 4.8 2.68
90 100
s s
x y z
n n
+ = + =
or ( 7.48 , 2.12 ). We are 98% confident that urban students have a mean score
2.12 to 7.48 points higher than that of the rural students.

10.7 We are to test the hypotheses
0 1 2 1 1 2
: 0 versus : 0 H H = with
0.05 = . Since the sample sizes are large, we employ the Z-test, and so, the test
statistic is
2 2
1 2
1 2
X Y
Z
S S
n n
=
+
. Since H
1
is two-sided, the rejection region is
0.05
2
: 1.96 R Z z = . From the sample data, we calculate the value of the observed z
to be
2 2
76.4 81.2 4.8
4.17
1.151
(8.2) (7.6)
90 100
z

= = =
+
, which lies in R. Hence, H
0
is
rejected at 0.05 = . Furthermore, the associated p-value is [ 4.17] 0.0001 P Z < ,
so the evidence against no mean difference is very strong.

289

1 2
is given by
( ) 0.02
2
2 2 2 2
1 2
1 2
(4.2) (1.9)
(12.6 9.5) 2.33 3.1 1.519
50 50
s s
x y z
n n
+ = + =
or (1.581, 4.619). We are 98% confident that the mean number of words per
sentence for magazine 1 is 1.581 to 6.419 greater than that of magazine 2.

10.9 (a) Since the assertion is that
1 2
> , we formulate the hypotheses

0 1 2 1 1 2
: 2 versus : 2 H H = > .
(b) Since the sample sizes
1
50 n = and
2
50 n = are large, we employ the Z-test, and
so, the test statistic is
2 2
1 2
1 2
X Y
Z
S S
n n
=
+
. Since H
1
is right-sided and 0.05 = ,
the rejection region is
0.05
: 1.645 R Z z = .
(c) From the sample data, we calculate the value of the observed z to be
( )
2
2
12.6 9.5 3.1
1.69
0.652
1.9 (4.2)
50 50
z

= = =
+
0
is rejected at
0.05 = . Furthermore, the associated p-value is [ 4.755] 0.0455 P Z = , so the
evidence is support of H
1
is very strong.

10.10 (a) From Example 4,
Sample mean for firefighters is 3.673
f
x =
Sample variance for firefighters is
2 2
(0.7235) 0.523
f
s = =

Sample mean for office supervisors is 3.547
o
x =
Sample variance for office supervisors is
2 2
0
(0.6089) 0.371 s = =
(b) Since we are given that
Sample mean for clergy is 3.79
c
x =
Sample variance for clergy is
2 2
(0.590) 0.3481
c
s = = ,
we see that
2 2
0
0.371 0.3481
0.0733
247 90 247 90
c
s s
+ = + = .
Thus, a large sample 95% confidence interval for
1 2
is given by

( ) 0.05
2
2 2
0
1 2
(3.79 3.547) 1.96(0.0733)
0.243 1.96(0.0733) 0.243 0.1437
c
c o
s s
x x z
n n
+ =
= =

or (0.0993, 0.3867).

10.11 The problem here is to test
0 1 2 1 1 2
: 0 versus : 0 H H = with
0.05 = . Because 0 is not included in the 95% confidence interval
(0.0993, 0.3867) (from Exercise 10.10), the conclusion is that H
0
is rejected at
0.05 = .

10.12 The summary statistics are:

Aggressive:
1
47 n = , 7.92 x = ,
1
3.45 s =
Non-aggressive:
2
38 n = , 5.80 y = ,
2
2.87 s =

Denote by
1
and
2
the population BPC scores for the two groups
aggressive and non-aggressive, respectively. Since the conjecture is that
1
is larger than
2
, we formulate the hypotheses
0 1 2 1 1 2
: 0 versus : 0 H H = > .
Since the sample sizes are large, we employ the Z-test, and so, the test statistic
is
2 2
1 2
1 2
X Y
Z
S S
n n
=
+
. From the sample data, we calculate the value of the
observed z to be
2 2
7.92 5.80 2.12
3.09
0.6856
(3.45) (2.87)
47 38
z

= = =
+
. The p-
value [ 3.09] P Z is smaller than 0.001. This extremely small p-value provides
a strong justification of the conjecture that
1
is larger than
2
.

10.13 (a) We first obtain:
5 x = 7 y =

( )
2 2 2 2
2
1
1 ( 1) 2 ( 2)
3.333
3
s
+ + +
= =
( )
2 2 2
2
2
1 2 ( 1)
3
2
s
+ +
= =
Consequently, the pooled variance is given by
( ) ( )
2 2
1 1 2 2 2
pooled
1 2
1 1
10 6
3.2
2 4 3 2
s n s n
s
n n
+
+
= = =
+ +
.
(b) We estimate the common sigma by
pooled
3.2 1.789 s = = .
(c) The t-statistic is
pooled
1 2
1 1
X Y
T
s
n n
=
+
with d.f. =
1 2
2 n n + . In the
present problem, we have
5 7
1.463
1 1
1.79
4 3
t

= =
+
, with d.f. = 5.

291


MALE FEMALE
6 x = 3 y =

( )
2 2 2
2
1
2 ( 2) 0
4
2
s
+ +
= =
( )
2 2 2
2
2
2 ( 2) 0
4
2
s
+ +
= =

2 2
2
pooled
1 2
( ) ( )
8 8
4
2 3 3 2
i i
x x y y
s
n n
+
+
= = =
+ +

.
(b) We estimate the common sigma by
pooled
4 2 s = = .
(c) The t-statistic is
pooled
1 2
1 1
X Y
T
s
n n
=
+
with d.f. =
1 2
2 n n + . In the
6 6
0
1 1
2
3 3
t

= =
+
, with d.f. = 4.

10.15 (a)
2 2
2
pooled
1 2
( ) ( )
28 32
2.4
2 14 13 2
i i
x x y y
s
n n
+
+
= = =
+ +

(b) We are to test the hypotheses
0 1 2 1 1 2
: 0 versus : 0 H H = >
with 0.05 = . We assume normal populations with equal variance, so the
test statistic is
pooled
1 2
1 1
X Y
T
s
n n
=
+
with d.f. =
1 2
2 n n +
Since H
1
is right-sided, the rejection region is
0.05
: 1.708 R T t = . From
the sample data, we calculate the value of the observed t is
9 17 8
13.40
0.597 1 1
2.4
14 13
t

= = =
+
,
which does not lie in R. So, H
0
is not rejected at 0.05 = .
(c) A corresponding 95% confidence interval has the form
( ) pooled
1 2
2
1 1
X Y t s
n n
+ .
Since
0.025
2.060 t = for d.f. = 25, the 95% confidence interval in this case
is 8 2.060(0.597) 8 1.230 = or (-9.23, -7.23).

10.16 (a) We are to test the hypotheses
0 1 2 1 1 2
: 0 versus : 0 H H =
with 0.05 = . We assume normal populations with equal variance, so the
test statistic is
pooled
1 2
1 1
X Y
T
s
n n
=
+
with d.f. =
1 2
2 n n +
Since H
1
0.05
2
: 2.110 R T t = (for d.f.
= 8 + 11 2 = 17) . From the sample data, we calculate the following:
73.9 91.9 18 x y = = ,
2
1
101.268 s = , and
2
2
153.291 s =
Consequently, we have

2 2
1 1 2 2
pooled
1 2
( 1) ( 1) (8 1)101.268 (11 1)153.291
11.483
2 8 11 2
n s n s
s
n n
+ +
= = =
+ +
.
As such, the value of the observed t is
73.9 91.9
3.37
1 1
11.483
8 11
t

= =
+
,
which lies in R. So, H
0
is rejected at 0.05 = . Furthermore, the
associated p-value is [ 3.37] 0.0036 P T = .
(b) A corresponding 95% confidence interval has the form
( ) pooled
1 2
2
1 1
X Y t s
n n
+ .
Since
0.025
2.110 t = for d.f. = 17, the 95% confidence interval in this case
is 18 2.110(11.483) 0.2159 18 11.3 = or ( 29.3, 6.7 ) pounds.
(c) We assume normal populations with equal variances. The sample
variances are not too different, and there are no obvious outliers in the
data.


Abused:
1
52 n = , 2.48 x = ,
1
1.94 s =
Non-abused:
2
67 n = , 1.57 y = ,
2
1.31 s =

Denote by
1
and
2
the population mean number of crimes for the abused
and non-abused groups, respectively. Since the conjecture is that
1
is larger
than
2
0 1 2 1 1 2
: 0 versus : 0 H H = > .
Since the sample sizes are large, we employ the Z-test, and so, the test statistic
293

is
2 2
1 2
1 2
X Y
Z
S S
n n
=
+
. Since H
1
is right-sided, the rejection region is of the form
: R Z c . Using the summary statistics, we obtain
2 2 2 2
1 2
1 2
2.48 1.57 0.91
(1.94) (1.31)
0.3130
52 67
x y
s s
n n
= =
+ = + =

Hence, the observed value of z is
0.91
2.907
0.3130
Z = = . The associated p-value
is [ 2.907] 0.00182 P Z = . This very low p-value signifies strong evidence in
support of H
1
.

1 2
is given by
( ) 0.01
2
2 2
1 2
1 2
s s
x y z
n n
+
Using the calculations in Exercise 10.17, this becomes
0.91 2.58(0.3130) or (0.102, 1.72).

10.19 (a) The summary statistics are:

Method 1:
1
10 n = , 19.1 x = ,
1
4.818 s =
Method 2:
2
10 n = , 23.3 y = ,
2
5.559 s =

Denote by
1
and
2
the population mean job times corresponding to
Method 1 and Method 2, respectively. Since the conjecture is that
1
is
smaller than
2
0 1 2 1 1 2
: 0 versus : 0 H H = < .
We assume normal populations with equal variance, so the test statistic is
pooled
1 2
1 1
X Y
T
s
n n
=
+
with d.f. =
1 2
2 n n +
Since H
1
is left-sided, the rejection region is
0.05
: 1.734 R T t = for d.f.
= 10 + 10 2 = 18. Using the summary statistics, we obtain

2 2 2 2
1 1 2 2
pooled
1 2
19.1 23.3 4.2
( 1) ( 1) 9(4.818) 9(5.559)
5.201
2 18
x y
n s n s
s
n n
= =
+ +
= = =
+

Hence, the observed value of t is given by

4.2
1.81
1 1
5.201
10 10
t

= =
+
,
0
is rejected at 0.05 = . We conclude that the mean
job time is significantly less for Method 1 than for Method 2.
(b) See part (a).
(c) A corresponding 95% confidence interval has the form
( ) pooled
1 2
0.05
2
1 1
X Y t s
n n
+ .
Since
0.025
2.101 t = for d.f. = 18, using the calculations in part (a), we
obtain that the 95% confidence interval in this case is
4.2 2.101(2.326) 4.2 4.89 = or ( 9.09, 0.69 ).

10.20 (a) Use the Z-test since the sample sizes are large.
(b) Use the t-test with pooling. The sample sizes are small, and
1
s and
2
s are
not too far apart (
2
1
1.3
s
s
= ). Assume normal populations with equal
variances.
(c) Use the conservative t-test without pooling. The sample sizes are small,
and
2
1
s
s
is larger than 2. Assume that the populations are normal.
(d) Use the Z-test since the sample sizes are large.

10.21 The 90% confidence interval is (3.6, 5.2). Because the null hypothesis value of
1 2
, namely 4.1, lies inside this 90% confidence interval, H
0
is not rejected
at 0.10 = .


Isometric Method:
1
10 n = , 2.4 x = ,
1
0.8 s =
Isotonic Method:
2
10 n = , 3.2 y = ,
2
1.0 s =

(a) Denote by
1
and
2
the population mean decrease in abdomen
measurements under the Isometric Method and Isotonic Method,
respectively. Since the intent is to demonstrate that the Isotonic Method is
more effective (that is,
1
is smaller than
2
), we formulate the hypotheses
0 1 2 1 1 2
: 0 versus : 0 H H = < .
pooled
1 2
1 1
X Y
T
s
n n
=
+
with d.f. =
1 2
2 n n +
295

Since H
1
is left-sided, the rejection region is
0.05
: 1.734 R T t = for d.f.

2 2 2 2
1 1 2 2
pooled
1 2
2.4 3.2 0.8
( 1) ( 1) 9(0.8) 9(1.0)
0.906
2 18
x y
n s n s
s
n n
= =
+ +
= = =
+


0.8 0.8
1.98
0.405 1 1
0.906
10 10
t

= = =
+
,
0
is rejected at 0.05 = . As such, the superiority of
the Isotonic Method is demonstrated by the data.
(b) Using the calculations from part (a), a corresponding 95% confidence
interval has the form

( ) pooled
1 2
0.05
2
1 1
0.8 2.101(0.405) 0.8 0.85 X Y t s
n n
+ = =
or ( 1.65, 0.05 ) centimeters.

10.23 (a) Age, eating habits, initial degree of obesity, other exercise activities, etc.
(b) Randomization prevents these extraneous factors from influencing the
response of either treatment group in a systematic manner. For instance, a
very overweight person has an equal chance of being assigned to each
exercise method.
(c) Assign identifying numbers 1 20 to the twenty executives and place
twenty similarly numbered marbles in an urn. After mixing, blindly draw
ten marbles and assign those numbered people to Treatment 1. Then, give
Treatment 2 to the others.

10.24 (a) The summary statistics are:

1
9 n = , 1.887 x = ,
2
1
0.0269 s =

2
9 n = , 0.670 y = ,
2
2
0.1133 s =
The more conservative test in Section 10.5.1 requires that we use
2
t
with
d.f. = 8 and 0.05 = , which is
0.025
2.306 t = . So, a 95% confidence
interval for
1 2
has the form
( )
2 2
1 2
1 2
0.025
S S
X Y t
n n
+ = 1.217 2.306(0.1247)
or (0.9295,1.5045) .
(b) The confidence interval from Example 12 is 0.94, 1.49). The confidence
interval in (a) is larger, which is reasonable since it is more conservative.

10.25 We test the hypotheses:
0 1 2 1 1 2
: 0 versus : 0 H H = <

We use the test statistic
2 2
1 2
1 2
0.05
X Y
T
S S
t
n n
=
+

with d.f. = the smaller of
1 2
1, 1 n n . Since
1
H is left-sided, the rejection
region is
0.05
: R T t . Since
1 2
9, 9, d.f . 8 n n = = = , this region becomes
: 1.860 R T . Using the sample statistics from Example 12 and Exercise
10.24, we see that the value of the test statistic is
1.887 0.670 1.217
5.248
1.860(0.1247) 0.2319
T

= = =
Since this value is not in the rejection region R, we do not reject
0
H ; there is
insufficient evidence to support the claim.


1
40 n = , 14.38 x = ,
1
7.34 s =

2
43 n = , 23.72 y = ,
2
5.78 s =

Since the sample sizes are large, neither the assumption of normal populations
nor the assumption of equal variances is needed. A large sample 99%
confidence interval for
1 2
is given by
( ) 0.01
2
2 2
1 2
1 2
s s
x y z
n n
+
.
Observe that

0.005
2.58 z =

2 2 2 2
1 2
1 2
14.38 23.72 9.34
7.34 5.78
1.457
40 43
x y
s s
n n
= =
+ = + =

So, the 99% confidence interval becomes
9.34 2.58(1.457) 9.34 3.759 = or (-13.099, -5.581).

10.27 (a) Since the p-value is less than 0.05, we reject H
0
.

(b) We test the hypotheses:
0 1 2 1 1 2
: 5 versus : 5 H H =

We use the test statistic
297

2 2
1 2
1 2
0.05
X Y
T
S S
t
n n
=
+

with d.f. = the smaller of
1 2
1, 1 n n . Since
1
H is two-sided, the rejection
region is
0.025
: R T t . Since
1 2
7, 10, d.f . 6 n n = = = , this region becomes
: 2.447 R T . Using the sample statistics we see that the value of the test
statistic is
2 2
73.9 91.9 18
2.461
7.313
10.1 12.4
2.447
8 11
T

= = =
+

Since this value is in the rejection region R, we reject
0
H .

10.28 From five slips labeled A, B, C, E, and J, we drew B and E, so Bob and Ellen
form the control group. (Answers may vary.)

10.29 We drew slips with , , and , so group 1 is {alpha, beta, tau}. (Answers
may vary.)

10.30 (a) Suppose you could randomly select the number of cigarettes each subject
would smoke per day for a period of 10 years. Some subjects would not
be allowed to smoke, while others would be assigned 2, 1, or 0.5 packs per
day. This would uncouple the genetic cause from observed disease rates.
(b) Of course, this would not be an ethical experiment because there is already
strong evidence that smoking is hazardous.

10.31 We must be careful. It is likely that mothers who are warmer toward everyone
have a much higher tendency to nurse their babies than mothers who have
colder personalities. There are many other reasons as well.

10.32 (a) Without randomization, the rats chosen for hormone treatment may have a
systematic difference from the control. Randomization gives the strongest
and the weakest rats an equal chance of receiving the treatment or control.
Those easiest to grab may be the weakest, most listless, or otherwise less
prone to gain weight.
(b) Put the marbles into a bowl, thoroughly mix, and select six, one-at-a-time.
These numbered rats will form the treatment group, while the others are
the control.

10.33 (a) The 6 n = paired differences d x y = are 3, 3, 5, 1, 2, -2.
Their mean and standard deviation are 2, 2.366
D
d s = = .
So,
2
2.071
2.366 / 6
t = = .
(b) d.f. = 1 5 n =

10.34 (a) The 5 n = paired differences d x y = are 0, 2, 4, 2, 2 .
Their mean and standard deviation are 2, 1.414
D
d s = = .
So,
2
3.163
1.414 / 5
t = = .
(b) d.f. = 1 4 n =

10.35 It is a matched pair sample because there may be considerable variation of
conditions in the different plants. The paired differences d = (before after) are
2,1, 1, 2, 3, 1 . We assume these differences constitute a random sample from
a normal distribution with mean . The null hypothesis of no change is
0
: 0 H = , and the alternative of more loss before than after is
1
: 0 H > . We
calculate
1.0
0.683
6
1.0, 1.673, 0.683, 1.46
D
s
D
d s t = = = = = , and d.f. = 5.
With 0.05 = , the rejection region is
0.05
: 2.015 R T t > = . Since the observed t
is less than 2.015, H
0
is not rejected at 0.05 = . As such, the claim of
effectiveness of the safety program is not demonstrated.

10.36 (a) Select five persons to wear the boots. Because the boots will be subjected
to different wear conditions, apply the major brand to one boot in the pair
and the manufacturers water-proofing to the other.
(b) Flip a coin for each pair. If Heads, then the left boot gets major brand M.
Otherwise, it gets the manufacturers water-proofing m. The following
represents one possible assignment of the two methods to each of the 5
pairs of boots.

m M M m M m m M M m

10.37 This is a matched pair design because from each farm a pair of milk specimens
are taken, and then one is treated with PC while the other is not.
(a) The paired differences d = (with PC without PC) are 7, 6, 2,10, 0, 8, 4 .
From this, we obtain the following summary statistics:
7, 4.714, and 4.348
D
n d s = = =
We assume that the population distribution of the Ds is normal, and
denote the population mean by . Because the conjecture is that the
mean response with PC is higher than without PC (that is, 0 > ), we
formulate the hypotheses
0
: 0 H = versus
1
: 0 H > .
The test statistic is
/
D
D
T
S n
= , d.f. = 1 n . Since H
1
is right-sided and
0.05 = , the rejection region is
0.05
: 1.943 R T t = (for d.f. = 6). Using
the summary statistics, we see that the observed t is
4.714
4.348/ 7
2.87 t = = ,
299

which lies in R. Hence, H
0
associated p-value is [ 2.87] 0.0142 P T = . As such, we conclude that
there is strong evidence in support of this conjecture.
(b) The corresponding 90% confidence interval for is given by

( )
4.35
0.05
7
4.71 1.943 4.71 3.19
D
s
n
d t = = or (1.52, 7.90).

10.38 (a) The paired differences d = (A B) are 0.9, 0.1, 0.8, 0, 0.1, 0.3 .
Assume that these constitute a random sample from a population with
distribution ( ) ,
D
N . Observe that
0.452
6
6, 0.3, 0.452, 0.1845
D
D
s
n
n d s = = = = = .
For 0.05 = and d.f. = 5, we have
0.025
2.571 t = . So, the 95% confidence
interval for is given by
( )
0.452
0.025
6
0.3 2.571 0.3 0.47
D
s
n
d t = = or ( 0.17, 0.77 ).
(b) For each person, the assignment of A and B should be made at random
between the two nights. This can be done, for instance, by tossing a coin
for each person. If this coin falls Heads, then give A the first night and B
the second; if Tails appear, then give B the first night and A the second.

10.39 This is a matched pair design. The paired differences d = (A B) are
5, 1, 2, 3, 0, 1, 4, 3, 3 . From this, we obtain the following summary statistics:
9, 1.333, and 2.693
D
n d s = = =
We assume that the population distribution of these differences is normal, and
denote the population mean by . We formulate the hypotheses
0
: 0 H = versus
1
: 0 H .
/
D
D
T
S n
= , d.f. = 1 n . Since H
1
is two-sided and
0.025
: 2.306 R T t = (for d.f. = 8). Using the
summary statistics, we see that the observed t is
1.333
2.693/ 9
1.49 t = = , which does
not lie in R. Hence, H
0
is not rejected at 0.05 = , and so we conclude that the
difference is not significant at this level.

10.40 (a) Observe that
10, 42.9, and 34.346
D
n d s = = = .
For 0.05 = and d.f. = 9, we have
0.025
interval for (the mean of D = previous year usage experimental year
usage) is given by
( )
34.346
0.025
10
42.9 2.262
D
s
n
d t = or (18.3, 67.5).
(b) We reject
0
: 0 H = for any value of the observed t which lies in the
rejection region
0.025
: 2.262 R T t = . Here,
42.9
34.346/ 10 /
3.950
D
d
s n
t = = = ,
which lies in R. Hence, we reject H
0
is favor of
1
: 0 H at 0.05 = .
(c) Some persons would have to be under the pricing policy the first year, and
others on it the second year. The allocation would be made at random.
(d) If July the previous year was much hotter than July in the experimental
year, then air conditioner electrical power requirements could be the cause
of the drop in usage. Without randomizing the price scheme over the two
summers, we cannot separate the two causes: temperature and price.

10.41 This is a matched pair design because the two plots in the same farm are alike
with respect to soil condition, rainfall, temperature, and other factors that affect
the yield. The paired differences d = (yield of A yield of B) are
5, 6, 2, 9, 0, 9, 7, 2 . From this, we obtain the following summary statistics:
8, 4.50, and 4.11
D
n d s = = =
denote the population mean by .
(a) In order to substantiate that A has a higher mean yield than B (that is,
0 > ), we formulate the hypotheses
0
: 0 H = versus
1
: 0 H > .
/
D
D
T
S n
= , d.f. = 1 n . Since H
1
is right-sided and
0.05
: 1.895 R T t = (for d.f. = 7). Using
4.50
4.11/ 8
3.10 t = = ,
0
is rejected at 0.05 = , and so we conclude
that A has a significantly higher mean yield than B.
(b) At each farm, the assignment of the two plots to strain A and strain B
should be randomized. Label the two plots 1 and 2, and toss a coin if a
Head turns up, assign plot 1 to strain A and plot 2 to strain B. If a Tail
turns up, the assign plot 1 to strain B and plot 2 to strain A. Repeat this
process for all the farms.

10.42 Here we have a situation of independent random samples because 16 subjects
(farms) were randomly divided into two groups of 8 each to be assigned to the
two treatments. We calculate the following summary statistics:
Strain A:
1
8 n = , 32.0 x = ,
1
8.83 s =
Strain B:
2
8 n = , 27.5 y = ,
2
8.37 s =
Denote by
1
and
2
the population mean yields of strain A and strain B,
respectively. Since the conjecture is that
1
is larger than
2
, we formulate the
hypotheses
0 1 2 1 1 2
: 0 versus : 0 H H = > .
301

pooled
1 2
1 1
X Y
T
s
n n
=
+
with d.f. =
1 2
2 n n +
Since H
1
is right-sided, the rejection region is
0.05
: 1.761 R T t = for d.f. = 8 +
8 2 = 14. Using the summary statistics, we obtain

2 2 2 2
1 1 2 2
pooled
1 2
4.50
( 1) ( 1) 7(8.83) 7(8.37)
8.60
2 14
x y
n s n s
s
n n
=
+ +
= = =
+


4.50 4.50
1.05
4.30 1 1
8.60
8 8
t = = =
+
,
0
is not rejected at 0.05 = . We conclude that
the claim of a higher mean yield for strain A fails to be demonstrated by the
data.
(b) The selection of 8 farms for assigning each treatment should be randomized.
Tag the 16 farms 1, 2, , 16, place 16 numbered marbles in an urn, and
randomly draw 8 marbles. Assign the farms corresponding to the selected
marbles to strain A and the others to strain B.
(c) Comparing the calculations in Exercises 10.41 and 10.42, we find that
4.50 d x y = = . So, the numerators of the two t-statistics are the same.
However, the estimated S.E.s are quite different:
For matched pair:
4.11
8
1.453
D
s
n
= =
For independent samples:
pooled
1 2
1 1
4.30 s
n n
+ =
A considerable reduction of variability in the yield data has been accomplished
by means of the matched pair design.

10.43 This is a matched pair design. The paired differences d = (left right) are
given, from which we obtain the following summary statistics:
10, 60.37, and 0.1418
D
n d s = = =
denote the population mean by .
(a) We formulate the hypotheses
0
: 60 H = versus
1
: 60 H .
/
D
D
T
S n

= , d.f. = 1 n . Since H
1
is two-sided and
0.025
: 2.262 R T t = (for d.f. = 9). Using
60.37 60
0.1418/ 10
8.251 t

= = ,
0
is rejected at 0.05 = .
(b) For 0.05 = and d.f. = 9, we have
0.025
( )
0.025
60.37 2.262 0.045 60.37 0.10179
D
s
n
d t = =
or (60.268, 60.472).

10.44 (a) From the data in Exercise 10.10, we see that
180 149
1 2 226 249
1 2
0.796, 0.598, so that
0.796 0.598 0.198
p p
p p
= = = =
= =

Estimated S.E. =
1 1 2 2
1 2
(0.796(0.204) (0.598)(0.402)
0.0408
226 249
p q p q
n n
+ = + =

So, a 95% confidence interval for
1 2
p p is given by
( ) 0.05
2
1 1 2 2
1 2
1 2

1.98 1.96(0.0408) 1.98 0.080
p q p q
p p z
n n
+ = =
or (1.90, 2.06 ).
(b) For such a test, the test statistic is
1 2
1 2

1 1

p p
Z
pq
n n
=
+
.
Observe that
Pooled estimate
1 1 2 2
1 2
226(0.796) 249(0.598)
0.692
226 249
n p n p
p
n n
+ +
= = =
+ +
,
Observed z is
1.98 1.98
46.698
0.0424 1 1
(0.692)(0.308)
226 249
= =
+
.
The associated p-value is [ 46.698] 0 P Z . This means that H
0
would
be rejected for small values of 0.05 < .

10.45 (a) Observe that
78
1 90
149
2 247
1 2
proportion of clergy 0.867,
proportion of office supervisors 0.598,
0.867 0.598 0.269
p
p
p p
= = =
= = =
= =

Estimated S.E. =
1 1 2 2
1 2
(0.867)(0.133) (0.598)(0.402)
0.047
90 247
p q p q
n n
+ = + =

1 2
p p is given by
( ) 0.05
2
1 1 2 2
1 2
1 2

0.269 1.96(0.047) 0.269 0.092
p q p q
p p z
n n
+ = =
or ( 0.177, 0.361).
303

(b) For such a test, the test statistic is
1 2
1 2

1 1

p p
Z
pq
n n
=
+
.
Observe that
Pooled estimate
1 1 2 2
1 2
90(0.867) 247(0.598)
0.670
90 247
n p n p
p
n n
+ +
= = =
+ +
,
Observed z is
0.269 0.269
4.719
0.057 1 1
(0.670)(0.330)
90 247
= =
+
.
The associated p-value is [ 4.719] 0.001 P Z < . This means that H
0

would be rejected for extremely small values of 0.001 < .

10.46 (a) We formulate the hypotheses:
0
:
A B
H p p = versus
1
:
A B
H p p < .
1 2
1 2

1 1

p p
Z
pq
n n
=
+
, where
1
p is identified as
A
p and
2
p is identified as
B
p . Since H
1
is left-sided and 0.05 = , the rejection
region is
0.05
: 1.645 R Z z = . We calculate the following:

50 88
1 2 120 150
0.417, 0.587 p p = = = =
Pooled estimate
1 1 2 2
1 2
120(0.417) 150(0.587)
0.511
120 150
n p n p
p
n n
+ +
= = =
+ +
,
Observed z is
0.417 0.587 0.17
2.778
0.0612 1 1
(0.511)(0.489)
120 150

= =
+
,
0
associated p-value is [ 2.778] 0.0028 P Z = . So, there is very strong
evidence that drug B has a higher cure rate than drug A. The rejection
region and p-value are illustrated below (in that order):
(b)
2 1
0.587 0.417 0.17 p p = =
Estimated S.E. =
1 1 2 2
1 2
(0.417)(0.583) (0.587)(0.413)
120 150
0.060
p q p q
n n
+ = +
=

1 2
p p is given by
( ) 0.05
2
1 1 2 2
1 2
1 2

0.17 1.96(0.060) 0.17 0.118
p q p q
p p z
n n
+ = =
or (0.05, 0.29).

10.47 Let p
1
and p
2
denote the probabilities of resistant for the HRL and LRL
groups, respectively. We formulate the hypotheses:
0 1 2
: H p p = versus
1 1 2
: H p p < .
1 2
1 2

1 1

p p
Z
pq
n n
=
+
. Since H
1
is left-sided, the rejection
region is of the form : R Z c . We calculate the following:

15 42
1 2 49 54
0.306, 0.778 p p = = = =
Pooled estimate
1 1 2 2
1 2
15 42
0.553
49 54
n p n p
p
n n
+ +
= = =
+ +
,
Observed z is
0.306 0.778
4.81
1 1
(0.553)(0.447)
49 54
=
+
.
The associated p-value is [ 4.81] P Z is less than 0.0001. So, there is very
strong evidence in support of H
1
.

10.48
1 2
0.306 0.778 0.472 p p = =
Estimated S.E. =
1 1 2 2
1 2
(0.306)(0.694) (0.778)(0.222)
0.087
49 54
p q p q
n n
+ = + =
1 2
p p is given by
( ) 0.01
2
1 1 2 2
1 2
1 2

0.472 2.58(0.087) 0.472 0.224
p q p q
p p z
n n
+ = =
or ( 0.70, 0.25 ).

10.49 Let p
1
and p
2
denote the probabilities of being a chronic offender for abused and
non-abused groups, respectively. We formulate the hypotheses:
0 1 2
: H p p = versus
1 1 2
: H p p > .
1 2
1 2

1 1

p p
Z
pq
n n
=
+
. Since H
1
is right-sided and 0.01 = ,
0.01

21 11
1 2 85 120
0.247, 0.092 p p = = = =
Pooled estimate
1 1 2 2
1 2
21 11
0.156
85 120
n p n p
p
n n
+ +
= = =
+ +
,
Observed z is
0.247 0.092
3.01
1 1
(0.156)(0.844)
85 120
Z

= =
+
,
305

0
is rejected at 0.01 = . Furthermore, the associated p-
value is [ 3.01] 0.0013 P Z = . So, there is very strong evidence in support of
H
1
.

10.50
1 2
0.247 0.092 0.155 p p = =
Estimated S.E. =
1 1 2 2
1 2
(0.247)(0.753) (0.092)(0.908)
0.054
85 120
p q p q
n n
+ = + =
1 2
p p is given by
( ) 0.05
2
1 1 2 2
1 2
1 2

0.155 1.96(0.054) 0.155 0.106
p q p q
p p z
n n
+ = =
or ( 0.05, 0.26 ).
10.51 Let p
1
and p
2
denote the probability of survival for the treated group (with
carbolic acid) and the control group (without carbolic acid), respectively. We
formulate the hypotheses:
0 1 2
: H p p = versus
1 1 2
: H p p .
1 2
1 2

1 1

p p
Z
pq
n n
=
+
. Since H
1
is two-sided and 0.05 = ,
0.05
2
: 1.96 R Z z = .
We calculate the following:

34 19
1 2 40 35
0.850, 0.543 p p = = = =
Pooled estimate
1 1 2 2
1 2
34 19
0.707
40 35
n p n p
p
n n
+ +
= = =
+ +
,
Observed z is
0.850 0.543
2.91
1 1
(0.707)(0.293)
40 35
=
+
,
0
is rejected at 0.05 = . Furthermore, the associated
p-value 2 [ 2.91] 2(0.0018) 0.0036 P Z = = . This means that H
0
would be
rejected with as small as 0.0036. As such, a difference in the survival rates is
strongly demonstrated by the data.

10.52
1 2
0.850 0.543 0.307 p p = =
Estimated S.E. =
1 1 2 2
1 2
(0.850)(0.150) (0.543)(0.457)
0.101
40 35
p q p q
n n
+ = + =
1 2
p p is given by
( ) 0.05
2
1 1 2 2
1 2
1 2

0.307 1.96(0.101) 0.307 0.198
p q p q
p p z
n n
+ = =
or (0.11, 0.51).

10.53 Let p
1
and p
2
denote the population proportions of 8 hours of sleep for the age
group 30 40, and the age group 60 70, respectively. We formulate the
hypotheses:
0 1 2
: H p p = versus
1 1 2
: H p p > .
1 2
1 2

1 1

p p
Z
pq
n n
=
+
. We calculate the following:

173 120
1 2 250 250
0.692, 0.480 p p = = = =
Pooled estimate
1 1 2 2
1 2
293
0.586
500
n p n p
p
n n
+
= = =
+
,
Observed z is
0.692 0.480
4.81
1 1
(0.586)(0.414)
250 250
=
+

1
.

10.54
1 2
0.692 0.480 0.212 p p = =
Estimated S.E. =
1 1 2 2
1 2
(0.692)(0.308) (0.480)(0.520)
0.043
250 250
p q p q
n n
+ = + =
1 2
p p is given by
( ) 0.05
2
1 1 2 2
1 2
1 2

0.212 1.96(0.043) 0.212 0.084
p q p q
p p z
n n
+ = =
or (0.13, 0.30).

10.55 (a) Let p
1
and p
2
denote the probability of having prominent wrinkles for
smokers and non-smokers, respectively. We formulate the hypotheses:
0 1 2
: H p p = versus
1 1 2
: H p p > .
1 2
1 2

1 1

p p
Z
pq
n n
=
+

95 103
1 2 150 250
0.633, 0.412 p p = = = =
Pooled estimate
1 1 2 2
1 2
95 103
0.495
400
n p n p
p
n n
+ +
= = =
+
,
Observed z is
0.633 0.412
4.28
1 1
(0.495)(0.505)
150 250
=
+

1
.
307

(b) A direct causal relation between smoking and wrinkled skin cannot be
readily concluded. Various psycho-physiological factors could influence
both the smoking habit and the presence of wrinkled skin.

10.56 For testing the hypotheses
0 1 2
: H p p = versus
1 1 2
: H p p , we use the
test statistic
1 2
1 2

1 1

p p
Z
pq
n n
=
+
.
(a) Let

1
p = proportion for 25-34 year olds, and

2
p =proportion for 35-44 year olds.
Since H
1
is two-sided and 0.05 = , the rejection region is

0.05
2
Pooled estimate
1 1 2 2
1 2
64(0.2031) 93(0.1505)
0.1719
64 93
n p n p
p
n n
+ +
= = =
+ +
,
Observed z is
0.2031 0.1505
0.858
1 1
(0.1719)(0.8281)
64 93
=
+
,
which does not lie in R. Hence, H
0
(b) Let

1
p = proportion for 19-24 year olds, and

2
p =proportion for 25-34 year olds.
Again, since H
1

0.05
2
Pooled estimate
1 1 2 2
1 2
47(0.0638) 64(0.2031)
0.1441
47 64
n p n p
p
n n
+ +
= = =
+ +
,
Observed z is
0.1393
2.065
1 1
(0.1441)(0.8559)
47 64
=
+
,
0

10.57 (a) Let p
1
and p
2
denote the probability of getting hepatitis for the vaccine
group and the placebo group, respectively. We formulate the hypotheses:
0 1 2
: H p p = versus
1 1 2
: H p p < .
1 2
1 2

1 1

p p
Z
pq
n n
=
+
. Since H
1
is left-sided and
0.01
: 2.33 R Z z = . We calculate the
following:

70 11
1 2 549 534
0.020, 0.131 p p = = = =
Pooled estimate
1 1 2 2
1 2
11 70
0.075
1083
n p n p
p
n n
+ +
= = =
+
,
Observed z is
0.020 0.131
6.93
1 1
(0.075)(0.925)
549 534
=
+
,
0
is rejected at 0.01 = . So, there is very
strong evidence that the vaccine is effective.
(b)
1 2
0.131 0.020 0.111 p p = =
Estimated S.E.=
1 1 2 2
1 2
(0.020)(0.980) (0.131)(0.869)
549 534
0.016
p q p q
n n
+ = +
=

1 2
p p is given by
( ) 0.05
2
1 1 2 2
1 2
1 2

0.111 1.96(0.016) 0.111 0.031
p q p q
p p z
n n
+ = =
or (0.08, 0.14).

10.58 (a) We formulate the hypotheses:
0
:
D C
H p p = versus
1
:
D C
H p p > .
1 2
1 2

1 1

p p
Z
pq
n n
=
+
, where we identify and
D C
p p as
1 2
and p p , respectively. Since H
1
is right-sided and 0.10 = , the
rejection region is
0.10

43 41
1 2 160 200
0.256, 0.215 p p = = = =
Pooled estimate
1 1 2 2
1 2
41 43
0.233
160 200
n p n p
p
n n
+ +
= = =
+ +
,
Observed z is
0.256 0.215
0.91
1 1
(0.233)(0.767)
160 200
=
+
,
0
(b) We formulate the hypotheses:
0
:
E C
H p p = versus
1
:
E C
H p p > .
1 2
1 2

1 1

p p
Z
pq
n n
=
+
, where we identify and
E C
p p as
1 2
and p p , respectively. Since H
1
is right-sided, the rejection region is of
the form : R Z c . We calculate the following:
309

78 43
1 2 150 200
0.520, 0.215 p p = = = =
Pooled estimate
1 1 2 2
1 2
78 43
0.346
150 200
n p n p
p
n n
+ +
= = =
+ +
,
Observed z is
0.520 0.215
5.94
1 1
(0.346)(0.654)
150 200
=
+

The associated p-value [ 5.94] P Z is less than 0.0002. Hence, the data
strongly substantiate H
1
.

10.59 (a)
78 39
150 160
0.520, 0.244
E H
p p = = = =
0.276
E H
p p =
Estimated S.E.=
1 1 2 2
1 2
(0.520)(0.480) (0.244)(0.756)
150 160
0.053
p q p q
n n
+ = +
=

E H
p p is given by
( ) 0.05
2
1 2

0.276 1.96(0.053) 0.276 0.104
E E H H
E H
p q p q
p p z
n n
+ = =
or (0.17, 0.38).
(b)
43 39
200 160
0.215, 0.244
C H
p p = = = = , 0.029
H C
p p =
Estimated S.E.=
1 2
(0.244)(0.756) (0.215)(0.785)
160 200
0.045
C C H H
p q p q
n n
+ = +
=

H C
p p is given by
( ) 0.10
2
1 2

0.029 1.645(0.045) 0.029 0.074
C C H H
H C
p q p q
p p z
n n
+ = =
or ( 0.045, 0.103 ).
(c) A 95% confidence interval for a population proportion is given by

1.96
pq
p
n
. Calculations for the individual groups are as follows:
Diabetes:
41
160
160, 0.256
D
n p = = =
The confidence interval for
D
p is given by:
(0.256)(0.744)
0.256 1.96 0.256 0.068
160
= or (0.19, 0.32)
Heart condition:
39
160
160, 0.244
H
n p = = =
H
p is given by:
(0.244)(0.756)
0.244 1.96 0.244 0.067
160
= or (0.18, 0.31)
Epilepsy:
78
150
150, 0.520
E
n p = = =
E
p is given by:
(0.520)(0.480)
0.520 1.96 0.520 0.080
150
= or (0.44, 0.60)
Control:
43
200
20, 0.215
C
n p = = =
C
p is given by:
(0.215)(0.785)
0.215 1.96 0.215 0.057
200
= or (0.16, 0.27)
10.60 (a) A large sample 98% confidence interval for
1 2
is given by

( ) 0.02
2
2 2 2 2
1 2
1 2
(1.609) (1.643)
(3.00 4.429) 2.33
18 28
1.429 2.33(0.4901) 1.429 1.142
s s
x y z
n n
+ = +
= =

or ( 2.571, 0.287 ).
(b) The conclusion of testing
0 1 2 1 1 2
: 0 versus : 0 H H = at
0.02 = can be readily drawn from part (a). Because 0 is inside the 98%
confidence interval, H
0

1 2
is given by

( ) 0.05
2
2 2 2 2
1 2
1 2
(1.1) (3.4)
(12.2 7.2) 1.96
40 40
5 1.96(0.565) 5 1.1074
s s
x y z
n n
+ = +
= =

or (3.893, 6.107).

(b) You need to make certain that the group of 80 is completely randomized,
in some way. Then, choose 40 at random from it to minimize any
potential bias.

1 2
is given by

( ) 0.05
2
2 2 2 2
1 2
1 2
235 (831.3)
(177.2 563.9) 1.96
18 48
386.7 1.96(132.156) 386.7 259.026
s s
x y z
n n
+ = +
= =

or ( 645.726, 127.674 ).

2 1
150 > + , we formulate the hypotheses

0 1 2 1 2 1
: 150 versus : 150 H H = > .
311

1
18 n = and
2
48 n = are sufficiently large, we
employ the Z-test, and so, the test statistic is
( )
2 2
1 2
1 2
150 Y X
Z
S S
n n

=
+
. Since H
1
is
right-sided and 0.05 = , the rejection region is
0.05
: 1.645 R Z z = .
( ) 563.9 177.2 150
1.829
132.156
z

= = , which lies in R. Hence, H
0
is rejected
at 0.05 = . Furthermore, the associated p-value is
[ 1.829] 0.0335 P Z = , so the evidence is support of H
1
is strong.

A B
> , we formulate the hypotheses

0 1
: 0 versus : 0
A B A B
H H = > .
1
55 n = and
2
58 n = are large, we employ the Z-
test, and so, the test statistic is
2 2
1 2
1 2
X Y
Z
S S
n n
=
+
. Since H
1
is right-sided
and 0.10 = , the rejection region is
0.10
: 1.28 R Z z = .
2 2
4.64 4.03
2.09
(1.25) (1.82)
55 58
z

= =
+
0
is rejected at
0.10 = . Furthermore, the associated p-value is [ 2.09] 0.0183 P Z = ,
so the evidence is support of H
1
is strong.

A B
is given by

( ) 0.10
2
2 2 2 2
1 2
1 2
(1.25) (1.82)
(4.64 4.03) 1.645
55 58
0.61 0.48
s s
x y z
n n
+ = +
=

or (0.13, 1.09). We are 90% confident that
A
is 0.13 to 1.09 hours
longer than
B
.
(b) The 95% confidence interval for
A
is given by

0.05
2
1
1
1.25
4.64 1.96 4.64 0.33
55
s
x z
n
| |
= =
|
\
or (4.31, 4.97).
10.66 We have:
1 2
13, 13 n n = = ,
1 2
1.58, 1.15 s s = =
2 2 2 2
2 1 1 2 2
pooled
1 2
( 1) ( 1) 12(1.58) 12(1.15)
1.909
2 13 13 2
n s n s
s
n n
+ +
= = =
+ +
.
8 x = 5 y =

2
( ) 20
i
x x =

2
( ) 14
i
y y =

2 2
2
pooled
1 2
( ) ( )
20 14
4.857
2 5 4 2
i i
x x y y
s
n n
+
+
= = =
+ +

.
(b) The t-statistic is
( )
pooled
1 2
2
1 1
X Y
T
s
n n

=
+
with d.f. =
1 2
2 n n + . In the
( ) 8 5 2
0.6764
1 1
4.857
5 4
t

= =
+
, with d.f. = 7.

New Spray:
1
12 n = , 249 x = ,
1
19 s =
Standard Spray:
2
15 n = , 233 y = ,
2
45 s =
Denote by
1
and
2
the population mean yields under the use of the new spray
and the standard spray, respectively. Since the intent is to demonstrate that
1

is larger than
2
0 1 2 1 1 2
: 0 versus : 0 H H = > .
The sample sizes are small. We assume normal populations, but the assumption
of equal variances is not reasonable since
2
1
s
s
is larger than 2. Therefore, we use
the conservative test without pooling. As such, the test statistic is
*
2 2
1 2
1 2
X Y
T
S S
n n
=
+
with d.f. = smaller of
1 2
1, 1 n n .
Since H
1
is right-sided and 0.05 = , the rejection region is
*
0.05
: 1.796 R T t =
for d.f. = 11. Using the summary statistics, we obtain

2 2 2 2
1 2
1 2
249 233 16
19 45
12.848
12 15
x y
s s
n n
= =
+ = + =

Hence, the observed value of t
*
is given by
16
1.25
12.848
t
= = , which does not

lie in R. So, H
0

313

10.69 Noting that for d.f. = 11,
0.025
2.201 t = , and using the calculations in Exercise
10.68, we obtain the following 95% confidence interval for
1 2
:
( )
2 2
1 2
0.025
1 2
(249 233) 2.201(12.848)
16 28.278
s s
x y t
n n
+ =
=

So, the confidence interval is (-12.278, 44.278).

10.70 We test the hypotheses
0 1 2 1 1 2
: 0 versus : 0 H H = .

(a) The sample sizes are small. We assume normal populations, but the
assumption of equal variances is not reasonable since
2
1
s
s
is larger than 2.
Therefore, we use the conservative test without pooling. As such, the test
statistic is
*
2 2
1 2
1 2
X Y
T
S S
n n
=
+
with d.f. = smaller of
1 2
1, 1 n n .
Since H
1
0.05
2
: 2.447 R T t =
for d.f. = 6. Using the given sample statistics, we obtain
2 2
12.2 7.2
3.724
1.1 3.4
8 7
t

= =
+
.
This is in R, so we reject H
0
at

0.05 = .

(b) For the approximate t-procedure, we use the same test statistic as in part (a), but
now with the following estimated d.f.:
Est. d.f. =
2
2 2
1 2
1 2
2 2
2 2
1 2
1 2
1 2
1 1
S S
n n
S S
n n
n n
| |
+
|
\
| | | |
| |
\ \
+

.
Substituting in the information provided in this formula yields Est. d.f. = 5.848,
so that we round down to 5.
Now, we proceed as in (a). Since H
1
is two-sided and 0.05 = , the rejection
region is
0.05
2
: 2.571 R T t = for d.f. = 5. This is even more conservative than
in (a), but even so, since t = 3.724 is still in R, we reject H
0
at 0.05 = .

(c) Assume normal populations with unequal variances.


City A:
1
75 n = , 37.8 x = ,
1
6.8 s =
City B:
2
100 n = , 43.2 y = ,
2
7.5 s =

(a) Denote by
1
and
2
the population mean age for City A and City B,
respectively. In order to demonstrate a difference between
1
and
2
, we
formulate the hypotheses

0 1 2 1 1 2
: 0 versus : 0 H H = .
Since the sample sizes are large, we employ the Z-test, and so, the test
statistic is
2 2
1 2
1 2
X Y
Z
S S
n n
=
+
. Since H
1
is two-sided and 0.02 = , the
rejection region is
0.02
2
: 2.33 R Z z = . Using the summary statistics, we
obtain
2 2 2 2
1 2
1 2
37.8 43.2 5.4
(6.8) (7.5)
1.086
75 100
x y
s s
n n
= =
+ = + =

Hence, the observed value of z is
5.4
4.97
1.086
Z

= = , which lies in R.
Hence, H
0
is rejected at 0.02 = , and so there is strong evidence that the
mean ages are different.
1 2
is given by

( ) 0.02
2
2 2 2 2
1 2
1 2
(6.8) (7.5)
(37.8 43.2) 2.33 5.4 2.53
75 100
s s
x y z
n n
+ = + =
.
or ( 7.93, 2.87 ) . We are 98% confident that the mean age for City B is 2.87
to 7.93 years higher than for City A.
(c) The individual 98% confidence intervals are given by:

( )
1
1
1
6.8
75
: 2.33 37.8 2.33
s
n
x = or (35.97, 39.63) years

( )
2
2
2
7.5
100
: 2.33 43.2 2.33
s
n
y = or (41.45, 44.95) years


Female:
1
36 n = , 52.92 x = ,
1
8.83 s =
Male:
2
25 n = , 59.70 y = ,
2
17.5 s =

315

Denote by
1 2
and the population mean neck sizes for female and male,
respectively. In order to determine if
1 2
and are significantly different, we
test the hypotheses
0 1 2 1 1 2
: 0 versus : 0 H H = with 0.03 = . We
assume normal populations with equal variance, so the test statistic is
pooled
1 2
1 1
X Y
T
s
n n
=
+
with d.f. =
1 2
2 n n +
Since H
1
0.03
2
: 2.23 R T t =
(for d.f. = 36 + 25 2 = 59). From the sample data, we calculate the following:
52.92 59.70 6.78 x y = =

2 2 2 2
1 1 2 2
pooled
1 2
( 1) ( 1) 35(8.83) 24(17.5)
13.070
2 59
n s n s
s
n n
+ +
= = =
+
.
6.78
1.993
1 1
13.070
36 25
t

= =
+
,
0

10.73 (a) Since the p-value is 0.067, we do not reject H
0
at 0.05 = .
1 2
is given by

( ) 0.05
2
2 2 2 2
1 2
1 2
(0.554) (0.390)
(2.643 2.945) 1.96
15 20
0.302 1.96(0.1675) 0.302 0.3283
s s
x y z
n n
+ = +
= =

or ( 0.6303 , 0.0263).

(c) Denote by
1 2
and the population mean CAS for female and male,
respectively. In order to determine if
1 2
and are significantly different,
we test the hypotheses
0 1 2 1 1 2
: 0.1 versus : 0.1 H H = with
0.05 = . We assume normal populations with equal variance, so the test
statistic is
pooled
1 2
1 1
X Y
T
s
n n
=
+
with d.f. =
1 2
2 n n +
Since H
1
0.05
2
: 2.039 R T t = (for d.f. = 15 + 20 2 = 33). From the sample data,
we calculate the following:
2.643 2.945 0.302 x y = =

2 2 2 2
1 1 2 2
pooled
1 2
( 1) ( 1) 14(0.554) 19(0.390)
0.467
2 33
n s n s
s
n n
+ +
= = =
+
.
0.302
1.89
0.3416(0.467)
t

= = ,
0

10.74 (a) The experimental units are the 12 SUVs. Since the gasoline could vary
considerably in different cars, a matched pair design is appropriate. With
each car, observe the mileage with a tank full of each brand of gasoline.
Randomize the order of the two brands independently for each SUV.
(b) A matched pair design would be effective. Use the two brands, one on
each half of wall. Randomize between the two halves. Each half of a
wall is an experimental unit.
(c) Each girl is an experimental unit. Since they are homogenous with regard
to age, the design of independent samples will be appropriate for
comparing the two teaching methods. Divide the 25 girls at random into
two groups of 12 and 13. Use Method 1 for one group and Method 2 for
the other group.

10.75 (a) This is a matched pair design. From the data on the paired differences d =
(lab A lab B), we obtain the following summary statistics:
9, 9.667, and 14.5
D
n d s = = =
We assume that the population distribution of these differences is normal,
and denote the population mean by . We formulate the hypotheses
0
: 0 H = versus
1
: 0 H .
/
D
D
T
S n
= , d.f. = 1 n . Since H
1
is two-sided and
0.01
: 2.896 R T t = (for d.f. = 8). Using
9.667
14.5/ 9
2.00 t = = ,
0
is not rejected at 0.02 = , and so we
conclude that the difference is not significant at this level.
(b) For 0.10 = and d.f. = 8, we have
0.05
( )
14.5
0.05
9
9.667 1.860 9.667 8.999
D
s
n
d t = = or ( 0.6789, 18.654 ).

10.76 (a) This is matched pair design. From the data on paired differences
(Scale 1 Scale 2), we obtain the following summary statistics:
5, 1.08, 0.8497
d
n d s = = =
317

Assume the populations are normal. Then, for 0.05 = , d.f. = 4, we have
0.025
2.776 t = . So, a 95% confidence interval for is:
0.025
0.8497
1.08 2.776 1.08 1.055
5
d
s
d t
n
| | | |
= =
| |
\ \

or (0.02495, 2.135).
(b) 0 is not inside the confidence interval in (a), so we reject H
0
.
(c) Randomly select people at the facility at different times of the day and of
different body builds.

10.77 This is a matched pair design, where each location gives a pair. We calculate
the paired differences d = (with additive without additive) as:
3, 3, 1, 1, 3, 1, 4, 2, 4, 4 . From this, we obtain the following summary
statistics: 10, 2.0, and 2.160
D
n d s = = =
denote the population mean by . We formulate the hypotheses
0
: 0 H = versus
1
: 0 H > .
/
D
D
T
S n
= , d.f. = 1 n . Since H
1
is right-sided and
0.05
: 1.833 R T t = (for d.f. = 9). Using the
summary statistics, we see that the observed t is
2.0
2.160/ 10
2.93 t = = , which lies in
R. Hence, H
0
is rejected at 0.05 = , and so we conclude that there is strong
evidence that the additive is effective.

10.78 (a) For each location, toss a coin. If a Head appears, choose East for the
additive, and if Tail appears, choose West.
(b) W W E W E E W E E E (Answers will vary.)

10.79 (a) This is a case of independent random samples. We calculate the following
summary statistics:
Path A:
1
6 n = , 12.5 x = ,
1
2.429 s =
Path B:
2
6 n = , 15.167 y = ,
2
2.317 s =
Denote by
1
and
2
the population mean travel times for Path A and
Path B, respectively. Since the conjecture is that
1
and
2
are different,
we formulate the hypotheses
0 1 2 1 1 2
: 0 versus : 0 H H = .
pooled
1 2
1 1
X Y
T
s
n n
=
+
with d.f. =
1 2
2 n n +
Since H
1
0.05
2
: 2.228 R T t = for d.f.

2 2 2 2
1 1 2 2
pooled
1 2
2.667
( 1) ( 1) 5(2.429) 5(2.317)
2.374
2 10
x y
n s n s
s
n n
=
+ +
= = =
+


2.667
1.947
1 1
2.374
6 6
t

= =
+
,
0
is not rejected at 0.05 = . We conclude
that the difference is not significant.
(b) Each driver could take both Path A and Path B. This would be a paired
comparison that removes differences between drivers.

10.80 (a) The paired differences d = (Method 1 Method 2) are 2, 3,1, 0,1 . For
0.05 = and d.f. = 4, we have
0.025
( )
1.8166
0.025
5
0.6 2.776 0.6 2.776(0.812)
D
s
n
d t = = or ( 2.85, 1.65 ).
(b) Method 1:
2
15.0, ( ) 20.0
i
x x x = =

Method 2:
2
15.6, ( ) 5.2
i
y y y = =

2
pooled
pooled
1 2
20.0 5.2
3.15
8
1 1 1 1
1.775 1.123
5 5
s
s
n n
+
= =
+ = + =

For d.f. = 8,
0.025
2.306 t = . Hence, the 95% confidence interval for
1 2
is given by
(15.0 15.6) 2.306(1.123) 0.6 2.59 = or ( 3.19, 1.99) .
(c) The analysis of a paired sample is appropriate if (i) in each batch the hemp
is homogeneous, but there is variation in different batches, or (ii) test
conditions vary from pair to pair, but are alike in each pair. Otherwise,
treating as independent samples will be appropriate as more d.f. will be
gained in t.

10.81 Let p
1
and p
2
denote the probabilities of rain from seeded clouds and non-seeded
clouds, respectively. We formulate the hypotheses:
0 1 2
: H p p = versus
1 1 2
: H p p < .
1 2
1 2

1 1

p p
Z
pq
n n
=
+
319

7 43
1 2 50 165
0.140, 0.261 p p = = = =
Pooled estimate
1 1 2 2
1 2
50
0.233
215
n p n p
p
n n
+
= = =
+
,
Observed z is
0.140 0.261
1.77
1 1
(0.233)(0.767)
50 165
=
+
,
The associated p-value is [ 1.77] 0.0384 P Z = . So, H
0
would be rejected for
as small as 0.0384. As such, there is fairly strong evidence in support of the
conjecture.

10.82
1 2
0.140 0.261 0.121 p p = =
Estimated S.E. =
1 1 2 2
1 2
(0.140)(0.860) (0.261)(0.739)
0.060
50 165
p q p q
n n
+ = + =
1 2
p p is given by
( ) 0.05
2
1 1 2 2
1 2
1 2

0.121 1.645(0.060) 0.121 0.099
p q p q
p p z
n n
+ = =
or ( 0.22, 0.02 ).

10.83 (a) Let p
1
and p
2
denote the population proportions of uremic and normal
patients, respectively, who are allergic to the antibiotic. We formulate the
hypotheses:
0 1 2
: H p p = versus
1 1 2
: H p p > .
1 2
1 2

1 1

p p
Z
pq
n n
=
+
. Since H
1
is right-sided and
0.01
2
: 2.33 R Z z = . We calculate the
following:

38 21
1 2 100 100
0.38, 0.21 p p = = = =
Pooled estimate
1 1 2 2
1 2
38 21
0.295
200
n p n p
p
n n
+ +
= = =
+
,
Observed z is
0.38 0.21
2.64
1 1
(0.295)(0.705)
100 100
=
+
,
0
is rejected at 0.01 = . The associated p-value
is [ 2.64] 0.0041 P Z = . As such, there is strong evidence of a higher
incidence of allergy in uremic patients.

(b)
1 2
0.38 0.21 0.17 p p = =
Estimated S.E. =
1 1 2 2
1 2
(0.38)(0.62) (0.21)(0.79)
0.063
100 100
p q p q
n n
+ = + =
1 2
p p is given by
( ) 0.05
2
1 1 2 2
1 2
1 2

0.17 1.96(0.063) 0.17 0.12
p q p q
p p z
n n
+ = =
or ( 0.05, 0.29 ).

10.84 (a) Let p
F
and p
M
denote the true proportion of females and males who donate
to charity respectively. We formulate the hypotheses:
0
:
F M
H p p = versus
1
:
F M
H p p > .

1 1

F M
F M
p p
Z
pq
n n
=
+
, where and
F M
p p are identified
with and
F M
p p . Since H
1
is right-sided and 0.05 = , the rejection
region is
0.05
2
: 1.96 R Z z = . Using the given sample statistics, we have
Pooled estimate
1 1 2 2
1 2
195(0.2462) 294(0.1769)
0.205
195 294
n p n p
p
n n
+ +
= = =
+ +
,
Observed z is
0.2462 0.1769
1.858
1 1
(0.205)(0.795)
195 294
=
+
,
0
(b) 0.0693
F M
p p =
Estimated S.E.
=
(0.2462)(0.7538) (0.1769)(0.8231)
0.0380
195 294
F F M M
F M
p q p q
n n
+ = + =
F M
p p is given by

( ) 0.05
2

0.0693 1.96(0.0380) 0.0693 0.07448
F F M M
F M
F M
p q p q
p p z
n n
+ = =
or (-0.00518, 0.14378).

10.85 (a) Since the p-value is 0.127, which is greater than 0.05, there is not strong
enough evidence to reject H
0
.
(b) Enter the data into columns C3 and C4 of a Minitab worksheet. The
output is as follows:

Two Sample T-Test and Confidence Interval
Two sample T for C3 vs C4

N Mean StDev SE Mean
C3 8 73.9 10.1 3.6
C4 11 91.9 12.4 3.7
321

97% CI for mu C3 - mu C4: ( -30.3, -5.8)
T-Test mu C3 = mu C4 (vs not =): T = -3.50
P = 0.0030 DF = 16

The 97% confidence interval is ( 30.3, 5.8 ).

10.86 Denote by
1 2
and the population mean for males and females, respectively.
The summary statistics are:

Male:
1
5 n = , 13.40 x = ,
1
8.82 s =
Female:
2
11 n = , 13.09 y = ,
2
5.52 s =

A 95% confidence interval for
1 2
has the form
( ) pooled
1 2
2
1 1
X Y t s
n n
+ .
Using the sample statistics, we obtain:

2 2
pooled
0.31
4(8.82) 10(5.52)
6.63
14
x y
s
=
+
= =

Since
0.025
2.145 t = for d.f. = 5 + 11 2 = 14, the 95% confidence interval in this
case is
1 1
(13.40 13.09) (2.145)(6.63) 0.31 2.145(3.58)
5 11
0.31 7.68
+ =
=

or ( 7.4, 8.0 ). The confidence interval includes 0, so this is one plausible
value for
1 2
. Males should have much more testosterone than females, so
we expected the difference of means to have a large positive value. The males
are apparently not healthy.

10.87 (a) Denote by
1 2
and the population mean for males at Lake Apopka and
Lake Woodfruff, respectively. The summary statistics are:

Lake Apopka:
1
5 n = , 13.40 x = ,
1
8.82 s =
Lake Woodruff:
2
9 n = , 50.44 y = ,
2
27.13 s =

We should not pool the variances since the second standard deviation is
more than 3 times the first.

(b) Using the conservative procedure, the 90% confidence interval for
1 2

has the form

( )
2 2
1 2
1 2
2
s s
X Y t
n n
+ , d.f. = smaller of
1 2
1, 1 n n .
Using the sample statistics, we obtain:

2 2 2 2
1 2
1 2
13.40 50.44 37.04
(8.82) (27.13)
9.87
5 9
x y
s s
n n
= =
+ = + =

Since
0.025
2.132 t = for d.f. = 4, the 90% confidence interval in this case is
37.40 (2.132)(9.87) 37.40 21.04 = or ( 58.0, 16.0 ).
(c) The mean for males in the control group at Lake Woodruff is from 16 to
58 units larger than the mean for males at Lake Apopka.

10.88 Enter the data into columns C3 and C4 of a Minitab worksheet. The output is as
follows:
Two Sample T-Test and Confidence Interval

Two sample T for Head width (males) vs Head width (females)

Head wid 25 32.11 5.24 1.0
Head wid 36 30.45 4.06 0.68

95% CI for mu Head wid - mu Head wid: ( -0.9, 4.17)
T-Test mu Head wid = mu Head wid (vs not =): T = 1.33
P = 0.19 DF = 43

From the Minitab output, we see that a 95% confidence interval for the difference
of means of male and female bear head width is given by ( 0.9, 4.17 ).
The result of the hypothesis test of
0 1 2
: 0 H = versus
1 1 2
: 0 H from
Minitab (not assuming equal variances) is to not reject H
0
since the p-value is
rather large, being 0.19.

10.89 Enter the data into columns C9 and C10 of a Minitab worksheet. The following
is the output:


C9 40 455.1 37.3 5.9
C10 40 429.1 41.0 6.5

95% CI for mu C9 - mu C10: ( 8.5, 43.4)
T-Test mu C9 = mu C10 (vs not =): T = 2.96 P = 0.0041 DF = 77

This was done without pooling the variance note that since the ratio of the
standard deviations is less than 1, one could have pooled the variance to get a
323

slightly improved test statistic. Nonetheless, the p-value is 0.0041 with the
more conservative estimate, so we reject H
0
and claim that the two means are
different. Note also the 95% confidence interval is given.

10.90 Enter the data into two columns, say C1 and C2, of a Minitab worksheet.. The
output is as follows:


C1 81 38.62 9.06 1.0
C2 81 45.05 9.64 1.1

95% CI for mu C1 - mu C2: ( -9.3, -3.5)
T-Test mu C1 = mu C2 (vs not =): T = -4.38 P = 0.0000 DF = 159

Note the 95% confidence interval, and the test is very significant. Also, note
that since the ratio of the standard deviations is close to 1, so we could have
pooled the variance for an even better estimate.

10.91 Enter the data into two columns of a Minitab worksheet. The output is as
follows:

Two sample T for Pre row vs Post row
Pre row 81 725.9 89.3 9.9
Post row 81 673.3 83.0 9.2
95% CI for mu Pre row - mu Post row: ( 25.9, 79.4)
T-Test mu Pre row = mu Post row (vs not =): T = 3.89
P = 0.0001 DF = 159
There is a highly significant difference between mean pre row time and post
row time.


MAT 540 Statistical Concepts For Research

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

MAT 540 Statistical Concepts For Research

Uploaded by

Copyright:

Available Formats

287

= = , which does not

You might also like