Anda di halaman 1dari 13

438 Chapter 8 Inferences about More Than Two Population Central Values

8.9 Exercises
8.1 Introduction
Med. 8.1 For the port-wine stains research study, answer the following:
a. What are the populations of interest?
b. What are some factors besides change in skin color that may be of interest to the
investigators?
Med. 8.2 For the port-wine stains research study, do the following:
a. Describe how the subjects in this experiment could have been selected so as to satisfy
the randomization requirements.
b. State several research hypotheses that may have been of interest to the researchers.

8.2 A Statistical Test about More Than Two Population Means: An Analysis of Variance
Theory 8.3 Four techniques of teaching algebra are to be compared in five high schools. Four mathe-
matics teachers are randomly selected in each of the five schools and the four techniques are
randomly assigned to the teachers. The students are given a standardized algebra exam at the
end of the semester with the average score in each classroom used as the measure of the effec-
tiveness of the teaching technique. Therefore, five measurements of the teaching effectiveness
are obtained for each of the four teaching techniques. Would it be appropriate to use the AOV
F test to evaluate whether there is a difference in the average scores of the four teaching
techniques?
Theory 8.4 In Example 8.2, suppose the psychologist wanted to compare method 1 to method 2. What
is the advantage of using a t test having s 2W in the denominator as opposed to using the conven-
tional pooled t test of Chapter 5, with s2p, the average of the sample variances from method 1 and
method 2 data?
Theory 8.5 For an experiment comparing t treatments with sample sizes n1, n2, . . . , nt and sample vari-
ances s21, s22 , . . . , s2t , consider the following questions.
a. If the sample sizes satisfy n1  . . .  nt, show the s 2W is the average of the t sample
variances, s21, s22 , . . . , s2t .
b. Does this hold if the sample sizes are not equal? If not, why not just use the average?
Ag. 8.6 A large laboratory has four types of devices used to determine the pH of soil samples. The
laboratory wants to determine whether there are differences in the average readings given by
these devices. The lab uses 24 soil samples having known pH in the study, and randomly assigns
six of the samples to each device. The soil samples are tested and the response recorded is the dif-
ference between the pH reading of the device and the known pH of the soil. These values, along
with summary statistics, are given in the following table.

Sample
Sample Standard
Device 1 2 3 4 5 6 Size Mean Deviation

A .307 .294 .079 .019 .136 .324 6 .1605 .1767


B .176 .125 .013 .082 .091 .459 6 .0947 .2091
C .137 .063 .240 .050 .318 .154 6 .1227 .1532
D .042 .690 .201 .166 .219 .407 6 .2735 .2492

a. Based on your intuition, is there evidence to indicate any difference among the mean
differences in pH readings for the four devices?
b. Run an analysis of variance to confirm or reject your conclusion of part (a). Use a  .05.
c. Compute the p-value of the F test in part (b).
d. What conditions must be satisfied for your analysis in parts (b) and (c) to be valid?
e. Suppose the 24 soil samples have widely different pH values. What problems may
occur by simply randomly assigning the soil samples to the different devices?
8.9 Exercises 439

Bus. 8.7 A cigarette manufacturer has advertised that it has developed a new brand of cigarette,
LowTar, that has a lower average tar content than the major brands. To evaluate this claim, a con-
sumer testing agency randomly selected 100 cigarettes from each of the four leading brands of
cigarettes and 100 from the new brand. The tar content (milligrams) of the cigarettes gave the fol-
lowing results:

Brand yi si ni

LowTar 9.64 .291 100


A 10.22 .478 100
B 10.77 .372 100
C 11.57 .352 100
D 13.59 .469 100

A boxplot of the data used to produce the table are given here.

Boxplots of tar content by 15


brand for Exercise 8.7
(means are indicated by 14
solid circles)
13
Tar content

*
12 **

11

10

9
LowTar A B C D
Brand

a. Based on the information contained in the boxplot, does the LowTar brand appear to
have a lower average tar content than the other brands?
b. Using the computer output shown here, is there a significant (a  .01) difference in
the average tar content of the five brands of cigarettes?
c. What is the p-value of the test statistic in (b)?
d. What are the practical consequences of making a Type I error with respect to your test
in (b)?

One-Way Analysis of Variance for Exercise 8.7

Analysis of Variance for Tar Cont


Source DF SS MS F P
Brand 4 941.193 235.298 1478.39 0.000
Error 495 78.784 0.159
Total 499 1019.976
Individual 95% CIs for Mean
Based on Pooled StDev
Level N Mean StDev -+---------+---------+---------+-----
1 100 9.644 0.291 (*)
2 100 10.221 0.478 (*)
3 100 10.775 0.372 (*)
4 1 00 11 . 57 0 0 . 3 52 (*)
5 100 1 3.5 9 2 0. 46 9 (*)
-+---------+---------+---------+-----
Pooled StDev = 0.399 9.6 10.8 12.0 13.2
440 Chapter 8 Inferences about More Than Two Population Central Values

8.3 The Model for Observations in a Completely Randomized Design


Theory 8.8 Four populations are to be compared based on differences in their means. Suppose the pop-
ulation means are given as follows:
m1  20 m2  25 m3  15 m4  35
Using the relationship mi  m  ti, compute the values of m and ti.
Consum. 8.9 Refer to Example 8.1. Apply the model yij  m  ti  eij to the data in this example by iden-
tifying the values of t, n1, n2, and n3. Also, estimate the values of m, ti, s from the observed data.
Med. 8.10 Refer to Example 8.2. Apply the model yij  m  eij to the data in this example by iden-
tifying the values of t, n1, n2, and n3. Also, estimate the values of m, ti, s from the observed data.

8.4 Checking on the AOV Conditions


Theory 8.11 Suppose that in a study for comparing five population means ni  10 for i  1, . . . , 5. The
data yield s2W  0. What can we conclude about the 50 residuals: eij  yij  yi?
Consum. 8.12 Refer to Example 8.1.
a. From the data set compute the 15 residuals.
b. Are the conditions for conducting the AOV F test satisfied by this data set?
Med. 8.13 Refer to Example 8.2. Are the conditions for conducting the AOV F test satisfied by this
data set?
Med. 8.14 Refer to Exercise 8.6. Are the conditions for conducting the AOV F test satisfied by this
data set?
Med. 8.15 Refer to Exercise 8.7. Are the conditions for conducting the AOV F test satisfied by this
data set?

8.5 An Alternative Analysis: Transformations of the Data


Envir. 8.16 Refer to Example 8.4.
a. Apply the AOV F test to the original measurements using a  .05.
b. Apply the AOV F test to the transformed data using a  .05
c. Did transforming the data alter your conclusion whether the oxygen content is
related to the distance to the mouth of the Mississippi River?
Pol. 8.17 Refer to Example 8.6.
a. Apply the AOV F test to the original measurements using a  .05.
b. Apply the AOV F test to the transformed data using a  .05
c. Did transforming the data alter your conclusion whether there is a difference in the
four geographical regions with respect to their opinion of the EPA regulations on air
pollution?
Engin. 8.18 Refer to Example 7.9. The consumer testing agency was interested in evaluating whether
there was a difference in the mean percentage increase in mpg of the three additives. In Exam-
ple 7.9, we showed that the data did not appear to have a normal distribution.
a. Apply the natural logarithm transformation to the data. Do the conditions for apply-
ing the AOV F test appear to hold for the transformed data?
b. Test for a difference in the means of the three additives using a  .05.
Biol. 8.19 Refer to Exercise 7.20.
a. The biologist hypothesized that the mean weight of deer raised in a zoo would differ
from the mean weight of deer raised either in the wild or on a ranch. Do the condi-
tions necessary for applying the AOV F test appear to be valid?
b. If the conditions for AOV F test are satisfied, then conduct the test to evaluate the
biologists claim. If not, then suggest a transformation, and conduct the test on the
transformed data.
8.9 Exercises 441

Edu. 8.20 Refer to Exercise 7.22.


a. The school administrators want to determine if there was a difference in the mean
increase in test scores between the four methods of instruction. Do the conditions
necessary for applying the AOV F test appear to be valid?
b. If the conditions for AOV F test are satisfied, then conduct the test to evaluate
whether there is a difference in the means for the four methods of instruction. If not,
then suggest a transformation and conduct the test on the transformed data.
Cons. 8.21 Refer to Exercise 3.55.
a. The state legislative committee in charge of allocations for food stamps wanted to de-
termine if there was a difference in the mean food expenditures among the five family
sizes. Do the conditions necessary for applying the AOV F test appear to be valid?
b. If the conditions for AOV F test are satisfied, then conduct the test to evaluate
whether there is a difference in the means for the four methods of instruction. If not,
then suggest a transformation and conduct the test on the transformed data.
8.22 Refer to Example 8.5. In many situations in which the difference in variances is not too
great, the results from the AOV comparisons of the population means of the transformed data are
very similar to those from the results that would have been obtained using the original data. In
these situations, the researcher is inclined to ignore the transformations because the scale of the
transformed data is not relevant to the researcher. Thus, confidence intervals constructed for the
means using the transformed data may not be very relevant. One possible remedy for this prob-
lem is to construct confidence intervals using the transformed data, and then perform an inverse
transformation of the endpoints of the intervals. Then we would obtain a confidence interval with
values having the same scale units of measurements as the original data.
a. Test the hypothesis that the mean hours of relief for patients on the three treatments
differs using a  .05. Use the original data.
b. Place 95% confidence intervals on the mean hours of relief for the three treatments.
c. Repeat the analysis in (a) and (b) using the transformed data.
d. Comment on any differences in the results of the test of hypotheses.
e. Perform an inverse transformation on the endpoints of the intervals constructed
in (c). Compare these intervals to the ones constructed in (b).

8.6 A Nonparametric Alternative: The KruskalWallis Test


Engin. 8.23 In a 1996 article published in Technometrics, 38, pp. 1122, the authors discuss the relia-
bility of nuclear power plant emergency generators. To control the risk of damage to the nuclear
core during accidents at nuclear plants, the reliability of emergency diesel generators (EDG) to
start on demand must be maintained at a very high level. At each nuclear power plant there are a
number of such generators. An overall measure of reliability is obtained by counting the number
of times the EDGs successfully work when needed. The table here provides the number of suc-
cessful demands for implementation of an EDG between each subsequent failure in an EDG for
all the EDGs at each of seven nuclear power plants. A regulatory agency wants to determine if
there is a difference in the reliability of the seven nuclear power plants.

Plant ni Number of Times EDG Works

A 34 28 50 193 55 4 7 174 76 10 0 10 84 0 9 1 0 62
26 15 226 54 46 128 4 105 40 4 273 164 7 55 41 26 6
B 15 2 11 75 6 1 12 4 6 64 3 0 3 1 20 78
C 17 142 110 3 273 54 32 3 40 23 30 17 7 12 6 12 7 5
D 8 64 29 1 3 8 29 4 60
E 12 139 21 214 67 174 1 9 2 119 237 110 71
F 7 18 108 9 8 17 88 28
G 10 0 6 0 16 1 58 13 36 33 19
442 Chapter 8 Inferences about More Than Two Population Central Values

a. Do the conditions necessary for conducting the AOV F test appear to be satisfied
by these data?
b. Because the data are counts of number of successes for the EDGs, the
Poisson model may be an alternative to the normal based analysis. Apply a
transformation to the data and then apply the AOV F test to the transformed
data.
c. As a second alternative analysis which has fewer restrictions, answer the agencys
question by applying the KruskalWallis test to the reliability data.
d. Compare your conclusions to parts (a)(c). Which of the three procedures do you
feel more confident with its conclusion?

Envir. 8.24 Refer to Example 8.4.


a. Apply the KruskalWallis test to determine if there is a difference in the distri-
butions of oxygen content for the various distances to the mouth of the Mississippi
River.
b. Does your conclusion differ from the conclusion reached in Exercise 8.16?

Med. 8.25 Refer to Example 8.5.


a. Apply the KruskalWallis test to determine if there is a difference in the distributions
of pain reduction for the three analgesics.
b. Does your conclusion differ from the conclusion reached in Exercise 8.22?

Med. 8.26 Refer to Example 8.6.


a. Apply the KruskalWallis test to determine if there is a difference in the distributions
of opinions across the four geographical regions.
b. Does your conclusion differ from the conclusion reached in Exercise 8.17?

Engin. 8.27 In the manufacture of soft contact lenses, the actual strength (power) of the lens needs to
be very close to the target value for the lenses to properly fit the customers needs. In the paper,
An ANOM-type test for variances from normal populations, Technometrics (1997), 39:
274 283, a comparison of several suppliers is made relative to the consistency of the power of the
lenses. The following table contains the deviations from the target power of lenses produced
using materials from three different suppliers:

Lens

Supplier 1 2 3 4 5 6 7 8 9

A 189.9 191.9 190.9 183.8 185.5 190.9 192.8 188.4 189.0


B 156.6 158.4 157.7 154.1 152.3 161.5 158.1 150.9 156.9
C 218.6 208.4 187.1 199.5 202.0 211.1 197.6 204.4 206.8

a. Using the appropriate tests and plots given here, assess whether the data meet the
necessary conditions to use an AOV to determine whether there is a significant
difference in the mean deviations for the three suppliers.
b. Conduct an AOV with a  .05 and compare your results with the conclusions from (a).
c. Apply the KruskalWallis test to evaluate the research hypothesis that the three
suppliers have different distributions of deviations.
d. Suppose that a difference in mean deviation of 20 units would have commercial
consequences for the manufacture of the lenses. Does there appear to be a practical
difference in the three suppliers?
8.9 Exercises 443

220
210
200

Deviations
190
180
170
160
150
A B C
Suppliers

.999
.99
Probability

.95
.80
.50
.20
.05
.01
.001

-10 0 10
Residuals
Average: -0.0000000 Anderson-Darling Normality Test
StDev: 5.53971 A-Squared: 0.718
N: 27 P-Value: 0.054

Ag. 8.28 The Agricultural Experiment Station of a university tested two different herbicides and
their effects on crop yield. From 90 acres set aside for the experiment, the station used herbicide
1 on a random sample of 30 acres, herbicide 2 on a second random sample of 30 acres, and they
used the remaining 30 acres as a control. At the end of the growing season, the yields (in bushels
per acre) were as follows:

Sample Standard
Sample Mean Deviation Sample Sizes

Herbicide 1 90.2 6.5 30


Herbicide 2 89.3 7.8 30
Control 85.0 7.4 30

a. Use these data to conduct a one-way analysis of variance to test whether there is a
difference in the mean yields. Use a  .05.
b. Construct 95% condence intervals on the mean yields mi.
c. Which of the mean yields appear to be different?
444 Chapter 8 Inferences about More Than Two Population Central Values

Hort. 8.29 Researchers from the Department of Fruit Crops at a university compared four different
preservatives to be used in freezing strawberries. The researchers prepared the yield from a straw-
berry patch for freezing and randomly divided it into four equal groups. Within each group they
treated the strawberries with the appropriate preservative and packaged them into eight small plas-
tic bags for freezing at 0C. The bags in group I served as a control group, while those in groups II,
III, and IV were assigned one of three newly developed preservatives. After all 32 bags of straw-
berries were prepared, they were stored at 0C for a period of 6 months. At the end of this time, the
contents of each bag were allowed to thaw and then rated on a scale of 1 to 10 points for discol-
oration. (Note that a low score indicates little discoloration.) The ratings are given here:

Group I 10 8 7.5 8 9.5 9 7.5 7

Group II 6 7.5 8 7 6.5 6 5 5.5

Group III 3 5.5 4 4.5 3 3.5 4 4.5

Group IV 2 1 2.5 3 4 3.5 2 2

a. Use the following plots of the residuals and a test of the homogeneity of variances to
assess whether the conditions needed to use AOV techniques are satised with this
data set.
b. Test whether there is a difference in the mean ratings using a  .05.
c. Place 95% condence intervals on the mean ratings for each of the groups.
d. Conrm your results with the computer output given here.

One-Way Analysis of Variance for Exercise 8.29

Analysis of Variance for Ratings


Source DF SS MS F P
Group 3 159.187 53.062 55.67 0.000
Error 28 26.687 0.953
Total 31 185.875
Individual 95% CIs for Mean
Based on Pooled StDev
Group N Mean StDev --+---------+---------+---------+----
I 8 8.3125 1.0670 (---*--)
II 8 6.4375 1.0155 (--*---)
III 8 4.0000 0.8452 (---*---)
IV 8 2.5000 0.9636 (---*--)
--+---------+---------+---------+----
Pooled StDev = 0.9763 2.0 4.0 6.0 8.0

Boxplots of ratings by group 10


for Exercise 8.29 (means are
9
indicated by solid circles)
8
7
Ratings

6
5
4
3
2
1
0
I II III IV
Group
8.9 Exercises 445
Normal probability plot of
residuals for Exercise 8.29 .999
.99
.95

Probability
.80
.50
.20
.05
.01
.001

-1 0 1
Residuals
Average: 0 Anderson-Darling Normality Test
StDev: 0.927840 A-Squared: 0.503
N: 32 P-Value: 0.191

8.30 Refer to Exercise 8.29. In many situations in which the response is a rating rather than an
actual measurement, it is recommended that the KruskalWallis test be used.
a. Apply the KruskalWallis test to determine whether there is a shift in the distribution
of ratings for the four groups.
b. Is the conclusion reached using the KruskalWallis test consistent with the conclusion
reached in Exercise 8.29 using AOV?
H.R. 8.31 Salary disputes and their eventual resolutions often leave both employers and employ-
ees embittered by the entire ordeal. To assess employee reactions to a recently devised salary
and fringe benets plan, the personnel department obtained random samples of 15 employees
from each of three divisions in the company: manufacturing, marketing, and research. The per-
sonnel staff asked each employee sampled to respond (in condence) to a series of questions.
Several employees refused to cooperate, as reected in the unequal sample sizes. The data are
given here:

Manufacturing Marketing Research

Sample size 12 14 11
Sample mean 25.2 32.6 28.1
Sample variance 3.6 4.8 5.3

a. Write a model for this experimental situation.


b. Use the summary of the scored responses to compare the means for the three divi-
sions (the higher a score, the higher the employee acceptance). Use a  .01.
Ag. 8.32 Researchers record the yields of corn, in bushels per plot, for four different varieties of
corn, A, B, C, and D. In a controlled greenhouse experiment, the researchers randomly assign
each variety to eight of 32 plots available for the study. The yields are listed here:

A 2.5 3.6 2.8 2.7 3.1 3.4 2.9 3.5


B 3.6 3.9 4.1 4.3 2.9 3.5 3.8 3.7
C 4.3 4.4 4.5 4.1 3.5 3.4 3.2 4.6
D 2.8 2.9 3.1 2.4 3.2 2.5 3.6 2.7
446 Chapter 8 Inferences about More Than Two Population Central Values

a. Write an appropriate statistical model.


b. Perform an analysis of variance on these data and draw your conclusions. Use
a  .05.
8.33 Refer to Exercise 8.32. Perform a KruskalWallis analysis of variance by ranks (with
a  .05) and compare your results to those in Exercise 8.32.
Edu. 8.34 Doing homework is a nightly routine for most school-age children. The article Family in-
volvement with middle-grades homework: effects of differential prompting [Journal of Experi-
mental Education, 66:31 48], examines the question of whether parents involvement with their
childrens homework is associated with improved academic performance. Seventy-four sixth
graders and their families participated in the study. Researchers assigned the students, similar in
student academic ability and background, in one of three mathematics classes taught by the same
teacher, and randomly assigned the classes to one of the three treatment groups.

Group I, student /family prompt: Students were prompted to seek assistance from a family
member and the family was encouraged to provide assistance to students.
Group II, student prompt: Students were prompted to seek assistance from a family
member but there was no specic encouragement of family members to provide
assistance to students.
Group III, no prompts: Students were not prompted to seek assistance from a family
member nor were family members encouraged to provide assistance to students.
Thus, one class was assigned to each of the three treatment groups. The researchers gave
the students a posttest, with the results given here:

Treatment Number of Mean Posttest


Group Students Score

Student /family prompt 22 68%


Student prompt 22 66%
No prompt 25 67%

The researchers concluded that higher levels of family involvement were not associated with
higher student achievement in this study.
a. What is the population of interest in this study?
b. Based on the data collected, to what population can the results of this study be
attributed?
c. What is the effective sample for each of the treatment groups; that is, how many
experimental units were randomly assigned to each of the treatment groups?
d. What criticisms would you have for the design of this study?
e. Suggest an improved design for addressing the research hypothesis that family
involvement improves student performance in mathematics classes.
Gov. 8.35 In a 1994 Senate subcommittee hearing, an executive of a major tobacco company
testied that the accusation that nicotine was added to cigarettes was false. Tobacco company
scientists stated that the amount of nicotine in cigarettes was completely determined by the size
of tobacco leaf, with smaller leaves having greater nicotine content. Thus, the variation in nico-
tine content in cigarettes occurred due to a variation in the size of the tobacco leaves and was not
due to any additives placed in the cigarettes by the company. Furthermore, the company argued
that the size of the leaves varied depending on the weather conditions during the growing sea-
son, for which they had no control. To study whether smaller tobacco leaves had a higher nico-
tine content, a consumer health organization conducted the following experiment. The major
factors controlling leaf size are temperature and the amount of water received by the plants dur-
ing the growing season. The experimenters created four types of growing conditions for tobacco
8.9 Exercises 447

plants. Condition A was average temperature and rainfall amounts. Condition B was lower than
average temperature and rainfall conditions. Condition C was higher temperatures with lower
rainfall. Finally, condition D was higher than normal temperatures and rainfall. The scientists
then planted 10 tobacco plants under each of the four conditions in a greenhouse where tem-
perature and amount of moisture were carefully controlled. After growing the plants, the scien-
tists recorded the leaf size and nicotine content, which are given here:

Plant A Leaf Size B Leaf Size C Leaf Size D Leaf Size

1 27.7619 4.2460 15.5070 33.0101


2 27.8523 14.1577 5.0473 44.9680
3 21.3495 7.0279 18.3020 34.2074
4 31.9616 7.0698 16.0436 28.9766
5 19.4623 0.8091 10.2601 42.9229
6 12.2804 13.9385 19.0571 36.6827
7 21.0508 11.0130 17.1826 32.7229
8 19.5074 10.9680 16.6510 34.5668
9 26.2808 6.9112 18.8472 28.7695
10 26.1466 9.6041 12.4234 36.6952

Plant A Nicotine B Nicotine C Nicotine D Nicotine

1 10.0655 8.5977 6.7865 9.9553


2 9.4712 8.1299 10.9249 5.8495
3 9.1246 11.3401 11.3878 10.3005
4 11.3652 9.3470 9.7022 9.7140
5 11.3976 9.3049 8.0371 10.7543
6 11.2936 10.0193 10.7187 8.0262
7 10.6805 9.5843 11.2352 13.1326
8 8.1280 6.4603 7.7079 11.8559
9 10.5066 8.2589 7.5653 11.3345
10 10.6579 5.0106 9.0922 10.4763

a. Perform a one-way analysis of variance to test whether there is a signicant difference


in the average leaf size under the four growing conditions. Use a  .05.
b. What conclusions can you reach concerning the effect of growing conditions on the
average leaf size?
c. Perform a one-way analysis of variance to test whether there is a signicant difference
in the average nicotine content under the four growing conditions. Use a  .05.
d. What conclusions can you reach concerning the effect of growing conditions on the
average nicotine content?
e. Based on the conclusions you reached in (b) and (d), do you think the testimony of
the tobacco companies scientists is supported by this experiment? Justify your
conclusions.
8.36 Using the plots given here, do the nicotine content data in Exercise 8.35 suggest violations
of the AOV conditions? If you determine that the conditions are not met, perform an alternative
analysis and compare your results to those of Exercise 8.35.
448 Chapter 8 Inferences about More Than Two Population Central Values
Boxplots of leaf size by group 50
for Exercise 8.36 (means are
indicated by solid circles)
40

Leaf size
30

20

10

0
A B C D
Group

Probability plot of residuals


leaf size for Exercise 8.36 .999
.99
.95
Probability

.80
.50
.20
.05
.01
.001

-10 0 10
Leaf size residuals

Average: 0.0000000 Anderson-Darling Normality Test


StDev: 4.75535 A-Squared: 0.205
N: 40 P-Value: 0.864

Boxplots of nicotine by group


13
for Exercise 8.36 (means are
indicated by solid circles) 12
11
10
Nicotine

9
8
7
6 *

A B C D
Group
8.9 Exercises 449
Probability plot of residuals
nicotine content for .999
Exercise 8.36 .99

Probability
.95
.80
.50
.20
.05
.01
.001

-4 -3 -2 -1 0 1 2 3
Nicotine content residuals

Average: 0.0000000 Anderson-Darling Normality Test


StDev: 1.62647 A-Squared: 0.443
N: 40 P-Value: 0.273

Ag. 8.37 Scientists conducted an experiment to test the effects of ve different diets in turkeys. They
randomly assigned six turkeys to each of the ve diet groups and fed them for a xed period of time.

Group Weight Gained (pounds)

Control diet 4.1, 3.3, 3.1, 4.2, 3.6, 4.4


Control diet  level 1 of additive A 5.2, 4.8, 4.5, 6.8, 5.5, 6.2
Control diet  level 2 of additive A 6.3, 6.5, 7.2, 7.4, 7.8, 6.7
Control diet  level 1 of additive B 6.5, 6.8, 7.3, 7.5, 6.9, 7.0
Control diet  level 2 of additive B 9.5, 9.6, 9.2, 9.1, 9.8, 9.1

a. Plot the data separately for each sample.


b. Compute y and s2 for each sample.
c. Is there any evidence of unequal variances or nonnormality? Explain.
d. Assuming that the ve groups were comparable with respect to initial weights of the
turkeys, use the weight-gained data to draw conclusions concerning the different
diets. Use a  .05.
8.38 Run a KruskalWallis test for the data of Exercise 8.37. Do these results conrm what you
concluded from an analysis of variance? What overall conclusions can be drawn? Use a  .05.
Hort. 8.39 Some researchers have conjectured that stem-pitting disease in peach tree seedlings might
be related to the presence or absence of nematodes in the soil. Hence, weed and soil treatment using
herbicides might be effective in promoting seedling growth. Researchers conducted an experiment
to compare peach tree seedling growth with soil and weeds using with one of three treatments:
A: Control (no herbicide)
B: Herbicide with Nemagone
C: Herbicide without Nemagone
The researchers randomly assigned 6 of the 18 seedlings chosen for the study to each treat-
ment group. They treated soil and weeds in the growing areas for the three groups with the
appropriate herbicide. At the end of the study period, they recorded the height (in centimeters)
for each seedling. Use the following sample data to run an analysis of variance for detecting
differences among the seedling heights for the three groups. Use a  .05. Draw your conclusions.
Herbicide A 66 67 74 73 75 64
Herbicide B 85 84 76 82 79 86
Herbicide C 91 93 88 87 90 86
450 Chapter 8 Inferences about More Than Two Population Central Values

8.40 Refer to the data of Exercise 8.37. To illustrate the effect that an extreme value can have
on conclusions from an analysis of variance, suppose that the weight gained by the fth turkey in
the level 2, additive B group was 15.8 rather than 9.8.
a. What effect does this have on the assumptions for an analysis of variance?
b. With 9.8 replaced by 15.8, if someone unknowingly ran an analysis of variance, what
conclusions would he or she draw?
8.41 Refer to Exercise 8.40. What happens to the KruskalWallis test if you replace the value
9.8 by 15.8? Might there be a reason to run both a KruskalWallis test and an analysis of vari-
ance? Justify your answer.
Engin. 8.42 A small corporation makes insulation shields for electrical wires using three different types of
machines. The corporation wants to evaluate the variation in the inside diameter dimension of the
shields produced by the machines. A quality engineer at the corporation randomly selects shields
produced by each of the machines and records the inside diameters of each shield (in millimeters).
She wants to determine whether the means and standard deviations of the three machines differ.

Shield Machine A Machine B Machine C

1 18.1 8.7 29.7


2 2.4 56.8 18.7
3 2.7 4.4 16.5
4 7.5 8.3 63.7
5 11.0 5.8 18.9
6 107.2
7 19.7
8 93.4
9 21.6
10 17.8

a. Conduct a test for the homogeneity of the population variances. Use a  .05.
b. Would it be appropriate to proceed with an analysis of variance based on the results
of this test? Explain.
c. If the variances of the diameters are different, suggest a transformation that may alle-
viate their differences and then conduct an analysis of variance to determine whether
the mean diameters differ. Use a  .05.
d. Compare the results of your analysis in (c) to the computer output given here, which
was an analysis of variance on the original diameters.
e. How could the engineer have designed her experiment differently if she knew that the
variance of machine B and machine C were so much larger than that of machine A?

One-Way Analysis of Variance for Exercise 8.42

Analysis of Variance
Source DF SS MS F P
Factor 2 4141 2071 2.73 0.094
Error 17 12907 759
Total 19 17048 Individual 95% CIs for Mean
Based on Pooled StDev
Level N Mean StDev --------+---------+---------+--------
Machine 5 8.32 6.52 (---------*----------)
Machine 5 16.78 22.43 (----------*---------)
Machine 10 40.70 34.52 (------*-------)
--------+---------+---------+--------
Pooled StDev = 27.55 0 25 50

8.43 The KruskalWallis test is not as highly affected by unequal variances as the AOV test.
Demonstrate this result by applying the KruskalWallis test to both the original and transformed
data and comparing the conclusions reached in this analysis for the data of Exercise 8.42.