03758_12_ch11_p425-481.

qxd 9/7/11 1:00 PM Page 452

452 ❍ CHAPTER 11 THE ANALYSIS OF VARIANCE

isolating the block variation must outweigh the loss of degrees of freedom for
error. Usually, though, if you suspect that the experimental units are not
homogeneous and you can group the units into blocks, it pays to use the
randomized block design!
• Finally, remember that you cannot construct confidence intervals for
individual treatment means unless it is reasonable to assume that the b blocks
have been randomly selected from a population of blocks. If you construct
such an interval, the sample treatment mean will be biased by the positive
and negative effects that the blocks have on the response.

11.8 EXERCISES

BASIC TECHNIQUES b. Do the data present sufficient evidence to indicate dif-

11.28 A randomized block design was used to com- ferences among the block means? Test using a ! .05.
pare the means of three treatments within six blocks. c. Rank the four treatment means using Tukey’s
Construct an ANOVA table showing the sources of method of paired comparisons with a ! .01.
variation and their respective degrees of freedom. d. Find a 95% confidence interval for the difference in
11.29 Suppose that the analysis of variance calcula- means for treatments A and B.
tions for Exercise 11.28 are SST ! 11.4, SSB ! 17.1, e. Does it appear that the use of a randomized
and Total SS ! 42.7. Complete the ANOVA table, block design for this experiment was justified?
showing all sums of squares, mean squares, and perti- Explain.
nent F-values. 11.34 The data shown here are observations
11.30 Do the data of Exercise 11.28 provide suffi- EX1134 collected from an experiment that compared
cient evidence to indicate differences among the treat- three treatments, A, B, and C, within each of five
ment means? Test using a ! .05. blocks, using a randomized block design:
Block
11.31 Refer to Exercise 11.28. Find a 95% confi-
dence interval for the difference between a pair of Treatment 1 2 3 4 5 Total
treatment means A and B if x!A ! 21.9 and !xB ! 24.2. A 2.1 2.6 1.9 3.2 2.7 12.5
B 3.4 3.8 3.6 4.1 3.9 18.8
11.32 Do the data of Exercise 11.28 provide suffi- C 3.0 3.6 3.2 3.9 3.9 17.6
cient evidence to indicate that blocking increased the
Total 8.5 10.0 8.7 11.2 10.5 48.9
amount of information in the experiment about the
treatment means? Justify your answer. MS Excel output for Exercise 11.34

11.33 The data that follow are observations Anova: Two-Factor Without Replication
EX1133 collected from an experiment that compared SUMMARY Count Sum Average Variance
A 5 12.5 2.5 0.265
four treatments, A, B, C, and D, within each of three B 5 18.8 3.76 0.073
blocks, using a randomized block design. C 5 17.6 3.52 0.167
1 3 8.5 2.833 0.443
Treatment 2 3 10 3.333 0.413
Block A B C D Total 3 3 8.7 2.9 0.79
4 3 11.2 3.733 0.223
1 6 10 8 9 33 5 3 10.5 3.5 0.48
2 4 9 5 7 25 ANOVA
3 12 15 14 14 55 Source of Variation SS df MS F P-value F crit
Total 22 34 27 30 113 Rows 4.476 2 2.238 79.929 0.000 4.459
Columns 1.796 4 0.449 16.036 0.001 3.838
a. Do the data present sufficient evidence to indicate Error 0.224 8 0.028
differences among the treatment means? Test using Total 6.496 14
a ! .05.
03758_12_ch11_p425-481.qxd 9/7/11 1:00 PM Page 453

11.8 THE ANALYSIS OF VARIANCE FOR A RANDOMIZED BLOCK DESIGN ❍ 453

Use the Excel ouput to analyze the experiment. Inves- c. Suppose that prior to looking at the data, you had
tigate possible differences in the block and/or treat- decided to compare the mean mileage per gallon for
ment means and, if any differences exist, use an formulations A and B. Find a 90% confidence inter-
appropriate method to specifically identify where the val for this difference.
differences lie. Has blocking been effective in this d. Use an appropriate method to identify the pairwise
experiment? Present your results in the form of a differences, if any, in the average mileages for the
report. three formulations.
11.35 The partially completed ANOVA table for a 11.37 Water Resistance in Textiles An
randomized block design is presented here: EX1137 experiment was conducted to compare the
Source df SS MS F effects of four different chemicals, A, B, C, and D, in
Treatments 4 14.2 producing water resistance in textiles. A strip of mate-
Blocks 18.9 rial, randomly selected from a bolt, was cut into four
Error 24 pieces, and the four pieces were randomly assigned to
Total 34 41.9 receive one of the four chemicals, A, B, C, or D. This
process was replicated three times, thus producing a
a. How many blocks are involved in the design? randomized block design. The design, with moisture-
b. How many observations are in each treatment resistance measurements, is as shown in the figure
total? (low readings indicate low moisture penetration).
c. How many observations are in each block total? Analyze the experiment using a method appropriate
d. Fill in the blanks in the ANOVA table. for this randomized block design. Identify the blocks
e. Do the data present sufficient evidence to indicate and treatments, and investigate any possible differ-
differences among the treatment means? Test using ences in treatment means. If any differences exist, use
a ! .05. an appropriate method to specifically identify where
the differences lie. What are the practical implications
f. Do the data present sufficient evidence to indicate
for the chemical producers? Has blocking been effec-
differences among the block means? Test using
tive in this experiment? Present your results in the
a ! .05.
form of a report.

APPLICATIONS Illustration for Exercise 11.37

11.36 Gas Mileage A study was conducted
EX1136 to compare automobile gasoline mileage for Blocks (bolt samples)
three formulations of gasoline. Four automobiles, all 1 2 3
of the same make and model, were used in the experi- C D B
9.9 13.4 12.7
ment, and each formulation was tested in each automo-
A B D
bile. Using each formulation in the same automobile 10.1 12.9 12.9
has the effect of eliminating (blocking out) automo- B A C
11.4 12.2 11.4
bile-to-automobile variability. The data (in miles per D C A
gallon) follow. 12.1 12.3 11.9

Automobile
11.38 Glare in Rearview Mirrors An experiment
Formulation 1 2 3 4 was conducted to compare the glare characteristics
A 25.7 27.0 27.3 26.1 of four types of automobile rearview mirrors. Forty
B 27.2 28.1 27.9 27.7 drivers were randomly selected to participate in the
C 26.1 27.5 26.8 27.8 experiment. Each driver was exposed to the glare
produced by a headlight located 30 feet behind the
a. Do the data provide sufficient evidence to indicate a rear window of the experimental automobile. The
difference in mean mileage per gallon for the three driver then rated the glare produced by the rearview
gasoline formulations? mirror on a scale of 1 (low) to 10 (high). Each of the
b. Is there evidence of a difference in mean mileage four mirrors was tested by each driver; the mirrors
for the four automobiles? were assigned to a driver in random order. An
03758_12_ch11_p425-481.qxd 9/7/11 1:00 PM Page 454

454 ❍ CHAPTER 11 THE ANALYSIS OF VARIANCE

analysis of variance of the data produced this MINITAB output for Exercise 11.39
ANOVA table: Two-way ANOVA: Growth versus Soil Prep, Location

Source df SS MS F Source DF SS MS F P
Soil Prep 2 38.000 19.0000 10.06 0.012
Mirrors 46.98 Location
Error
3
6
61.667
11.333
20.5556
1.8889
10.88 0.008

Drivers 8.42 Total 11 111.000

Error S = 1.374 R-Sq = 89.79% R-Sq(adj) = 81.28%
Total 638.61 Individual 95% CIs For Mean Based on
Pooled StDev
a. Fill in the blanks in the ANOVA table. Soil Prep
1
Mean
12.5
---------+---------+---------+---------+--
(-------*-------)
2 16.0 (-------*-------)
b. Do the data present sufficient evidence to indicate 3 12.0 (-------*-------)
---------+---------+---------+---------+-
differences in the mean glare ratings of the four 12.0 14.0 16.0 18.0

rearview mirrors? Calculate the approximate Individual 95% CIs For Mean Based on
Pooled StDev
p-value and use it to make your decision. Location Mean ------+---------+---------+---------+-----
1 12.0000 (-------*-------)
c. Do the data present sufficient evidence to indicate 2
3
15.0000
16.3333
(-------*-------)
(------*-------)
that the level of glare perceived by the drivers var- 4 10.6667 (-------*------)
-----+---------+---------+---------+-----
ied from driver to driver? Use the p-value approach. 10.0 12.5 15.0 17.5

d. Based on the results of part b, what are the practical

11.40 Digitalis and Calcium Uptake A study
implications of this experiment for the manufactur-
EX1140 was conducted to compare the effects of three lev-
ers of the rearview mirrors?
els of digitalis on the levels of calcium in the heart mus-
11.39 Slash Pine Seedings An experiment cles of dogs. Because general level of calcium uptake
EX1139 was conducted to determine the effects of three varies from one animal to another, the tissue for a heart
methods of soil preparation on the first-year growth of muscle was regarded as a block, and comparisons of the
slash pine seedlings. Four locations (state forest lands) three digitalis levels (treatments) were made within a
were selected, and each location was divided into three given animal. The calcium uptakes for the three levels of
plots. Since it was felt that soil fertility within a loca- digitalis, A, B, and C, were compared based on the heart
tion was more homogeneous than between locations, a muscles of four dogs and the results are given in the
randomized block design was employed using locations table. Use the Excel printout to answer the questions.
as blocks. The methods of soil preparation were A (no Dogs
preparation), B (light fertilization), and C (burning). 1 2 3 4
Each soil preparation was randomly applied to a plot
A C B A
within each location. On each plot, the same number of
1342 1698 1296 1150
seedlings were planted and the average first-year B B A C
growth of the seedlings was recorded on each plot. Use 1608 1387 1029 1579
the MINITAB printout to answer the questions. C A C B
1881 1140 1549 1319
Location
Soil a. How many degrees of freedom are associated
Preparation 1 2 3 4 with SSE?
A 11 13 16 10 b. Do the data present sufficient evidence to indicate a
B 15 17 20 12 difference in the mean uptakes of calcium for the
C 10 15 13 10 three levels of digitalis?
a. Conduct an analysis of variance. Do the data pro- c. Use Tukey’s method of paired comparisons with
vide evidence to indicate a difference in the mean a ! .01 to rank the mean calcium uptakes for the
growths for the three soil preparations? three levels of digitalis.
b. Is there evidence to indicate a difference in mean d. Do the data indicate a difference in the mean
rates of growth for the four locations? uptakes of calcium for the four heart muscles?
c. Use Tukey’s method of paired comparisons to rank e. Use Tukey’s method of paired comparisons with
the mean growths for the three soil preparations. a ! .01 to rank the mean calcium uptakes for the
Use a ! .05. heart muscles of the four dogs used in the
d. Use a 95% confidence interval to estimate the dif- experiment. Are these results of any practical value
ference in mean growths for methods A and B. to the researcher?
03758_12_ch11_p425-481.qxd 9/7/11 1:00 PM Page 455

11.8 THE ANALYSIS OF VARIANCE FOR A RANDOMIZED BLOCK DESIGN ❍ 455

f. Give the standard error of the difference between the 21st Fireman’s State
mean calcium uptakes for two levels of digitalis. Location Century Geico AAA Fund Farm
g. Find a 95% confidence interval for the difference in West Hollywood 3922 4073 3663 4075 3876
mean responses between treatments A and B. Laguna Beach 2378 2512 2478 3056 2508
Redlands 2560 2476 2549 2756 2614
MS Excel output for Exercise 11.40 Riverside 2584 2759 2494 2940 2714
Source: www.insurance.ca.gov
Anova: Two-Factor Without Replication
a. What type of design was used in collecting these
SUMMARY Count Sum Average Variance data?
A 4 4661 1165.25 16891.583
B 4 5610 1402.5 20261.667 b. Is there sufficient evidence to indicate that insur-
C 4 6707 1676.75 72681.583 ance premiums for the same type of coverage dif-
1 3 4831 1610.333 72634.333
2 3 4225 1408.333 78182.333
fers from company to company?
3 3 3874 1291.333 67616.333 c. Is there sufficient evidence to indicate that insur-
4 3 4048 1349.333 46700.333
ANOVA
ance premiums vary from location to location?
Source of Variation SS df MS F P-value F crit
d. Use Tukey’s procedure to determine which insur-
Digitalis 524177.167 2 262088.583 258.237 0.000 5.143 ance companies listed here differ from others in
Dogs 173415 3 57805 56.955 0.000 4.757 the premiums they charge for this typical client.
Error 6089.5 6 1014.917
Total 703681.667 11
Use a ! .05.
e. Summarize your findings.
11.43 Where to Shop? Do you shop at the
11.41 Bidding on Construction Jobs A EX1143 grocery store closest to home or do you look for
EX1141 building contractor employs three construction the store that has the best prices? We compared the
engineers, A, B, and C, to estimate and bid on jobs. regular prices at four different grocery stores for eight
To determine whether one tends to be a more conserva- items purchased on the same day.
tive (or liberal) estimator than the others, the contractor Stores
selects four projected construction jobs and has each Stater
estimator independently estimate the cost (in dollars per Items Vons Ralphs Bros WinCo
square foot) of each job. The data are shown in the table:
Salad mix, 12 oz. bag 3.99 2.79 1.99 1.78
Construction Job Hillshire Farm® Beef Smoked 4.29 4.29 3.99 2.50
Sausage, 14 oz.
Estimator 1 2 3 4 Total
Kellogg’s Raisin Bran®, 25.5 oz. 4.49 5.49 4.49 3.15
A 35.10 34.50 29.25 31.60 130.45 Kraft® Philadelphia® 2.99 3.19 2.79 1.48
B 37.45 34.60 33.10 34.40 139.55 Cream Cheese, 8 oz.
C 36.30 35.10 32.45 32.90 136.75 Kraft® Ranch Dressing, 16 oz. 3.19 3.49 3.49 1.48
Total 108.85 104.20 94.80 98.90 406.75 Treetop® Apple Juice, 64 oz. 2.99 3.49 3.49 1.58
Dial® Bar Soap, Gold, 8–4 oz. 5.99 6.49 5.79 5.14
Analyze the experiment using the appropriate methods. Jif® Peanut Butter, Creamy, 28 oz. 5.15 5.49 4.79 4.34
Identify the blocks and treatments, and investigate any a. What are the blocks and treatments in this experiment?
possible differences in treatment means. If any differ- b. Do the data provide evidence to indicate that there
ences exist, use an appropriate method to specifically are significant differences in prices from store to
identify where the differences lie. Has blocking been store? Support your answer statistically using the
effective in this experiment? What are the practical ANOVA printout that follows.
implications of the experiment? Present your results in
c. Are there significant differences from block to
the form of a report.
block? Was blocking effective?
11.42 Premium Equity? The cost of auto Two-way ANOVA: Price versus Item, Store
EX1142 insurance varies by coverage, location, and the Source DF SS MS F P
driving record of the driver. The following are esti- Item 7 40.2184 5.74548 29.99 0.000
mates of the annual cost for standard coverage as of Store 3 14.6695 4.88982 25.53 0.000
January 1, 2011 for a male driver with 6–8 years of Error 21 4.0230 0.19157
Total 31 58.9108
experience, driving a Honda Accord with no accidents
or violations.3 (These are quotes and not premiums.) S = 0.4377 R-Sq = 93.17% R-Sq(adj) = 89.92%
03758_12_ch11_p425-481.qxd 9/7/11 1:00 PM Page 456

456 ❍ CHAPTER 11 THE ANALYSIS OF VARIANCE

11.44 Where to Shop?, continued Refer to a. What is the appropriate value of q.05(k, df ) for test-
Exercise 11.43. The printout that follows provides the ing for differences among stores?
average costs of the selected items for the k ! 4 stores. MSE
Store Mean "#
b. What is the value of v ! q.05(k, df ) ""?
b
Vons 4.1350
Ralphs 4.3400
c. Use Tukey’s pairwise comparison test among stores
Stater 3.8525 used to determine which stores differ significantly
WinCo 2.5975 in average prices of the selected items.

THE a ! b FACTORIAL EXPERIMENT:

A TWO-WAY CLASSIFICATION
11.9
Suppose the manager of a manufacturing plant suspects that the output (in number of
units produced per shift) of a production line depends on two factors:
• Which of two supervisors is in charge of the line
• Which of three shifts—day, swing, or night—is being measured

That is, the manager is interested in two factors: “supervisor” at two levels and “shift”
at three levels. Can you use a randomized block design, designating one of the two
factors as a block factor? In order to do this, you would need to assume that the ef-
fect of the two supervisors is the same, regardless of which shift you are considering.
This may not be the case; maybe the first supervisor is most effective in the morning,
and the second is more effective at night. You cannot generalize and say that one su-
pervisor is better than the other or that the output of one particular shift is best. You
need to investigate not only the average output for the two supervisors and the aver-
age output for the three shifts, but also the interaction or relationship between the
two factors. Consider two different examples that show the effect of interaction on
the responses in this situation.

EXAMPLE 11.11 Suppose that the two supervisors are each observed on three randomly selected days
for each of the three different shifts. The average outputs for the three shifts are shown
in Table 11.4 for each of the supervisors. Look at the relationship between the two
factors in the line chart for these means, shown in Figure 11.10. Notice that supervi-
sor 2 always produces a higher output, regardless of the shift. The two factors behave
independently; that is, the output is always about 100 units higher for supervisor 2,
no matter which shift you look at.
Now consider another set of data for the same situation, shown in Table 11.5. There
is a definite difference in the results, depending on which shift you look at, and the
interaction can be seen in the crossed lines of the chart in Figure 11.11.

TABLE 11.4
● Average Outputs for Two Supervisors on Three Shifts
Shift
Supervisor Day Swing Night
1 487 498 550
2 602 602 637