Case Notes 1

ANSWERS TO CASES

Chapter 1
DiGiorno Pizza: Introducing a Frozen Pizza to Compete with Carry-Out

In conducting research for the launching of a new product it is imperative that the target
population be identified. In this case, who are the people most likely to be interested in
purchasing and consuming frozen pizzas in lieu of carry-out pizzas? How are these
people to be identified for sampling (Chapter 7 refers to this group as the frame)?
Should a test market city or area be used? Why or why not? What mode of survey such
as telephone, mail, or personal interview should be used? When should these people be
surveyed? Does time of day, day of the week, or season of the year make any difference?
What types of measurements should be taken? Some possible measurements might
include dollar amounts spent per week of pizza per family, number of pizzas purchased
per month, percentage of family pizza consumption that is frozen pizza, and total amount
spent per month on take out food.

1. One population that was identified was pizza lovers. These people may have been
previously identified by market researchers based on number of pizzas purchased per
month, use of coupons, or from previous surveys. Another population mentioned in the
case was women ages 25 to 54. The advertisements shown on national TV were likely
aimed at the general population because Kraft was attempting to achieve broader goals
such as brand name recognition and a dissemination of the fresh-baked taste message.

In each of the research efforts, the market research company selected only a sample of the
population. SMI-Alcott sent out 1,000 surveys to pizza lovers, the Loran Marketing
Group conducted focus groups (which usually have no more than 15 people per group)
with women 25 to 54, and Product Dynamics used focus groups to conduct blind taste
tests.

The market research companies (SMI-Alcott, the Loran Marketing Group, and Product
Dynamics), took various measurements on sample members and from these
measurements likely computed statistics. Some of these measurements may have
included the ranking of various frozen pizza brands based on taste or status, numerical
ratings of various types of pizza in terms of taste (perhaps, for example, on a scale from
1 to 7), amount of time sample members are willing to spend cooking a pizza, amount of
money spent per month on pizzas, and percentage of sample who recognize the DiGiorno
name. Using these measurements, sample statistics such as average amount of money
spent per
 Case Notes 2

month on pizzas per family or proportion of the sample that recognized the DiGiorno
name can be computed. From these sample statistics, population parameters can be
estimated such as the percentage of all adults in the country who recognize the DiGiorno
name; and the average amount a family spends on take-out pizza per month? This is a
good place to introduce the estimation concepts of Chapter 8 intuitively. One can discuss
point estimates (sample statistics) and the notion of sampling error.

Kraft likely used known descriptive market statistics in their product decision making
such as total annual amount of dollars spent in the U.S. on frozen pizza; population
demographics of the U.S. including age, number and size of households, average
household income; and number of competitors in the frozen pizza market.

2. a. number of pizzas per week ratio level

b. age of purchaser ratio level
c. zip code nominal level
d. dollars spent per month ratio level
e. time between purchases ratio level
f. rating of taste ordinal level (but some
researchers treat as interval)
g. ranking of four brands ordinal level
h. geographic location nominal level
i. quality rating ordinal level
j. identification number nominal level
k. gender nominal level
 Case Notes 3

Chapter 2
Soap Companies Do Battle

The pie chart is useful in displaying the market shares in one device adjacent to each
other. Many decision makers are used to viewing pie charts in connection which budgets
and therefore might feel more at ease with a pie chart. On the other hand, as mentioned
in the text, when percentages are close such as with Dial and Others in 1983, it can be
difficult to discern the difference using the pie chart slices. In this case, the bar chart
shown above is more desirable.

1. Shown below are pie charts for the 1983, the 1991, and the 1999 market shares.
 Case Notes 4

An examination of the pie charts from 1983 through 1999 reveals that the slice
sizes of Unilever have grown and the sizes of the Procter & Gamble slices have shrunk.
Shown below are the actual percentage figures for the three time periods so that you have
the option of displaying the data in different ways:

Company 1983 Share 1991 Share 1999 Share

Procter & Gamble 37.1 30.5 28.4

Unilever 24.0 31.5 38.5
Dial 15.0 19.0 14.8
Colgate-Palmolive 6.5 8.0 9.3
Others 17.4 11.0 9.1

2. Shown below is a histogram of the weekly sales of bars of soaps over the year.
The histogram was constructed using 10 classes. In MINITAB, the student has the option
of trying several different values for the number of intervals. In Excel, students can
explore various bin options. The shape of the histogram will somewhat change according
to the number of class intervals. Note the shape of this histogram is mound shaped with
some skewness to the right. The center of the distribution appears to be near to 20
million as would be expected since Procter & Gamble sells about 20 million bars per
week. Note, however that some weeks actually average as much as 39 million bars per
week and others only 12 million bars. What inventory, production, and human resource
implications might this have? How does a company cope with such fluctuations?
 Case Notes 5

The stem and leaf plot for these data is shown below. The advantage of the stem and leaf
over histograms, pie charts, bar charts, and others is that the stem and leaf retains the
original data in case the researcher wants to calculate other statistics on the numbers.
Production people would likely find the histogram the most interesting because it
displays to them where the bulk of production occurs and the magnitude of the unusual
size runs.

Stem Leaf
12 2
13 6
14 7
15 0, 4, 4, 5
16 8
17 0, 1, 1, 4, 5
18 2, 3, 4, 5, 7
19 1, 3, 6, 9
20 0, 3, 3, 4, 4, 6, 7, 9
21 3, 4, 4
22 5, 8
23 1, 4, 8, 9
24 0, 3
25 2, 2
26 2, 2, 3, 3, 6, 9
27
28
29
30 6
31
32 8
 Case Notes 6

33
34
35
36
37
38
39 8
 Case Notes 7

Chapter 3
Coca-Cola Goes Small in Russia

Shown below is MINITAB output describing the sample of 50 bottle fills. Note that the
mean fill is 200.12 oz. with a standard deviation of 0.42. The minimum fill is 199.20 oz.
and the maximum is 201.10 oz. The median fill is 200.15 oz. The measure of skewness
(.0367) demonstrates a very slight positive skewness. However the histogram and the
high p-value associated with the normality test indicate that the data are approximately
normally distributed. The mean fill of 200.12 oz. indicates that, on average, the sample
fills are very near to 12 oz. and are, if anything, giving a slight amount of free product
away to the consumer. Under the empirical rule, using = 200.12 and = .42, 95% of
the fill should be within 200.12 + .42 or between 199.70 and 200.54 oz.

Descriptive Statistics: Bottle Fills

Variable N Mean Median TrMean StDev SE Mean

Bottle F 50 200.12 200.15 200.12 0.42 0.06

Variable Minimum Maximum Q1 Q3

Bottle F 199.20 201.10 199.80 200.40

Descriptive Statistics
Variable: Bottle Fills

Anderson-Darling Normality Test

A-Squared: 0.217
P-Value: 0.833

Mean 200.118
StDev 0.419
Variance 0.175384
Skewness 3.67E-02
Kurtosis -1.4E-01
N 50
199.2 199.6 200.0 200.4 200.8 201.2
Minimum 199.200
1st Quartile 199.800
Median 200.150
3rd Quartile 200.400
95% Confidence Interval for Mu Maximum 201.100
95% Confidence Interval for Mu
199.999 200.237
199.95 200.05 200.15 200.25 95% Confidence Interval for Sigma
0.350 0.522
95% Confidence Interval for Median
95% Confidence Interval for Median
199.967 200.233
 Case Notes 8

2. The bottles have a label that claims there are 20 oz. of fluid therein. This sample of
150 bottles has an average of 20.003 oz. with a median of 20.005 and a mode of 20.004.
All three of these statistics indicate that, if anything, the bottles are slightly overfilled
overall. The standard deviation of fills is .027 oz. The MINITAB normality statistics
and histogram overlaid with the normal curve indicate that the data are approximately
normally distributed. We can apply the empirical rule. Approximately 68% of the fills
are within 20.003 + 1(.027), from 19.976 to 20.03 oz.; and 95% within 20.003 + 2(.027),
from 19.949 to 20.057.

The measure of skewness (-.085) indicates a slight negative skewness. The box plot
indicates that there is an outlier on both the lower and the upper end (extreme underfilled
bottle and extreme overfilled bottle). An examination of the minimum value shows that
there is a bottle with 19.92 oz. which has a Z score of 3.07 or over three standard
deviations below the mean. The maximum value is 20.09 which has a Z score of 3.22 or
almost three-and-a-quarter standard deviations above the mean. Production and quality
management people can better interpret these extreme fills in light of company goals and
specifications. Overall, the fills are averaging more than 20 oz., the fills are
approximately normally distributed, and the standard deviation of fills is about .03 of an
ounce.
 Case Notes 9

Chapter 4
Colgate-Palmolive Makes a Total Effort

1. Two probabilities are given in this case. The first is the marginal probability that a
U.S. household had purchased Total for the first time:

P(T1) = .21

The second is the conditional probability that a household purchased Total (for a second
time) given that it had purchased it before:

P(T2 T1) = .43

The percentage of U.S. household that purchased Total at least twice can be
computed as:

P(T1 T2) = P(T1) P(T2T1) = (.21)(.43) = .0903

A little over nine percent (9.03%) of U.S. households purchased Total at least
twice during this time period.

2. If age is independent of the willingness to try Total, then the marginal

probability of an age category should equal the conditional probability that someone is in
that age category given that they are willing to try Total. Here, for example, the
probability that someone is in the 45-64 age category is:

P(45-64) = .20

The conditional probability that someone is in the 45-64 age category given their
willingness to try Total is:

P(45-64T) = .24

If age is independent of willingness to try Total, then the following must be true:

P(45-64) = P(45-64T)

but .20 .24.

Therefore, age is not independent of the willingness to try Total.

 Case Notes 10

3. The probability that a person is either in the 45-64 age category or purchased Total is a
union probability and is computed as:

P(45-64 T) = P(45-64) + P(T) - P(45-64 T) =

.20 + .21 - .0903 = .3197

The probability that a person purchased Total given that they are in the 45-64 age
category can be computed as:

P (T 45 64) .0903
P(T45-64) = = = .4515
P ( 45 64) .20

4. Let S = saw the commercial, N = didnt see the commercial, Pi = prior probability

P(S) = .32 P(N) = .68

P(TS) = .40 P(TNS) = .1206

Event Prior P(TPi) P(T Pi) P(Pi T)

S .32 .40 (.32)(.40) = .128 .61

N .68 .1206 (.68)(.1206) = .082 .39

P(T) = .210
 Case Notes 11

Chapter 5
Fuji Film Introduces APS

1. n = 30, p = .40.

The expected number is:

= np = 30(.40) = 12

The probability that six or fewer purchase an APS camera is:

Prob(x < 6 n = 30 and p = .40) =

30 C6 (.40) 6 (.60) 24 +30 C5 (.40)5 (.60) 25 + +30 C0 (.40)0 (.60)30 =

.0115 + .0041 + .0012 + .0003 + .0000 + .0000 + .0000 = .0171

If the market share actually is 40% (p = .40), the expected number of purchasers from a
sample of 30 is 12. Six or fewer are considerably less than the expected number (12).
How often would one get six or fewer out of thirty when twelve is expect less than 2%
(.0171) of the time. Therefore, if six or fewer out of thirty actually purchase, it is likely
that the market share is not 40%.

2. Let = 2.4 complaints/100,000 rolls. Shown below are some of the values for
the Poisson distribution with = 2.4 from MINITAB:

x Probability
0 .0907
1 .2177
2 .2613
3 .2090
4 .1254
5 .0602
6 .0241
7 .0083
8 .0025
9 .0007
10 .0002
11 .0000
12 .0000
 Case Notes 12

With a Poisson distributed average rate of 2.4 complaints/100,000 rolls, the probability of
getting 7 complaints/100, 000 rolls by chance is .0083 or less than 1%. Often researchers
like to use the cumulative probability of x > 7 on a problem like this. The probability of
randomly getting 7 or more complaints/100,000 rolls when the average is only 2.4 is .
0117 (.0083 + .0025 + .0007 + .0002). This is still quite unlikely to happen by chance. If
seven complaints are actually logged, then managers would have to strongly consider the
possibility that the average rate of 2.4 complaints has increased. This is, as has been
stated, unacceptable to management.

3. This is a hypergeometric problem with:

N = 52, n = 10, A = 18 (customer satisfaction), and x = 7

The probability is computed as:

18 C7 34 C3
= .0120
52 C10

The probability that seven out of the ten successful products were created for customer
service when only eighteen of the fifty-two original products were created for customer
service is about 1% (.012). Since this is unlikely to happen by chance, it is likely that
there is something inherently more successful about creating a product for customer
service reasons than for revenue growth.
 Case Notes 13

Chapter 6
Mercedes Goes after Younger Buyers

1. Prob(x > 42,000 = 43,215 and = 2981):

x 42,000 43,215
z= = = -0.41
2981

From Table A.5, the area for z = -0.41 is .1591.

Prob(x > 42,000) = .5000 + .1591 = .6591 = 65.91%

Almost 66% of Mercedes dealers would be priced out of competition with this BMW
model.

Prob (x > 43,215 = 34,990 and = 2367):

x 43,215 34,990
z= = = 3.47
2367

From Table A.5, the area for z = 3.47 is .4997.

Prob(x > 43,215) = .5000 - .4997 = .0003 = 0.03%

Virtually none of the BMW dealers are pricing the 328 ci more than the average
price of the Mercedes CLK320.

Prob(x < 34,990 = 43,215 and = 2981):

x 34,990 43,215
z= = = -2.76
2981

From Table A.5, the area for z = -2.76 is .4971.

Prob(x < 34,990) = .5000 - .4971 = .0029 = 0.29%

About .3% of the Mercedes dealers are pricing CLK320 less than the average price
of the BMW 320 ci.

Prob(x < 37,059 = 43,215 and = 2981):

 Case Notes 14

x 37,059 43,215
z= = = -2.07
2981

From Table A.5, the area for z = -2.07 is .4808.

Prob(X < 37,059) = .5000 - .4808 = .0192 = 1.92%

Less than 2% of the Mercedes dealers price the CLK320 less than $37,059.

Conclusion: There is little overlap in the prices of the two cars and it could be
concluded that they are not really competing with each other pricewise.

2. CLK: a = 24 b = 34 x1 = 26 x2 = 30

30 26
Prob. = = .4 = 40%
34 24

328is: a = 25 b = 35 x1 = 26 x2 = 30

30 26
Prob. = = .4 = 40%
35 25

The same proportion of 328is cars fall into this category (26-30 mpg). However, an
examination of the end points of the mileage distributions of each car reveals that the
upper end for 328s is 1 mpg. higher than for the 328iss and the lower end for CLKs is
1 mpg. lower. Both cars have very close gas mileage figures.

Each car more than 30 mpg.:

For CLK:

34 30
Prob. = = .4 = 40%
34 24

For 328is:

35 30
Prob. = = .5 = 50%
35 25

A higher proportion of 328s are in the more than 30 mpg. category than CLKs.
 Case Notes 15

3. = 1.37 cars/3 hours, = 1/1.37 = .73 of 3 hours = 2.19 hours

For 1 hour:

1 hour = .333 of 3 hours. x0 = 0.333. The cumulative probability of this time interval
is .3663. This means that there is a 36.63% chance that there will be less than one hour
between sales.

For 12 hours:

12 hours = 4 times 3 hours. x0 = 4. The cumulative probability for this time interval is .
9958 meaning that there is a 99.58% chance that there will be less than 12 hours between
sales. The complement of this is that there is a 1 - .9958 = .0042 = 0.42% chance that
there will be more than 12 hours between sales.

Managers know that there is an almost 37% chance of a sale within every hour. They
need to determine how much staffing it takes to sell a car every hour or less. Given that it
takes several potential buyers and often multiple visits to the dealership to sell one car
and that it is relatively likely (probability almost 75%) that they will close a sale every 3
hours (x0 = 1), the dealership should never go without having salespeople around and
may have to have several employees around all the time.

By having good exponential and Poisson distribution data, one can, to some extent, track
the impact of advertising on sales by testing values of using random arrival data in time
periods following advertising to determine if has increased. For example, if = 1.37
every 3 hours but following a advertising campaign, there is a randomly selected 3 hours
period and 5 cars are sold, then management might be able to statistically justify that the
has increased and then conclude that the advertising campaign was the cause.

In many businesses, the value of lambda changes with time of day, day of the week, and
season of the year. In the car business, there may be an increase in sales on the weekend,
in the evening, or perhaps in the fall when new models arrive. Students should always be
cautioned about using the same value of lambda for all time periods. Many students
know intuitively that lambda varies over time.
 Case Notes 16

Chapter 7
Shell Attempts to Return to Premier Status

1. The answers to this question will vary. Students should select one of the four types of
random sampling or a hybrid (eg. area stratified) to use in their sampling plan. The
target population is that group of people that the researcher want to be able to infer to.
For example, if Shell wants to impact all adults in the U.S., then all adults in the U.S.
should be their target population. The frame is the list or roster of this target population
from which the researchers sample. What is the frame of all adults in the U.S.? This is a
difficult question. Many national lists involve some specialty group such as registered
Republicans, Visa card users, Internet users, Catholic church members, etc. which really
do not capture all U.S. adults. Instead of searching for a national frame, the researcher
may want to use some form of area sampling such as selecting a test market city which is
thought to be similar in demographics to the U.S. This might make frame identification
easier because some test market city directory such as the phone book or voter
registration list might be accessible and include most adults.

As an example of two-stage sampling, researchers could randomly select a few test

market cities and then sample every 100th name from the phone book or interview
everyone from a few randomly selected blocks within the test market city.

2. Shell contracted researchers appear to have stratified on age, ethnicity, household

location, occupation, and previous employment with Shell. With regard to opinions
about Shell as a premier company, some strata that might make sense are age,
ethnicity, economic class, education, occupation, gender, and geographic location. It is
important to Shell that all segments of the U.S. adult population be reached. In order to
test to determine the effectiveness of marketing campaigns, Shell needs to recognize that
there will likely be differences in the perceptions of young people and those of old people
of what a premier company should be because of their life experiences and the types of
messages that appeal to them. The same thing is true for different ethic groups (different
cultural values may appeal to different groups), economic class (the economics of the
household may determine what types of premier messages appeal to their needs),
education (at what level of education should the messages be targeted), occupation (how
does Shell impact various occupations differently? eg. environment, reliability of
products, availability of products, pricing), gender (often men and women seek different
outcomes from firms), and geographic location (state and regional cultures vary and thus
consumer messages should be targeted so as to parallel geographic interests).
 Case Notes 17

3. In 1979, p = .12. = .25.

A new survey of n = 350 resulted in p

> .25 n = 350 and p = .12):

Prob.( p

p p .25 .12
z= =
p q (.25)(.75) = 5.62
n 350

From Table A.5, the area for z = 5.62 is .5000.

> .25) = .5000 - .5000 = .0000

Prob.( p

It is virtually impossible to randomly select 350 people and have 25% declare that Shell
is a premier company if in the population only 12% believe that Shell is a premier
company. This is strong statistical evidence that the 12% figure is no longer true and that
the actual population figure is greater. This is a nice segue into section 8.3 of chapter 8 in
which the sample data can be used to estimate the actual proportion who now believe that
Shell is a premier company. It also lays the groundwork for the hypothesis testing to
come in chapter 9.

4. Prob.( x > 2.0 = 1.8, = .7, and n = 35):

x 2.0 1.8
z= =
.7 = 1.69
n 35

From Table A.5, the area for z = 1.69 is .4545

Prob.( x > 2.0) = .5000 - .4545 = .0455

There is only a 4.55% probability that the sample mean of 2.0 was obtained by chance. It
is likely that the population mean is no longer 1.8 and indeed, is now higher. This is a
good place for the instructor to mention .05 as a common standard for low probability
and begin the groundwork for chapter 9.
 Case Notes 18

Prob.( x > 2.5 = 1.8, = .7, and n = 35):

x 2.5 1.8
z= =
.7 = 5.92
n 35

From Table A.5, the area for z = 5.92 is .5000.

Prob.( x > 2.5) = .5000 - .5000 = .0000

It is virtually impossible to randomly obtain a sample mean of 2.5 or more from a sample
of 35 with this standard deviation if the population mean is only 1.8. This is, of course,
even more conclusive evidence than obtaining a sample mean of 2.0. A discussion of
the .0455 probability above and the .0000 value here may result in a preliminary
understanding of the notion of p-value introduced in chapter 9.
 Case Notes 19

Chapter 8
Thermatrix

1. n = 115 For 95% confidence, z = 1.96

p q
Use: p z
n

63
1) Yes: p = = .5475
115

(.5475)(.4522)
.5475 + 1.96 = .5478 + .0910
115

.4568 < p < .6388

86
2) Yes: p = = .7478
115

(.7478)(.2522)
.7478 + 1.96 = .7478 + .0794
115

.6684 < p < .8272

101
3) Yes: p = = .8783
115

(.8783)(.1217)
.8783 + 1.96 = .8783 + .0598
115

.8185 < p < .9381

 Case Notes 20

105
4) Yes: p = = .9130
115

(.9130)(.0870)
.9130 + 1.96 = .9130 + .0515
115

.8615 < p < .9645

2. n = 9 df = 8 For 95% confidence, t.025,8 = 2.306

s
Use: x t
n

1) x = 3.79 s = .86

(.86)
3.79 + 2.306 = 3.79 + .66
9

3.13 < < 4.45

2) x = 2.74 s = 1.27

(1.27 )
2.74 + 2.306 = 2.74 + .98
9

1.76 < < 3.72

3) x = 4.18 s = .63

(.63)
4.18 + 2.306 = 4.18 + .48
9

3.70 < < 4.66

 Case Notes 21

4) x = 3.34 s = .81

(.81)
3.34 + 2.306 = 3.34 + .62
9

2.72 < < 3.96

5) x = 3.95 s = .21

(.21)
3.95 + 2.306 = 3.95 + .16
9

3.79 < < 4.11

 Case Notes 22

Chapter 9
Frito-Lay Targets the Hispanic Market

1. a) H0: p = .63
Ha: p .63

Let = .05 For a two-tailed test, /2 = .025

z.025 = + 1.96 (critical value)

n = 850 x = 575

575
p = = .6765
850

p p .6765 .63
z= =
pq (.63)(.37) = 2.81
n 850

Since the observed z = 2.81 > z.025 = 1.96, the decision is to reject the null
hypothesis. The proportion of Hispanics that are Mexican Americans is not
.63. The sample data indicate that the proportion is higher.

b) H0: p = .93
Ha: p < .93

Let = .05 z.05 = - 1.645 (critical value)

n = 689 x = 606

606
p = = .8795
689

p p .8795 .93
z= =
pq (.93)(.07) = -5.20
n 689

Since the observed z = -5.20 < z.05 = -1.645, the decision is to reject the null
hypothesis. The proportion of Hispanic grocery shoppers that are women is
less than .93.
 Case Notes 23

c) H0: p = .83
Ha: p .83

Let = .05 For a two-tailed test, /2 = .025

n = 438

p = .792237

Since the p-value of .042 < = .05 (the p-value was adjusted by MINITAB
for the two-tailed test), the decision is to reject the null hypothesis. The
proportion of Hispanics who listen to advertisements in Spanish is not .83.
The sample data indicate that the proportion may now be less than .83.

2. a) H0: = 31
Ha: 31 = .01

x = 28.81

The observed t = -1.52

For a two-tailed test, t23,.005 = -2.07

Since the observed t = -1.52 > t23,.005 = -2.07, the decision is to fail to
reject the null hypothesis. There is not enough evidence to say that the
average age is different from 31 years. Marketing decision makers must assume
that the average age has not changed.

b) H0: = 45
Ha: < 45 = .05 n = 18 df = 17 t.05, 17 = -1.740

x = 35.67 s = 19.26

x 35.67 45
t= =
s 19.26 = -2.06
n 18

Since the observed t = -2.06 < t.05,17 = -1.740, the decision is to reject the null
hypothesis. We conclude that the average expenditure of Hispanic customers
on chips per year is less than $ 45.
 Case Notes 24

Chapter 10
Seitz Corporation Producing Quality Gear-Driven and Linear-Motion Products

1. Comparing last years mean transactions to this years:

n1 = 20 x 1 = 2300 s1 = 500
n2 = 25 x 2 = 2450 s2 = 540

H0: 1 - 2 = 0
Ha: 1 - 2 0

df = n1 + n2 - 2 = 20 + 25 - 2 = 43 = .05 /2 = .025

t.025,43 = + 2.021

( x1 x 2 ) ( 1 2 )
2 2
t = s1 (n1 1) + s 2 (n2 1) 1 1 =
+
n1 + n 2 2 n1 n2

( 2300 2450) (0)

(500) (19) + (540) 2 (24)
2
1 1 = -0.96
+
20 + 25 2 20 25

Because t = -0.96 > t.025,43 = -2.021, the decision is to fail to reject the null
hypothesis. There is not enough evidence here to say that there is any difference
in the average dollar amount of sales between this year and last.
 Case Notes 25

2. Comparison of tractors at two plants using a confidence interval:

n1 = 45 x1 = 18 n2 = 51 x2 = 12

18 12
p1 = = .4000 p 2 = = .2353
45 51

p 1 q1 p 2 q 2
( p 1 p 2 ) z +
n1 n2

(.40)(.60) (.2353)(.7647)
(.40 - .2352) 1.96 + = .1647 + .1845
45 51

-.0198 < p1 - p2 < .3492

The point estimate of the difference in quality of tractors at the two plants is 16.47%.
However, due to the relatively small samples, the error of the interval is 18.45% which is
greater than the point estimate. Combining the error of the interval with the point
estimate results in the confidence interval shown above. Note that zero is in the interval
indicating that there is a possibility that there is no difference in the quality ratings of
tractors produced at the two plants. If this were a hypothesis testing problem, then the
decision would be to fail to reject the null hypothesis based on the confidence intervals
inclusion of zero.
 Case Notes 26

3. 2000 vs. 2001:

t = -1.83 with a p-value of .07. This is not significant at = .05. There is no significant
difference in the mean ratings between 2000 and 2001. This is underscored by the
confidence interval that includes zero. However, if = .10 were used, there would be a
significant difference. Examining the means reveals that the mean score for 2002 was
higher. The sample sizes were 75 for 2000 and 93 for 2001.

4. Comparison of variances for week 1 and week 5:

H0: 12 = 22
Ha: 12 22

Let = .05 /2 = .025 df1 = n1 - 1 = 5 df2 = n2 - 1 = 6

F.025,5,6 = 5.99 F.975,6,5 = 1/5.99 = 0.167

Week 1: n1 = 6 s1 = 1.17

Week 5: n2 = 7 s2 = 1.68

2
s1 (1.17) 2
F= 2
= = 0.485
s2 (1.68) 2

Since the observed value of F = 0.485 is > the left tail critical value of
F = 0.167, the decision is to fail to reject the null hypothesis. The variances of
product being produced these two weeks are not significantly different. Management
would probably like this because this indicative of consistent
production patterns. Wide swings in variance would be of concern because it would
indicate that some weeks the variability is more out-of-control than others and a less
consistent product is being produced.
 Case Notes 27

Chapter 11
J. R. Clarkson Company

1. The two by three factorial design is analyzed using a two-way ANOVA. There
are two independent variables, temperature and supplier. Temperature has three
treatment levels: 70o, 110o, and 150o. Supplier has two classifications levels:
supplier A and supplier B. The dependent variable is strength of the valve as
measured in psi. Shown below is Excel output for this analysis.

ANOVA
Source of Variation SS df MS F P-value F crit
Supplier 20.056 1 20.056 1.11 0.313385 4.75
Temperature 800.111 2 400.056 22.09 0.000095 3.89
Interaction 84.778 2 42.389 2.34 0.138598 3.89
Within 217.333 12 18.111

Total 1122.278 17

First, we examine the observed F for interaction which is 2.34 with a p-value of
.1386. Since interaction is not significant at any commonly used alpha, we proceed
to examine main effects. There is no significant difference between the two suppliers
(F = 1.11, p-value = .31339). There is a significant difference in the strength of the
valves by temperature at = .0001. The mean psi for 70o is 159.83, for 110o is 158.33,
and for 150o is 145 psi. It appears that at 150o the valves are not as strong.
Using MINITAB, Tukeys multiple comparison tests were done to determine
if there were any significant differences in valve strength by temperature. The results
are:
Tukey's pairwise comparisons

Family error rate = 0.0500

Individual error rate = 0.0203

Critical value = 3.67

Intervals for (column level mean) - (row level mean)

1 2

2 -5.444
8.444

3 7.890 6.390
21.777 20.277

These results show that there were significant differences between 70o and 150o
and between 70o and 150o. The confirms what we observed above with the
Excel output.
 Case Notes 28

2. The data are analyzed using a one-way ANOVA. The independent variable is country
with four classifications: Canada, Columbia, Taiwan, and the U.S. The dependent
variable is the dollar cost to replace a seal. Shown below is the MINITAB output for this
one-way analysis with multiple comparisons:

One-way ANOVA: Cost versus Country

Analysis of Variance for Cost

Source DF SS MS F P
Country 3 64331 21444 32.68 0.000
Error 24 15750 656
Total 27 80081
Individual 95% CIs For Mean
Based on Pooled StDev
Level N Mean StDev -----+---------+---------+---------+-
1 7 244.29 31.01 (---*---)
2 7 323.57 30.65 (---*---)
3 7 195.00 20.82 (---*---)
4 7 222.14 17.04 (---*---)
-----+---------+---------+---------+-
Pooled StDev = 25.62 200 250 300 350

Tukey's pairwise comparisons

Family error rate = 0.0500

Individual error rate = 0.0110

Critical value = 3.90

Intervals for (column level mean) - (row level mean)

1 2 3

2 -117.0
-41.5

3 11.5 90.8
87.0 166.3

4 -15.6 63.7 -64.9

59.9 139.2 10.6

The results show that there is a significant difference in the cost of seal replacement
between countries (F = 32.68, p-value = .000). An examination of the means shows
that there is potential significant differences between countries (Canada - $244.29,
Columbia - $323.57, Taiwan - $195.00, and U.S. - $222.14). The results of the Tukey
multiple comparison analysis shows that there are significant pairwise differences
between Canada and Columbia, between Canada and Taiwan, between Columbia and
Taiwan, and between Columbia and the U.S. Clarkson might be very effective in
marketing to Columbian companies because in Columbia the cost of replacing the
seal is significantly higher than in other countries.
 Case Notes 29

3. This is a randomized block design. The main independent variable of interest

is type of valve. The blocking variable was day of the week. Below is Excel
output for a two-way ANOVA without replication.

ANOVA
Source of Variation SS df MS F P-value F crit
Day of the Week 0.7767 4 0.1942 4.98 0.00595206 2.87
Type of Valve 6.0107 5 1.2021 30.85 0.000000010 2.71
Error 0.7793 20 0.0390

Total 7.5667 29

A highly significant observed F value was obtained for type of valve

(F = 30.85 with a p-value of .000000001). The various valve types
and their associated mean lead times are:
Safety 1.64
Butterfly 2.12
Clack 1.32
Slide 1.66
Poppet 2.18
Needle 0.88
Since there is a significant difference in valve type and since the means appear
to be quite different, Tukeys multiple comparisons were used to determine
which pairs of means, if any, are significantly different. The MINITAB output for
this analysis is given below:
Tukey's pairwise comparisons

Family error rate = 0.0500

Individual error rate = 0.00501

Critical value = 4.37

Intervals for (column level mean) - (row level mean)

1 2 3 4 5

2 -0.9776
0.0176

3 -0.1776 0.3024
0.8176 1.2976

4 -0.5176 -0.0376 -0.8376

0.4776 0.9576 0.1576

5 -1.0376 -0.5576 -1.3576 -1.0176

-0.0424 0.4376 -0.3624 -0.0224

6 0.2624 0.7424 -0.0576 0.2824 0.8024

1.2576 1.7376 0.9376 1.2776 1.7976

The results of these multiple comparison tests is that lead times for the
following pairs of valves are significantly different:
 Case Notes 30

Safety and Poppet, Safety and Needle, Butterfly and Clack, Butterfly and
Needle, Clack and Poppet, Slide and Poppet, Slide and Needle, and Poppet and
Needle.

The study was attempting to control for day of the week as a blocking variable.
The blocking variable produced an F value that was significant at = .01.
Lead times differ by type of valve. The needle and clack valves have the shortest
lead times and the butterfly and poppet valves have the longest lead times. A cursory
examination of the mean lead times by day of the week indicates that Mondays and
Fridays produce the longest lead times.
 Case Notes 31

Chapter 12
Foot Locker in the Shoe Mix

1. Has the distribution of shoe sales by Price Category changed from the year 2000 to
the year 2001? A chi-square goodness-of-fit test can be used to test this. Let the
2000 distribution be the expected frequencies and the 2001 values be the observed
frequencies. Neither MINITAB nor Excel directly computes a chi-square goodness-
of-fit test. Utilizing MINITABs calculator capability (could have used Excels), the
following work is obtained.

2000 2001 chi-sq.

115 126 1.05217 The observed chi-square value is 1.85380

38 40 0.10526
37 35 0.10811 Using an = .05, and df of k-1 = 7, the critical value is:
30 27 0.30000 7,.05 = 14.0671
22 20 0.18182
21 20 0.04762
11 11 0.00000
17 18 0.05882
2 = 1.85380

Since the observed chi-square is less than the critical chi-square, the decision is to fail
to reject the null hypothesis. There is not enough evidence to declare that the 2001
distribution of shoe sales is any different than the 2000 distribution of shoe sales.
Implications to Foot Locker and Nike is that the target market has not changed. This
might mean that marketing attempts to change the target market have not been effective.
It might also mean that the production schedule for various types of shoes need not
change. Demand for shoes at the various price levels remains constant. If the companies
desire to sell more high end shoes, they need to make a renewed effort to effect changes
in these distributions. This result basically tells them that they are where they were.
 Case Notes 32

2. A chi-square test of independence is used to determine if gender is independent of

gender. Shown below is MINITAB output for the analysis of this question.

Chi-Square Test: Male, Female

Expected counts are printed below observed counts

Male Female Total

U.S. West 29 43 72
41.27 30.73

U.S. South 48 20 68
38.98 29.02

U.S. East 52 61 113

64.77 48.23

U.S. North 28 25 53
30.38 22.62

Europe 78 32 110
63.05 46.95

Australia 47 29 76
43.56 32.44

Total 282 210 492

Chi-Sq = 3.647 + 4.898 + 2.090 + 2.806 + 2.517 + 3.380 +

0.186 + 0.250 + 3.545 + 4.761 + 0.272 + 0.365 = 28.716
DF = 5, P-Value = 0.000

An observed chi-square of 28.716 was obtained with an associated p-value of .000 on this
question. This indicates that gender is not independent of location when it comes to the
number of formal suggestions. A cursory examination of the observed values reveals that
more suggestions were submitted by females than males in the U.S. West and the U.S.
East. However, more suggestions were submitted by males than females in all other
regions. In the U.S. South and Europe the ratio of formal suggestions of males to females
was over two to one. What this result says is when it comes to making suggestions, the
regional culture has an impact on who makes the suggestions. Foot Locker might make
a concerted effort to increase participation by the gender with fewer suggestions in each
region and perhaps make an attempt to change corporate culture so that regional
tendencies do not effect the employee suggestion program.

3 According to sources, Foot Locker has a 19.4% share of the sneaker market. Foot
Locker believes that it holds a higher share of the market in the U.S. West. To test this
notion, Foot Locker hires a market research company that randomly samples 1,000
people in the U.S. West who have just purchased new sneakers. Of these, 230 have
purchased their sneakers at Foot Locker. The hypotheses being tested are:

H0: p = .194
Ha: p > .194
 Case Notes 33

230
The sample information is: n = 1,000 x = 230 p = = .230
1000
Let alpha be .01.
Chapter 9 techniques can be used to analyze these data. However, the chi-square
goodness-of-fit can also be used. Using the .194 as the expected proportion and .230 as
the observed proportion, a 2x2 table can be set up and analyzed using the chi-square
goodness-of-fit. The observed frequency is calculated by taking .194 of 1,000 (the
market share that Foot Locker holds nationally).

( fo fe )2
fe fo
fe

Foot Locker 194 230 6.6804

Other Store 806 770 1.6079

The resulting observed chi-square is 8.2884. With 1 degree of freedom, and an = .01,
the critical chi-square value is 6.6349. The decision is to reject the null hypothesis.
This indicates that proportion of the market share in the west is significantly higher
than in other locales. Perhaps Foot Locker ought to study the clientele, the stores, the
sales people, and the sale methods to determine why they are doing better in the West and
attempt to implement things that they learn in other locales.
 Case Notes 34

Chapter 13
Delta Wire Uses Training as a Weapon

1. Shown below is MINITAB output for correlation and regression analysis for the
education and sick day data. The correlation between hours of education and number of
sick days is a relatively strong negative correlation. This indicates that the more hours of
education received, the fewer the sick days. There may be several reasons for this. One
explanation may be that as workers participate in an educational process about their
work, they become more interested in what they are doing because they understand more
about it and can see more potential. If the training is interesting, they may look forward
to the time in the classroom. In addition, they may feel more a part of the team by having
been included in training and feel more self worth for being selected for the training.

Correlation of HrsEduc and SickDays = -0.773

The regression equation is

SickDays = 7.75 - 0.0795 HrsEduc

Predictor Coef Stdev t-ratio p

Constant 7.7455 0.7980 9.71 0.000
HrsEduc -0.07946 0.01536 -5.17 0.000

s = 2.455 R-sq = 59.8% R-sq(adj) = 57.5%

Analysis of Variance

SOURCE DF SS MS F p
Regression 1 161.26 161.26 26.75 0.000
Error 18 108.49 6.03
Total 19 269.75

The F test for the overall model is significant at .001 as is the t value testing the slope.
The r2 of .598 indicates that almost 60% of the variation of the sick days is accounted for
by the hours of education. The standard error of the estimate is 2.455 days which is a
modest error. The regression equation has an intercept of 7.75 which indicates that the
model predicts that a worker will have an average of 7.75 sick days if the worker has
participated in no hours of education. Notice the negative slope. As the number of hours
of education increase, the regression model is deduct days from the intercept resulting in
a prediction of fewer sick days.

Shown below is a regression plot of this line and the data:

 Case Notes 35

2. The F value for the regression model indicates significant overall regression. The
t-ratio, which here is the square root of F, also yields a p-value of .000 indicating
that the population slope is different from zero. The value of r2 = 90.1% is extremely
high denoting a strong predictability in the model. The regression equation has a positive
slope which indicates that the higher the satisfaction scores, the more sales there are.
 Case Notes 36

3. Shown below is Excel regression output for this problem. This output shows that hours of
training is highly predictive of productivity (r2 = .976). The overall F value for this
regression is extremely large and significant (F = 662.28, p-value of .000000).
In addition, the standard error (1005.644) is quite small in relation to the five-figure
productivity numbers that have a range of 20,000. The t test of the slope
is also highly significant (t = 179.65, p-value of .000000) further underscoring
the significance of hours of training as a predictor. Overall, this is a very strong
regression model with much predictability.

SUMMARY OUTPUT

Regression Statistics
Multiple R 0.988
R Square 0.976
Adjusted R Square 0.975
Standard Error 1005.644
Observations 18

ANOVA
df SS MS F Significance F
Regression 1 669781231.5 669781231.5 662.28 1.9027E-14
Residual 16 16181129.59 1011320.599
Total 17 685962361.1

Coefficients Standard Error t Stat P-value

Intercept 70880.252 394.546 179.65 7.138E-28
Hours 5.093 0.198 25.73 1.903E-14
 Case Notes 37

Chapter 14
Starbucks Introduces Debit Card

1. This model uses four independent variables in an effort to predict the amount of money
people spend on their debit card. Overall, the model has modest to good predictability
with an R2 of .755 and a standard error of $22.15. This standard error indicates that
about 95% of the time, the model will be within +2($22.15) or +$44.30 of the actual
figurewhich is not particularly good.. While the overall test of the model is significant
(F = 15.38, p-value = .000007), an examination of the t tests and their associated p-values
shows that only one of the predictors, income (t = 6.69, p-value = .000002) is significant.
None of the other variables are even close. Had a simple regression model been
developed using just income to predict the amount of the prepaid card, the R2 would be .
723, the t value for income would increase to 7.74, the standard error of the estimate
would reduce to $21.96, and the overall F test would increase to 59.90.

SUMMARY OUTPUT

Regression Statistics
Multiple R 0.869
R Square 0.755
Adjusted R Square 0.706
Standard Error 22.1483
Observations 25

ANOVA
df SS MS F Significance F
Regression 4 30175.0423 7543.7606 15.38 0.000007
Residual 20 9810.9577 490.5479
Total 24 39986

Coefficients Standard Error t Stat P-value

Intercept -83.826 22.494 -3.73 0.0013
Age 0.237 0.576 0.41 0.6852
Days 1.190 1.474 0.81 0.4291
Cups 1.422 2.631 0.54 0.5949
Income 2.407 0.360 6.69 0.000002
 Case Notes 38

2. This model attempts to predict the number of days per month that a customer frequents
Starbucks. The predictor variables are age, income, and number of cups of coffee per
day. The Excel results of this analysis are shown below. The model is modest to weak
with an R2 of just .416. The standard error of 3.28 days indicates that the model would
predict within + 2(3.28) or + 6.56 days about 95% of the time. A perusal of the data
shows that the range of number of days is 16 days. The relatively large size of the
standard error to this range is further evidence of the models weakness. A study
of the t statistics reveals that the predictor variable, cups, is the only significant predictor
(t = 3.40, p-value .0027). The number of cups of coffee that a person drinks per day
seems to be a good predictor of the number of times per month the person frequents
Starbucks. Heavy coffee drinkers come often (the coefficient indicates a positive
relationship between cups and frequency). In attempting to increase store traffic,
Starbucks could target their marketing efforts at the more heavy coffee drinkers or
develop and market products that might lure lighter coffee drinkers to their outlets for
different reasons. If a simple regression model is used to predict number of days by
cups of coffee, the R2 is .345 and the standard error is 3.32.
 Case Notes 39

SUMMARY OUTPUT

Regression Statistics
Multiple R 0.645
R Square 0.416
Adjusted R Square 0.332
Standard Error 3.2791
Observations 25

ANOVA
df SS MS F Significance F
Regression 3 160.7593 53.5864 4.98 0.0091
Residual 21 225.8007 10.7524
Total 24 386.56

Coefficients Standard Error t Stat P-value

Intercept 5.9684 3.0651 1.95 0.0650
Cups 1.0644 0.3127 3.40 0.0027
Income 0.0716 0.0509 1.41 0.1742
Age -0.0785 0.0835 -0.94 0.3578
 Case Notes 40

3. Shown below is the output from an Excel multiple regression analysis to predict sales
revenue by number of stores, number of drinks, and average weekly earnings. The
predictability is extremely high with an R2 of .9998. In predicting sales revenues that
range from 400 to 2600, the standard error of the estimate is only 16.69. The overall
F of 4539.21 is significant at alpha = .00001. While number of stores is not a significant
predictor (t = -0.95, p-value = .41145), both number of drinks (t = -7.47, p-value =
.00497) and average weekly earnings (t = 13.70, p-value .00084) are significant at
= .01. Notice that for the predictor, number of drinks, both the t value and the
coefficient are negative. This indicates that, at least in this model with other variables
in the model, there is a negative relationship between number of drinks and sales
revenue. However, a cursory examination of the raw data shows that as sales revenues
increase so do the number of drinks. This points out one of the dangers in over
interpreting the regression coefficients (discussed in Chapter 15 in section on
multicollinearity). When a simple regression model is run using number of drinks as
the sole predictor of sales revenue, the r2 is .929, and more importantly the regression
coefficient is positive as is the t statistic. This might serve as an informal/intuitive
introduction to the notion of collinearity. The correlation between the two
significant predictors in the multiple regression model, number of drinks and weekly
earnings, is .984.

SUMMARY OUTPUT

Regression Statistics
Multiple R 0.9999
R Square 0.9998
Adjusted R Square 0.9996
Standard Error 16.689
Observations 7

ANOVA
df SS MS F Significance F
Regression 3 3792735.879 1264245.293 4539.21 0.000006
Residual 3 835.55 278.52
Total 6 3793571.429

Coefficients Standard Error t Stat P-value

Intercept -13500.237 946.164 -14.27 0.00075
Stores -0.0264 0.0278 -0.95 0.41145
Drinks -75.20448212 10.07005905 -7.47 0.00497
Earnings 38.989 2.846675638 13.70 0.00084
 Case Notes 41

Chapter 15
Virginia Semiconductor

1. Shown below is a multiple regression analysis which contains a model to predict

the size of a companys purchase by four other predictors. Below that is a
stepwise regression analysis of the same data. The full multiple regression model has an
R2 of 77.3%. However, the adjusted R2 is only 69% indicating that there are some
nonsignificant predictors in the model. The stepwise regression analysis confirms this by
showing that an R2 of 77.0% can be obtained with only two predictors, size of company
and whether or not the company has a central purchasing agent. The overall F value is
significant at alpha = .01. The standard error of the estimate is 94.43. An examination of
the p-values of the t ratios to test the slopes confirms the stepwise regression output. In
the full multiple regression model, only size of the company has a significant t value with
a p-value of .01. The variable, central purchasing agent, has a p-value of .081 which is
significant if alpha = .10. In the stepwise regression analysis, size of company was the
strongest single predictor yielding an R2 of 69.44%. The inclusion of central purchasing
agent as step two brings the R2 up to 77%. These are the only two significant predictors
included in the stepwise analysis. Both analyses result in positive regression coefficients
for these two significant predictors. The larger the company, the larger the size of
purchase tends to be which makes sense. Also, having a central purchaser tends to result
in a larger size of purchase. The distance that a customer company is away from Virginia
Semiconductor does not seem to impact size of purchase nor does the percent of a
customer companys imports.

Regression Analysis

The regression equation is:

CustPurc = - 1.8 + 1.37 SizeCo - 0.32 Co%Imp + 0.111 MilesAwy +

110 CentPur?

Predictor Coef Stdev t-ratio p

Constant -1.78 69.22 -0.03 0.980
SizeCo 1.3735 0.4412 3.11 0.010
Co%Imp -0.321 2.065 -0.16 0.879
MilesAwy 0.1110 0.3789 0.29 0.775
CentPur? 110.43 57.45 1.92 0.081

s = 94.43 R-sq = 77.3% R-sq(adj) = 69.0%

 Case Notes 42

Analysis of Variance

SOURCE DF SS MS F p
Regression 4 333650 83413 9.35 0.002
Error 11 98097 8918
Total 15 431748

Stepwise Regression

F-to-Enter: 4.00 F-to-Remove: 4.00

Response is CustPur? on 4 predictors, with N = 16

Step 1 2
Constant 24.316 -7.771

SizeCo 1.82 1.45

T-Ratio 5.64 4.26

CentPur? 109
T-Ratio 2.07

S 97.1 87.4
R-Sq 69.44 77.00

2. The regression analysis attempts to develop a model which can be used to predict sales
figures with the predictors of average hours worked per week and number of customers.
Shown below are scatter plots of each of these predictors with sales. Note the number of
customers has a slight linear shape but is more like the upper left quadrant in Tukeys 4-
quadrant approach. Hours per week produces a graph that is quite scattered. In an effort
to examine each of these predictors together and with squared and interaction terms
included, a full quadratic model was develop. The results follow. Note that none of
the predictors are significant as measured by the t test for slope. Yet the R2 value is over
90%. The stepwise regression analysis that follows reveals that there is significant
interaction between the two predictors. The interaction predictor variable accounts for
over 68% of the variance of sales by itself. In combination with the interaction variable,
the variable, squared number of customers is also significant bringing the total R2 to over
88% with just these two variables. This analysis demonstrates how predictors by
themselves may not be significant by that the interaction of 2 or more predictors might
be.
Case Notes 43
 Case Notes 44

Regression Analysis

The regression equation is

Sales = 41 - 2.22 Hrs/Wk + 0.67 No.Custs + 0.0316 HrsSq - 0.0196 CustSq +
0.0093 IntHrCus

Predictor Coef Stdev t-ratio p

Constant 41.5 117.9 0.35 0.739
Hrs/Wk -2.221 5.099 -0.44 0.681
No.Custs 0.668 3.373 0.20 0.851
HrsSq 0.03163 0.05644 0.56 0.599
CustSq -0.01965 0.02981 -0.66 0.539
IntHrCus 0.00933 0.06959 0.13 0.899

s = 1.097 R-sq = 90.1% R-sq(adj) = 80.3%

Analysis of Variance

SOURCE DF SS MS F p
Regression 5 55.026 11.005 9.15 0.015
Error 5 6.016 1.203
Total 10 61.042

Stepwise Regression

F-to-Enter: 4.00 F-to-Remove: 4.00

Response is Sales on 5 predictors, with N = 11

Step 1 2
Constant 7.089 3.139

IntHrCus 0.0115 0.0247

T-Ratio 4.41 6.27

CustSq -0.0132
T-Ratio -3.71

S 1.46 0.942
R-Sq 68.41 88.38
 Case Notes 45

3. Shown below is a MINITAB scatter plot of sales and number of employees.

Notice how the graph rises and then levels out. This fits fairly closely with the
upper left quadrant of Tukeys 4-quadrant approach. From this, we attempted to
use the log of number of employees as a second predictor. The result is shown below in
a stepwise regression analysis.

Stepwise Regression

F-to-Enter: 4.00 F-to-Remove: 4.00

Response is Sales2 on 2 predictors, with N = 10

Step 1 2
Constant -94.68 -883.10

logEmpl 25.40 220.40

T-Ratio 3.58 4.79

No.Empl. -1.24
T-Ratio -4.26

S 3.89 2.19
R-Sq 61.57 89.29

Notice that log of number employees enters the analysis first and accounts for almost
62% of the variation of sales. At the second step, number employees enters the process
adding another 28%. It was likely worthwhile to recode the data using logs. According
to this, sales increases are associated with the log of number employees and not as much
with the straight linear increase. The company should exercise caution in merely hiring
more people as sales increase.
 Case Notes 46

Chapter 16
DeBourgh Manufacturing Company

1. The decomposition analysis shows several things. A study of the original data
shows that there is a general upward trend. This is underscored by the graph showing the
trend line fit through the data. There appear to be important seasonal effects in these
data. Note that for January, February, and March, the seasonal indices are less than 100.
From April through August, seasonal indices are more than 100 peaking in the summer
months of June, July, and August. This would indicate that these are the strongest
months of locker sales. This might make sense because schools might be purchasing
lockers and installing them in the summertime in preparation for the opening of school.
The months of October, November, and December have low seasonal indexes.
The upward trend marks good news for DeBourgh demonstrating a consistent growth.
The seasonal indices can be used to help DeBourgh plan for both production and
shipping. By examining lead time and production times, DeBourgh can more closely
schedule raw materials from suppliers and do human resource planning.

2. Using MINITAB, several forecasting techniques were explored. Three different

moving averages were used. These were unweighted moving averages of 4 years, 3
years, and 2 years producing MADs of 5.1825, 5.2461, and 4.0717. The 2-year moving
average produced the least MAD and seemed to make the best fit. The MINITAB
graphical output for the 2-year moving average model is shown below.

Next, single exponential smoothing models were examined with various values
of . For = .3, the MAD was 7.3010. For = .6, MAD was 7.0447. For
= .9, MAD was 3.6254. Thus, the higher the value of alpha, the better the forecast with
a big improvement as alpha moved from .6 to .9. Since alpha weights the actual previous
value, the exponential smoothing works better here when the model actually shadows the
previous value. Since there seems to be a downward overall trend in the data, tracking
the previous value may be a good strategy. An optimal search for alpha resulted in an
alpha of .987. The value of MAD for this model was 2.9569. The MINITAB graphical
output for this optimal value of alpha is shown below.

Trend analysis was preformed on these data by fitting a line through the data. MINITAB
trend analysis resulted in MAD of 4.7155. A quadratic fit resulted in MAD = 2.8426.
 Case Notes 47

Overall, the best fitting model to these data was the quadratic trend analysis model. The
model produced by this analysis is 91.6691 - 6.21916 t + 0.266113 t2. Using t = 15 to
represent the year 2002, the predicted per-unit labor cost is: 91.6691 - 6.21916(15) +
0.266113 (15)2 = 58.26. Shown below is the graphical analysis for the quadratic trend
model produced using MINITAB.
DeBourgh can become more productive and produce better quality lockers by
reducing costs (particularly those do to poor work, scrap, rework, poor raw materials,
etc.).

Moving Average

Actual
85
Predicted
Actual
Predicted

75
Cost

Moving Average
65
Length: 2

MAPE: 6.4362
MAD: 4.0717
55 MSD: 51.6762

0 5 10 15
Time
 Case Notes 48

Single Exponential Smoothing

Actual
85
Predicted
Actual
Predicted

75
Labor Cost

Smoothing Constant
65
Alpha: 0.987

MAPE: 4.6176
MAD: 2.9569
55 MSD: 37.2189

0 5 10 15
Time

Trend Analysis for Labor Cost

Quadratic Trend Model
Yt = 91.6691 - 6.21916*t + 0.266113*t**2

Actual
85
Fits
Actual
Fits

75
Labor Cost

65

MAPE: 3.9710
MAD: 2.8426
55 MSD: 16.7767

0 5 10 15
Time
 Case Notes 49

Chapter 17
Schwinn

1. Since two independent samples are being compared and the shape of the population
distribution is unknown, the Mann-Whitney U test is used to analyze the data rather
than the t test for two independent samples. The null hypothesis is that there is no
difference between the age of Schwinn customers in Colorado Springs (near the
mountain) and the age of Schwinn customers in St. Louis. The MINITAB computer
output for the Mann-Whitney test is shown below. The analysis reveals that the median
age for a customer in Colorado Springs was 31 years of age and the median age for a
customer in St. Louis is 14 years. It appears that the target market in Colorado Springs is
young adult versus St. Louis where it is children and early teens. Since one of the sample
sizes is greater than 10, a large sample Mann-Whitney U test is the appropriate
test. As noted in the text, the MINITAB Mann-Whitney test does not yield a z
statistic but rather gives the value of W and a p-value. Since the p-value is .0030,
the null hypothesis is rejected at = .01. There is a significant difference between the
age of Schwinn customers in Colorado Springs and those in St. Louis. The median ages
support the theory that a much older group of customers purchase Schwinn
bikes in Colorado Springs perhaps to be used in mountain biking. The St. Louis
population appears to be a youth market.

Mann-Whitney Test and CI: C.S., St. Louis

C.S. N = 11 Median = 31.00

St. Loui N = 9 Median = 14.00
Point estimate for ETA1-ETA2 is 17.00
95.2 Percent CI for ETA1-ETA2 is (9.00,23.00)
W = 155.0
Test of ETA1 = ETA2 vs ETA1 not = ETA2 is significant at 0.0030
The test is significant at 0.0030 (adjusted for ties)

2. Since three independent groups are being compared, it would appear that this is a
completely randomized design and that a one-way ANOVA could be used to analyze the
data. However, it is uncertain whether the data are normally distributed or not, so a
Kruskal-Wallis test is used to analyze the data. The independent variable is supplier with
three classifications: supplier 1, supplier 2, and supplier 3. The dependent variable is the
weight of the handle bar. The null hypothesis is that there is no difference in the weight
of handle bars supplied by the three suppliers. The alternative hypothesis is that there is a
difference in the weights of handle bars by supplier. MINITAB was used to analyze
these data. The results, given below, show that there is no significant difference in the
weights of handle bars according to supplier. The H value (3.09) is MINITABs
equivalent to the Kruskal-Wallis K value. The associated p-value (.213) denotes that
there is no significant difference even at = .10. The differences in the medians, shown
in the MINITAB output, are merely due to chance. To Schwinn this might mean that, at
least on handle bar weight, these suppliers are interchangeable.
 Case Notes 50

Kruskal-Wallis Test: Weight versus Supplier

Kruskal-Wallis Test on weight

C5 N Median Ave Rank Z

Supplier 1 8 201.9 12.4 1.57
Supplier 2 6 198.2 9.5 -0.26
Supplier 3 5 195.2 6.8 -1.48
Overall 19 10.0

H = 3.09 DF = 2 P = 0.213

3. The MINITAB output for the Runs Test contains a p-value of .9083 that is not
significant (underscored by MINITABs statement Cannot reject at alpha 0.05).
This indicates that the paint flaws are occurring in a random manner. In this sample,
there are 19 observations above K and 56 below (out of 75 bicycles). Since the data
were coded with a 1 to indicate at least one flaw and a 0 to indicate no flaws, there
are 19 bicycles with at least one flaw out of the 75 bicycles. The proportion of flawed
19
bicycles is = .2533 (shown as K in the output). While Schwinn management
75
may
by happy to know that there appears to be no systematic pattern of flaws, they are likely
to be unhappy about having at least one flaw in 25.33% of the bikes. Perhaps, an
intensive, manufacturer-wide quality effort could be made to reduce the percentages of
flaws. After, studying chapter 18 (Statistical Quality Control), the student might be able
to recommend the use of a Pareto Chart to prioritize the types of flaws that are occurring
and a Fishbone diagram to assist managers and production engineers in searching for
causes of the flaws which might include such things as raw materials, machinery,
technique, and workers.
 Case Notes 51

Chapter 18
Robotron

1.

Produce Shipping
Billing
Item
Stop
Beginning
of Assembly
Line
(no)
Being Tag
Manufactured?(yes) Item

Plant
Clerk
(no)
Mail Order Standard (no)
Room Processing Item? (yes) Warehouse Available? (yes)

2. Part 173:

The xbar chart looks pretty good with the exception of sample number 13 which has a
mean that is above the upper control limit. Most of the rest of the means seem to
fluctuate randomly around the centerline mostly in the inner 1/3 of the limit area. The R
chart indicates that the ranges are in control. None of the ranges are outside the control
limits and most lie near the centerline with no obvious pattern occurring.

Part 248:

The x bar chart indicates a process that is in control. None of the points are outside of the
limits. In fact, only two of the spikes are in the outer 2/3 of the limits. There seems to be
little or no pattern with points occurring randomly. However, the range chart indicates
some potential problems. The ranges for samples 9 and 17 are above the upper control
 Case Notes 52

limits. These samples and their origin should be investigated. However, the rest of the
range values seem to be in order and well under control.

3. p chart:

The centerline of the p chart is .03640. On average, 3.64% of the items in each sample
are in nonconformance. Yet, there are 4 of the 100 samples that produced sample
proportions above the upper control limit which is established at .1159 or 11.59%.
Management has to be concerned about samples that have almost 12% nonconforming
items when the overall average is only 3.64%. There are many samples that are on the
lower control limit. However, since the lower values of p for nonconformance indicate
few items in nonconformance, these values are good. In fact, 19 of the 100 samples had
no items nonconforming. Overall, the patterns seem to be random. Managers should
investigate to determine why, seemingly out-of-the-blue, samples contain unacceptable
proportions of nonconforming items.
 Case Notes 53

Chapter 19
Fletcher-Terry: On the Cutting Edge

1. There are several decisions that management had to make during this time including 1)
whether or not to invest in technology, 2) expand its line of offerings through imports, 3)
conduct a significant planning process, 4) attempt to increase market share, 5) develop
new products, 6)create greater employee involvement, 7) invest in employee education,
8) invest in plant improvements, and 8) implement a participatory management system.

Several states of nature occurred which could have effected Fletcher-Terrys outcomes.
Some of these include: 1) its largest customers decided to introduce their own private-
label cutters made overseas, 2) the technology that Fletcher-Terry invested in would not
work, 3) dollar weakened, 4) slow-down in demand for cutters, 5) employees do not
respond to company efforts.

2.

$ Up $ Same $ Down
(.25) (.35) (.40)
Decision Import $350,00 $275,00 -$555,000
Alternative 0 0
Not Import -$22,700 -$22,700 -$22,000

The expected monetary value (EMV) for each of these alternatives is:

Import: (.25)($350,000) + (.35)($350,000) + (.40)(-$555,000) = -$63,000

Not Import: (.25)(-$22,700) + (.35)(-$22,700) + (.40)(-$22,700) = -$22,700

The EMVer would choose the highest of these alternatives which is to not import and
take a $22,700 loss. The risk avoider would also choose to not import. However, a risk
taker might decide to import gambling that the dollar does not go down.
 Case Notes 54

The decision table and the expected values are displayed below in a decision tree.