CHAPTER

Introduction to Data Analy

and Decision Making

USING QUANTITATIVE METHODS TO SOLV

BUSINESS PROBLEMS
s you embark on your study of data analysis and decis
might question the usefulness of quantitative methods
world.” A front-page article in the December 31,1997,e
entitled “Higher Math Delivers Formula for Success” pr
ing evidence of the applicability of the methods you w
heading of the article,“Businesses turn to algorithms t
problems,” says it all.Today’s business problems tend
the past,many managers and executives used a “by th
approach to solve problems—that is,they used their bu
their intuition,and some thoughtful guesswork to obtai
mon sense and intuition go only so far in the solution o
lems businesses now face.This is where data analysis
and the algorithms mentioned in the title of the article
the methods in this book are implemented in user-frien
ware packages and are then applied to complex proble
amazing.Robert Cross,whose company,DFI Aeronomics
based systems to airlines,states it succinctly:“It’s like
and spinning money out of it.”
The power of the methods in this book is that they are
so many problems and environments.The article ment
 “success stories” where quantitative analysis has been applied;others will
throughout this book.
1. United Airlines installed one of DFI’s systems,which cost between
$20 million.United expects the system to add $50 million to $100
to its revenues.
2. The Gap clothing chain uses quantitative analysis to determine e
employees should staff each store during the holiday rush.
3. Quantitative analysis has helped medical researchers test potent
drugs on fewer people with better results.
4. IBM obtained a $93-million contract to build a computer system f
Department of Energy that would do a once-impossible task:mak
models of atomic blasts.It won the contract—and convinced the D
tem was cost effective—only by developing quantitative methods
processing time by half.
5. Hotels,airlines,and television broadcasters all use quantitative an
a new method called “yield management.” In this method,differe
charged to different customers,depending on their willingness to
that more customers are attracted,and revenues increase.
The article concludes by stating that Microsoft’s Excel spreadsheet softwar
mini-optimization program called Solver.This is a key statement.Many of th
enable the successes discussed in the article are very complex mathemati
beyond the grasp of the typical user,including most readers of this book.Ho
no longer need to understand all of the details behind the algorithms.They
know how to model business problems so that appropriate algorithms can
then how to apply them with user-friendly software.For example,we see in
and 15 how to apply Excel’s Solver to a variety of complex problems.You w
intricacies of how Solver does its optimization,but you will learn how to use
productively.The same statement applies to the other methods discussed i
might not understand exactly what is happening in the computer’s “black b
forms its calculations,but you will learn how to become a very effective pro
taking advantage of powerful software.■

1 INTRODUCTION
We are living in the age of technology. This h
entering the business world. First, technology
of data. Retailers collect point-of-sale data on
action occurs; credit agencies have all sorts o
obtain credit; investment companies have a li
terns of stocks, bonds, and other securities; a
nomic trends, the environment, social welfa
everything else we can imagine. It has becom
result, data are plentiful. However, as many org
is quite a challenge to analyze and make s
A second important implication of technolog
power and responsibility to analyze data and
analysis. Those entering the business world can
sis to the “quant jocks,” the technical speciali
crunching. The vast majority of employees now
2 Chapter 1 Introduction to Data Analysis and Decision Making

disposal, they have access to relevant data, and they have been trained in easy-to-use soft-
ware, particularly spreadsheet and database software. For these employees, statistics and
other quantitative methods are no longer forgotten topics they once learned in college.
Quantitative analysis is now an integral part of their daily jobs.
A large amount of data already exists and will only increase in the future. Many com-
panies already complain of swimming in a sea of data. However, enlightened companies
are seeing this expansion as a source of competitive advantage. By using quantitative
methods to uncover the information in the data and then acting on this information—again
guided by quantitative analysis—they are able to gain advantages that their less enlight-
ened competitors are not able to gain. Several pertinent examples of this follow.

■ Direct marketers analyze enormous customer databases to see which customers are
likely to respond to various products and types of promotions. Marketers can then
target different classes of customers in different ways to maximize profits—and give
their customers what the customers want.
■ Hotels and airlines also analyze enormous customer databases to see what their cus-
tomers want and are willing to pay for. By doing this, they have been able to devise
very clever pricing strategies, where not everyone pays the same price for the same
accommodations. For example, a business traveler typically makes a plane reserva-
tion closer to the time of travel than a vacationer. The airlines know this. Therefore,
they reserve seats for these business travelers and charge them a higher price (for the
same seats). The airlines profit, and the customers are happy.
■ Financial planning services have a virtually unlimited supply of data about security
prices, and they have customers with widely differing preferences for various types
of investments. Trying to find a match of investments to customers is a very chal-
lenging problem. However, customers can easily take their business elsewhere if
good decisions are not made on their behalf. Therefore, financial planners are under
extreme competitive pressure to analyze masses of data so that they can make
informed decisions for their customers.
■ We all know about the pressures U.S. manufacturing companies have faced from foreign
competition in the past couple of decades. The automobile companies, for example,
have had to change the way they produce and market automobiles to stay in business.
They have had to improve quality and cut costs by orders of magnitude. Although the
struggle continues, much of the success they have had can be attributed to data analysis
and wise decision making. Starting on the shop floor and moving up through the organi-
zation, these companies now measure almost everything they do, analyze these measure-
ments, and then act on the information from these measurements.
 We talk about companies analyzing data and making decisions. However, companies don’t
really do this; people do it. And who will these people be in the future? They will be you! We
know from experience that students in all areas of business, at both the undergraduate and
graduate level, will soon be required to describe large complex data sets, run regression
analyses, make quantitative forecasts, create optimization models, and run simulations. You
are the person who will soon be analyzing data and making important decisions to help gain
your company a competitive advantage. And if you are not willing or able to do so, there will
be plenty of other technically trained people who will be more than happy to replace you.
Our goal in this book is to teach you how to use a variety of quantitative methods to
analyze data and make decisions. We plan to do so in a very hands-on way. We discuss a
number of quantitative methods and illustrate their use in a large variety of realistic busi-
ness problems. As you will see, this book includes many examples from finance, market-
ing, operations, accounting, and other areas of business. To analyze these examples, we
take advantage of the Microsoft Excel spreadsheet package, together with a number of

powerful Excel add-ins. In each example we will provide step-by-step details of the
method and its implementation in Excel.
This is not a “theory” book. It is also not a book where you can lean comfortably back
in your chair, prop your legs up on a table, and read about how other people use quantita-
tive methods. It is a “get your hands dirty” book, where you will learn best by actively fol-
lowing the examples throughout the book at your own PC. In short, you will learn by
doing. By the time you have finished, you will have acquired some very useful skills for
today’s business world.

1 AN OVERVIEW OF THE BOOK

This book is packed with quantitative method
covered in any single course. Therefore, we
chapter brief so that you can get on with the an
the methods you will be learning and the tools
an overview of the methods covered in this boo
them. Then in the next section we preview so
detail in later chapters. Finally, we present a b
process. Our primary purpose at this point is to

1.2.1 The Methods

 This book is rather unique in that it combines topics from two separate fields: statistics and
management science. In a nutshell, statistics is the study of data analysis, whereas man-
agement science is the study of model building, optimization, and decision making. In the
academic arena these two fields traditionally have been separated, sometimes widely.
Indeed, they are often housed in separate academic departments. However, from a user’s
standpoint it makes little sense to separate them. Both are useful in accomplishing what the
title of this book promises: data analysis and decision making.
Therefore, we do not distinguish between the “statistics” and “management science”
parts of this book. Instead, we view the entire book as a collection of useful quantitative
methods that can be used to analyze data and help make business decisions. In addition,
our choice of software helps to integrate the various topics. By using a single package,
Excel, together with a number of add-ins, we see that the methods of statistics and man-
agement science are similar in many important respects. Most importantly, their combina-
tion gives us the power and flexibility to solve a wide range of business problems.
Three important themes run through this book. Two of them are in the title: data analysis
and decision making. The third is dealing with uncertainty.1 Each of these themes has sub-
themes. Data analysis includes data description, data inference, and the search for
relationships in data. Decision making includes optimization techniques for problems with
no uncertainty, decision analysis for problems with uncertainty, and structured sensitivity
analysis. Dealing with uncertainty includes measuringuncertainty and modelinguncertainty
explicitly into the analysis. There are obvious overlaps between these themes and subthemes.
When we make inferences from data and search for relationships in data, we must deal with
uncertainty. When we use decision trees to help make decisions, we must deal with uncer-
tainty. When we use simulation models to help make decisions, we must deal with uncertainty
and we often make inferences from the simulated data.

1The fact that the uncertainty theme did not find its way into the title of this book does not detract from its impor-
tance. We just wanted to keep the title reasonably short!

4 Chapter 1 Introduction to Data Analysis and Decision Making

Figure 1.1 shows where you will find these themes and subthemes in the remaining chap-
ters of this book. In the next few paragraphs we discuss the book’s contents in more detail.

Themes Subthemes
Figure 1.1
Themes and
Subthemes
We begin in Chapters 2 and 3 by illustrating a number of ways to summarize the infor-
mation in data sets. These include graphical and tabular summaries, as well as numerical
summary measures such as means, medians, and standard deviations. The material in these
two chapters is elementary from a mathematical point of view, but it is extremely impor-
tant. As we stated at the beginning of this chapter, organizations are now able to collect
huge amounts of raw data. The question then becomes, What does it all mean? Although
there are very sophisticated methods for analyzing data sets, some of which we cover in
later chapters, the “simple” methods in Chapters 2 and 3 are crucial for obtaining an initial
understanding of the data. Fortunately, Excel and available add-ins now make what was
once a very tedious task quite easy. For example, Excel’s pivot table tool for “slicing and
dicing” data is an analyst’s dream come true. You will be amazed at the complex analysis it
enables you to perform—with almost no effort!
After the analysis in Chapters 2 and 3, we step back for a moment in Chapter 4 to see
how we get the data we need in the first place. We know from experience that many students
and businesspeople are able to perform appropriate statistical analysis once they have the
data in a suitable form. Often the most difficult part, however, is getting the right data, in the
right form, into a software package for analysis. Therefore, in Chapter 4 we present a num-
ber of extremely useful methods for doing this within Excel. Specifically, we discuss meth-
ods for using Excel’s built-in filtering tools to perform queries on Excel data sets, for using
Microsoft Query (part of Microsoft Office) to perform queries on external databases (such
as Access) and bring the resulting data into Excel, for importing data directly into Excel
from Web sites, and for “cleansing” data sets (getting rid of “bad” data values). This chapter
provides tools that many analysts need but are usually not even aware of.
Uncertainty is a key aspect of most business problems. To deal with uncertainty, we need
a basic understanding of probability. We provide this understanding in Chapters 5 and 6.
 Chapter 5 covers basic rules of probability and
cept of probability distributions. Chapter 6 foll
the most important probability distributions, th
briefly discusses the Poisson and exponential d
probability
We have found that one of the best ways to
and easier to understand is by using computer
mon theme that runs through this book, begi
chapters of the book are devoted entirely to si
tion early and often to illustrate
In Chapter 7 we apply our knowledge of pro
tainty. These types of problems—faced by all
terized by the need to make a decision now, e
demand for a product or returns from investme
rial in Chapter 7 provides a rational basis fo
illustrate do not guarantee perfect outcomes—
ently than we had expected—but they do enab
of the given circumstances. Additionally, the s
allows us, with very little extra work, to see
inputs. This is crucial because the inputs to ma
guesses. Finally, we examine the role of risk a
In Chapters 8, 9, and 10 we discuss samplin
problem is to estimate one or more characteris
time consuming to learn about the entire popu
a random sample from the population and the
the characteristics of the population. We see t
the results of various polls. We also see it in m
typically sample only a fraction of a company’
of the entire population of records from the re
company has been following acc
In Chapters 11 and 12 we discuss the extrem
which is used to study relationships between v
its generality. Every part of a business has va
regression can often be used to estimate possib
managerial accounting, regression is used to es
labor hours and production volume. In market
volume depends on advertising and other ma
used to estimate how the return of a stock de
studies, regression is used to estimate how t
assessed valuation of the house and character
square footage. Regression analysis finds per
any method in
From regression, we move to times series an
topic is particularly important for providing i
example, manufacturing companies must fore
sible decisions about quantities to order from
rants must forecast customer arrivals, sometim
so that they can staff their r
There are many approaches to forecasting,
involve regression-based methods, in which o
forecast the variable of interest, whereas othe
extrapolation method the historical patterns

6
6 Chapter 1 Introduction to Data Analysis and Decision Making

demand or customer arrivals, are studied carefully and are then “extrapolated” into the
future to obtain forecasts. A number of extrapolation methods are available. In Chapter 13
we study both regression and extrapolation methods for forecasting.
Chapters 14 and 15 are devoted to spreadsheet optimization, with emphasis on linear
programming. We assume a company must make several decisions, and there are con-
straints that limit the possible decisions. The job of the decision maker is to choose the
decisions such that all of the constraints are satisfied and an objective, such as total profit
or total cost, is optimized. The solution process consists of two steps. First, we build a
spreadsheet model that relates the decision variables to other relevant quantities by means
of logical formulas. In this first step there is no attempt to find the optimal solution; all we
want to do is relate all relevant quantities in a logical way. The second step is then to find
the optimal solution. Fortunately, Excel contains a Solver add-in that performs this step.
All we need to do is specify the objective, the decision variables, and the constraints;
Solver then uses powerful algorithms to find the optimal solution. As with regression, the
power of this approach is its generality. An enormous variety of problems can be solved by
spreadsheet optimization.
Finally, Chapters 16 and 17 illustrate a number of computer simulation models. This is
not our first exposure to simulation—it is used in a number of previous chapters to illustrate
statistical concepts—but here it is studied in its own right. As we discussed previously, most
business problems have some degree of uncertainty. The demand for a product is unknown,
future interest rates are unknown, the delivery lead time from a supplier is unknown, and so
on. Simulation allows us to build this uncertainty explicitly into spreadsheet models.
Essentially, some cells in the model contain random values with given probability distribu-
tions. Every time the spreadsheet recalculates, these random values change, which causes
“bottom-line” output cells to change as well. The trick then is to force the spreadsheet to
recalculate many times and keep track of interesting outputs. In this way we can see which
output values are most likely, and we can see best-case and worst-case results.
Spreadsheet simulations can be performed entirely with Excel’s built-in tools. However,
this can be quite tedious. Therefore, we use a spreadsheet add-in to streamline the process. In
particular, we learn how the @RISK add-in can be used to run replications of a simulation,
keep track of outputs, create useful charts, and perform sensitivity analyses. With the inher-
ent power of spreadsheets and the ease-of-use of such add-ins as @RISK, spreadsheet simu-
lation is becoming one of the most popular quantitative tools in the business world.

1.2.2 The Software

The topics we have just discussed are very important. Together, they can be used to solve a
wide variety of business problems. However, they are not of much practical use unless we
have the software to do the number crunching. Very few business problems are small
enough to be solved with pencil and paper. They require powerful software.
The software included in new copies of this book, together with Microsoft Excel, pro-
vides you with a powerful software combination that you will not use for one course and
then discard. This software is being used—and will continue to be used—by leading com-
panies all over the world to solve large, complex problems. We firmly believe that the
experience you obtain with this software, through working the examples and problems in
this book, will give you a key competitive advantage in the marketplace.
It all begins with Excel. All of the quantitative methods that we discuss are imple-
mented in Excel. We cannot forecast the state of computer software in the long-term
future, but as we are writing this book Excel is the most heavily used spreadsheet package
on the market, and there is every reason to believe that this state will persist for many
years. Most companies use Excel, most employees and most students have been trained in
Excel, and Excel is a very powerful, flexible, and easy-to-use package.

Built-in Excel Features

Virtually everyone in the business world knows the basic features of Excel, but relatively few
know many of its more powerful features. In short, relatively few people are the “power
users” we expect you to become by working through this book. To get you started, the file
Excel Tutorial.doc on the CD-ROM inside new copies of this book explains some of the
“intermediate” features of Excel—features that we expect you to be able to use. These
include the SUMPRODUCT, VLOOKUP, IF, NPV, and COUNTIF functions. They also
include range names, the Data Table command, the Paste Special command, the Goal Seek
command, and a few others. Finally, although we assume you can perform routine spread-
sheet tasks such as copying and pasting, we include a few tips to help you perform these tasks
more efficiently.
In the body of the book we describe several of Excel’s advanced features in more
Although the tutorial is detail. In Chapters 2 and 3 we introduce pivot tables, the Excel tool that enables you to
presented in a Word summarize data sets in an almost endless variety of ways. (Excel has many useful tools, but
file, it contains we personally believe that pivot tables are the most ingenious and powerful of all. We
“embedded” Excel won’t be surprised if you agree.) Beginning in Chapter 5, we introduce Excel’s RAND
spreadsheets that
function for generating random numbers. This function is used in all spreadsheet simula-
allow you to practice
tions (at least those that do not take advantage of an add-in).
spreadsheet techniques
within Word.
Solver Add-in
In Chapters 14 and 15 we make heavy use of Excel’s Solver add-in. This add-in, developed
by Frontline Systems (not Microsoft), uses powerful algorithms—all behind the scenes—
to perform spreadsheet optimization. Before this type of spreadsheet optimization add-in
was available, specialized (nonspreadsheet) software was required to solve optimization
problems. Now we can do it all within a familiar spreadsheet environment.

StatTools Add-in
 Much of this book discusses basic statistical an
decision as we developed the book. A number o
the market, including Minitab, SPSS, SAS, StatG
now user-friendly Windows versions of these p
have found through our own experience that st
ages, regardless of their inherent quality, so we w
(We briefly discuss SPSS and SAS in Chapter
book.) Unfortunately, Excel’s built-in statistic
ToolPak (developed by a third party) that sh
Therefore, we developed an add-in called
StatTools is powerful, easy to use, and capab
interpretable form. We do not believe you sho
to produce some statistical output. This might
but after that it acts as a strong incentive not to
should be able to generate output quickly and
the output, and it also allows you to
A good illustration involves the construction
graphs, discussed in Chapter 2. All of these extre
forward way with Excel’s built-in tools. But by th
“dress up” the charts exactly as you want them, y
process again. StatTools does it all quickly and

2Users of the previous edition of the book will note the change from StatPro to StatTools. Palisade Corporation
has redeveloped StatPro as a commercial package under the name StatTools. The user interface has changed con-
siderably (for the better), but the statistical functionality is virtually the same.

8 Chapter 1 Introduction to Data Analysis and Decision Making

resulting charts, but that’s up to you.) Therefore, if we advise you in a later chapter, say, to look
at several scatterplots as a prelude to a regression analysis, you can do so in a matter of seconds.
SolverTable Add-in
An important theme throughout this book is sensitivity analysis: How do outputs change
when inputs change? Typically these changes are made in spreadsheets with a data table, a
built-in Excel tool. However, data tables don’t work in optimization models, where we
would like to see how the optimal solution changes when certain inputs change. Therefore,
we include an Excel add-in called SolverTable to perform this type of sensitivity analysis.
It works almost exactly like Excel’s data tables, and it is included with this book. In
Chapters 14 and 15 we explain how to use SolverTable.

Decision Tools Suite

In addition to StatTools, SolverTable, and built-in Excel add-ins, we also have included in
this book a slightly scaled-down version of Palisade Corporation’s powerful Decision
Tools suite. Most of the items in this suite are Excel add-ins—so the learning curve isn’t
very steep. There are six separate packages in this suite: @RISK, PrecisionTree, TopRank,
RISKOptimizer, BestFit, and RISKview. The first two are the most important for our pur-
poses, but all are useful for certain tasks.

@RISK
 The simulation add-in @RISK enables us to run as many replications of a spreadsheet sim-
ulation as we like. As the simulation runs, @RISK automatically keeps track of the outputs
we select, and it then displays the results in a number of tabular and graphical forms.
@RISK also enables us to perform a sensitivity analysis, so that we can see which inputs
have the most effect on the outputs. Finally, @RISK provides a number of spreadsheet func-
tions that enable us to generate random numbers from a variety of probability distributions.

PrecisionTree
The PrecisionTree add-in is used in Chapter 7 to analyze decision problems with uncer-
tainty. The primary method for performing this type of analysis is to draw a decision tree.
Decision trees are inherently graphical, and they have always been difficult to implement
in spreadsheets, which are based on rows and columns. However, PrecisionTree does this
in a very clever and intuitive way. Equally important, once the basic decision tree has been
built, it is easy to use PrecisionTree to perform a sensitivity analysis on the model inputs.

TopRank
Although we will not use the other Palisade add-ins as extensively as @RISK and
PrecisionTree, they are all worth investigating. TopRank is the most general of them. It
starts with any spreadsheet model, where a set of inputs are used, along with a number of
spreadsheet formulas, to produce an output. TopRank then performs a sensitivity analysis
to see which inputs have the largest effect on the output. For example, it might tell us
which input affects after-tax profit the most: the tax rate, the risk-free rate for investing, the
inflation rate, or the price charged by a competitor. Unlike @RISK, TopRank is used when
uncertainty is not explicitly built into a spreadsheet model. However, it considers uncer-
tainty implicitly by performing sensitivity analysis on the important model inputs.

RISKOptimizer
RISKOptimizer combines optimization with simulation. There are often times when we
want to use simulation to model some business problem, but we also want to optimize a
summary measure, such as a mean, of an output distribution. This optimization can be
performed in a trial-and-error fashion, where we try a few values of the decision vari-
able(s) and see which provides the best solution. However, RISKOptimizer provides a
more automatic (and time-intensive) optimization procedure.

BestFit
BestFit is used to determine the most appropriate probability distribution for a spreadsheet
model when we have data on some uncertain quantity. For example, a simulation might
model each week’s demand for a product as a random variable. What probability distribu-
tion should we use for weekly demand: the well-known normal distribution or possibly
some skewed distribution? If we have historical data on weekly demands for the product,
we can feed them into BestFit and let it recommend the distribution that best fits the data.
This is a very useful tool in real applications. Instead of guessing a distribution that we
think might be relevant, we can let BestFit point us to a distribution that fits historical data
well. We discuss BestFit briefly in Chapter 6.

RISKview
Finally, RISKview is a drawing tool that complements @RISK. A number of probability dis-
tributions are available in @RISK and can be used in simulations. Each has an associated
@RISK function, such as RiskNormal, RiskBinomial, and so on. Before selecting any of
these distributions, however, it is useful (especially for beginners) to see what these distribu-
tions look like. RISKview performs this task easily. For any selected probability distribution
(and any selected parameters of this distribution), it creates a graph of the distribution, and it
allows us to find probabilities for the distribution in a completely intuitive, graphical manner.
We use RISKview in Chapter 16 to help learn about potential input probability distributions
for simulation models.
Palisade Corporation Finally, RISKview is a drawing tool that complements @RISK. A number of probability dis-
originally marketed tributions are available in @RISK and can be used in simulations. Each has an associated
BestFit and RISKview @RISK function, such as RiskNormal, RiskBinomial, and so on. Before selecting any of
as separate products. these distributions, however, it is useful (especially for beginners) to see what these distribu-
Although they still exist tions look like. RISKview performs this task easily. For any selected probability distribution
as separate products, (and any selected parameters of this distribution), it creates a graph of the distribution, and it
their functionality is now allows us to find probabilities for the distribution in a completely intuitive, graphical manner.
included in @RISK. We use RISKview in Chapter 16 to help learn about potential input probability distributions
for simulation models.

Software Guide
Figure 1.2 provides a guide to where these various add-ins appear throughout the book. We
don’t show Excel explicitly in this figure for the simple reason that Excel is used exten-
sively in all chapters.
With Excel and the add-ins included in this book, you have a wealth of software at
your disposal. The examples and step-by-step instructions throughout this book will help
you to become a power user of this software. Admittedly, this takes plenty of practice and
a willingness to experiment, but it is certainly within your grasp. When you are finished,
we will not be surprised if you rate “improved software skills” as the most valuable thing
you have learned from this book.

Developer Add-In
Figure 1.2
Software Guide StatTools

@RISK

RISKview

10 Chapter 1 Introduction to Data Analysis and Decision Making

1 A SAMPLING OF EXAMPLES
Perhaps the best way to illustrate what you will be learning in this book is to preview a few
examples from later chapters. Our intention here is not to teach you any methods; that will
come later. We only want to indicate the types of problems you will learn how to solve.
Each example below is numbered as in the chapter where it appears.

EXAMPLE 4
 he Spring Mills Company produces and distributes a wide variety of manufactured
goods. Because of its variety, it has a large number of customers. The company classi-
fies these customers as small, medium, and large, depending on the volume of business
each does with Spring Mills. Recently, Spring Mills has noticed a problem with its
accounts receivable. It is not getting paid back by its customers in as timely a manner as it
would like. This obviously costs Spring Mills money. If a customer delays a payment of
$300 for 20 days, say, then the company loses potential interest on this amount. The com-
pany has gathered data on 280 customer accounts. For each of these accounts, the data set
lists three variables: Size, the size of the customer (coded 1 for small, 2 for medium, 3 for
large); Days, the number of days since the customer was billed; and Amount, the amount
the customer owes. What information can we obtain from these data?

Objective To use charts, summary

accounts receivable at Spring Mills.

Solution
It is always a good idea to get a rough sense of the data first. We do this by calculating sev-
eral summary measures for Days and Amount, a histogram of Amount, and a scatterplot of
Amount versus Days. The next logical step is to see whether the different customer sizes
have any effect on either Days, Amount, or the relationship between Days and Amount.
There is obviously a lot going on here. We point out the following: (1) there are far fewer
large customers than small or medium customers; (2) the large customers tend to owe
considerably more than small or medium customers; (3) the small customers do not tend
to be as long overdue as the medium or large customers; and (4) there is no relationship
between Days and Amount for the small customers, but there is a definite positive relation-
ship between these variables for the medium and large customers. If Spring Mills really
wants to decrease its receivables, it might want to target the medium-size customer group,
from which it is losing the most interest. Or it could target the large customers because
they owe the most on average. The most appropriate action depends on the cost and effec-
tiveness of targeting any particular customer group. However, the analysis presented here
gives the company a much better picture of what’s currently going on.

This example from Chapter 3 is a typical example of trying to make sense out of a large
data set. Spring Mills has 280 observations on each of three variables. By realistic stan-
dards, this is not a large data set, but it still presents a challenge. We examine the data from
a number of angles and present several tables and charts. For example, the scatterplots in
Figures 1.3 through 1.5 clearly indicate that there is a positive relationship between the
amount owed and the number of days since billing for the medium- and large-size cus-
tomers, but that no such relationship exists for the small-size customers. As we will see,
graphs such as these are very easy to construct in Excel, regardless of the size of the data set.

Figure 1.3 Scatterplot of Amount versus Days for Small-size Customers

 Figure 1.4 Scatterplot of Amount versus Days for Medium-size Customers

12 Chapter 1 Introduction to Data Analysis and Decision Making

Figure 1.5 Scatterplot of Amount versus Days for Large-size Customers

EXAMPLE 7
ciTools Incorporated, a company that specializes in scientific instruments, has been
invited to make a bid on a government contract. The contract calls for a specific num-
ber of these instruments to be delivered during the coming year. The bids must be sealed
(so that no company knows what the others are bidding), and the low bid wins the contract.
SciTools estimates that it will cost $5000 to prepare a bid and $95,000 to supply the instru-
ments if it wins the contract. On the basis of past contracts of this type, SciTools believes
that the possible low bids from the competition, if there is any competition, and the associ-
ated probabilities are those shown in Table 1.1. In addition, SciTools believes there is a
30% chance that there will be no competing bids.

Table 1.1 Probabilities of Low Bids from Competi

Low Bid Probability
Less than $115,000
Between $115,000 and $120,000
Between $120,000 and $125,000
Greater than $125,000

Solution
This is a typical example of decision making
SciTools has to make decisions now (whether
knowing what the competition is going to do.
 outcome, but it can make a rational decision in light of the uncertainty it faces. We will see
how decision trees, produced easily with the PrecisionTree add-in to Excel, not only lay
out all of the elements of the problem in a logical manner but also indicate the best solu-
tion. The completed tree for this problem is in Figure 1.6, which indicates that SciTools
should indeed prepare a bid, for the amount $115,000.

Figure 1.6 Decision Tree for SciTools

EXAMPLE 10
n auditor wants to determine the proportion of invoices that contain price errors—that
is, prices that do not agree with those on an authorized price list. He checks 93 ran-
domly sampled invoices and finds that two of them include price errors. What can he con-
clude, in terms of a 95% one-sided confidence interval, about the proportion of all invoices
with price errors?

Solution
 This is an important application of statistical in
try to determine what is true about a populatio
by examining a relatively small sample from
limit so that he is 95% confident that the ove

14 Chapter 1 Introduction to Data Analysis and Decision Making

greater than this upper limit. We show the spreads

that the auditor can be 95% confident that the over
no greater than 6.6% (see cell B10).

Figure 1.7
Analysis of Auditing
Example

EXAMPLE 11
he Bendrix Company manufactures various types o
ager of the factory wants to get a better understand
head costs include supervision, indirect labor, supp
depreciation, and a number of miscellaneous items
tion, insurance, utilities, and janitorial and mainten
costs are “fixed” in the sense that they do not vary
being done, whereas others are “variable” and do v
The fixed overhead costs tend to come from the su
neous categories, whereas the variable overhead co
supplies, payroll taxes, and overtime premiums ca
a clear line between the fixed and variable overhea
The Bendrix manager has tracked total overhead c
help “explain” these, he has also collected data on
amount of work done at the factory. These variable

■ MachHrs:
■ ProdR
The first of these is a direct measure of the amoun
second, we note that Bendrix manufactures parts in
sponds to a production run. Once a production run
for the next production run. During this setup there
machinery is reconfigured for the part type schedu
Therefore, the manager believes both of these vari
ways) for variations in overhead costs. Do scatterp

Solution
 This is a typical regression example, in a cost-acco
see what type of relationship, if any, there is betwe
tory variables: number of machine hours and numb
requested appear in Figures 1.8 and 1.9. They do i
tionship between overhead and the two explanator

Figure 1.8 Scatterplot of Overhead versus Machine Hours

Figure 1.9 Scatterplot of Overhead versus Production Runs

16 Chapter 1 Introduction to Data Analysis and Decision Making

However, regression goes well beyond scatterplots. It estimates an equation relating

the variables. This equation can be determined from regression output such as that shown
in Figure 1.10. This output implies the following equation for predicted overhead as a
function of machine hours and production runs:
Predicted Overhead 5 3997 1 43.5

Figure 1.10 Multiple Regression Output for Bendrix Example

 The positive coefficients of MachHrs and ProdRuns indicate the effects these variables
have on overhead. We will not take the example any further at this point but will simply indic
that it is easy to generate the output in Figure 1.10 with StatTools. The challenge is learning
how to interpret it. We spend plenty of time in Chapters 11 and 12 on interpretation issues. ■

EXAMPLE 13
he file PCDevices.xls contains quarterly sales data (in millions of dollars) for a chip-
manufacturing firm from the beginning of 1990 through the end of 2004. Are the com-
pany’s sales growing exponentially through this entire period?

Solution
This example illustrates a regression-based trend curve, one of several possible forecasting
techniques for a time series variable. A time series graph of the company’s quarterly sales
appears in Figure 1.11. It indicates that sales have been increasing steadily at an increasing
rate. This is basically what an exponential trend curve implies. To estimate this curve we
use regression analysis to obtain the following equation for predicted quarterly sales as a
function of time:

Predicted Sales 5 6
This equation implies that the company’s sales are increasing by approximately 6.6%
per quarter during this period, which translates to an annual percentage increase of about
29%! As we see in Chapter 13, this is the typical approach used in forecasting. We look at
a time series graph to discover trends or other patterns in historical data and then use one
of a variety of techniques to fit the observed patterns and extrapolate them into the future.

Figure 1.11
Time Series Graph
of Quarterly Sales
at a PC Chip
Manufacturer

EXAMPLE 16
 t the present time, the beginning of year 1, the Barney-Jones Investment Corporation
has $100,000 to invest for the next 4 years. There are five possible investments,
labeled A through E. The timing of cash outflows and cash inflows for these investments is
somewhat irregular. For example, to take part in investment A, cash must be invested at the
beginning of year 1, and for every dollar invested, there are returns of $0.50 and $1.00 at
the beginnings of years 2 and 3. Similar information for the other investments are as fol-
lows, where all returns are per-dollar invested:

■ Investment B: Invest at the beginning of year 2, receive returns of $0.50 and $1.00 a
the beginnings of years 3 and 4.
■ Investment C: Invest at the beginning of year 1, receive return of $1.20 at the begin-
ning of year 2.
■ Investment D: Invest at the beginning of year 4, receive return of $1.90 at the begin
ning of year 5.
■ Investment E: Invest at the beginning of year 3, receive return of $1.50 at the begin-
ning of year 4.
We assume that any amounts can be invested in these strategies and that the returns are the
same for each dollar invested. However, to create a diversified portfolio, Barney-Jones
decides to limit the amount put into any investment to $75,000. The company wants an
investment strategy that maximizes the amount of cash on hand at the beginning of year 5.
At the beginning of any year, it can invest only cash on hand, which includes returns from
previous investments. Any cash not invested in any year can be put in a short-term money
market account that earns 3% annually.

Solution
This is one of many optimization examples w
situation is that a company such as Barney-Jo
certain constraints, that optimize some objectiv
the amounts to invest, subject to some constrain

18 Chapter 1 Introduction to Data Analysis and Decision Making

now. Our job is to formulate a spreadsheet model, similar to the one shown in Figure 1.12,
that relates the various elements of the problem.
The investment amounts in row 26 are the decision variables, called “changing cells”
in Excel’s terminology. When we formulate the model, we can enter any values in these
changing cells; we do not need to guess “good” values. Then we turn it over to Excel’s
Solver add-in. The Solver uses a powerful algorithm to find the optimal values in the
changing cells—that is, the values that optimize the objective while satisfying the con-
straints. The values shown in Figure 1.12 are actually the optimal values. They imply that
Barney-Jones can end with final cash of $286,792 by investing as indicated in row 26.

Figure 1.12 Investment Model

EXAMPLE 18
e assume that there are two dominant companies in the soft drink industry: “us” and
“them.” For this example, we will view everything from the point of view of “us.”
We start with a 45% market share. During each of the next 20 quarters, each company pro-
motes its product to some extent. To make the model simple, we will assume that each
company each quarter either promotes at a “regular” level or at a “blitz” level. Depending
on each company’s promotional behavior in a given month, the change in our market share
from this month to the next is triangularly distributed, with parameters given in Table 1.2.
For example, if we blitz and they don’t, then we could lose as much as 1% market share,
we could gain as much as 6% market share, and our most likely outcome is an increase of
2% market share. We want to develop a simulation model that allows us to gauge the long-
term change in our market share for any pattern of blitzing employed by us and them.

Table 1.2 Parameters of Market Share Change

Blitzer Minimum
Neither -0.03
Both -0.05
Only us -0.01
Only them -0.06

Solution
 This is a typical example of computer simulation. We make a number of assumptions,
build a spreadsheet model around these assumptions, explicitly incorporate uncer-
tainty into some of the cells, and see how this uncertainty affects “bottom-line” out-
puts. The simulation model appears in Figure 1.13. Several cells in this model are
random, including all of the numerical values in rows 21 through 24. Therefore, the
numbers you see in this figure represent just one possible scenario of how market
shares might evolve through time. By generating new random values, we see different
scenarios.
Our job is to build the logic and randomness into the spreadsheet model. Then we can
use Excel’s built-in tools or an add-in such as @RISK to replicate the model and keep track
of selected outputs. A typical result from @RISK appears in Figure 1.14. It shows a time
series graph of how our market share might evolve, given a certain strategy of blitzing by
our company and theirs.

Figure 1.13 Spreadsheet Simulation for Market Share Example

20 Chapter 1 Introduction to Data Analysis and Decision Making

Figure 1.14 Summary Chart of Our Market Share for One Set of Strategies
1 MODELING AND MODELS
We have already used the term model several times in this chapter. In fact, we have shown
several spreadsheet models in the previous section. Models and the modeling process are
key elements throughout this book, so we explain them in more detail in this section.3
A model is an abstraction of a real problem. A model tries to capture the essence and
key features of the problem without getting bogged down in relatively unimportant details.
There are different types of models, and, depending on an analyst’s preferences and skills,
each can be a valuable aid in solving a real problem. We describe three types of models
here: (1) graphical models, (2) algebraic models, and (3) spreadsheet models.

1.4.1 Graphical Models

Graphical models are probably the most intuitive and least quantitative type of model. They
attempt to portray graphically how different elements of a problem are related—what affects
what. A very simple graphical model appears in Figure 1.15. It is called an “influence
diagram.” (It can be constructed with the PrecisionTree add-in discussed in Chapter 7, but we
will not use influence diagrams in this book.)

Figure 1.15
Influence Diagram
for Souvenir
Example

3Management scientists tend to use the terms model and modeling more than statisticians. However, many tradi-
tional statistics topics such as regression analysis and forecasting are clearly applications of modeling.
This particular influence diagram is for a company that is trying to decide how many
souvenirs to order for the upcoming Olympics. The essence of the problem is that the com-
pany will order a certain supply, customers will request a certain demand, and the combi-
nation of supply and demand will yield a certain payoff for the company. The diagram
indicates fairly intuitively what affects what. As it stands, the diagram does not provide
enough quantitative details to enable us to “solve” the company’s problem. But this is usu-
ally not the purpose of a graphical model. Instead, its purpose is usually to show the impor-
tant elements of a problem and how they are related. For complex problems this can be
very helpful and enlightening information for management.

1.4.2 Algebraic Models

Algebraic models are at the opposite end of the spectrum. By means of algebraic equations
and inequalities, they specify a set of relationships in a very precise way, and their precise-
ness and lack of ambiguity are very appealing to people with a mathematical background.
In addition, algebraic models can usually be stated concisely and with great generality.
A typical example is the “product mix” problem we discuss in Chapter 14. A company
can make several products, each of which contributes a certain amount to profit and con-
sumes certain amounts of several scarce resources. The problem is to select the product
mix that maximizes profit subject to the limited availability of the resources. All product
mix problems can be stated algebraically as follows:

max ^pjxj
j51

subject to ^aijxj # bi,

j51

0 # xj # uj,

Here xj is the amount of product j produced, uj is an upper limit on the amount of product j
that can be produced, pj is the unit profit margin for product j, aij is the amount of resource i
consumed by each unit of product j, bi is the amount of resource i available, n is the number
of products, and m is the number of scarce resources. This algebraic model states very con-
cisely that we should maximize total profit [expression (1.1)], subject to consuming no
more of the resources than is available [inequalities (1.2)], and all production quantities
should be between 0 and the upper limits [inequalities (1.3)].
Algebraic models such as this appeal to mathematically trained analysts. They are
concise, they spell out exactly which data are required (we would need to estimate the uj’s,
the pj’s, the aij’s, and the bi’s from company data), they scale well (a problem with
500 products and 100 resource constraints is just as easy to state as one with only 5 prod-
ucts and 3 resource constraints), and many software packages accept algebraic models in
essentially the same form as shown here, so that no “translation” is required. Indeed, alge-
braic models were the preferred type of model for years—and still are by many analysts.
Their main drawback is that they require an ability to work with abstract mathematical
symbols. Some people have this ability, but many perfectly intelligent people do not.

1.4.3 Spreadsheet Models

A fairly recent alternative to algebraic modelin
ing various quantities with algebraic equations
sheet with cell formulas. This process is much
experience). One of the primary reasons for
22 Chapter 1 Introduction to Data Analysis and Decision Making

spreadsheets. If you enter a formula incorrectly, it is often immediately obvious (from error
messages or unrealistic numbers) that you have made an error, which you can then go back
and fix. Algebraic models provide no such immediate feedback.
A specific comparison might help at this point. We already saw a general algebraic
model of the product mix problem. Figure 1.16, taken from Chapter 14, illustrates a
spreadsheet model for a specific example of this problem. The spreadsheet model should
be fairly self-explanatory. All quantities in shaded cells are inputs to the model, the quanti-
ties in row 16 are the decision variables (they correspond to the xj’s in the algebraic model),
and all other quantities are created through appropriate Excel formulas. To indicate con-
straints, we enter inequality signs in appropriate cells.

Figure 1.16 Optimal Solution for Product Mix Example

A B C D E
1 Product mix model
2
3 Input data
4 Hourly wage rate $8.00
5 Cost per oz of metal $0.50
6 Cost per oz of glass $0.75
7
8 Frame type 1 2 3 4
9 Labor hours per frame 2132
10 Metal (oz.) per frame 4212
11 Glass (oz.) per frame 6212
12 Unit selling price $28.50 $12.50 $29.25 $21.50
13
14 Production plan
15 Frame type 1 2 3 4
16 Frames produced 1000 800 400 0
17 <= <= <= <=
18 Maximum sales 100020005001000
19
20 Resource constraints Used Available
21 Labor hours 4000 <= 4000
22 Metal (oz.) 6000 <= 6000
23 Glass (oz.) 8000 <= 10000
24
25 Revenue, cost summary
26 Frame type 1 2 3 4
27 Revenue $28,500 $10,000 $11,700 $0
28 Costs of inputs
29 Labor $16,000 $6,400 $9,600 $0
30 Metal $2,000 $800 $200 $0
31 Glass $4,500 $1,200 $300 $0
32 Profit $6,000 $1,600 $1,600 $0
Although a well-designed and well-documented spreadsheet model such as the one in
Figure 1.16 is undoubtedly more intuitive for most people than its algebraic counterpart,
the art of developing good spreadsheet models is not easy. Obviously, they must be correct.
The formulas relating the various quantities must have the correct syntax, the correct cell
references, and the correct logic. In complex models this can be quite a challenge.
However, correctness is not enough. If spreadsheet models are to be used in the busi-
ness world, they must also be well designed and well documented. Otherwise, no one other
than you (and maybe not even you after a few weeks have passed) will be able to under-
stand what your models do or how they work. The strength of spreadsheets is their

flexibility—you are limited only by your ima

liability in spreadsheet modeling unless you
Note the clear design in Figure 1.16. Mos
the spreadsheet. All of the financial calculati
constraints, the two sides of the constraint
range B21:D23). Borders, colors (which app
shading are used for added clarity. Descrip
imposes none of these “rules,” but yo
We have made a conscious effort to establish
this book. We have designed and redesigned ou
as possible. This does not mean that you have t
develop their own spreadsheet style—but our m
Just remember that in the business world, you
then up to you to develop a model that is not on
others. This takes a lot of practicing and a lot of

1.4.4 The Seven-Step Modeling Process

Most of the modeling you will do in this book is only part of the overall modeling process
typically done in the business world. We portray it as a seven-step process, as discussed
here. Of course, not all problems require all seven steps. For example, the analysis of sur-
vey data might entail primarily steps 2 (data analysis) and 5 (decision making), without the
formal model building discussed in steps 3 and 4.

The Modeling Process

1. Define the problem. Typically, a company does not develop a model unless it
believes it has a problem. Therefore, the modeling process really begins by identify-
ing an underlying problem. Perhaps the company is losing money, perhaps its mark
share is declining, or perhaps its customers are waiting too long for service. Any
number of problems might be evident. However, as several people have warned [see
Miser (1993) and Volkema (1995), for example], this step is not always as straight-
forward as it might appear. The company must be sure that it has identified the righ
problem before it spends time, effort, and money trying to solve it.
 For example, Miser cites the experience of an analyst who was hired by the military
investigate overly long turnaround times between fighter planes landing and taking
again to rejoin the battle. The military was convinced that the problem was caused b
inefficient ground crews; if they were sped up, turnaround times would decrease. Th
analyst nearly accepted this statement of the problem and was about to do classical
time-and-motion studies on the ground crew to pinpoint the sources of their ineffici
However, by snooping around, he found that the problem obviously lay elsewhere. T
trucks that refueled the planes were frequently late, which in turn was due to the ine
cient way they were refilled from storage tanks at another location. Once this latter p
lem was solved—and its solution was embarrassingly simple—the turnaround times
decreased to an acceptable level without any changes on the part of the ground crew
the analyst had accepted the military’s statement of the problem, the real problem m
never have been located or solved.

2. Collect and summarize data. This crucial step in the process is often the most
tedious. All organizations keep track of various data on their operations, but these d
are often not in the form an analyst requires. They are also typically scattered in dif-
ferent places throughout the organization, in all kinds of different formats. Therefor
one of the first jobs of an analyst is to gather exactly the right data and summarize th

24 Chapter 1 Introduction to Data Analysis and Decision Making

data appropriately—as we discuss in detail in Chapters 2 and 3—for use in the model.
Collecting the data typically requires asking questions of key people (such as the
accountants) throughout the organization, studying existing organizational databases,
and performing time-consuming observational studies of the organization’s processes.
In short, it entails a lot of leg work.
3. Formulate a model. This is the step we emphasize, especially in the latter chapters
of the book. The form of the model varies from one situation to another. It could be a
graphical model, an algebraic model, or a spreadsheet model. The key is that the
model should capture the key elements of the business problem in such a way that it
is understandable by all parties involved. This latter requirement is why we favor
spreadsheet models, especially when they are well designed and well documented.
4. Verify the model. Here the analyst tries to determine whether the model developed
in the previous step is an accurate representation of reality. A first step in determining
how well the model fits reality is to check whether the model is valid for the current
situation. This verification can take several forms. For example, the analyst could use
the model with the company’s current values of the input parameters. If the model’s
outputs are then in line with the outputs currently observed by the company, the ana-
lyst has at least shown that the model can duplicate the current situation.
A second way to verify a model is to enter a number of input parameters (even if
they are not the company’s current inputs) and see whether the outputs from the
model are reasonable. One common approach is to use extreme values of the inputs
to see whether the outputs behave as they should. If they do, then we have another
piece of evidence that the model is reasonable.
If certain inputs are entered in the model, and the model’s outputs are not as expected,
there could be two causes. First, the model could simply be a poor representation of
reality. In this case it is up to the analyst to refine the model until it provides reason-
ably accurate predictions. The second possible cause is that the model is fine but our
intuition is not very good. In this case the fault lies with us, not the model.
 A typical example of faulty intuition occurs with random sequences of 0’s and 1’s,
such as might occur with successive flips of a fair coin. Most people expect that heads
and tails will alternate and that there will be very few sequences of, say, four or more
heads (or tails) in a row. However, a perfectly accurate simulation model of these flips
will show, contrary to what most people expect, that fairly long runs of heads or tails
are not at all uncommon. In fact, one or two long runs should be expected if there are
enough flips.
The fact that outcomes sometimes defy intuition is an important reason why models
are important. These models prove that our ability to predict outcomes in complex
environments is often not very good.
5. Select one or more suitable decisions. Many, but not all, models are decision mod-
els. For any specific decisions, the model indicates the amount of profit obtained, the
amount of cost incurred, the level of risk, and so on. If we believe the model is work-
ing correctly, as discussed in step 4, then we can use the model to see which deci-
sions produce the best outputs.
6. Present the results to the organization. In a classroom setting you are typically fin-
ished when you have developed a model that correctly solves a particular problem. In
the business world a correct model, even a useful one, is not always enough. An ana-
lyst typically has to “sell” the model to management. Unfortunately, the people in
management are sometimes not as well trained in quantitative methods as the analyst,
so they are not always inclined to trust complex models.

There are two ways to mitigate this pro

people throughout the company in the m
that everyone has an understanding o
Second, it helps to use a spreadsheet
designed and documented properly. A
comfortable with spreadsheets, so sprea
7. Implement the model and update it over time.Again, there is a big difference
between a classroom situation and a business situation. When you turn in a classroo
assignment, you are typically finished with that assignment and can await the next o
In contrast, an analyst who develops a model for a company can usually not pack up
bags and leave. If the model is accepted by management, the company will then nee
implement it company-wide. This can be very time consuming and politically diffic
especially if the model’s prescriptions represent a significant change from the past. A
the very least, employees must be trained how to use the model on a day-to-day bas

In addition, the model will probably have to be updated over time, either because of
changing conditions or because the company sees more potential uses for the model
as it gains experience using it. This presents one of the greatest challenges for a
model developer, namely, the ability to develop a model that can be modified as the
need arises. Keep this in mind as you develop models throughout this book. Always
try to make them as general as possible.

2 CONCLUSION
 In this chapter we tried to convince you that the skills in this book are important for you to
know as you enter the business world. The methods we discuss are no longer the sole
province of the “quant jocks.” By having a PC on your desk that is loaded with powerful
software, you incur a responsibility to use this software to solve business problems. We
have described the types of problems you will learn to solve in this book, along with the
software you will use to solve them. We also discussed the modeling process, a theme that
runs throughout this book. Now it’s time for you to get started!

26 Chapter 1 Introduction to Data Analysis and Decision Making

C ASE 1.1 ENTERTAINMENT ON A CRUISE SHIP

ruise ship traveling has become big business.

Many cruise lines are now competing for cus-
tomers of all age groups and socioeconomic levels.
They offer all types of cruises,from relatively inex-
pensive 3- to 4-day cruises in the Caribbean,to
12- to 15-day cruises in the Mediterranean,to
several-month around-the-world cruises.Cruises
have several features that attract customers,many
of whom book 6 months or more in advance:(1)
they offer a relaxing,everything-done-for-you way
to travel;(2) they serve food that is plentiful,usu-
ally excellent,and included in the price of the
cruise;(3) they stop at a number of interesting
ports and offer travelers a way to see the world;
and (4) they provide a wide variety of entertain-
ment,particularly in the evening.
This last feature,the entertainment,presents a
difficult problem for a ship’s staff. A typical cruise
might have well over 1000 passengers,including
elderly singles and couples,middle-aged people with
or without children,and young people,often honey-
mooners. These various types of passengers have
varied tastes in terms of their after-dinner prefer-
ences in entertainment.Some want traditional
ruise ship traveling has become big business. weekly,and va
Many cruise lines are now competing for cus- ans played oc
tomers of all age groups and socioeconomic levels. Although this
They offer all types of cruises,from relatively inex- passengers,m
pensive 3- to 4-day cruises in the Caribbean,to dance.The nig
12- to 15-day cruises in the Mediterranean,to who were con
several-month around-the-world cruises.Cruises often had less
have several features that attract customers,many ence—sometim
of whom book 6 months or more in advance:(1) singers,dance
they offer a relaxing,everything-done-for-you way ences,but ther
to travel;(2) they serve food that is plentiful,usu- spite of this,th
ally excellent,and included in the price of the and they stuck
cruise;(3) they stop at a number of interesting short-term fina
ports and offer travelers a way to see the world; ence.The perf
and (4) they provide a wide variety of entertain- one was in the
ment,particularly in the evening. already paid (
This last feature,the entertainment,presents a of the cost of t
difficult problem for a ship’s staff. A typical cruise tunity cost to
might have well over 1000 passengers,including the loss of liqu
elderly singles and couples,middle-aged people with the entertainm
or without children,and young people,often honey- tainers was no
mooners. These various types of passengers have houses—but th
varied tastes in terms of their after-dinner prefer- their hours we
ences in entertainment.Some want traditional getting paid to
dance music,some want comedians,some want If you were in
rock music,some want movies,some want to go ship,how woul
back to their cabins and read,and so on.Obviously, tainment:Is it
cruise entertainment directors want to provide the low-quality en
variety of entertainment their customers desire— entertainment
within a reasonable budget—because satisfied cus- How might yo
tomers tend to be repeat customers. The question straints might
is how to provide the right mix of entertainment. keep a strict s
On a cruise one of the authors and his wife took cruise director
a few years ago,the entertainment was of high quality Would you gat
and there was plenty of variety.A seven-piece show What data wo
band played dance music nightly in the largest lounge, cial considerat
two other small musical combos played nightly at two be long-term o
smaller lounges,a pianist played nightly at a piano bar
in an intimate lounge,a group of professional singers
4There was also a mod
and dancers played Broadway-type shows about twice
attract the same peopl
the ship was in port.
Data Analysis
king

METHODS TO SOLVE REAL

data analysis and decision making,you

of quantitative methods to the “real
he December 31,1997,edition of USA Today
Formula for Success” provides some convinc-
y of the methods you will be learning.The sub-
ses turn to algorithms to solve complex
business problems tend to be very complex.In
xecutives used a “by the seat of your pants”
hat is,they used their business experience,
htful guesswork to obtain solutions.But com-
y so far in the solution of the complex prob-
is where data analysis and decision making—
in the title of the article—are so useful.When
mplemented in user-friendly computer soft-
pplied to complex problems,the results can be
ompany,DFI Aeronomics,sells algorithm-
s it succinctly:“It’s like taking raw information

his book is that they are applicable to

ments.The article mentions the following
 1

een applied;others will be discussed

ems,which cost between $10 million and

add $50 million to $100 million annually

analysis to determine exactly how many

the holiday rush.
researchers test potentially dangerous
.
ild a computer system for the
ce-impossible task:make exact real-time
ct—and convinced the DOE that its sys-
ng quantitative methods that would cut the

s all use quantitative analysis to implement

” In this method,different prices are
g on their willingness to pay.The effect is
venues increase.
cel spreadsheet software contains a
ey statement.Many of the algorithms that
ery complex mathematically.They are well
readers of this book.However,users
nd the algorithms.They need only to
opriate algorithms can be applied and
.For example,we see in Chapters 14
omplex problems.You will not learn the
ou will learn how to use Solver very
her methods discussed in this book.You
the computer’s “black box” as it per-
ome a very effective problem solver by

the age of technology. This has two important implications for everyone
ness world. First, technology has made it possible to collect huge amounts
s collect point-of-sale data on products and customers every time a trans-
redit agencies have all sorts of data on people who have or would like to
vestment companies have a limitless supply of data on the historical pat-
bonds, and other securities; and government agencies have data on eco-
the environment, social welfare, consumer product safety, and virtually
e we can imagine. It has become relatively easy to collect the data. As a
entiful. However, as many organizations are now beginning to discover, it
allenge to analyze and make sense of all the data they have collected.
tant implication of technology is that it has given many more people the
onsibility to analyze data and make decisions on the basis of quantitative
ntering the business world can no longer pass all of the quantitative analy-
jocks,” the technical specialists who have traditionally done the number
ast majority of employees now have a desktop or laptop computer at their
However, companies don’t
ture? They will be you! We
oth the undergraduate and
data sets, run regression
, and run simulations. You
rtant decisions to help gain
g or able to do so, there will
han happy to replace you.
quantitative methods to
nds-on way. We discuss a
e variety of realistic busi-
les from finance, market-
lyze these examples, we
gether with a number of

1 Introduction 3

y-step details of the

ean comfortably back

ther people use quantita-
learn best by actively fol-
ort, you will learn by
me very useful skills for

ked with quantitative methods and examples, probably more than can be
single course. Therefore, we purposely intend to keep this introductory
hat you can get on with the analysis. Nevertheless, it is useful to introduce
will be learning and the tools you will be using. In this section we provide
e methods covered in this book and the software that is used to implement
e next section we preview some of the examples we cover in much more
apters. Finally, we present a brief discussion of models and the modeling
mary purpose at this point is to stimulate your interest in what is to follow.
parate fields: statistics and
analysis, whereas man-
d decision making. In the
d, sometimes widely.
However, from a user’s
in accomplishing what the

anagement science”
on of useful quantitative
s decisions. In addition,
sing a single package,
ds of statistics and man-
mportantly, their combina-
usiness problems.
in the title: data analysis
ach of these themes has sub-
and the search for
echniques for problems with
and structured sensitivity
nty and modelinguncertainty
these themes and subthemes.
s in data, we must deal with
we must deal with uncer-
we must deal with uncertainty,

k does not detract from its impor-

es in the remaining chap-

k’s contents in more detail.

Subthemes Chapters Where Emphasized

 o summarize the infor-
ies, as well as numerical
ions. The material in these
ut it is extremely impor-
are now able to collect
es it all mean? Although
me of which we cover in
cial for obtaining an initial
ns now make what was
able tool for “slicing and
d at the complex analysis it

nt in Chapter 4 to see
xperience that many students
alysis once they have the
getting the right data, in the
hapter 4 we present a num-
ecifically, we discuss meth-
n Excel data sets, for using
n external databases (such
data directly into Excel
d” data values). This chapter
aware of.
with uncertainty, we need
ding in Chapters 5 and 6.

1 An Overview of the Book 5

 basic rules of probability and then discusses the extremely important con-
y distributions. Chapter 6 follows up this discussion by focusing on two of
nt probability distributions, the normal and binomial distributions. It also
he Poisson and exponential distributions, which have many applications in
probability models.
d that one of the best ways to make probabilistic concepts “come alive”
derstand is by using computer simulation. Therefore, simulation is a com-
runs through this book, beginning in Chapter 5. Although the final two
book are devoted entirely to simulation, we do not hesitate to use simula-
on early and often to illustrate difficult statistical concepts.
we apply our knowledge of probability to decision making under uncer-
es of problems—faced by all companies on a continual basis—are charac-
eed to make a decision now, even though important information (such as
duct or returns from investments) will not be known until later. The mate-
7 provides a rational basis for making such decisions. The methods we
guarantee perfect outcomes—the future could unluckily turn out differ-
d expected—but they do enable us to proceed rationally and make the best
umstances. Additionally, the software we use to implement these methods
very little extra work, to see how sensitive the optimal decisions are to
ucial because the inputs to many business problems are, at best, educated
we examine the role of risk aversion in these types of decision problems.
9, and 10 we discuss sampling and statistical inference. Here the basic
imate one or more characteristics of a population. If it is too expensive or
to learn about the entire population—and it usually is—we instead select
e from the population and then use the information in the sample to infer
cs of the population. We see this continually on news shows that describe
ous polls. We also see it in many business contexts. For example, auditors
only a fraction of a company’s records. Then they infer the characteristics
ulation of records from the results of the sample to conclude whether the
mpany has been following acceptable accounting standards.
and 12 we discuss the extremely important topic of regression analysis,
study relationships between variables. The power of regression analysis is
Every part of a business has variables that are related to one another, and
ften be used to estimate possible relationships between these variables. In
unting, regression is used to estimate how overhead costs depend on direct
production volume. In marketing, regression is used to estimate how sales
s on advertising and other marketing variables. In finance, regression is
e how the return of a stock depends on the “market” return. In real estate
sion is used to estimate how the selling price of a house depends on the
on of the house and characteristics such as the number of bedrooms and
Regression analysis finds perhaps as many uses in the business world as
any method in this book.
n, we move to times series analysis and forecasting in Chapter 13. This
arly important for providing inputs into business decision problems. For
acturing companies must forecast demand for their products to make sen-
about quantities to order from their suppliers. Similarly, fast-food restau-
ast customer arrivals, sometimes down to the level of 15-minute intervals,
so that they can staff their restaurants appropriately.
ny approaches to forecasting, ranging from simple to complex. Some
on-based methods, in which one or more time series variables are used to
able of interest, whereas other methods are based on extrapolation. In an
method the historical patterns of a time series variable, such as product
 An Overview of the Book 7

es of Excel, but relatively few

w people are the “power
To get you started, the file
ook explains some of the
be able to use. These
TIF functions. They also
l command, the Goal Seek
n perform routine spread-
help you perform these tasks

features in more
ool that enables you to
l has many useful tools, but
d powerful of all. We
roduce Excel’s RAND
n all spreadsheet simula-

-in. This add-in, developed

—all behind the scenes—
heet optimization add-in
ed to solve optimization
vironment.
k discusses basic statistical analysis. Here we needed to make an important
eloped the book. A number of excellent statistical software packages are on
ng Minitab, SPSS, SAS, StatGraphics, and many others. Although there are
Windows versions of these packages, they are not spreadsheet-based. We
gh our own experience that students resist the use of nonspreadsheet pack-
their inherent quality, so we wanted to use Excel as our “statistics package.”
ss SPSS and SAS in Chapter 4, but they are not used anywhere else in the
ately, Excel’s built-in statistical tools are rather limited, and the Analysis
loped by a third party) that ships with Excel has significant limitations.
we developed an add-in called StatTools that accompanies this book.2
werful, easy to use, and capable of generating output quickly in an easily
m. We do not believe you should have to spend hours each time you want
statistical output. This might be a good learning experience the first time,
cts as a strong incentive not to perform the analysis at all! We believe you
o generate output quickly and easily. This gives you the time to interpret
tput, and it also allows you to try different methods of analysis.
ion involves the construction of histograms, scatterplots, and time series
Chapter 2. All of these extremely useful graphs can be created in a straight-
xcel’s built-in tools. But by the time you perform all the necessary steps and
s exactly as you want them, you will not be very anxious to repeat the whole
Tools does it all quickly and easily. (You still might want to “dress up” the

tatTools. Palisade Corporation

e user interface has changed con-

n a later chapter, say, to look

do so in a matter of seconds.
ions of a spreadsheet sim-
keeps track of the outputs
ar and graphical forms.
we can see which inputs
umber of spreadsheet func-
f probability distributions.

An Overview of the Book 9

tribution for a spreadsheet

ple, a simulation might
What probability distribu-
distribution or possibly
demands for the product,
ion that best fits the data.
g a distribution that we
ion that fits historical data

A number of probability dis-

s. Each has an associated
. Before selecting any of
s) to see what these distribu-
cted probability distribution
ph of the distribution, and it
y intuitive, graphical manner.
put probability distributions
A number of probability dis-
s. Each has an associated
. Before selecting any of
s) to see what these distribu-
cted probability distribution
ph of the distribution, and it
y intuitive, graphical manner.
put probability distributions

ar throughout the book. We

hat Excel is used exten-

alth of software at
ghout this book will help
kes plenty of practice and
When you are finished,
s the most valuable thing

Chapter(s) Where Used

6, 16–17

is book is to preview a few

you any methods; that will
will learn how to solve.
ears.
y of manufactured
. The company classi-
the volume of business
a problem with its
n as timely a manner as it
mer delays a payment of
on this amount. The com-
ese accounts, the data set
mall, 2 for medium, 3 for
and Amount, the amount
data?

measures, and pivot tables to understand data on

do this by calculating sev-

mount, and a scatterplot of
different customer sizes
ween Days and Amount.
ng: (1) there are far fewer
ustomers tend to owe
ll customers do not tend
there is no relationship
a definite positive relation-
s. If Spring Mills really
ium-size customer group,
rge customers because
nds on the cost and effec-
e analysis presented here
ng on.

ake sense out of a large

ables. By realistic stan-
We examine the data from
ample, the scatterplots in
lationship between the
m- and large-size cus-
tomers. As we will see,
ess of the size of the data set.

1 A Sampling of Examples 11

omers
Customers
tomers

struments, has been

ls for a specific num-
The bids must be sealed
e low bid wins the contract.
95,000 to supply the instru-
s type, SciTools believes
ompetition, and the associ-
Tools believes there is a

s of Low Bids from Competition

Probability
0.2
0.4
0.3
0.1

l example of decision making under uncertainty, the topic of Chapter 7.

make decisions now (whether to bid and, if so, how much to bid), without
e competition is going to do. The company can’t assure itself of a perfect
 1 A Sampling of Examples 13

tainty it faces. We will see

n to Excel, not only lay
o indicate the best solu-
indicates that SciTools

ain price errors—that

He checks 93 ran-
e errors. What can he con-
e proportion of all invoices
ant application of statistical inference in the auditing profession. Auditors
what is true about a population (in this case, all of a company’s invoices)
relatively small sample from the population. The auditor wants an upper
is 95% confident that the overall proportion of invoices with errors is no

er limit. We show the spreadsheet solution in Figure 1.7, which shows

e 95% confident that the overall proportion of invoices with errors is
(see cell B10).

y manufactures various types of parts for automobiles. The man-

ants to get a better understanding of overhead costs. These over-
pervision, indirect labor, supplies, payroll taxes, overtime premiums,
umber of miscellaneous items such as charges for building deprecia-
ies, and janitorial and maintenance expenses. Some of these overhead
he sense that they do not vary appreciably with the volume of work
others are “variable” and do vary directly with the volume of work.
osts tend to come from the supervision, depreciation, and miscella-
ereas the variable overhead costs tend to come from the indirect labor,
s, and overtime premiums categories. However, it is not easy to draw
he fixed and variable overhead components.
r has tracked total overhead costs over the past 36 months. To
he has also collected data on two variables that are related to the
e at the factory. These variables are

MachHrs: number of machine hours used during the month

ProdRuns: number of separate production runs during the month
direct measure of the amount of work being done. To understand the
Bendrix manufactures parts in fairly large batches. Each batch corre-
on run. Once a production run is completed, the factory must “set up”
on run. During this setup there is typically some downtime while the
gured for the part type scheduled for production in the next batch.
ger believes both of these variables might be responsible (in different
n overhead costs. Do scatterplots support this belief?
ession example, in a cost-accounting setting. The manager is trying to
ionship, if any, there is between overhead costs and the two explana-
er of machine hours and number of production runs. The scatterplots
Figures 1.8 and 1.9. They do indeed indicate a positive and linear rela-
rhead and the two explanatory variables.

1 A Sampling of Examples
an equation relating
output such as that shown
edicted overhead as a

icted Overhead 5 3997 1 43.54MachHrs 1 883.62ProdRuns

 effects these variables
his point but will simply indicate
s. The challenge is learning
12 on interpretation issues. ■

of dollars) for a chip-

of 2004. Are the com-

everal possible forecasting

company’s quarterly sales
ng steadily at an increasing
o estimate this curve we
icted quarterly sales as a

Predicted Sales 5 61.376e0.0663Time

approximately 6.6%
entage increase of about
in forecasting. We look at
cal data and then use one
olate them into the future.

1 A Sampling of Examples 17

■
estment Corporation
le investments,
ws for these investments is
ash must be invested at the
ns of $0.50 and $1.00 at
r investments are as fol-

e returns of $0.50 and $1.00 at

e return of $1.20 at the begin-

e return of $1.90 at the begin-

e return of $1.50 at the begin-

and that the returns are the

ortfolio, Barney-Jones
The company wants an
at the beginning of year 5.
hich includes returns from
put in a short-term money

any optimization examples we present in Chapters 14 and 15. The typical

a company such as Barney-Jones must make several decisions, subject to
s, that optimize some objective. In this case, Barney-Jones needs to decide
vest, subject to some constraints, to maximize its ending cash 4 years from

one shown in Figure 1.12,

lled “changing cells”

nter any values in these
turn it over to Excel’s
optimal values in the
hile satisfying the con-
al values. They imply that
s indicated in row 26.
 ■

k industry: “us” and

nt of view of “us.”
uarters, each company pro-
will assume that each
“blitz” level. Depending
change in our market share
meters given in Table 1.2.
uch as 1% market share,
outcome is an increase of
lows us to gauge the long-
oyed by us and them.

1 A Sampling of Examples 19

ers of Market Share Change Distributions

Most Likely Maximum
0.00 0.03
0.00 0.05
0.02 0.06
-0.02 0.01
 mber of assumptions,
ncorporate uncer-
“bottom-line” out-
in this model are
h 24. Therefore, the
io of how market
alues, we see different

t model. Then we can

te the model and keep track
ure 1.14. It shows a time
ain strategy of blitzing by

of Strategies
 ■

er. In fact, we have shown

he modeling process are
detail in this section.3
pture the essence and
tively unimportant details.
st’s preferences and skills,
e three types of models
heet models.

itative type of model. They

em are related—what affects
s called an “influence
iscussed in Chapter 7, but we

tisticians. However, many tradi-

lications of modeling.

1 Modeling and Models 21

 to decide how many
he problem is that the com-
demand, and the combi-
ompany. The diagram
agram does not provide
s problem. But this is usu-
usually to show the impor-
x problems this can be

eans of algebraic equations

ise way, and their precise-
mathematical background.
d with great generality.
hapter 14. A company
mount to profit and con-
is to select the product
he resources. All product

max ^pjxj (1.1)

j51

# bi, 1#i#m (1.2)

0 # xj # uj, 1#j#n (1.3)

on the amount of product j

is the amount of resource i
i available, n is the number
raic model states very con-
bject to consuming no
ll production quantities

analysts. They are

d need to estimate the uj’s,
ll (a problem with
as one with only 5 prod-
cept algebraic models in
is required. Indeed, alge-
ill are by many analysts.
abstract mathematical
ligent people do not.

ternative to algebraic modeling is spreadsheet modeling. Instead of relat-

ities with algebraic equations and inequalities, we relate them in a spread-
rmulas. This process is much more intuitive to most people (at least in our
ne of the primary reasons for this is the instant feedback available from
 diately obvious (from error
which you can then go back

a general algebraic
pter 14, illustrates a
preadsheet model should
ts to the model, the quanti-
xj’s in the algebraic model),
mulas. To indicate con-

E F G H I

Range names used:

Frames_produced =Model!$B$16:$E$16
Maximum_sales =Model!$B$18:$E$18
Profit =Model!$F$32
Resources_available =Model!$D$21:$D$23
4 Resources_used =Model!$B$21:$B$23

4
0
<=

4 Totals
$0 $50,200

$0 $32,000
$0 $3,000
$0 $6,000
$0 $9,200
del such as the one in
ts algebraic counterpart,
ously, they must be correct.
ct syntax, the correct cell
quite a challenge.
o be used in the busi-
d. Otherwise, no one other
d) will be able to under-
readsheets is their

1 Modeling and Models 23

are limited only by your imagination. However, this flexibility can be a

eadsheet modeling unless you plan the design of your models carefully.
ar design in Figure 1.16. Most of the inputs are grouped at the top of
. All of the financial calculations are done at the bottom. When there are
he two sides of the constraints are placed next to each other (as in the
). Borders, colors (which appear on the screen but not in this book), and
sed for added clarity. Descriptive labels are used liberally. Excel itself
s none of these “rules,” but you should impose them on yourself.
a conscious effort to establish good habits for you to follow throughout
e designed and redesigned our spreadsheet models so that they are as clear
does not mean that you have to copy everything we do—everyone tends to
spreadsheet style—but our models should give you something to emulate.
at in the business world, you typically start with a blank spreadsheet. It is
develop a model that is not only correct but is also intelligible to you and to
lot of practicing and a lot of editing, but it is a skill well worth developing.

Modeling Process
overall modeling process
ep process, as discussed
mple, the analysis of sur-
ision making), without the

evelop a model unless it

cess really begins by identify-
sing money, perhaps its market
too long for service. Any
veral people have warned [see
step is not always as straight-
that it has identified the right
g to solve it.
who was hired by the military to
ter planes landing and taking off
hat the problem was caused by
und times would decrease. The
and was about to do classical
nt the sources of their inefficiency.
em obviously lay elsewhere. The
hich in turn was due to the ineffi-
her location. Once this latter prob-
simple—the turnaround times
on the part of the ground crews. If
e problem, the real problem might

process is often the most

n their operations, but these data
also typically scattered in dif-
of different formats. Therefore,
he right data and summarize the
 1 Modeling and Models 25

e two ways to mitigate this problem. First, it is helpful to include relevant

oughout the company in the modeling process—from beginning to end—so
eryone has an understanding of the model and feels an ownership for it.
it helps to use a spreadsheet model whenever possible, especially if it is
and documented properly. Almost everyone in today’s business world is
e with spreadsheets, so spreadsheet models are more likely to be accepted.
n, there is a big difference
. When you turn in a classroom
ment and can await the next one.
mpany can usually not pack up his
nt, the company will then need to
nsuming and politically difficult,
ificant change from the past. At
he model on a day-to-day basis.

d over time, either because of

e potential uses for the model
greatest challenges for a
el that can be modified as the
throughout this book. Always
ok are important for you to
are no longer the sole
is loaded with powerful
business problems. We
his book, along with the
eling process, a theme that
weekly,and various professional singers and comedi-
ans played occasional single-night performances.4
Although this entertainment was free to all of the
passengers,much of it had embarrassingly low atten-
dance.The nightly show band and musical combos,
who were contracted to play nightly until midnight,
often had less than a half dozen people in the audi-
ence—sometimes literally none.The professional
singers,dancers,and comedians attracted larger audi-
ences,but there were still plenty of empty seats.In
spite of this,the cruise staff posted a weekly schedule,
and they stuck to it regardless of attendance.In a
short-term financial sense,it didn’t make much differ-
ence.The performers got paid the same whether any-
one was in the audience or not,the passengers had
already paid (indirectly) for the entertainment as part
of the cost of the cruise,and the only possible oppor-
tunity cost to the cruise line (in the short run) was
the loss of liquor sales from the lack of passengers in
the entertainment lounges.The morale of the enter-
tainers was not great—entertainers love packed
houses—but they usually argued,philosophically,that
their hours were relatively short and they were still
getting paid to see the world.
If you were in charge of entertainment on this
ship,how would you describe the problem with enter-
tainment:Is it a problem with deadbeat passengers,
low-quality entertainment,or a mismatch between the
entertainment offered and the entertainment desired?
How might you try to solve the problem? What con-
straints might you have to work within? Would you
keep a strict schedule such as the one followed by this
cruise director,or would you play it more “by ear”?
Would you gather data to help solve the problem?
What data would you gather? How much would finan-
cial considerations dictate your decisions? Would they
be long-term or short-term considerations? ■

4There was also a moderately large onboard casino, but it tended to

attract the same people every night, and it was always closed when
the ship was in port.

Case 1.1 Entertainment on a Cruise Ship 27

A Sampling of Examples 15