Introduction to

Decision Analysis
H
ave you ever had a difficult decision to make? If so, did you wish for a straight-
forward way to keep all of the different issues clear? Did you end up making
the decision based on your intuiti on or on a "hunch" that seemed correct? At one
time or anot her, all of us have wished that a hard deci sion was easy to make. The sad
fact is that hard deci sions are just that-hard. As individuals we run into such diffi -
cult decisions frequent ly_ Business executives and governmental policy makers
struggle with hard problems all the time. For example, consider the following prob-
lem faced by the Oregon Department of Agriculture (ODA) in 1985.
GYPSY MOTHS AND THE ODA
In the winter of 1985, the ODA grappled with the problem of gypsy moth infeslal ion
in Lane County in western Oregon. Forest industry representali ves argued strongly
for an aggressive eradication campaign using potent chemical insecticides. The ODA
instead proposed a plan that involved sprayi ng mosL of the affected area with BT
(Bacillus thuringiensis), a bacterial insecticide known to be (1) target-spec ifi c (that
is, it does little damage to organi sms other than moths), (2) ecologicall y safe, and (3)
reasonably effect ive. As well as using BT, the ODA proposed sprayi ng three smaller
areas near the city of Eugene with the chemical spray Orthene. Although Orthene
. was registered as an acceptable insecticide for home garden use, there was some
doubt as to its ul timate ecological effects as well as its danger to humans. Forestry
officials argued that the chemi cal insecticide was more potent than BT and was nec-
2 CHAPTER 1 INTRODUCTION TO DECISION ANALYSIS
essary to ensure eradication in the most heavily infested areas. Environmentali sts ar-
gued that the potential danger from the chemi cal spray was too great to warranl its
li se. Some individuals argued that spraying would not help because the infestation
already was so advanced that no program would be successful. Others argued that an
aggressive spray program could solve the problem once and for all , but only if done
immediately. Clearl y, in making its final decision the ODA would have to deal with
many issues.
The ODA has an extremely complex problem on its hands. Before deciding ex-
act ly what course of action to take, the agency needs to consider many issues, in-
cl uding the values of different constituent groups and the uncertai nti es involving the
effectiveness and risks of the pest icides under consideration. The aDA must con-
sider these issues carefull y and in a balanced way-but how? There is no escaping
the problem: This hard decision requires hard thinki ng.
Decision analysis provides structure and guidance for thinking systematicall y
about hard decisions. With decision analysis, a decision maker can take action with
confidence gained through a cl ear understanding of the problem. Along wi th a con-
ceptual framework for thinking about hard problems, decision anal ysis provides an-
alytical tools that can make the required hard thinking easier.
Why Are Decisions Hard?
What makes decisions hard? Certainly different problems may involve different and
often special difficulti es. For example, the ODA's problem requires it to think about
the interests of various groups as well as to consider onl y limited information on the
possible effects of the sprays. Although every decision may have its own special
probl ems, there are four basic sources of difficulty. A decision-analysis approach call
help a decision maker with all four.
First, a decision can be hard simply because of its compl exity. In the case of the
gypsy moths, the ODA must consider many different individual issues: the uncer-
tainty surrounding the different sprays, the values held by different community
groups, the different possible courses of action, the economic impact of any pest-
control program, and so 011. Simply keeping all of the issues in mind at one time is
nearly impossible. Decision analysis provides effective methods for organi zing a
complex probl em into a structure that can be anal yzed. In pru1icular, elements of a de-
cision's structure include the possible courses of action, the possible outcomes that
could result, the likelihood of those outcomes, and eventual consequences (e.g., costs
and benefit s) to be derived from the different outcomes. Structuring tools that we will
consider include decision trees and influence diagrams as well as procedures for ana-
lyzing these structures to find solutions and for answering "what if" questi ons.
Second, a decision can be di ffic ult because of the inherent uncertainty in lhe situa-
tion. In the gypsy moth case, (he major uncertainties are the effectiveness of the differ-
ent sprays in reducing the moth population and their potential for detrimental ecologi-
cal and health effect s. In some decisions the main issue is uncertainty. For example,
"
WHY STUDY DECISION ANALYSIS?
3
imagine a finn trying to decide whether to introduce a new product. The size of the
market, the market price, eventual competiti on, and manufacturing and distributi on
casts all may be uncertain to some extent, and all have some impact on the fi nn' s even-
payoff. Yet the decision must be made without knowing for sure what these lIncer-
tam values will be. A decision-analysis approach can help in ident ifying important
of and representing that uncellainty in a systematic and useful way.
Thud, a d.eclslOn may be interested in working toward multiple objectives,
but progress 111 one direction may impede progress in others. In such a case a deci-
sion maker trade off benefit.s in one area against costs in another. In th'e gypsy
mot h example, Important trade-ofts must be made: Are the potential economic bene-
fit s to from spraying Orthene worth the potential ecological damage and
health n sk? In I.Ovestment decisions a trade-off that we uSllal1y must make is be-
tween expected return and riski ness. Decision analysis again provides both a frame-
work and specific tools for dealing with mult ipl e objecti ves.
. Fourth, and fInall y, a problem may be difficult if different perspectives lead to
?lfferent conclUSIOns. Or, even from a single perspective, slight changes in certain
mputs may lead to different choices. This SOurce of difficulty is particularl y pertinent
when more than one person is involved in making the decision. Different individuals
may at the problem from different perspect ives, or they may disagree 0 11 the Ull -
certaillty or value of the varioliS olltcomes. The lise of the decision-analysis frame-
work and tools can help SOil through and resolve these differences whether the deci-
sion maker is an indi vidual or a group of stakeholders with diverse opinions.
Why Study Decision Analysis?
obvious reason for studyi ng decision analysis is that carefully applying its tech-
can lea.d better deci si ons. But what is a good deci sion? A simple answer
mIght be 1l IS the one that gives the best outcome. Thi s answer, however, con-
fuses Idea .of a lucky outcome wi th a good decisi on. Suppose that you are inter-
ested mlnvestmg an inheritance. After carefully considering all the options avai lable
and with investment specialist s and financial planners, you decide to in-
In stocks. Tf you purchased a portfolio of stocks in 1982, the investment most
tllmed out to be a good one, because stock values increased dramaticall y dur-
lllg the 1980s. On the other hand, if your stock purchase had been in earl y 1929, the
stock market crash and the followi ng depression would have decreased the value of
your portfolio drasticall y.
Was the investment decision a good one? It certainly could have been if it was made
after careful consideration of the avail able informati on and thorough deli beration about
the goals and possible outcomes. Was the outcome a good one? For the 1929 investor
the answer is no. This example illustrates the difference between a good decision and
lucky outcome: You can make a good decision but still have an unlucky outcome. Of
course, you may prefer to have lucky outcomes rather than make good decisions!
Although decision analysis cannot improve your luck, it can help you to understand
better the problems you face and thus make better decisions. That understanding must
4
CHAPTER 1 INTRODUCTION TO DECI SION ANALYSIS
include the structure of the probl em as well as the uncertainty and trade-off's inherent in
the altclllati ves and outcomes. You may then improve your chances of enjoying a beHer
outcome; more impol1ant, you will be less likely to experience unpl easant surprises in
the form of unlucky outcomes that were either unforeseen or not full y understood. in
other words, you wilJ be making a decision with your eyes open.
The preceding di scussion suggests that decision analysis allows people to make
effective decisions more consistentl y. Thi s idea itself wanants di scussion. Decision
analysis is intended to help people deal with difficult decisions. It is a "prescripti ve
approach designed for normall y intell igent people who want to Ihink hard and .sys-
temati cally abont some importanl real problems" (Keeney and R",ffa 1976, p. Vtl).
Thi s prescriptive view is the most appropriate way to think about decision analy-
sis. It gets acrosS the idea that although we are not perfect decision we ,ca,n
do better through more structure and guidance. We will see that decISIOn analysIs IS
not an idealized theory designed for superrational and omni scient beings. Nor does it
describe how people actually make decisions. In fac t, ample experimental evidence
from psychology shows that people generall y do not process informati on and
decisions in ways that are consistent with the decision-analysis approach. (It they
did then there would be no need for decision analysis; why spend a lot of lime
studying decision analysis if it suggests that you do what you already do?) Instead,
using some fundament al principl es, and informed by what we know about human
frailti es in judgment and deci sion making, decision analysis offers guidance to nor-
mal people working on hard deci sions.
Although decision analysis provides structure and guidance for systematic thinking
in difficult situations, it does not claim to recommend an al ternative that must be blindly
accepted. Indeed, afler the hard thinking thai decision analysis foslers, there should be
no need for blind acceptance; the decision maker should understand the situation thor-
oughly. Instead of providing solutions, decision analysis isy erhaps best 0: as
simply an infonnation source, providing insight about the situation,
tives, and trade-offs, and possibly yielding a reconunended course of actloll . Thus, deCI-
sion analysis does not usurp the decision maker's job. According to another author,
The basic presumption of decision analysis is not at all to replace the decision
maker's intui tion, to relieve him or her of the obli gations in facing the problem,
or to be, worst of all , a competitor to the decision maker's personal style of
analysis, but to complement, augment, and generally .work the decision
maker in exemplifyi ng the nature of the problem. Ultimately, It IS 01 most value
if the decision maker has actually learned something about the problem and hi s
or her own decision-making attitude through the exercise (Bunn 1984, p. 8).
We have been di scussi ng decision analysis as if it were al ways used to help an in-
dividual make a decision. Indeed, this is what it is designed for, but its techniques
have many other uses. For example. one might use decision-anal ysis methods to solve
complicated inference problems (that is, answering questions such as "What conclu-
sions can be drawn from the available evidence?"). Structuring a decision problem
may be useful for understanding its precise nature, for generating alternative
of action, and for identifying important objectives and trade-offs. Understanding
u'ade-offs can be crucial for making progress in negotiation setti ngs. Finally, decision
analysis can be lI sed to justify why a previously chosen action was appropriate.
THE DECISION-ANALYSIS PROCESS 5
Subjective Judgments and Decision Making
Personal judgments about uncertainty and values are important inputs for decision
analysis. It will become clear through thi s text that di scovering and developing these
judgment s involves thinking hard and systematicall y about important aspects of a
deci sion.
Managers and policy makers frequentl y compl ain that analytical procedures
from management science and operations research ignore subj ective judgments.
Such procedures orten purport to generate "optimal " actions on the basis of purely
obj ective inputs. But the deci sion-analysis approach allows the inclusion of subj ec-
tive judgments. rn fact, decision analysis requires personal judgments; they are im-
portant ingredients for making good decisions.
At the same time. it is important to reali ze that human beings are imperfect in-
formation processors. Personal insights about uncertainty and preferences can be
both limited and mi sleading, even while the indi vidual making the judgments may
demonstrate an amazing overconfidence. An awareness of human cognitive limita-
tions is critical in developing the necessary judgmental inputs, and a decision maker
who ignores these problems can magnify rather than adjust for human frail ties.
Much CUlTen! psychological research has a direct bearing on the practice of deci-
sion-analysis techniques. In the chapters that follow, many of the results from thi s
research will be di scussed and related to deci sion-analysis techniques . The spirit of
the discussion is Ihat understanding the problems peopl e face and carefull y apply-
decisi on-anal ysis techniques can lead to better j udgments and improved deci-
SIOns.
The Decision-Analysis Process
Figure 1.] shows a flowchart for the decision-analysis process. The first step is for
the deci sion maker to identify the decision situati on and to understand hi s or her ob-
jecti ves in that situation. Although we usually do not have troubl e finding deci sions
to make or problems to solve. we do sometimes have trouble identifying the exact
probl em, and thus we sometimes treat the wrong problem. Such a mi stake has been
call ed an "error of the third kind." Careful identifi cation of the decision at hand is al-
ways important. For example, perhaps a surface problem hides the real issue. For ex-
ample. in the gypsy moth case, is the decision which insecticide to lise to control the
insects, or is it how to mollify a vocal and ecologically minded minority?
Understanding one's objectives in a decision situati on is also an important first
step and involves some introspection. What is important? What are the objectives?
. Minimi zing cost? Maximizing profit or market share? What about minimi zing
ri sks? Does risk mean the chance of a monetary loss, or does it refer to conditions
potenliall y damaging to health and the environment ? Getting a clear understanding
of the crucial objectives in a decision situation must be done before much more can
be accompli shed. In Ihe next step, knowledge of objectives can help in identifying
6
CHAPTER 1 INTRODUCTION TO DECISION ANALYSIS
Figure 1.1
A decision-analysis
process flowchart.
Identify the decision
situation and
understand objectives.
Decompose and
model the problem:
[. Model of problem
structure.
2. Model of
uncertaint y.
3. Model of
preferences.
Is further Yes
analysis needed?
alternatives, and beyond that the objeclives indicate how outcomes must be mea-
sured and what kinds of uncertainties should be considered in the analysis.
Many authors argue that Ihe first thing to do is to identify the problem and then 10
figure out the appropri ate objectives to be used in addressing the problem. But Keeney
(1992) argues the opposite; it is far better, he claims, to spend a lot of effort under-
standing one's central values and objectives, and then looking for ways-decision op-
pOltunities-to achieve those objectives. The debate notwithstanding, the fact is that
decisions come in many fOlms. Sometimes we are lucky enough to shape our decision-
making future in the way Keeney suggests, and other times we find ourselves in diffi-
THE DECISIONANALYSIS PROCESS 7
cult situat ions that we Illay not have anticipated. Tn either case, establi shing the precise
nature of the deci sion situation (which we will later caU the decision context) goes
hand in hand with identifying and understanding one's objectives in that situation.
With the deci sion situation and pertinent objectives establi shed, we nlfIlto the di s-
covery and creation of alternati ves. Often a careful examination and analysis of objec-
tives can reveal altelllat ives that were not obvious at the outsel. This is an impOitant ben-
efi t of a decision-analysis approach. rn addition, research in the area of creativi ty has led
to a number of techniques that can improve the chance of tinding new alternatives.
The next two steps, whi ch mi ght be called "modeling and solution," form the he8J1
of most textbooks 0 11 decision analysis, incl uding this one. Much of this book will
focus on decomposi ng problems to understand their structures and measure uncer-
tainty and value; indeed, decompositi on is the key to decision analysis. The approach
is to "divide and conquer." The first level of decomposition call s for stmcturing the
probJem in smaller and more manageable pieces. Subsequent decomposition by the de-
cision maker may entail careful consideration of elements of uncertainty in different
parts of the problem or careful thought about different aspects of the objectives.
The idea of modeling is cri tical in decision analysis, as it is in most quantitative
or anal ytical approaches to problems. As indi cated in Figure 1.1, we will use model s
in several ways. We will use influence diagrams or decision trees to create a repre-
sentation or model of the decision problem. Probability will be ll sed to build model s
of the uncertainty inherent in the problem. Hi erarchical and nctwork models will be
used to understand the relationshi ps among multiple objcctives, and we will assess
utility functions in order to model the way in whi ch decision makers val ue different
outcomes and trade off competing objectives. These models are mathematical and
graphical in nat ure, allowing one to find insights that may not be apparent on the sur-
face. Of course, a key advantage from a decision-making perspective is that the
mathematical representation ofa decision can be subjected to analysis, which can in-
di cate a "preferred" alternati ve.
Decis ion analysis is typicalJy an iterative process. Once a model has been built ,
sensitivity lInalysis is perfOlmed. Such analysis answers "what if" questions: "If we
make a sl ight change in one or more aspects of the model , does the opti mal deci sion
change?" If so, the decision is said to be sensitive to these small changes, and the de-
cision maker may wish to reconsider more carefull y those aspects to which the deci -
sion is sensiti ve. Virtually any part of a decision is fair game for sensitivity analysi s.
The arrows in Figure J.l show that the decision maker may return even to the identi -
fication of the problem. It may be necessary to refine the definition of objectives or
include obj ectives that were not previously included in the model. New alternatives
may be identified, the model structure may change, and the models of uncertainty
and preferences may need to be refined. The term cycle best de-
scribes the overall process, which may go through several iterat ions before a sati s-
factory soluti on is found.
In thi s iterative process, the decision maker's percepti on of the problem changes,
. beliefs about the li kelihood of various uncertain eventualiti es may develop and
change, and preferences for outcomes not previously considered may mature as
more time is spent in retl ection. Decision analysis not only provides a struct ured way
to think about decisions, but also more fundamentall y provides a structure within
8
CHAPTER 1 INTRODUCTION TO DECISION ANALYSIS
which a decision maker can develop beliefs and fee lings, those subjective judgments
that are critical for a good solution.
Requisite Decision Models
Phillips (1982, 1984) has introduced the term requisite decision modeling. Thi s mar-
velous term captures the essence of the modeling process in decision analysis. In
Phillips's words, "a model can be considered requisite onl y when DO new inwi tions
emerge about the problem" (1984, p. 37), or when it contains everything that is essen-
ti al for solving the problem. That is, a model is requisite when the deci sion maker's
thoughts about the problem, beliefs regarding ullcel1ainty, and preferences arc fully de-
veloped. For example, consider a firsHime mutual-fund investor who finds hi gh, over-
all long- term returns appealing. Imagine, though, that in the process of researching the
funds the investor begins to understand and become wary of highly volatile stocks and
mutual funds . For this investor, a deci sion model that selected a fund by maximizing
the average return in the long run would not be requisite. A requisite model would have
to incorporate a trade-off between long-term retullls and volati lity.
A careful decision maker may cycle through the process shown in Figure 1. 1 sev-
eral times as the analysis is refined. Sensit ivity analysis at appropriate times can help
the decision maker choose the next modeling steps to take in developing a requisite
model. Successful decision analysts artistically use sensitivity analysis to manage
the iterative development of a decision model. An important goal of this book is that
you begin to acquire this artistic ability through familiarity and practice with the
concepts and tools of deci sion analysis.
Where Is Decision Analysis Used?
Decision analysis is widely llsed in bus iness and government decision making.
Perusing the literature reveals appli cations that include managing research-and-
development programs, negot iating for oil and gas leases, forecasting sales for new
products, understanding the world oil market, deciding whether to launch a new
product or new venture, and developing ways to respond to environmental risks, to
name a few. And some of the largest firms make use of decision analysis, including
General Motors, Chevron, and Eli Lill y. A palticul arly important arena for decision-
analysis applications has been in publi c utilities. especialJ y electric power genera-
tion. In part this is because the problems utiliti es face (e.g., si te selection, power-
generation methods, waste cleanup and storage, pollution control) are particularly
appropriate for treatment with decision-analysis techniques; they involve long time
frames and hence a hi gh degree of uncertainty. In addition, mUltipl e obj ectives mList
be considered when a decision affects many different stakeholder groups.
In the literature, many of the reported appli cati ons relate to public-poli cy prob-
lems and relatively few to commercial decisions, partly because public-policy prob-
lems are of interest to such a wide audience. It is perhaps more closely related to the
WHERE DOES THE SOFTWARE fiT IN? 9
fact that commercial applications often are proprietary; a good decision analysis
can create a competiti ve advantage for a finn. which may not appreciate havi ng its
advantage revealed in the open literature. Important public-policy applications have
included regulat ion in the energy (espec ially nucl ear) industry and standard setting
in a variety of different situat ions ranging from regul ations for air and water poll u-
tion to standards for safety features on new cars.
Another impOItant area of application for decision analysis has been in medi cine.
Deci sion analysis has helped doctors make speciti c diagnoses and individual s to un-
derstand the risks of different treatments. Institutional -level studi es have been done,
sLich as studying the optimal inventory or usage of blood in a blood bank or the deci-
sion of a firm regarding different kinds of medi cal insurance to provide its employees.
On a grander scale, studies have examined poticies such as widespread testing for var-
ious forms of cancer or the impact on society of different treatment recommendations.
This discussion is by no means exhaust ive; the intent is only to give you a feel for
the breadth of possible applicat ions of decision analysis and a glimpse at some of the
things that have been done. Many other applicat ions are described in cases and ex-
amples throughout the book; by the time you have fin ished, you should have a good
understanding of how decision analysis can be (and is) used in many different are-
nas. And if YOll feel the need for more, articles by Ulvil a and Brown (1982) and
Corner and Kirkwood ( 199 l) describe many differen t appl ications.
Where Does the Software Fit In?
Included with the text is a CD containing Pali sade' s DecisionTools suite, whi ch is a
set of computer programs designed La help you complete the modeling and solution
phase of the decision process. The suite consists of five programs (PrecisionTree,
RISKview, BestFit, TopRank, and @RISK), each intended for different steps in the
decision process. As you work your way through the text learning the different steps,
we introduce the programs that will heJp you complete each step. We supply detailed
instructions on how to use the program and how to interpret the output at the end of
certain chapters. Table 1. 1 shows where in the decision process each of the five pro-
grams is used and the chapter where the instructi ons appear.
One of the best aspects about the DecisionTools suite is that the programs work
together as one program wi thin Excel. When either Precision Tree, TopRank, or
@RISK are opened, their functions are added directly to Excel's toolbar. Thi s allows
us to model and solve complex decision probl ems within an ordinary spreadsheet.
These programs are designed to extend the capability of the spreadsheet to handle
the types of models used in decision maki ng. If you are already familiar with Excel,
then you will not need to learn a completely different system to use Preci sionTree,
TopRank, and @RISK.
The programs RISKview and BestFit can be run either as independent programs
or as components of@RISK. They are specialized programs specificall y designed to
help the decision maker determine how best to model an uncertainty. The output
10 CHAPTER 1 INTRODUCTION TO DECISION ANALYSIS
Table 1.1
The DccisionTools
Programs
DecisionTool s
Program
PrecisionTree
TopRank
RI SKview
BestFit
@RISK
Where It Is Used in the
Decision Process
Structuring the decision
Solving the decision
Sensitivity analysis
Value of information
Modeling preferences
Sensitivity analysis
Modeling uncertainty
Using data to model uncertainty
Simulation modeling
Where in Text
Chapter 3
Chapter 4
Chapter 5
Chapter 12
Chapter 13
Chapter 5
Chapters 8 and 9
Chapter 10
Chapter I [
from these two programs is lIsed as input to the decision model. Although these pro-
grams do not operate within Excel, they do export their output to Excello be used by
the other programs in the suite. As you become more hlmiliar with the theory and the
DecisionTools suite, you will see how the five programs are interrelated and form an
organized whol e.
There is also a strong interconnection between the DecisionTools programs and
Excel. These programs do more than use the spreadsheet as an analytical engine. For
example, you can link your decision tree to a spreadsheet model. The links wi ll be
dynamic; changes made in the spreadsheet are immediately reflected in the decis ion
tree. These dynamic links will pull input val ues from the tree to the spreadsheet, cal-
culate an output value, and send the output back to the tree.
Linking is one of many ways that Excel and the DecisionTools programs work
together. In future chapters we wiLl see other connections. We will also see how flex-
ible and useful electronic spreadsheets can be when constructing decision models.
Much of what we will learn about usi ng Excel will extend beyond decision analysis.
It is hard to overemphasize the power of modern spreadsheets; these programs can
do virtuall y anything that requires calculations. Spreadsheets have become one of
the most versat ile and powerful quantitative tool s available to business managers.
Virtuall y all managers now have personal computers on their desks with the ability
to run these sophi st icated spreadsheet programs, which suggests that aspiri.ng man-
agers would be well advised to become profi cient in the use of this flexible tooL
The software can help you learn and understand the concepts presented in the
chapter. Reading about the concepts and examples will provide a theoretical under-
standing. The programs will test the understanding when you apply the theory to ac-
tual probl ems. Because the program will carry out your instructions exactly as you
input them, it will refl ect how well you understand the theory. You will find that your
understanding of the concepts greatly increases because the programs force you to
think carefull y throughout the constructi on and analysis of the model.
SUMMARY 11
Where Are We Going from Here?
S UMMARY
Thi s book is divided into three main sections. The first is titled "Modeling Decisions,"
and it introduces influence diagrams and decision trees as methods for building models
of decision problems. The process is sometimes called structuring because it specifi es
the elements of the decision and how the elements are interrelated (Chapters 2 and 3).
We also introduce ways to organize a dec ision maker's values into hierarchies and net-
works; doing so is useful when multipl e objectives must be considered. We will find
out how to analyze our decision models (Chapter 4) and how to conduct sensiti vity
analysis (Chapter 5). In Chapter 6 we discuss creativity and decision making.
The second section is "Modeling Uncertainty." Here we del ve into the use of
probability for modeling uncertai nty in decision problems. First we review basic
probability concepts (Chapter 7). Because subjective judgments playa central role in
decision analysis. subj ecti ve assessments of uncertainty are the topic of Chapter 8.
Other ways to use probability include theoretical probability models (Chapter 9),
dala-based models (Chapter 10), and simulation (Chapter II ). Chapter 12 closes the
section with a discussion of information and how to val ue it in the context of a prob-
ability model of uncertainty withi n a decision problem.
"Modeling Preferences" is the fmal section. Here we fUm to the development of a
mathematical representation of a decision maker's preferences, includi.ng the identifica-
tion of desirable objectives and trade-oft-s between conflicting objectives. A fundamen-
tal issue that we often must conti-ont is how to trade off riskiness and ex pected value.
Typically, if we want to increase our chances at a better outcome, we must accept a
simultaneous lisk of loss. Chapters 13 and 14 delve into the problem of modeling a de-
cision maker's attitude toward risk. Chapters 15 and 16 complete the section with a
treatment of other conflicting objectives. In these chapters we will complete the discus-
sion of multiple objectives begun in Section 1, showing how to construct a mathemati-
cal model that reflects subjective judgments of relative importance among competing
objectives.
By the end of the book, you will have learned all of the basic techniques and con-
cepts that are central to the practice of modern decision analysis. Thi s does not mean
that your hard decisions will suddenl y become easy! But with the decision-analysis
framework, and with tools for modeling decisions, uncertai nty, and preferences, you
will be able to approach your hard decisions systematicall y. The understanding and
insight gained from such an approach wi ll give you confidence in your act ions and
allow for better decisions in difficult situations. That is what the book is about-an
approach that will help you to make hard decisions.
The purpose of decision analysis is to help a decision maker think systemati call y
about complex problems and to improve the quality of the resulting decisions. In this
regard, it is important to distinguish between a good decision and a lucky outcome.
A good decision is one that is made on the basis of a thorough understanding of the
12
CHAPTER 1 INTRODUCTION TO DECI SION ANALYSIS
problem and careful thought regarding the important issues. Outcomes, on the other
hand, may be lucky or unl ucky, regardl ess of decision quality.
In general, decision analysis consists of a framework and a tool kit for dealing
with diffi cult decisions. The incorporati on of subjective j udgments is an important
aspect of decision analysis, and to a great extent mature judgments develop as the
decision maker reflects on the decision at hand and develops a working model of the
problem. The overall strategy is to decompose a complicated probl em into smaller
chunks that can be more readiJ y analyzed and understood. These small er pi eces can
then can be brought together to create an overall representati on of the decision situa-
tion. Finall y, the decision-analysis cycle provides the framework within which a de-
ci sion maker can construct a requi site decision model, one that contains the essenti al
elements of the problem and from which the decision maker can take action.
QUESTIO N S AN D PROB L EM S
1. 1 Give an example of a good decision that you made in the face of some uncei1ainty. Was
the outcome lucky or unlucky? Can you give an example of a poorly made decision
whose outcome was lucky?
1.2 Explain how modeling is used in decision analysis. What is meant by the term "requisite
decision model"?
1.3 What role do subjective judgments play in decision analysis?
1.4 At a dinner party, an acquaintance asks whether you have read anything interesting lately,
and you mention that you have begun to read a text on decisi on anal ysis. Your.friend asks
what decision analysis is and why anyone would want to read a book about It, let alone
write one! How would you answer?
1.5 Your fri end in Question 1.4, upon hearing your answer, is deli ghted! "Thi s is marvelous,"
she exclaims. "I have thi s very difficult choice to make at work. I'll tell you the facts, and
you can tell me what I should do!" Explain to her why you cannot do the analysis for her,
1.6 Give an exampl e in which a decision was compli cated because of di fficult preference
trade-offs, Give one that was complicated by uncertainly.
1.7 In the gypsy moth example, what are some of the issues that you woul d consider in making
this decision? What are the altemat ive courses of action? What issues involve uncertainty,
and how could you get infOlmation to help resolve that uncertainty? What are the values
held by opposing groups? How might your decision trade off these values?
1.8 Can you think of some different alternatives that the ODA might consider for controll ing
the gypsy moths?
1.9 Describe a decision that you have had to make recently that was difficult. What were the
major issues? What were your alternatives? Did you have to deal with uncertainty? Were
there important trade-offs to make?
1.10 "Socially responsible investing" first became fashionabl e in the 198,Os. Such
volves consideration of the kinds of businesses that a fum engages m and selection of 111-
vestments that are as consistent as possible with the investor' s sense of ethical and moral
business activity. What trade-offs must the sociall y responsible investor make? How are
CASE STUDI ES 13
these trade-offs more complicated than those that we normally consider in maki ng in-
vestment decisions?
1. 11 Many decisions are simpl e, preprogrammed, or already solved, For example, retailers do
not have to think long to decide how to deal with a new customer. Some operations-
research models provide "ready-made" decisions, such as fi ndi ng an optimal inventory
level using an order-quantity formula or determining an optimal production mi x using lin-
ear programming. Contrast these decisions with unstructured or sb'ategic decisions, such as
choosing a career or locating a nuclear power pl ant. What kinds of decisions are appropri-
ate for a decision-analysis approach? COl1lillent on the statement, "Decision making is what
you do when YOli don' t know what to do." (For marc discll ssion, sec Howard 1980,)
1. 12 The argument was made that beliefs and preferences can change as we explore and learn,
Thi s even holds for learning about decision analysis! For example, what was your im-
pression of thi s book before reading the fi rst chapter? Have your beliefs about the value
of decision analysis changed? How nught tlus affect your decision about reading more of
Ihe book?
CASE STUDIES
DR . /O YCELYN ELD ERS AND THE W A R ON D RU GS
After the Nancy Reagan slogan, "Just Say No," and 12 years of Republi can adminis-
n'ation efforts to fi ght ill egal drug use and trafficki ng, on December 7, 1993, then
Surgeon General Dr. Joycelyn Elders made a startling statement. In response to a re-
porter's questi on, she indicated that, based on the experiences of other countri es, the
cri me rate in the United States mi ght actualJ y decrease if drugs were legalized, She
conceded thai she did not know all of Ihe ramifi cati ons and suggested that perhaps
some studi es should be done.
The nation and especi all y the Clinton admini stration were shocked to hear thi s
statement. What heresy after all the efforts to control illegal drugs ! Of course, the White
House immediately went on the defensive, making sure that everyone understood that
President Clinton was not in favor of legalizing drugs. And Dr. Elders had to cl arify her
statement; it was her personal opini on, not a statement of administration policy,
Questions
1 What decision situati on did Dr. Elders identi fy? What speci fi c values would be im-
plied by choosing to study the legali zation of drugs?
2 From a decision-making perspective, which makes more sense: Nancy Reagan's
"Just Say No" poli cy or Elders's suggesti on that the issue of legalization be stud-
ied? Why?
. 3 Consider Elders's deci sion to suggest studying the legalization of drugs, Was her
decision to respond to the reporter the way she did a good decision with a bad out-
come? Or was it a bad decision in the fi rst place?
14 CHAPTER 1 INTRODUCTION TO DECISION ANALYSIS
4 Why was Elders's suggestion a politi cal hot potato for Clinton's admini stration?
What, if any, are the impl icalions for dec ision analysis in politi cal situations?
LLOYD BENTSEN FOR VICE PRESIDENT?
In the summer of 1988, Michael Dukakis was the Democratic Party's presidenti al
nominee. The son of Greek immi grants, hi s political career had flouri shed as gover-
nor of Massachusetts, where he had demonstrated excel1 ent administrati ve and ti scal
skill s. He chose Lloyd Bentsen, U.S. Senator from Texas, as his running mate. In an
analysis of Dukaki s's choice, E. J. Di onne of The New York Times (July l3, 1988)
made the following point s:
1 The main job of the vice presidenti al nominee is to carry his or her home slate.
Could Bent sen carry Texas? The Republican presidential nominee was George
Bush, whose own adopted state was Texas. Many people thought that Texas
would be very dimcult for Dukakis to win, even with Bentsen's help. If Dukaki s
could wi n Texas's 29 electoral votes, however, the gamble would payoff dramat-
icall y, depriving Bush of one of the largest states that he might have taken for
gran led.
2 Bentsen was a conservative Democrat. Jesse Jackson had run a strong race and
had assembled a slrang following of liberal voters. Would the Jackson supporters
be disappointed in Dukakis's choice? Or would they ultimately come back to the
fold and be hlithful to the Democratic Party?
3 Bentsen' s ties with bi g business were unusual for a Democratic nominee. Would
Democratic voters accept him? The other side of this gambl e was that Bentsen
was one of the best fund raisers around and might be able to eliminate or even re-
verse the Republicans' traditional financi al advantage. Even if some of the more
liberal voters were di senchanted, Be nt sen could appeal to a more business-
oriented constituency.
4 The safer choice for a runni ng mate would have been Senator John Glenn from
Ohio. The polls suggested thai with Glenn as hi s running mate, Dukaki s would
have no troubl e winning Ohio and it s 23 electoral votes.
Questions
1
2
3
4
Why is choosing a running mate a hard deci sion?
What objectives do you think a presidenti al nominee should consi der in making the
choi ce?
What elements of risk are involved?
The title of Di onne's article was "Bentsen: Bold Choice or Risky Gamble?" In
what sense was Dukaki s's decision a "bold choice," and in what sense was it a
"risky gamble"?
'I
REFERENCES 1S
DUPONT AND CHLOROFLUOROCARBONS
REFERE NCES
Chlorofluorocarbons (CFCs) are chemicals used as refri gerants in air conditioners
and ot her cooling appliances, propellant s in aerosol sprays, and in a v3.1iety of other
applications. Scientifi c evidence has been accumulating for some time that CFCs re-
leased into the at mosphere can destroy ozone molecules in the ozone layer 15 miles
above the earth 's surface. Thi s layer shields the earth from dangerous ultraviolet ra-
diati on. A large hol e in the ozone layer above Antarctica has been found and attrib-
uted 10 CFCs, and a J 988 reporl by 100 scienli sts concluded Ihat the ozone shield
above the mid-Northern Hemisphere had shrunk by as much as 3% since 1969.
Moreover, depletion of the ozone layer appears to be irreversible. Fut1her destruction
of the ozone layer could lead to crop failures, damage to marine ecology, and possi-
bly dramatic changes in global weather patterns.
Environmentalists estimate that approximately 30% of the CFCs released into
the atmosphere come from aerosols. In 1978, the U.S. government banned their use
as aerosol propellants, but many forei gn governments sti ll permit them.
Some $2.5 billion of CFCs are sold each year, and DuPont Chemical Corporation
is responsible for 25% of Ihal amount. III early 1988, DuPonl announced that the
company would gradually phase out its product ion of CFCs and that replacements
would be developed. Already DuPont claims to have a CFC substitute for automobil e
air conditi oners, although the new substance is more expensive.
Quest ions
Imagine that you are a DuPont executive charged with making the decision regarding
continued production of CFCs.
1 What issues would you take into account?
2 What major sources of uncerta.inty do you face?
3 Whal corporate objectives would be important for you to consider? Do you think
that DuPont's corporate objectives and the way the company views the problem
mi ght have evolved since the mid- I 970s when CFCs were just beginni ng to be-
come an issue?
SOl/l'ces: "A Gaping Hole in the Sky," Nell',fll'eek, July I J, 1988, pp. 21-23; A. M. Louis (1988),
"DuPont to Ban Products That Hann Ozone," Sail Fl'lIlIcisco Chl'ollic/e. March 25, p. I.
The decision-analysis view is distinctl y prescriprille. That is, decision analysis is interested
in helping people make better decisions; in contrast, a descriptive view of decision making
focll ses on how people acnmlly make decisions. Keeney and Raiffa (1976) explain the pre-
scriptive view as well as anyone. For an excellent summary of the descriptive approach, see
Hogat1h ( 1987). Bell , Raiffa, and Tversky (1988) provide many readings on these topics.
16 CHAPTER 1 INTRODUCTION TO DECISION ANALYSIS
A fundamental element of the prescripti ve approach is discerning and accepting the
difference between a good decision and a lucky outcome. Thi s issue has been discussed
by many authors, bOlh academics and practitioners. An excellent recent reference is Vlek
et al. ( 1984).
Many other books and articles describe the process, and each
seems to have its own twist. Thi s chapter has drawn heavi ly from Ron Howard's
thoughts; his 1988 arli cle summarizes hi s approach. Other books worth consulting in-
clude Behn and Vaupel ( 1982), Bunn (1984), Holloway (1979), Keeney (1992), Lindley
( 1985), Raiffa ( 1968), Samson (1988), and von Winterfeldt and Edwards (1986).
PhiHips 's (1982, 1984) idea of a requi site decision model is a fu ndamental concept
that we wiD use throughout the text. For a related view, see Watson and Buede (1987).
Behn, R. D. , and 1. D. Vaupel (1982) Quick Analysis for Busy Decision Makers. New
York: Basic Books.
Bell , D. , H. Raiffa, and A. Tversky (1988) Decision Making: Descriptive, Normative,
and Prescriptive Interactiolls. Cambridge, MA: Cambridge Universi ty Press.
Bunn, D. (1984) Applied Decision Analysis. New York: McGraw- Hill.
Corner, 1. L. , and C. W. Kirkwood (1991) "Decision Analysis Applications in the
Operations Research Literature, 1970- 1989." Operations Research, 39,206-219.
Hogarth, R. (1987) JI/dgement and Choice, 2nd ed. New York: Wi ley.
Holl oway, C. A. ( 1979) Decision Making under Uncertainty: Models and Choices.
Englewood Cliffs, NJ: Prentice-Hall.
Howard, R. A. ( 1980) "An Assessment of Decision Analysis." Operations Research, 28,
4-27.
Howard, R. A. (1988) "Decision Analysis: Practice and Promise," Management Science,
34,679- 695.
Keeney, R. (1992) Vallie-Focused Thinkillg. Cambridge, MA: Harvard University Press.
Keeney, R., and H. Raiffa (1976) Decisions with Multiple Objectives. New York: Wiley.
Lindl ey, D. V. ( 1985) Making Decisiol/s, 2nd ed. New York: Wiley.
Phillips, L. D. (1982) " Requisi te Dec ision Modelling." Journal of the Operationai
Research Society, 33,303-312.
Philli ps, L. D. (1984) "A Theory of Requisite Decision Models." Acta Psychologic", 56,
29-48.
Raiffa, H. (1968) Decision Analysis. Reading, MA: Addison-Wesley.
Samson, D. ( 1988) Managerial Decision Al1alysis. Homewood, ll..: Irwin.
Ulvil a, 1. W., and R. V. Brown ( 1982) " Decision Analysis Comes of Age." Harvard
Business Revie\v, September-October 1982, 130-14 1.
Vlek, c., W. Edwards, l. Kiss, G. Majone, and M. Toda ( 1984) "What Consti tutes a Good
Deci sion?" Acta PsychoLogica, 56,5- 27.
von Winterfeldt, D., and W. Edwards ( 1986) Decision Analysis and Behavioral Research.
Cambridge: Cambri dge University Press.
Watson, S., and D. Buede ( 1987) Decision Synthesis. Cambridge: Cambridge University
Press.
EPILOGUE

EPILOGUE 17
What did the aDA decide? Its directors decided to use onl y BT on all 227,000 acres,
which were sprayed 011 three separate occasions in Jare spring and earl y Slimmer 1985. At
the time, this was the largest gypsy moth-controI program ever auempled in Oregon. In
1986, 190,000 acres were sprayed, also wi th BT. Most of the areas sprayed the second
year had not been treated the first year because aDA had found later that the gypsy moth
infestation was more widespread than flrst thought. In the SUlIuner of 1986, gypsy mot h
traps throughout the area indi cated that the population was almost completely controlled.
In the spri ng of 1987, the ODA used BT to spray only 7500 acres in 10 isolated pockets
of gypsy moth populations on the fringes of the previously sprayed areas. By 1988, the
spray program was reduced to a few isolated areas near Eugene, and official s agreed that
the gypsy moth population was under control .
Modeling Decisions
T
hi S first section is about modeling decisions.
Chapter 2 presents a short discussion on the ele-
ments of a decision. Through a series of simple exam-
pl es, the basic elements are illustrated: values and objec-
tives , decisions to be made, upcoming uncertain events,
and consequences. The focus is on identifyi ng the basic
elements. This skill is necessary for modeling decisions
as described in Chapters 3, 4, and 5.
In Chapter 3, we learn how to create graphical struc-
tures for decision models. First we consider values and
objectives, di scuss ing in depth how multiple objectives
can be organi zed in hierarchies and networks that can
provide insight and help to generate creative alterna-
tives. We also develop both influence diagrams and de-
cision trees as graphical modeling tool s for representing
the basic structure of decisions. An intluence diagram is
parti cularl y useful for developing the structure of a
complex decision problem because it allows many as-
pects of a problem to be displayed in a compact and in-
tuitive form. A decision-tree representation provides an
alternative picture of a deci sion in whi ch more of the de-
tail s can be di splayed. Both graphical techniques can be
used to represent single-objective decisions, but we
show how they can be used in multiple-objective situa-
tions as well . We end Chapter 3 with a di scussion of
measurement , presenting concepts and techniques that
can be used to ensure that we can adequately measure
achievement of our objecti ves, whet her those objectives
are straightforward (e.g., maximi zing dollars or saving
time) or more difficult to quantify (e.g., minimizing en-
vironmental damage).
Chapters 4 and 5 present the basic tool s available to
the deci sion maker for analyzing a decision model.
Chapter 4 shows how to solve decision trees and influ-
ence diagrams. The basic concept presented is expected
vaLue. When we are concerned with monetary out-
comes, we call this expected monetwy value and abbre-
viate it as EMY. In analyzing a decision, EMV is calcu-
lated for each of the available alternatives. In many
decision situations it is reasonable to choose the alter-
native with the highest EMY. In addition to the EMV
criterion, Chapter 4 also looks briefly at the idea of risk
analysis and the uses of a stochastic-dominance Clite-
rion for making decisions. Finally, we show how ex-
pected value and ri sk analysis can be used in multiple-
objective decisions.
In Chapter 5 we learn how to use sensitivity-analysis
tools in concert with EMV calculations in the iterative
deci sion-structuring and analysis process. After an ini-
tial basic model is built, sensitivity analysis can tell
which of the input variables really matter in the deci-
sion and deserve more attention in the model. Thus,
with Chapter 5 we bring the discussion of modeling de-
cisions full circle, showing how structuring and analysis
are intertwined in the decision-analysis process.
Finally, Chapter 6 delves into issues relating to cre-
ati vity and decision making. One of the clitical aspects of
constructing a model of a decision is the detelmination
of viable alternatives. When searching for alternati ve
actions in a deci sion situation, though, we are subject to
a variety of creative blocks that hamper our search for
new and different possibilities. Chapter 6 describes these
blocks to creativity, discusses creativity from a psycho-
logical perspective, and shows how a careful understand-
ing of one's objectives can aid the search for creative
alternatives. Several creativity-enhancing techniques are
described.
Elements of
Decision Problems
G
iven a complicated probl em, how should one begin? A critical first step is to
identify the elements of the situation. We will classify the various elements into
(I ) values and objectives, (2) decisions to make, (3) uncertain events, and (4) conse-
quences. In thi s chapter, we will di scuss bri efly these four basic elements and illus-
trate them in a seri es of examples.
Values and Objectives
Imagine a farmer whose trees are laden with fruit that is nearly ripe. Even without an
obvious problem to solve or decision to make, we can consider the farmer's objec-
tives. Certainly one objective is to harvest the fruit successfully. This may be impor-
tant because the fruit can then be sold, providing money to keep the farm operating
and a profit that can be spent for the welfare of the family. The falmer may have
other underlying objectives as well , such as maximizing the use of organic farming
methods.
Before we can even talk about making decisions, we have to understand values
and objectives. "Val ues" is an overused term that can be somewhat ambiguous; here
we usc it in a general sense to refer to things that matter to you. For example, you
may want to learn how to sail and take a trip around the world. Or you may have an
. objective of learning how to speak Japanese. A scientist may be interested in resolv-
ing a specific scientific question. An investor may want to make a lot of money or