CHAPTER 3

Structuring Decisions
Notes
Chapters 3 and 4 might be considered the heart of Making Hard Decisions with DecisionTools. Here is
where most of the action happens. Chapter 3 describes the process of structuring objectives and building
influence diagrams and decision trees. Chapter 4 discusses analysis. The chapter begins with a
comprehensive discussion of value structuring and incorporates value-focused thinking throughout.
Constructing influence diagrams and decision trees to reflect multiple objectives is demonstrated, and the
chapter contains a discussion of scales for measuring achievement of fundamental objectives, including
how to construct scales for objectives with no natural measures.

Understanding one’s objectives in a decision context is a crucial step in modeling the decision. This section
of the chapter shows how to identify and structure values, with an important emphasis on distinguishing
between fundamental and means objectives and creating hierarchies and networks, respectively, to
represent these. The fundamental objectives are the main reasons for caring about a decision in the first
place, and so they play a large role in subsequent modeling with influence diagrams or decision trees.

Students can generally grasp the concepts of influence diagrams and how to interpret them. Creating
influence diagrams, on the other hand, seems to be much more difficult. Thus, in teaching students how to
create an influence diagram for a specific situation, we stress basic influence diagrams, in particular the
basic risky decision and imperfect information. Students should be able to identify these basic forms and
modify them to match specific problems. The problems at the end of the chapter range from simple
identification of basic forms to construction of diagrams that are fairly complex.

The discussion of decision trees is straightforward, and many students have already been exposed to
decision trees somewhere in their academic careers. Again, a useful strategy seems to be to stress some of
the basic forms.

Also discussed in Chapter 3 is the matter of including in the decision model appropriate details. One issue
is the inclusion of probabilities and payoffs. More crucial is the clarity test and the development of scales
for measuring fundamental objectives. The matter of clarifying definitions of alternatives, outcomes, and
consequences is absolutely crucial in real-world problems. The clarity test forces us to define all aspects of
a problem with great care. The advantage in the classroom of religiously applying the clarity test is that the
problems one addresses obtain much more realism and relevance for the students. It is very easy to be lazy
and gloss over definitional issues in working through a problem (e.g., “Let’s assume that the market could
go either up or down”). If care is taken to define events to pass the clarity test (“The market goes up means
that the Standard & Poor’s 500 Index rises”), problems become more realistic and engaging.

The last section in Chapter 3 describes in detail how to use PrecisionTree for structuring decisions. The
instructions are intended to be a self-contained tutorial on constructing decision trees and influence
diagrams. PrecisionTree does have an interactive video tutorial along with video tutorials on the basics and
videos from experts. These videos along with example spreadsheets and the manual all can be found in the
PrecisionTree menu ribbon under Help, then choosing Welcome to PrecisionTree.

Please note that if you enter probability values that do not sum to 100% for a chance node, then the
program uses normalized probability values. For example, if a chance node has two branches and the
corresponding probabilities entered are (10%, 80%), then the model will use (11.11%, 88.89%) – i.e.,
0.1/(0.1+0.8) and 0.8/(0.1+0.8).

Because this chapter focuses on structuring the decision, many of the problems do not have all of the
numbers required to complete the model. In some cases, the spreadsheet solution provides the structure of
the problem only, and the formulas were deleted (for example, problem 3.10). In other cases, the model is

14
completed with representative numbers (for example, problem 3.5). In the completed model, you will see
expected values and standard deviations in the decision trees and influence diagrams. These topics are
discussed in Chapter 4.

Topical cross-reference for problems

Branch pay-off formula 3.25 – 3.28
Calculation nodes 3.14, Prescribed Fire
Clarity test 3.4, 3.5, 3.12, 3.21, Prescribed Fire
Constructed scales 3.3, 3.14, 3.20, 3.21
Convert to tree 3.11, 3.26, 3.28, Prescribed Fire
Decision trees 3.5 - 3.7, 3.6 - 3.11, 3.13, 3.20 - 3.28,
Prescribed Fire, S.S. Kuniang, Hillblom
Estate
Imperfect information 3.9, 3.11
Influence diagrams 3.4, 3.6 - 3.9, 3.11, 3.14, 3.16, 3.20, 3.21,
3.26, 3.28, Prescribed Fire, S.S. Kuniang
Net present value 3.24, 3.25
Objectives 3.1 - 3.3, 3.7, 3.10, 3.14 - 3.19, 3.21, 3.23,
Prescribed Fire
PrecisionTree 3.5, 3.9, 3.24 - 3.28, Prescribed Fire, S.S.
Kuniang
Sensitivity analysis 3.20
Umbrella problem 3.9

Solutions
3.1. Fundamental objectives are the essential reasons we care about a decision, whereas means objectives
are things we care about because they help us achieve the fundamental objectives. In the automotive safety
example, maximizing seat-belt use is a means objective because it helps to achieve the fundamental
objectives of minimizing lives lost and injuries. We try to measure achievement of fundamental objectives
because we want to know how a consequence “stacks up” in terms of the things we care about.

Separating means objectives from fundamental objectives is important in Chapter 3 if only to be sure that
we are clear on the fundamental objectives, so that we know what to measure. In Chapter 6 we will see that
the means-objectives network is fertile ground for creating new alternatives.

3.2. Answers will vary because different individuals have different objectives. Here is one possibility.
(Means objectives are indicated by italics.)

Best Apartment

Minimize Maximize Maximize

Travel time Ambiance Use of leisure
time

To School To Shopping To Leisure-time

Activities Alone Friends Neighbors

Maximize
Discretionary $$
Centrally Maximize
Parking features (e.g.,
located at apartment pool, sauna,
laundry)
Maximize
windows, light Minimize
Rent

15
3.3. A constructed scale for “ambiance” might be the following:

Best Many large windows. Unit is like new. Entrance and landscape are clean and inviting
with many plants and open areas.
-- Unit has excellent light into living areas, but bedrooms are poorly lit. Unit is clean and
maintained, but there is some evidence of wear. Entrance and landscaping includes some
plants and usable open areas but is not luxurious.
-- Unit has one large window that admits sufficient light to living room. Unit is reasonably
clean; a few defects in walls, woodwork, floors. Entrance is not inviting but does appear
safe. Landscaping is adequate with a few plants. Minimal open areas.
-- Unit has at least one window per room, but the windows are small. Considerable wear.
Entrance is dark. Landscaping is poor; few plants, and small open areas are not inviting.
Worst Unit has few windows, is not especially clean. Carpet has stains, woodwork and walls are
marred. Entrance is dark and dreary, appears unsafe. Landscaping is poor or nonexistent;
no plants, no usable open areas.

3.4. It is reasonable in this situation to assume that the bank’s objective is to maximize its profit on the
loan, although there could be other objectives such as serving a particular clientele or gaining market share.
The main risk is whether the borrower will default on the loan, and the credit report serves as imperfect
information. Assuming that profit is the only objective, a simple influence diagram would be:

Credit Default?
report

Make
Loan? Profit

Note the node labeled “Default” Some students may be tempted to call this node something like “Credit
worthy?” In fact, though, what matters to the bank is whether the money is paid back or not. A more
precise analysis would require the banker to consider the probability distribution for the amount paid back
(perhaps calculated as NPV for various possible cash flows).

Another question is whether the arrow from “Default” to “Credit Report” might not be better going the
other way. On one hand, it might be easier to think about the probability of default given a particular credit
report. But it might be more difficult to make judgments about the likelihood of a particular report without
conditioning first on whether the borrower defaults.

Also, note that the “Credit Report” node will probably have as its outcome some kind of summary measure
based on many credit characteristics reported by the credit bureau. It might have something like ratings that
bonds receive (AAA, AA, A, and so on). Arriving at a summary measure that passes the clarity test could
be difficult and certainly would be an important aspect of the problem.

If the diagram above seems incomplete, a “Credit worthiness” node could be included and connected to
both “Credit report” and “Default”:

16
 Credit
worthiness

Credit Default?
report

Make Profit
Loan?

Both of these alternative influence diagrams are shown in the Excel file “Problem 3.4.xlsx.” Two different
types of arcs are used in the diagrams: 1) value only and 2) value and timing, and these are explained in the
text. A value influence type influences the payoff calculation and a timing type exists if the outcome
precedes that calculation chronologically (or is known prior to the event).

3.5. This is a range-of-risk dilemma. Important components of profit include all of the different costs and
revenue, especially box-office receipts, royalties, licensing fees, foreign rights, and so on. Furthermore, the
definition of profits to pass the clarity test would require specification of a planning horizon. At the
specified time in the future, all costs and revenues would be combined to calculate the movie’s profits. In
its simplest form, the decision tree would be as drawn below. Of course, other pertinent chance nodes could
be included.

Make Revenue
Profit = Revenue-Cost
movie

Don't make movie Value of best

alternative

The revenue for the movie is drawn as a continuous uncertainty node in the above decision tree.
Continuous distributions can be handled two ways in PrecisionTree either with a discrete approximation
(see Chapter 8 in the text) or with simulation (see Chapter 11 in the text). This decision tree with a discrete
approximation of some sample revenue values is shown in the Excel file “Problem 3.5.xlsx.” A potentially
useful exercise is to have the students alter the sample values to see the effect on the model and specifically
the preferred alternative.

3.6.
Influence Diagrams Decision Trees
Strengths Compact Display details
Good for communication Flexible representation
Good for overview of large problems

Weaknesses Details suppressed Become very messy

for large problems

17
Which should he use? I would certainly use influence diagrams first to present an overview. If details must
be discussed, a decision tree may work well for that.

3.7. This problem can be handled well with a simple decision tree and consequence matrix. See Chapter 4
for a discussion of symmetry in decision problems.

Best
Representation

Max
Communication

Overview Details Max flexibility

Influence Excellent for Details Best for symmetric

Diagram large problems hidden decision problems

Decision Poor, due to Details Very flexible

Tree complexity displayed for assymetric
decisions

3.8.

Win Senate
election? Yes
No

Run for Outcome

Senate?
Run for Win Senate? Outcome
Run for Senate
Senate Yes US Senator
Run for House
No Lawyer
House Yes US Representative
No US Representative

Note that the outcome of the “Win Senate” event is vacuous if the decision is made to run for the house.
Some students will want to include an arc from the decision to the chance node on the grounds that the
chance of winning the election depends on the choice made:

18
 Win
election?
Win
Lose

Run for Outcome

Senate?

Run for Senate Run for Win Election? Outcome

Run for House Senate Yes US Senator
No Lawyer
House Yes US Representative

Note that it is not possible to lose the House election.

The arc is only to capture the asymmetry of the problem. To model asymmetries in an influence diagram,
PrecisionTree uses structure arcs. When a structural influence is desired, it is necessary to specify how the
predecessor node will affect the structure of the outcomes from the successor node. By using a structure
arc, if the decision is made to run for the house, the “Win election?” node is skipped. This influence
diagram is shown in the Excel file “Problem 3.8.xlsx.”

3.9. (Thanks to David Braden for this solution.) The following answers are based on the interpretation that
the suit will be ruined if it rains. They are a good first pass at the problem structure (but see below).

suit not ruined, plus a sense of relief

rain

no rain

take umbrella
suit not ruined, but inconvenience incurred

suit ruined
do not take umbrella

rain

no rain

suit not ruined

(A) Decision tree

19
 Weather
Rain Rain
Forecast

Take Take
Satisfaction Satisfaction
Umbrella? Umbrella?

(B) Basic Risky Decision (C) Imperfect Information

The Excel solution “Problem 3.9.xlsx” shows a realization of this problem assuming the cost of the suit is
$200, the cost of the inconvenience of carrying an umbrella when it is not raining is $20, the probability of
rain is 0.25, and the weather forecaster is 90% accurate.

Note that the wording of the problem indicates that the suit may be ruined if it rains. For example, the
degree of damage probably depends on the amount of rain that hits the suit, which is itself uncertain! The
following diagrams capture this uncertainty.
suit not ruined,
plus a sense of relief
rain

no rain
suit not ruined, but
take umbrella inconvenience incurred
suit ruined suit ruined

do not take umbrella

suit not
ruined suit not ruined, but some
rain effort spent to avoid ruining
the suit
no rain

suit not ruined

(A) Decision tree

20
 Weather
Rain Rain
Forecast

Take Take Ruin

Satisfaction Ruin Satisfaction
Umbrella? Suit Umbrella? Suit

(C) Imperfect Information

(B) Basic Risky Decision

3.10. (Thanks to David Braden for this solution.)

The decision alternatives are (1) use the low-sodium saline solution, and (2) don’t use the low-sodium
saline solution. The uncertain variables are: (1) The effect of the saline solution, consequences for which
are patient survival or death; (2) Possibility of court-martial if the saline solution is used and the patent
dies. The possible consequences are court-martial or no court-martial. The decision tree:

patient dead
do not use
saline solution

patient patient survives and the

saved use of saline solution justified
for other patients
use saline
solution patient dead and
patient dead doctors suffer

court-martial

no court-martial

patient dead

This decision tree is drawn in the Excel file “Problem 3.10.xlsx.”

3.11. a.

Sunny
Best
Weather
Outdoors Rainy Terrible

Indoors
Good
Party
Satisfaction
decision
No party
Bad

21
This influence diagram is drawn in the Excel file “Problem 3.11.xlsx” with some sample values assumed
(on a utility scale, a sunny party outside is worth 100, an indoors party is worth 80, no party is worth 20, a
party outside in the rain is worth 0, and the probability of rain is 0.3). A structure only arc is added in the
file between party decision and weather to include the asymmetries to skip the weather uncertainty if the
decision is made to have no party or have one indoors.
The second worksheet in the file shows the default decision tree created by the “Convert to Tree” button on
the influence diagram settings dialog box. (Click on the name of the influence diagram “Problem 3.11a” to
access the influence diagram settings. The Convert to Decision Tree button creates a decision tree from the
current influence diagram. This can be used to check the model specified by an influence diagram to insure
that the specified relationships and chronological ordering of nodes are correct. Conversion to a decision
tree also shows the impacts of any Bayesian revisions made between nodes in the influence diagram.
Once a model described with an influence diagram is converted to decision tree, it may be further edited
and enhanced in decision tree format. However, any edits made to the model in decision tree format will
not be reflected in the original influence diagram.
b. The arrow points from “Weather” to “Forecast” because we can easily think about the chances
associated with the weather and then the chances associated with the forecast, given the weather. That is, if
the weather really will be sunny, what are the chances that the forecaster will predict sunny weather? (Of
course, it is also possible to draw the arrow in the other direction. However, doing so suggests that it is easy
to assess the chances associated with the different forecasts, regardless of the weather. Such an assessment
can be hard to make, though; most people find the former approach easier to deal with.)

Forecast Weather

Party
Satisfaction
decision

22
 Sunny
Best

Outdoors Rainy
Terrible

Indoors
Good
Forecast
= “Sunny” No party
Bad

Sunny
Best

Outdoors Rainy
Forecast Terrible
= “Rainy”
Indoors
Good

No party
Bad

The influence diagram including the weather forecast is shown in the third worksheet and the associated
default decision tree created by the “Convert to Tree” function is shown in the fourth worksheet.
Additionally, we assumed that the weather forecaster is 90% accurate.
3.12. The outcome “Cloudy,” defined as fully overcast and no blue sky, might be a useful distinction,
because such an evening outdoors would not be as nice for most parties as a partly-cloudy sky. Actually,
defining “Cloudy” to pass the clarity test is a difficult task. A possible definition is “At least 90% of the sky
is cloud-covered for at least 90% of the time.”

The NWS definition of rain is probably not as useful as one which would focus on whether the guests are
forced indoors. Rain could come as a dreary drizzle, thunderstorm, or a light shower, for instance. The
drizzle and the thunderstorm would no doubt force the guests inside, but the shower might not.

One possibility would be to create a constructed scale that measures the quality of the weather in terms that
are appropriate for the party context. Here are some possible levels:

(Best) Clear or partly cloudy. Light breeze. No precipitation.

-- Cloudy and humid. No precipitation.
-- Thunderclouds. Heavy downpour just before the party.
-- Cloudy and light winds (gusts to 15mph). Showers off and on.
(Worst) Overcast. Heavy continual rain.

3.13.

23
 Change product Behind schedule

Engineer
says “Fix #3.” Replace
#3 #3 Defective On schedule, costly

#3 not Behind scedule,costly

defective

Change product Behind schedule

Engineer says
“#3 OK.”
Replace
#3 #3 Defective On schedule, costly

#3 not
Behind scedule,costly
defective
This decision tree is drawn in the Excel file “Problem 3.13.xlsx.”

3.14.

Forecast Hurricane
Path

Safety
Decision

Evacuation Consequence
Cost

Note that Evacuation Cost is high or low depending only on the evacuation decision. Thus, there is no arc
from Hurricane Path to Evacuation Cost.

This influence diagram is drawn in the Excel file “Problem 3.14.xlsx.” Because PrecisionTree allows only
one payoff node per influence diagram, the “Safety” and “Evaluation Cost” nodes are represented by
calculation nodes. A calculation node (represented by a rounded blue rectangle) takes the results from
predecessor nodes and combines them using calculations to generate new values. These nodes can be used
to score how each decision either maximizes safety or minimizes cost.

The constructed scale for safety is meant to describe conditions during a hurricane. Issues that should be
considered are winds, rain, waves due to the hurricane’s storm surge, and especially the damage to
buildings that these conditions can create. Here is a possible constructed scale:

(Best) Windy, heavy rain, and high waves, but little or no damage to property or infrastructure.
After the storm passes there is little to do beyond cleaning up a little debris.
-- Rain causes minor flooding. Isolated instances of property damage, primarily due to
window breakage. For people who stay inside a strong building during the storm, risk is
minimal. Brief interruption of power service.

24
 -- Flooding due to rain and storm surge. Buildings within 100 feet of shore sustain heavy
damage. Wind strong enough to break many windows, but structural collapse rarely
occurs. Power service interrupted for at least a day following the storm.
-- Flooding of roads and neighborhoods in the storm’s path causes areas with high property
damage. Many roofs are severely damaged, and several poorly constructed buildings
collapse altogether. Both electrical power and water service are interrupted for at least a
day following the storm.
(Worst) Winds destroy many roofs and buildings, putting occupants at high risk of injury or
death. Extensive flooding in the storm’s path. Water and electrical service are interrupted
for several days after the storm. Structural collapse of older wood-frame buildings occurs
throughout the region, putting occupants at high risk.

3.15. Answers to this question will depend largely on individual preferences, although there are some
typical responses. Some fundamental objectives: improve one’s quality of life by making better decisions,
help others make better decisions, improve career, graduate from college, improve one’s GPA (for one who
is motivated by grades). Some means objectives: satisfy a requirement for a major, improve a GPA (to have
better job opportunities), satisfy a prerequisite for another course or for graduate school. Note that “making
better decisions” is itself best viewed as a means objective because it can provide ways to improve one’s
life. Only a very few people (academics and textbook writers, for example), would find the study of
decision analysis to be its own reward!

The second set of questions relates to the design of the course and whether it is possible to modify the
course so that it can better suit the student’s objectives. Although I do not want to promote classroom
chaos, this is a valuable exercise for both students and instructor to go through together. (And the earlier in
the term, the better!) Look at the means objectives, and try to elaborate the means objectives as much as
possible. For example, if a class is primarily taking the course to satisfy a major requirement, it might make
sense to find ways to make the course relate as much as possible to the students’ major fields of study.

3.16. While individual students will have their own fundamental objectives, we based our hierarchy on a
study titled "Why do people use Facebook?" (Personality and Individual Differences, Volume 52, Issue 3,
February 2012, Pages 243–249). The authors, Ashwini Nadkarni and Stefan G. Hofmann from Boston
University, propose that Facebook meets two primary human needs: (1) the need to belong and (2) the need
for self-presentation. Thus, a particular student’s objectives may be a variation on these themes. Students
may find it interesting to see how their fundamental objectives for their own Facebook page compare and
contrast to the study.

The fact that one’s private area can sometimes be seen by outsiders, particularly employers or future
employers, interferes with presenting your private self. If the private area were truly private, then a user
could be more truthful discussing their revelries and celebrations or even their private thoughts. If you
believe someone is eavesdropping on your conversation, then you are naturally more guarded with your
speech. Thus, Facebook does not provide a good means to expressing your private self.

Maximize Socialization

Maximize Need for Maximize Need for

Belonging Self Presentation

Maximize Maximize Minimize Maximize Maximize

Self Esteem Self Worth Loneliness Public Persona Private Persona

25
Clearly, each individual will have their own FOH, but there are some facts that students may want to take
into consideration. First, it is naïve to think that future employers will not “Google” or “Facebook” you.
Such concerns are not usually on the mind of a college freshman or particularly a high-school student, but
there are a host of problems that can arise from indiscriminately sharing private information. Even if there
is no illegal activity being shown (underage drinking, drug use, etc.), different audiences will have different
norms, and pictures of drinking, dancing, and partying could be considered compromising or
unprofessional.

Second, a future employer or even your current employer may be interested in your postings. They may
want to know what religion you practice, what your interests are, what organizations you belong to, such as
the NRA. All of these could bias them, good or bad, towards you. Also, discretion might be an important
aspect of your position, and employers might view your postings to determine if you can be trusted with
proprietary information.

Posted Information can also be used by competing firms either to gain a direct benefit or more nefariously
to befriend you, and thereby learn more about their competitor. Your personal profile may include job
details and thus provide an opening by unscrupulous ‘cyber sharks’, or by competing businesses hoping to
learn from the eyes and ears of the opposition. You may even have posted confidential information without
realizing its sensitive nature.

Facebook, Inc. itself has rights to your private information and has wide latitude on how it can use your
info. Facebook’s privacy agreement states "We may use information about you that we collect from other
sources, including but not limited to newspapers and Internet sources such as blogs, instant messaging
services, Facebook Platform developers and other users of Facebook, to supplement your profile.”
Facebook can also sell a user's data to private companies, stating: "We may share your information with
third parties, including responsible companies with which we have a relationship."

There are also many data mining and identify theft issues that could result from even the public areas of
your Facebook page. Searches can be performed to discover what movies you like, what music you listen
to, what organizations you belong to, etc. This information can be used for marketing, recruiting, or even
identity theft.

Finally, a student’s short-term and long-term objectives may differ. Short term, the focus will most likely
be on being connected and being cool, i.e., partying, sexual prowess, being wild-n-crazy, etc. Many high-
school and college students compete using Facebook to see who can have the most friends. Having 500 to
1000 Facebook friends is actually common. For these users, their objectives are to attract their friend’s
friends and present a profile attractive to large swaths of the population. Long term, the focus will most
likely be on being professional and staying connected to their college friends.

3.17. Here are my (Clemen’s) fundamental-objectives hierarchy and means-objective network (italics) in
the context of purchasing or building a telescope. The diagram does provide insight! For example, many
astronomers focus (so to speak) on image quality, and so there is a tendency to overemphasize aperture and
quality of eyepieces. But for me, an important issue is enjoying myself as much as possible, and that can
mean taking the telescope out of the city. All of the means objectives related to “Maximize enjoyment of
viewing sessions” (which is intended to capture aspects other than enjoying the quality of the images
viewed) push me toward a smaller, lighter, less expensive telescope. Thus, it is arguable that the basic
question I must ask is whether I just want to get out at night and enjoy seeing a few interesting sights, or
whether my interest really is in seeing very faint objects with as much clarity as possible.

Of course, the creative solution would be to find an inexpensive and highly transportable telescope with
large aperture, excellent optics, and very stable mount. Unfortunately, such a telescope doesn’t exist; all of
the desired physical features would lead to a very expensive telescope!

26
 Best Telescope

Maximize image Maximize Maximize quality

clarity enjoyment of of astrophotography
viewing sessions
Maximize Maximize Maximize
brightness image stability
defnition

Max
aperture Max quality Max Max quality
of optics stability of of tracking device
mount
Max additional
viewing accessories

Min light
pollution in sky

Min cost
Max visits to of telescope
dark-sky site
Min
total weight
Max
transportability
Minimize
telescope size

27
3.18. Objectives are, of course, a matter of personal preferences, and so answers will vary considerably.
a. Here is an objectives hierarchy for the decision context of going out to dinner:

Time Driv ing time

Minimize Ordering/serv ing

cost

Dollar cost

Atmosphere
Maximize
Location
experience
Menu

Quality of f ood

b. A simple hierarchy for deciding from among different trips:

Company (f riends)
Maximize
Learning
experience
Relax

Time
Preparation

Minimize Expense
cost
Trav el cost

Some means objectives might be going to a particular kind of resort; maximizing time spent shopping,
golfing, or on the beach; maximizing nights spent in youth hostels; using a travel agent (to reduce time
spent in preparation); maximizing time in a foreign country (to maximize language learning, for example).

c. Here is a possible fundamental-objectives hierarchy for choosing a child’s name:

Similarity of name
Maximize
f amily ties
Closeness of namesake

by child
Ease of by child’s play mates
use/learning
by relativ es

Nicknames
Minimize negativ e
potential
Teasing by “f riends”

3.19. Responding to this question requires considerable introspection and can be very troubling for many
students. At the same time, it can be very enlightening. I have had students face up to these issues in

28
analyzing important personal decisions such as where to relocate after graduation, whether to make (or
accept) a marriage proposal, or whether to have children. The question, “What is important in my life?”
must be asked and, if answered clearly, can provide the individual with important insight and guidance.

3.20. a. These influence diagrams and decision trees are drawn in the Excel file “Problem 3.20.xlsx.”

Recover Long, healthy

Surgery
fully life
results Have
surgery
Die Death

Have Quality
surgery? Don’t Progressive
of life
have surgery debilitation

b.

Surgery
results Complications

Have Quality
surgery? of life

Long,
Recover healthy Full recovery Long healthy life after
fully life difficult treatment
Have Complications Partial recovery Invalid after difficult
surgery treatment
Die Death
Death Death after difficult
treatment
Don’t Progressive
have surgery debilitation

Given the possibility of complications and eventual consequences, the surgery looks considerably less
appealing.

c. Defining this scale is a personal matter, but it must capture important aspects of what life would be like
in case complications arise. Here is one possible scale:

29
 (Best) No complications. Normal, healthy life.
-- Slight complications lead to minor health annoyances, need for medication, frequent
visits. Little or no pain experienced. Able to engage in most age-appropriate activities.
-- Recovery from surgery requires more than two weeks of convalescence. Pain is intense
but intermittent. Need for medication is constant after recovery. Unable to engage in all
age-appropriate activities.
-- Recovery requires over a month. Chronic pain and constant need for medication.
Confined to wheelchair 50% of the time.
(Worst) Complete invalid for remainder of life. Restricted to bed and wheelchair. Constant pain,
sometimes intense. Medication schedule complicated and occasionally overwhelming.

3.21. This question follows up on the personal decision situation that was identified in problem 1.9.

3.22 This decision tree is drawn in the Excel file “Problem 3.22.xlsx.”

“Not to be”
(commit suicide) “What dreams may
come” (What comes
after death?)

“To be” “Bear fardels”

(continue to live) (burdens of life)

3.23. This decision tree is drawn in the Excel file “Problem 3.23.xlsx.”

Aircraft hostile Crew safe

Shoot

Aircraft not Crew safe

hostile Civilians killed

Aircraft hostile Harm to

crew
Don't shoot

Aircraft not Crew safe

hostile Civilians safe

Rogers’s most crucial objectives in this situation are to save lives, those of his crew and of any civilians
who are not involved. It is not unreasonable to consider objectives of saving his ship or improving the
relationship with Iran, but in the heat of action, these were probably not high on Rogers’ list.

The risk that Rogers faces is that the blip on the radar screen may not represent a hostile aircraft. The main
trade-off, of course, is the risk to his crew versus possibly killing innocent civilians.

As usual, there are a lot of ways this decision tree could be made more elaborate. For example, a “Wait”
alternative might be included. The tree above assumes that if the decision is to shoot, the incoming aircraft
would be hit (generally a safe assumption these days), but one might want to include the possibility of
missing.

30
3.24. This is a straightforward calculation of NPV. Assuming that all the cash flows happen at the end of
the year, the following table shows the cash flows:

Cash Stop Continue Continue Continue Continue Continue

Flows No Patent Patent Patent Patent Patent
License Develop Develop Develop
Year Dem. High Dem. Med Dem. Low
0 0 0 0 0 0 0
1 0 -2 -2 -2 -2 -2
2 0 0 0 0 0 0
3 0 0 5 -5 -5 -5
4 0 0 5 -5 -5 -5
5 0 0 5 11 6.6 3
6 0 0 5 11 6.6 3
7 0 0 5 11 6.6 3
8 0 0 0 11 6.6 3
9 0 0 0 11 6.6 3

Present values are calculated by applying the appropriate discount rate to each cash flow; the discount rate
1
is for the cash flows in year i. Finally, NPV is the sum of the present values. Also, the NPV function
1.15i
in Excel can be used for the calculations as shown in the Excel file “Problem 3.24.xls.”

Present Stop Continue Continue Continue Continue Continue

Values No Patent Patent Patent Patent Patent
Discount License Develop Develop Develop
Year Factor Dem. High Dem. Med Dem. Low
0 1 0.00 0.00 0.00 0.00 0.00 0.00
1 0.8696 0.00 -1.74 -1.74 -1.74 -1.74 -1.74
2 0.7561 0.00 0.00 0.00 0.00 0.00 0.00
3 0.6575 0.00 0.00 3.29 -3.29 -3.29 -3.29
4 0.5718 0.00 0.00 2.86 -2.86 -2.86 -2.86
5 0.4972 0.00 0.00 2.49 5.47 3.28 1.49
6 0.4323 0.00 0.00 2.16 4.76 2.85 1.30
7 0.3759 0.00 0.00 1.88 4.14 2.48 1.13
8 0.3269 0.00 0.00 0.00 3.60 2.16 0.98
9 0.2843 0.00 0.00 0.00 3.13 1.88 0.85

NPV 0.00 -1.74 10.93 13.20 4.76 -2.14

In file “Problem 3.24.xlsx”, the decision tree references the NPV calculations to demonstrate the process of
choosing to continue or stop development. The ability to build these trees in Excel and reference cells as
done in this problem makes this a powerful program. The payoff for each branch of the tree is a formula
that corresponds to the correct cell in the NPV calculations worksheet.

Alternative assumptions can be made about the timing of the cash flows. For example, it would not be
unreasonable to believe that the expenses must be paid at the beginning of the year and that revenue arrives
at the end of the year. The most realistic scenario, however, is that all cash flows are evenly spread out over
the year for which they are specified.

31
3.25. This decision tree is relatively complex compared to the ones that we have seen so far. Buying the
new car does not involve any risk. However, the used car has an uncertainty each year for the next three
years. The decision tree is shown below. Note that it is also possible to calculate the NPVs for the ends of
the branches; the natural interest rate to use would be 10%, although it would be best to use a rate that
reflects what you could earn in another investment. This decision tree representation does not discount the
values.

Down Maintenance and loan payments Net

Payment Year 1 Year 2 Year 3 Salvage Value
New Car -5500 -2522.20 -2722.20 -2722.20 3626 -9841

Repairs Repairs Repairs Net

Purchase Year 1 Year 2 Year 3 Salvage Value
-500 (0.2) 2000 -5350
-700 -1500 (0.6) 2000 -6350
(0.2) -2500 (0.2) 2000 -7350

-500 (0.2) 2000 -6350

-650 -1700 -1500 (0.6) 2000 -7350
(0.2) (0.6) -2500 (0.2) 2000 -8350

-500 (0.2) 2000 -7350

-2700 -1500 (0.6) 2000 -8350
(0.2) -2500 (0.2) 2000 -9350

-500 (0.2) 2000 -6350

-700 -1500 (0.6) 2000 -7350
(0.2) -2500 (0.2) 2000 -8350

-500 (0.2) 2000 -7350

Used Car -5500 -1650 -1700 -1500 (0.6) 2000 -8350
(0.6) (0.6) -2500 (0.2) 2000 -9350

-500 (0.2) 2000 -8350

-2700 -1500 (0.6) 2000 -9350
(0.2) -2500 (0.2) 2000 -10350

-500 (0.2) 2000 -7350

-700 -1500 (0.6) 2000 -8350
(0.2) -2500 (0.2) 2000 -9350

-500 (0.2) 2000 -8350

-2650 -1700 -1500 (0.6) 2000 -9350
(0.2) (0.6) -2500 (0.2) 2000 -10350

-500 (0.2) 2000 -9350

-2700 -1500 (0.6) 2000 -10350
(0.2) -2500 (0.2) 2000 -11350

32
The Excel file “Problem 3.25.xlsx” contains two solutions for this problem. The workbook consists of 3
worksheets. The first worksheet is titled Data & Formulas, and contains the input data for the problem (car
costs, loan value, etc.) along with some formulas. The formulas in cells G29 and H29 calculate the 3-year
net cost of the new and used car respectively. Remember to include the initial $5,500 payment. The
formulas in M12 and L12 also calculate the 3-year net cost but incorporate the time value of money. We
used 10% as the interest rate.

The next two worksheets show the decision trees, with and without incorporating the time value of money.
Both of these trees are linked trees, which we introduce in Chapter 4. There is a method for students to
solve the problem not using linked trees, and this is explained in the text box as Option 1. Essentially,
students will need to create 27 formulas for 3-year net cost, one for each unique combination of
maintenance costs across the 3 years. Once these 27 formulas have been created, it is a simple matter to use
Excel referencing to reference the end node with its corresponding formula. Assigning this problem in
Chapter 3 will help the students realize the flexibility of linked decision trees when they encounter them in
Chapter 4. Additional hints might be needed.

Finally, this problem can also be used to exhibit good-modeling techniques to the students via separating
out the data inputs and placing all of them into one area. Not only does this help with constructing the
model, but it also facilitates running a sensitivity analysis.

3.26a. The influence diagram from Exercise 3.11(a) is shown here drawn in PrecisionTree.

Weather

Party decision

Satisfaction

In order for the “Convert To Tree” button to automatically adjust for the asymmetries, a structure arc is
needed from “Party Decision” to “Weather” (represented by the dotted arrow). This influence diagram and
the corresponding converted decision tree are shown in the first two worksheets in the Excel file “Problem
3.26.xlsx.” Some assumed values for outcomes and probabilities are shown in the converted decision tree.

Sunny 70.0% 0
0 100
Outdoors FALSE Weather
0 70
Rainy 30.0% 0
0 0
Converted Problem 3.26 (a) Party decision
80
Indoors TRUE 1
0 80
No party FALSE 0
0 20

b. Adding the arrow from “Weather” to “Party Decision” means that the information regarding the weather
is known before the time of the decision. Therefore, in the Converted Decision Tree, the “Weather” chance
events will appear in the tree prior to the “Party Decision.”

33
 Weather

Party Decision Satisfaction

Outdoors TRUE 0.7

100 100
Sunny 70.0% Party Decision
0 100
Indoors FALSE 0
80 80
No party FALSE 0
20 20
Converted Problem 3.26 (b) Weather
94
Outdoors FALSE 0
0 0
Rainy 30.0% Party Decision
0 80
Indoors TRUE 0.3
80 80
No party FALSE 0
20 20
c.

Forecast Weather

Party decision Satisfaction

If the arrow went from "Party" to "Forecast", then you would have to make the party decision before you
got the forecast. If an arrow started at "Forecast" and went to "Weather", we would be stating that somehow
the forecast influences the weather.

34
 Sunny 95.5% 0.63
0 100
Outdoors TRUE Weather
0 95.455
Rainy 4.5% 0.03
0 0
"Sunny" forecast 66.0% Party decision
0 95.455
Indoors FALSE 0
0 80
No party FALSE 0
0 20
Converted Problem 3.26 (c) Forecast
90.2
Sunny 20.6% 0
0 100
Outdoors FALSE Weather
0 20.588
Rainy 79.4% 0
0 0
"Rainy" forecast 34.0% Party decision
0 80
Indoors TRUE 0.34
0 80
No party FALSE 0
0 20

3.27. This problem is more challenging than some previous ones as it pushes the students to incorporate
numerical values into the decision model, and to do so, using formulas. Also, the data were provided in two
different formats (annual and monthly) requiring the students to pick one format for the consequence
measure. The solution in the spreadsheet “Problem 3.27.xlsx” uses monthly values. A portion of the tree is
shown below.

The formula for Jameson’s monthly take home if he stays the course is
$2,000 + (Actual Revenue – VJ’s Cost) x (1 – Tax Rate)
Subtracting the required $2,400 is the surplus/deficit Jameson faces, and is the consequence measure we
used.

If the student uses the revenue values, then he or she will need to subtract $2,400 after the analysis. This
only works because the same value is being subtracted from all end nodes. Also, students will run into
PrecisionTree adding the values along the branches (cumulative method). For example, using monthly
revenue values of $2,300 in year 1 and $2,400 in year 2 has a cumulative value of $4,700. In this case,
$4,800 would need to be subtracted. Better is to use the consequence measure that fits the problem.

The formula for Jameson’s monthly take home if he pursues an MSW in 2 years is
(Annual salary/12) x (1 – Tax Rate) – Loan Payment
Again, we subtracted $2,400 for the consequence measure.

35
So what should Jameson do? The future looks bleak for him, but bleaker if he stays the course. Over the
next 3 years, we see him having a monthly deficit of $570 on average if he stays the course, but only a $75
deficit if he gets his MSW. Either way, he is falling behind every month, and expenses associated with
raising children grow almost as fast as the children do.

3.28. From an income maximization point of view, Susan should not quit. There are no financial advantages
of her quitting. As a matter of fact, the only financial ramification is the loss of 6 months of salary or $12,000.
It could be that Susan had set her mind on quitting and cannot process not quitting based on something that
might only happen. She is anchored to quitting, and is probably excited to start a new life.

Susan is not as concerned with income as she is with running through their savings and being evicted. To
model this, we consider all the different scenarios that were presented for Susan and Denzel, and for each
calculate what would be left in their savings account. If this ever drops below zero, then they run the risk of
eviction. The timeframe for them is clearly 6 months because both Susan and Denzel will be on their feet
by then with their new careers.

The solution, shown below and in the file “Problem 3.28.xlsx,” shows the structure of the problem and the
values we used for the end nodes. To help understand the formulas, the spreadsheet cells have been named
and thus the formulas are of the form:
=Savings+6*(Denzel’s Contr if Laid Off + Assistance) - Six_Months_Req.
This formula reports their end savings account balance when Denzel is laid off, when Susan is not
contributing, and they do receive assistance. See the file for complete details. Please note that the above
equals a negative value (-$2,000), but we are not saying the savings account balance can go negative.
Rather, this measures by amount of their deficit.

Susan should definitely not quit her job at the coffee shop. If she does, there is a 70% chance they will have
at least a $2,000 deficit. Any deficit is to be avoided. If she stays working at Joes, then no matter what
happens to Denzel, they will have at least $4,000 left in their savings account.

To exhibit the iterative natures of modeling, we extended the model by adding a third alternative, namely,
staying at the coffee shop for a few months. Cell E7 allows you to enter any value between 0 and 6 for the

36
number of Months Susan stays at Joe’s Coffee. By making this a cell reference, she can dynamically
change it value to view intermediate alternatives. For example, if she stays at Joe’s for 3 months, then the
probability of a deficit drops to 8%.

Case Study: Prescribed Fire

1. Appropriate objectives for the agency in the context of this decision relate to costs. In particular, they
would like to minimize the cost of disposing of the material, and they would like to minimize the cost of
containing fires that go out of control. The latter could include damages if a fire escapes.

2. Many influence-diagram representations are possible. Here is one popular alternative:

Fire Problem
behavior cost

Burn or YUM Treatment Total cost of

& Burn cost containment

Total
cost

A possible decision tree:

Total cost
Success

High cost
Problems Med cost
Burn
Low cost

Escape

Success

High cost
YUM & Problems Med cost
Burn
Low cost

Escape

To pass the clarity test, the costs must be precisely defined. Also, fire behavior must be defined. For
example, it would be necessary to distinguish between and escaped fire and a fire that develops problems
that can be brought under control.

Excel file “Prescribed fire case study.xlsx” shows the above influence diagram with one additional
structural arc. In order for the “Problem Cost” uncertainty to occur only if the “Fire behavior” results in
problems as shown in the above decision tree, an additional structural arc was added between these two
nodes. The file also contains the resulting decision tree from automatically converting the influence
diagram. While the above decision tree treats the resulting multiple objectives at the outcome of each

37
branch, we used calculation nodes for the “Treatment cost” and “Total cost of containment” and are thus
represented as additional nodes in the converted decision tree.

Case Study: The S.S. Kuniang

1. Again, many different influence-diagram representations are possible here. A popular alternative is:

Highest
Bid competitor
amount bid

Winning
bid

Coast
Guard
judgment
Cost

Winning CG
Bid Bid judgment Cost
- - - -
6 6 H 6 + refit cost
5 6 H Cost of next best
5 6 L
} alternative
- - - -

This representation shows that the decision maker has to think about chances associated with the highest
bid from a competitor and the Coast Guard’s judgment. An arc might be included from “Winning bid” to
“Coast Guard judgment” if the decision maker feels that different winning bids would affect the chances
associated with the coast guard’s judgment (which is a distinct possibility). Also, it would be reasonable to
include a separate decision node for purchasing alternative transportation if the bid fails. The representation
above includes that consideration implicitly in the cost table.

A possible decision tree:

Coast Guard Cost of

judgment S.S. Kuniang
Win bid

Bid
amount
Lose bid Cost of next
best alternative

Additional nodes might be added. As with the influence diagram, a decision regarding what to do if the
bid is lost could be included. Also, a decision to keep or sell Kuniang might be included in the event of a
high Coast Guard judgment. The file “SSKuniang I.xlsx” shows representations of both the above influence
diagram and the decision tree.

Case Study: The Hillblom Estate, Part I

No solution is provided for this case. We recommend that the instructor consult the article by Lippman and
McCardle, (2004), “Sex, Lies and the Hillblom Estate,” Decision Analysis, 1, pp 149–166.

38