Structuring Decisions
Dr. Yan Liu
Department of Biomedical, Industrial & Human Factors Engineering
Wright State University
Introduction
Step 1: Identifying and Structuring the Values and Objectives
Step 2: Structuring the Elements of Decision Situation into a
Logical Framework
Identifying issues that matter
Listing objectives and separating means and fundamental objectives
Specifying measures of fundamental objectives
Structuring logic and time sequence among decisions, uncertain events, and
consequences
Tools: influence diagrams and decision trees
Step 3: Refining and Precisely Defining the Elements
The decisions to be made and the available alternatives
Probability distributions of uncertain events through a combination of data
analysis and expert judgment
Measures of consequences and tradeoffs
2
Identifying Values and Objectives
Techniques for Identifying Objectives (Table 3.2 at page 45)
Develop a wish list (What do/should we want and value?)
Determine strategic objectives (What are our ultimate goals?)
Determine generic objectives (what are our objectives for customers, our
family, or ourselves?)
Identify alternatives (what are perfect/terrible/reasonable alternatives and their
good/bad sides?)
Consider problems and shortcomings (what is wrong/right? what needs
fixing?)
Predict consequences (what might occur to what we care about?)
Identify goals, constraints, and guidelines (what are our aspirations and
limitations placed on us?)
Consider different perspectives (what would be our competitors concern?)
3
Structuring Values and Objectives
Categorizing Objectives
Sort the list of objectives and group them into categories
Removing Irrelevant Objectives
Separating Means and Fundamental Objectives
Means objectives are those helping to achieve other objectives
Fundamental objectives are those reflecting what we want to accomplish
ultimately
e.g. One objective of taking this class is to maximize your learning of decision
analysis process
e.g. One objective of going for a vacation is to maximize relaxation
Whether an objective is a means or fundamental objective can be a subjective
judgment
Structuring Values and Objectives (Cont.)
Fundamental Objectives Hierarchy
Back
Upper levels represent more general objectives
Lower levels describe important elements of the more general levels
Lowest-level fundamental objectives are the basis on which consequences are
measured
A Fundamental Hierarchy of Vehicle Regulation
Structuring Values and Objectives (Cont.)
Means Objectives Networks
A means objective can be connected to several other objectives to indicate it
helps achieve them
A Means Objective Network of Vehicle Regulation
Techniques for Organizing Objectives
Example
7
Summer Intern Example
The Peach Tree Consumer Products Inc. has an opening for a summer
intern. Working under the supervision of a senior employee in the
marketing group, the intern would focus primarily on the development
of a market survey for certain of the companys products. The problem
is how to find an appropriate individual to fill this slot.
List of Objectives
Maximize quality of market survey;
Sell more consumer products;
Build market share;
Identify new market niches for companys
products;
Minimize cost of survey design;
Try out prospective permanent employee;
Establish relationship with local college;
Provide assistant to senior employee;
Free up an employee to be trained for new
assignment;
Learn updated techniques from intern;
Expose intern to real-world business experience;
Maximize profit;
Provide financial assistance to college students
9
Categorized Objectives:
Boost Business Performance
Sell more products, maximize profit, increase market share, identify market
niche
Improve Work Environment
Bring in new energy, assist senior employee
Improve the quality and efficiency of marketing activities
Maximize survey quality, minimize survey cost
Better Personnel and Corporate Development
Lean updated techniques, free up employees for new assignment, try out
prospective employee
Increase the Engagement of Community Service
Financial aid, expose intern to real world, relationship with local colleges
10
11
Structuring Decision Problems
Step 1: Identifying and Structuring the Values and Objectives
Step 2: Structuring the Elements of Decision Situation into a Logical
Framework
Step 3: Refine and Precisely Define the Elements
12
Influence Diagrams
Decision Node
(decisions to be made)
Venture
Succeeds
or Fails
Invest?
Return on
Investment
Chance Node
(uncertain events)
Computer
Industry
Growth
Computation Node
(intermediate consequence)
Overall
Satisfaction
Payoff Node
(final consequence)
Influence Diagram of a Venture Capitalists Decision Problem
13
Influence Diagrams (Cont.)
Relationships between nodes are symbolized with arrows or directed arcs
Distinctions are made here between sequence and dependence arcs only for
teaching purposes. Once you are familiar with the differences, you can use
solid arcs throughout the influence diagram like the convention used in the
textbook
14
Influence Diagrams (Cont.)
Influence Diagrams and Fundamental Objectives Hierarchy
The Payoff node corresponds to the most general objective (located at the
upper-most level) in the fundamental-objectives hierarchy
The computation nodes correspond to the objectives at the lower levels in
the hierarchy
15
Basic Influence Diagrams
Basic Risky Decision
Whether the potential gain in the risky choice is worth the risk
Investment Example
You have $2,000 to invest and the objective is to earn as high a return on your
investment as possible. There are two alternatives: investing in a friends business
or keeping the money in a savings account with a fixed interests rate. If you invest
in the business, your return depends on the success of the business. You figure
there could be two possible outcomes: the business is either widely successful
earning you $3,000 beyond your initial investment (hence leaving you $5,000 in
total) or a total flop, in which case you will lose all your money. On the other
hand, if you put your money into a saving account, you will earn $200 in interest
regardless of your friends business.
16
Influence Diagram of the Investment Decision Problem
Decision Tree
17
Basic Influence Diagrams (Cont.)
Imperfect Information
Imperfect information about some uncertain event (e.g. forecast and estimate) will
affect the eventual payoff
Evacuation Example
Suppose you live in Miami. A hurricane near the Bahama Islands threatens to cause
severe damage. As a result, the authorities recommend everyone to evacuate.
Although the evacuation is costly, you would be safe. On the other hand, staying is
risky. You could be injured or even killed if the storm comes ashore within 10 miles
of your home. If the hurricanes path changes, however, you would be safe without
having incurred the cost of evacuating. The two fundamental objectives are to
maximize your safety and to minimize your costs.
Undoubtedly, you will pay close attention to the weather forecasters who would
predict the course of the storm. However, the weather forecasters are not perfect
predictors because not everything is known about hurricanes.
18
Forecasts
Will hit Miami
Will miss Miami
Forecast
dependence Hurricane
Path
Outcomes
Hits Miami
Misses Miami
sequence
Alternatives
Evacuate
Stay
Decision
Consequence
Choices Outcomes
Consequences
Hits Miami Safety, High Cost
Evacuate
Stay
Misses Miami Safety, High Cost
Hits Miami Danger, Low Cost
Misses Miami Safety, Low Cost
Influence Diagram of the Evacuation Decision Problem
Decision Tree
19
Basic Influence Diagrams (Cont.)
Sequential Decisions
Two or more decisions that need to be made in sequence
Evacuation Example
Suppose in the example of hurricane-evacuation decision, you are waiting
anxiously for the forecast as the hurricane is bearing down. Should you keep
waiting for the forecast or leave immediately? In this case, you are facing a
sequential decision situation. If you decide to wait for the forecast, then your next
decision is whether you should evacuate or stay based on the forecast information.
20
Forecast
Hurricane
Path
Evacuate?
Consequence
sequence
Wait for
Forecast?
Influence Diagram of the Sequential Evacuation Decision Problem
21
Basic Influence Diagrams (Cont.)
Computation Nodes (Intermediate Calculations )
Emphasizing the structure of the influence diagram, especially when a node
receives inputs from many other nodes
Used in the same way as payoff nodes
Their values can be calculated directly from inputs of predecessor nodes
Product Example
Suppose a firm is considering introducing a product, and its fundamental objective
is to maximize the profit.
Revenue
1st Version
Introduce
Product?
Cost
Profit
22
Price/unit?
2nd Version
Introduce
Product?
Price/unit?
3rd Version
Units Sold
Fixed Cost
Profit
Units Sold
Fixed Cost
Revenue
Introduce
Product?
Variable Cost
Variable Cost
Cost
Profit
23
Constructing an Influence Diagram
No set strategy is given; a good approach is to put together a simple
version of the diagram first and then add details as necessary
Steps for Constructing Influence Diagram
1. Identify the decisions to be made. If there are more than one decision,
determine their time sequence and draw sequence arcs to connect the decision
nodes
2. Structure fundamental objectives hierarchy and convert the fundamental
objectives into payoff or computation nodes in the influence diagram
3. Identify relevance relationships between the decision nodes and computation
nodes or payoff node and draw corresponding arcs
4. Identify all the uncertain events
5. Identify the sequence relationships between the chance nodes and decision
nodes and draw corresponding arcs
6. Identify the relevance relationships between the chance nodes and draw
corresponding arcs
24
Constructing an Influence Diagram
Steps for Constructing Influence Diagram (Cont.)
7. Identify the relevance relationships between the chance nodes and
computation nodes or payoff node and draw corresponding arcs
8. Check the appropriateness of the influence diagram (any missing and/or
irrelevant information)
25
EPA Example
The Environmental Protection Agency (EPA) often must decide whether to
permit the use of an economically beneficial chemical that may induce cancer
(carcinogenic). Furthermore, the decision often must be made without perfect
information about either the long-term benefits or health hazards. Alternative
courses of actions are to permit the use of the chemical, restrict its use, or to
ban it all together. Tests can be run to learn something about the carcinogenic
potential, and survey data can give an indication of the extent to which
people are exposed when they do use the chemical. These pieces of
information are both important in making the decision. For example, if the
chemical is only mildly toxic and the exposure rate is minimal, then restricted
use may be reasonable. On the other hand, if the chemical is only mildly
toxic but the exposure rate is high, then banning its use may be imperative.
26
Influence Diagram of the EPA Decision Problem
Usage
Decision?
Lab
Test
Survey
Exposure
Rate
Carcinogenic
Potential
Economic
Value
Overall
Satisfaction
Cancer
Cost
Overall Satisfactio
Econ
omic
Value
Canc
er
Cost
27
Economic
Value
Usage
Decision?
Lab
Test
Survey
Exposure
Rate
Cancer
Cost
Overall
Satisfaction
Cancer
Risk
Carcinogenic
Potential
Influence Diagram of the EPA Decision Problem
(adding a computation node)
28
Comments on Influence Diagrams
NOT a flowchart of the decision process
Should NEVER contain cycles (no feedbacks)
Very compact notations that hide lots of information
Interpreting an influence diagram is generally easy
A snapshot of the decision situation at a particular time
Sequencing is implied
Good for conveying model design to others
Creating influence diagrams can be difficult
29
Decision Trees
Decision Trees Display A Decision Problem in Detail
Decision trees explicitly identify the sequence of decisions/events (from left to
right)
Decision trees show all possible future scenarios
One branch for each decision alternative
One branch for each outcome of an uncertain event (outcomes must be mutually
exclusive and collectively exhaustive)
30
Decision Trees (Cont.)
Decision
Alternative
Decision
Node
Chance
Node
Consequence
Widely
Success
$3,000
Business
Business
Result
Flop
-$2,000
Outcome of
Uncertain
Event
Investment
Choice
Savings
$200
Decision Tree of the Investment Decision Problem
Influence Diagram
31
Basic Decision Trees
Basic Risky Decision
Politician Example
The fundamental objective of a politician is to have a career that provides leadership
for the country and representation for her constituency. She can do so to a varying
degrees by serving in Congress. She might have two options: 1) running for reelection
to her U.S. House of Representatives seat, in which case her reelection is virtually
assured; and 2) running for a Senate seat, in which case there is a chance of losing. If
she loses, she could return to her old job as a lawyer (the worst possible outcome). The
best possible outcome is to win the Senate place in terms of her objective of providing
leadership and representation.
32
Running
Decision
Election
Result
Decision Tree of the Politicians Basic Risk Decision
33
Basic Decision Trees (Cont.)
Double-Risk Decision Dilemma
Decide between two risky prospects
Election
Result
Running
Decision
Election
Result
The Politicians Double- Risk Decision Dilemma
34
Basic Decision Trees
Range-of-Risk Decision Dilemma
The outcomes of the chance events are a range of values
Insurance Example
An individual has sued for damages of $450,000 because of injury. The insurance
company has offered to settle for $100,000. The plaintiff must decide whether to
accept the settlement or go to court.
Court
Result
Decision Tree of the Insurance Example
35
Basic Decision Trees
Imperfect Information
Placing the corresponding chance node prior to the decision that it affects
Forec
ast
Evacua
tion
Decisio
n
Evacua
tion
Decisio
n
Decision Tree of the Evacuation Decision Problem
Influence Diagram
36
Basic Decision Trees
Sequential Decisions
Order decisions in decision trees from left to right
Wait
Decis
ion
Evacua
tion
Decisi
on
Evacua
tion
Decisi
on
Decision Tree of the Sequential Evacuation Decision Problem
37
Basic Decision Trees
Schematic Representation of Sequential Decisions
In problems with many decisions involved, the sizes of full-blown decision
trees can increase exponentially
38
Compare Influence Diagrams and
Decision Trees
Both influence diagrams and decision trees have strength and
weakness and can complement each other
Tools
Pros
Cons
Compact representation and easy to
Influence understand; Particularly valuable for the
Hide many details
Diagrams structuring phase of problem solving and
representing large problems
Decision
Trees
Display many details of the decision
The size of the tree increases
problem; Useful during careful reflection
fast as the decision problems
and sensitivity analysis on specific
become more complicated
probability and value inputs
39
Structuring Decision Problems
Identifying and Structuring the Values and Objectives
Step 2: Structuring the Elements of Decision Situation into a
Logical Framework
Step 3: Refine and Precisely Define the Elements
40
Decision Details
Define Elements of the Decision Clearly
Every Element of the Decision Model Needs to Pass the Clarity Test
e.g. In the Environmental Protection Agency example, one fundamental
objective is to minimize the social cost of cancer. How will the cancer cost be
measured, in terms of incremental lives cost or incremental cases of cancer?
Include both treatable and fatal? One uncertain event is rate of exposure. What
are the possible outcomes? How to measure? The number of people exposed to
the chemical per day or per hour?
Various people involved in the decision think about the decision elements in
exactly the same way; no misunderstandings regarding the definitions of the
basic decision elements
Cash Flows and Probabilities
Specific chances associated with each outcome of uncertain events
Specific cash flows at different times
41
Research-and-Development Example
A company needs to decide whether to spend $2M to continue with a
particular research project. The success of the project (measured by
obtaining a patent) is not assured. At this point, the decision maker
judges only a 70% chance of getting the patent. If the patent is
awarded, the company can either license the patent for an estimated
$25M or invest an additional $10M to create a production and
marketing system to sell the product directly. If the company chooses
the latter, it faces uncertainty of demand and associated profit from
sales.
Decision Tree
42
Continue
Development
Patent
Awarded
License
Technology
No
Patent
Stop
Development
$23
M
(p=0
$2
.3)
M
$
0
Demands
High
(p=0.$5
25)Med.5M
Sell Product Market (p=0.$3
Result 55)Low3M
$10
(p=0 $1
M
.20) 5M
(p=0
Develop Production
Production
.7)
and Marketing to
Developmen Decision
$2 t Result
M
Development
Decision
$25
M
$4
3M
$2
1M
$3
M
A Decision Tree Representation (With Cash Flows and Probabilities
Specified) of the Research-and-Development Decision Problem
Problem
43
Decision Details (Cont.)
Defining Measurement Scales for Fundamental Objectives
Objectives with natural attribute scales can be measured objectively
e.g. monetary values, time, length, weight, etc.
Objectives without natural attribute scales
e.g. public image, quality of life, etc.
Measured indirectly with proxies
e.g. GPA as a measure of a persons intelligence
Measured subjectively using an attribute rating scale
e.g. The quality of life can be measured using a five-point Likert scale
questionnaire (best, better, satisfactory, worse, and worst)
44
Exercise
A dapper young decision maker has just purchased a new suit for
$200. On the way out the door, the decision maker considers taking
an umbrella. With the umbrella on hand, the suit will be protected
in the event of rain. Without the umbrella, the suit will be ruined if
it rains. On the other hand, if it does not rain, carrying the umbrella
is an unnecessary inconvenience.
1. Draw an influence diagram of this situation
2. Draw a decision tree of this situation
3. Before deciding, the decision maker considers listening to the
weather forecast on the radio. Draw an influence diagram and
decision tree that take into account the weather forecast
1.
Whether to
bring
umbrella?
Weather
Overall Satisfaction
Protect
Suit
Overall
Satisfaction
Minimize
Inconvenience
Rain
2.
Prote
ct
Suit
Minimiz
e
Inconve
nience
Suit Ruined; Convenience
No
Weather
Whether to
bring
umbrella?
Yes
No Rain
Suit Protected; Convenience
Suit protected; Inconvenience
46
3.
Whether to
bring
umbrella?
Forecast
Weather
Protect
Suit
Overall
Satisfaction
Minimize
Inconvenience
Rain
Suit Ruined; Convenience
No
Rain
Whether to
bring
umbrella?
Weather
No Rain
Yes
Rain
Forecast
Suit Protected; Convenience
Suit protected; Inconvenience
Suit Ruined; Convenience
No
No Rain
Whether to
bring
umbrella?
Weather
Yes
No Rain
Suit Protected; Convenience
Suit protected; Inconvenience
47