You are on page 1of 48

Probability and Decision

Analysis

Structuring Decisions

1
This chapter will cover
• Structuring values and objectives
• Measurement scales
• Influence diagrams
• Decision trees
• Non-probabilistic decision rules

2
Structuring values and objectives
• Recall that:
– Values are things that matter to us
– Objectives are what we specifically want to achieve (to
reach our values)
• Once our high-level values are defined, we should:
– Define our objectives
– Structure them
– Define their measurement scales

3
Defining our objectives
Follow creative brainstorming approaches to define what you care about achieving.

4
Example: objectives when hiring a
summer intern

5
Clean-up the objectives given the
decision context

6
Categorize the objectives
• For the hiring summer intern example:
– Improve work environment
– Develop personnel and company resources
– Provide community service

7
Objectives can be categorized into
fundamental and means
• Fundamental objectives: reflect what we
really want to achieve.
• Means objectives: identify objectives that help
achieve other fundamental objectives.

This step is essential in increasing the clarity of our objectives and our ability to
later measure them and assess what affects them.

8
Example: Fundamental objectives
Organize fundamental objectives in a hierarchy.

9
Example: Means objectives
Organize means objectives in a network.

Fundamental objective

10
Identifying fundamental and means
objectives

11
Back to the summer intern example

Fundamental
objectives
organized as a
hierarchy

Means
objectives
organized as a
network (italic)

12
Measuring fundamental objectives
• In complex decision settings, multiple
objectives are considered.
• Ideally, the aim is to be able to measure each
of the lower level fundamental objectives in
the hierarchy.
• How to measure these fundamental
objectives?

13
Defining measurement scales
• We differentiate between:
– Objectives with natural attribute scales: The scales
describe well-defined and understood units of
measurements:
• Dollars
• Hours
• kg…
– Objectives without natural attribute scales:
• “Quality” measures for example

14
Examples of objectives with natural
attributes

15
How to deal with objectives without
natural attributes?
• Look for a proxy that has a natural attribute:
– Means objectives might be good candidates.
– For the intern example, to assess the quality of
the survey, we might want to look at:
– The intern GPA in market research and statistics courses

• Construct an attribute scale:


– Make sure that the scale has meaningful levels.
– Each level is well-described to make it easy to
classify a performance to a corresponding level.
16
A constructed scale for survey quality

17
After defining the objectives, the decision
elements should be structured.
• Two tools are relevant for this step:
– Influence diagrams
– Decision trees

• The purpose is to develop a complete model


where all the following elements are well-
defined:
– Relevant objectives
– Decisions to be made
– Uncertainties and probabilities
– Consequences
18
Influence Diagrams
• Provide simple graphical representations of
decision situations using four basic shapes
– Decision node
– Chance node
– Payoff node
– Consequence or calculation node
– Connectors
– Beginning node of an arc is called a predecessor.
– The node at the end of an arc is a successor.
19
Influence Diagram - Example

20
Influence Diagram & Fundamental
Objectives Hierarchy - Example
Venture
capitalist’s
decision with two
objectives

21
Influence Diagram & Fundamental
Objectives Hierarchy - Example

Multiple
objectives in
selecting a bomb
detection system

22
Using Arcs in Influence Diagrams

Arcs are used to


represent
relationships among
the nodes, and
represent either
relevance or
sequence.

23
Some Basic Influence Diagrams
Basic risky decision: There is one decision to make and one
uncertain event.

24
Some Basic Influence Diagrams
Imperfect information: The eventual payoff is affected by some
type of imperfect information about an uncertain event, e.g.,
weather forecast, expert diagnosis, and computer estimate.

25
Some Basic Influence Diagrams
Imperfect information: The eventual payoff is affected by some
type of imperfect information about an uncertain event, e.g.,
weather forecast, expert diagnosis, and computer estimate.

26
Some Basic Influence Diagrams
Sequential decisions: a series of interdependent decisions linked
together by sequence arcs. The sequence of decisions is shown
by the arcs.

27
Some Basic Influence Diagrams
Sequential decisions: a series of interdependent decisions linked
together by sequence arcs. The sequence of decisions is shown
by the arcs.

Arcs connecting forecasts


and decisions for
previous days (not only
the n-1 day) are omitted
for simplicity. They are
called non-forgetting arcs
indicating that the
decision maker does not
forget previous events
and their outcomes.

28
Some Basic Influence Diagrams
Intermediate calculation: A node that aggregates results from
certain predecessor nodes. It provides a convenient summary.

Basic More detailed

Intermediate
calculations shown

29
Common Mistakes in Influence
Diagrams
• Interpreting influence diagrams as flowcharts
– A flowcharts depicts the sequential nature of a
particular process where each node represents an
event or activity.
– whereas an influence diagrams is a snapshot of
the decision situation at a particular point in
time.
• Including cycles, i.e. circular paths among
nodes
30
Decision trees can display more
details than an influence diagram.

Branches from circle:


possible outcomes of
a chance event
Phrases at end of the
branches:
Square: consequences
decision to
be made

Branches from square:


choices available to
decision maker

31
Interpreting Decision Trees
1. Options shown by branches from a decision
node indicate that the decision maker can
choose only one option.
2. Branches from each chance node form a set
that is:
– Mutually exclusive – only one outcome can
happen
– And collectively exhaustive – no other possible
outcomes exist.
32
Interpreting Decision Trees
3. The decision tree shows all of the possible paths
that the decision maker might follow.
– All possible decision alternatives
– All possible chance outcomes
4. Often helpful to view the nodes as occurring
chronologically
– Beginning on the left side of the tree: this is the first
thing to happen, usually a decision.
– Moving to the right: consequent decisions or chance
events happen.

33
Decision Trees & Fundamental
Objectives Hierarchy
• List all relevant consequences at the end of the
branches. Often shown as a consequence matrix

Objective A Objective B Objective C Objective D

34
Some Basic Decision Trees
Basic risky decision

35
Some Basic Decision Trees
Imperfect information
(and sequential decisions)

36
Some Basic Decision Trees
Range-of-risk dilemma: outcome of a chance event can take on
a value within a range of possible values

37
Some Basic Decision Trees
Sequential decisions

38
Non-probabilistic Decision Rules
• We will cover probabilistic approaches in detail
later.
• Let’s take a look at some possible non-
probabilistic rules:
– Maximax
– Maximin
– Minimax regret

• None of these rules is always best and they all


have their own weaknesses.
39
Example
• Consider the following decision situation:

Land Purchased Airport is Built at Location


at Location(s) A B
A $13 ($12)
B ($8) $11 Payoff matrix
A&B $5 ($1)
None $0 $0
40
Maximax Decision Rule
• Identify the maximum payoff of each alternative.
• Choose the alternative with the largest maximum
payoff.
Land Purchased Airport is Built at Location
at Location(s) A B
A wins A $13 ($12)
B ($8) $11
A&B $5 ($1)
None $0 $0

41
Weakness?

Land Purchased Airport is Built at Location


at Location(s) A B
A wins again! A $13 ($10000)
B ($8) $11
A&B $5 ($1)
None $0 $0

42
Maximin Decision Rule
• Identify the minimum payoff of each alternative.
• Choose the alternative with the largest minimum
payoff.
Land Purchased Airport is Built at Location
at Location(s) A B
A $13 ($12)
B ($8) $11
A&B $5 ($1)
None wins None $0 $0

43
Weakness?

Land Purchased Airport is Built at Location


at Location(s) A B
A $13 ($12)
B ($8) $11
A&B $10000 ($1)
None wins again! None $0 $0

44
Minimax Regret Rule
• Compute the possible regret for each
alternative under each state of nature.
• Identify the maximum possible regret for each
alternative.
• Choose the alternative with the smallest
maximum regret.

45
Minimax Regret Rule

Airport is Built at
Land Purchased Location
at Location(s) A B Max
A $0 $23 $23
B $21 $0 $21
A&B wins A&B Regret matrix
$8 $12 $12
None $13 $11 $13

46
Anomalies with Minimax Regret Rule
 Consider the following payoff matrix
State of Nature
Decision 1 2
A 9 2
B 4 6
 The regret matrix is:
State of Nature
Decision 1 2 MAX
A 0 4 4 ←minimum
B 5 0 5
 Note that we prefer A to B.

47
Adding an alternative
 Alternative C is added.
State of Nature
Decision 1 2
A 9 2
B 4 6
C 3 9

 The regret matrix is:


State of Nature
Decision 1 2 MAX
A 0 7 7
B 5 3 5 ←minimum
C 6 0 6

 Now we prefer B to A? 48

You might also like