Decision Analysis

• Decision analysis was created to incorporate the uncertainty

into the decision making.

• What is the problem to solve?

• What are the alternatives available?
• Which uncertain events are relevant, and what are their
outcomes?
• What are the possible consequences from decisions and
uncertain events, and which do you prefer?
 Example
• Cricket Match in Delhi
o May get cancelled in case of rain.
o Should you go or not?
• Decision Alternative:
o Go
o Don't Go
• States of Nature:
o Rain
o No Rain
• Payoff Table:
Outcomes
Alternatives Rain No Rain
Go -100 500
Don’t Go 0 0
 Thompson Lumber Co. Example
• Problem:
o Whether or not to make and sell storage sheds
• Decision Alternatives:
o Build a large plant
o Build a small plant
o Do nothing
• States of Nature:
o Demand for sheds will be high, moderate, or low

Outcomes (Demand)
Alternatives High Moderate Low
Large plant 200,000 100,000 -120,000
Small plant 90,000 50,000 -20,000
No plant 0 0 0
 Five Steps in Decision Making

1. Clearly define the problem

2. List all possible alternatives
o Decision variable
3. Identify all possible outcomes for each alternative
o Chance variable
• A particular event that is relevant to the decision but is beyond our
control.
• Event may have several (finite) possible outcomes.
• Outcomes are called states of nature.
1. Identify the payoff for each alternative & outcome
combination
o Payoff function
2. Use a decision modeling technique to choose an
alternative
How to decide which decision alternative to
select ?
✓ Decision Making under uncertainty
 Decision Making Under Uncertainty

• Probabilities of the possible outcomes are not known

• Probabilities of outcomes are available

How to decide which decision alternative to

select ?
• Several Ways to decide
• Depends on the outlook of decision maker
• No universal right decision
• Pessimistic vs. Optimistic vs. Rationalistic
• Several criteria for decision making are possible
• Several strategies are available for decision making
 Decision making without probabilities

• Maximin Payoff Criterion

• Maximax Payoff Criterion
• Equally Likely Criterion
• Criteria of Realism
• Minimax Regret Criterion
 The Maximin payoff criterion

• Best guaranteed outcome

• Picks the decision alternative whose minimum
payoff gives the maximum.
Outcomes
Alternatives Rain No Rain
Go -100 500
Don’t Go 0 0

• The alternative selected is Not to Go

 The Maximin payoff criterion

Outcomes (Demand)
Alternatives High Moderate Low
Large plant 200,000 100,000 -120,000
Small plant 90,000 50,000 -20,000
No plant 0 0 0

• Choose no plant (best payoff)

 The Maximax payoff criterion
• Decision alternative which maximizes the maximum
return
• Optimistic Approach

Outcomes
Alternatives Rain No Rain
Go -100 500
Don’t Go 0 0

• The alternative selected is to Go

 The Maximax payoff criterion

Outcomes (Demand)
Alternatives High Moderate Low
Large plant 200,000 100,000 -120,000
Small plant 90,000 50,000 -20,000
No plant 0 0 0

• Choose the Large Plant (Best Payoff)

 Equally Likely Criterion

• Assume all outcomes are equally likely

• Choose the alternative which maximizes the average
payoff
Outcomes Average
Alternatives Rain No Rain Payoff
Go -100 500 200
Don’t Go 0 0 0

• The alternative selected is to Go

 Equally Likely Criterion

Outcomes (Demand) Average

Alternatives High Moderate Low Payoff
Large plant 200,000 100,000 -120,000 60,000
Small plant 90,000 50,000 -20,000 40,000
No plant 0 0 0 0

• Choose Large Plant

 Criterion of Realism

• Middle ground between the extremes posed by the

optimist and pessimist criteria
• Use the coefficient of realism (α) to estimate the
decision maker’s optimism
• 0<α<1
• Realism payoff for alternative =
➢ α x (max payoff for alternative) + (1- α) x (min payoff for alternative)
• Choose the alternative with Maximum Realism
payoff
 Criterion of Realism

• Suppose α = 0.25

Outcomes Realism
Alternatives Rain No Rain Payoff
Go -100 500 50
Don’t Go 0 0 0

• The alternative selected is to Go

 Criterion of Realism

• Suppose α = 0.45

Outcomes (Demand) Realism

Alternatives High Moderate Low Payoff
Large plant 200,000 100,000 -120,000 24,000
Small plant 90,000 50,000 -20,000 29,500
No plant 0 0 0 0

• Choose Small Plant

 Minimax Regret Criterion
• Decision alternative which minimizes the maximum regret
• Regret =
(payoff for the best decision for that state of nature) – (payoff for that decision)

Outcomes Regret Table Max

Alternatives Rain No Rain Rain No Rain Regret
Go -100 500 100 0 100
Don’t Go 0 0 0 500 500

• The alternative selected is to Go

 Minimax Regret Criterion
Outcomes (Demand)
Alternatives High Moderate Low
Large plant 200,000 100,000 -120,000
Small plant 90,000 50,000 -20,000
No plant 0 0 0

Regret Table Max

Alternatives High Moderate Low Payoff
Large plant 0 0 120,000 120,000
Small plant 110,000 50,000 20,000 110,000
No plant 200,000 100,000 0 200,000

• Choose Small Plant

 Need for Probabilities

• For better decision making, knowledge of

probabilities is required.
• Probability: Quantitative measure of chance
• EMV (Expected Monetary Value) Criterion
 EMV Criterion

• Expected Monetary Value (EMV) or Expected Payoff uses the

probabilities to calculate the weighted average payoff for
each alternative
– EMV(di) = Σj r(di,cj)pj
• Selects a decision that maximizes the EMV
Outcomes
Alternatives Rain (c1) No Rain (c2) EMV
Go (d1) -100 500 = 0.75 × -100 + 0.25 × 500 = 50
r(di,cj)
Don’t Go (d2) 0 0 = 0.75 × 0 + 0.25 × 0 = 0
p1= 0.75 p2= 0.25

• The alternative selected is to GO

 EMV Criterion

Outcomes (Demand)
Alternatives High Moderate Low EMV
Large plant 200,000 100,000 -120,000 86,000
Small plant 90,000 50,000 -20,000 48,000
No plant 0 0 0 0
p1= 0.3 p2= 0.5 p3= 0.2

• Choose Large Plant

 EOL Criterion

• Expected Opportunity Loss (EOL) uses the probabilities to

calculate the weighted average regret for each alternative
– EOL(di) = Σj l(di,cj)pj
• Selects a decision that minimizes the EOL

Regret Table
Alternatives Rain (c1) No Rain (c2) EOL
Go (d1) 100 0 = 0.75 × 100 + 0.25 × 0 = 75
l(di,cj)
Don’t Go (d2) 0 500 = 0.75 × 0 + 0.25 × 500 = 125
p1= 0.75 p2= 0.25

• The alternative selected is to GO

 EOL Criterion

Regret Table
Alternatives High Moderate Low EOL
Large plant 0 0 120,000 24,000
Small plant 110,000 50,000 20,000 62,000
No plant 200,000 100,000 0 110,000
p1= 0.3 p2= 0.5 p3= 0.2

• Choose Large Plant

 Relationship between EMV and EOL

• Both decision criteria (EMV and EOL) pointed to the same

action alternative
• Will this always be the case?
– Yes
• Suppose Perfect Information(forcast) is available.
– Perfect Information would tell us with certainty which outcome
is going to occur
– Having perfect information before making a decision would
allow choosing the best payoff for the outcome
• Perfect Forecaster would predict
– High 30% of the time
– Medium 50% of the time
– Low 20% of the time
 Relationship between EMV and EOL

Outcomes (Demand)
Alternatives High Moderate Low
Large plant 200,000 100,000 -120,000
Small plant 90,000 50,000 -20,000
No plant 0 0 0
p1= 0.3 p2= 0.5 p3= 0.2

• Make the decision with Highest possible payoffs under the

different market-demand conditions
• Taking the expected value of these best-possible payoffs:
EMV* = EMV(d*) = 0.3 x 200,000 + 0.5 x 100,000 + 0.2 x 0 = 110,000
• Perfect forecasting Max EMV (EMV*)
 Relationship between EMV and EOL

• Note that this idealized expected payoff (or expected

payoff given perfect forecasts) arises because the
corresponding expected opportunity loss is zero.
– Clearly, Max EMV can only be obtained with Min EOL. Thus,
both criteria must point to the same decision alternative
• Now, any expected opportunity loss that is incurred
must come out of forgone expected payoffs.
• EMV + EOL = 110,000
• EMV + EOL = Expected Payoff given Perfect Forecasts
• EV|PI (EVwPI)
= ∑ (probability of outcome) x ( best payoff of outcome)
 The minimum EOL

Regret Table
Alternatives High Moderate Low
Large plant 0 0 120,000
Small plant 110,000 50,000 20,000
No plant 200,000 100,000 0
p1= 0.3 p2= 0.5 p3= 0.2

• Make the decision with zero opportunity loss under the

different market-demand conditions
• Taking the expected value of these best-possible payoffs:
EOL* = EOL(d*) = 0.3 x 0 + 0.5 x 0 + 0.2 x 0 = 0
• Perfect forecasting Zero EOL (EOL*)
Expected Value of Perfect Information (EVPI)

• The amount by which perfect information would

increase our expected payoff
• Provides an upper bound on what to pay for
additional information
EVPI = EV|PI – EMV
EV|PI = Expected value given perfect information
EMV = the best EMV without perfect information
 Expected Value of Perfect Information

EVPI = EV|PI – EMV

= $110,000 - $86,000 = $24,000 = EOL

• EOL = Best (min) EOL without perfect information

• The “perfect information” increases the expected
value by $24,000
• Would it be worth $30,000 to obtain this perfect
information for demand?
 Decision Tree

• Visual/pictorial display of the problem

• Can be used instead of a table to show alternatives,
outcomes, and payoffs
• Chronologically depicts the sequence of actions and
outcomes as they unfold
• Helpful when sequence of decisions are to be made
 Constructing the Decision Tree

• Decision node:
– A decision to be made at that point
– represented by a square

• Event (Chance) node:

– A random event occurs
at that point
– represented by a circle
Example

Representation of a decision problem

in the form of a tree
 Doing the Calculations

Rain
-$100
$50 (0.75)

No rain
$500
$50
(0.25)
a

Rain
$0
$0 (0.75)
c

No rain
$0
(0.25)

Node b: 0.75 × -100 + 0.25 × 500 = 50

Node c: 0.75 × 0 + 0.25 × 0 = 0
 Folding Back a Decision Tree

• For identifying the best decision in the tree

• Work from right to left
• Calculate the expected payoff at each event
node
• Choose the best alternative at each decision
node (based on expected payoff)
 Decision Tree for Thompson Lumber
High Demand (0.30)
$200,000
$86,000
Moderate Demand (0.50)
b $100,000

Low Demand (0.20) -$120,000

High Demand (0.30) $90,000

$86,000 $48,000
Small Plant Moderate Demand (0.50)
a c $50,000

Low Demand (0.20) -$20,000

$0
All Demands
d $0

Node b: 200000 × 0.3 + 100000 × 0.5 - 120000 × 0.2 = $86000

Node c: 90000 × 0.3 + 50000 × 0.5 - 20000 × 0.2 = $48000
 Decision Making with Experimentation Sequential decision making

• Can we get better information??

• Experimentation is the process of collecting more
information in order to obtain a better estimate of
probabilities.
– Better the estimate → better the decision.
• Experiment:
– Call a friend in Delhi to find out if it is cloudy.
– Conduct a market survey
• Experiments revise the estimate of probabilities. The
expected payoff is calculated again with revised
probabilities to get the final decision.
 Expanded Thompson Lumber Example

• Suppose Thompson Lumber has to make two decisions,

with the second decision dependent upon the outcome of
the first.
1. Whether to conduct a market survey at cost $4000 or not.
2. Whether to build a large, small or no plant.
• Survey results will be imperfect (not perfect)
• Depending on survey result decide whether to build a large
plant, small plant, or no plant
 Expanded Thompson Lumber Example

• The results of survey are uncertain: Positive result with

probability 0.57 and Negative result with probability
0.43.
• Additional data is given:
P(High Demand|PS)= 0.509, P(Mod Demand|PS)= 0.468,
P(Low Demand|PS)= 0.023, P(High Demand|NS)= 0.023,
P(Mod Demand|NS)= 0.543, P(Low Demand|NS)= 0.434.
 Thompson Lumber: Optimal Strategy

• Conduct the survey

• If the survey results are positive, then build the
large plant (EMV = $141,840)
• If the survey results are negative, then build the
small plant (EMV = $16,540)
Expected Value of Sample Information (EVSI)
• Experiment provides sample information (not perfect
information)
• What is the value/worth of this sample information?
• EVSI = Gross EV|SI - EMV (w/o any information)
• Why use the Gross EV|SI and not the Net EV|SI?
– Similar to the case of EV|PI where cost of perfect information was not
included.
• EVSI < EVPI
• Efficiency of sample information = EVSI/EVPI
• For Thompson Lumber Co Example
• EV|SI = 87961 + 4000 = 91961
• EVSI = 91961 – 86000 = 5961
• Efficiency of sample information = EVSI/EVPI
= 5961/24000 = 0.248
Expected net gain of experimenting (ENGE)

• ENGE = EVSI - cost of experiment

• If ENGE > 0 conduct the experiment.
• If ENGE < 0 don't conduct the experiment.
• If ENGE = 0 technically, one is indifferent about
conducting the experiment.
 Known Prior Probabilities

P(HD) = 0.30
P(MD) = 0.50
P(LD) = 0.20

• How do we find the revised probabilities where the

survey result is given?
• For example: P(HD|PS) = ?
 Baye's Theorem

• Apply the Conditional probability formula:

P(A|B) = P(A and B) / P(B)
• P(A|B) is the probability of event A occurring, given
that event B has occurred
• When P(A|B) ≠ P(A), this means the probability of
event A has been revised based on the fact that
event B has occurred
Recap of total probability:
• Given an event B, and a set of mutually exclusive and
collectively exhaustive events (B1,B2,…,Bn),
P(B) = P(B/B1).P(B1) + P(B/B2).P(B2) + .... +
P(B/Bn).P(Bn)
• The marketing research firm provided the following
probabilities based on its track record of survey
accuracy:
• P(PS|HD) = 0.967 P(NS|HD) = 0.033
• P(PS|MD) = 0.533 P(NS|MD) = 0.467
• P(PS|LD) = 0.067 P(NS|LD) = 0.933

• Here the demand is “given,” but we need to reverse

the events so the survey result is “given”
• Finding probability of the demand outcome given the
survey result:
P(HD|PS) = P(HD and PS)/ P(PS)
= P(PS|HD) x P(HD)/P(PS)
• Known probability values are in blue, so need to find
P(PS)

P(PS|HD) x P(HD) 0.967 x 0.30

+ P(PS|MD) x P(MD) + 0.533 x 0.50
+ P(PS|LD) x P(LD) + 0.067 x 0.20
P(PS) = = 0.57
• Now we can calculate P(HD|PS):

P(HD|PS) = P(PS|HD) x P(HD) / P(PS)

= 0.967 x 0.30 / 0.5
= 0.509

• The other five conditional probabilities are found in

the same manner
• Notice that the probability of HD increased from 0.30
to 0.509 given the positive survey result