Decision Analysis

Problem Formulation
simplifying complex problem
Decision Making without Probabilities
Decision Making with Probabilities
Problem Formulation
first step in the decision analysis process
verbal statement of the problem
1. identify decision altenatives
2. state of nature (uncertain future events)
3. payoff (consequences) associataed with each combination of:
decision altenative
state of nature

3 decision alternatives
d1 using building design A
d2 using building design B
d3 using building design C

3 state of nature
s1 40
s2 60
s3 80

Payoff table
the consequence resulting from a specific combinatio
a table showing payoffs
payoffs can be expressed in tems of profit, cost, time
Payoff table (payoffs are profit per week)
Average Number of Customers Per Hour
s1=40 s2=60
d1 10,000 15,000
d2 8,000 18,000
d3 6,000 16,000

Influence Diagram
graphical decvise showing relationships among decisi
squares or rectangles for decision nodes (DA)
circles or ovals depic state of nature chance nodes
diamonds for consequence or payoffs
line or arcs connecting the nodes and direction
 Decision Tree
chronological representation of the decision problem
a decision tree has 2 types of nodes:
round nodes
square nodes
branches leaving round node
branches leaving square nodes
at the end are payoffs

Decision Making without Probabilities

criteria for decision making
three commonly used criteria for decision when prob of states of nature is unavailable
1. maximax approach optimistic
2. maximin approach conzervative
3. minimax regret approach

Optimistic Approach
used by an optimistic decision maker
choose alternative that will give the overall largest payoff if the goal is maximizing profit
if the payoff table is in tems of costs, then minimum yung pipiliin
 d3-s3 21,000

Conservative Approach
used by a conservative decision maker
for each decision, identify the minimum payoff is listed
maximum of the minimum payoff is selected
if payoffs are intems of cost, then baliktad

worst payoffs
d1 10,000
d2 8,000
d3 6,000

Minimax regret approach

requires the construction of a regret table or an opportunity loss table
done by calculating for each state of nature the difference between each payoff and tehe largest payoff for
then using the regret table, the maximum regret for each possible decision is listed
the decision corresponding
Decision Making with Probabilities
Assigning probabilities
once we defined DA and SN for the chance events, we focus on detemining prob of SN

Classical
no. events can happen/ all possible events
Relative frequency
using past data
Subjective
what we feel

Expected Value Approach

identify the best or recommended decision alternative
expected value of each decision alternative is calculated
decision altenative yielding the best expected value if chosen
profit largest
cost smallest

EV(d1) 12600
EV (d2) 11,600
EV (d3) 14,000
Expected Value of Perfect Information
frequently information is available which can improve the probability estimates for the states of nature
the expected value of perfect information (EVPI)

EVwoPI 14,000
EVwPI ?

1 optimal returns
s1 10,000
s2 18,000
s3 21,000
2 expected value of optimal returns
s1 10,000 0.4 4000
s2 18,000 0.2 3600
s3 21,000 0.4 8400
EVwPI 16000
EVwoPI 14,000
EVPI 2,000
increase in EV if we have PI
2,000 is the increase in the expected profit that would result
 using building design A
using building design B
using building design C

ulting from a specific combination of a decision altenative and state of nature

ssed in tems of profit, cost, time, distance or any other appropriate measure
are profit per week)
Customers Per Hour
s3=80
14,000
12,000
21,000

owing relationships among decisions, chance events, consequences

s for decision nodes (DA)
state of nature chance nodes
quence or payoffs
ng the nodes and direction
entation of the decision problem
types of nodes:
chance events
decisions
leaving round node state of nature
leaving square nodes alternatives
d are payoffs

imizing profit
ayoff and tehe largest payoff for that state of nature
g prob of SN
mates for the states of nature

t that would result

Burger Prince Restaurant Problem
Payoff Table (Payoffs are Profit Per Week)

Average Number of Customers Per Hour sum product

s1=40 s2=60 s3=80
Probability 0.4 0.2 0.4 Row Maximum Row Minimum Expected Value
Design A (d1) 10,000.00 15,000.00 14,000.00 15,000.00 10,000.00 12,600.00
Design B (d2) 8,000.00 18,000.00 12,000.00 18,000.00 8,000.00 11,600.00
Design C (d3) 6,000.00 16,000.00 21,000.00 21,000.00 6,000.00 14,000.00
Column Maxim 10,000.00 18,000.00 21,000.00

Regret/Loss Table
Average Number of Customers Per Hour min the max regret
s1=40 s2=60 s3=80 Decision Approach
Probability 0.4 0.2 0.4 Row Maximum Expected Loss Minimax Reg
Design A (d1) - 3,000.00 7,000.00 7,000.00 3,400.00 Not optimal
Design B (d2) 2,000.00 - 9,000.00 9,000.00 4,400.00 Not optimal
Design C (d3) 4,000.00 2,000.00 - 4,000.00 2,000.00 Design C (d3)

Expected Value without Perfect Information (EVwoPI) 14,000.00

Expected Value with Perfect Information (EVwPI) 16,000.00
Expected Value Of Perfect Information (EVPI) 2,000.00
 best of the best best of the worst payoffs
Decision Approach
Optimistic Conservative Expected Value
Not optimal Design A (d1) Not optimal
Not optimal Not optimal Not optimal
Design C (d3) Not optimal Design C (d3)

the max regret

Decision Approach
Expected Loss
Not optimal
Not optimal
Design C (d3)
minimum of the maximum regret
A decision maker
Payoff Table (Payoffs are Profit Per Week)

Demand sum product

Low s1 Medium s2 High s3
Probability 0.2 0.5 0.3 Row Maximum Row Minimum Expected Value
d1 12.00 15.00 15.00 15.00 12.00 14.40
d2 9.00 25.00 35.00 35.00 9.00 24.80
d3 6.00 35.00 60.00 60.00 6.00 36.70
d4 3.00 45.00 90.00 90.00 3.00 50.10
Column Maxim 12.00 45.00 90.00

Regret/Loss Table
Average Number of Customers Per Hour min the max regret
Low s1 Medium s2 High s3 Decision Approach
Probability 0.2 0.5 0.3 Row Maximum Expected Loss Minimax Reg
d1 - 30.00 75.00 75.00 37.50 Not optimal
d2 3.00 20.00 55.00 55.00 27.10 Not optimal
d3 6.00 10.00 30.00 30.00 15.20 Not optimal
d4 9.00 - - 9.00 1.80 d4
Column Maxim 9.00 30.00 75.00
Expected Value without Perfect Information (EVwoPI) 50.10
Expected Value with Perfect Information (EVwPI) 51.90
Expected Value Of Perfect Information (EVPI) 1.80
 best of the best best of the worst payoffs
Decision Approach
Optimistic Conservative Expected Value
Not optimal d1 Not optimal
Not optimal Not optimal Not optimal
Not optimal Not optimal Not optimal
d4 Not optimal d4

the max regret

Decision Approach
Expected Loss
Not optimal
Not optimal
Not optimal
d4
 Decision Alternatives
d1 stock 10 cases minimum stock option
d2 stock 11 cases
d3 stock 12 cases
State of Nature
s1 demand for 10 cases
s2 demand for 11 cases
s3 demand for 11 cases

s1 s2 s3
10 cases 11 cases 12 cases cost per case 3
stock 10 $50 $50 $50 selling price 8
stock 11 $47 $55 $55 profit 5
stock 12 $44 $52 $60
Beth's Problem
Profit Table

Daily Demand sum product

s1=40 s2=60 s3=80 Row Row Expected
Probability 0.23 0.43 0.33 Maximum Minimum Value
Stock 10 (d1) 50.00 50.00 50.00 50.00 50.00 50.00
Stock 11 (d2) 47.00 55.00 55.00 55.00 47.00 53.13
Stock 12 (d3) 44.00 52.00 60.00 60.00 44.00 52.80
Column Maximum 50.00 55.00 60.00

Regret/Loss Table
Average Number of Customers Per Hour min the max regret
s1=40 s2=60 s3=80 Row Decision Approach
Probability 0.233333333 0.433333333 0.33333333333 Maximum Expected Loss Minimax Reg
Stock 10 (d1) - 5.00 10.00 10.00 5.50 Not optimal
Stock 11 (d2) 3.00 - 5.00 5.00 2.37 Stock 11 (d2)
Stock 12 (d3) 6.00 3.00 - 6.00 2.70 Not optimal

Expected Value without Perfect Information (EVwoPI) 53.13

Expected Value with Perfect Information (EVwPI) 55.50
Expected Value Of Perfect Information (EVPI) 2.37
 best of the best pbest of the worst payoffs
Decision Approach
Optimistic Conservative Expected Value
Not optimal Stock 10 (d1) Not optimal
Not optimal Not optimal Stock 11 (d2)
Stock 12 (d3) Not optimal Not optimal

n the max regret

Decision Approach
Expected Loss
Not optimal
Stock 11 (d2)
Not optimal