Decision Analysis

1
Problem needs decision
Refurbish Profit ?

An old power plant

Sale Profit ?

Reliability ? Continue
Profit ?
operation

2
Problem formulation
• Influence diagram – a graphical method to show the relationships
among the decisions, the chance of events, and the consequences for
a decision problem.

• Payoff table – a table showing payoffs of all combination of decision

alternative and the states of natures.

• Decision tree – a graphical presentation of the decision-making

process

3
Problem Formulation - Influence diagrams
States of nature
High (s1)
Reliability of
Low (s2)
power plant

Step forward Profit

Decision alternatives Consequence

Refurbish (d1) Profit
Sale (d2)
Continue operation (d3) 4
Problem Formulation – Payoff Tables

Decision alternatives States of nature (Profit in million of RM)

High reliability (s1) Low reliability (s2)
Refurbish (d1) 80 -40
Sale (d2) 50 50
Continue operation (d3) 120 -20

5
Problem Formulation - Decision trees
High (s1)
Refurbish
(d1) 2
Low (s2)

High (s1)
Sale (d2)
1 3
Low (s2)

High (s1)
Continue operation 4
(d3)
Low (s2)
6
Decision Making Without Probabilities
• Optimistic approach
• Conservative approach
• Minimax regret approach

7
Decision Making Without Probabilities
- Optimistic and conservative approach

Decision alternatives States of nature (Profit in million of RM)

High reliability (s1) Low reliability (s2)

Refurbish (d1) 80 -40

Sale (d2) 50 50
Continue operation 120 -20
(d3)

Optimistic approach Conservative approach

- Choose the highest profit from - Choose the highest profit
the maximum payoff column from the minimum payoff
column
8
Decision Making Without Probabilities
- Minimax regret approach
Decision States of nature
alternatives
Strong demand Opportunity Weak demand for Opportunity loss /
for electricity loss / regret electricity (s2) regret
(s1)
Refurbish 80 120 – 80 = 40 -40 50 – (-40) = 90
(d1)
Sale (d2) 50 120 – 50 = 70 50 50 – 50 = 0
Continue 120 120 – 120 = 0 -20 50 – (-20) = 70
operation
(d3)

Maximum payoff Decision alternatives Maximum regret Maximum payoff

Refurbish (d1) 90
Sale (d2) 70
Continue operation (d3) 70
Minimum of
maximum regret
9
Decision Making with Probabilities – Prior
probabilities
If probability of States of Nature are given as P(s1) = 0.2 and P(s2) = 0.8, the Expected
Value without Perfect Information (EVwoPI) can be calculated by:

EVwoPI (d1) = 0.2*80 + 0.8*-40 = -16

EVwoPI (d2) = 0.2*50 + 0.8*50 = 50

EVwoPI (d3) = 0.2*120 + 0.8*-20 = 8

Decision alternatives States of nature (Profit in million of RM)

High reliability (s1) Low reliability (s2)
Refurbish (d1) 80 -40
Sale (d2) 50 50
Continue operation (d3) 120 -20
10
Decision Making with Probabilities – Prior probabilities decision tree
80
High (s1)
Refurbish P(s1) = 0.2
(d1) 2
P(s2) = 0.8
Low (s2)
-40
50
High (s1)
Sale (d2) P(s1) = 0.2
1 3
P(s2) = 0.8
Low (s2)
50
120
High (s1)
P(s1) = 0.2
Continue operation 4
(d3) P(s2) = 0.8
Low (s2)
-20
11
Decision Making with Probabilities – Expected Value without Perfect Information
(EVwoPI) approach using decision tree

Refurbish EVwoPI (d1) = 0.2*80 + 0.8*-40 = -16

(d1) 2

Sale (d2)
1 3 EVwoPI (d2) = 0.2*50 + 0.8*50 = 50

Continue operation 4 EVwoPI (d3) = 0.2*120 + 0.8*-20 = 8

(d3)

12
Decision Making with Probabilities – Expected Value with Perfect Information

If probability of States of Nature determined as P(s1) = 0.2 and P(s2) = 0.8, the
Expected Value with Perfect Information (EVwPI) can be calculated by:

EVwPI = 0.2*120 + 0.8*50= 64 EVwoPI (d1) = 0.2*80 + 0.8*-40 = -16

EVwoPI (d2) = 0.2*50 + 0.8*50 = 50

EVPI = = 64 – 50 = 14
EVwoPI (d3) = 0.2*120 + 0.8*-20 = 8

Decision alternatives States of nature (Profit in million of RM)

High reliability (s1) Low reliability (s2)
Refurbish (d1) 80 -40
Sale (d2) 50 50
Continue operation (d3) 120 -20
13
Decision Making with Probabilities – Expected Opportunity Loss
EOL (d1) = 0.2*40 + 0.8*90 = 80

EOL (d2) = 0.2*70 + 0.8*0 = 14 Best decision

alternative
EOL (d3) = 0.2*0 + 0.8*70= 56
Decision States of nature
alternative
s High reliability Opportunity Low reliability (s2) Opportunity loss /
(s1) loss / regret regret

Refurbish 80 120 – 80 = 40 -40 50 – (-40) = 90

(d1)
Sale (d2) 50 120 – 50 = 70 50 50 – 50 = 0
Continue 120 120 – 120 = 0 -20 50 – (-20) = 70
operation
(d3)
14
 Risk analysis
If we choose Sale (d2):
- the probability to get RM50 million for s1 is 0.2
- the probability to get RM50 million for s2 is 0.8

In both cases (s1 and s2), the profit of RM 50 million is confirmed!

Are we satisfied with this profit?

Decision alternatives States of nature (Profit in million of RM)
High reliability (s1) Low reliability (s2)
Refurbish (d1) 80 -40
Sale (d2) 50 50
Continue operation (d3) 120 -20

15
 Sensitivity Analysis
Continue operation (d3)
Plot EV(dx) for x = 1, 2 and 3 with 0≤Ps1≤1
140 Sale (d2)

120 EV (d1) = Ps1*80 + (1-Ps1)*-40

100 EV (d2) = Ps1*50 + 0.8*50

80 EV (d3) = Ps1*120 + 0.8*-20

Expected Value (EV)

60
d1
EV (d1)
40
d2
EV (d2)
20 d3
EV (d3)

0
0 0.2 0.4 0.6 0.8 1
-20
Do we want to get more
-40 information on the probability
of high reliability and low
-60 reliability of the power plant?
ps1
16
 Decision Analysis
Quiz 1
To absorb some short-term excess production capacity at its Arizona plant, Special Instrument Products is
considering a short manufacturing run for either of two new products, a temperature sensor or a pressure
sensor. The market for each product is known if the products can be successfully developed. However, there
is some chance that it will not be possible to successfully develop them. Revenue of $1,000,000 would be
realized from selling the temperature sensor and revenue of $400,000 would be realized from selling the
pressure sensor. Both of these amounts are net of production cost but do not include development cost. If
development is unsuccessful for a product, then there will be no sales, and the development cost will be
totally lost. Development cost would be $100,000 for the temperature sensor and $10,000 for the pressure
sensor.
To illustrate the analysis approach, Sketch the decision tree used in this exercise to help make a decision.

17
 Decision Making with Probabilities – Posterior probability
Appoint a consultant to conduct study and produce a report. The report can be:
1. Favorable report
2. Unfavorable report

Report from Reliability of

States of nature
consultant power plant High (s1)
Low (s2)

Research by consultant Step forward Profit

Decision alternatives Consequence

Refurbish (d1)
Sale (d2) Profit
Continue operation (d3)
18
Decision Making with Probabilities – Expected Value approach using decision tree EV(Node 6) = 0.95(80)+0.05(-40) =74
High (s1) 80 EV(Node 7) = 0.95(50)+0.05(50) = 50
Refurbish (d1) EV(Node 8) = 0.95(120)+0.05(-20) = 113
6
Favorable report Low (s2) EV(Node 9) = 0.25(80)+0.75(-40) = -10
-40 EV(Node 10) = 0.25(50)+0.75(50) = 50
(P = 0.7)
50 EV(Node 11) = 0.25(120)+0.75(-20) = 15
3 Sale (d2) High (s1)
EV(Node 12) = 0.2(80)+0.8(-40) = -16
Report by 7 EV(Node 13) = 0.2(50)+0.8(50) = 50
Low (s2) EV(Node 14) = 0.2(120)+0.8(-20) = 8
consultant 50
2 Continue operation 120 80
High (s1) High (s1)
(d3) Refurbish (d1)
8 9
Low (s2) Low (s2) -40
-20
Unfavorable report
50
P = 0.3 4
Sale (d2) High (s1)
1 10
Low (s2)
50
120
80 High (s1)
High (s1)
Refurbish (d1) 11
12 Low (s2)
-20
Low (s2)
No report by -40
PFavorable report = 0.7
consultant 5 High (s1) 50
Sale (d2) PUnfavorable report = 0.3
13
Low (s2)
50 P(High reliability given Favorable report) = 0.95
Continue operation P(Low reliability given Favorable report) = 0.05
120
(d3) High (s1)
14 P(High reliability given Unfavorable report) = 0.25
Low (s2) P(Low reliability given Unfavorable report) = 0.75
-20
If no report from consultant, P(s1) = 0.2, P(s2) = 0.8 19
Decision Making with Probabilities – Expected Value approach using decision tree EV(Node 6) = 0.95(80)+0.05(-40) =74
EV(Node 7) = 0.95(50)+0.05(50) = 50
Refurbish (d1) EV(Node 8) = 0.95(120)+0.05(-20) = 113
6 74
Favorable report EV(Node 9) = 0.25(80)+0.75(-40) = -10
(P = 0.7) EV(Node 10) = 0.25(50)+0.75(50) = 50
EV(Node 11) = 0.25(120)+0.75(-20) = 15
3 Sale (d2) EV(Node 12) = 0.2(80)+0.8(-40) = -16
50
Report by 7 EV(Node 13) = 0.2(50)+0.8(50) = 50
EV(Node 14) = 0.2(120)+0.8(-20) = 8
consultant
2 Continue operation
(d3) Refurbish (d1)
8 113 9 -10

Unfavorable report
P = 0.3 4
Sale (d2)
1 10 50

Continue operation
(d3)
Refurbish (d1) 11 15
12 -16
No report by
PFavorable report = 0.7
consultant 5
Sale (d2) PUnfavorable report = 0.3
13 50
P(High reliability given Favorable report) = 0.95
Continue operation P(Low reliability given Favorable report) = 0.05
(d3)
14 8 P(High reliability given Unfavorable report) = 0.25
P(Low reliability given Unfavorable report) = 0.75

If no report from consultant, P(s1) = 0.2, P(s2) = 0.8 20

Decision Making with Probabilities – Expected Value approach using decision tree EV(Node 6) = 0.95(80)+0.05(-40) =74
EV(Node 7) = 0.95(50)+0.05(50) = 50
EV(Node 8) = 0.95(120)+0.05(-20) = 113
Favorable report EV(Node 9) = 0.25(80)+0.75(-40) = -10
(P = 0.7) EV(Node 10) = 0.25(50)+0.75(50) = 50
EV(Node 11) = 0.25(120)+0.75(-20) = 15
3
EV(Node 12) = 0.2(80)+0.8(-40) = -16
Report by EV(Node 13) = 0.2(50)+0.8(50) = 50
EV(Node 14) = 0.2(120)+0.8(-20) = 8
consultant
2 Continue operation
(d3)
8 113

Unfavorable report
P = 0.3 4
Sale (d2)
1 10 50

No report by
PFavorable report = 0.7
consultant 5
Sale (d2) PUnfavorable report = 0.3
13 50
P(High reliability given Favorable report) = 0.95
P(Low reliability given Favorable report) = 0.05

P(High reliability given Unfavorable report) = 0.25

P(Low reliability given Unfavorable report) = 0.75

If no report from consultant, P(s1) = 0.2, P(s2) = 0.8 21

Decision Making with Probabilities – Expected Value approach using decision tree

Expected Value of Sample Information (EVSI)

EVSI = = 79.1 – 50 = 29.1

Report by Continue operation (d3)
consultant
2 EV (Node 2) =0.7*113 = 79.1 Efficiency of sample information

E = EVSI/EVPI = 29.1/14 =2.08 or 208%

No report by
consultant PFavorable report = 0.7
5
Sale (d2) PUnfavorable report = 0.3
13 50
P(Strong demand given Favorable report) = 0.95
P(Weak demand given Favorable report) = 0.05

P(Strong demand given Unfavorable report) = 0.25

P(Weak demand given Unfavorable report) = 0.75
22
If no market study, P(s1) = 0.2, P(s2) = 0.8
Decision Making with Probabilities – Risk analysis

Favorable report Profit Probability

(P = 0.7) 120 0.665
3 -20 0.035
50 0.075
Report by
50 0.225
consultant
2 Continue operation 120
(d3) High (s1)
8
Low (s2)
-20
Unfavorable report
50
P = 0.3 4
Sale (d2) High (s1)
1 10
0.7 Low (s2)
50
0.6
0.5
Probability

0.4
0.3 PFavorable report = 0.7
PUnfavorable report = 0.3
0.2
P(Strong demand given Favorable report) = 0.95
0.1 P(Weak demand given Favorable report) = 0.05
0
P(Strong demand given Unfavorable report) = 0.25
-20 30 80 P(Weak demand given Unfavorable report) = 0.75
Profit
Payoff
With Perfect Information With Sample Information If no market study, P(s1) = 0.2, P(s2) = 0.8 23