College of Science
Department of Statistics & OR
OR 122
Biostatistics
Chapter 8:
Decision Theory
Structure of Decision-Making Problem
1) The decision maker: who is charged with the responsibility for making
decision-selection of one from a set of possible courses of action.
2) Acts: Are the alternative courses of action or strategies, that are available to
decision maker.
⇒ The decision maker has a control over choice of these acts.
3) States of nature (events): Determine the level of success for a given act.
⇒ The decision-maker has no control on them.
4) Uncertainty: It is indicated in terms of probabilities assigned to events.
Structure of Decision-Making Problem
5) Types of information:
I) Perfect information on problem.
⇒ In this case, we have decision making under certainty.
II) Partial or imperfect information.
⇒ In this case, we have two types of decision-making situations that are:
i) Decision under risk.
ii) Decision under uncertainty.
Structure of Decision-Making Problem
6) Payoff table: it takes the following form
States of nature
Alternatives 𝑆1 𝑆2 ⋯ 𝑆𝑛
𝑃1 𝑃2 ⋯ 𝑃𝑛
𝐴1 𝑟11 𝑟12 ⋯ 𝑟1𝑛
𝐴2 𝑟21 𝑟22 ⋯ 𝑟2𝑛
⋮ ⋮ ⋮ ⋮
𝐴𝑚 𝑟𝑚1 𝑟𝑚2 ⋯ 𝑟𝑚𝑛
Where 𝑃𝑗 ; 𝑗 = 1,2, … , 𝑛, is the probability to each state of nature, and 𝑟𝑖𝑗 ; 𝑖 =
, 1,2, … , 𝑚, 𝑗 = 1,2, … , 𝑛 is the payoff due to selecting the alternative 𝑖.
Example 1
A manger has three devices that are defect A, B, and C. He has only three
technicians to fix these devices. Experiences show that the following table
presents the time taken to fix these devices by the three technicians. Suppose
the manger will assign only one for each device. So, what is the best choice for
him to get a minimum time of fixing the devices:
Devices
Technicians A B C
J 3 7 4
G 4 6 6
M 3 8 5
Solution
This is a problem of perfect information. So, we have the following alternatives:
𝑎1 : 𝐽 𝐴 , 𝐺 𝐵 , 𝑀 𝐶 ⇒ 𝑡1 = 3 + 6 + 5 = 14
𝑎2 : 𝐽 𝐴 , 𝐺 𝐶 , 𝑀 𝐵 ⇒ 𝑡2 = 3 + 6 + 8 = 17
𝑎3 : 𝐽 𝐵 , 𝐺 𝐴 , 𝑀 𝐶 ⇒ 𝑡3 = 7 + 4 + 5 = 16
𝑎4 : 𝐽 𝐵 , 𝐺 𝐶 , 𝑀 𝐴 ⇒ 𝑡4 = 7 + 6 + 3 = 16
𝑎5 : 𝐽 𝐶 , 𝐺 𝐵 , 𝑀 𝐴 ⇒ 𝑡5 = 4 + 6 + 3 = 13
𝑎6 : 𝐽 𝐶 , 𝐺 𝐴 , 𝑀 𝐵 ⇒ 𝑡6 = 4 + 4 + 8 = 16
So, the best alternative is to assign 𝐽 → 𝐶, 𝐺 → 𝐵, 𝑀 → 𝐴 with minimum time 𝑡5 = 13.
Decisions under risk:
In this situation, the decision-maker possesses some measures to identify the
best alternative.
1) The expected payoff criterion (E)
2) The expected opportunity loss criterion (EOL)
3) The criterion of most likelihood.
Example 2
An investor wants to invest an amount of money during a year. He has the
following alternatives and the states of nature with their probability:
State of nature
Inflation (0.2) Recession (0.3) Growth (0.5)
Furniture (𝑎1 ) 7 8 12
Stock market (𝑎2 ) -2 10 25
Cars (𝑎3 ) 6.5 8.5 16.5
What is the best alternative according to the expected payoff criterion.
Find the expectation, then take the max value if the question about profit or
take the min value if the question about cost.
Solution
𝐸 𝑎1 = 7 0.2 + 8 0.3 + 12 0.5 = 9.8
𝐸 𝑎2 = −2 0.2 + 10 0.3 + 25 0.5 = 15.1
𝐸 𝑎1 = 6.5 0.2 + 8.5 0.3 + 16.5 0.5 = 12.1
So, the best alternative for the investor is to invest the money in stock market as
𝐸 𝑎2 = 15.1 is the greatest value
Example 3
Solve the previous example using the other two criteria.
Solution:
State of nature State of nature
𝑆1 (0.2) 𝑆2 (0.3) 𝑆3 (0.5) 𝑆1 (0.2) 𝑆2 (0.3) 𝑆3 (0.5)
𝑎1
𝑎2
7-7
7-(-2)
10-8
10-10
25-12
25-25 ⇒ 𝑎1
𝑎2
0
9
2
0
13
0
𝑎3 7-6.5 10-8.5 25-16.5 𝑎3 0.5 1.5 8.5
Expected opportunity loss criterion: Construct the regret table. Then, find the
expectation, then take the min value if the question about profit, or cost.
Solution
𝐸𝑂𝐿 𝑎1 = 0 0.2 + 2 0.3 + 13 0.5 = 7.1
𝐸𝑂𝐿 𝑎2 = 9 0.2 + 0 0.3 + 0 0.5 = 1.8
𝐸𝑂𝐿 𝑎1 = 0.5 0.2 + 1.5 0.3 + 8.5 0.5 = 4.8
According to this measure we have the best alternative 𝑎2 as it has the minimum EOL.
Solution
Using the third criterion (most likelihood):
Consider the highest probability column, then take the max for the profit table or take
the min for the cost table.
𝑆3 (0.5)
𝑎1 12
𝑎2 25
𝑎3 16.5
The highest probability is 0.5 for the growth state, from the opposite table we have 𝑎2
with the highest payoff. Then, 𝑎2 is the best alternative.
Example 4
For the following cost table find the best alternative?
State of nature
𝑆1 (0.3) 𝑆2 (0.1) 𝑆3 (0.4) 𝑆4 (0.2)
𝑎1 8 9 5 12
𝑎2 10 12 6 12
𝑎3 17 5 8 15
Solution
𝐸 𝑎1 = 8 0.3 + 9 0.1 + 5 0.4 + 12 0.2 = 7.70
𝐸 𝑎2 = 10 0.3 + 12 0.1 + 6 0.4 + 12 0.2 = 9
𝐸 𝑎1 = 17 0.3 + 5 0.1 + 8 0.4 + 15 0.2 = 11.80
According to the expected payoff criterion 𝑎1 is the best alternative.
Solution
State of nature
𝑆1 (0.3) 𝑆2 (0.1) 𝑆3 (0.4) 𝑆4 (0.2)
𝑎1 0 4 0 0
𝑎2 2 7 1 0
𝑎3 9 0 3 3
𝐸𝑂𝐿 𝑎1 = 0 0.3 + 4 0.1 + 0 0.4 + 0 0.2 = 0.4
𝐸𝑂𝐿 𝑎2 = 2 0.3 + 7 0.1 + 1 0.4 + 0(0.2) = 1.7
𝐸𝑂𝐿 𝑎1 = 9 0.3 + 0 0.1 + 3 0.4 + 3(0.2) = 4.50
According to the EOL we have 𝑎1 is the best alternative.
Solution
Using the third criterion (most likelihood):
𝑆3 (0.4)
𝑎1 5
𝑎2 6
𝑎3 8
We have 𝑎1 is the best alternative, because it has the minimum cost.
Suppose the following matrix, find the best alternative by using the most
likelihood criterion.
𝑆1 (0.4) 𝑆2 (0.4) 𝑆3 (0.2)
𝑎1 1 3 2
𝑎2 1 4 3
𝑎3 2 5 7
We need to find the average for the first and second columns then take the max
for the profit table or take min for the cost table [𝑎1 (cost), 𝑎3 (profit)].
Decision under uncertainty
❖ The decision-maker in this types does not have any information about the state of nature (events).
Instead, he has only these states.
❖Examples are the new phenomena such as increasing prices of Gold, Oli, Gas, …, etc.
❖For this type we have 5 criteria that are:
1) Laplace criterion: the probability of all states are equal.
For profit we use maximin.
2) Pessimistic criterion:
For cost we use minimax.
For profit we use maximax.
3) Optimistic criterion:
For cost we use minimin.
4) Hurwicz criterion.
5) Savage criterion: we apply minimax on the regret table.
Hurwicz Criterion
He gives a probability of optimistic (𝛼) and then evaluate the expected value for
each alternative as follows:
maximum profit low profit
𝑉(𝑎𝑖 ) = α ∗ + (1 − α)
minimum cost maximum cost
The following examples show how can we use these criteria of decision under
uncertainty.
Example 5
What is the best alternative for the following cost table.
𝑆1 𝑆2 𝑆3 𝑆4
𝑎1 5 8 3 1
𝑎2 7 4 5 2
𝑎3 3 6 6 4
Solution
1) Using Laplace:
Let 𝑃 𝑆1 = 𝑃 𝑆2 = 𝑃 𝑆3 = 𝑃 𝑆4
1
⇒ 𝑃 𝑆𝑖 = ; 𝑖 = 1,2,3,4.
4
1 17
𝐸 𝑎1 = 5+8+3+1 = = 4.25
4 4
1 18
𝐸 𝑎2 = 7+4+5+2 = = 4.50
4 4
1 19
𝐸 𝑎3 = 3+6+6+4 = = 4.75
4 4
According to Laplace 𝑎1 is the optimum alternative.
Solution
2) Using the pessimistic criterion:
Worst (cost)
𝑎1 8
𝑎2 7
𝑎3 6
This means that minimax = 6. So, 𝑎3 is the optimum alternative.
Solution
3) Using the optimistic criterion:
Low (cost)
𝑎1 1
𝑎2 2
𝑎3 3
This means that minimin = 1. So, 𝑎1 is the optimum alternative.
4) Using Hurwicz with 𝛼 = 0.4 ⇒ 1 − 𝛼 = 0.6
V 𝑎1 = 1 0.4 + 8 0.6 = 5.2
V 𝑎2 = 2 0.4 + 7 0.6 = 5
V 𝑎3 = 3 0.4 + 6 0.6 = 4.8
So, 𝑎3 is the optimum alternative.
5) Using Savage criterion:
We construct the regret table first.
𝑆1 𝑆2 𝑆3 𝑆4 𝑆1 𝑆2 𝑆3 𝑆4
Max
𝑎1 5-3 8-4 3-3 1-1 𝑎1 2 4 0 0
⇒ ⇒
𝑎1 4
𝑎2 7-3 4-4 5-3 2-1 𝑎2 4 0 2 1 𝑎2 4
𝑎3 3
𝑎3 3-3 6-4 6-3 4-1 𝑎3 0 2 3 3
So, 𝑎3 is the optimum alternative.
Example 6
What is the optimum alternative for the following profit table.
𝑆1 𝑆2 𝑆3
𝑎1 7 2 -1
𝑎2 3 6 2
𝑎3 0 3 8
Solution
1) Using Laplace:
Let 𝑃 𝑆1 = 𝑃 𝑆2 = 𝑃 𝑆3
1
⇒ 𝑃 𝑆𝑖 = ; 𝑖 = 1,2,3.
3
1 8
𝐸 𝑎1 = 7 + 2 − 1 = ≅ 2.7
3 3
1 11
𝐸 𝑎2 = 3+6+2 = ≅ 3.7
3 3
1 11
𝐸 𝑎3 = 0+3+8 = ≅ 3.7
3 3
Both 𝑎2 and 𝑎3 are optimum alternatives.
Solution
2) Using the pessimistic criterion:
Worst (profit)
𝑎1 -1
𝑎2 2
𝑎3 0
This means that maximin = 2. So, 𝑎2 is the optimum alternative.
Solution
3) Using the optimistic criterion:
High (profit)
𝑎1 7
𝑎2 6
𝑎3 8
This means that maximax = 8. So, 𝑎3 is the optimum alternative.
4) Using Hurwicz with 𝛼 = 0.4 ⇒ 1 − 𝛼 = 0.6
V 𝑎1 = 7 0.4 + (−1) 0.6 = 2.2
V 𝑎2 = 6 0.4 + 2 0.6 = 3.6
V 𝑎3 = 8 0.4 + (0) 0.6 = 3.2
So, 𝑎2 is the optimum alternative.
5) Using Savage criterion:
We construct the regret table first.
𝑆1 𝑆2 𝑆3 𝑆1 𝑆2 𝑆3
Max
𝑎1 7-7 6-2 8-(-1) 𝑎1 0 4 9
⇒ ⇒
𝑎1 9
𝑎2 7-3 6-6 8-2 𝑎2 4 0 6 𝑎2 6
𝑎3 7
𝑎3 7-0 6-3 8-8 𝑎3 7 3 0
This means that, 𝑎2 is the optimum alternative.
Example 7
Hurwicz Criterion:
Consider the following profit table:
𝑆1 𝑆2 𝑆3
𝑎1 3 6 -1
𝑎2 8 5 4
𝑎3 -4 7 12
What is the probability of optimism that makes 𝑎1 the optimum alternative?
𝛼 = ?? 0 ≤ 𝛼 ≤ 1 ⇒ (1)
V 𝑎1 = 6 𝛼 + −1 1 − 𝛼 = 7𝛼 − 1
V 𝑎2 = 8 𝛼 + 4 1 − 𝛼 = 4𝛼 + 4
V 𝑎3 = 12 𝛼 + (−4) 1 − 𝛼 = 16 𝛼 − 4
𝐴∗ = 𝐴1
⇒ 𝑉 𝐴1 > 𝑉 𝐴2 ⇒ 7𝛼 − 1 > 4𝛼 + 4 ⇒ 3𝛼 > 5 ⇒ 𝛼 > 1.67 ⇒ (2)
and, 𝑉 𝐴1 > 𝑉 𝐴3 ⇒ 7𝛼 − 1 > 16 𝛼 − 4 ⇒ 9𝛼 < 3 ⇒ 𝛼 < 0.33 ⇒ (3)
There is no value for 𝛼 that makes 𝐴1 optimum solution. There is no 𝛼 that
satisfy (1), (2), and (3)
0 0.33 1 1.67
What is the probability of optimism that makes 𝑎3 the optimum alternative?
𝛼 = ?? 0 ≤ 𝛼 ≤ 1 ⇒ (1)
V 𝑎1 = 6 𝛼 + −1 1 − 𝛼 = 7𝛼 − 1
V 𝑎2 = 8 𝛼 + 4 1 − 𝛼 = 4𝛼 + 4
V 𝑎3 = 12 𝛼 + (−4) 1 − 𝛼 = 16 𝛼 − 4
𝐴∗ = 𝐴3
⇒ 𝑉 𝐴3 > 𝑉 𝐴1 ⇒ 16 𝛼 − 4 > 7𝛼 − 1 ⇒ 9𝛼 > 3 ⇒ 𝛼 > 0.33 ⇒ (2)
and, 𝑉 𝐴3 > 𝑉 𝐴2 ⇒ 16 𝛼 − 4 > 4𝛼 + 4 ⇒ 12𝛼 > 8 ⇒ 𝛼 < 0.67 ⇒ (3)
For all 0.67 < 𝛼 ≤ 1 ⇒ 𝐴∗ = 𝐴3
Example 8
Consider the following cost table:
𝑆1 𝑆2 𝑆3 𝑆4
𝑎1 8 9 5 12
𝑎2 10 12 6 12
𝑎3 17 5 8 15
What is the probability of optimism that makes 𝑎1 the optimum alternative?
𝛼 = ?? 0≤𝛼≤1
V 𝑎1 = 5 𝛼 + 12 1 − 𝛼 = −7𝛼 + 12
V 𝑎2 = 6 𝛼 + 12 1 − 𝛼 = −6𝛼 + 12
V 𝑎3 = 5 𝛼 + 17 1 − 𝛼 = −12𝛼 + 17
𝐴∗ = 𝐴1
⇒ 𝑉 𝐴1 < 𝑉 𝐴2 ⇒ −7𝛼 + 12 < −6𝛼 + 12 ⇒ −𝛼 < 0 ⇒ 𝛼 > 0
and, 𝑉 𝐴1 < 𝑉 𝐴3 ⇒ −7𝛼 + 12 < −12𝛼 + 17 ⇒ 5𝛼 < 5 ⇒ 𝛼 < 1
For all 0 < 𝛼 < 1 ⇒ 𝐴∗ = 𝐴1
What is the probability of optimism that makes 𝑎2 the optimum alternative?
𝛼 = ?? 0≤𝛼≤1
V 𝑎1 = 5 𝛼 + 12 1 − 𝛼 = −7𝛼 + 12
V 𝑎2 = 6 𝛼 + 12 1 − 𝛼 = −6𝛼 + 12
V 𝑎3 = 5 𝛼 + 17 1 − 𝛼 = −12𝛼 + 17
𝐴∗ = 𝐴2
⇒ 𝑉 𝐴2 < 𝑉 𝐴1 ⇒ −6 𝛼 + 12 < −7𝛼 + 12 ⇒ 𝛼 < 0
and, 𝑉 𝐴2 < 𝑉 𝐴3 ⇒ −6𝛼 + 12 < −12𝛼 + 17 ⇒ 6𝛼 < 5 ⇒ 𝛼 < 0.83
There is no solution that makes 𝐴2 optimum solution.
