This action might not be possible to undo. Are you sure you want to continue?

**Decision Analysis I November 8th. 2004
**

Page 1 oI 12 HW#4 Solutions

Homework Assignment #4 - Solutions

0

5

10

15

20

25

-INF - 0 0.5 - 1 1.5 - 2 2.5 - 3 3.5 - 4 4.5 - 5 5.5 - 6 6.5 - 7 7.5 - 8 8.5 - 9 9.5 - 10

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

1.00

Number of Grades within Range Cumulative

Score on HW Question (./1)

-2.00 -1.50 -1.00 -0.50 0.00 0.50 1.00

Question 1

10% 50%90%

Question 2

10% 50% 90%

Question 3

10% 50% 90%

Question 4

10% 50%90%

Question 5

10% 50% 90%

Question 6

10% 50% 90%

Question 7

10% 50% 90%

Question 8

10% 50%90%

Question 9

10% 50% 90%

Question 10

10% 50% 90%

MS&E 252 Handout #16

Decision Analysis I November 8th. 2004

Page 2 oI 12 HW#4 Solutions

Distinctions

These distinctions were prepared by the teaching team and reIlect our best belieI oI the

meanings oI these terms.

Prospect: a Iully deIined possible Iuture state oI the world. a particular Iuture.

your liIe going Iorward with a particular event or possibility occurring. A prospect

by deIinition must not involve any uncertainty.

Deal: A deal is consists oI a set oI Iuture prospects and their associated

probabilities.

Test - assessed Iorm: Prior: The probability distribution oI the distinction oI

interest conditioned on your inIormation. Likelihood: Conditional distribution

oI the observed distinction given the distinction oI interest.

Test - inIerential Iorm: Posterior: Conditional distribution oI the distinction oI

interest given the observed distinction. Pre-posterior: The probability

distribution oI the observed distinction (e.g.. test results)

PreIerence Probability: The preIerence probability Ior prospect P among a given

ordered set oI prospects B~.~W (B and W are the best and worst prospects.

respectively) is the probability p at which the decision maker is indiIIerent

between receiving prospect P Ior sure and receiving a deal on prospects B and W

with probability p oI B and probability (1-p) oI W. Notes: 1) People with same

preIerence probabilities will make the same decisions (though maybe diIIerent

VOI which depends on risk attitude). 2) preIerence probabilities are not really

probabilities. but can be treated as such (substitution rule).

Five Rules oI Actional Thought: the Iive rules can be remembered by using the

phrase "POE'S Choice." since the Iive rules consist oI the probability rule. the

order rule. the equivalence rule. the substitution rule. and the choice rule.

The probability rule: states that we shall consider the world (or all inIormation

relevant to a decision making) in terms oI possible prospects (elemental

possibilities) and their associated probabilities.

The order rule: states that we can make a list oI possible prospects such that any

prospect higher on the list is at least as good as ones lower on the list. This

implies transitivity oI prospects (iI A is preIerred to B and B is preIerred to C.

then A is preIerred to C). When you violate the order rule you create a potential

Ior a money pump. This means that someone can create a situation in which they

can make an inIinite amount oI money Irom you.

The equivalence rule: states that. given three prospects A. B. and C. where A is

preIerred to B. and B is preIerred to C. there exists a probability p such that

MS&E 252 Handout #16

Decision Analysis I November 8th. 2004

Page 3 oI 12 HW#4 Solutions

1-p

p

B ∼

C

A

In other words. there is some p that makes you indiIIerent between receiving B Ior

certain and receiving a deal with probability p oI prospect A and 1-p oI prospect

C. This probability p is a preference probability and is based on your

preIerences. not an event that can be Ioretold by the clairvoyant. This rule states

that a certain equivalent can be Iound.

The substitution rule: states that. given the situation described in the equivalence

rule. you are willing to exchange the prospect B with the A or C` deal.

The choice rule: states that. given the choice between the Iollowing two deals

1-p

p

B

A

1-q

q

B

A

you will choose the Iirst deal iI p~q. and you will choose the second deal iI q~p.

In other words. you must preIer the deal with the highest probability oI the

prospect you like best.

Money pump: A person is a money pump iI they exhibit behavior that allows you

to get as much money Irom them as you want. In particular. a person who violates

the Order Rule by having preIerences that are not transitive is a money pump.

Such a person preIers A to B and B to C. but also preIers C to A. Ior some

prospects A. B. and C. Then. Ior example. iI this person started with prospect A.

we could get him to pay us something to exchange A Ior C. some more to

exchange C Ior B. and some more to exchange B Ior A. We could repeat this

cycle indeIinitely and make a bundle.

e-value: is a probability weighted average oI a measure. The weighted sum oI

some measure associated with each prospect. where the weights are iust the

elemental probabilities oI each prospect and the sum is taken over all possible

prospects. We usually talk about the e-value oI the u-value or the e-value oI the

monetary measures oI our prospects.

u-value: is a preIerence probability that has been scaled and shiIted by means oI a

positive linear transIormation to a convenient scale. In other words. we can take a

preIerence probability. add or subtract any number and multiply or divide by any

positive number and get a u-value.

MS&E 252 Handout #16

Decision Analysis I November 8th. 2004

Page 4 oI 12 HW#4 Solutions

Probabilistic questions

1) Solution: a

You believe that each ball is equally likely to be drawn. so you assign a 5/10 or 1/2

chance to selecting a red ball. Your belieIs should not change aIter Samir takes one ball

out because you do not have any new inIormation about what is inside the urn. The ball

he selected could have been red. but equally likely. it could have been black.

This can also be seen on a probability tree. with two distinctions: 'color oI the ball drawn

by Samir¨. and 'color oI the ball drawn by the speaker¨.

2) Solution: c

Let pA. pI and pH be the preIerence probabilities Ior the Audi. the InIiniti. and the

Honda. respectively. Based on the inIormation provided. we know that:

i) pA ~ pI ~ pH

ii) .4pA ¹ .6pI ~ .9pI ¹ .1pH ( pA ~ .75pI ¹ .25pH)

iii) pI ~ .3pA ¹ .7pH

These statements do not contradict each other (take pA ÷ 1. pI ÷ .5 and pH ÷ 0 Ior

example) and so statement I is NOT true. Statement II is true (Iollows Irom statements i

and iii): pI ~ .3pA ¹ .7pH ~ .25pA ¹ .75pH). Statement III is true as well: preIerences can

change over time.

3) Solution: c

Statement I violates the Probability Rule: probability ranges are not allowed by the

probability rule. Statement II does not violate any oI the 5 rules. Adding or deleting

alternatives should not change the preIerence order oI the other alternatives. Statement III

violates the Order Rule. as there is no consistent ordering oI Silvia`s preIerences.

4) Solution: a

I. is Ialse You may not remove an arrow Irom a relevance diagram without Iurther

inIormation (at the level oI numbers);

II. is Ialse Adding an arrow Irom 'Fever` to 'Outdoor Activities` would create a cycle

in the relevance diagram: 'Outdoor Activities` ÷ 'Cold` ÷'Fever` ÷ 'Outdoor

Activities`;

III. is Ialse In order to Ilip this arrow without any Iurther manipulation oI the diagram.

'Pneumonia` and 'Cough` would need to be based on the same state oI inIormation.

Here. there is an arrow Irom 'Outdoor Activities` to 'Pneumonia` but no arrow Irom

'Outdoor Activities` to 'Cough`.

MS&E 252 Handout #16

Decision Analysis I November 8th. 2004

Page 5 oI 12 HW#4 Solutions

5) Solution: b

I. is Ialse Since relevance cannot be read oII a relevance diagram (only irrelevance can).

we do not even need to proceed to any arrow manipulations to see that statement I. cannot

be inIerred;

II. is true PerIorm the Iollowing operations:

• Add an arrow Irom 'Outdoor Activities` to 'Sneezing`. which is allowed since

this does not create any cycle in the diagram;

• Flip the arrow Irom 'Cold` to 'Sneezing`. which is allowed since both are now

conditioned on the same state oI inIormation. namely 'Outdoor Activities`.

Then 'Pneumonia` and 'Sneezing` are conditioned on 'Outdoor Activities` only and

we can see that there is no arrow between them. ThereIore. 'Pneumonia` and

'Sneezing` are irrelevant given 'Outdoor Activities`;

III. is Ialse PerIorm the Iollowing operations on the graph you obtained aIter going

through the two steps detailed above:

• Flip the arrow Irom 'Outdoor Activities` to 'Pneumonia`. which is allowed

since both are now conditioned on the same state oI inIormation. namely &;

• In order to Ilip the last arrow. Irom 'Outdoor Activities` to 'Sneezing`. we Iirst

need to ensure that both oI them are conditioned on the same state oI inIormation.

UnIortunately. because oI the arrow that goes Irom 'Pneumonia` to 'Outdoor

Activities`. the only way we can Ilip 'Outdoor Activities` to 'Sneezing` is by

adding an arrow Irom 'Pneumonia` to 'Sneezing`.

The existence oI an arrow Irom 'Outdoor Activities` to 'Sneezing` shows that we

cannot inIer that statement II. is valid iust by manipulating the structure oI this relevance

diagram;

IV. is Ialse The two nodes we are interested in are already conditioned on 'Outdoor

Activities`. We now want to condition them on 'Fever` as well. and so we need to Ilip

the two arrows that go Irom 'Fever` to 'Pneumonia` and 'Cold`. UnIortunately. the

only way we can do this is iI we Iirst add an arrow that links 'Pneumonia` and 'Cold`.

We can thus not inIer that IV. is valid.

6) Solution: a

We need to renormalize the preIerence probabilities so that 'Hawaii and Snorkeling¨

corresponds to 1.0 and 'Hawaii and Sailing¨ corresponds to 0.0. The preIerence

probability Ior 'Australia and Snorkeling¨ is then equal to

6 . 0

4 . 0 9 . 0

4 . 0 7 . 0

) Sailing" & Hawaii P(" - ) " Snorkeling & Hawaii P("

) Sailing" & Hawaii P(" - ) " Snorkeling & Australia P("

=

−

−

=

7) Solution: c

We can work with either the original preIerence probabilities or with the renormalized

preIerence probabilities oI the last question Ior this problem because all the preIerence

probabilities are linearly rescaled. the outcome will be the same in terms oI which

outcome Emily preIers. We will work with the original probabilities oI question 6.

MS&E 252 Handout #16

Decision Analysis I November 8th. 2004

Page 6 oI 12 HW#4 Solutions

Emily`s preIerence probability Ior 'Australia & Bungee Jumping¨ can be computed as

0.5 (0.8)(0.4) (0.2)(0.9) ) Sailing" & Hawaii (0.8)P(" ) " Snorkeling & Hawaii P(" ) 2 . 0 ( = + = +

Since her original preIerence probability Ior 'Australia & Sailing¨ is also 0.5. she is

indiIIerent between the two options.

8) Solution: a

Based on Jason`s values Ior the simple deals pictured in the problem. we can rewrite the

complex deal in question as:

This can now be reduced to the Iollowing simple deal:

This is Deal B. and we know Irom the problem statement that Jason`s certain equivalent

is $20 Ior this deal.

9) Solution: b

Susan is indiIIerent between participating in the contest (and winning one oI the prizes)

and $40 Ior sure. This implies that Susan`s preIerence probability Ior the contest. and her

preIerence probability Ior $40 Ior sure must equal:

0.5 ÷ q* 0.8 ¹ (1-q)*0.3

Solving Ior q gives you 0.4. ThereIore the correct answer is b).

0.2

0.5

$100

0.8

0.4

0.1

$0

0.9

0.1

$0

$100

0.6

0.4

$0

$100

0.7

$100

$0

0.3

MS&E 252 Handout #16

Decision Analysis I November 8th. 2004

Page 7 oI 12 HW#4 Solutions

10) Solution: d

Answer a) is wrong because what Gael had paid Ior the Ilowers is a sunk cost. and should

not be taken into account when setting the sale price Ior the remaining roses. Answer b) is

also wrong because. according to Gael`s belieIs. his revenues Irom this point onward will

be higher iI he reduces the price. Answer c) is another common mistake made by people

who Iorget the distinction between a good decision and good outcome. ThereIore all oI

the above answers are wrong and the correct answer is d).

Quantitative Problems

1) Relevance

a.

¦B'A&} ÷ 0.5 ÷ ¦B'A`&} thereIore A and B are not relevant given &.

¦C'BA&} ≠ ¦C'B`A&} thereIore B and C are relevant given A&.

¦C'AB&} ≠ ¦C'A'B&} thereIore A and C are relevant given B&.

The relevance diagram is the Iollowing:

A

C

B

b.

MS&E 252 Handout #16

Decision Analysis I November 8th. 2004

Page 8 oI 12 HW#4 Solutions

A

.38

.9

.63

.37

C

C'

A

A'

A'

.1

.62

B

.67

B'

.33

B

.57

B'

.43

B

.43

B'

.57

B

.33

B'

.67

0.02

0.04

0.24

0.32

0.08

0.06

0.16

0.08

¦A'C&} ≠ ¦A'C`&} thereIore C and A are relevant given &.

¦B'AC&} ≠ ¦B'A`C&} thereIore A and B are relevant given C&.

¦B'AC&} ≠ ¦B'AC'&} thereIore C and B are relevant given B&.

The relevance diagram is the Iollowing:

A

C

B

2) Soojin`s prospects

Deal B can be compared to Deal A by replacing the $35 with the equivalent uncertain

deal using the Substitution Rule. Deal B becomes:

=

$0

.5

.5

$80

.7

.3

$80

$0

.35

.65

$0

by the Choice Rule. A ~ B. Using the Probability Rule. we can simpliIy Deal C to:

MS&E 252 Handout #16

Decision Analysis I November 8th. 2004

Page 9 oI 12 HW#4 Solutions

$0

$20

.5

$80

.1

.4

This is better than A because C has a .5 chance oI $20 instead oI $0 and Ed preIers

more money to less. Hence Ed`s preIerence ordering is C ~ A ~ B.

3) Carolina`s certain equivalence.

By the Substitution Rule one can replace the $50 and $35 with their equivalent

uncertain deals:

.125

.25

.25

.375

$0

$100

.25

.75

$100

$0

$0

.50

.50

$100

but this is equivalent to the Iollowing deal (by collecting like terms in the

outcomes).

$0

.50

.50

$100

Carolina has already identiIied she is indiIIerent between owning this deal or $35.

hence her certain equivalent is $35.

4) Allais Big Deal

a. Rik`s preIerences A ~ B and D ~ C are inconsistent with the rules oI actional

thought.

By the order rule. 5~1~0

By the equivalence rule. Ior some

p .

5

0

p

1p

1 ~

MS&E 252 Handout #16

Decision Analysis I November 8th. 2004

Page 10 oI 12 HW#4 Solutions

Consider A and B:

Thc 5ubstItutInn Ru!c

5

O

O.1O

O.O1

1

O.89

1

ß

A

O.1O

O.O1

O.89

O

ß

A

5

1-

O

5

1-

5

O

~

O

5

O.1O + O.89

O.9O - O.89

ß

A

~

5

1-

O

By the choice rule. A ~ B

p ~ 0.10 ¹ 0.89p

0.11p ~ 0.10

p ~ 10/11

Now consider D and C:

Thc 5ubstItutInn Ru!c

O

1

O.11

O.89

C

D

5

O.1O

O.9O

O

O

O.11

O.89

C

D

~

5

O.1O

O.9O

O

5

1-

O

O

5

O.11

1-O.11

C

D

5

O.1O

O.9O

O

~

By the choice rule. D ~ C

0.10 ~ 0.11p

p · 10/11;

Thus we have a contradiction. i.e.. iI p ~ 10/11. the choice rule is violated.

The only consistent preIerences are: A ~ B

.

C ~ D

MS&E 252 Handout #16

Decision Analysis I November 8th. 2004

Page 11 oI 12 HW#4 Solutions

B ~ A

.

D ~ C

A ~ B

.

D ~ C.

Hence. Rik`s choices oI A and D violated the choice rule.

Since Rik is risk neutral and preIers more money to less. he should choose B over

A and D over C. since the e-values are 1.14 ~ 1 and .25 ~ .11

Your answer will depend on your aversion to risk. but to be consistent with the

choice rule. you must hold one oI the three preIerences listed in the solution to (a).

OI course this assumes that you preIer more money to less. but keep in mind. you

could always give money to charity.

5) Taxi Cabs

a. Here is a relevance diagram showing all the inIormation in the problem except Ior

what the witness actually said. Color oI cab is given unconditionally. and our

belieIs on what the witness will say are conditioned on the color oI the cab he

saw.

witness

says

color of

cab

'Accuracy oI witness¨ or 'distribution oI cab colors¨ are bad distinctions Ior

decision-making. The distinctions we draw should always be ones which bear

concretely on the issue at hand. We include our understanding oI accuracy oI

witness in the probabilities inside the node 'witness says¨; we include our prior

distribution oI cab colors in the probabilities on color oI this particular cab.

Employing observable distinctions like what the witness says. allows us to

perIorm probabilistic inIerence and update our belieIs.

b.

Blue

Green

Says Blue

Says Green

Says Blue

Says Green

Blue, Says Blue

Blue, Says Green

Green, Says Blue

Green, Says Green

.8

.2

.2

.8

.72

.18

.02

.08

.9

.1

c. Here is a Ilipped tree. Now we are prepared to incorporate the evidence.

Previously the branches which had 'witness says green¨ were scattered

throughout the tree; now they are all in the 'says green¨ sub tree.

MS&E 252 Handout #16

Decision Analysis I November 8th. 2004

Page 12 oI 12 HW#4 Solutions

Says Blue

Blue

Green

Blue

Green

Says Blue, Blue

Says Blue, Green

Says Green, Blue

Says Green, Green

.74

.26

.97

.03

.69

.31

.72

.02

.18

.08

Says Green

d. In the relevance diagram corresponding to the tree Ilip. the arc has been reversed.

witness

says

color of

cab

e. Our conditional probability oI the cab being green is 0.31. Conditioning

probabilities on evidence is the correct way to incorporate evidence into one`s

probabilistic belieIs.

f. It may seem surprising that a person with 80° accuracy would have only a 31°

chance oI being right. but this number is also a Iunction oI the low ratio oI green

cabs in the city.

- ProblemSession7
- ProblemSession1
- ProblemSession3
- ProblemSession5
- ProblemSession4
- HO15-HW5
- ProblemSession6
- Midterm
- ProblemSession2
- HO10-PracticeMidterm
- HO28-HW7Solutions
- HO29-OtherDAClasses
- MidtermSolutions
- HO26-PracticeFinal1Sol
- HO8-HW2Solutions
- HO25-PracticeFinal1
- HO19-HW5Solutions
- HO2-HW1
- HO12-HW3Solutions
- HO5-HW2 v2
- HO17-CaseInstructions
- HO23-HW6Solutions
- HO26b-PF1Sols Erratum
- Final Instructions Part2
- HO27-PracticeFinal2WS

Sign up to vote on this title

UsefulNot usefulMS&E 252

MS&E 252

- HO13-HW4
- HO18-HW6
- HO28-HW7Solutions
- HO19-HW5Solutions
- HO7-HW3
- HO22-HW7
- HO8-HW2Solutions
- HO5-HW2 v2
- HO23-HW6Solutions
- HO15-HW5
- HO10-PracticeMidterm
- MidtermSolutions
- HO12-HW3Solutions
- HO26-PracticeFinal1Sol
- HO17-CaseInstructions
- Midterm
- HO25-PracticeFinal1
- HO27-PracticeFinal2WS
- ProblemSession3
- HO29-OtherDAClasses
- HO9-MidtermLogistics
- ProblemSession8
- ProblemSession4
- HO20-Decision Diagrams
- ProblemSession7
- ProblemSession1
- ProblemSession5
- Final instructions Part1
- HO3-ProbQ
- ProblemSession6
- HO16-HW4Solutions