BASIC DECISION ANALYSIS

Nur Aini Masruroh

The party problem

Kim ingin mengadakan pesta

ulang tahun. Dia
mempertimbangkan 3 lokasi
tempat: outdoor, indoor, teras
(porch)

Decision node

Chance node

Nodes

Decision node:
Berbentuk

persegi
Melambangkan titik dalam tree yang menyatakan titik
pengambilan keputusan, decision maker mempunyai
kebebasan penuh untuk mengambil keputusan

Chance node:
Berbentuk

bulat
Melambangkan uncertain variable, decision maker
tidak mempunyai kontrol terhadap outcome variable ini

Solving party problem

Dengan menggunakan 5 rules:
Ordering rule
Equivalence rule
Substitution rule
Decomposition rule
Choice rule

Ordering rule

Possible outcome:
Outdoor sunny
Outdoor rainy
Porch sunny
Porch rainy
Indoor sunny
Indoor rainy

Objective: memaksimumkan kepuasan

Best outcome: outdoor -- sunny
Worst outcome: outdoor -- rainy

Equivalence rule

Semua
intermediate
outcome
ditentukan
equivalensinya
terhadap best
outcome dan
worst outcome.

Membuat decision tree

Misalkan:
Probability cuaca
besok cerah = 0.4
Probability cuaca
besok hujan = 0.6

Substitution rule decomposition rule

Choice rule

Jadi dipilih lokasi pesta indoor, memaksimalkan kemungkinan untuk

mendapatkan best outcome

Expected utility untuk setiap alternatif

Equivalent Monetary or Dollars Values

Cara lain selain

menggunakan utility
value adalah
menggunakan equivalent
monetary value untuk
setiap outcome.

Dollar value vs utility value

Utility function u(x)

Certain equivalent

The certainty equivalent (CE) is the amount in which a

person is just indifferent between receiving it for sure
and an uncertain or risky prospect that might either
pays more or less than this amount.

The Certainty Equivalent of a deal is the Personal

Indifferent Selling Price (PISP)
To find the CE of an alternative, we first compute its
expected utility and then take its inverse to convert it
back into equivalent dollar value.

Contoh
Outdoors:
Expected utility = 0.4.
Hence certainty
equivalent = u-1( 0.4 )
= $26
Porch:
Expected utility = 0.57.
Hence certainty
equivalent = u-1( 0.57 )
= $40

Note that the best decision is also the one with the
highest certainty equivalent.

Indoors:
Expected utility = 0.63.
Hence equivalent
= u-1( 0.63 ) = $46

Mengunakan utility function untuk kasus

lain

Misalkan Kim menghadapi deal sebagai berikut:

EU = 0,5
PISP = u-1(0,5) = $32 (dari grafik utility vs dollar
value)

Risk Neutral Decision Maker

Utility value vs dollar value

The optimal decision for Jane

is to hold the party in the
porch
Notice that this is different
from Kims optimal choice
which is indoors. This is due
the difference in preferences
between the two.

Comparing utility curves

When the decision maker is risk neutral, there is no need to use a utility function.
Just work on the dollar values on the decision tree. This is equivalent to using the
function u(x) = x

Selling Price for the Coin Tossing Game

Jane: Expected utility = 0.5. Certainty equivalent = u-1(0.5) = $50.

Kim: Expected utility = 0.5. Certainty equivalent = u-1(0.5) = $34.

Kim is more averse to risk than Jane. She is willing to sell off the
deal at a lower price than Jane.

VALUE OF INFORMATION
ANALYSIS

The Value of Clairvoyance (perfect

information) on Weather

Lets use our previous party problem

Suppose that a clairvoyant offers to tell Kim
whether the weather will sunny or rainy tomorrow.
However, he charges a fee of $15.
Should Kim buy the information about the weather
tomorrow from the clairvoyance for a fee of $15?

Decision tree
Conclusion: At a cost of $15,
Kim should buy the information.
Suppose that the cost of the
information is $16 instead?
Should Kim buy it or not?
What if the fee is $20 or
$25?

Impact of Changing the Cost of Clairvoyance

We determine, by
resolving the decision
tree repeatedly, the
expected utilities for
No clairvoyance and
Buy Clairvoyance for
$x) for values of x in
the range $0 to $50.

Certainty equivalent
Kim is indifferent between
Buy clairvoyance and No
clairvoyance when the cost
of
clairvoyance is about $20.
The Value of Clairvoyance or
Expected Value of Perfect
Information on an
uncertain variable is the cost
of clairvoyance at which the
decision maker is just
indifferent
between buying and not
buying the information.

Interpretation of Value of Information

The value of clairvoyance represents the maximum amount

one should be willing to pay for the perfect information
The value of clairvoyance provides a benchmark against
which to compare any information gathering scheme that may
be proposed
If the cost of the scheme exceeds the value of clairvoyance,
then there is no need to examine the scheme in any further
detail.
The expected value of perfect information or clairvoyance
for Kim is $20. Thus, no other sources of information about
the weather could be worth more than $20 to her.

Expected Value of Imperfect Information

Suppose, instead of clairvoyance, Kim was offered the

service of an Acme Rain Detector which will indicate
either Rainy or Sunny with an accuracy of only
80%.
That is, if the actual weather is going to be sunny, it will
read sunny with probability 0.8, and if the actual
weather is going to be rainy, it will read rainy with
probability 0.8.
The fee for using the Acme detector is $12, a 20%
discount on the $15 asking price of the clairvoyant.
Should Kim pay $12 to use the Acme detector which is
80%?

Probability tree
Rain detector performance

Flip the tree

Decision tree

Conclusion: It is not worth

paying $12 for the use of
the Acme rain detector.

Impact of Changing the Cost of Detector

(Imperfect Information)

Kim is indifferent between Use detector and No detector when the cost of
detector is about $8.80 Expected Value of Imperfect Information is $8.80
The Expected Value of (Imperfect) Information (EVI) of an uncertain event is
the amount payable for which a decision maker is just indifferent between having and
not having information on the event.

Value of Clairvoyance for Risk Neutral Case

Expected Value
Indifference Method
When Jane is
indifferent between
no information and
$x-clairvoyance:
70 x = 48 x =
70 - 48 = $22
Hence Value of
Clairvoyance =
$22.00

Difference of Certainty Equivalents Method

We have thus found that Kim and Jane have

different value of clairvoyance for weather.
These differences arise solely from differences in
taste (preference), not from differences in
structure or information.

Expected Value of Imperfect Information

for Risk Neutral case

Janes expected value of information for rain detector is $59.20 - $48.00 = $11.20

Again, we have found that Kim and Jane have different expected value of imperfect
information for the same rain detector (accuracy=80%).

These differences arise solely from differences in preference, not from differences in structure
nor information

Sensitivity Analysis

Sensitivity of Kims Expected Utility to Probability of

Sunshine
Let p be the probability of sunshine
The expected utilities for the three
alternatives as a function of p are:
EU(outdoors) = p
EU(porch) = 0.950p + 0.323(1-p) = 0.323
+ 0.627p
EU(indoors) = 0.586p + 0.667(1-p) = 0.667
0.099p

Sensitivity of expected utility to

probability of sunshine

Probability of
sunshine, p Best
decision
0 p 0.47
Indoors
0.47 p 0.87
Porch
0.87 p 1
Outdoors

Sensitivity of Kims Expected Utility for Free

Clairvoyance to Probability of Sunshine

If the probability of
sunshine is p, the expected
utility of free clairvoyance
is
p (1) + (1- p)(0.667) =
0.667 + 0.333 p

Sensitivity of Kims Certainty Equivalent

to Probability of Sunshine

Sensitivity Analysis for Jane

Comparing Kims and Janes decision thresholds

Probability of sunshine, p
Kim
Indoors

0 p 0.47
Porch 0.47 p 0.87
Outdoors 0.87 p 1

Jane
Indoors

0 p 0.375
Porch 0.375 p 0.667
Outdoors 0.667 p 1

Acknowledgment
Assoc. Prof. Poh Kim Leng
Dept Industrial and Systems Engineering
NUS