11

Decision Theory and Decision Tree

Analysis
Introduction
taryfield of life one has to make decisions in
differentalternative courses of action.
Decision making is needed whenever an individual or an organisation (private or public) is
psdwith a situation of selecting an optimal (or best) course of action among several
available
lhermatives. For example, an individual may have todecide whether to invest his money in stock.
honds or debentures; whether to build a house or to purchase a flat or live in a rented
anommodation; whether to join a service or tostart own business; which company's car/scooter
should be purchased, etc. Similarly, abusiness firm may have to decide which product it should
produce, among various products; the type of technique to be used in production; what is thÁ most
apropriate method of advertising its product, etc.(Decision making is aprocess of choosing an
optimal course of action out of several alternative courses for the purpose of achieving a 'goal or
goals' .Statistical decision theory consists of alarge number of quantitative techniques which helps
in analysing adecision situation and enable us to assure at a conclusion which is the best under given
CITCumstances of the case.

Elements of aDecision Problem

Ihere are taree elements of any decision problem. These are acts, states of nature and pay off
matrix andegret matrix which are discussed below:
Acts: The decision always involves a choice among sveral altermatives. These several
allermatives are called acts. For example, a management is faced with the problem of choosing one
One three products X. Y and Z for manufacturing. It means that there are three acts out of one act is
to
De Chosen. Thus, acts are the several alternative course of action or strategies, that are available
adecision maker. These are denoted by Aj, Ag, A3, ...A,.
beyond
)Sates of Nature (or Events): In every act, there are events which are uncertain and
he control of a decision maker. These events are outside the firm and not under its control. In the
above example where the management is faced with the problem of choosing one of three products
the potential demandfor the product may turn out to be good, moderate or poor.
manufacutring,
The consumer s demand for the products are the events which are uncertain and beyond the control
 380
Quantitative Methods For Business
of the decision maker. Thus, events which are beyond the
control of the decision maker are called
states of nature. These are denoted byS,S,,S3,..Sp:
(3) Pay off Table (or Pay off Matrix): When the value of each
event in the act are
calculated directly in terms of gains or losses expressed in money, it is called a pay off. Each
combination of a course of action in Acts and States of Nature is associated with pay off, which
measures the net benefit to the decision maker. A pay off matrix (or a table) consists of the
following two things (i) alternative acts like Aj, A2,A3, .. A, and (ii)various states of nature
like S, S, S3, .., S, involved in every act.
The format of a pay eff tablepay off matrix is given below:
States of Natare Acts

A1 Az Ag
S X1 X12 X13
S X21 X2 X23
S3 X31 X32 X33
In the above table column represents acts and row represents events (or States of Nature).
A. Opportunity Loss Table (or Regret Matrix): Opportunity loss is the loss in carried as a
consequence of the failure to take the best possible decisions. For any given state of nature (S;) the
opportunity loss for any given act (A;), is defined as the difference between the maximum possible
pay off over different acts for a given state of nature and the actual pay off for the act. (A; ) over that
state of nature. A specimen of opportunity loss table (or regret matrix) is as follows:
Opportunity Loss Table
States of Nature Acts

A A2 A3
S M- X1 M- X12 M- X13
S2 M-X1 M- X21 M-X3
S3 M- X31 M- X31 M-X33

Here M=Maximum possible pay of.

Decision Making Environments
Decisions are made under three types of enviroments:
(1) Decision making under conditions of certainty: In this environment, only one state of
nature exists i.e., there is complete certainty about the future. It is easy to analyse the situation and
make good decisions.
(2) Decision making under conditions of uncertainty: Here, more than one states o a
exist but the decision maker lacks sufficient knowledge toassign probabilities to the various stales
Oa)Me.NOTE? PR:
Al QUAD CAMERA
 Theory and Decision Tree Analysis
Decision making under
the
)deCision
maker has conditions of risk:
sufficient knowledge to assign
Here alsO, more
than one states of
381

Decision Criteria probabilities to each of nature exist

thest states of
Everv decision maker has to make
sllCs ot
nature Different criteria are choice
used among the best course of action (or
enNrONments, which, from the for making act)
view point of decision under different under various
ollowngcharts:
convenience study have been present
of decision making
ed by the
Decision Criteria

Decision Making Under Uncertainty

Decision Making Under Risks
Maximin Maximax Minimax regret||LaplaceHurwicz
Expected Expected
monetary opportunity
value (EMV loss EOL)

ADecision Making Under Uncertainty

For decision making under uncertainty without the use of probability, the following different
criteria are usually adopted:
(1)Maximin Criterion
(2) Maximax Criterion
(3) Minimax Regret Criterion
(4) Hurwicz Criterion
(5) Laplace Criterion
by Wald. It is based on extreme
(1) Maximin Criterion: The maximin.criterion was introduced
possible is going to happen.
pressimism. This decision criterion assumes that worse of the minimum monetary pay
that maximises the
Iherefore, it is designed toselect the action alternative
determine the minimum pay off for each action,
that the decision maker has to (i)
o11. This implies
änd (i1) select that act which maximises
the minimum pay offs.
The maximax criterion is based on extreme optimism. lt suggests that
(2) Maximax Criterion: which it is possible for him to receive the
particular act under
uie decision maker should select
that
the monetary pay ofts). This implies the
that
t lavourable pay off
(the action that maximises off for each action, and (ii)selects that act
determine the maximum pay
SiOn maker has to(i)
which maximises the maximum pay offs. criterionwas developed by Savage. He pointe
Minimax Regret Criteríon: This decision after the decision has been made andthe state
)
Outthat the decision maker might experience
regret
to minimise the regret(minimax) befor
the decision maker should attempt decision maker has to (i) transformthe
cure occurred. Thus. that the
particular action. This implies
actually selecting a
 382 Quantitative Methods For Business
pay off matrix in to a Regret Matrix. Thiscan be done by subtracting each of the values of the act
from the argest pay off of that act for agiven state of nature (ii) identifies the maximum regret for
each act, and (iii) selects that act which minimises the maximum regret.
(4) Hurwicz Alpha Criterion: Leonid Hurwicz has developed a criterion which is a
combination of maximax (optimistic) and maximin (pessimistic) decision criterion. This criterion is
basedon the assumption that adecision maker has adegree of optimism, which is represented by the
coeff. of optimism a.The maximum pay off of each act is multiplied by degree of optimus a and
minimum pay off by the degree of pessimism(1-«). This implies that the decision maker has to (i)
choose an appropriate degree of optimism, a so that (1-a)represents the degree of pessimism, (ii)
determine the maximum as well as minimum of eaçh alternative and obtain P =aX maximum +
(l-a) xminimum for each act, and (ii) choose the act that yield the maximum value of weighted
pay off, denoted by P.
(5) Laplace Criterion: Laplace criterion of decision making is applicable in those cases in
which all the events (or states of nature) have equal opportunity. So equal probabilities are assigned
to all the events. This implies that the decision maker has to (i) determine the average pay off for
1
each act by using the formula -(p, +p, t... p,) where n denotes the number of events and P
n

denotes the pay offs and (i) selects the act which results in maximum average pay off.
Spplications of Decision Making Under Uncertainty
The applications relating to decision making under uncertainty are studied under the following
heads:

(1) When the pay offs matrix with profit data is given
(2) When the pay offs matrix with cost data is given.
(1) When the pay offsmatrix with profit data is given: When we are given pay off matrix
with profit data, the uses of different decision criteria can be illustrated by following examples:
VExample 1. Given the following pay off matrix:
Acts States of Nature
A A, A3
S1 700 500 s00
S2 300 450 300

S3 15) 100 303

Determine the best act to be chosen under:

(i) Maximin Cxiterion
(i) Maximax Criterion and
(iiü) Mipimax Regret Criterion
Solution.
) Maximin Criterion: When this criterion is adopted, we select that act which maximises
the minimum pay off:
388 Quantitative Methods For BuSiness
6. Find the best act by Hurwicz criterion from the following pay off table.
Pay off Table
Acts/States of Nature S S
A -2 -3 4 3
-1 7
A3 -3 4 9

[Take the coefficient of optimism a=0-7]

[Ans. A is the best act with a =0.7I

(2)Decision Making Under Risks with Probability

In this case, the decision maker developedgood probabilityestimates for the different states of
nature. The following two criteria are usually adopted in decision making under risks with
probability:
(1) Expected Monetary Value (EMV) Criterion
(2) Expected Opportunity Loss (EOL)Criterion
(3) Expected yalueof Perfect Informations (EVPI)
( Expected Monetary Value (EMV) Criterion: This criterion Fequires the calculation of
expeted monetary value (EMV} of each act which is obtainied by multiplying the conditional pay
offs for that act by the assigned probabilities of various states of nature. The decision maker selects
that act that yields the highest EMV. The Expected Monetary Value (EMV) for acourse of action X
is given by:
EMV(X) =PiX11 +P2X1 +P3X31
where, XË|;X2; and X31 denote pay off of act X for S,S) and S; event or states of nature and
P1 P2 and p; denote the probability of occurrence of S,S,, S, event (or states of nature).
Procedure
The calculation of EMV consistsof following steps:
(i) Construct a pay off table listing the alternative course of action and the various states of
nature, if not given. Enter the conditional profit for each decision act-event combination
along with the associated probailities.
(iü) Calculate thé EMV for each decision act (or alternative) by multiplying the conditional
profit by assigned probabilities and adding the resulting conditional values.
the act (or alternative) that yields the highest EMV.
(2)'Expected Opportunity Loss (EOL) Criterion: An alternative criterion (o maximis1ng
EMV approach) is to minimise expected opportunity loss (EOL). Expected opportunity loss (0r
willbe reduced
expected value ofiregrets) represents the amount by which naximum possible profitreductions isthe
under várious possible actions. The course of action that minimises these lossesanoract is obtained by
optimal decision act(or alternative). The expected loss (EOL) of
opportunity

8
 Decisiorn Theory
and Decision Tree Analysis
389
multiplyingthe conditional opportunity loss for that act by the assigned
nature. For acourse of action X, the probabilities of various
statesof expected opportunity loss(EOL) is given by:
EOL (X) =PL41 +P2L21 + P3L31
,1 . Lo1 and L31 enote OPPortunity loss of act X for S;, S and Sa
event or states of
natureand Pi P2 and pa denote the probability of occurrence of S;, ST
and S, event (or states of
nature).
Procedure
The calculation of EOL consists of the following steps:
(1). Construct a conditional pay off table for each act- event combination, if not given along
with the associated probailities.
Do each event (or states of nature) determine the conditional opportunity loss
(EOL) by
eubtraction the pay off from the maximum pay off for that event (or states or nature).
(2 Calculate the expected opportunity loss (EOL) for each decision alternative (or act) by
multiplying the conditional loss by the associated probability and then adding the values.
(4) Selectthe alternative (or act) that yields the lowest EOL.
Note: EOL criterion is similar to EMV criterion except the opportunity losses are considered
instead of profits.
(3) Expected Value of Perfect Information (EVP): Under this ceriteion, it is assume that the
decision maker has authenticand perfect information about the future. With perfect information, the
retailer (decision maker) would know in advance the demand for each day and will store the exact
number as per demand. The expected value of perfectinformation (EVPI) is the difference between
expected pay off with perfect information (EPPI) and expected pay off with uncertainted (or EMV
of Best Action). Symbolically:
EVPI = EPPIEMV of Best Action
Where, EPPI = (Best pay off for lst state of nature xprobability of lst state of nature ) + (Best
pay off for 2nd state of nature xprobability of 2nd state of nature) +...+ (Best pay off last state of
nature x probability of last state of nature)
Applications of Decision-Makiig Under Risks
The applications of decision making under risks with probability are studied under the
following heads:
(1)When the conditional pay off matrix with profit data is given.
(2) When the conditional pay off matrix with profit data is not given.
(1) When the conditional pay off matrix with profit data is given: When we are given
the associated
conditional profit table for EMV
different acts and the various states of nature along with
the following
probabilities, the uses of criterion and EOL criterion can be illustrated by
examples:
402 Quantitative Methods For Business
Decision Trees
Decision making involves several stages and at each stage, each of the chocies open will resule
in adifferent payoff. For thesake of simplicity, these stages canbe represented by analtermaltiye
method called tree formation listing out allevents and the resultant outs comes. The tree diagramie
alsoknown as decision tree. Adecision tree is a graphic device of a decision making process.It
consists of nodes, branches, probabilities and the resultant pay-ofs. Decision trees have standard
symbols. Square indicates adecision node. These are anumberof branchesleading from this square.
These branchesindicate various courses of action available to the decision maker. At the end of each
branch there is a circle which represents chance node (or state of nature). Various outcomes emerge
out of the chance node with their associate probability estimates. The net result of cach outcome is
indicated against each circle. The branches that are drawn from the decision node are named as
decision branches while the branches drawn from chance node are nanmed as chance branches
Following diagram gives the structure of the decision tree.
Specimen of adecision Tree
Outcomes
E
Q11

Q12

A1 E1
Qz1
A2
A3 Q22
E
Q31

Decision Node Q32

Chance Node

application of decision tree.

The following examples illustrate the and new. The new machine is
more
two packaging machines: old
Example l: An organisation has machine performs
good quality, on the other hand the old
efficient if the materials are of
quality. The following information are given:
better if the materials are of poor
and 20% of poor quality.
() 80% materials have been of good quality
Aii) The profit position is as under:
ka) Using old machine
2000
-If the materials are good
1600
If the materials are poor