Lecture 4. Decision Theory - Decision Tables and Decision Trees, Game Theory.

1. Introduction and Basic Terms of Decision Theory.

1.2. Decision Making under Certainty.
1.3. Decision Making under Uncertainty.
1.4. Decision Making under Risk.
2. Characteristic and Format of Decision Trees.
3. Solution to Conflict Decision Problems by Use of Game Theory.

1. Introduction and Basic Terms of Decision Theory

Decision Theory - a theoretical technique utilizing a group of related constructs to describe or prescribe
how individuals or groups of people choose a course of action when faced with several alternatives and a
variable amount of knowledge about the determinants of the outcomes of those alternatives.

Decision theory in management is concerned with identifying the values, uncertainties and other issues
relevant in a given decision, its rationality, and the resulting optimal decision. It is closely related to the
field of game theory as to interactions of agents with at least partially conflicting interests whose
decisions affect each other.

Decision theory represents a generalized approach to decision making. It enables the decision maker:
 to analyze a set of complex situations with many alternatives and many different possible
consequences
 to identify a course of action consistent with the basic economic and psychological desires of the
decision maker

Decision theory problems are characterized by the following:

 a decision criterion. A decision criterion represents the decision maker´s point of view for
selecting the best alternative. Usually, it is derived from the decision goal. It may be maximization
(for example gains) or minimization (for example costs).

 a list of alternatives. The alternative represents an acceptable solution to the decision problem.
Success of developing suitable alternatives depends on the experience and creativity of the
decision maker.

 a list of possible future events (states of nature). Events represent possible future situations that
will be the primary determinants of the eventual consequence of the decision. The situations must
be mutually exclusive (no two or more events can occur simultaneously) and collectively
exhaustive (the events must cover all the possibilities).

 payoffs associated with each combination of alternatives and events . These payoffs may be
profits, revenues, costs, or other measure of value. Usually the measures are financial. They may
be weekly, monthly or annual amounts, or they might represent values of future cash flows.
Usually, payoffs are estimated values. The more accurate these estimates, the more likely it is that
the decision maker will choose an appropriate alternative. If the number of alternatives is m and
the number of states of nature is n, m x n possible payoffs must be determined.

 the degree of certainty of possible future events. This degree can range from complete knowledge
about which state will occur to partial knowledge (the probabilities of states of nature are known)
and to no knowledge (complete uncertainty).

There is a wide range of management decision problems. Among them there are capacity and order
planning, product and service design, equipment selection, location planning, and so on.

1
Lecture 4. Decision Theory - Decision Tables and Decision Trees, Game Theory.

Some examples of alternatives and possible events for these alternatives are shown in Table 4.1.

 
Alternatives Events
To order 10, 11, … units of a product Demand for the product may be 0, 1, … units
To make or to buy a product The cost of making may be 20, 22, …
$thousands
To buy or not to buy accident An accident may occur, or may not occur
insurance

Table 4.1 “Examples of Alternatives and Events”

Various properties of decision problems enable a classification of decision problems. Solution to any
decision problem consists of these steps:

1. Identify the problem

2. Specify objectives and the decision criteria for choosing a solution
3. Develop alternatives
4. Analyze and compare alternatives
5. Select the best alternative
6. Implement the chosen alternative
7. Verify that desired results are achieved

Decisions problems that involve a single decision are usually best handled through payoff tables, whereas
problems that involve a sequence of decisions, are usually best handled using decision trees.

Rows of a payoff table (called also decision matrix) relate to the alternatives, columns relate to the states
of nature. Elements of a decision matrix represent the payoffs for each alternative under each possible
event.

We will construct a payoff table for the following simple example.

A grocer solves a problem of how much pastry to order every day. His profit depends on a demand that
can be low, moderate, or high. Values of the profit per day (in $) for these situations and for small,
medium or large order are shown in Table 4.2. The body of this table represents a decision matrix.

Demand
Low Moderate High
Order 
Small 50 50 50
Medium 42 52 52
Large 34 44 54

Table 4.2 “Payoff Table for Order Planning”

2
Lecture 4. Decision Theory - Decision Tables and Decision Trees, Game Theory.

The environments in which decisions are made can be classified according to the degree of certainty.
There are three basic categories: certainty, risk, and uncertainty. The importance of these three decision
environments is that they require different techniques of analysis.

1.2. Decision Making under Certainty

When the decision maker knows for certain which state of nature will occur, he will choose the
alternative that has the highest payoff (or the smallest loss) under that state of nature.

In the above grocer´s decision problem, suppose that it is known with certainty that demand will be low.
The highest profit for low demand is $50 (see Table 4.2) and therefore the small order should be elected.

1.3. Decision Making under Uncertainty

Under complete uncertainty, either no estimates of the probabilities for the occurrence of the different
states of nature are available, or the decision maker lacks confidence in them. For that reason,
probabilities are not used at the choice of the best alternative.

Most of the rules for decision making under uncertainty express a different degree of decision maker´s
optimism.

We will describe the most used rules for the choice of the best alternative supposing that the decision
criterion expresses the requirement of maximization (the modification of these rules for minimization
criteria is not difficult). We will use all the described rules to solution of the order planning problem
given in Table 4.2.

The maximax rule is appropriate for extreme optimists that expect the most favourable situation (they
choose the alternative that could result in the maximum payoff). Under this rule, the decision maker will
find the largest payoff in the decision matrix and select the alternative associated with it (the largest
payoff is determined for each alternative and then the largest payoff from these values is selected;
therefore “maximax”). This procedure for the order planning problem is shown in Table 4.3.

Demand
Low Moderate High Row Maximum
Order 
Small 50 50 50 50
Medium 42 52 52 52
Large 34 44 54 54  Maximum

Table 4.3 “Maximax Solution for Order Planning Problem”

The best overall profit is $54 in the third row. Hence, the maximax rule leads to the large order (the
grocer hopes that the demand will be high).

The maximin rule (Wald criterion) represents a pessimistic approach when the worst decision results
are expected. The decision maker determines the smallest payoff for each alternative and then chooses the
alternative that has the best (maximum) of the worst (minimum) payoffs (therefore “maximin”).

3
Lecture 4. Decision Theory - Decision Tables and Decision Trees, Game Theory.

In the Table 4.3, the smallest numbers in the rows are 50, 42, 34. Since 50 is the largest, the low order
should be chosen (if order is low, the $50 grocer‘s profit is guaranteed).

The Hurwicz -criterion represents a compromise between the optimistic and the pessimistic approach
to decision making under uncertainty. The measure of optimism and pessimism is expressed by an
optimism - pessimism index ,  [0;1] . The more this index is near to 1, the more the decision maker
is optimist. By means of the index  , a weighted average of the best payoff (its weight =  ) and the
worst payoff (its weight = 1- ) is computed for each alternative and the alternative with the largest
weighted average should be chosen.

Formula for Hurwicz Criterion:

The largest weighted average of the best and the worst payoff in the rows of a payoff table has the
following value: where
a = the optimism-pessimism index;
vij = the payoff for alternative Ai under state of nature Sj.

If  =1, the above rule is the maximax criterion, whereas if  =0, it is the maximin rule.

If we choose  = 0.7 at determining the best size of the order in Table 1.2, the weighted average (WA) of
the largest and the smallest profit for each size of the order has the following values:

WA (small) = 0.7* 50 + 0.3 * 50 = 50;

WA (medium) = 0.7* 52 + 0.3 * 42 = 49;

WA (large) = 0.7* 54 + 0.3 *34 = 48.

Maximizing the weighted average of the largest and the smallest profit, the small order should be
selected.

The maximax and maximin rules and the Hurwicz criterion can be criticized because they focus only on
extreme payoffs and exclude the other payoffs. An approach that does take all payoffs into account is the
minimax regret rule (Savage criterion). This rule represents a pessimistic approach used for an
opportunity loss table.

The opportunity loss reflects the difference between each payoff and the best possible payoff in a column
(it can be defined as the amount of profit foregone by not choosing the best alternative for each state of
nature). Hence, opportunity loss amounts are found by identifying the greatest payoff in a column and,
then, subtracting each of the other values in the column from that payoff.

The values in an opportunity loss table can be viewed as potential ”regrets” that might be suffered as the
result of choosing various alternatives. Minimizing the maximum possible regret requires identifying the
maximum opportunity loss in each row and, then, choosing the alternative that would yield the minimum
of those regrets (this alternative has the “best worst”).

To illustrate the described procedure, we will recall the decision problem given in Table 4.2 and first
construct the corresponding regret table.

4
Lecture 4. Decision Theory - Decision Tables and Decision Trees, Game Theory.

Payoff table:

Demand
Low Moderate High
Order 
Small 50 50 50
Medium 42 52 52
Large 34 44 54

In the first column of the payoff matrix, the largest number is 50, so each of the three numbers in that
column must be subtracted from 50. In the second column, we must subtract each payoff from 52 and in
the third column from 54. The results of these calculations are summarized in Table 4.4. A column with
the maximum loss in each row is added to this table.
 

Demand
Low Moderate High Maximum Loss
Order 
Small 0 2 4 4  Minimum
Medium 8 0 2 8
Large 16 8 0 16

Table 4.4 “Regret Table for Order Planning”

Minimizing the maximum loss, the small order should be chosen.

The minimax regret criterion disadvantage is the inability to factor row differences. It is removed in the
further rule that incorporates more of information for the choice of the best alternative.

The principle of insufficient reason (Laplace criterion) assumes that all states of nature are equally
likely. Under this assumption, the decision maker can compute the average payoff for each row (the sum
of the possible consequences of each alternative is divided by the number of states of nature) and, then,
select the alternative that has the highest row average. This procedure is illustrated by the following
calculations with the data in Table 4.3.

EMV (small) = (50 + 50 + 50)/3 = 50;

EMV (medium) = (42 + 52 + 52)/3 = 48,7;

EMV (large) = (34 + 44 + 54)/3 = 44.

Since the profits at the small order have the highest average, that order should be realized.

1.4. Decision Making under Risk

5
Lecture 4. Decision Theory - Decision Tables and Decision Trees, Game Theory.

In this case, the decision maker doesn’t know which state of nature will occur but can estimate the
probability of occurrence for each state. These probabilities may be subjective (they usually represent
estimates from experts in a particular field), or they may reflect historical frequencies.

A widely used approach to decision making under risk is expected monetary value criterion.

The expected monetary value (EMV) of an alternative is calculated by multiplying each payoff that the
alternative can yield by the probability for the relevant state of nature and summing the products. This
value is computed for each alternative, and the one with the highest EMV is selected.

Expected Monetary Value

The expected monetary value for the i-th alternative in a decision problem with m alternatives and n states
of nature is:

where
EMVi = the expected monetary value for the i-th alternative; i = 1,2,…,m
pj = the probability of the j-th state of nature; j = 1,2,…,n
vij = the payoff for the i-th alternative under the j-th state of nature

In the discussed grocer´s decision problem, suppose that the grocer can assign probabilities of low,
moderate and high demand on the basis of his experience with sale of pastry. The estimates of these
probabilities are 0.3, 0.5, 0.2, respectively. We will recall the payoff table for the considered problem.

Payoff table:

Demand →
Low Moderate High
Order 
Small 50 50 50

Medium 42 52 52

Large 34 44 54

The EMV for various sizes of the order are as follows.

EMV (small) = 0.3*50 + 0.5*50 + 0.2*50 = 50;

EMV (medium) = 0.3*42 + 0.5*52 + 0.2*52 = 49;

EMV (large) = 0.3*34 + 0.5*44 + 0.2*54 = 43.

Therefore, in accordance with the EMV criterion, the small order should be chosen.

The expected value approach (the calculation of the EMV) is particularly useful for decision making
when a number of similar decisions must be made; it is a “long-run” approach. For one-shot decisions,
especially major ones, approaches for decision making under uncertainty may be preferable.

6
Lecture 4. Decision Theory - Decision Tables and Decision Trees, Game Theory.

2. Characteristic and Format of Decision Trees

Decision tree is a graphic tool for describing the actions available to the decision maker, the events that
can occur, and the relationship between these actions and events. Decision trees are particularly useful for
analyzing situations that involve sequential decisions.

The term ”decision tree” gets its name from the treelike appearance of the diagram. A decision tree can be
deterministic or stochastic. As a prototype of decision tree, a tree with deterministic and stochastic
elements is considered.

A decision tree is composed of nodes and branches (arcs).The terminology of nodes and arcs comes
from network models which have a similar pictorial representation.

A decision tree has three types of nodes: decision nodes, chance event nodes, and terminating nodes.

Decision nodes are denoted by squares. Each decision node has one or more arcs beginning at the node
and extending to the right. Each of those arcs represents a possible decision alternative at that decision
point.

Chance event nodes are denoted by circles. Each chance event node has one or more arcs beginning at
the node and extending to the right. Each of those arcs represents a possible event at that chance event
point. The decision maker has no control over these chance events. The events associated with branches
from any chance event node must be mutually exclusive and all events included. The probability of each
event is conditional on all of the decision alternatives and chance events that precede it on the decision
tree. The probabilities for all of the arcs beginning at a chance event node must sum to 1.

A terminating node represents the end of the sequence of decisions and chance events. No arcs extend to
the right from a terminating node. No geometric picture is used to denote terminating nodes. Terminating
nodes are the starting points for the computations needed to analyze the decision tree.To construct a
decision tree, we must list the sequence of decision alternatives and events that can occur and that can
affect consequences of decisions

Every decision problem that is modelled by means of a decision table can be structured and pictured as a
decision tree (in general, the reverse statement is not true). It is shown in Fig. 4.1 that represents a
decision tree for the order planning problem given in Table 4.5. We recall this table.
 

7
Lecture 4. Decision Theory - Decision Tables and Decision Trees, Game Theory.

Fig. 4.1. “Decision Tree for Order Planning”

After the tree has been drawn, it is analyzed from right to left. The aim of this analysis is to determine the
best strategy of the decision maker, that means an optimal sequence of the decisions.

To analyze a decision tree, we must know a decision criterion, probabilities that are assigned to each
event, and revenues and costs for the decision alternatives and the chance events that occur.

For the above order planning problem, the use of a decision table in comparison with the use of a decision
tree may seem easier and simpler. However, as the decision problem becomes more complex, the decision
tree becomes more valuable in organizing the information needed to make the decision. This is especially
true if the decision maker must make a sequence of decisions, rather than a single decision, as the next
example illustrates.

Decision trees have many advantages. They make possible to obtain a visual portrayal of sequential
decisions, i. e. they picture a series of chronological decisions. They are universal, they make more
accurate the structure of the decision process (they have an “arranging” function), they facilitate a
communication among solvers of the decision problem, they force the decision maker to appreciate all
consequences of his decisions. Construction and analysis of decision trees by means of computers makes
possible to experiment with decision trees and quickly to establish the impact of changes in the input
parameters of the tree on the choice of the best strategy.

Like all models, the decision tree is an abstraction and simplification of the real problem. Only the
important decisions and events are included; otherwise, the tree becomes too “bushy” and the number of
the calculations – in spite of that they are not difficult – is too large.

We cannot use decision trees if the chance event outcomes are continuous. Instead, we must redefine the
outcomes so that there is a finite set of possibilities.

If we are not risk indifferent, it is more appropriate to use utility values instead of monetary values.

8
Lecture 4. Decision Theory - Decision Tables and Decision Trees, Game Theory.

The most important result of the analysis of a decision tree is a selection of the best alternative in the first
stage of the decision process. After this stage, some changes in the decision situations can come, an
additional information can be obtained, and usually, it is necessary to actualize the decision tree and to
determine a new optimal strategy. This procedure is necessary before every further stage.

3. Solution to Conflict Decision Problems by Use of Game Theory

The Nature of Game Theory Problems and Basic Terms

Game theory deals with decision making under conflict or competition. Decision making of this type
appears in parlour games (from this area some terms in game theory were adopted), nevertheless, there
are many examples of real-life game theory problems: international military conflicts, choice of
marketing strategies, labour-management negotiations, potential mergers and so on.

Game theory has its beginning in the 1920’s, but its greatest advance occurred in 1944, when John von
Neumann and Oskar Morgenstern, both at Princeton University, published their landmark book “Theory
of Games and Economic Behavior”.

The main characteristic of games is that two or more decision makers with conflicting objectives are
involved and the consequences of the decisions (payoffs) to each depend on the courses of action taken by
all. Each decision maker is usually trying to maximize his welfare at the expense of the others.

The following basic terms are used in game theory:

Play is a one-shot decision in a conflict situation.

Game is a series of repetitive decisions (plays).

Player is an active participant of the game (it may be a single person or a group of persons with the same
interests). A player can be rational, intelligent (if he aims at the best result of the game) or non-intelligent,
indifferent towards the result of the game (usually such a player represents a chance mechanism).

Strategy is a predetermined plan for selecting a course of action. A set of strategies for a player forms a
space of strategies for this player.

Payoff is a numerically expressed consequence of the decisions of the players. The payoff depends on the
choice of the strategies of all players and therefore we speak about payoff function.

Value of a game is an average payoff per play. A game whose value is zero is called a fair game.

A solution to game problems provides us with answers to these two questions:

 what strategy should each player follow to maximize his welfare

 what will the payoff to each player be if the recommended strategy is followed.

Matrix Games and Their Solution

The simplest type of games are two-person zero-sum games with a finite number of strategies for each
player. In these games, the sum of payoffs for both players after each play is zero (a win of one player is a
loss of the other player). Since interests of both players with regard to the outcome of the game are
diametrically opposed, these games are often called games of pure conflict or antagonistic games.

9
Lecture 4. Decision Theory - Decision Tables and Decision Trees, Game Theory.

Let us consider player A with strategies A1, A2, …, Am and player B with strategies B1, B2, …, Bn. Let us
denote the payoff function of player A for strategies Ai, Bj by M1 (Ai, Bj) and the payoff function of player
B by M2 (Ai, Bj). The equation M1 (Ai, Bj) + M2 (Ai, Bj) = 0 results in the relationship M2 (Ai, Bj) = -
M1(Ai, Bj). Therefore only one payoff function marked M (Ai, Bj)is sufficient. Its values can be arranged
in a m x n payoff matrix (table) with elements aij = M (Ai,Bj), i=1,2,…,m; j=1,2,…,n. Therefore finite
two-person zero-sum games are called matrix games.

Rows of a payoff matrix relate to player A, columns relate to player B. Payoff matrices are usually
constructed from player A’s standpoint. Under this assumption, a positive number aij indicates a win for
player A and a loss for player B, whereas a negative number aij indicates a loss for player A and a win for
player B.

A game payoff table is similar to a decision payoff table. The difference between the two is that in a
decision table, there is only one decision maker who makes decisions under conditions expressed as
“states of nature”. In a game table, there are two decision makers, one on the left and the other at the top
of the table. For this reason, a decision table is sometimes viewed as a “one-player game against nature”.

A game payoff table is illustrated by the following example.

Players A and B show simultaneously 1 or 2 or 3 fingers. If the sum of the shown fingers is an even
number, player A receives from player B so many coins how many fingers the player B showed. If the
sum of the fingers is an odd number, the player B receives from player A so many coins how many
fingers player A showed.

The payoff matrix of this game is contained in Table 4.6.

 
Table 4.6. “Payoff Table of the Morra Game”
An important term of game theory is dominance of strategies. Successive elimination of dominated
strategies results in reduction of payoff matrix size.
A strategy is said to dominate another when all payoffs in the row (or the column) of that strategy are as
good as and at least one is better than the corresponding payoffs of the other row (or column). It is
apparent that optimal behavior of players will never require the use of dominated strategies.
In Table 4.6, strategy “1 finger” is for player A as good as or better than strategy “3 fingers”, no matter
what player B does (numbers in first row of the payoff matrix are equal or greater than the corresponding
numbers in third row). Therefore strategy “3 fingers” is dominated by strategy “1 finger” and player A,
being rational, will not select it. A similar dominance of strategies exists for player B (numbers in first
column of the payoff matrix are equal or smaller than the corresponding numbers in third column and
therefore player B will not choose his third strategy).
A matrix game can be solved in pure or in mixed strategies.

10
Lecture 4. Decision Theory - Decision Tables and Decision Trees, Game Theory.

Solution in pure strategies means that only one strategy is repeatedly recommended to each player, in
contrast to solution in mixed strategies when players change from strategy to strategy if the game is
repeated.
Optimal pure strategy of player A (let us mark it A0) ensures this player a maximal win regardless of
what the other player does. Similarly, optimal pure strategy of player B (marked B0) ensures this player a
minimal loss regardless of what the other player does. In other words, any player deviating from the
optimal strategy will find either no improvement or (usually) a worsening of payoffs, if the other player
chose his optimal strategy.
The value of payoff function M (A0,B0) is called value of the game and is marked v.
The triplet (A0,B0,v) represents solution of the game in pure strategies.
If both players know their optimal pure strategies, playing the game using these strategies is rather boring.
If a matrix game has no optimal pure strategies, we solve it in mixed strategies.

Solution to Games with a Saddle Point

Rational players take the minimax approach which maximizes the minimum possible gain (for player A)
and minimizes the maximum possible loss (for player B). In other words, both players determine the
worst possible payoff associated with each of their strategies and then they select that strategy that yields
the best of these worst payoffs.

The minimax approach is illustrated by a game with the payoff matrix involved in the following table.

Player A will choose the largest of the row minimums (in the above case strategy A1), and player B will
choose the smallest of the column maximums (in the above case strategy B2). Since the maximin (the
maximum of the minimum values of the rows) is equal to the minimax (the minimum of the maximum
values of the columns), the game has solution in pure strategies (it is a pure strategy game or a strictly
determined game).
Optimal pure strategies in the considered game are A0=A1, B0=B2.
The value of payoff function M (A0, B0) is a saddle point of the payoff table (it is the number that is the
smallest in its row and at the same time the greatest in its column). The value of the saddle point
represents the value of the pure strategy game (in the solved game v = 1).

Table 4.7 “Multiple Solution Game”

11
Lecture 4. Decision Theory - Decision Tables and Decision Trees, Game Theory.

Payoff matrices can contain several saddle points. For example, the game presented in Table 4.7 has a
saddle point with the value 1. This point determines the following optimal pure strategies for both
players: (A1,B2). For both these strategies the value of the game is 1. An optimal strategy of player A (B)
is determined by any row (column) containing a saddle point.

12