Decision Theory and Decision Trees 1

11.7 DECISION TREE ANALYSIS

Decision-making problems discussed earlier were limited to arrive at a decision over a fixed period of
time. That is, payoffs, states of nature, courses of action and probabilities associated with the
occurrence of states of nature were not subject to change.
However, situations may arise when a decision-maker needs to revise his previous decisions due to
availability of additional information. Thus he intends to make a sequence of interrelated decisions over
several future periods. Such a situation is called a sequential or multiperiod decision process. For example,
in the process of marketing a new product, a company usually first go for ‘Test Marketing’ and other
alternative courses of action might be either ‘Intensive Testing’ or ‘Gradual Testing’. Given the various
possible consequences – good, fair, or poor, the company may be required to decide between redesigning
the product, an aggressive advertising campaign or complete withdrawal of product, etc. Based on this
decision there might be an outcome that leads to another decision and so on.
A decision tree analysis involves the construction of a diagram that shows, at a glance, when decisions Decision tree is
are expected to be made – in what sequence, their possible outcomes, and the corresponding payoffs. the graphical display
of the progression of
A decision tree consists of nodes, branches, probability estimates, and payoffs. There are two types
decision and random
of nodes: events.
 Decision (or act) node: A decision node is represented by a square and represents a point of time
where a decision-maker must select one alternative course of action among the available. The
courses of action are shown as branches or arcs emerging out of decision node.
 Chance (or event) node: Each course of action may result in a chance node. The chance node
is represented by a circle and indicates a point of time where the decision-maker will discover the
response to his decision.
Branches emerge from and connect various nodes and represent either decisions or states of nature. There
are two types of branches:
 Decision branch: It is the branch leading away from a decision node and represents a course of
action that can be chosen at a decision point.
 Chance branch: It is the branch leading away from a chance node and represents the state of
nature of a set of chance events. The assumed probabilities of the states of nature are written
alongside their respective chance branch.
2 Operations Research: Theory and Applications

 Terminal branch: Any branch that makes the end of the decision tree (not followed by either a
decision or chance node), is called a terminal branch. A terminal branch can represent either a
course of action. The terminal points of a decision tree are supposed to be mutually exclusive
points so that exactly one course of action will be chosen.
The payoff can be positive (i.e. revenue or sales) or negative (i.e. expenditure or cost) and it can be
associated either with decision or chance branches.
An illustration of a decision tree is shown in Fig. 11.2. It is possible for a decision tree to be
deterministic or probabilistic. It can also further be divided in terms of stages – into single stage (a decision
under condition of certainty) and multistage (a sequence of decisions).

Fig. 11.2
Decision Tree

The optimal sequence of decisions in a tree is found by starting at the right-hand side and rolling
backwards. At each node, an expected return is calculated (called position value). If the node is a chance
node, then the position value is calculated as the sum of the products of the probabilities or the branches
emanating from the chance node and their respective position values. If the node is a decision node, then
the expected return is calculated for each of its branches and the highest return is selected. This procedure
continues until the initial node is reached. The position values for this node corresponds to the maximum
expected return obtainable from the decision sequence.
 Decision Theory and Decision Trees 3
Remark Decision trees versus probability trees Decision trees are basically an extension of probability
trees. However, there are several basic differences:
(i) The decision tree utilizes the concept of ‘rollback’ to solve a problem. This means that it starts
at the right-hand terminus with the highest expected value of the tree and works back to the
current or beginning decision point in order to determine the decision or decisions that should
be made. It is the multiplicity of decision points that make the rollback process necessary.
(ii) The probability tree is primarily concerned with calculating the probabilities, whereas the
decision tree utilizes probability factors as a means of arriving at a final answer.
(iii) The most important feature of the decision tree, is that it takes time differences of future earnings
into account. At any stage of the decision tree, it may be necessary to weigh differences in
immediate cost or revenue against differences in value at the next stage.
Example 11.20 You are given the following estimates concerning a Research and Development programme:
Decision Probability of Decision Outcome Probability of Payoff Value
Di Di Given Research R Number Outcome xi Given Di of Outcome, xi
P(Di | R ) P(xi| Di ) (Rs ’000)
Develop 0.5 1 0.6 600
2 0.3 – 100
3 0.1 0

Do not develop 0.5 1 0.0 600

2 0.0 – 100
3 1.0 0

Construct and evaluate the decision tree diagram for the above data. Show your workings for evaluation.
Solution The decision tree of the given problem along with necessary calculations is shown in Fig. 11.3.

Fig. 11.3
Decision Tree

Example 11.21 A glass factory that specializes in crystal is developing a substantial backlog and for
this the firm’s management is considering three courses of action: To arrange for subcontracting ( S1), to
begin overtime production ( S2 ), and to construct new facilities (S3). The correct choice depends largely
upon the future demand, which may be low, medium, or high. By consensus, management ranks the
respective probabilities as 0.10, 0.50 and 0.40. A cost analysis reveals the effect upon the profits. This is
shown in the table below:
4 Operations Research: Theory and Applications

Demand Probability Course of Action

S1 S2 S3
(Subcontracting) (Begin Overtime) (Construct Facilities)

Low (L) 0.10 10 – 20 – 150

Medium (M ) 0.50 50 60 20
High (H ) 0.40 50 100 200

Show this decision situation in the form of a decision tree and indicate the most preferred decision and
its corresponding expected value.
Solution A decision tree that represents possible courses of action and states of nature is shown in Fig.
11.4. In order to analyze the tree, we start working backwards from the end branches.
The most preferred decision at the decision node 0 is found by calculating the expected value of each
decision branch and selecting the path (course of action) that has the highest value.

Fig. 11.4
Decision Tree

Since node 3 has the highest EMV, therefore, the decision at node 0 will be to choose the course of
action S3, i.e. construct new facilities.
Example 11.22 A businessman has two independent investment portfolios A and B, available to him,
but he lacks the capital to undertake both of them simultaneously. He can either choose A first and then
stop, or if A is not successful, then take, B or vice versa. The probability of success of A is 0.6, while for
B it is 0.4. Both investment schemes require an initial capital outlay of Rs 10,000 and both return nothing
if the venture proves to be unsuccessful. Successful completion of A will return Rs 20,000 (over cost) and
successful completion of B will return Rs 24,000 (over cost). Draw a decision tree in order to determine
the best strategy. [Delhi Univ., MBA, 2000, AMIE, 2006]
 Decision Theory and Decision Trees 5
Solution The decision tree based on the given information is shown in Fig. 11.5. The evaluation of each
chance node and decision is given in Table 11.33.

Decision Point Outcome Probability Conditional Value Expected Value

(Rs)
D3 (i) Accept A Success 0.6 20,000 12,000
Failure 0.4 –10,000 – 4,000
8,000
(ii) Stop – – – 0
D2 (i) Accept B Success 0.4 24,000 9,600
Failure 0.6 –10,000 – 6,000
3,600
(ii) Stop – – – 0
D1 (i) Accept A Success 0.6 20,000 + 3,600 = 23,600 14,160
Failure 0.4 –10,000 – 4,000
10,160
(ii) Accept B Success 0.4 24,000 + 8,000 = 32,000 12,800
Failure 0.6 –10,000 – 6,000 Table 11.33
6,800 Evaluation of
(iii) Do nothing – – – 0 Decision and Chance
Nodes

Fig. 11.5
Decision Tree

Since the EMV = Rs 10,160 at node D1 is highest, therefore the best strategy is to accept course of
action A first and if A is successful, then accept B.
6 Operations Research: Theory and Applications

Example 11.23 The Oil India Corporation (OIC) is wondering whether to go for an offshore oil drilling
contract that is to be awarded in Bombay High. If OIC bid, value would be Rs 600 million with a 65 per
cent chance of gaining the contract. The OIC may set up a new drilling operation or move the already
existing operation, which has already proved successful for a new site. The probability of success and
expected returns are as follows:

Outcome New Drilling Operation Existing Operation

Probability Expected Revenue Probability Expected Revenue
(Rs million) (Rs million)
Success 0.75 800 0.85 700
Failure 0.25 200 0.15 350

If the Corporation do not bid or lose the contract, they can use Rs 600 million to modernize their
operation. This would result in a return of either 5 per cent or 8 per cent on the sum invested with
probabilities 0.45 and 0.55. (Assume that all costs and revenue have been discounted to present value.)
(a) Construct a decision tree for the problem showing clearly the courses of action.
(b) By applying an appropriate decision criterion recommend whether or not the Oil India Corporation
should bid the contract. [Delhi Univ. MBA, AMIE, 2001, 2005]

Solution The decision tree based on the given information is shown in Fig. 11.6. The evaluation of each
chance node and decision node is given in Table 11.34.

Fig. 11.6
Decision Tree
 Decision Theory and Decision Trees 7

Decision Point Outcome Probability Conditional Value Expected Value

(Rs)
D3 (i) Modernize 5% return 0.45 600 × 0.05 = 30 30 × 0.45 = 13.5
8% return 0.55 600 × 0.08 = 48 48 × 0.55 = 26.4
39.9
D2 (i) Undertake new Success 0.75 800 600
drilling operation Failure 0.25 200 50
650
(ii) Move existing Success 0.85 700 595
operation Failure 0.15 350 52.5
647.5

D1 (i) Modernize 5% return 0.45 600 × 0.05 = 30 30 × 0.45 = 13.5

8% return 0.55 600 × 0.08 = 48 48 × 0.55 = 26.4
39.9
(ii) Bid Success 0.65 650 422.50 Table 11.34
Failure 0.35 39.9 13.96 Evaluation of
Decision and Chance
436.46 Nodes

Since EMV, Rs 436.46 at event node 2 is highest, therefore the best decision at decision node D1 is
to decide for bid and if successful establish a new drilling operation.
Example 11.24 A large steel manufacturing company has three options with regard to production: (i) produce
commercially (ii) build pilot plant (iii) stop producing steel. The management has estimated that their pilot plant,
if built, has 0.8 chance of high yield and 0.2 chance of low yield. If the pilot plant does show a hight yield,
management assigns a probability of 0.75 that the commercial plant will also have a high yield. If the pilot
plant shows a low yield, there is only a 0.1 chance that the commercial plant will show a high yield. Finally,
management’s best assessment of the yield on a commercial-size plant without building a pilot plant first has a 0.6
chance of high yield. A pilot plant will cost Rs. 3,00,000. The profits earned under high and low yield conditions
are Rs. 1,20,00,000 and – Rs. 12,00,000 respectively. Find the optimum decision for the company. [Punjab
Tech Univ., BE, 2006]
Solution A decision tree representing possible courses of action and states of nature are shown in Fig. 11.7.
In order to analyse the tree, we proceed backward from the end branches

Fig. 11.7