Decision-Making
Self-Instructional Material 133
NOTES
UNIT 4 DECISION-MAKINGStructure
4.0 Introduction4.1 Unit Objectives4.2 Decision-Making Environment
4.2.1 Decision-Making under Certainty4.2.2 Decision-Making under Uncertainty4.2.3 Decision-Making under Risk
4.3 Decision Tree Analysis4.4 Integer Programming and Dynamic Programming: Concepts
and Advantages4.4.1 Importance of Integer Programming Problems4.4.2 Applications of Integer Programming4.4.3 Decision Tree and Bellman’s Principle of Optimality4.4.4 Characteristics of DPP4.4.5 Dynamic Programming Algorithm4.4.6 Advantages of DPP
4.5 Summary4.6 Key Terms4.7 Answers to ‘Check Your Progress’4.8 Questions and Exercises4.9 Further Reading
4.0 INTRODUCTION
Decision-making is an everyday process in life. It is the major job of a manager too.A decision taken by a manager has far-reaching consequences on the business. Rightdecisions will have a salutary effect and wrong ones may prove to be disastrous.
Decisions may be classified into two categories, tactical and strategic. Tacticaldecisions are those which affect the business in the short run. Strategic decisionsare those which have a long-term effect on the course of business.
These days, in every organization—large or small, the person at the top has totake the crucial decisions, knowing that certain events beyond his control may occurand make him regret the decision. He is uncertain as to whether or not theseunfortunate events will happen. In such situations the best possible decision can bemade with the use of statistical tools which try to minimize the degree to which theperson is likely to regret the decision he takes for a particular problem.
The problem under study may be represented by a model in terms of the followingelements:
(i) The decision-maker The decision-maker is charged with theresponsibility of taking a decision; he has to select one from a set ofpossible courses of action.
(ii) Acts These are the alternative courses of action or strategies that areavailable to the decision-maker. The decision involves a selection amongtwo or more alternative courses of action. The problem is to choose thebest of these alternatives to achieve an objective.
Operations Research
134 Self-Instructional Material
NOTES
(iii) Events These are the occurrences which affect the achievement of theobjectives. They are also called states of nature or outcomes. The eventsconstitute a mutually exclusive and exhaustive set of outcomes whichdescribe the possible behaviour of the environment in which the decisionis made. The decision-maker has no control over which event will takeplace and can only attach a subjective probability of occurence of each.
(iv) Pay-off table It represents the economics of a problem, i.e., the revenueand costs associated with any action with a particular outcome. It is anordered statement of profit or costs resulting under a given situation. Apay-off can be interpreted as the outcome in quantitative form if thedecision-maker adopts a particular strategy under a particular state ofnature.
(v) Opportunity loss table An opportunity loss is the loss incurred becauseof the failure to take the best possible action. Opportunity losses arecalculated separately for each state of nature that might occur. Giventhe occurence of a specific state of nature, we can determine the bestpossible act. For a given state of nature, the opportunity loss of an act isthe difference between the pay-off of that act and the pay-off for thebest act that could have been selected.
4.1 UNIT OBJECTIVES
After going through this unit, you will be able to:● Analyse the nature of a decision-making environment● Understand the various choices available for decision-making under
uncertainty● Describe how decisions are taken under risk● Describe the procedure of decision tree analysis● Understand the importance and applications of integer programming● Describe the characteristic features of dynamic programming problems
4.2 DECISION-MAKING ENVIRONMENT
In any decision problem, the decision-maker is concerned with choosing from amongthe available alternative courses of action, the one that yields the best result. If theconsequences of each choice are known with certainty, the decision-maker can easilymake decisions. But in most of real life problems, the decision-maker has to dealwith situations where uncertainty of the outcomes prevail.
The decision-making problems can be discussed under the following heads onthe basis of their environments:
1. Decision-making under certainty2. Decision-making under uncertainty3. Decision-making under risk4. Decision-making under conflict
Decision-Making
Self-Instructional Material 135
NOTES
4.2.1 Decision-Making under Certainty
In this case, the decision-maker knows with certainty the consequences of everyalternative or decision choice. The decision-maker presumes that only one state ofnature is relevant for his purpose. He identifies this state of nature, takes it forgranted and presumes complete knowledge as to its occurence.
4.2.2 Decision-Making under Uncertainty
When the decision-maker faces multiple states of nature but he has no means toarrive at probability values to the likelihood of occurence of these states of nature,the problem is a decision problem under uncertainty. Such situations arise when anew product is introduced in the market or a new plant is set up. In business, thereare many problems of this ‘nature’. Here, the choice of decision largely depends onthe personality of the decision-maker.
The following choices are available before the decision-maker in situations ofuncertainty.
(a) Maximax Criterion(b) Minimax Criterion(c) Maximin Criterion(d) Laplace Criterion (Criterion of equally likelihood)(e) Hurwicz Alpha Criterion (Criterion of Realism)( a ) The maximax decision criterion (criterion of optimism) The term
‘maximax’ is an abbreviation of the phrase maximum of the maximums,and an adventurous and aggressive decision-maker may choose to takethe action that would result in the maximum pay-off possible. Supposefor each action there are three possible pay-offs corresponding to threestates of nature as given in the following decision matrix:
DecisionsStates ofnature A1 A2 A3
S1 220 180 100S2 160 190 180S3 140 170 200
Maximum under each decision are (220, 190, 200). The maximum of thesethree maximums is 220. Consequently, according to the maximax criteria the decisionto be adopted is A1.
(b) The minimax decision criterion Minimax is just the opposite ofmaximax. Application of the minimax criteria requires a table of lossesinstead of gains. The losses are the costs to be incurred or the damagesto be suffered for each of the alternative actions and states of nature.The minimax rule minimizes the maximum possible loss for each courseof action. The term ‘minimax’ is an abbreviation of the phrase minimumof the maximum.Under each of the various actions, there is a maximumloss and the action that is associated with the minimum of the variousmaximum losses is the action to be taken according to the minimaxcriterion. Suppose the loss table is
Operations Research
136 Self-Instructional Material
NOTES
DecisionsState ofnature A1 A2 A3
S1 0 4 10S2 3 0 6
S3 18 14 0
It shows that the maximum losses incurred by the various decisionsA1 A2 A3
18 14 10and the minimum among three maximums is 10 which is under actionA3. Thus, according to minimax criterion, the decision-maker shouldtake action A3.
(c) The maximin decision criterion (criterion of pessimism) Themaximin criterion of decision-making stands for choice betweenalternative courses of action assuming pessimistic view of nature. Takingeach act in turn, we note the worst possible results in terms of pay-offand select the act which maximizes the minimum pay-off. Suppose thepay-off table is
ActionState ofnature A1 A2 A3
S1 –80 –60 –20S2 –30 –10 –2S3 30 15 7S4 75 80 25
Minimum under each decision are respectively–80 –60 –20The action A3 is to be taken according to this criterion because it is themaximum among minimums.
(d) Laplace criterion As the decision-maker has no information about theprobability of occurrence of various events, he makes a simpleassumption that each probability is equally likely. The expected pay-offis worked out on the basis of these probabilities. The act havingmaximum expected pay-off is selected.Example 4.1:
ActEvents
A1 A2 A3
E1 20 12 25E2 25 15 30E3 30 20 22
Decision-Making
Self-Instructional Material 137
NOTES
Solution: We associate equal probability for each event say, 1/3.Expected pay-offs are:
1
2
3
1 1 1 75A 20 25 30 253 3 3 31 1 1 47A 12 15 20 15.673 3 3 31 1 1 77A 25 30 22 25.673 3 3 3
⎛ ⎞ ⎛ ⎞ ⎛ ⎞→ × + × + × = =⎜ ⎟ ⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠ ⎝ ⎠⎛ ⎞ ⎛ ⎞ ⎛ ⎞→ × + × + × = =⎜ ⎟ ⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠ ⎝ ⎠⎛ ⎞ ⎛ ⎞ ⎛ ⎞→ × + × + × = =⎜ ⎟ ⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠ ⎝ ⎠
Since A3 has the maximum expected pay-off, A3 is the optimal act.(e) Harwicz alpha criterion This method is a combination of maximum
criterion and maximax criterion. In this method, the decision-maker’sdegree of optimism is represented by α ,the coefficient of optimism. αvaries between 0 and 1. When α = 0, there is total pessimism and whenα = 1, there is total optimism.we find D1, D2, D3, etc. connected with all strategies where Di = a Mi +(1 – a) mi where Mi is the maximum pay-off of ‘i’ the strategy and mi isthe minimum pay-off of ‘i’ th strategy. The strategy with highest of D1,D2, ...... is chosen. The decision-maker will specify the value ofα depending upon his level of optimism.Example 4.2:
ActEvents
A1 A2 A3
E1 20 12 25E2 25 15 30E3 30 20 22
Solution: Let α = .6for A1 max. pay-off = 30Min. pay-off = 20
( ) ( ) 26206.1306.1 =−+×=∴D
Similarly, 2 .6 30 1 .6 12D 22.8
3 .6 30 1 .6 22 26.8D
Since D3 is max, select the act A3.
4.2.3 Decision-Making under Risk
In this situation, the decision-maker has to face several states of nature. But he hassome knowledge or experience which will enable him to assign probability to theoccurence of each state of nature. The objective is to optimize the expected profit,or to minimize the opportuiniy loss.
For decision problems under risk, the most popular methods used are ExpectedMonetary Value (EMV) criterion, Expected Opportunity Loss (EOL) criterion orExpected Value of Perfect Information (EVPI).
Check Your Progress
1. What are the two categoriesof decisions?
2. What are events?3. List the various choices
available for the decision-maker under uncertainsituations.
Operations Research
138 Self-Instructional Material
NOTES
(a) EMV Criterion When probabilities can be assigned to the various statesof nature, it is possible to calculate the statistical expectation of gain foreach course of action.The conditional value of each event in the pay-off table is multiplied byits probability and the product is summed up. The resulting number isthe EMV for the act. The decision-maker then selects from the availablealternative actions, the action that leads to the maximum expected gain(that is the action with highest EMV).Consider the following example. Let the states of nature be S1and S2and the alternative strategies be A1 and A2. Let the pay-off table be asfollows.
A1 A2
S1 30 20S3 35 30
Let the probabilities for the states of nature S1and S2be 6 and 4respectively .Then,EMV for A1 = (30 × .6) + (35 × .4) = 18 + 14 = 32EMV for A2 = (20 × .6) + (30 × .4) = 12 + 12 = 24Since EMV for A1 is greater, the decision-maker will choose the strategyA1.
(b) EOL Criterion The difference between the greater pay-off and the actualpay-off is known as opportunity loss. Under this criterion, the strategywhich has minimum Expected Opportunity Loss (EOL) is chosen. Thecalculation of EOL is similar to that of EMV.The following is an opportunity loss table. A1and A2 are the strategiesand S1and S2 are the states of nature.
A1 A2
S1 0 10S2 2 –5
Let the probabilities for two states be .6 and .4.EOL for A1 = (0 × .6) + (2 × .4) = 0.8EOL for A2 = (10 × .6) + (–5 × .4) = 6 – 2 = 4EOL for A1 is least. Therefore, the strategy A1may be chosen.
(c) EVPI Method The expected value of perfect information (EVPI) is theaverage (expected) return in the long run, if we have perfect informationbefore a decision is to be made.In order to calculate EVPI, we choose the best alternative with theprobability of their state of nature. EVPI is the expected outcome withperfect information minus the outcome wth max EMV.∴ EVPI = Expected value with perfect information – max. EMVConsider the following example.
Decision-Making
Self-Instructional Material 139
NOTES
Example 4.3: A1, A2, A3 are the acts and S1, S2, S3 are the states of nature. Also,known that P(S1) = .5, P(S2) = .4 and P(S3) = .1
Solution: Pay-off table is as follows:Pay off table
State of natureA1 A2 A3
S1 30 25 22S2 20 35 20S3 40 30 35
EMV for ( ) ( ) ( ) 274815401.204.305.A1 =++=×+×+×=
EMV for ( ) ( ) ( ) 5.293145.12301.354.255.A2 =++=×+×+×=
EMV for ( ) ( ) ( ) 5.225.3811351.204.225.A3 =++=×+×+×=
The highest EMV is for the strategy A2 and it is 29.5.Now to find EVPI, work out the expected value for maximum pay-off under all
states of nature.Max. profitof each state Probability Expected value
( = Prob. × Profit)S1 30 .5 15S2 35 .4 14S3 40 .1 4
∴ Expected pay-off with perfect information = 33.∴ Thus, the expected value of perfect information (EVPI) = Expected Value
with Perfect Information – Maximum EMV = 33 – 29.5 = 3.5.Example 4.4: You are given the following pay-offs of three acts A1, A2 and A3,
and the states of nature S1, S2 and S3. States of nature A1 A2 A3
S1 25 –10 –125S2 400 440 400S3 650 740 750
The probabilities of the states of nature are respectively, .1, .7 and .2. Calculateand tabulate the EMV and conclude which of the acts can be chosen as the best.
Solution:Act A1
Prob. × Pay off Act A2
Prob. × Pay off Act A3
Prob. × Pay off .1 × 25 = 2.5
.7 × 400 = 280
.2 × 650 = 130
.1 × –10 = –1
.7 × 440 = 308
.2 × 740 = 148
.1 × –125 = –12.5 .7 × 400 = 280 .2 × 750 = 150
∴ EMV for A1 = 412.5, EMV for A2 = 455, EMV for A3 = 417.5Since EMV is maximum for A2 choose the Act A2.
Operations Research
140 Self-Instructional Material
NOTES
Example 4.5: A management is faced with the problem of choosing one of theproducts for manufacturing. The probability matrix after market research for thetwo products was as follows.
State of natureAct Good Fair PoorProduct A 0.75 0.15 0.10Product B 0.60 0.30 0.10
The profit that the management can make for different levels of market accept-ability of the products are as follows.
State of nature Profit (in Rs) if market is
Acts Good Fair PoorProduct A 35,000 15,000 5,000Product B 50,000 20,000 Loss of 3,000
Calculate the expected value of the choice of alternatives and advise themanagement.
Solution: Let us put this information in a pay-off matrix with probabilitiesassociated with the states of nature.
State ofnature
Product AProfit × Probability
Product BProfit × Probability
GoodFairPoor
35000 × 0.75 = 26250 1500 × 0.15 = 2250 5000 × 0.10 = 500
50000 × 0.60 = 3000020000 × 0.30 = 6000–3000 × 0.10 = –300
EMV 29000 35700
Since the expected pay-off (EMV) of product B is greater, product B should bepreferred by the management.
Preparation of pay-off table
Example 4.6: An ink manufacturer produces a certain type of ink at a totalaverage cost of Rs 3 per bottle and sells at a price of Rs 5 per bottle. The ink isproduced over the weekend and is sold during the following week. According topast experience, the weekly demand has never been less than 78 or greater than 80bottles in his place.
You are required to formulate the pay-off table.Solution: The different states of nature are the demand for 78 units, 79 units or
80 units. Call them S1, S2, S3.The alternative courses of action are selling 78 units, 79 units or 80 units. Call
them A1, A2, A3.Selling price of ink = Rs 5- per bottleCost price = Rs 3- per bottleCalculation of pay-offs (Pay-off stands for the gain)Sale quantity× price – Production quantity× Cost
Decision-Making
Self-Instructional Material 141
NOTES
160240400380580SA158237395379579SA156234390378578SA155240395380578SA158237395379579SA156234390378578SA150240390380578SA153237390379578SA156234390378578SA
33
32
31
23
22
21
13
12
11
=−=×−×==−=×−×==−=×−×==−=×−×==−=×−×==−=×−×==−=×−×==−=×−×==−=×−×=
(Explanation: A1S1 means selling quantity is 78 and manufacturing quantity is78. A2S1 means sales 78, production 79, and so on.)
Pay-off Table
Act (strategy)State of nature(events) A1 A2 A3
S1
S2
S3
156156156
153158158
150155160
Note: We shall show the states of nature in rows and acts in columns.
Preparation of loss table
Example 4.7: An ink manufacturer produces a certain type of ink at a totalaverage cost of Rs 3 per bottle and sells at a price of Rs 5 per bottle. The ink isproduced over the weekend and is sold during the following week. According topast experience, the weekly demand has never been less than 78 or greater than 80bottles in his place.
You are required to formulate the loss table.Solution: Calculation of regret (opportunity loss)A1S1= 0 (since production and sales are of equal quantity say 78)A2S1 = 1× 3 = 3 (since one unit of production is in excess whose cost = Rs 3)A3S1 = 2× 3 = 6 (since 2 units of production are in excess whose unit cost is
@ Rs 3)A1S2 = 1× 2 = 2 (since the demand of one unit is more than produced, the profit
for one unit is Rs 2)Similarly, A2S2 = 0 (since units of production = units of demand)A3S2 = 1 × 3 = 3 A1S3 = 2× 2 = 4A2S3 = 2× 1 = 2 and A3S3 = 0
Opportunity loss table
ActionState of nature(events) A1 A2 A3
S1
S2
S3
024
302
630
Operations Research
142 Self-Instructional Material
NOTES
From the pay-off table oportunity loss table can also be prepared.Method: Let every row of the pay-off table represent a state of nature and every
column represent a course of action. Then from each row select the highest pay-offand substract all pay-offs of that row from it. They are the opportunity losses.
See the following examples.Example 4.8: The following is a pay-off table. From it, form a regret
(opportunity loss) table.Pay-off table
States of nature A1 A2 A3
E1 156 153 150 E2 156 158 155 E3 156 158 160
Solution:Opportunity Loss Table
ActionState ofnature A1 A2 A3
E1
E2
E3
156 –156 = 0158 – 156 = 2160 – 156 = 4
156 – 153 = 3158 – 158 = 0160 – 158 = 2
156 – 150 = 6158 – 155 = 3160 – 160 = 0
Example 4.9: The following is a pay-off table.Event (State of nature)
Action E1 E2 E3 E4 A1 50 300 –150 50 A2 400 0 100 0 A3 –50 200 0 100 A4 0 300 300 0
Suppose that the probabilities of the events in this table are P(E1) = 0.15; P(E2)= 0.45; P(E3) = 0.25; P(E4) = 0.15.
Calculate the expected pay-off. Prepare the opportunity loss table (Regret table)and calculate the expected loss of each action.
Solution:Hint: Rewrite the question with E1, E2, E3, E4 as rows and A1, A2, A3, A4 as
columns.Acts
EventsA1 A2 A3 A4
E1
E2
E3
E4
50300
–15050
4000
1000
– 50200
0100
0300300
0
Decision-Making
Self-Instructional Material 143
NOTES
Calculation of Expected Pay-off (EMV)A1
Pay-off × Prob. A2
Pay-off × Prob. A3
Pay-off × Prob. A4
Pay-off × Prob. 50 × .15 = 7.5
300 × .45 = 135 –150 × .25 = –37.5
50 × .15 = 7.5
400 × .15 = 60 0 × .45 = 0
100 × .25 = 25 0 × .15 = 0
– 50 × .15 = 75 200 × .45 = 90
0 × .25 = 0 100 × .15 = 15
0 × .15 = 0 300 × .45 = 135 300 × .25 = 75
0 × .15 = 0 EMV = 112.5 EMV = 85 EMV = 112.5 EMV = 210
Opportunity Loss Table
A1 A2 A3 A4
E1
E2
E3
E4
400 – 50 = 350300 – 300 = 0300 + 150 = 450100 – 50 = 50
400 – 400 = 0300 – 0 = 300300 – 100 = 200100 – 0 = 100
400 + 50 = 450300 – 200 = 100300 – 0 = 300100 – 100 = 0
400 – 0 = 400300 – 300 = 0300 – 300 = 0100 – 0 =100
Calculation Expected Loss (EOL)A1
Loss × Prob.A2
Loss × Prob.A3
Loss × Prob.A4
Loss × Prob.350 × .15 = 52.5
0 × .45 = 0450 × .25 = 112.5
50 × .15 = 7.5
0 × .15 = 0300 × .45 = 135200 × .25 = 50100 × .15 = 15
450 × .15 = 67.5100 × .45 = 45300 × .25 = 75
0 × .15 = 0
400 × .15 = 600 × .45 = 00 × .25 = 0
100 × .15 = 15EOL = 172.5 EOL = 200 EOL = 187.5 EOL = 75
Example 4.10: A newspaper boy has the following probability of selling amagazine:
No. of copies sold Probability10 0.1011 0.1512 0.2013 0.2514 0.30
The cost of a copy is 30 paise and sale price is 50 paise. He cannot returnunsold copies. How many copies should he order?
Solution: We can apply either EMV criterion or EOL criterion. Let us applyEMV criterion for which we have to calculate the pay-off. Number of copies orderedare the different courses of action. The copies ordered may be 10, 11, 12, 13, 14.Denote them by A1, A2, A3, A4, A5.
Similarly, number of copies demanded may be 10, 11, 12, 13 or 14. Thesedemands may be D1, D2, D3, D4, D5. These are events. The pay-off values arecalculated as follows:
Selling price of each item = 50 paise and cost of a copy = 30 paiseA1D1 = (10 × 50) – (10 × 30) = 200A2D1 = (10 × 50) – (11 × 30) = 170A3D1 = (10 × 50) – (12 × 30) = 140A4D1 = (10 × 50) – (13 × 30) = 110, and so on.
Operations Research
144 Self-Instructional Material
NOTES
The 25 pay-off values can thus be calculated. They are shown as follows:A1 A2 A3 A4 A5
D1 200 170 140 110 80D2 200 220 190 160 130D3 200 220 240 210 180D4 200 220 240 260 230D5 200 220 240 260 280
The given probabilities are .10, .15, .20, .25, .30.Calculation of EMV for all the acts.
A1
Pay off × Prob.A1
Pay off × Prob.A1
Pay off × Prob..10 × 200 = 20.15 × 200 = 30.20 × 200 = 40.25 × 200 = 50.30 × 200 = 50
.10 × 170 = 17
.15 × 220 = 33
.20 × 220 = 44
.25 × 220 = 55
.30 × 220 = 66
.10 × 140 = 20.15 × 190 = 28.5.20 × 240 = 48.25 × 240 = 60.30 × 240 = 72
200 215 222.5
A4
Pay off × ProbA5
Pay off × Prob.10 × 110 = 20.15 × 160 = 30.20 × 210 = 40.25 × 260 = 50.30 × 260 = 50
.10 × 80 = 8.15 × 130 = 19.5
.20 × 180 = 36
.25 × 230 = 55
.30 × 280 = 84 220 205
∴ EMV for the acts A1, A2, A3, A4, and A5 are respectively 200, 215, 222.5, 220and 205.
EMV for A3 is greater and therefore A3 is optimal act.∴No. of copies to be ordered = 12.
Example 4.11: A grocery store with a bakery department is faced with theproblem of how many cakes to buy in order to meet the day’s demand. The grocerprefers not to sell day-old goods in competition with fresh products. Leftover cakesare, therefore, a complete loss. On the other hand, if a customer desires a cake andall of them have been sold, the disappointed customer will buy elsewhere and thesales will be lost. The grocer has, therefore collected information on the past salesin a selected 100-day period as shown in the following table:
Sales per day No. of days Probability25 10 0.1026 30 0.3027 50 0.5028 10 0.10
100 1.00
Construct the pay-off table and the opportunity loss table. What is the optimalnumber of cakes that should be bought each day? Apply both EMV and EOL criteria.
Decision-Making
Self-Instructional Material 145
NOTES
Also find (and interpret) EVPI (Expected Value of Perfect Information). A cakecosts Rs 0.80 and sells for Re 1.
Solution: Let A1, A2, A3, and A4 for strategies and S1, S2, S3, S4 stands for statesof nature.
Then A1, A2, A3, A4 respectively stand for stocking 25, 26, 27,28 cakes. S1, S2,S3, S4 respectively stand for demands for 25, 26, 27, 28 cakes.
Conditional pay-off values can be obtained as explained earlier. The pay-offvalues thus obtained are as follows.
Pay-off table
Alternative strategiesState of nature(Demand) A1(25) A2(26) A3(27) A4(28)
S1(25)S2(26)S3(27)S4(28)
5.005.005.005.00
4.205.205.205.20
3.404.405.405.40
2.603.604.605.60
Probabilities are .10, .30, .50, .10
EMV for Act A1 = ( ) ( ) ( ) ( ) 5.10.550.530510.5 =×+×+×+×
Similarly, EMV values A2, A3, A4 are 5.10, 4.9, 4.2 respectively.The maximum EMV is seen strategy A2. Thus, according to the EMV decision
criterion, the store would stock 26 cakes (A2).Regret (Opportunity loss) table
Alternative strategiesState of nature(Demand) A1(25) A2(26) A3(27) A4(28)
S1(25)S2(26)S3(27)S4(28)
00.20.40.6
0.80
0.20.4
1.60.80
0.2
2.41.60.80
Probabilities are 0.10, 0.30, 0.50, 0.10
Expected opportunity loss for A1 = ( ) ( )+×+× 302100 ... ( ) ( ) ...... 32106504 =×+×
Similarly for A2, A3, A4 expected opportunity losses are 0.22, 0.42 and 1.12respectively.
EOL is least for A2. \ A2 is the optimal Act.Hence, 26 is the optimal number to buy.To find EVPI, select the highest pay-off in each row and find the expected
value. Then we havePay-off Prob. Payoff× Prob.
5.00 0.10 0.505.20 0.30 1.565.40 0.50 2.705.60 0.10 0.56
Total 5.32
Operations Research
146 Self-Instructional Material
NOTES
∴Expected value with perfect information = 5.32
Max. EMV is for Act 2A which is equal to 5.10.
Expected value of perfect information (EVPI) = Expected Value with PerfectInformation – Highest EMV (EMV of A2) = 5.32 – 5.10 = .32
Example 4.12: A factory produces 3 varieties of fountain pens. The fixed andvariable costs are as follows:
Fixed Cost Variable CostType 1 Rs 2,00,000 Rs 10Type 2 Rs 3,20,000 Rs 8Type 3 Rs 6,00,000 Rs 6
The likely demands under three situations are given as follows:Demand UnitsPoor 25,000Moderate 1,00,000High 1,50,000
If the price of each type is Rs 20, prepare the pay-off table after showing necessarycalculations.
Solution: Let T1, T2, T3 stand for Type 1, Type 2, and Type 3 and D1, D2 and D3for poor demand, moderate demand and high demand respectively.
The pay-off (in thousands) = Sales revenue from the estimated demand – Totalvariable cost – Fixed cost
15006009003000600)1506()15020(148032012003000320)1508()15020(
130020015003000200)15010()15020(8006006002000600)1006()10020(8803208002000320)1008()10020(
80020010002000200)10010()10020(250600150500600)256()2520(20320200500320)258()2520(
50200250500200)2510()2520(
33
32
31
23
22
21
13
12
11
+=−−=−×−×=+=−−=−×−×=+=−−=−×−×=
+=−−=−×−×=+=−−=−×−×=+=−−=−×−×=
−=−−=−×−×=−=−−=−×−×=+=−−=−×−×=
DTDTDTDTDTDTDTDTDT
The pay-off table (in ’000s)
T1 T2 T3
D1
D2
D3
50800
1300
–20880
1480
–250800
1500Note:
(i) Expected value of sales = Sum of the products of various values of sales and probabili-ties
(ii) Expected Monetary value of an act = Sum of the products of various values of the payoff of the act and probabilities
(iii) Expected value of cost = Sum of the products of the various values of the cost andprobabilities
(iv) Expected value of loss of an act = Sum of the products of the losses of the acts andProbabilities
Decision-Making
Self-Instructional Material 147
NOTES
Example 4.13: RM (Mudland) Limited has recently installed new machinerybut has not yet decided on the appropriate number of a certain spare part requiredfor repairs.
Spare parts cost Rs 2,000 each but are only available if ordered now. If the plantfailed and there was no spare part available, the cost to the business of mending theplant rises to Rs 15,000. The plant has an estimated life of ten years and the probabilitydistribution of failure during this time, based on the experience with similar plant,is as follows.
Number of failure over ten-year period Probability0 0.11 0.42 0.33 0.14 0.15 and over Nil
(For the purpose of this question, ignore any discounting for time value of money)You are required to calculate:
(a) The expected number of failures in the ten-year periods(b) The optimal number of spares that should be purchased now: (Include
in your workings a table to show the cost of alternative ordering policiesand failures)
(c) The current cost of the ordering policy chosen;(d) The value of perfect information of the number of failures in the ten-
year life.Solution:(a) Expected number of failures
= ( ) ( ) ( ) ( ) ( ) 71104103302401100 ...... =×+×+×+×+×
(b) Pay-off values are calculated as follows:
Alternative strategies (Spare partspurchased) (in ‘000)State of nature
(Number offailure) A1
(0)A2
(1)A3
(2)A4
(3)A5
(4)E1(0)E2(1)E3(2)E4(3)E5(4)
015304560
22
173247
444
1934
6666
21
88888
Explanation for calculation of pay-off values: Calculation ofA1E1 = (0 × 2000) + ( 0) 15000 = 0A1E2 = (0 × 2000) + (1 – 0) 15000 = 15000A1E2 = (2 × 2000) + (0 × 15000) = 4000A3E4 = (2 × 2000) + (3 – 2) 15000 = 19000, etc.
Operations Research
148 Self-Instructional Material
NOTES
EMV for each A1, A2, A3, A4 and A5 are 25.5, 14, 8.5, 7.5, 8 respectively selectingthe act with least expected value, Optimal Act is A3.
∴ Purchase three parts(c) If three spare parts are purchased, the expected cost is a minimum and the
cost current of the ordering policy chosen = 3× 2,000 = Rs 6000.(d) If one had perfect information, in advance, of the number of failures that
would occur one would purchase the corresponding number of spare partsand expected cost under perfecet information (ECPI).
(0.1 0) (0.4 2000) (0.3 4000)(0.1 6000) (0.1 8000) Rs 3400
The expected value of perfect information = Optimal EMV – ECPI = Rs 7500– Rs 3,400 = 4,100.
Example 4.14: A food products compay is counterplanning the introduction ofa revolutionary new product with new packing to replace the existing product atmuch higher price (S1) or a moderate change in the composition of the existingproduct with a new packaging at a small increase in price (S2) or a small change inthe composition of the existing expect the word, ‘New with a negligible increase inthe price (S3). The three possible states of nature of events are: (i) high increase insales (N1). (ii) no change in sales (N2) and (iii) decrease is sales (N3). The marketingdepartment of the company worked out the pay-offs in terms of yearly new profitsfor each of the strategies or these events (expected sales). This is represented in thefollowing table.
Pay-offsState of nature
Strategies N1 N2 N3
S1 700 300 150 S2 500 450 0 S3 300 300 300
Which strategy should the executive concerned choose on the basis of:(a) Maximum criterion (b) Maximax criterion (c) Minimax regret criterion
(d) Laplace criterion?Solution: Writing the pay-off table properly
StrategiesState of nature
S1 S2 S3
N1
N2
N3
700300150
500450
0
300300300
(a) Maximin criterionMinimum pay offs
S1 150S2 0S3 300
Max of these Minima = 300The executive should choose strategy S3.
Decision-Making
Self-Instructional Material 149
NOTES
(b) Maximax criterionMaximum pay-offs
S1 700S2 500S3 300
Maximum of maxima = 700∴ The executive can choose Act S1.
(c) Minimax regret criterionOpportunity loss table
S1 S2 S3
N1
N2
N3
700 – 700 = 0450 – 300 = 150300 – 150 = 150
700 – 500 = 200450 – 450 = 0300 – 0 = 300
700 – 300 = 400450 – 300 = 150300 – 300 = 0
Maximum Opportunity lossfor S1 – 150for S2 – 300for S3 – 400
The executive should choose strategy S1 since it minimizes the maximum of thelosses.
(d) Laplace criterionAssigning equal probability (say 1/3) to each state of nature calculate the expected
monetary value for each act.Prob. × Pay-off:
3003
90030031300
31300
31
6663163
950031450
31500
31
333833
115015031300
31700
31
==⎟⎠⎞
⎜⎝⎛ ×+⎟
⎠⎞
⎜⎝⎛ ×+⎟
⎠⎞
⎜⎝⎛ ×
==⎟⎠⎞
⎜⎝⎛ ×+⎟
⎠⎞
⎜⎝⎛ ×+⎟
⎠⎞
⎜⎝⎛ ×
==⎟⎠⎞
⎜⎝⎛ ×+⎟
⎠⎞
⎜⎝⎛ ×+⎟
⎠⎞
⎜⎝⎛ ×
.
.
Since the EMV is highest for strategy 1, the executive should select strategy S1.Example 4.15: The research department of consumer products division has
recommended to the marketing department to launch a soap with three differentperfumes. The marketing manager has to decide the type of perfume to launchunder the following estimated pay-off for the various levels of sales.
Estimated level of sales (units)
Types of perfume 20,000 10,000 2,000I 250 15 10II 40 20 5III 60 25 3
Examine which type can be chosen under maximax, minimax, maximin andLaplace criterion.
Operations Research
150 Self-Instructional Material
NOTES
Solution: Rewriting the pay off table Types of perfume
Levels of Sales I II III
20000 250 40 60 10000 15 20 25 2000 10 5 3
(1) Maximax criterionMaximum for Type I = 250Maximum for Type II = 40Maximum for Type III = 60Maximum of maximum = 250∴ Select type I perfume
(2) Minimax criterionLoss table is
Type Sales
I II III 20,000 0 210 190 10,000 10 5 0 2,000 0 5 7
Maximum losses under I, II, III are respectively 10, 210, 190 (from the pay-offtable)
Minimum of these is 10.∴ Type I is preferred.(3) Maximin criterionMinimum pay-off under each type I, II, III are respectively 10, 5, 3 (from the
pay-off table)Maximum of these is 10.∴ Type I is preferred.
(4) Laplace criterionLet the probability for each levels of sales be taken as 1/3 each.Expected pay-offs are:
33.293
88313
3125
3160III Type
67.21365
315
3120
3140II Type
67.913
2753110
3115
31250I Type
==⎟⎠⎞
⎜⎝⎛ ×+⎟
⎠⎞
⎜⎝⎛ ×+⎟
⎠⎞
⎜⎝⎛ ×→
==⎟⎠⎞
⎜⎝⎛ ×+⎟
⎠⎞
⎜⎝⎛ ×+⎟
⎠⎞
⎜⎝⎛ ×→
==⎟⎠⎞
⎜⎝⎛ ×+⎟
⎠⎞
⎜⎝⎛ ×+⎟
⎠⎞
⎜⎝⎛ ×→
Maximum expected pay-off is for Type I, so choose Type I perfume.
Decision-Making
Self-Instructional Material 151
NOTES
Example 4.16: A company has an opportunity to computerize its recordsdepartment. However, the existing personnel have job security under unionagreement. The cost of the three alternative programmes for the changeover dependupon the attitude of the union and is estimated as follows:Attitude of union General retraining Selective retraining Hire new employeesAntagonist 940 920 900
Passive 810 800 820
Enthusiastic 700 710 860
(i) Construct the opportunity losses table.Solution: Opportunity loss for row I is obtained by subtracting 900
(least element of row I) from all the elements of I row are940 – 900 = 40, 920 – 900 = 20, 900 – 900 = 0
Similarly, for row II the opportunity losses are810 – 800 =10, 800 – 800 = 0 and 820 – 800 = 20
Similarly for row III, they are700–700 = 0, 710–700 = 10, 860-700 = 160
Thus, the opportunity loss table is
State of actionAttitude of the nation
General Selective Hire newAntagonistPassiveEnthusiastic
40100
200
10
020
160
4.3 DECISION TREE ANALYSIS
A ‘Decision tree’ is one of the tools used for the diagramatic presentation of thesequential and multidimensional aspects of a particular decision problem forsystematic analysis and evaluation. Here, the decision problem, the alternativecourses of action, the states of nature and the likely outcomes of alternatives arediagramatically or graphically depicted as branches and sub-branches of a horizontaltree.
The decision tree consists of nodes and branches. The nodes are of two types,decision nodes and chance nodes. Courses of action (or strategies) originate fromthe decision nodes as the main branches. At the terminal of each main branch thereare chance nodes. From these chance nodes, chance events emanate in the form ofsub-branches. The respective pay-offs and probabilities associated with alternativecourses and chances events are shown along the sub-branches. At the terminal ofthe sub-branches are shown the expected values of the outcome.
Operations Research
152 Self-Instructional Material
NOTES
Fig 4.1 Decision Tree
Here A1, A2, A3 and A4 are strategies E1, E2, E3 are events O11, O12, O21, O22, O31,O32 are outcomes.
A decision tree is highly useful to a decision-maker in multistage situationswhich involve a series of decisions each dependent on the preceding one. Workingbackward, from the future to the present, we are able to eliminate unprofitablebranches and determine optimum decisions. The decision tree analysis allows oneto understand, simply by inspection, various assumptions and alternatives in a graphicform which is much more easier to understand than the abstract analytical form.
The advantages of the decision tree structure are that complex managerialproblems and decisions of a chain-like nature can be systematically and explicitlydefined and evaluated.
Example 4.17: A firm owner is seriously considering to drill a farm well. Inthe past, only 70% of wells drilled were successful at 200 feet of depth in the area.Moreover, on finding no water at 200 ft., some persons drilled it further up to 250feet but only 20% struck water at 250ft. The prevailing cost of driling is Rs 50 perfoot. The farm owner has estimated that in case he does not get his own wells, hewill have to pay Rs 15,000 over the next 10 years, in present value (PV) term, tobuy water from the neighbour. The following decisions can be optimal.
(i) Do not drill any well (ii) drill up to 200ft and (iii) if no water is found at 200ft. drill further up to 250ft.
Draw an appropriate decision tree and determine the farm owner’s strategyunder EMV approach.
Solution:
Decision Tree for the Firm Owner
At D2 pointDecision : (a) drill upto 250feet (b) Do not drillEvent : (a) No water (b) Water
Decision-Making
Self-Instructional Material 153
NOTES
Probabilities are .2, .8EMV for drill up to 250 feet.
= (12500 × .2) + (27500 × .8) = 24500EMV for do not drill = 25000 (from the tree)EMV is smaller for the act drill upto 250 feet. So it is an optimal act.At D2 pointThe decisions are drill upto 200 feet and do not drill. Events are same as those
of D2 point.Probabilities are .7, .3EMV for drill up to 200 feet= (10000 × .7) + (24500 × .3) = 14350EMV for do not drill = 15,000 from the tree.The optimal decision is drill upto 200 feet (as the EMV is small).Therefore, combining D1and D2 the optimal strategy is to drill the well upto 200
feet and if no water is struck, then further drill it to 250 feet.Example 4.18: A firm is planning to develop and market a new drug. The
cost of extensive research to develop the drug has been estimated at Rs. 100000.The manager of the research programme has found that there is a 60% chance thatthe drug will be developed successfully. The market potential has been assessed asfollows:
Market condition Prob. Present value of profit (Rs)
Large Market potential 0.1 50,000
Moderate Market potential 0.6 25,000
Low Market potential 0.3 10,000
The present value figures do not include the cost of research. While the firm isconsidering this proposal, a second proposal almost similar comes up forconsideration. The second one also requires an investment of Rs 1,00,000 but thepresent value of all profits is Rs 12,000. The return on investment in the secondproposal is certain.
(i) Draw a decision tree indicating all events and choices of the firm.(ii) What decision should the firm take regarding the investment of
Rs 1,00,000?Solution:
Decision Tree for the Drug Firm
Operations Research
154 Self-Instructional Material
NOTES
At point D2
Decisions are (1) Enter market (2) Do not enter market.(a) Enter market
EMV = Expected P V= (50000 × .1) + (25000 × .6) + (10000 ×.3)= 5000 + 15000 + 3000 = 23000
(b) Do not enter marketEMV = Expected PV = 0 × 1 = 0Decision: Enter the market since EMV is more.
At point D1
Decisions are (1) Develop new drug (2) Accept proposal II
(a) Develop new drug
EMV = Expected PV = (23000 × .6) + (0 × .4) = (13800 + 0) =13800
(b) Accept Proposal II
EMV = Expected PV = 12000 × 1 = 12000
Using EMV criterion, the optimal decision at D1 is to develop and market thenew drug.
4.4 INTEGER PROGRAMMING AND DYNAMICPROGRAMMING: CONCEPTS ANDADVANTAGES
A linear programming problem in which all or some of the decision variables areconstrained to assume non-negative integer values is called an Integer ProgrammingProblem. (IPP).
In a linear programming problem if all variables are required to takeintegral values then it is called the pure (all) integer programming problem (PureIPP).
If only some of the variables in the optimal solution of an IPP are restricted toassume non-negative integer values while the remaining variables are free to takeany non-negative values, then it is called a mixed integer programming problem(Mixed IPP).
Further, if all the variables in the optimal solution are allowed to take values 0or 1, then the problem is called the 0–1 programming problem or standard discreteprogramming problem.
The general integer programming problem is given by Max Z = CX.Subject to the constraintsA x < bx > 0 and some or all variables are integers
Decision-Making
Self-Instructional Material 155
NOTES
4.4.1 Importance of Integer Programming Problems
In IPP, all the decision variables were allowed to take any non-negative real valuesas it is quite possible and appropriate to have fractional values in many situations.There are several frequently occurring circumstances in business and industry thatlead to planning models involving integer valued variables. For example, inproduction, manufacturing is frequently scheduled in terms of batches, lots or runs.In allocation of goods, a shipment must involve a discrete number of trucks andaircrafts. In such cases, the fractional values of variables like 13/3 may be meaninglessin the context of the actual decision problem.
This is the main reason why integer programming is so important for marginaldecisions.
4.4.2 Applications of Integer Programming
Integer programming is applied in business and industry. All assignment andtransportation problems are integer programming problems, as in assignment andtravelling salesmen problem, all the decision variables are either zero or one.
i.e., xij = 0 or 1Other examples are capital budgeting and production scheduling problems. In
fact, any situation involving decisions of the type ‘either to do a job or not to do’ canbe viewed as an IPP In all such situations
xij = 1 if the jth activity is performed,0 if the jth activity is not performed.In addition, allocation problems involving the allocation of men, and machines
give rise to IPP, since such communities can be assigned only in integers and not infractions.
Note: If the non-integer variable is rounded off, it violates the feasibility andalso there is no guarantee that the rounded off solution will be optimal. Due to thesedifficulties, there is a need for developing a systematic and efficient procedure forobtaining the exact optimal integer solution to such problems.
Many decision-making problems involve a process that takes place in such away that at each stage, the process is dependent on the strategy chosen. Such type ofproblems are called Dynamic Programming Problem (DPP). Thus, DPP is concernedwith the theory of multistage decision process. Mathematically, a DPP is a decisionmaking problem in n variables, the problem being subdivided into n sub-problems(segments), each sub-problem being a decision-making problem in one variableonly. The solution to a DPP is achieved sequentially, starting from one (initial)stage to the next, till the final stage is reached.
4.4.3 Decision Tree and Bellman’s Principle of Optimality
A multistage decision system in which each decision and state variable can takeonly finite number of values can be represented graphically by a decision tree.
Operations Research
156 Self-Instructional Material
NOTES
Stage 1
Stage 2
Stage 3
Circles represent nodes corresponding to stages and lines between circles denotearcs corresponding to decisions.
The dynamic programming technique deals with such situations by dividingthe given problem into sub-problems or stages.
Bellman’s principle of optimality states that ‘An optimal policy (set ofdecisions) has the property that whatever the initial state and decisions are, theremaining decisions must constitute an optimal policy with regard to the stateresulting from the first decision.’
The problem which does not satisfy the principle of optimality cannot be solvedby the dynamic programming method.
4.4.4 Characteristics of DPP
The basic features which characterize the dynamic programming problem are asfollows.
1. The problem can be divided into stages with a policy decision requiredat each stage.
2. Every stage consists of a number of states associated with it. The statesare the different possible conditions in which the system may find itselfat that stage of the problem.
3. Decision at each stage converts the current stage into the state associatedwith the next stage.
4. The state of the system at a stage is described by a set of variablescalled state variables.
5. When the current state is known, an optimal policy for the remainingstages is independent of the policy of the previous ones.
6. The solution procedure begins by finding the optimal policy for eachstate to the last stage.
7. A recursive relationship which identifies the optimal policy for eachstate with n stages remaining, given the optimal policy for each statewith (n[1) stages left.
8. Using recursive equation approach, each time the solution proceduremoves backward stage by stage for obtaining the optimum policy ofeach state for the particular stage, till it attains the optimum policybeginning at the initial stage.
Note: A stage may be defined as the portion of the problem that possesses a setof mutually exclusive alternatives from which the best alternative is to be selected.
Check Your Progress4. Mention any two methods
which are used duringdecision-making under risk.
5. Define a decision tree.6. State an advantage of
decision tree analysis.
Decision-Making
Self-Instructional Material 157
NOTES
4.4.5 Dynamic Programming Algorithm
The solution of a multistage problem by dynamic programming involves thefollowing steps.
Step 1: Identify the decision variables and specify the objective function to beoptimized under certain limitations, if any.
Step 2: Decompose the given problem into a number of smaller sub-problems.Identify the state variable at each stage.
Step 3: Write down the general recursive relationship for computing the optimalpolicy. Decide whether forward or backward method is to be followed to solve theproblem.
Step 4: Construct appropriate stages to show the required values of the returnfunction at each stage.
Step 5: Determine the overall optimal policy or decisions and its value at eachstage. There may be more than such optimal policy.
4.4.6 Advantages of DPP
1. Generally the solution of a recursive equation involves two types ofcomputations depending upon whether the system is continuous or discrete.In the first case, the optimal decision at each stage is obtained by using theusual classical method of optimization. In the second case, a tabularcomputational scheme is followed.
2. The D.P.P is solved by using the recursive equation, starting from the first tothe last stage i.e. obtaining the sequence f1 f2 f3 ...fn of the optimal solution.This computation is called the forward computational procedure. If therecursive equations are formulated to obtain a sequence fn fn–1 ...f1 then thecomputation is known as the backward computational procedure.
Example 4.19: Use the principle of optimality to find theMaximum value of Z = b1x1 + b2x2 + ... + bnxn
when x1 + x2 + x3 + ... + xn = Cx1x2 ... xn ≥ 0
Solution: The problem can be considered to divide the positive quantity C in nparts x1x2 ... xn so that b1x1 + b2x2 + ... + bnxn is maximum.Let fn (C) = b1x1 + b2x2 + ... + bnxn .
Recursive equation
If Zi be the ith part (i = 1, 2 ... n) of the quantity then the recursive equation of theproblem are
f1(x1) = max b1Z1 = b1x1
Z1 = x1
and fi(xi) = max [biZi + fi – 1 (xi – Zi)]0 ≤ Zi ≤ xi[where i = 1, 2 ... n]
Operations Research
158 Self-Instructional Material
NOTES
Solution of Recursive Equations
For one stage problem i = 1f1(x1) = b1x1
This gives, 1 1( )f C b C= which is trivially true.
For two stage problem, where i = 1, 2f2(x2) = max (b2Z2 + f1(x2 – Z2)
0 ≤ Z2 ≤ x2
f2(C) = max (b2Z2 + f1(C – Z)0 ≤ Z2 ≤ C
for x2 = C1, Z2 = Z= max b2Z + b1(C – Z)
0 ≤ Z ≤ C= max (b2 – b1) Z + b1C
0 ≤ Z ≤ CIf b2 – b1 is positive then this is maximum for Z = C, otherwise, it will be
minimum.
Thus 2 2( )f C b C=
Similarly for three stage problem ( i = 1, 2 3)f3(x3) = max [(b3Z3 + f2(x3 – Z3)]
0 ≤ Z3 ≤ x3
f3(C) = max [(b3Z + f2(C – Z)[0 ≤ Z3 ≤ C= max [b3Z + b2(C – Z)]
0 ≤ Z ≤ C= max (b3 – b2) C + b2C
0 ≤ Z ≤ CAgain if b3 – b2 is positive, then it gives maximum value for Z = C otherwise, it
gives the minimum value.
Thus 2 3( )f C b C=
From the results of the three stages 1, 2, 3 it can be easily shown by inductionmethod that
( )n nf C b C=
Hence, the optimal policy will be(0, 0, 0 ... xn = C) with f1(C) = bnC
Decision-Making
Self-Instructional Material 159
NOTES
Example 4.20: Use dynamic programming to show thatZ = p1 log p1 + p2 log p2 + .... + pn log pn
Subject to the constraints p1 + p2 + .... + pn = 1 and
Pj ≥ 0 (j = 1, 2 .. n) is minimum, where p1 = p2 =... pn = 1n
Solution. The problem here is to divide unity into n parts so as to minimize thequantity
Σpi log pi
Let fn(1) denote the minimum attainable sum ofp1 log p1 (i = 1, 2, .... n)
For n = 1 (Stage 1)f1(1) = min (p1 log p1) = 1 log 10 < x ≤ 1
as unity is divided only into p1 = 1 partFor n = 2, the unity is divided into two parts p1 and p2 such that p1 + p2 = 1If p1 = x, p2 = 1 – x, then
f2(1) = min (p1 log p1 + p2 log p2)0 ≤ x ≤ 1= min [x log x + (1 – x) log (1 – x)]0 ≤ x ≤ 1= min [x log x f1(1 – x)]0 ≤ x ≤ 1
In general for an n stage problem the recursive equation isfn(1) = min (p1 log p1 + p2 log p2+ .... + pn log pn)
0 ≤ x ≤ 1= min[x log x f1 (1 – x)]0 ≤ x ≤ 1
We solve this recursive equationFor n = 2, (Stage 2)The function x log x + (1 – x) log (1 – x) attains its minimum value at
x = 12
satisfying the condition 0 < x ≤ 1
f2(1) = 1 1 1 1 1 1log 1– log 1– 2 log2 2 2 2 2 2
⎛ ⎞ ⎛ ⎞ ⎛ ⎞+ =⎜ ⎟ ⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠ ⎝ ⎠
Similarly for stage 3, the minimum value of the recusive equation is obtained asf3(1) = min [x log x + f2 (1 – x)]
0 ≤ x ≤ 1
Operations Research
160 Self-Instructional Material
NOTES
= min [x log x + 21 1–log
2 2x x−⎛ ⎞ ⎛ ⎞
⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠
]
0 ≤ x ≤ 1
Now, since the minimum value of
x log x + 21 1–log
2 2x x−⎛ ⎞ ⎛ ⎞
⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠
is
attained at x = 13 satisfying 0 < x £ 1 we have
f3(1) = 1 1 1 1 1 1log 2 log 3 log3 3 3 3 3 3
⎛ ⎞ ⎛ ⎞+ =⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠
∴ optimal policy is P1 = P2 = P3 = 13
In general for n stage problem, we assume the optimal policy to be
P1 = P2 = ... Pn =1n and fn(1) = n
1 1logn n⎡ ⎤⎢ ⎥⎣ ⎦
This can be shown easily using mathematical induction.For n = m + 1, the recursive equation is
fm + 1 (1) = min [x log x + fm(1 – x)]0 ≤ x ≤ 1
= min1 1log logx xx x m
m m⎡ − − ⎤⎛ ⎞ ⎛ ⎞+ ⎜ ⎟ ⎜ ⎟⎢ ⎥⎝ ⎠ ⎝ ⎠⎣ ⎦
0 ≤ x ≤ 1
= 1 1 1 1log log
1 1 1 1m
m m m m⎛ ⎞+ ⎜ ⎟+ + + +⎝ ⎠
= m + 1 1 1log
1 1m m⎛ ⎞⎜ ⎟+ +⎝ ⎠
Since minimum x log 1 1logx xxx m− −
+ is attained at x = 1
1m +.
Hence, the required optimum policy is 1 1 1, ....n n n
⎛ ⎞⎜ ⎟⎝ ⎠
with 1 1(1) lognf nn n
⎛ ⎞= ⎜ ⎟⎝ ⎠
Decision-Making
Self-Instructional Material 161
NOTES
4.5 SUMMARY
In this unit, you have learned about decision-making and its applications. In anydecision problem, the decision-maker is concerned with choosing the method whichyields the best result from among the various alternative courses of action. The unithas explained that when the decision-maker faces multiple states of nature and hasno means to arrive at probability values of the likelihood of occurrence of thesestates of nature, the problem is a decision problem under uncertainty. You have alsolearned that during decision-making under risk, the decision-maker has someknowledge which enables him to assign probability to the occurrence of each stateof nature. You have also learned how to prepare a pay-off table. It has also describeddecision tree analysis, where you have learned that the decision problem, alternativecourses of action, states of nature and the likely outcomes of alternatives are depictedas if they are branches and sub-branches of a horizontal tree. The unit has alsoexplained the characteristics of integer programming and dynamic programming.
4.6 KEY TERMS
● Tactical decisions: Decisions which affect a business in the short run.● Strategic decisions: Decisions which have far-reaching effects on the course
of business.● Pay-off table: It represents the economics of a problem, i.e., revenue and
costs associated with any action with a particular outcome. It is an orderedstatement of profit or costs resulting under the given situation.
● Opportunity loss table: It is the loss incurred because of failure to take thebest possible action. Opportunity losses are calculated separately for eachstate of nature that might occur.
● Decision-making under certainty: In this case, the decision-maker knowswith certainty the consequences of every alternative or decision choice. Thedecision-maker presumes that only one state of nature is relevant for hispurpose.
● Decision-making under uncertainty:When the decision-maker faces multiplestates of nature but he has no means to arrive at probability values to thelikelihood of occurence of these states of nature, the problem is a decisionproblem under uncertainty.
● Decision tree: A diagrammatic presentation of the sequential andmultidimensional aspects of a particular decision problem for systematicanalysis and evaluation.
● Pure IPP: In a linear programming problem if all variables are required totake integral values then it is called the pure (all) integer programming problem(Pure IPP).
● Mixed IPP: If only some variables in the optimal solution of an IPP arerestricted to assume non-negative integer values while the remaining variablesare free to take any non-negative values, then it is called a mixed integerprogramming problem (Mixed IPP).
Operations Research
162 Self-Instructional Material
NOTES
4.7 ANSWERS TO ‘CHECK YOUR PROGRESS’
1. Tactical and strategic.2. Events are the occurrences which affect the achievement of objectives.
They are also called states of nature or outcomes.3. (i) Maximax criterion, (ii) Minimax criterion, (iii) Maximin criterion,
(iv) Laplace criterion and (v) Hurwicz alpha criterion4. Expected Monetary Value criterion and Expected Opportunity Loss
criterion.5. A decision tree is a diagrammatic presentation of the sequential and
multidimensional aspects of a particular decision problem for systematicanalysis and evaluation.
6. Complex managerial problems can be systematically and explicitlydefined and evaluated.
4.8 QUESTIONS AND EXERCISES
Short-Answer Questions
1. What do you understand by ‘decision theory’?2. What are EMV and EOL criteria?3. What are pay-off tables and regret tables?
Long-Answer Questions
1. Describe some methods which are useful for decision-making underuncertainty.
2. Explain the terms (i) Expected Monetary Value (ii) Expected Value ofPerfect Information.
3. Write notes on the Bayesian decision theory.4. Write a note on decision tree.5. Write a note on action space.6. Explain (a) Maximax (b) Minimax (c) Maximin decision criteria.7. From the following pay-off table, decide which is the optimal act.
StrategiesEvents A1 A2 A3
S1 20 40 60 S2 15 –10 –15 S3 35 25 –20P(S1) = .4 P(S2) = .5 P(S3) = .1
EMV for A1 = 19, for A2 = 13.5, for A3 = 13.5 So A1]
Decision-Making
Self-Instructional Material 163
NOTES
4.9 FURTHER READING
Bierman, Harold, Charles Bonini and W.H. Hausman. 1973. Quantitative Analysisfor Business Decisions. Homewood (Illinois): Richard D. Irwin.Gillett, Billy E. 2007. Introduction to Operations Research. New Delhi: TataMcGraw-Hill.Lindsay, Franklin A. 1958. New Techniques for Management Decision Making.New York: McGraw-Hill.Herbert, G. Hicks. 1967. The Management of Organisations. New York: McGraw-Hill.Massie, Joseph L. 1986. Essentials of Management. New Jersey: Prentice-Hall.Ackoff, R.L. and M.W. Sasieni. 1968. Fundamentals of Operations Research. NewYork: John Wiley & Sons Inc.Enrick, Norbert Lloyd. 1965. Management Operations Research. New York: Holt,Rinehart & Winston.