CHAPTER 5

Decision Analysis
Supplement Outline

SUPPLEMENT TO

Introduction and Decision Tables, 2 Decision Trees and Influence Diagrams, 4 Value of Information, 7 Sensitivity Analysis, 7 OM in Action: Oglethorpe Power, 9 Summary, 12

Key Terms, 12 Solved Problems, 12 Discussion and Review Questions, 14 Internet Exercises, 15 Problems, 15 Mini-Case: Southern Company, 21 Mini-Case: ICI Canada, 22

Learning Objectives After completing this supplement, you should be able to:

Model a single-stage decision problem as a decision table and use the expected value approach to solve it. LO2 Construct a decision tree and use it to analyze a multistage problem; describe an influence diagram. LO3 Calculate the expected value of perfect and imperfect (sample) information. LO4 Conduct sensitivity analysis on a simple decision problem.
LO1

PART THREE System Design

LO1

Introduction and Decision Tables

Decision analysis is a scientific and practical method for making important decisions. It was introduced in the 1960s (see http://decision.stanford.edu/library/ronald-a.-howard/ Decision%20Analysis-%20Applied%20Decision%20Theory.pdf). Decision analysis involves identification, clear representation, and formal assessment of important aspects of a decision and then determination of the best decision by applying the maximum expected value criterion. Decision analysis is suitable for a wide range of operations management decisions where uncertainty is present. Among them are capacity planning, product design, equipment selection, and location planning. Decisions that lend themselves to a decision analysis approach tend to be characterized by the following elements: 1. There is at least one random variable with a set of possible values (or future conditions or states of nature) that will have a bearing on the result of the decision. 2. There is a decision with a list of alternatives for the manager/decision maker to choose from. 3. There is a known payoff (or value) for each alternative under each possible future condition. To use this approach, a decision maker would employ the following process: 1. Determine the goal or objective, e.g., maximize expected profit or net present value, or minimize expect cost. 2. Develop a list of possible alternatives for the decision in order to achieve the goal. 3. Identify possible future conditions or states of nature for each random variable (e.g., demand will be low, medium, or high; the number of contracts awarded will be one, two, or three; the competitor will or will not introduce a new product) that will affect the goal. 4. Determine or estimate the payoff (or value) associated with each alternative for every possible future condition. 5. Estimate the likelihood of each possible future condition for each random variable. 6. Evaluate alternatives according to the goal or decision criterion, and select the best alternative.
payoff table Table showing the payoffs (values) for each alternative in every possible state of nature of a onerandom-variable problem.

The information for a decision can be summarized in a payoff table, which shows the payoffs (value) for each alternative under the various possible states of nature of a one random-variable problem. These tables are helpful in choosing among alternatives because they facilitate comparison of alternatives. Consider the following payoff table, which illustrates a capacity planning problem.
Possible Future Demand Alternatives Small facility Medium facility Large facility Low $10* 7 (4) Moderate $10 12 2 High $10 12 16

*Net present value in $ millions.

The payoffs are shown in the body of the table. In this instance, the payoffs are in terms of net present values, which represent equivalent current dollar values of estimated future income less costs. This is a convenient measure because it places all alternatives on a comparable basis. If a small facility is built, the payoff will be $10 million for allthree possible states of nature. For a medium facility, low demand will have net present value of $7 million, whereas both moderate and high demand will have net present values of

SUPPLEMENT TO CHAPTER 5 Decision Analysis

$12million. Alarge facility will have loss of $4 million if demand is low, net present value of $2million if demand is moderate, and net present value of $16 million if demand is high. The problem for the decision maker is to select one of the alternatives, taking the net present values into account. In order to do so, one needs to determine the probability of occurrence for each state of nature (unless one of the alternatives dominates the rest, e.g., if the payoff of the medium facility for low demand was $10 million, then the medium alternative would dominate the other two alternatives, and therefore is the best). Because the states of nature should be mutually exclusive and collectively exhaustive for a random variable, these probabilities must add to 1.00. These probabilities are usually subjective opinions of experts. The decision analyst interviewing an expert should be careful not to ask leading questions and should be aware of psychological biases such as overconfidence or intentional underestimation by marketing/sales managers (to make meeting these goals easier). Given the probabilities for each state of nature, a widely used approach is the expected value method. The expected value is calculated for each alternative, and the one with the highest expected value is selected (when more is better such as for NPV). The expected value is the sum of the payoffs multiplied by probabilities for an alternative.
Example S-1

Using the expected value method, identify the best alternative for the following payoff table given these probabilities: low=.30, moderate=.50, and high=.20.
Possible Future Demand Alternatives Small facility Medium facility Large facility Low $10* 7 (4) Moderate $10 12 2 High $10 12 16

*Net present value in $ millions.

Find the expected value of each alternative by multiplying the probability of occurrence for each state of nature by the payoff of the alternative for that state of nature and summing them: EVsmall=.30($10)+.50($10)+.20($10)=$10 EVmedium=.30($7)+.50($12)+.20($12)=$10.5 EVlarge=.30(4)+.50($2)+.20($16)=$3 Hence, choose the medium facility because it has the highest expected net present value.

Solution

The expected value method is most appropriate when a decision maker is neither risk averse nor risk seeking, but is risk neutral. Typically, well-established organizations with numerous decisions of this nature tend to use the expected-value method because it provides the long-run average payoff. That is, the expected-value amount (e.g., $10.5 million in Example S-1) is not an actual payoff but an expected or average amount that would be approximated if a large number of identical decisions were to be made. Hence, if a decision maker applies this method to a large number of similar decisions, the expected payoff for the total will approximate the sum of the individual expected payoffs. If the decision maker is not risk neural, then he tends to consider the spread or risk of payoffs, and even individual payoffs. For a risk-averse decision maker, the minimum payoff, if the medium facility is built, $7 million, may still be acceptable. If it is not, he may look for a

PART THREE System Design FIGURE 5S-1

natur e1
Choos
2 Possible second decisions Payoff 1

Format of a decision tree

oo s
1 eA

of State
State

of na

e A1

Ch

ture 2

Payoff 2

Choos

Initial decision

e A2

Payoff 3

Ch

oo

Choos

e A3

Payoff 4

se

o State
State

f nat

ure 1

Choos

e A4

Payoff 5

of na

ture 2

Payoff 6

Decision point

Chance event

new alternative (e.g., seek more information) or choose to build a small facility that has a higher minimum payoff ($10 million). An approach to deal with risk-averse decision makers is called the utility theory. More complex decision problems will have more than one decision variable (i.e., a multistage decision problem) and more than one random variable. In this case, the payoffs cannot be represented by a table. Instead, decision trees are used to graphically model the decision problem.

LO2

Decision Trees and Influence Diagrams

decision tree A graphical representation of the decision variables, random variables, probabilities, and payoffs.

A decision tree is a graphical representation of the decision variables, random variables and their probabilities, and the payoffs. The term gets its name from the treelike appearance of the diagram (see Figure 5S-1). Decision trees are particularly useful for analyzing situations that involve sequential or multistage decisions. For instance, a manager may initially decide to build a small facility but she has to allow for the possibility that demand may be higher than anticipated. In this case, the manager may plan to make a subsequent decision on whether to expand or build an additional facility. A decision tree is composed of a number of nodes that have branches emanating from them (see Figure 5S-1). Square nodes denote decision points, and circular nodes denote chance events. Read the tree from left to right. Branches leaving square nodes represent alternatives; branches leaving circular nodes represent chance events (i.e., the states of nature). After the tree has been drawn, it is analyzed from right to left; that is, starting with the last decision that might be made it is rolled back. For each decision, choose the alternative that will yield the greatest return (or the lowest cost). For each chance node, calculate the expected value of the payoffs of its states of nature. If chance events follow a decision, choose the alternative that has the highest expected value (or lowest expected cost).

Example S-2

A manager must decide on the size of a video arcade to construct. The manager has narrowed the choices to two: large or small. Information has been collected on payoffs, and the following decision tree has been constructed. Analyze the decision tree and determine

SUPPLEMENT TO CHAPTER 5 Decision Analysis

which initial alternative (build small or build large) should be chosen in order to maximize expected value.
a dem nd (.4)
$40

Low

uil

m ds

all

High

dem

and

(.6)

ot on

hin

$40

Exp

and
oth ing

$55 ($10)

n Do

Bu

ild

lar

ge

Lo

em wd

and

(.4)

Red

uce

pric

es

$50

High

dem

and

(.6)

$70

The dollar amounts at the branch ends indicate the estimated payoffs if the sequence of decisions and chance events occurs. For example, if the initial decision is to build a small facility and it turns out that demand is low, the payoff will be $40 (thousand). Similarly, if a small facility is built, and demand turns out high, and a later decision is made to expand, the payoff will be $55. The figures in parentheses on branches leaving the chance nodes indicate the probabilities of those states of nature. Hence, the probability of low demand is .4, and the probability of high demand is .6. Payoffs in parentheses indicate losses. Analyze the decisions from right to left: 1. Determine which alternative would be selected for each possible second decision. For a small facility with high demand, there are two choices: do nothing, or expand. Because expand has higher payoff, you would choose it. Indicate this by placing a double slash through do nothing alternative. Similarly, for a large facility with low demand, there are two choices: do nothing or reduce prices. You would choose reduce prices because it has the higher expected value, so a double slash is placed on the other branch. 2. Determine the product of the chance probabilities and their respective payoffs for the remaining branches: Build small Low demand .4($40)=$16 High demand .6($55)=33 Build large Low demand .4($50)=20 High demand .6($70)=42 3. Determine the expected value of each initial alternative: Build small $16+$33=$49 Build large $20+$42=$62 Hence, the choice should be to build the large facility because it has a larger expected value than the small facility.

Solution

PART THREE System Design

The above problem was entered in the decision analysis software DPL, from Syncopation Software (www.syncopationsoftware.com). The solution is shown below:
Low Small Build [62.000] Demand [62.000] Demand [49.000] 40% High 60% React to low demand [50.000] [40.000] 40.000 React to Demand [55.000] Do nothing Expand Do nothing Reduce prices [70.000] 70.000 [40.000] 40.000 [55.000] 55.000 [10.000] 10.000 [50.000] 50.000 High 60%

Low 40%

Large

Briefly, in the DPLs window, the top pane is for the influence diagram (described next), and the bottom pane is for the decision tree (see the snapshot below). The decision tree was drawn by first adding decision nodes (by clicking on the square icon at the top and moving it into the top pane) and discrete chance nodes (by clicking on the green circle icon at the top and moving it into the top pane), and then moving them to the bottom pane in order to draw the decision tree. Each node is then double-clicked and its properties defined (name, alternatives/outcomes, probabilities, nature of nodes branches: symmetric vs. asymmetric). Next, each payoff is specified by double-clicking each branch. Finally, the expected value is computed by clicking on Run (at the top) and then Decision Analysis.

React to Demand Build Demand React to low demand

Demand

Low 40 Build Small High Expand 55 React to Demand Do nothing 40

Large Demand Low Do nothing 10 Reduce prices 50 High 70

SUPPLEMENT TO CHAPTER 5 Decision Analysis

Influence Diagrams
Influence diagrams (see example below) can graphically represent complex decision problems that have many random variables (events) and one or more decision variables. Influence diagrams are more concise than decision trees because they do not show the alternatives branches coming out of the decision nodes and the states of nature branches coming out of the chance nodes. Constructing and validating an influence diagram improves communication and consensus building at the beginning of the decision modelling process. The following is an example of the influence diagram representing the decision of whether or not to introduce a new product. The green circles show the random variables (chance events) and the rounded yellow squares show the payoff or part of it. This influence diagram for a new product decision also involves a pricing decision. The key uncertainties (i.e., crucial random variables) are units sold, which are affected by the pricing decision, fixed cost, and variable cost. Profit is the ultimate payoff, which is influenced by the total cost and revenue. Decision analysis software such as DPL is able to measure the effect of variability of each random variable on the payoff (profit) and identify those crucial random variables that have the greatest influence on profit. The result is plotted in a vertical bar chart called a Tornado diagram, named as such because the more crucial random variables, those that affect the payoff most and hence have a longer bar, are drawn on the top, making the chart look like a tornado (see e.g., http://en.wikipedia.org/wiki/Tornado_diagram).
Units Sold Fixed Cost

Price?

Variable Cost Revenue Total Cost

Introduce Product? Profit

Source: K. Chelst, Cant See the Forest Because of the Decision Trees: A Critique of Decision Analysis in Survey Texts, Interfaces (28)2, MarchApril 1998, pp. 8098.

LO3

Value of Information

In certain situations, it is possible to know with more certainty which state of nature of the crucial random variables will actually occur in the future. For instance, the choice of location for a restaurant may weigh heavily on whether a new highway will be constructed nearby or whether a zoning permit will be issued. A decision maker may have probabilities for these random variables; however, it may be possible to delay a decision until it is more clear which state of nature will occur. This might involve taking an option to buy the land. If the state of nature is favourable, the option can be exercised; if it is unfavourable, the option can be allowed to expire. The question to consider is whether the cost of the option will be less than the expected gain due to delaying the decision. Other possible ways of obtaining information about a random variable depend somewhat on the nature of the random variable. Information about consumer preferences might come from market research, additional information could come from product testing, or legal experts might be called on. If the information is perfect (i.e., it is 100 percent accurate and will pinpoint the value of the random variable), then the expected gain in payoff is called the expected value of perfect information, or EVPI.

PART THREE System Design

expected value of perfect information (EVPI) The difference between the expected payoff with perfect information and the expected payoff without the information.

Expected value of perfect information (EVPI) is the difference between the expected payoff with perfect information (under certainty) and the expected payoff without the information (under risk).

Expected value of Expected payoff Expected payoff = perfect information under certainty under risk

(5S-1)

Example S-3

Using the information from Example S-1, determine the expected value of perfect information using Formula 5S-1. First, calculate the expected payoff under certainty. To do this, identify the best payoff under each state of nature. Then combine these by weighting each payoff by the probability of that state of nature and adding the amounts. Thus, the best payoff under low demand is $10, the best under moderate demand is $12, and the best under high demand is $16. The expected payoff under certainty is, then: .30($10)+.50($12)+.20($16)=$12.2
Possible Future Demand Alternatives Small facility Medium facility Large facility Low $10* 7 (4) Moderate $10 12 2 High $10 12 16

Solution

*Net present value in $ millions.

The expected payoff under risk, as calculated in Example S-1, is $10.5. The EVPI is the difference between these: EVPI=$12.2$10.5=$1.7 This figure indicates the upper limit on the amount the decision maker should be willing to spend to obtain information in this case. Thus, if the cost exceeds this amount, the decision maker would be better off not spending additional money and simply going with the alternative that has the highest expected payoff. In most cases, the information that can be obtained about random variables is useful but not perfect. In these cases, a statistical result called Bayes Rule can be used to update the (prior) probability distribution of the random variable. The difference between the expected payoff with imperfect (sample) information and the expected payoff without sample information is called the expected value of sample information (EVSI). exhaustive states of nature. Let P(B1) and P(B2) be the prior probabilities of the states of nature, and P(A | B1) and P(A | B2) be the likelihood of observing A given B1 and B2, respectively (i.e., the conditional probabilities). Then, the posterior probabilities P(B1 | A) and P(B2 | A) are: P ( B 1 -|| A ) = P ( A | B 1) P ( B 1) P ( B 2 -|| A ) = 1 P ( B 1 -|| A )
Bayes Rule: Let A be an information event, and let B1 and B2 be mutually exclusive and

expected value of sample information (EVSI) The difference between the expected payoff with sample (imperfect) information and the expected payoff without sample information.

) ( P ( A | B 1) P ( B 1) + P ( A | B 2 ) P ( B 2 ))

Note: P(A)=P(A | B1) P(B1)+P(A | B2) P(B2).

SUPPLEMENT TO CHAPTER 5 Decision Analysis

A manager is considering random drug testing of his employees. However, the test is not 100 percent accurate. The conditional probabilities are as follows: If a person uses drugs, then the test will be positive 93 percent of the time, whereas if he is not, the test will be negative 97 percent of the time. Suppose that 5 percent of the employees are drug users. a. What are the posterior probabilities? b. If the cost of not identifying (and dismissing) a drug user is $1,000, cost of falsely accusing a non-user is $200, and other costs are zero, what is the maximum the manager should be willing to pay for a drug test, i.e., EVSI? a. Consider a random employee. Let B1=the event that he is a drug user, B2=the event that he is not a drug user. We have P(B1)=.05 and hence P(B2)=1 .05=.95. Let A + = the event that test result is positive. We have P(A+| B1)=.93 and P(A+| B2) = 1 .97=.03. Then, P ( B 2 -|| A + ) = 1 .62 = .38. P ( B 1 -|| A + ) = ( .93 ( .05 ) ) ( .93 ( .05 ) + .03 ( .95 ) ) . = .62 and

Example S-4

Solution

Also, P(A+)=P(A+| B1) P(B1)+P(A+| B2) P(B2)=.93(.05)+.03(.95)=.075, And P(A)=1 P(A+)=1 .075=.925 Note the relatively low (62 percent) probability that a drug user will be identified and a relatively high (38 percent) probability that a non-drug user will be implicated. Similarly (A | B1)=1 .93=.07 and P(A | B2)=.97, and P ( B 1 -|| A ) = ( .07 (.05 ) ) ( .07 (.05 ) + .97 (.95 ) ) . = .0038 and

P ( B 2 -|| A ) = 1 .0038 = .9962.

b. The answer is represented in the following decision tree. Note that negative payoffs are costs.
Positive Yes Test [9.215] Test result [9.215] 7.50% Employee 2 [3.800] Employee 1 [76.000] User 62.00% Non-user 38.00% User 0.38% Non-user 99.62% No Employee 3 [50.000] User 5.00% Non-user 95.00% [1000.000] 1000.000 [0.000] 0.000 Negative 92.50% [0.000] 0.000 [200.000] 200.000 [1000.000] 1000.000 [0.000] 0.000

The maximum that the manager should pay for a drug test (i.e., EVSI) is $50-$9.215=$40.785.

LO4

Sensitivity Analysis

Generally speaking, both the payoffs and the probabilities in a decision problem are estimated values. Consequently, it can be useful for the decision maker to have some indication

10

PART THREE System Design

sensitivity analysis Determining the range of probability for which an alternative has the best expected payoff.

of how sensitive the choice of an alternative is to changes in one or more of these values. Unfortunately, it is impossible to consider all possible combinations of every variable in a typical problem. Nevertheless, there are certain things a decision maker can do to judge the sensitivity of the chosen alternative to probability estimates. Sensitivity analysis provides a range of probability over which an alternative has the best expected payoff. The approach illustrated here is useful when there are two states of nature. It involves constructing a graph and then using algebra to determine a range of probabilities for which a given alternative is best. In effect, the graph provides a visual indication of the range of probability over which the various alternatives are optimal, and the algebra provides exact values of the endpoints of the ranges. Example S-5 illustrates the procedure. Given the following (revenue) payoff table, determine the range of probability for state of nature #2, that is, P2, for which each alternative is optimal using the expected value method.
State of Nature #1 A Alternative B C 4 16 12 #2 12 2 8

Example S-5

Solution

First, plot each alternative relative to P2. To do this, plot the #1 payoff on the left side of the graph and the #2 payoff on the right side. For instance, for alternative A, plot 4 on the left side of the graph and 12 on the right side. Then connect these two points with a straight line. The three alternatives are plotted on the graph as shown below. The graph shows the range of values of P2 over which each alternative is optimal. Thus, for low values of P2 [and thus high values of P1, since P1+P2=1.0], alternative B will have the highest expected value; for intermediate values of P2, alternative C is best; and for higher values of P2, alternative A is best. To find exact values of the ranges, determine where the upper parts of the lines intersect. Note that at the intersections, the two alternatives represented by the lines would be equivalent in terms of expected value. Hence, the decision maker would be indifferent between the two at that point. To determine the intersections, you must obtain the equation of each line. This is relatively simple to do. Because these are straight lines, they have the form y=a+bx, where a is the y-intercept value at the left axis, b is the slope of the line, and x is P2. Slope is defined as the change in y for a one-unit change in x. In this type of problem, the distance between the two vertical axes is 1.0. Consequently, the slope of each line is equal to the right-hand value minus the left-hand value. The slopes and equations are:
16 14 12 10 #1 Payoff 8 6 4 2 0 B best .2 .4 P (2) C best .6 A best .8 1.0 C B A 16 14 12 10 #2 8 Payoff 6 4 2

SUPPLEMENT TO CHAPTER 5 Decision Analysis

#1 A B C 4 16 12 #2 12 2 8 Slope 124=+8 216=14 812=4 Equation 4+8 P 2 1614 P 2 124 P 2

11

From the graph, we can see that alternative B is best from the point P2=0 to the point where that straight line intersects the straight line of alternative C, and that begins the region where C is better. To find that point, solve for the value of P2 at their intersection. This requires setting the two equations equal to each other and solving for P2. Thus, 1614 P2=124 P2 Rearranging terms yields 4=10 P2 Solving yields P2=.40. Thus, alternative B is best from P2=0 up to P2=.40. B and C are equivalent at P2=.40. Alternative C is best from that point until its line intersects alternative As line. To find that intersection, set those two equations equal and solve for P2. Thus, 4+8 P2=124 P2 Rearranging terms results in 12 P2=8 Solving yields P2=.67. Thus, alternative C is best from P2>. 40 up to P2=.67, where A and C are equivalent. For values of P2 greater than .67 up to P2=1.0, A is best. Note: If a problem calls for ranges with respect to P1, you could find the P2 ranges as above, and then subtract each P2 from 1.00 (e.g., .40 becomes .60, and .67 becomes .33).

OM in Action

Oglethorpe Power

glethorpe Power is an electric generation and distribution cooperative supplying 20 percent of the electricity used in Georgia. Most of the remaining demand for electricity in Georgia is supplied by Georgia Power Co. Oglethorpe and Georgia Power jointly own most of the power transmission lines in Georgia, and supply their excess electricity to Florida. Some years ago, Florida Power and Light (FPL) indicated to Oglethorpe that they were interested in constructing another major transmission line between Florida and Georgia. Oglethorpe invited the consulting company, Applied Decision Analysis, Inc., to assist it in making a decision. A team was formed to investigate the decision variables, random variables, and their payoffs. Brainstorming resulted in an influence diagram and identification of the three decision variables: line (joint with Georgia Power, alone, or no line), nature of control (Oglethorpe, or FPL), and whether to upgrade the associated facilities (joint with Georgia Power, alone, or no upgrade). Five random variables were identified: construction cost (low, medium, or high), competitive situation in Florida (good, fair, or bad), Florida demand (low, medium, or high), Oglethorpes

share of demand (very low, low, medium, high, or very high), and future spot price for electricity (low, medium, or high). The payoffs were measured in terms of net present value. The team estimated the probability of random events (the states of nature for each random variable) and estimated the payoffs for each combination of decision alternatives and states of nature. Using the DPL software, the alternative (in brackets) for each decision variable with highest expected value was: line (alone), control (Oglethorpe), and upgrade (no). Then the risk profile (the distribution of NPVTornado diagram) of this solution was determined by DPL. Because this showed considerable possible negative NPV, the team identified the random variable that most affected the downside risk: the competitive situation in Florida. The next step was to collect further information on this random variable in order to reduce the range of values for the payoff. The team reported all its findings to top management, which started negotiating with FPL. Applied Decision Analysis, Inc. is now part of PricewaterhouseCoopers.
Source: A. Borison, Oglethorpe Power Corporation Decides about Investing in a Major Transmission System, Interfaces (25)2, March April 1995, pp. 2536.

12

PART THREE System Design

Summary

Decision analysis is a formal approach to decision making that is useful in operations management. Decision analysis provides a framework for the analysis of decisions. It involves identifying the decision alternatives, chance events (random variables), and payoffs. A simple decision problem can be represented as a decision table. The common selection method is the expected value method. Two visual tools useful for representing decision problems are decision trees and influence diagrams. Information can be used to reduce the variability of the important random variables. Two tools that can be used in deciding whether or not to buy information are expected value of perfect information and expected value of sample information. Graphical sensitivity analysis can be used to determine the effect of variability of probabilities assigned to states of nature on the best alternative. decision tree, 4 expected value of perfect information (EVPI), 8 expected value of sample information (EVSI), 8 payoff table, 2 sensitivity analysis, 10

Key Terms

Xtags error: Style name ambiguous: tag @...:Solved

Solved Problems

The following solved problems refer to this payoff (profits) table:

New Bridge Built Alternative capacity for new store A B C 1 2 4 No New Bridge 14 10 6

where A=small, B=medium, and C=large.

Problem 1 Solution

Using graphical sensitivity analysis, determine the probability for No New Bridge state of nature for which each alternative would be optimal. Plot a straight line for each alternative. Do this by plotting the payoff for new bridge built on the left axis and the payoff for no new bridge on the right axis and then connecting the two points. Each line represents the expected profit for an alternative for the entire range of probability of no new bridge. Because the lines represent expected profit, the line that is highest for a given value of P(no new bridge) is optimal. Thus, from the graph, you can see that for low values of this probability, alternative C is best, and for higher values, alternative A is best (B is never the highest line, so it is never optimal, i.e., it is dominated).
Payoff if new bridge A 14 Payoff if no new bridge 10

C 4 2 1 0

C best .27

A best 1.0 P (no new bridge)

SUPPLEMENT TO CHAPTER 5 Decision Analysis

13

The dividing line between the ranges where C and A are optimal occurs where the two lines intersect. To find that probability, first formulate the equation for each line. To do this, let the intersection with the left axis be the y intercept; the slope equals the right-side payoff minus the left-side payoff. Thus, for C you have 4+(64)P, which is 4+2P . For A, 1+(141)P, which is 1+13P . Setting these two equal to each other, you can solve for P: 4+2P=1+13P Solving, P=.27. Therefore, the ranges for P(no new bridge) for each alternative to be best are: A: .27<P1.00 B: never optimal C: 0<P.27 Using the probabilities of .60 for a new bridge and .40 for no new bridge, calculate the expected value of each alternative, and identify the alternative that would be selected under the expected value approach. A: .60(1)+.40(14)=6.20 [best] B: .60(2)+.40(10)=5.20 C: .60(4)+.40(6)=4.80 Calculate the EVPI using the information from the previous problem. Using Formula 5S-1, the EVPI is the expected payoff under certainty minus the maximum expected value. The expected payoff under certainty involves multiplying the best payoff in each column by the column probability and then summing those amounts. The best payoff in the first column is 4, and the best in the second is 14. Thus, Expected payoff under certainty=.60(4)+.40(14)=8.00 Then EVPI=8.006.20=1.80 Suppose that the values in the payoff table represent costs instead of profits. a. Using sensitivity analysis, determine the range of P(no new bridge) for which each alternative would be optimal. b. If P(new bridge)=.60 and P(no new bridge)=.40, find the alternative chosen to minimize expected cost. a. The graph is identical to that shown in Solved Problem 1. However, the lines now represent expected costs, so the best alternative for a given value of P(no new bridge) is the lowest line. Hence, for very low values of P(no new bridge), A is best; for intermediate values, B is best; and for high values, C is best. You can set the equations of A and B, and B and C, equal to each other in order to determine the values of P(no new bridge) at their intersections. Thus, A=B: 1+13P=2+8P; solving, P=.20 B=C: 2+8P=4+2P; solving, P=.33 Hence, the ranges are: A best: 0P<.20 B best: .20<P.33 C best: .33<P1.00

Problem 2

Solution

Problem 3 Solution

Problem 4

Solution

14

PART THREE System Design

b. Expected value calculations are the same whether the values represent costs or profits. Hence, the expected payoffs for costs are the same as the expected payoffs for profits that were calculated in Solved Problem 1. However, now you want the alternative that has the lowest expected payoff rather than the one with the highest payoff. Consequently, alternative C is the best because its expected payoff is the lowest of the three.
Cost with new bridge A 14 Cost with no new bridge 10

C 4 2 1 0 A best B best .20

C best 1.0

.33 P (no new bridge)

Problem 5

A diagnostic test of a certain disease has 95 percent sensitivity (i.e., probability that the test will be positive for a patient with the disease) and 96 percent specificity (i.e., probability that the test will be negative for a healthy person). Only 1 percent of the population has the disease in question. If the diagnostic test reports that a person chosen at random from the population tests positive, what is the conditional probability that the person does, in fact, have the disease? P(disease|+test) =
( P( +test | | disease ) P(disease )) / ( P( +test | | disease ) P(disease ) + P( +test | | no disease ) P( no disease ) .95(.01) = 0.1935 .95(.01) + (1 .96)(1 .01)

Solution

or 19.35 percent, not a large probability.

Discussion and Review Questions

1. List the steps in the decision analysis process. (LO1) 2. What information is contained in a payoff table? (LO1) 3. Under what circumstances is expected value method appropriate? When isnt it appropriate? (LO1) 4. What information does a decision maker need in order to perform an expected value analysis of a problem? What options are available to the decision maker if the probabilities of the states of nature are unknown? Can you think of a way you might use sensitivity analysis in such a case? (LO1) 5. Suppose a manager is using the expected-value method as a basis for making a capacity decision and, in the process, obtains a result in which there is a virtual tie between two of the seven alternatives. How is the manager to make a decision? (LO1) 6. What is the difference between a decision tree and an influence diagram? When should each be used? (LO2) 7. Will expected value of perfect information be always greater than or equal to expected value of sample information? Briefly explain. (LO3) 8. What is sensitivity analysis, and how can it be useful to a decision maker? (LO4)

SUPPLEMENT TO CHAPTER 5 Decision Analysis

15

1. View the video http://www.youtube.com/chevron#p/u/12/JRCxZA6ay3M and describe how decision analysis has helped Chevron. (LO14) 2. Visit http://gunston.gmu.edu/healthscience/730/IntroductionToDecisionAnalysis. asp?E=0#Organization_of_the_Book, browse through the Introduction to Decision Analysis section, find out why some decisions are hard to make, and list them. (LO14) 3. Visit http://www.treeage.com/resources/caseStudies.html, choose a case study, and summarize how decision analysis is being used. (LO14) 4. Visit http://www.palisade.com/cases/bucknell.asp?caseNav=byIndustry, choose a case study, and summarize how decision analysis is being used. (LO14) 5. Visit http://www.syncopation.com/casestudies.html, choose a case study, and summarize how decision analysis is being used. (LO14) 6. Visit http://www.lionhrtpub.com/orms/surveys/das/das.html and find out the educational price of DPL. (LO14)

Internet Exercises

1. A small building contractor has recently experienced two successive years in which work opportunities exceeded the firms capacity. The contractor must now make a decision on capacity for next year. Estimated profits under each of the two possible states of nature for next years demand are as shown in the table below. (LO13)
Next Years Demand Alternative Do nothing Expand Subcontractor *Profit in $ thousands. Low $50* 20 40 High $60 80 70

Problems

Suppose after a certain amount of discussion, the contractor is able to subjectively assess the probabilities of low and high demand: P(low)=.3 and P(high)=.7. a. Determine the expected profit of each alternative. Which alternative is best? b. Analyze the problem using a decision tree. Show the expected profit of each alternative on the tree. c. Calculate the expected value of perfect information. How could the contractor use this knowledge? 2. Refer to Problem 1. Construct a graph that will enable you to perform sensitivity analysis on the problem. Over what range of P(high) would each of the alternative be best? (LO4) 3. A company that plans to expand its product line must decide whether to build a small or a large facility to produce the new products. If it builds a small facility and demand is low, the net present value after deducting for building costs will be $400,000; If demand is high, the company can either maintain the small facility or expand it. Expansion would have a net present value of $450,000, and maintaining the small facility would have a net present value of $350,000. (LO24) If a large facility is built and demand is high, the estimated net present value is $800,000; If demand turns out to be low, the net present value will be $10,000. The probability that demand will be high is estimated to be .60, and the probability of low demand is estimated to be .40. a. Analyze this decision problem using a decision tree. b. Calculate the EVPI. How could this information be used? c. Determine the range of P(demand low) over which each alternative would be better.

16

PART THREE System Design

4. Determine the course of action that has the highest expected payoff for the decision tree below. (LO2)
$1.0*
d an (.4 )

Sm

al

m de

Do n

othin

$1.3

M Larg e

md ediu

ema

nd (.

5)

Expa

nd

$1.3
g

dem

and

(.1)

Do n

othin

$1.5 $1.6 $1.8

Expand

Su bc on tra ct

Build

Do n

othin

$0.7 $1.5 $1.0 $1.6

Expand

Sm

all d Medium demand (.5)

ed ema

n ema

4) d (.

Other use #1
#2

Othe

r use

Larg

nd

(.1)

Do n

othin

$1.6 $1.5 $1.7

Subcontract

Build

* Net present value in millions

5. The lease of an amusement park is about to expire. Management must decide whether to renew the lease for another 10 years or to relocate near the site of a proposed resort. The town planning board is currently debating the merits of granting approval to the resort. A consultant has estimated the net present value of amusement parks two alternatives under each state of nature as shown below. (LO13)
Options Renew (current location) Relocate near resort Resort Approved $3,000,000 5,000,000 Resort Rejected $4,000,000 2,000,000

Suppose that the management of the amusement park has decided that there is a 75% probability that the resorts application will be approved. a. If the management uses the expected value method, which alternative should it choose?

Bu ild
Do n
Sm Med ium dem and (.5) La rg ed em an d (.1 ) all d nd ema (.4)
Othe

othin

($0.9) $1.4 $1.0 $1.0 $1.1 $0.9 $2.4

Other use #1
#2

r use

Do n
Othe

othin

Other use #1
#2

r use

SUPPLEMENT TO CHAPTER 5 Decision Analysis

17

b. Represent this problem in the form of a decision tree and solve. c. If the management has been offered the option of a temporary lease while the town planning board considers the resorts application, would you advise the management to sign the lease? The temporary lease will cost $200,000. 6. Construct a graph that can be used for sensitivity analysis of the preceding problem. (LO1&4) a. How sensitive is the solution to the problem in terms of the probability estimate of.75? b. Suppose that, after consulting with a member of the town planning board, management decides that the estimate of approval should be .60. How sensitive is the solution to this revised estimate? Explain. c. Suppose that management is confident of all the estimated payoffs except for $5 million. If the probability of approval is .60, for what range of payoff for relocate/approved will the alternative selected remain the same? 7. A company must decide whether to construct a small, medium, or large plant. A consultants report indicates a .20 probability that demand will be low and a .80 probability that demand will be high. If the company builds a small facility and demand turns out to be low, the net present value will be $42 million; if demand turns out to be high, the company can either subcontract some production and realize the net present value of $43 million or expand for a net present value of $48 million. The company could build a medium-size facility: if demand turns out to be low, its net present value is estimated at $22 million; if demand turns out to be high, the company could do nothing and realize a net present value of $46 million, or it could expand and realize a net present value of $50 million. If the company builds a large facility and demand is low, the net present value will be $20 million, whereas high demand will result in a net present value of $72 million. (LO24) a. Analyze this problem using a decision tree. What is the best alternative? b. Calculate the EVPI and interpret it. c. Perform sensitivity analysis on P(high). 8. A manager must decide how many machines of a certain type to buy. The machines will be used to manufacture a new gear for which there is increased demand. The manager has narrowed the decision to two alternatives: buy one machine or buy two. If only one machine is purchased and demand is more than it can handle, a second machine can be purchased at a later time. However, the cost per machine would be lower if the two machines were purchased at the same time. The estimated probability of low demand is .30, and the estimated probability of high demand is .70. The net present value associated with the purchase of two machines initially is $75,000 if demand is low and $130,000 if demand is high. The net present value for one machine and low demand is $90,000; if demand is high, there are three options: one option is to do nothing, which would have a net present value of $90,000; a second option is to subcontract some production, which would have a net present value of $110,000; the third option is to purchase a second machine, which would have a net present value of $100,000. How many machines should the manager purchase initially? Use a decision tree to analyze this problem. (LO2) 9. Determine the course of action that has the highest expected value for the following decision tree. (LO2)

18

PART THREE System Design

1/3 1/3 1/3

0 60 90 40

.30 .50 .20

1/3 1/3 1/3

Alte

tiv rna

eA

44 60

(45) 45 99 40

10. The director of social services of a province has learned that a new act has mandated additional information requirements. This will place additional burden on the agency. The director has identified three alternatives to handle the increased workload. One alternative is to reassign present staff members, the second is to hire and train two new workers, and the third is to redesign current practice so that workers can readily collect the information with little additional effort. An unknown factor is the caseload for the coming year when the new data will be collected on a trial basis. The estimated costs for various options and caseloads are shown in the following table: (LO13)
Caseload Moderate Reassign staff New staff Redesign collection *Cost in $ thousands. $50* 60 40 High 60 60 50 Very High 85 60 90

The director of social services has decided that reasonable caseload probabilities are .10 for moderate, .30 for high, and .60 for very high. a. Which alternative will yield minimum expected cost? b. Construct a decision tree for this problem and solve it. c. Determine the expected value of perfect information. d. Suppose the director of social services has the option of hiring an additional staff member if one staff member is hired initially and the caseload turns out to be high or very high. Under that plan, the cost if demand is moderate will be 40, the cost if demand is high will be 70, and the cost if demand is very high will be 70. Construct a decision tree that shows this additional alternative and determine which alternative will minimize expected cost.

Al ter na e tiv B

.30 .50

50 30
1/2 1/ 2

.20

40 50

SUPPLEMENT TO CHAPTER 5 Decision Analysis

19

11. A manager has compiled estimated profits for various capacity alternatives but is reluctant to assign probabilities to the states of nature. The payoff table is: (LO4)
State of Nature #1 A Alternative *In $ thousands. B C $20* 120 100 #2 140 80 40

a. Plot the expected-value lines on a graph. b. Is there any alternative that would never be appropriate in terms of maximizing expected profit? Explain on the basis of your graph. c. For what range of P(#2) would alternative A be the best choice if the goal is to maximize expected profit? d. For what range of P(#1) would alternative A be the best choice if the goal is to maximize expected profit? 12. Repeat all parts of Problem 11, assuming the values in the payoff table are estimated costs and the goal is to minimize expected costs. (LO4) 13. Consider the following payoff table of estimated NPV for four alternatives: (LO4)
State of Nature 1 #1 Alternative #2 #3 #4 *In $ thousands. $10* 8 5 0 State of Nature 2 2 3 5 7

Relative to the probability of state of nature 2, determine the range of probability for which each of the alternatives would maximize expected NPV. 14. Given the following payoff table: (LO4)
State of Nature #1 A Alternative B C D *In $ thousands. $120* 60 10 90 #2 20 40 110 90

a. Determine the range of P(#2) for which each alternative would be best, treating the payoffs as profits. b. Answer part a, treating the payoffs as costs. 15. A law firm is representing a company that is being sued by a customer. The decision is whether to go through litigation or settle the case out of court. The cost of losing the case is estimated to be $1 million, whereas the cost of settling the case is $200,000. Winning the case in court would result in no loss. What is the minimum chance of winning for which the company should contest the case? (LO1 & 4) 16. An 18year-old boy youth just arrived at a hospital complaining of abdominal pain. The medical findings are consistent with appendicitis but not totally typical of appendicitis. The lab and X-ray test results are not clear. The surgeon is wondering whether to operate or wait 12 hours

20

PART THREE System Design

(in case it is not appendicitis). In this case, if the pain does not recede, then it must be appendicitis and the surgeon will operate. The probability that it is appendicitis is 56 percent using past experience with similar cases. Also, from similar cases, the probability that the appendicitis will perforate after 12 hours of wait is 6 percent. The payoffs are measured in terms of death rate. The death rate of operating when appendicitis is present is 0.09 percent; it is 0.04 percent when appendicitis is not present. The death rate of operating a perforated appendicitis is 0.64 percent. Draw the decision tree and determine the best course of action in this case.1 (LO2) 17. An oil company has some land that may contain oil.2 The oil reserve and its probability (based on experience in the area) are:
Oil reserve (in 1000 barrels) Probability 500 .1 200 .15 50 .25 0 .50

The company has to decide whether to drill for oil or lease the land. Cost of drilling will be $500,000 if it is a dry well or $750,000 if it is oil producing. The profit per barrel will be $45. The revenue for leasing the land will be $3,000,000. Determine the best decision for the oil company. 18. The board of governors of Santa Clara University is contemplating mandatory testing of their student athletes.3 The test is not 100 percent accurate. The conditional probabilities are as follows: If an athlete uses drugs, then the test will be positive 94 percent of the time, whereas if he does not (i.e., he is a non-user), the test will be negative 98 percent of the time. Suppose the board suspects that 4 percent of the athletes use drugs. (LO2 & 3) a. Calculate the posterior probabilities. b. If a test costs $50, the cost of not identifying (and barring) a drug user is $1,000, the cost of falsely accusing a non-user is $100, and other costs are zero, should the university test any athlete? 19. An electric utility is trying to decide whether to replace the PCB transformer in a generating station with a new and safer transformer.4 To evaluate this decision, the utility needs information about the likelihood of an incident, such as a fire, the cost of such an incident, and the cost of replacing the transformer. Suppose that the total cost of replacement is $75,000. If the transformer is replaced, there is virtually no chance of a fire. However, if the current transformer is retained, the probability of a fire is assessed to be 0.0025. If a fire occurs, then the clean-up cost could be high ($80 million) or low ($5 million). The probability of a high clean-up cost, given that a fire occurs, is assessed to be 0.2. (LO2 & 4) a. Should the company replace the transformer? b. Perform sensitivity analysis on the probability of a fire. Does the optimal decision from part (a) remain optimal for a wide range of values for this probability? 20. A 59year-old executive has just completed his annual physical exam.5 (The exam included a test on a sample of his stool. The purpose of this test is to screen for cancer of the large intestine (colon cancer). The test looks for small amounts of blood in the stool. The executive is informed that the test showed presence of blood in his stool. The executive is quite anxious. The likelihood of any 59year-old man having colon cancer is 1 in 1,000. The test detects 85 percent of cases of colon cancer, but 2 percent of patients without cancer have a small amount of blood in their stool. What is the probability that he has colon cancer? (LO3) 21. An offshore underwater structure has tested positive for oil by two exploratory wells drilled near the crest of the structure.6 Management is now trying to determine whether it is wise
1

 . Clarke, The Application of Decision Analysis to Clinical Medicine, Interfaces 17(2), MarchApril 1987, R pp. 2734. 2 F. S. Hillier and G. J. Lieberman, Operations Research, 2nd ed, 1974, San Francisco: Holden-Day.  3 C. D. Feinstein, Deciding Whether to Test Student Athletes for Drug Use, Interfaces 20(3), 1990, pp.8087.  4 W. E. Balson et al, Using Decision Analysis and Risk Analysis to Manage Utility Environmental Risk,  Interfaces 22(6), 1992, pp. 126139. 5 P. H. Hill et al, Making Decisions, Massachusetts: Addison-Wesley, 1980, p. 170.  6 P. D. Newendorp, Decision Analysis for Petroleum Exploration, Tulsa, Oklahoma: PennWell Publishing, 1975,  p. 524.

SUPPLEMENT TO CHAPTER 5 Decision Analysis

21

to make a definite commitment to begin building a platform, and, if so, what size, or to defer the decision to allow for the drilling of an additional exploratory well. The principal uncertainty is the size of the field. If the structure is nearly filled with oil, the area of the field will be about 15 square miles and it would require installation of a large platform costing $50 million. If the structure is only partially filled with oil, the area will be much smaller and the field could be depleted with fewer wells and a smaller platform costing $30 million. Based on the subjective judgment of companys geologists, the likelihood of the structure being completely filled is 0.4 (i.e., field is large). If a small platform is built and the field turns out to be large, a second small platform will have to be built at the same cost. If another exploratory well is drilled along the flanks of the structure, there is a 10 percent chance that the result will be erroneous. However, management will base its platform size decision on the result of the exploratory well. The exploratory well will cost $2 million. (LO2 & 3) a. Calculate the posterior probabilities. b. Which decision has the lowest expected cost? 22. The management of a printed circuit boards (PCB) manufacturer is concerned about the significant number of bad (defective) PCBs that are sent to the companys customers.7 The quality control procedure for the PCBs is as follows: each PCB is sent to a testing station where it is tested using an electronic instrument; those that pass are sent to the customers; those that fail are retested; those that pass the second test are sent to the customers; those that fail the second time are retested again; those that pass the third test are sent to customers, and those that fail the third time are discarded. A consultant is brought in to examine the situation. She studied 1,136 PCBs. Of these, 1,087 passed the quality test at the first attempt. The 49 that failed were retested, and 21 passed. The remaining 28 PCBs all failed the third time and were discarded. The consultant asked for retesting of those that passed the first time. A sample of 200 was retested and 7 failed. (LO3) a. Calculate the probability of a good (non-defective) PCB passing the test the first time and the probability of a bad (defective) PCB failing the test the first time. Also, calculate the proportion of PCBs that are good (non-defective) and the proportion that are bad (defective). b. Propose a testing procedure that improves the quality of PCBs sent to the customers.
Mini-Case

Southern Company

leet managers have a large pool of cars and trucks to maintain.8 One approach to the vehicle maintenance is to use oil analysis, where the oil from the engine and transmission are subjected periodically to a test. These tests can sometimes signal an impending failure (for example, too much iron in the oil), and preventive maintenance is then performed (at a relatively low cost), eliminating the risk of failure (failure would result in a relatively high cost). However, oil analysis costs money, and it is not perfectit can indicate that a unit is defective when in fact it is not, and it can indicate that a unit is non-defective when in fact it is. As a possible substitute for oil analysis, the company could simply change the oil periodically, thereby reducing the probability of failure. The fleet manager for the Southern Company, an electrical utility

based in Atlanta, has four alternatives: (1)do nothing, (2) use oil analysis only, (3) replace oil only, or (4) replace oil and do oil analysis. For option (1) the probability of failure is 0.1, and the cost of failure is $1,200 (includes cost of an oil change). For option (2), the probability of failure remains at 0.1. If the unit is about to fail, the oil analysis will indicate this with probability 0.7; if the unit is not about to fail, the oil analysis will indicate this with probability 0.8. The oil analysis itself costs $20, and if it indicates that failure is about to occur, the oil will be changed at the cost of $14.80 and preventive maintenance will be performed. The cost of preventative maintenance to restore a unit that is about to fail is $500, whereas the cost of maintenance for a unit that is not about to fail is $250. The only difference between options (3) and (4) is that probability of failure decreases from 0.1 to 0.04 if oil is changed periodically. Analyze this decision problem.

 . C. Bell, Management Science/Operations Research: A Strategic Perspective, Cincinnati: South-Western P College Publishing, 1999, p. 90. 8 J. M. Mellichamp, The Southern Company Uses a Probability Model for Cost Justification of Oil Sample  Analysis, Interfaces 23(3), 1993, pp. 118124.

22

PART THREE System Design

Mini-Case

ICI Canada

Probability
Significant market and technical feasibility (P1) Board sanctions plant (P2) Commercial success (P3) 0.360.09 0.80.2 0.80.2

CIs Canadian subsidiary (now part of AkzoNobel) discovered a new, but unpatentable, application for a chemical agent to reduce pulp-mill water pollution.9 However, everything was quite uncertain, and the management was trying to decide whether to go ahead with its R&D or abandon the product. The following questions indicate the primary risks: Would market tests confirm that there is a significant market for the product and could the company develop a new process for making this productthat is, is it technically feasible? After a production process is developed, would the companys board sanction production on a commercial scale? Would the venture turn out to be commercially successful? The management team assumed that each of these questions had a yes or no answer. The probabilities of yes answers are shown below. The plus-or-minus value indicates managements uncertainty about the true probabilities.

The primary economic factors and their expense/gain (in million dollars) were the following: The marketing development cost to determine whether there is a significant market and research expenses to identify a new production process for the product (C1): $125% The process development costs, including pre-sanction engineering & commercial development (C2): $3.525% The commercial development costs after the boards sanction (C3): $1.025% The venture value (net present value) if successful (R): $25.050% Again, the plus-or-minus values indicate managements onsiderable uncertainty about the values. Should management c go-ahead with R&D for this product?

 . W. Hess, Swinging on the Branch of a Tree: Project Selection Applications, Interfaces 23(6), 1993, S pp.512).