DECISION MAKING MODELS SUPPLEMENTA 59 EXAMPLE A.7 Decisions under Risk Reconsider the payotf matrix in Example A. For the expected value decision rule, which isthe best alterna tive if te probably of smal damand is estimated to be 0.4 and the probabilty of large demand is estimated to boos? SOLUTION ‘The expected value for each alternative is as follows: ‘ternative Expected Value Smal facity | 0.47200) + 0.6(270) = 242 Lage city | 0.4160) + 0.6(800) = 544 DECISION POINT Management would choose a larg facili it used this expected value dacision rule, because it provides the best ong term resuts if consistenty applied overtime, MyOMLab Tutor AS in MyOMLa trate decisions under tik Decision Trees The decision tree method is a general approach to a wide range of processes and supply chain decisions, such as product planning, process analysis, process capacity, and location. It is par- ticularly valuable for evaluating different capacity expansion alternatives when demand is un- certain and sequential decisions are involved. For example, a company may expand a facility in 2013 only to discover in 2016 that demand is much higher than forecasted. In that case, a second. decision may be necessary to determine whether to expand again or build a second facility. ‘A decision tree is a schematic model of alternatives available to the decision maker, along ‘with their possible consequences. The name derives from the tree-like appearance of the model, Iteonsists of a number of square nodes, representing decision points, which are left by branches (which should be read from left to right), representing the alternatives. Branches leaving circu- lar, or chance, nodes represent the events. The probability of each chance event, P(E), is shown, above each branch. The probabilities for all branches leaving a chance node must sum to 1.0. ‘The conditional payoff, which is the payof for each possible alternative-event combination, is shown at the end of each combination. Payoffs are given only at the outset, before the analysis begins, for the end points of each alternative-event combination, In Figure A, for example, payoff 1 is the financial outcome the manager expects if alternative 1 is chosen and then chance event 1 occurs. No payoff can be associated yet with any branches farther to the left, such as alternative 1 as a whole, because itis followed by a chance event and is not an end point. Payofis often are 5 (PE Patt IPE (PIE Patt 2 Potts Passi 2d cecision (PE BIPIE ‘AE; = Probably of event decision tree A schematic model tives avalable maker, along with ther possible 4 FIGURE AS Ded on Tee Model60 PART 1 COMPETING WITH OPERATIONS. expressed as the present value of net profits. Ifrevenues are not affected by the decision, the pay- offs expressed as net costs. ‘After drawing a decision tree, we solve it by working from right to left, calculating the expected payoff for each node as follows: 1, Foran event node, we multiply the payoff of each event branch by the event's probability. We add these products to get the event node's expected payoff. 2, Foradecision node, we pick the alternative that has the best expected payoff. Ifan alternative leads to an event node, its payoff is equal to that node's expected payoff (already calculated). We “saw off,” or “prune,” the other branches not chosen by marking two short lines through them. The decision node's expected payoff is the one associated with the single remaining unpruned branch. We continue this process until the leftmost decision node is reached. The unpruned branch extending from it is the best alternative to pursue. If multistage decisions are involved, we must await subsequent events before deciding what to do next. Ifnew prob- ability or payoff estimates are obtained, we repeat the process. ‘Various software is available for drawing decision trees. PowerPoint can be used to draw de- cision trees, although it does not have the capability to analyze the decision tree. More extensive capabilities, in addition to POM for Windows, are found with SmartDraw (www.smartdraw.com), Precision(Tree decision analysis from Palisade Corporation (www.palisade.com), and TreePlan, (www.treeplan.com/treeplan.htm). PCat) MyOMLab Ace Mosel Asin Iyoluas provides ‘Analyzing a Decision Tree ‘Arataler must decide whether to bull a small or large facility at a new location. Demand at the location can be ether low or high, wth probabilties estimated to be 0.4 and 0.6, respectively a smal facity is bult land demand praves to be high, the manager may choose not to expand (payott = $223,000) or to expand (payott = $270,000). I a smal fcity is built and demand is low, there is no reason to expand and the payatt is $200,000 fa large fact is bult and demand proves to be low, the choice isto do nothing ($40,000) orto stimulate demand through local advertising, The response to advertising may be ether modest or sizable, with ther probabilities estimated to be 0.3 and 0.7, respective. It is modest, the payotis estimated to be only $820,000; the payoff grows to $220,000 i the response is sizable, Fina, fa large fact is built anc demand tums out to be high, the payot is $800,000. Draw a decision tree. Then analyze to determine the expected payoft for each decision and event node. Which alternative—bulding a small fact or bulding a large facity has the higher expected payot?? SOLUTION “The decision tree in Figure A. shows the event probabilty and the payoff for each ofthe seven alternative event combinations. The frst decision is whether to buld a small ora lage facility, Its node is shown fst, to the lat, because i isthe decision the retaller must make naw. Tha second decision node whether to expand ata later date—Is reached only Ifa smal facilty is bulk and demand turns out to be high. Finally, the third decision oint—whether to advertise—is reached ony ifthe retaler bulds a large facility and demand turns, out to be low. Analysis ofthe decision tree begins with calculation of the expected payots from right to let, shown on Figure A. beneath the appropriate event and decision nodes, 1. For the event node dealing with advertising, the expected payott is 160, or the sum of each event's payott weighted by its probability (0.3(20) + 0.7(220)) 2. The expected payotf for decision node 3 Is 160 because Advertise (160) Is better than Do nothing 40), Prune the Do nothing alternative, 3, The payott for decision node 2 is 270 because Exoand (270) is better than Do not expand (223). Prune De not expand. 4. The expected payoff for the event node dealing with demand, assuming that a small facity is built is 242 [or 0.41200) + 0.61270} 5. The expected payott forthe event node dealing with demand, assuming that a large tact is bul, 1s 544 for 0.4(160) + 0.6(800), 8. The expected payoff for decision node 1 is 544 because the large facilty's expected payor Is largest. Prune Smal facil.