Decision Analysis-Decision Trees • A decision tree is a graphical representation of every possible sequence of decision and random outcomes (states of nature) that can occur within a given decision making problem. • A decision tree is composed of a collection of nodes (represented by circles and squares) interconnected by branches (represented by lines). -73- HMP654 Decision Analysis-Decision Trees General Form of a Decision Tree -74- HMP654 Decision Analysis-Decision Trees • A square node is called a decision node because it represents a decision. Branches emanating from a decision node represent the different alternatives for a particular decision. Alternative A Alternative B Alternative C Decision Node -75- HMP654 Decision Analysis-Decision Trees • A circular node in a decision tree is called an event node because it represents an uncertain event. The branches emanating from an event node correspond to the possible states of nature or the possible outcomes of an uncertain event. State of Nature 1 State of Nature 2 State of Nature 3 Event Node -76- HMP654 Decision Analysis-Decision Trees Case Problem - (A) p. 38 (continued) -77- HMP654 Decision Analysis-Decision Trees -78- HMP654 Decision Analysis-Decision Trees Evaluation of Nodes V1 V2 V4 V3 V4 = MAX(V1, V2, V3, .....) • In a maximization problem, the value assigned to a decision node is the maximum of the values of the adjacent nodes. -79- HMP654 Decision Analysis-Decision Trees Evaluation of Nodes p1 V4 p2 p3 V1 V2 V3 V4 = V1 x p1 + V2 x p2 + V3 x p3 • The value assigned to an event node is the expectation of the values that correspond to adjacent nodes. -80- HMP654 Decision Analysis-Decision Trees -81- HMP654 Decision Analysis-Decision Trees Case Problem (A) p. 64 The agency has been in operation for more than a year and is now reassessing its performance and staffing. In reviewing demand data compiled in its information system, the agency learns that monthly demand has actually been slightly different from what had originally been anticipated. In fact, the agency now feels that monthly demand is more realistically modeled by the following probability distribution: Monthly Demand 30 90 140 150 Probability 0.10 0.27 0.33 0.30 The home health agency now has several things to consider as it plans how it will provide physical therapy services for its clients in the coming year. First of all, a new independent contractor has approached the agency offering to provide PT services for a flat rate of $55 per visit. No fringe benefits or other costs would be incurred. In addition, this contractor has also developed a new marketing program that it has successfully applied in a number of other cities. This program consists of an intensive month-long campaign to recruit additional clients followed by a brief market research study to determine the success of the effort. The agency has the option of purchasing this marketing program whether or not it hires the contractor to provide PT services. The agency has surveyed a number of organizations that have utilized this marketing program. The results of this survey indicate that the contractor has a 72 percent success rate in increasing demand for PT services. However, in the remaining 28 percent of the cases there was actually a decrease in demand for PT services because of the negative reaction by potential clients to the contractor's hard-sell marketing approached. The home health agency carefully analyzes the results of this quick but methodical survey and derives two additional probability distributions for demand for PT servicesone that is expected to hold if the marketing campaign to recruit additional clients is successful, and a second distribution applicable if the marketing campaign is a failure. The distribution of demand created by a successful marketing campaign is given below: Monthly Demand 140 150 Probability 0.5 0.5 On the other hand, when the marketing campaign is not successful, the demand is expected to be described by the following distribution: Monthly Demand 30 90 Probability 0.5 0.5 The home health agency now has several decisions to make. First of all, it must decide whether to negotiate with the new independent contractor to perform the -82- HMP654 Decision Analysis-Decision Trees marketing campaign and follow-up market research study. the cost of this program is $300 per month (for the 12-month planning period currently under study). If the home health agency does decide to contract for the marketing program, then it will receive a marketing research report indicating whether the marketing program was a success or a failure. The agency must decide for each reported outcome whether it will continue utilizing its salaried PT or utilize the contractor to provide the PT services. The costs associated with these two options are the same as those outlined above. In all cases, the average payment for a PT home visit is $75 per visit, and the agency is trying to maximize expected net profit. The home health agency realizes that the optimum approach is dependent upon the cost of the marketing program (currently set at $300 per month), so another objective is to investigate the sensitivity of the solution to this cost. Upon realizing that they must perform a multistage decision analysis, the agency staff turns their attention to the details of constructing an appropriate decision tree model. -83- HMP654 Decision Analysis-Decision Trees 0.1 Dem and = 30 0.27 Dem and = 90 Salaried PT 0.33 Dem and = 140 0.3 Dem and = 150 No Cam paign 1 0.1 Dem and = 30 0.27 Dem and = 90 Purchase PT Serv ices 0.33 Dem and = 140 0.3 Dem and = 150 1 0.5 D em and = 30 S alaried P T 0.5 D em and = 90 0.28 C am paign is a F ailure 1 0.5 D em and = 30 P urchase P T S erv ices 0.5 D em and = 90 C am paign 0.5 D em and = 140 S alaried P T 0.5 D em and = 150 0.72 C am paign is a S uccess 1 0.5 D em and = 140 P urchase P T S erv ices 0.5 D em and = 150 -84- HMP654 Decision Analysis-Decision Trees 0.1 Demand = 30 -2360 2040 -2360 0.27 Demand = 90 1720 6120 Salaried PT -4400 1720 0.33 Demand = 140 3658 5120 9520 5120 0.3 Demand = 150 5800 10200 No Campaign 5800 1 0 0.1 Demand = 30 3658 600 600 600 0.27 Demand = 90 1800 Purchase PT Services 0 1800 1800 0.33 Demand = 140 2370 2800 2800 2800 0.3 Demand = 150 3000 3000 3000 2 3967.2 0.5 Dem and = 30 -2660 Salaried PT -4400 2040 -620 0.28 Cam paign is a Failure -2660 0.5 Dem and = 90 1420 6120 1420 2 0 900 0.5 Dem and = 30 300 Purchase PT Services 0 900 600 300 0.5 Dem and = 90 1500 Cam paign -300 1800 3967.2 1500 0.5 Dem and = 140 4820 Salaried PT -4400 9520 5160 0.72 Cam paign is a Success 4820 0.5 Dem and = 150 5500 10200 5500 1 0 5160 0.5 Dem and = 140 2500 Purchase PT Services 0 2600 2800 2500 0.5 Dem and = 150 2700 3000 -85- 2700 HMP654 Decision Analysis - Treeplan Ctrl-t activates Treeplan Decision 1 0 0 0 1 0.5 Event 3 0 0 Decision 2 0 0 0 0 0.5 Event 4 0 0 -86- 0 HMP654 Decision Analysis - Treeplan -87- HMP654 Decision Analysis - Probability Frequency Table CB CR SS 120 10 130 ST 15 85 100 135 95 230 Joint Probability Distribution CB CR SS 0.52 0.04 0.57 ST 0.07 0.37 0.43 p CB ST 15 230 0.59 0.41 1 p CR 130 p SS 230 -88- 95 230 HMP654 Decision Analysis Conditional Probability Conditional Probabilities Color given Shape CB CR SS 0.92 0.08 1 pCR ST 0.37 pCR ST pST 0.43 ST 0.15 0.85 1 Compare to p(CR) = 0.41 Shape given color CB CR SS 0.89 0.11 ST 0.11 0.89 1 1 pST CR Compare to p(ST) = 0.43 -89- pCR ST 0.37 pCR 0.41 HMP654 Decision Analysis Perfect Information Perfect Information Frequency Table CB CR SS 135 0 135 ST 0 95 95 Joint Probability Distribution 135 95 230 CB CR Color given Shape CB CR SS 1 0 SS 0.59 0.00 0.59 ST 0.00 0.41 0.41 0.59 0.41 1 Shape given Color ST 0 1 CB CR -90- SS 1 0 ST 0 1 HMP654 Decision Analysis No Information No Information Frequency Table CB CR SS 413 287 700 ST 177 123 300 Joint Probability Distribution 590 410 1000 CB CR Color given Shape CB CR SS 0.59 0.41 SS 0.41 0.29 0.70 ST 0.18 0.12 0.30 0.59 0.41 1 Shape given Color ST 0.59 0.41 CB CR -91- SS 0.7 0.7 ST 0.3 0.3 HMP654 Decision Analysis Perfect Information 0.59 Draw blue 10 Predict blue 0 0 3.85 10 0.41 Draw red -5 Don't use shape info 0 -5 1 0 3.85 0.59 Draw blue -5 Predict red 0 0 1.15 -5 0.41 Draw red 10 0 10 1 Draw blue 10 Predict blue 0 10 2 10 0 10 0 Draw red 0.59 Draw square -5 0 -5 1 0 10 1 Draw blue -5 Predict red 0 0 -5 -5 0 Draw red 10 Use shape info 0 0 10 10 0 Draw blue 10 Predict blue 0 0 -5 10 1 Draw red 0.41 Draw triangle -5 0 -5 2 0 10 0 Draw blue -5 Predict red 0 0 10 -5 1 Draw red 10 0 -92- 10 HMP654 Decision Analysis No Information 0.59 Draw blue 10 Predict blue 0 0 3.85 10 0.41 Draw red -5 Don't use shape info 0 -5 1 0 3.85 0.59 Draw blue -5 Predict red 0 0 1.15 -5 0.41 Draw red 10 0 10 0.59 Draw blue 10 Predict blue 0 10 1 3.85 0 3.85 0.7 Draw square 0.41 Draw red -5 0 -5 1 0 3.85 0.59 Draw blue -5 Predict red 0 0 1.15 -5 0.41 Draw red 10 Use shape info 0 0 3.85 10 0.59 Draw blue 10 Predict blue 0 0 3.85 0.3 Draw triangle 10 0.41 Draw red -5 0 -5 1 0 3.85 0.59 Draw blue -5 Predict red 0 0 1.15 -5 0.41 Draw red 10 0 -93- 10 HMP654 Decision Analysis Imperfect Information 0.59 Draw blue 10 Predict blue 0 0 3.85 10 0.41 Draw red -5 Don't use shape info 0 -5 1 0 3.85 0.59 Draw blue -5 Predict red 0 0 1.15 -5 0.41 Draw red 10 0 10 0.92 Draw blue 10 Predict blue 0 10 2 8.3485 0 8.8 0.57 Draw square 0.08 Draw red -5 0 -5 1 0 8.8 0.92 Draw blue -5 Predict red 0 0 -3.8 -5 0.08 Draw red 10 Use shape info 0 0 8.3485 10 0.15 Draw blue 10 Predict blue 0 0 -2.75 0.43 Draw triangle 10 0.85 Draw red -5 0 -5 2 0 7.75 0.15 Draw blue -5 Predict red 0 0 7.75 -5 0.85 Draw red 10 0 -94- 10 HMP654 Decision Analysis Bayes Theorem pSS pSS CB pSS CR p SS CB pSS CB pSS CB p SS CB pCB pCB p SS CR pSS CR pSS CR p SS CR pCR pCR pSS p SS CB pCB p SS CR pCR In a similar wa y, it can be shown that pST p ST CB pCB pST CR pCR then p CB SS p SS CB pCB pSS CB pSS pSS CB pCB p SS CR pCR etc. -95- HMP654 Decision Analysis-Decision Trees Modified Case Problem - Imperfect Information • Assume that it is possible for the market research report to be wrong. Thus, the content of the report does not provide the decision maker with certain knowledge about the true outcome of the campaign. Outcome of Marketing Research Report Result is really a success (S) Result is really a failure (F) Report says “success” (RS) 0.85 0.25 Report says “failure” (RF) 0.15 0.75 Conditional probabilities of ‘report outcomes’ given ‘actual outcomes’ -96- HMP654 Decision Analysis-Decision Trees Modified Case Problem - Imperfect Information D em and = 30 D em and = 90 S alaried P T D em and = 140 D em and = 150 No C am paign 1 D em and = 30 D em and = 90 P urc hase P T S ervic es D em and = 140 D em and = 150 D em and = 30 C am paign is a failure D em and = 90 S alaried P T 1 D em and = 140 C am paign is a suc c ess D em and = 150 R eport say s "F ailure" 1 D em and = 30 C am paign is a failure D em and = 90 P urc hase P T S ervic es D em and = 140 C am paign is a suc c ess D em and = 150 C am paign D em and = 30 C am paign is a failure D em and = 90 S alaried P T D em and = 140 C am paign is a suc c ess D em and = 150 R eport say s "S uc c ess" 1 D em and = 30 C am paign is a failure D em and = 90 P urc hase P T S ervic es D em and = 140 C am paign is a suc c ess D em and = 150 -97- HMP654 Decision Analysis-Decision Trees Modified Case Problem - Imperfect Information p RS S 0.85 pS 0.72 p RF S 0.15 pF 0.28 p RS F 0.25 p RF F 0.75 pRS p RS S pS p RS F pF pRF p RF S pS p RF F pF p S RS p F RS p RS S pS pRS p RS F pF pRS -98- p S RF p F RF p RF S pS pRF p RF F pF pRF HMP654 Decision Analysis-Decision Trees Modified Case Problem - Imperfect Information Probabilities of “report outcome” given “actual outcome” S RS RF F 0.85 0.25 0.15 0.75 0.72 p(S) 0.28 p(F) 0.682 p(RS) 0.318 p(RF) Probabilities of “actual outcome” given “report outcome” S F RS 0.8974 0.1026 RF 0.3396 0.6604 -99- HMP654 Decision Analysis-Decision Trees Modified Case Problem - Imperfect Information 0.1 Demand = 30 -2,360 2,040 -2,360 0.27 Demand = 90 1,720 Salaried PT -4,400 6,120 3,658 1,720 0.33 Demand = 140 5,120 9,520 5,120 0.3 Demand = 150 5,800 No Campaign 10,200 5,800 1 0 3,658 0.1 Demand = 30 600 600 600 0.27 Demand = 90 1,800 Purchase PT Services 0 2,370 1,800 1,800 0.33 Demand = 140 2,800 2,800 2,800 0.3 Demand = 150 3,000 3,000 3,000 Next Page -100- HMP654 Decision Analysis-Decision Trees Modified Case Problem- Imperfect Information 0.5 Demand = 30 0.6604 Campaign is a failure Previous Page 0 -620 -2,660 2,040 -2,660 0.5 Demand = 90 1,420 Salaried PT 6,120 1,420 1 3,658 -4,400 1,343 0.5 Demand = 140 0.3396 Campaign is a success 0 5,160 0.318 Report says "Failure" 4,820 9,520 4,820 0.5 Demand = 150 5,500 10,200 5,500 2 0 1,477 0.5 Demand = 30 0.6604 Campaign is a failure 0 900 300 600 300 0.5 Demand = 90 1,500 Purchase PT Services 0 1,800 1,477 1,500 0.5 Demand = 140 0.3396 Campaign is a success 0 2,600 2,500 2,800 2,500 0.5 Demand = 150 2,700 Campaign -300 3,000 3,584 2,700 0.5 Demand = 30 0.1026 Campaign is a failure 0 -620 -2,660 2,040 -2,660 0.5 Demand = 90 1,420 Salaried PT -4,400 6,120 4,567 1,420 0.5 Demand = 140 0.8974 Campaign is a success 0 5,160 0.682 Report says "Success" 4,820 9,520 4,820 0.5 Demand = 150 5,500 10,200 5,500 1 0 4,567 0.5 Demand = 30 0.1026 Campaign is a failure 0 900 300 600 300 0.5 Demand = 90 1,500 Purchase PT Services 0 1,800 2,426 1,500 0.5 Demand = 140 0.8974 Campaign is a success 0 2,600 2,500 2,800 2,500 0.5 Demand = 150 2,700 3,000 -101- 2,700 HMP654 Decision Analysis-Decision Trees Imperfect Information-Sensitivity Analysis Probabilities of “report outcome” given “actual outcome” S RS RF F 0.90 0.15 0.10 0.85 0.72 p(S) 0.28 p(F) 0.69 p(RS) 0.31 p(RF) Probabilities of “actual outcome” given “report outcome” S F RS 0.9391 0.0609 RF 0.2323 0.7677 -102- HMP654 Decision Analysis-Decision Trees Imperfect Information-Sensitivity Analysis 0.1 Demand = 30 -2,360 2,040 -2,360 0.27 Demand = 90 1,720 Salaried PT -4,400 6,120 3,658 1,720 0.33 Demand = 140 5,120 9,520 5,120 0.3 Demand = 150 5,800 No Campaign 10,200 5,800 1 0 3,658 0.1 Demand = 30 600 600 600 0.27 Demand = 90 1,800 Purchase PT Services 0 2,370 1,800 1,800 0.33 Demand = 140 2,800 2,800 2,800 0.3 Demand = 150 3,000 3,000 3,000 Next Page -103- HMP654 Decision Analysis-Decision Trees Imperfect Information-Sensitivity Analysis 0.5 Demand = 30 Previous Page 0.7677 Campaign is a failure 0 -620 -2,660 2,040 -2,660 0.5 Demand = 90 1,420 6,120 Salaried PT 1,420 2 -4,400 3,719 0.5 Demand = 140 723 0.2323 Campaign is a success 0 5,160 4,820 9,520 4,820 0.5 Demand = 150 5,500 0.31 Report says "Failure" 10,200 5,500 2 0 0.5 Demand = 30 1,295 0.7677 Campaign is a failure 0 900 300 600 300 0.5 Demand = 90 1,500 1,800 Purchase PT Services 0 1,500 0.5 Demand = 140 1,295 0.2323 Campaign is a success 0 2,600 2,500 2,800 2,500 0.5 Demand = 150 2,700 3,000 Campaign -300 3,719 2,700 0.5 Demand = 30 0.0609 Campaign is a failure 0 -620 -2,660 2,040 -2,660 0.5 Demand = 90 1,420 Salaried PT -4,400 6,120 4,808 1,420 0.5 Demand = 140 0.9391 Campaign is a success 0 5,160 0.69 Report says "Success" 4,820 9,520 4,820 0.5 Demand = 150 5,500 10,200 5,500 1 0 4,808 0.5 Demand = 30 0.0609 Campaign is a failure 0 900 300 600 300 0.5 Demand = 90 1,500 Purchase PT Services 0 1,800 2,496 1,500 0.5 Demand = 140 0.9391 Campaign is a success 0 2,600 2,500 2,800 2,500 0.5 Demand = 150 2,700 3,000 -104- 2,700 HMP654