Multi-Criteria Decision Making Decision Making and Risk, Sp2006: Session 6 Notebook Computer Decision Consumer is interested in buying a notebook computer. Goes to electronics store. Sees the following options. How should she go about buying one? Choices Brand and Model Price Processor Speed RAM HD Screen Model “A” 699 1.8 GHz 256 MB 100 GB 15.4 Model “B” 749 1.9 GHz 256 MB 100 GB 15 Model “C” 599 1.4 GHz 512 MB 60 GB 12 Model “D” 799 1.7 GHz 256 MB 80 GB 14.1 How would you proceed if you were the one making the choice? Types of MCDM Rules Compensatory: Simple compensatory Linear (integration and valuation) Non-linear (linear integration, non-linear valuation) Non-compensatory Screening Selection Conjunctive Disjunctive Elimination by aspects Lexicographic Heuristics Simple Compensatory Rule Alternative with the most number of top of class performance. Brand and Model Price Processor Speed RAM HD Screen Model “A” 699 1.8 GHz 256 MB 90 GB 15.4 1 Model “B” 749 1.9 GHz 256 MB 100 GB 15.4 3 Model “C” 599 1.4 GHz 512 MB 60 GB 12 2 Model “D” 799 1.7 GHz 256 MB 80 GB 14.1 0 Approach #1 Imagine “n” objects, with “m” attributes each. For the ith object: Declare attribute importance for each of the “m” attributes (wj). Determine how much you like the level of the attribute in the ith product (aij) Combine attribute importance with attractiveness of the attribute level featured in the product. Add the above combinations. For the ith object, Vi = Σj=1 to n (wj * aij), akin to expected utility. Repeat this for all i objects. Pick “i” such that, i = Max (V1, V2, V3…..Vn) Illustration – Linear Compensatory Decision Rules Price CPU RAM HD Screen A 3 3 1 3 4 3.1 B 2 4 1 4 4 2.8 C 4 1 4 1 1 2.5 D 1 2 1 2 2 1.5 0.3 0.25 0.2 0.15 0.2 Wt Non-linear Compensatory Decision Rules Valuation of additional unit is not identical across the range of the levels. 60G to 80G is more consequential compared to 100 to 120G However, integration across attributes of the alternatives is still linear. Summary of Compensatory Simple compensatory: No relative valuation of attribute levels other than best/not best, equal weight for all attributes. Linear Compensatory: Equal valuation of attributes levels, linear combination of valuation through attribute-specific weights. Non-linear Compensatory: Unequal valuation of attribute levels, linear combination of valuation through attributespecific weights. Non-Compensatory Decision Rules Screening Rules Conjunctive decision rule Disjunctive decision rule Choice Rules Elimination by aspects Lexicographic decision rule Elimination by Aspects Elimination By Aspects Start with minimum cutoff on most important attribute. Eliminate those that do not clear cutoff. Take the next most important attribute, and repeat steps above. Stop when you have one brand. Elimination rule, rather than selection rule. Conjunctive Eliminate alternatives that don’t meet/exceed cutoff on every attribute. Brand Price CPU RAM HD Screen Size Compaq A 3 3 1 4 4 Compaq B 2 4 1 3 3 Compaq C 4 1 4 1 1 Compaq D 1 2 1 2 2 Min Cutoff 2 2 1 2 2 × × Disjunctive Accept any alternative that exceeds minimum cutoff on at least one attribute. Brand Price CPU RAM HD Screen Size Compaq A 3 3 1 3 4 Compaq B 2 4 1 3 4 Compaq C 4 1 4 1 1 Compaq D 1 2 1 4 2 Cutoff 3 2 1 2 2 Elimination By Aspects Brand Price CPU RAM HD Screen Size Compaq A 3 3 1 3 4 Compaq B 2 4 1 3 4 Compaq C 4 1 4 1 1 Compaq D 1 2 1 4 2 Cutoff <=2 N/A <=2 × × × Lexicographic Select best option on the most important attribute. Brand Price CPU RAM HD Screen Size Compaq A 3 3 1 3 4 Compaq B 2 4 1 3 4 Compaq C 4 1 4 1 1 Compaq D 1 2 1 4 2 Weight .3 .2 .2 .1 .2 Comparing Compensatory and Noncompensatory Decision Rules Compensatory Decision Rules Non-Compensatory Decision Rules Strengths of one attribute can overcome weakness of another. Selection rules Effortful Strength of one attribute cannot overcome weakness of another. Elimination rules Easier Often decision makers use hybrid rules Highlights What happens when decision options come with multiple attributes? Reduce them to a single attribute Deal with multiple attributes Compensatory Non-compensatory Weighted attribute utility approach (compensatory) EBA (sequential elimination) Lexicographic (selection by reduction to single attribute…repeat if necessary) Conjunctive (inclusion based on thresholds for every attribute) Assists in narrowing the consideration set Disjunctive (inclusion based on threshold for at least one attribute) Assists in broadening the consideration set Different strategies at different stages. Vary in effort and data required. Regret may be a function of the type of decision strategy. Thresholds are the result of past experiences, negatives leave a stronger imprint. Heuristics Decision rules, shortcuts. Sometimes, they are meta-decisions. Some Examples What I did the last time around? What does the expert think? The price-quality relationship is: Bogus, so, look for the relatively less expensive option. Valid, so, look for the relatively more expensive option. Minimize decision effort/cost. Minimize regret. Maximize effectiveness.