Transcript Slide 1
A Study on the Compatibility between Decision Vectors Claudio Garuti Universidad Federico Santa María, Chile [email protected] Valério Salomon Sao Paulo State University, Brazil [email protected] June 6, 2010 1 A Study on the Compatibility between Decision Vectors 1. Introduction 2. Compatibility between two vectors 3. Examples of compatibility indexes utilization 4. Concluding remarks References June 6, 2010 2 1. Introduction x = wD x: decision vector, D: decision matrix, w: weights for the criteria Alternative June 6, 2010 Criterion 1 Criterion 2 … Criterion j … Criterion n 1 d11 d12 … d1j … d1n 2 d21 d22 … d2j … d2n … … … … … … … i di1 di2 … dij … din … … … … … … … m dm1 dm2 … dmj … dmn 3 1. Introduction When we can say that x is a good decision vector? “Garbage in, Garbage out” Or else, if w and D were good, then, x must also be good. With AHP, the most widely used MCDM method (Wallenius et al. 2008), we try to answer that question with the consistency checking: A w = lMAX w m = (lMAX – n)/(n – 1) must be less than 10%. June 6, 2010 4 1. Introduction An example of consistency checked Relative electric consumption of household appliances Alternatives Electric range (A) Refrigerator (B) TV (C) Dishwasher (D) Iron (E) Hair-dryer (F) Radio (G) A 1 1/2 1/5 1/8 1/7 1/9 1/9 B 2 1 1/4 1/5 1/5 1/7 1/9 C D E F 5 8 7 9 4 5 5 7 1 2 5 6 1/2 1 4 9 1/5 1/4 1 5 1/6 1/9 1/5 1 1/8 1/9 1/9 1/5 G 9 9 8 9 9 5 1 w Actual .393 .392 .261 .242 .131 .167 .110 .120 .061 .047 .028 .028 .016 .003 Consistency checking: m = 0.02, OK! But, perhaps more important than that, w and Actual are close to each other. June 6, 2010 5 1. Introduction Decision vectors from applications of three different MCDM methods Alternative AHP MACBETH MAUT Hire more personnel .15 24 .22 Keep the same personnel .37 64 .61 Outsource .48 71 .72 Criteria like costs and quality from the service were considered. Different MCDM methods applications result on different cardinal decision vectors. But, they result in the same ordinal decision vector. June 6, 2010 6 2. Compatibility between 2 vectors When two vectors are close to each other, we can say that they are compatible . Saaty (2006) proposes a compatibility index, S, based on the Hadamard Product. Garuti (2007) proposes another compatibility index, G, based on the inner product between two vectors. Despite Compatibility is a new theme in MCDM, some limitations of S has already been identified. June 6, 2010 7 2. Compatibility between 2 vectors Saaty ‘s compatibility index, between x and y 1 T S = 2 e A •BTe n When: aij = xi/xj, bij = yi/yj, e is a row-matrix with all components equal to 1, and A • B is the cellwise or Hadamard Product: aij • bij= aij bij If x = y, then S = 1 If S > 1.1, then Saaty (2006) proposes that x and y were considered as not compatible June 6, 2010 8 2. Compatibility between 2 vectors The inner product x ∙ y = |x| |y| cosa = S(xi yj) x y a = 0° → total projection → → total vector similarity → total compatibility x∙y=1 x June 6, 2010 a = 90° → no projection → → no vector similarity → total incompatibility y x∙y=0 9 2. Compatibility between 2 vectors Garuti‘s compatibility index, between x and y min(xi , yi ) xi + yi G=∑ max(xi , yi ) 2 If x = y, then G = 1 If G < .9, then Garuti (2007) proposes that x and y were considered as not compatible June 6, 2010 10 3. Examples of S and G utilization Relative electric consumption of household appliances Alternatives Electric range Refrigerator TV Dishwasher Iron Hair-dryer Radio June 6, 2010 w Actual .393 .392 .261 .242 .131 .167 .110 .120 .061 .047 .028 .028 .016 .003 We have: S = 1.455 G = .92 So w and Actual are: • not compatible with S • compatible with G 11 3. Examples of S and G utilization Normalizing the data from Slide 6 Alternative AHP MACBETH MAUT Hire more personnel .15 .15 .14 Keep the same personnel .37 .40 .39 Outsource .48 .45 .47 For AHP and MACBETH we have S = 1.003 and G = .94 For AHP and MAUT we have S = 1.002 and G = .96 For MACBETH and MAUT we have S = 1.002 and G = .96 June 6, 2010 12 4. Concluding remarks With the compatibility index, G, we can answer: “When we can say that x is a good decision vector?” Instead of S, G is a weighted index, since it uses the average between vectors components or even the weights for the criteria, w. As Compatibility is a new theme in MCDM, more study and applications will be necessary to prove this theory in Decision Making. June 6, 2010 13 References Garuti, C. Measuring compatibility (closeness) in weighted environments: when close really means close? Int. Symposium on AHP, 9, Vina del Mar, 2007. Saaty, T. L. Fundamentals of decision making and priority theory. 2 ed. Pittsburgh : RWS, 2006. Wallenius, J., et al. Multiple criteria decision making, multiattribute utility theory: recent accomplishments and what lies ahead. Management Science, 7, 2008, Vol. 54, pp. 1336-1349. Whitaker, R. 2007. Validation examples of the Analytic Hierarchy Process and Analytic Network Process. Mathematical and Computer Modelling. 2007, Vol. 46, pp. 840-859. June 6, 2010 14 A Study on the Compatibility between Decision Vectors Claudio Garuti thanks Sociedad Argentina de Informática (SADIO) and Universidad Federico Santa María Valério Salomon thanks the Sao Paulo Research Foundation (FAPESP) for financial support June 6, 2010 15