Characterization of the Effectiveness of Several Outranking-Based Multi-Criteria Sorting Methods

Abstract

ELECTRE TRI-C, ELECTRE TRI-nC and THESEUS are recently proposed outranking-based multi-criteria sorting methods. The effectiveness of each method is here characterized by (i) its capacity for suggesting precise and appropriate assignments, (ii) the probability of suggesting imprecise assignments, and (iii) the probability of suggesting wrong assignments. We are also interested in studying to what extent the effectiveness of the methods is influenced by the number of criteria and categories, the cardinality of the reference set (in case of ELECTRE TRI-nC and THESEUS), and the way in which this set is chosen or populated. Our conclusions are supported by extensive computer experiments. The results underline the importance of (a) the number of reference actions and their distribution per category, (b) the number of criteria, and (c) the way in which the reference set is populated. These results help to identify contexts for which a particular method may be more appropriate.