Evaluating the effects of uncertainty in interval pairwise comparison matrices (Q6064251)

The present work is placed in the context of multicriteria decision-making (MCDM) approaches. At the heart of MCDM approaches is the adoption of the pairwise comparison matrices (PCMs). In order to handle uncertainty, PCMs affected by uncertainty are formally represented as matrices of intervals, the so-called interval pairwise comparison matrices (IPCMs). Further, including the concept of expert's uncertainty in the definition, additional data about the expert's uncertainty are considered. By increasing the level of uncertainty of the decision-maker (i.e., the size of the given intervals), the probability of sampling comparison matrices that lead to different ordinal rankings increases. In this context, the article concerns the identification of an absolute ranking. In particular, the authors analyze the relations between the uncertainty of the expert and the presence of comparison matrices associated to different ordinal rankings. Most approaches from the literature analyze a set of random IPCMs which are generated using Monte Carlo simulation starting from the uncertainty intervals in the IPCM. For each matrix in the set the absolute utility vector is computed in order to verify the existence of altered ordinal rankings for some matrix. In order to reduce or avoid the required computational effort, under the hypothesis that all the relative scores are affected by the same degree of uncertainty, and using a mixed-integer quadratic programming problem with linear equality and inequality constraints, able to evaluate the effects of decision maker's uncertainty in the decision-making process in case of IPCM, this work provides an upper bound on the uncertainty (i.e., the minimum effort required to alter the ranking) that guarantees to obtain the same ordinal ranking for each comparison matrix compatible with the given uncertainty degree.