Interpreting the Determinants of Sensitivity in MCDM Methods with a New Perspective: An Application on E-Scooter Selection with the PROBID Method

Abstract

It is not a desirable situation when input parameters excessively affect the results of a system as well as imply unwarranted drift and inefficiency. This situation, which expresses dependence or sensitivity to inputs, is also considered a problem in the multi-criteria decision-making (MCDM) methodology family, which has more than 200 members. A newly produced MCDM method is first subjected to sensitivity tests. MCDM methods are generally evaluated for their sensitivity to weighting methods. Sensitivity is affected by many different parameters such as data, normalization, fundamental equation, and distance type. The common methodical approach for sensitivity analysis is to check whether the best alternative changes with the alteration of weight coefficients. It is problematic to identify sensitivity only in the situation where the ranking position of the best alternative changes. In this study, the sensitivity of the entire ranking is based on a holistic view. Moreover, in the classical method, there is no reference point for sensitivity. Each different MCDM result is compared to each other and it is claimed that the method that produces rankings that are significantly different from the others is poor. We reinterpret sensitivity using the relationship between dynamic MCDM-based performance and static price towards the selection of an environmentally friendly, traffic-saving performance electric scooter. Two PROBID variants as well as the CODAS method are used in this study to deepen the accuracy in the comparison. Additionally, how four types of weighting methods and six types of normalization types affected MCDM sensitivity is measured with a different statistical framework. The finding from a total of 72 different MCDM rankings is striking: If the sensitivity of an MCDM method is generally high, the correlation between that MCDM method and the external anchor (price) is low. Conversely, if sentiment is low, a high correlation with price results. These matching patterns are a unique discovery of this work.