Визначення індексу узгодженості оцінок експертів з урахуванням їхньої компетентності при підтримці прийняття групових рішень

Abstract

The issue of assessment of expert estimates’ consistency in cases of group polling taking into ac-count experts’ competence has been considered. The development of an appropriate method should proceed from a range of particular basic postulates. Namely, it is taken into consideration that expert estimates always assume the existence of a «true» estimate corresponding to the average value of an array of individual expert estimates. Such an array can be presented on a continuous or a discrete numerical scale limited on both sides. The maximum level of consistency (1.0) can be achieved only if all experts provide the same estimate. The consistency index shouldn’t be dependent on estimate shifts along the axis. For the same set of estimates, increasing the scale leads to an increased consistency index, and vice versa. If a set of estimates has a higher level of consistency compared to another on a certain scale, it also has greater consistency on any other scale. In case of linear changes (simultaneous proportional increase/decrease) in scale parameters and the values of all estimates, the consistency index remains unchanged. So, the index should be scalable. The higher the level of an expert’s relative competence, the higher this estimate’s impact on the aggregate level of consistency is. The article considers two approaches, when analyzing the level of experts’ competence is reduced to calculating the consistency index without taking into consideration experts’ competence, and when the index is to be calculated as the normalized value of the amount of distances between expert estimates for all possible estimate pairs multiplied by the product of respective experts’ weight coefficients. Imitation modeling has been carried out, and an evaluation has been offered for a threshold consistency value above which aggregation of expert estimates becomes possible. The proposed method has been practically implemented and tested within a system of distributed collection and processing of expert data for decision support systems.