Research On Rank Reversal In ANP And AHP Method

ANP/AHP method as a typical multi-attribute decision model has been widelyused in various fields, and has the advantages of easy to understand, easy to operate,flexibility and practicality. However, due to the AHP algorithm has some drawbacks,its rank reversal phenomenon has been the focus of the academic debate. As ageneralization of the AHP, we can easily find that the ANP method also has theproblem of rank reversal. The so-called selective reversal problem (also known asrank reversal problem) refers to changes of alternatives’ order of preference when analternative is added to or removed from the decision problem. Rank reversalphenomenon has led to a change in the decision-making result, we will be unable todetermine which decision result is the correct, but also reduces the effect andefficiency of the decision-making model.To solve this problem, this paper attempts to introduce behavioral decisiontheory into ANP/AHP model, by analyzing the characteristics of decision makerbehavioral factors, to extract the behavioral variables, and introducing behavioralvariables into the decision-making model. We will establish an ANP model whichincludes psychological and behavioral characteristics of decision-makers to overcomethe reverse problem.Firstly, this paper introduces the rank reversal phenomenon in AHP model,summarizes several typical improved methods proposed by predecessors, finds outthree kinds of effective and feasible order preserving methods. Then through a lot ofsimulation, we find numerical examples for rank reversal in the ANP method andanalyse the cause of rank reversal. Finally, we propose an approach to avoid theproblem based on the ideal point, and verify the validity of our proposed approach.Only when certain conditions are met, the improved method based on ideal pointcan solve the rank reversal problem in ANP model. So the improved method alsorequires in-depth study and continue to improve. But this opens up a new directionfor the study of the reversal problems for ANP/AHP. I believe this improvementevolving thought would make ANP/AHP model more universal in the actualdecision-making.