The purpose of this paper is to investigate the range sensitivity of the analytic hierarchy process (AHP) and evaluate the effectiveness of using a bottom‐up approach to mitigate the possible range insensitivity bias in the AHP.
An experiment was conducted to test the normative range‐sensitivity of four different methods: the AHP with bottom‐up evaluation; direct ratio weights; swing weights; and trade‐off weights. Also, the significance of the range‐sensitivity effects and the differences among weighting approaches were rigorously tested using various statistical models.
Results show that the range sensitivities of AHP and direct ratio weights are significantly less than the range sensitivities of swing weights and tradeoff weights, suggesting that the bottom‐up evaluation approach might not be a feasible solution for the range‐insensitivity problem. This finding is consistent with the value‐comparison hypothesis proposed in an earlier study, and is partially supported by the theory of the multi‐dimensionality of attribute importance.
It is concluded that treating the attribute weights and performance scoring scales separately in the AHP or other multi‐attribute decision analysis models might lead to an arbitrary final ranking of alternatives. Therefore, it may be necessary to incorporate better elicitation procedures into the AHP models to ensure that attribute weights properly reflect the range or scale of measurement.
This study provides new evidence and issues words of warning of the range‐sensitivity effects in the multi‐attribute decision analysis.
