英文摘要 |
The analytic hierarchy process (AHP) is one of the most commonly applied multiattributes decision analysis techniques for the reason that its implementation steps are very common to the people. It is even more popular in cases that quantitative and qualitative attributes combine. Lin et al. (2009) set out from the meaning of a judgment matrix, pointed out that its coefficients are perturbed measurement data of a preference structure, and then showed that Saaty’s eigenvector method has some theoretical weakness and practical disadvantages. On contrast, the methods based on statistical regression not only have nice properties in the decision theory, but also have several practical advantages. Therefore, analyzing the judgment matrix by regression is more proper. In additions, they defined the graphical consistency of a judgment matrix and the graphical consistency constraints for solving the priority vector, then considered the theoretical properties of the logarithm least squares method and the goal programming method with the graphical consistency constraints. It is shown that these two methods not only have nice properties but also several practical advantages. Since finding the priority vector is a problem of statistical estimation, the behavior and the ability of the estimators are as important as their theoretical properties. In this study, we perform a factorial experiment to compare the related methods, for the case that the logarithms of error terms are normally distributed. From the experimental results, it is found that LLSM-GC and GPM-GC perform better than the traditional methods in finding the priority order. In additions, in this experiment we have also found examples that the priority order found by EM is not the true order, and testing the consistency of a judgment matrix by using CR-value is not a serious method. |