多屬性決策方法評估基礎之設計

Design of Evaluation Benchmarks for Multiattribute Decision Making Techniques

賴世剛(Shih-Kung Lai) ； 許仁成(Jen-Cheng Hsu)

《國立台灣大學建築與城鄉研究學報》 8期 (1996/06) Pp. 105-115

https://doi.org/10.6154/JBP.1996.8.009

摘要

基於多屬性決策方法不易有效表達受測者偏好的課題（Lai及Hopkins，1995），本論文研擬一套評估多屬性決策方法的基礎、或實驗設計，以決定何種方法較能表達受測者偏好。該評估基礎的實驗設計是結合Hopkins（1984）內部評審（internal judges）及外部評審（external judges）兩種評估方法的優點，以比較不同的多屬性決策方法。內部評審主要由受測者本身判斷決策方法的有效性，可表達受測者的偏好。但受測者須使用兩種以上的決策方法，且評估方法使用順序不同所造成的學習效果不易控制。外部評審由外部專家產生彼此一致的標準，受測者只須使用一種決策方法，方法較簡單。但外部專家所產生的標準，無法代表受測者的偏好。因此，本文假設以受測者因群體彼此學習影響，而產生群體穩定價值以作為外部評審。此群體穩定價值亦可解釋為一種共識的達成。本文根據MAVT（Measurable Additive Value Function Theory），或可衡量加法價值函數，提出一種基礎方法（benchmark），作為評估基礎產生的依據。首先，由偏好強度的理論定義價值為偏好強度，使其明確且易操作。以可衡量加法價值函數理論決定受測者初始偏好結構，包括屬性權重及價值函數。以平均強度或加總強度，經由特徵向量法求取受測者的影響力權重，以計算群體偏好強度。本文證明此兩種衡量方式，不管在判斷一致或不一致的情況下，均可求得相同之穩定權重。最後，將受測者不同的影響力權重乘以受測者初始偏好結構，求得穩定的群體偏好強度，作為評估基礎或外部評審。實驗結果的評估，是以數個受測群體使用標準方法所產生的數個評審的偏好結構，與該受測群體的受測者使用決策方法的偏好結構差距比較。兩者的變異係數愈小，代表方法愈有效或受測者及評審群體之共識的差異愈小。在本文所提出的實驗設計中，受測者只須使用標準方法與該組的決策方法，因此，不但較內部評審簡單，亦能解決外部評審無法表達受測者偏好的缺點。

關鍵字

多屬性決策 ； 集體決策 ； 特徵向量 ； 偏好結構

並列摘要

It is not clear whether multiattribute decision making techniques can express effectively decision makers' preferences. This paper presents an experimental design to evaluate these techniques in terms of ability to express decision makers' preferences. The design incorporates the internal and external judges designs as proposed by Hopkins(1984). Internal judges are the subjects who apply these techniques and thus evaluate the effectiveness of these techniques based on their own preferences. In this design, the subjects must apply more than two techniques in a sequence. Learning across techniques is thus difficult to control. External judges are the experts other than subjects who reach a collective consensus for evaluation based on the subjects' performance. In this design, each subject applies only one technique and it is simpler than the internal judges design. The consensus resulting from external judges does not, however, represent the subjects' own preferences for these techniques. We propose a procedure taking into account interactions among subjects resulting in converged, final group preferences as the benchmark for evaluating these techniques.Based on the measurable additive value function theory (MAVT), we propose a benchmark method to derive the evaluation basis. We first define values in concrete and operational terms as preference intensities or strengths of preference suggested by Dyer and Sarin(1979). The subjects' initial preference structures are then represented by MAVT, including attribute weights and value functions. Based on Saaty's computations of average intensities or total intensities in a network and the eigenvector approach(1986), we can then derive among the subjects the influence weights to compute the group preference structure as an evaluation basis. We prove that the two computation methods, average and total intensities, result in consistent weights. The evaluation basis is computed by multiplying the subjects' initial preference structures with the associated influence weights respectively and summing up the weighted preference structures across the subjects.The effectiveness of each technique is determined by the difference between the preference structures expressed by the subjects using that technique and the method as an evaluation basis or external judge. The smaller the difference between the subjects' and the judge's preference structures, the greater the effectiveness of the technique. In our design, subjects apply only the benchmark technique and the technique under consideration; thus it is simpler than the internal judges design while solving the problem of the external judges design for not being able to express subjects' own preferences among techniques.

並列關鍵字

Multiattribute decision making ； group decision making ； eigenvector approach ； preference structures ； AHP ； MAVT