應用資料包絡分析模式(data envelopment analysis, DEA)於績效評估時,每一個決策單位都會選擇對其最有利的一組權數,來計算其效率分數。其缺點是每一個決策單位用來計算效率分數的變數權數會有很大的差異。為了減低各評估變數權數的差異,本研究提出以調整上下界限的方式,對各評估變數的權數做限制。此方法可避免求解DEA模式時找不到可行解,或產生過多有效率的決策單位導致區別不佳的情形。本方法亦可應用於整合諸專家意見,或折衝專家意見與原始DEA模式對變數權數的歧異。本文舉二個例子來說明本研究方法,結果顯示本研究方法在彈性與效率的區別力上,有較佳的效果。其中,第二個例子更說明了針對決策單位作排序時,不能使用無效率決策單位的變數權數,否則,在求解DEA模式時,可能產生沒有效率分數為1的決策單位之窘境。
In data envelopment analysis (DEA), each decision-making unit (DMU) incorporates its most favorable set of weights to calculate its efficiency. This results in a set of weights that varies for each DMU. To reduce the variation in factor weight for all DMUs, this study introduces a statistic approach to construct the lower and upper bounds of factor weights. This approach enables the weight bounds be adjusted to avoid the absence of efficient DMU or reduce the number of efficient DMUs. It can also be used to integrate expert opinions or resolve the conflict between the results obtained using DEA model and the experience of experts if expert information is available. Two examples illustrate that this method has merits in both flexibility and discrimination in performance evaluation. Moreover, the second example shows that the weights of inefficient DMUs are not suitable for ranking.