對於一群利用多投入因子以生產多項產出之同質性決策單位需要被評估確認學習標竿時，一個以線性規劃為基礎的方法論：資料包絡分析，已被認定是實用的標竿評選工具。然而，產出項指標經常被考慮取自於二個不同屬性構面，例如：「營運構面與品質構面」或「營運構面與獲利構面」。當雙構面產出項指標被直接摻合執行資料包絡分析進行標竿評選時，在其中一個構面具有低產出指標的決策單位可能被誤認為學習標竿；更確切而言，學習標竿的績效表現可能在其中一個構面被所對應的無效率決策單位凌駕。儘管在近年來已有各種標竿方法被提出來增強傳統資料包絡分析的標竿評選能力，但是這些標竿方法仍有部份缺失值得改善：求解過程涉及價值判斷或先驗資訊、納入最終評選的決策單位可能僅保有原始決策單位的部份成員、學習標竿的績效表現可能在其中一個構面被所對應的無效率決策單位凌駕。 本研究提出雙構面資料包絡分析增強傳統資料包絡分析的標竿評選能力，並且克服近來標竿方法的缺點。此外，雙構面資料包絡分析被修訂成延伸型雙構面資料包絡分析，在具有負面產出的情況下進行標竿選取。雙構面資料包絡分析與延伸型雙構面資料包絡分析被各別應用到二個實際案例論證實務上的貢獻。第一個案例為美國麻薩諸塞州護理之家，考慮產出項指標取自於「營運構面」與「品質構面」，此例論證雙構面資料包絡分析如何客觀地找尋最適學習標竿，亦即學習標竿的績效表現不會在營運構面或品質構面被所對應的無效率決策單位凌駕。第二個案例為臺灣垃圾焚化爐，考慮產出項包含了「正面產出(電力生產)」與「負面產出(空氣污染)」，此例論證延伸型雙構面資料包絡分析如何提供市府固態垃圾焚化爐二個有關能源與環境政策制定的替選方案。 Data envelopment analysis (DEA), a linear-programming-based methodology, has been recognized as a useful benchmarking tool for circumstances in which a set of homogeneous decision making units (DMUs) uses multiple inputs to produce multiple outputs that need to be assessed to identify benchmarks. However, DEA output measures are frequently considered as derived from two constructs, e.g., operating and quality constructs or operating and profitability constructs. When two-construct output measures are directly mixed together to execute DEA for benchmark selection, DMUs with low output measures of either construct may be identified as benchmark DMUs. More specifically, the performance of benchmark DMUs may be dominated by that of the corresponding inefficient DMUs under either construct. Despite their development in recent years to enhance the capability of benchmark selection of traditional DEA, various benchmarking methods have some of the following shortcomings: solution procedures involve value judgments or a prior information; DMUs remaining in the final evaluation may be some of the original ones; and performance of benchmark DMUs may be dominated by that of the corresponding inefficient DMUs under either construct. This dissertation proposes a Two-Construct DEA (TCDEA) to enhance the capability of benchmark selection of traditional DEA as well as overcome the shortcomings of recently developed benchmarking methods. Moreover, TCDEA is modified as an Extended-TCDEA (E-TCDEA) for benchmark selection in the presence of undesirable outputs. The proposed TCDEA and E-TCDEA are applied, respectively, to two real case examples in order to demonstrate its practical contributions. The first case example is nursing homes in Massachusetts of the United States where the output measures are derived from operating and quality constructs. This case example demonstrates how TCDEA is applied to objectively seek the most appropriate benchmark DMUs whose performance is not dominated by the corresponding inefficient DMUs under either an operating or quality construct. The second case example is municipal solid waste incinerators (MSWIs) in Taiwan, where the output measures include desirable (electricity generation) and undesirable (air pollution) items. This case example demonstrates how E-TCDEA provides two alternatives for energy and environmental policy making for MSWIs.
