資料包絡分析法(Data Envelopment Analysis,DEA)與主成份分析(Principal Component Analysis,PCA)分別為經濟指標與多變量統計分析應用於排名評估之方法,此兩種方法皆透過權重加權所產生的績效分數來做排名。DEA利用每個決策單位元(Decision Making Unit,DMU)的投入(Input)與產出(Output)來評估績效並產生一組對自己最有利的權重,而PCA利用整體觀點給予所有受評單位的各變數一權重值。 本研究將探討此兩種方法的各種構面,以應用於績效評估。首先,將針對權重產生方式的不同,來比較此兩種方法排名是否具有相似處。研究結果顯示,以DEA與PCA來排名之結果並不相同,尤其是當有效率DMU過多時,其排名之差異更顯著。其次,本研究利用PCA來分析DEA的權重,經由此方法能夠了解各DMU的權重特性,同時可透過群集分析了解權重之群聚性。最後,本研究將使用主成份分析法,將DEA資料,在特定比例之成份值下,計算出DEA分數,以提高鑑別度。
Data envelopment analysis(DEA)and principal component analysis(PCA)are economic index and multivariate statistical analysis and are applied to performance measurement. The two methods evaluate a decision-making unit (DMU) by the obtained weights. The weights found with DEA by maximizing the score of the target DMU. In PCA, weights are given by considering all observations instead of single DMU. This study investigates DEA and PCA for performance assessment. Firstly, we compare the ranking results of DEA and PCA. The results reveal that the ranking of DEA is not similar to that of PCA. Especially when there are too many efficient DMUs found in DEA, the difference in ranking between DEA and PCA is considerable. Secondly, this study analyzes the weights of DEA by using PCA and interprets the characteristic of each DMU. Moreover, the grouping of DMUs is explored by clustering analysis. Finally, this study improves the discrimination of DEA with PCA under certain level of explaining variation in the original dataset.