蘋果比橘子：應用貝氏統計校正消費者與產品異質性

Abstract

在資料富饒的時代下，超量的資訊促使消費者對於他人評價（口碑）的依賴性大幅提高，藉以快速過濾數以萬計的產品資訊並從中獲得符合需求的產品或服務。然而，電子口碑中的評價不僅包含消費者填答習慣，更重要的是每項商品的評價是由不同人按其自身評估標準組合而成的。若直接將不同評價者評價作為比較基礎，就好比拿蘋果與橘子兩種本質上為截然不同的產品做比較。此種不客觀的產品評比結果，將導致決策的錯誤。本研究提出產品與消費者異質性評分調整模型及尺度使用異質性調整模型，針對多產品多問項之評分情況，發展出一個完整的統計模型，以解決（1）評價者之間的差異所造成非標準化評價效應；（2）產品特性的差異對於評價者評價之影響；（3）評價者的個人特質對多產品評分時所造成的差異；及（4）評價者利用多項式問卷對多項產品進行評分時，尺度使用異質性所造成的影響。本文應用台灣某大學教師評鑑資料庫進行實證研究，以瞭解學生（評價者）在進行教師教學成效評比時，將前述四項特性對於評價得分之影響與以消弭。研究結果顯示，經本模型調整後之教師評鑑名次與調整前名次呈現顯著差異。如何客觀的呈現每一位教師的評鑑值，才是能夠真實反應教師貢獻的衡量機制。本文最後提出相關之管理意涵及未來研究限制建議，希冀對實務應用及後續相關研究有所助益。

The pervasiveness of reviewers rating information (electronic WOM) allows decision-makers to quickly screen out abundant products/services options and identify the desired alternatives. A major drawback of employing these scores to make a comparison is that they are not evaluated by the same group of reviewers with identical evaluation criteria. This is because the observed scores consist of unobserved individual preference for various product features, scale usage behavior, and, more importantly, the heterogeneous assessment criteria. Scores that fail to distinguish these effects might give rise to a dilemma of comparing apples and oranges for decision makers. This study proposed a rating score adjustment model utilizing the Bayesian statistical approach that controls for the heterogeneous scale usage pattern of individual reviewers in cross-products/services comparison, as well as the characteristic of products/services per se. The model was empirically tested using a student evaluation of instruction (SEI) dataset from a university in Taiwan. The empirical results show a significant difference in the ranking of instructors’ performance before and after the adjustment. Managerial implications and future research suggestions are discussed.