醫院平衡計分卡之單向度與多閾值分祈

Unidimensionality and Threshold Analysis of Balanced Scorecard Implemented In Hospitals

許多醫院在推動平衡計分卡，每月如何有效率地挑出異常的指標列入檢討與追蹤，是本研究之目的。傳統上以指標達成率為依歸，甚難列出適合各月及各指標屬性的客觀標準。本研究利用試題反應理論的Rasch分析之(1)簡單二元計分模式(RM)、多元評定量尺(RSM)及部份給分模式(PCM)，分別檢測某個案醫院4年來的顧客面14項指標，並以標準化殘差分析圖示系統來解析客觀且有效率地挑出異常指標列入檢討與追蹤。研究發現，原列入顧客面的抱怨件數（指標1）及上報次數（指標2）不符Rasch模式資料的預期。多點計分會較二元計分之信度來得高，PCM較RSM有較高的擬合度，業務性指標應自顧客面中排除，否則會失真顧客面的意義。作者建議以後的研究可再偵測指標是否存在有季節間的功能差異性，題目多分選項是否合宜也應檢測。電腦化殘差分析的資訊模組可以迅速且客觀地挑出當月指標的異常，未來尚可再進行其他構面的單向度分析及信效度的研究。

The purpose of this study was to help hospitals effectively implement balanced scorecard (BSC) and detect aberrant responded indicators for further discussion and improvement. How to decide objective cut-off values of thresholds is still controversial. The 14 items of customer satisfaction prospective implemented in a specified hospital were examined with the Rasch model (RM), rating scale model (RSM) and partial credit model (PCM) to check whether they constitute a one-dimensional construct and form an interval scored scale. The standardized residual analysis with a computer module was adopted to swiftly identify distorted indicators. The results show that those two items regarding complain frequency and messages in daily newspapers were deemed as mis-fitting to Rasch model's requirement. The reliability derived from polytomous BSC was higher than that from dichotomous one. The PCM had a better fit and a higher reliability than the RSM. Volumes of operation onto customers should be excluded from the prospective of customer satisfaction of BSC to make the dimensional meaningful. It was suggested that practitioners can apply the PCM to fit BSC data and a computer module with the feature of Z-score detection can detect aberrant responses across indicators objectively at a very early stage. Future studies can focus on detecting differential item functioning (DIF) among season groups, investigating scoring rubrics of item responses, and integrating the existing workflows via computer sciences. Analysis of other aspects of BSC is also recommended.