社會投資報酬率（SROI）之財務代理值既有文獻多聚焦於代理值「選擇」層次之爭議，較少檢視代理值選定後實際計算所得之財務代理指數數列本身，其統計分布特性應如何被客觀檢定與處理；當同一成果存在多筆財務代理觀察值時，採用平均數易受極端值拉動，直接刪除異常值易造成資訊損失，本研究旨在於不否定任何一筆觀察值真實性之前提下，建立能找出更具代表性之財務代理指數代表值的檢定與處理框架。本研究依偏態、峰度與變異係數三項指標判定資料分布情境，並分別建立單年度資料檢定與處理框架與多年度資料檢定與處理框架，以因應實務上組織資料年數不一之限制。以某基金會受益團體資料進行實務驗證，結果顯示原始資料以平均數計算之SROI為6.98，以中位數計算則為1.63，兩者差距懸殊；經本研究單年度與多年度框架處理後，SROI分別為3.14與3.90，兩項結果皆落於原始平均數與中位數之間，初步顯示處理結果並未因框架不同而發散。本研究提出之分布特性檢定與資料處理框架，為SROI財務代理指數之統計穩健性提供了一套可操作且不需否定原始資料真實性之解決方案。;Existing literature on Social Return on Investment (SROI) has predominantly focused on the subjectivity involved in selecting financial proxies， with limited attention paid to how the statistical distribution characteristics of the resulting financial proxy data should be objectively examined and processed. When multiple financial proxy observations exist for a single outcome， using the arithmetic mean as the representative value renders it vulnerable to distortion by extreme values， whereas outright deletion of outliers risks information loss and sample selection bias. Without negating the authenticity of any individual observation， this study aims to establish a diagnostic and data-processing framework capable of identifying a financial proxy representative value that more accurately reflects the typical experience of the majority of samples. Based on skewness， kurtosis， and coefficient of variation， this study develops distinct frameworks for single-year and multi-year financial proxy data， in order to accommodate the varying lengths of data histories across organizations in practice. An empirical case study using beneficiary group data from a foundation′s online public-benefit support program was conducted to validate the frameworks. Results show that the original SROI ratio calculated using the mean was 6.98， compared to 1.63 using the median. After processing through the proposed single-year and multi-year frameworks， the SROI ratios converged to 3.14 and 3.90 respectively， both falling between the original mean and median values， suggesting that the processed results did not diverge substantially despite the different frameworks applied. The proposed distribution diagnostic and data-processing framework offers an operable solution for enhancing the statistical robustness of SROI financial proxy indices without requiring the authenticity of the original data to be negated.
