專家異質性評估：基於臺灣公共治理主觀指標調查的檢證

Expert Heterogeneity Assessment: Verification Based on the Subjective Indicators Survey of Public Governance in Taiwan

有鑑於許多公共議題具有高度專業性，一般民眾難以熟悉和提供具體的意見或評價，因此，許多國際組織採用的公共治理調查，多依賴菁英或專家來獲取各種評分或建議，我國行政院國家發展委員會自2008年起，委外執行的臺灣公共治理主觀指標調查亦是如此。由文獻檢閱可知，目前仍少有實證研究就專家異質性對調查結果的影響進行評估，使我們對專家調查的測量品質了解相對有限。因之，本研究以2012年和2013年臺灣公共治理指標專家調查結果為例，利用Q-Q圖、蒙地卡羅模擬，以及多種統計檢定等量化分析方法，從方法論的角度探討：1.無反應與趨中選答的專家答題模式；2.隨機專家假設的抽樣模擬和假設檢定；3.不同專業領域的專家異質性比較。
研究結果顯示，適度減少專家的回答量（如七大面向中允許避答三個以內）並不影響推論結果，尤其學界代表的回覆普遍呈現較高同質性。此外，不同參與次數的專家在「無反應」與「中間值」的填答模式上亦展現一致性。本研究不僅有助於理解專家異質性對公共治理評估成效的影響，亦提出專家調查設計的改進建議，包括：針對首次參與者舉辦調查說明會，以協助理解調查目的與釐清較長問卷內容；在受訪者配置上，平衡政府內、外部專家比例，以降低偏誤；適度減少學界專家樣本並增納企業與公民社會的多元觀點，以提升推論結果的包容性；專家填答方式則可透過隨機減少部分必答面向，或依據其專長分配作答，仍可維持整體評估品質之穩定性。

Due to the high level of expertise required in many public issues, the general public often finds it difficult to fully understand these topics and provide concrete evaluations. Consequently, many international organizations rely on elite or expert surveys to collect scoring data and professional opinions for governance assessments. Since 2008, Taiwan’s National Development Council has commissioned the Taiwan Public Governance Indicator Survey, conducted by the Taiwan Public Governance Research Center, for this purpose. However, existing domestic literature reveals a lack of empirical studies assessing the impact of expert heterogeneity on survey outcomes, which constrains our understanding of the measurement quality of expert surveys. This study draws on expert survey data from the 2012 and 2013 Taiwan Public Governance Indicators and applies Q-Q plots, Monte Carlo simulations, and a series of statistical tests to examine three methodological issues: (1) patterns of expert responses involving non-response and central tendency bias, (2) sampling simulations and hypothesis testing under the random expert assumption, and (3) heterogeneity among experts from different professional fields.
The results indicate that moderately reducing expert response requirements (e.g., allowing respondents to skip up to three of the seven major dimensions) does not affect inferential validity, particularly as academic experts tend to provide more homogeneous responses than experts from other fields. Moreover, experts with different participation frequencies over the two years demonstrated consistent patterns in terms of non-response and midpoint selection. Beyond contributing to the understanding of expert heterogeneity in governance evaluations, this study offers practical recommendations for improving the design of expert surveys: organizing briefings for first-time participants to clarify survey objectives and questionnaire content, balancing the ratio of internal government and external experts to reduce bias, limiting highly homogeneous academic samples while incorporating more diverse perspectives from business and civil society, and adopting simplified or specialized response assignments to ensure stability in overall assessment quality.