調查方法與選舉預測：混合調查模式之趨勢與理由

Abstract

隨著科技與社會變遷，民意調查已成為選舉預測的重要工具，尤其我國大型選舉頻繁，過去以傳統市內電話調查為主要的調查方法，然而手機普及化後逐漸面臨涵蓋率下降、訪問成功率降低及樣本代表性不足等挑戰，選舉民調預測失準頻繁發生。目前國際上對於民意調查方法的趨勢，已由單一的模式趨向於採用多種混合調查模式，在COVID-19大流行期間為減少群聚，以便繼續完成調查工作。 國內文獻對於選舉預測的調查方法研究，多半聚焦在傳統市內電話調查相關的分析與建議，近年來陸續雖增加手機樣本的雙底冊應用探討，但是討論的範圍仍是以單一調查方法為主。因此，本研究以2020年總統選舉、2022年縣市長選舉及2024年總統選舉等民調數據個案，並採取次級資料分析法，驗證不同調查方法的自變項對於選舉預測準確度的依變項影響，同時也進一步比較不同調查方法取得的成功樣本結構與母體差異。 研究結果顯示，傳統市內電話調查的誤差較為明顯，且成功樣本結構年齡偏高及教育程度偏低，即便採用市內電話及手機雙底冊調查，其成功樣本結構仍有年齡及男性比例偏高的問題。相比之下，整合市內電話及網路調查的混合調查模式，不僅能夠有效提升涵蓋率與回應率，由於透過混合調查模式所獲得的加權前成功樣本，其人口分佈更趨近於母體特徵，進而改善選舉預測的準確度。 本研究的貢獻在於首次藉由跨選舉個案分析歸納單一調查方法與混合調查模式的準確度比較，並提供實證資料支持混合調查模式應用於選舉預測的趨勢與理由。最後，本研究建議未來民調機構可進一步發展混合調查模式，以提升選舉預測的準確度，並且公開加權前的原始數據與資料處理步驟，增強社會對民意調查結果的信賴。

With technological advancements and social changes, opinion surveys have become a crucial tool for election forecasting, particularly in Taiwan, where large-scale elections are held frequently. Traditionally, telephone surveys served as the primary method for conducting opinion polls. However, with the widespread adoption of mobile phones, traditional telephone surveys have encountered increasing challenges, including declining coverage, lower response rates, and the underrepresentation of key demographic groups. These issues have frequently resulted in inaccuracies in election polling forecasts. In response, the global trend in survey methodology has shifted from single-mode to mixed-mode approaches, a transition that accelerated during the COVID-19 pandemic to minimize in-person contact while maintaining data collection efforts. Most domestic studies on election forecasting have focused on analyzing and improving traditional telephone surveys. While the inclusion of Dual-Frame sampling—incorporating both landline and mobile phone respondents—has gained traction in recent years, discussions have largely remained centered on single-mode methodologies. To address this gap, this study examines polling data from Taiwan’s 2020 and 2024 presidential elections and the 2022 special municipality and county (city) mayoral elections. Using secondary data analysis, this research evaluates how different survey methodologies impact the accuracy of election forecasts. Furthermore, it compares the demographic composition of successful survey samples obtained through different polling methods, highlighting variations in representation. This study makes a significant contribution by being the first to systematically compare the accuracy of single-mode and mixed-mode surveys across multiple elections. It provides empirical evidence supporting the shift toward mixed-mode approaches in election forecasting. Finally, this study recommends that polling organizations further develop mixed-mode survey techniques to enhance the accuracy of election predictions. Additionally, it advocates for increased transparency by making raw data and pre-weighting processing steps publicly available, thereby fostering greater public trust in opinion poll results.