民意調查中效果分析的展示與詮釋：以民主文教基金會「美中台」關係民調為例

Using “Predictive Differences in Probability” to Present and Interpret Effect Analyses in Public Opinion Polls

在民意調查的教學與實務研究中，經常需要處理的一個問題是如何衡量某個自變數對於依變數的效果，特別是當依變數為一個類別變數（例如：投票對象或政黨支持），如何展示與詮釋不同自變數（例如：不同議題）的影響效果，是民意調查實務研究中一個非常重要的課題。本文介紹「預測機率差異變化」此一概念，並運用實際民調資料進行不同議題論述對於支持國民黨的效果分析，期能協助民調實務工作者找到影響臺灣民眾政治支持的關鍵議題。

In Taiwan, analyses of public opinion polls predominantly rely on basic frequency distributions and cross-tabulations, which limit their capacity to yield strategic insights applicable to political or electoral decision-making. Traditional methodologies often inadequately identify, in comparison to other issues, which topics most effectively enhance public satisfaction with a specific political stance or bolster support for particular candidates or parties. Furthermore, these methods fall short in evaluating which discourse strategies, when amplified, are most likely to reinforce positive political support. This study introduces the concept of predictive differences in probability as an analytical tool to assess the potential mobilizing effects of policy discourse and electoral issues on various forms of political support, including party identification, vote choice, and policy approval. By employing this approach, the study seeks to address the analytical shortcomings of conventional polling reports and provide methodological insights for both practitioners and scholars engaged in public opinion research.