選舉民調與選戰動態分析---貝式統計模型的應用

Abstract

本研究以 2008 年總統大選前近三個月內各媒體所發佈的民調數據為基礎，利用貝式統計模型分析時間序列，以評估媒體民調包括TVBS/年代、中國時報、遠見、聯合報、及蘋果日報等的機構效應—即是否媒體民調結果有所謂偏藍偏綠的傾向。實証研究發現，並非每個媒體機構所發佈的民調結果都有所謂「機構效應」存在。此外，透過時間序列模型及Kalman Filtering and Smoothing Algorithm的應用，本研究整合了過去的選舉結果、實際的選舉結果、及即時的民調資料來描繪馬英九的兩黨支持度在競選期間每日的變化。

This analysis uses Bayesian modeling approach to assess time series of media’s poll data and to explore the so-called “house effect”. Specifically, we collected data three months prior to the 2008 Presidential Election from five media institutions—namely, TVBS/ERA, China Times, Global Vision Monthly, United Daily News, and Apple Daily to analyze whether their poll results biased toward Pan-blue or Pan-green camps. Our findings are twofold: first, while some media poll results have a consistent “house effect”, some do not have such systematic bias. Second, by utilizing Kalman Filtering and Smoothing Algorithm and combining past and recent election results as well as updated poll numbers, this analysis successfully depict day-to-day changes of Ma ying-jeou’s two-party vote during the campaign period.