|
Since the implementation of democratic elections in Taiwan in 1996, opinion polls have played a crucial role in the electoral process starting from the year 2000, with various polling organizations employing scientific sampling methods to provide accurate forecasts. This study explores the application of meta-analysis in election polling, using the 2024 Taiwanese presidential election as a case study to analyze the accuracy of poll data and their predictive power regarding the final election outcome. Meta-analysis aggregates data from multiple polling organizations and employs random-effects models to account for variations between different organizations and the inherent errors in the polls themselves. This study observes the limitations of the final pre-election polls and expands data collection to include poll data after candidates register for the election. The meta-analysis then examines the discrepancy between poll support and actual vote shares. The results indicate that, with Ko Wen-je as the baseline, the support for the Democratic Progressive Party candidate Lai Ching-te was generally underestimated in most polls, with the actual vote share significantly higher than the predicted aggregate results. However, in Taiwan, the presence of a polling blackout period, along with the potential impact of unforeseen events, cross-strait relations, and political maneuvering among parties, could lead to distortions in poll data. The study aims to demonstrate the advantages of meta-analysis in integrating multiple data sources and reducing sampling errors to enhance predictive accuracy. |