Please use this identifier to cite or link to this item: https://ah.nccu.edu.tw/handle/140.119/114339


Title: 探索選民的投票行爲變化:應用機率分配模型的預測方法
Change in Voting Behaviour: Applying an Election Forecasting Model of Probability Distributions to Modify the Accuracy of Poll Outcomes
Authors: 張順全
莊文忠
Chang, Shun-Chuan
Keywords: 選舉預測;機率分配模型;民意調查;投票行爲;貝氏統計
Public Polling;Voting Behavior;Election Forecasting;Mixture Distribution;Bayesian Statistics
Date: 2008-11
Issue Date: 2017-11-02 16:55:24 (UTC+8)
Abstract: 過去許多有關選舉預測的研究發現,選舉民調中的贏家最後卻未能勝出,箇中隱藏不少值得深入探究的問題。本研究從方法論的觀點,嘗試利用過去選舉經驗資料,改進單純使用民意調查結果所做之選舉預測。就預測方法論而言,本文提出的選舉預測模型乃符合貝氏統計架構,此一架構所欲掌握的是選民投票行爲中所潛藏的異質性,據以探索選民的投票行爲昨日、今日、明日的變化。綜言之,本篇文章的研究所得價值有三:(1)成功應用貝塔-二項機率分配(beta-binomial distribution)模式改進單純使用民意調查結果,發展輔助民調預測模型,及提供作爲評估該次選舉是與過去開票結果不同的「變天型」抑或是「維持型」的預警指標;(2)同時利用此一機率分配模型處理未表態資料,預測選舉民調中未表態受訪者的投票意向;(3)將本研究所建立之模型應用在國內外不同選舉,檢證本模型的適用情境,無論是2004年美國總統大選、2006年台灣高雄市長選舉、抑或是2008年台灣總統選舉的候選人得票率預估,都獲得相當準確的效果。此外,本研究結果也同時指出若干有待思考的問題,說明未來值得繼續發展的方向。
Due to undervotes, misvotes, or switchvotes bias, many polling data users felt frustrated in using the past polling outcome to forecast the new election. It is commonplace for voters to note an early frontrunner in polls will be doomed to fall in the real election outcome. A beta-binominal distribution is suggested to model the accuracy of early poll outcome which strategically influences the polling data users such as political parties, candidates, and mass media in implementing the election campaign. We demonstrate the advantages of probabilistic distribution and Bayesian reasoning, and how to estimate the parameters from past data, in modifying the accuracy of prior poll outcomes. In comparison with the traditional frequency approach, beta-binominal mixture distribution imposes a statistical-adjusting framework with ability to proportionate a coherent mechanism that synthesizes the performances of prior votes. The empirical data sets include the 2004 US presidential election in Atlas Web and TVBS polls in 2006 Kaohsiung mayor election and 2008 presidential election in Taiwan. This paper describes the general fitting of beta-binomial distribution on both datasets and discusses fruitful avenues for future research.
Relation: 選舉研究 , 15(2) , 91-117
Data Type: article
DOI 連結: http://dx.doi.org/10.6612%2ftjes.2008.15.02.91-117
Appears in Collections:[選舉研究 TSSCI] 期刊論文

Files in This Item:

File Description SizeFormat
15(2)(091-117).pdf1821KbAdobe PDF114View/Open


All items in 學術集成 are protected by copyright, with all rights reserved.


社群 sharing