Logit模型預測2006年高雄市長選舉

Forecast Kaohsiung mayor election by logit model in 2006

黃羿菁(Yi-Ching Huang)

指導教授 ： 張紘炬

淡江大學/管理學院/管理科學研究所碩士班/碩士(2007年)

https://doi.org/10.6846/TKU.2007.00269

摘要

本研究主要是在探討高雄市長選舉，利用Logit模型來預測各候選人的得票率，並以人口統計變數如：行政區、性別、年齡、籍貫、教育程度與政黨傾向為研究變項，探討高雄市各選民區隔在人口統計變數上是否有顯著性。分析結果如下：
1. Logit模型?堙A人口統計變數中行政區、性別、籍貫、教育程度與政黨傾向五項解釋變數與選民投票傾向具有相關性。
2.候選人本身的形象會影響選民的支持，勞工運動出身的陳菊獲得較多女性與非高學歷的選民支持，學者出身的黃俊英在高學歷的選民中獲得的支持較多。
3.省籍確實會影響選情，本省籍對兩位候選人支持程度伯仲之間，大陸各省選民卻明顯支持黃俊英。
4.在樣本結構分析中發現，各個分層中隱性選民的比例甚高，大多在三到四成，預測的準確性受到隱性選民影響。
5.投票的實際結果，陳菊小勝黃俊英，預測結果黃俊英小勝陳菊，其差距來自於電訪中，陳菊的支持者較不願意表態支持以及投票前馬英九特別費案影響選情。

關鍵字

選舉預測 ； Logit模型

並列摘要

This study adopts Logit model to perform data analysis and interpret the results. Logit model is conducted by using the sampling survey method and the study range includes the Kaohsiung city area. We set our study target as the voters of the election for Kaohsiung city magistrate in 2006. we apply Logit model method to forecast the predicted voting rate for every candidate, and apply the population variances to understand the diverse voting situations of different voter groups. Results in this study:
1.In the analysis of Logit model method, five explanatory variables, the administrative area and gender, and party partial, and education, and the province and county of one's family register are apparently distinguished from voting tendency.
2.The image of the candidate itself will influence his/her support from the voters. With the background of labor activity, Chu, Chen gain more supports from female voters and voters with less education background. Chun-ying, Huang , as a scholar, gains more support from voters with higher education background on the other hand.
3.Provincialism can indeed affect the vote. Native voters support the two candidates equally and immigrant voters obviously incline to vote . Chun-ying, Huang
4.According to the sample structure analysis, the hidden ballots in each one of the category take a high percentage at around thirty to forty percent. The accuracy of the prediction will be affect greatly by the hidden ballot
5.The actual result of the vote showed that the vote to Chu, Chen slightly surpass the vote to. Chun-ying, Huang. Nevertheless, the prediction was . Chun-ying, Huang would gain more vote than Chu, Chen. The difference came from the fact the Chu, Chen’s supporters were less willing to show their support before the vote and also some after effect due to the investigation to Ying-jeou, Ma’s special expense case.

並列關鍵字

Election Forecasting ； Logit model

參考文獻

2.俞振華;蔡佳泓(2006)，如何利用全國性民調推估地方民意?多層次貝式定理估計模型與分層加權的應用，台灣政治學刊，10卷1

1.周應龍;盛杏湲(2005)，選樣偏誤模型在選舉預測上的應用，選舉

3.盛治仁(2000)，總統選舉預測探討－以情感溫度計預測未表態選民的應用， 選舉研究，7卷2期，p75-107。

7.張世英，選舉預測方法的實證研究，成功大學統計學研究所碩士論文，民國88年。

5.范凌嘉，台灣縣市長選舉預測模型之研究: 一個基礎模型的建立及其應用，政治大學政治學研究所碩士論文，民國88年。