本文使用分類與迴歸樹此項決策樹分析方法,探討影響選舉結果的各項條件與指標變項。首先使用2012年台灣的總統選舉資料,對台灣的選舉結果進行探索式分析。分析結果顯示,除了區域因素之外,族群、鄉鎮市長黨籍與村里長黨籍等因素,皆為影響選舉結果的重要條件性質。其次在選舉預測上,本文運用分類與迴歸樹法能夠依據不同條件因素,對於樣本進行切割的特點,針對不同葉部節點探討其迴歸係數之差異,並依據不同節點的迴歸係數,對於2016年民進黨各村里的總統選舉得票率進行預測分析。分析結果顯示,使用分類與迴歸樹決策法則進行得票率之預測,其預測結果較OLS線性迴歸模型的預測結果更為準確,顯示分類與迴歸樹演算法對於選舉預測有所助益。
This paper aims to investigate the classification effects of election results in Taiwan with classification and regression trees analysis. The classification and regression tree analysis can distinguish the effects of different groups level of independent variables in the classification mechanism. We apply this method to Taiwan's political bloc analysis and find the significant effect of the region, ethnic group, political party, age, education and income. This study takes the political bloc as the research hypothesis to analyze the different group level effects on the election results. Moreover, the effects of the ethnic membership, township mayor and village chief party membership are going to be discussed. The analysis results show that in addition to regional factors, the ethnic membership, township mayor and village chief party membership factors should be important conditions affecting the election results. Finally, this paper uses the classification and regression tree model to cut samples based on different conditional factors, and discusses the difference in regression coefficients for different leaf nodes. The analysis results show that using the classification and regression tree decision rule to predict the voting rate, the prediction results are more accurate than the prediction results of the OLS linear regression model.