選制改革對投票穩定與變遷的影響：臺灣五次立委選舉的實證分析

Impact of Electoral Reforms on Stability and Change in Voting Behavior: Elections to the Legislative Yuan, 1998-2012

本文利用「動態追蹤資料模型」(dynamic panel data model)中的「兩階段最小平方法」(two-stage least squares)，檢證影響臺灣投票穩定與變遷的各項因素；其中，本文特別關注，2008年第七屆立法委員選舉制度改革，對於投票行為的衝擊。觀測時間為1998年至2012年五次立委選舉，研究對象為全臺364個鄉鎮市區，樣本數為1,820筆觀察值。資料來源有二：一是中央選舉委員會建置的選舉資料庫；二是各縣市政府出版的《統計年報》。實證結果顯示，選舉制度改革、選舉制度類型、性別、教育，以及公職人員比例等五個變數，是影響「投票變動指數」(vote volatility index)的重要因素。此外，總體經濟環境、投票習慣與地理區位，也是關鍵影響因素。整體而言，本文的量化實證結果，支持「杜弗傑法則」(Duverger's Law)與「自我實現的預言」(selffulfilling prophecy)，即選制改革對投票穩定與變遷，具有長期且深遠的影響。本文的學術貢獻有三：第一，本文論證選舉制度、政黨體制，以及投票穩定與變遷三者間的互動關係。第二，測量臺灣立委選舉的投票穩定與變化度，可供政治實際運作參考。第三，迥異於既有文獻採用「質化與受限依變數模型」(qualitative and limited dependent variable model)，本文引入動態追蹤資料模型，檢證選制改革對投票穩定與變遷的具體效應。換言之，本文的研究發現可作為選舉研究的補充。

This study uses the method of two-stage least squares of dynamic panel data models to examine the variables closely related to electoral stability and change in Taiwan, especially focusing on the impact of electoral reform of the Legislative Yuan in 2008 on voting behavior. To account for causal effects, the analysis includes 1,820 observations consisted of the 364 township and village-level units of five legislative elections from 1998 to 2012. The data sources come from the archive of election outcomes officially released by the Central Election Commission, and the statistical yearbooks published by county and city governments. The findings reveal that the variables of electoral reforms, types of electoral system, gender ratio, the level of educational attainment, and proportion of civil servants present statistically significant associations with the index of vote volatility. In addition, the factors of macroeconomic conditions, voting patterns, and geographical contexts emerge as statistically significant and in the anticipated directions. The empirical results demonstrate the Duverger's law and also the self-fulfilling prophecy; i.e., the reforms of electoral system exert a profound effect of the stability and change of voting behavior. There are at least three academic implications that can be drawn from the outcome of this study. First, this study verifies the relationship between electoral system, party system, and electoral stability and change. Second, it empirically measures the vote volatility index which might be valuable for campaign strategies. Last, different from the qualitative and limited dependent variable models in the previous studies of political participation, this study takes advantage of a dynamic panel data model to assess the impact of electoral reforms on stability and change in voting behavior. Developing an electoral model that is both concise and accurate awaits future research.