一九九八年台北市市長選舉中，影響選舉結果最大的是外省籍選民及新黨選民「分裂投票」現象。大量新黨選民在市長選舉中一面倒地投票給國民黨候選人馬英九，不過市議員、立法委員仍舊投給新黨則是導致這種現象出現的主因。常態下，擁有高滿意度的在位者，常能連任。而在任內施政滿意度常常保持在七成以上的市長陳水扁落選，是相當令人意外的。筆者認為這種現象值得作有系統的分析。 本文利用政治大學選舉研究中心的抽樣調查面訪資料，針對一九九八年台北市選舉中選民分裂投票的行為進行深入分析。本文歸納影響選民分裂投票的六個模型，利用六個模型經過對數成敗比模型的比較，在控制人口基本背景因素之後（性別、年齡、教育程度及籍貫），發現模型影響選民分裂投票的情況：一是政黨認同強度減弱的政黨解組作用導致選民分裂投票。雖然政黨認同在模型中具有明顯的影響力，但台北市近年來並未有政黨解組的趨勢，因此政黨認同減弱並非分裂投票的主因。二為選民的省籍，尤其是外省籍選民分裂投票行為相當明顯，其中年輕世代比年紀大的世代更形嚴重。第三則是統獨議題及相關的中國人/台灣人認同。研究結果發現統獨議題對選民分裂投票不具顯著影響力，但中國人/台灣人則對於市長與立法委員之間的分裂投票有明顯影響力；第四是候選人形象中的裙帶效應並不顯著，對分裂投票與否之間不具明顯差異。五為選民分立制衡觀念。在相同層級的市府及議會之間選民具有分立制衡的觀念，想藉政黨之間相互制衡的方式保護民眾的最大福祉，故採取分裂投票的行為。中央及地方制衡的觀念則不具影響力。第六則為新黨的選民策略投票是影響選民分裂投票的重要因素。新黨選民對於尋求連任的民進黨市長候選人陳水扁相當反感，且新黨市長候選人王建□當選機會不大的情況下，選擇轉投給國民黨馬英九之策略投票，並進而形成分裂投票。從六個模型最後推論的結果，真正決定選民分裂與否最重要的因素是選民的省籍背景導致的情感反應及新黨選民的策略投票。而這兩者的關聯又非常密切。 / Split ticket voting by New Party identifiers and Mainlanders had a crucial influence on the outcome of the 1998 Taipei City mayoral election. Large numbers of voters who identified themselves as New Party supporters voted for the KMT mayoral candidate. However, in the concurrent Legislative Yuan and City Council elections, they voted for New Party candidates. Normally, an incumbent with extremely high approval ratings will win re-election easily, but in this case Chen Shui-bian lost the mayoral race. This surprising result is worth researching. This thesis uses face-to-face interview data collected after the 1998 election. To analyze split ticket voting behavior, we consider six different logit models. After controlling basic demographic variables, including sex, age, education, and ethnic background, it is possible to discover what influences split ticket voting. First, weakening party identification can induce a dealignment process. This, in turn, gives rise to split ticket voting. In the data, the intensity of party identification has an obvious effect on split ticket voting. However, there has been no dealignment. Thus, weakening party identification is not the main reason for split ticket voting. The second factor is ethnic background. Mainlanders, especially young and middle aged voters, are more likely to split their tickets. Third, the related questions of unification or independence and Taiwanese or Chinese identification also have influence. The effect of the unification/independence position is not significant, but the respondents’ ethnic identity is. A fourth finding is that there is no significant coattail effect influencing split ticket voting. Fifth, many voters wish to balance the parties against each other, and so they split their votes. However, this only holds for offices at the same level (mayor and city council) and not for different levels (local and national). Sixth, strategic voting by New Party supporters was very important. New Party supporters were very opposed to Chen Shui-bian, and the New Party nominee did not have much chance of winning. As a result, many split their votes, voting strategically for the KMT mayoral candidate. After examining the six models, we find that the most important factors influencing split ticket voting are ethnic background and strategic voting by New Party supporters. These two factors are intimately connected, of course.