本文旨在從理性選擇的觀點，以2020年高雄市長罷免案為例，討論選民在罷免案中的投票行為。在理論模型和實證分析整合架構（EITM）下，本研究從工具性與消費性動機的觀點，解析罷免投票行為是否會因為選舉與罷免制度的差異而有所不同。在加總可分與有序層次偏好兩種不同的實證設計下，利用政治大學選舉研究中心提供的電訪調查資料，透過機率單元模型與中介模型，驗證選民支持或不支持罷免案的假設。研究結果發現選民的政黨認同與對被罷免人的綜合評價有其影響力，而且綜合評價是個中介變數。政黨認同除了具有影響罷免投票行為的直接效果之外，還會透過綜合評價間接影響選民支不支持罷免案。又，選民在罷韓案中具有只問藍綠的特性，表示政黨動員支持者對罷免案的成敗具有影響力。同時，選民透過綜合評價決定罷免投票行為的效果顯著，代表掌握選民綜合評價的資訊，政黨才能真正判斷罷免案成敗的可能性。另外，政黨認同的直接效果小、間接效果大，選民戴著有色眼鏡評價被罷免人的中介比例已經超過一半。無論加總可分模型還是有序層次偏好模型均顯示，政黨動員黨員和掌握資訊的能力是推動罷免案的關鍵因素，且直接效果與間接效果的相對影響力大小值得關注。

This study aims to analyze voter behavior in the 2020 Kaohsiung mayoral recall election from the perspective of rational choice. Using an integrated framework of the theoretical models and empirical analysis (EITM), this study examines whether voter behavior varies due to differences in the electoral and recall systems, taking into account instrumental and consumptive motivations. By applying the settings of additively separable utility and lexicographic preferences, probit model and mediation models are used to analyze whether voters voted to support the recall or not. Data used is provided by the Election Study Center of National Chengchi University. Empirical analysis indicates that party identification and the evaluation of mayoral performance are statistically significant in explaining voter's decisions. In addition, the evaluation of mayoral performance is a mediator as well. Party identification not only exerts a direct effect on recall voting behavior but also indirectly influences whether voters support or oppose the recall through their evaluation. Furthermore, the study highlights that voters primarily consider party affiliation, indicating that party mobilization plays a significant role in the success or failure of recall efforts. Moreover, the effect of comprehensive evaluation on recall voting behavior is substantial, highlighting that political parties can accurately assess the likelihood of recall success only when they possess information about voters' evaluations. Additionally, the direct effect of party identification is relatively small, while the indirect effect is substantial, indicating that more than half of the voter evaluations is mediated by partisan biases. The results of both empirical models suggest that an effective party mobilization and a compressive understanding of public opinion are keys to the success of a political party in a recall election, and the relative magnitudes of their direct and indirect effects warrant attention.