選民在歷屆的選舉中，究竞是傾向於把票投給同一政黨的候選 人，還是把票投給不同政黨的候選人，不僅攸關個別政黨與候選人的 選舉成敗，而且還影響到政黨之間勢力的起伏消長，甚至會牵動政黨 體系的整體變遷，其重要性，不言而喻。也正因如此，政治學者亟思 理出歷屆選舉各黨勢力消長的軌跡，描述並説明選民投票的穩定與變 遷（electoral stability and change )。儘管有關選民投票變動的研究 已經卷帙浩繁，然而其方向與幅度究竟應如何估算、分析，在學界卻 仍無定見。本文的目的，是將方法學中研究「常與變」的一般原则應用到 「投票穩定與變遷」這個重要的主題，釐清總變量（gross change ) 與淨變量（net change )的差異，整理出文獻中使用的幾種資料形態 與分析方法、比較其優缺點。由於定群追蹤的個體資料（panel data)可兼顧總變量與淨變量之估計，是很理想的數據資料型態，而 爲了彰顯此一特色，在分析方法上則又以「間斷時間暨間斷空間之馬 可夫鍊模型 J ( discrete-time discrete-space Markov chain models )最 適合，因爲其「移轉機率」（transition probabilities )的參數和「固 票、挖票、跑票」等耳熟能詳的選舉策略語彙、以及「選票穩定度、 選票流入或流失之變遷率J等學理概念相當契合，既能捕捉類別變數 .隨著間隔的時間點前後相依、與時推移的變化軌跡，又能進而同時估 計總變量與淨變量的多寡，更能進一步以母群的異質性説明其動態演變模式。本文最後以日本選舉研究的三波定群追縱民調爲基礎，舉例 説明如何應用馬可夫鍊模型分析自民黨在1993、1996、2000年三次衆 院選舉中選票之穩定與變遷。

How voters cast their votes in successive elections determines not only the fate of candidates but also the rise and fall of political parties and sometimes even causes party system changes. The subject of electoral stability and change, due to its significance in theory and practice, has long attracted the attention of political scientists around the world. Despite the voluminous publications cumulated so far, however, there are still heated debates regarding how best to model this dynamic electoral process and to estimate the amount of changes. The purpose of this paper is two-folds. First, it clarifies some confusion in the literature caused by its failing to distinguish gross change from net change and to recognize the strengths and weaknesses of various types of data in evaluating these two changes. After pointing out how panel data prevail over repeated cross-sections and aggregate, data in estimating both forms of changes, we then proceed to identify a statistical model that best fits the categorical measurement of electoral changes dominant in panel surveys. The second part of this paper, therefore, pinpoints discrete- time discrete-state Markov chain models as ideal tools for describing the dynamic electoral process and further analyzing the sources of change patterns. The transition probabilities of Markov models coincide with the theoretical concepts of flow-of-the-votes and reflect the way state dependence shapes the three-waves of electoral changes. Finally, we apply a mixed Markov model to the three-wave Japanese Election Study (JES) panel data set to illustrate the potential of this technique.