我國民意代表選舉所採取的「多席次選區單記非讓渡投票制」，使黨派屬性相近的候選人往往成為彼此主要的競爭對手。候選人間的選票區隔因此成為研究台灣選舉的主要課題之一。然而，這類研究往往需要分析候選人選票在不同地域單位間的分布。由於人與地皆屬類別性變數，且多席次選舉的資料矩陣必然龐大，常用的統計方法大多難以處理這樣的資料。本文主張，「對應分析法」(correspondence Analysis ) 是解決這個問題的妥當工具。此法將任一n×m矩陣視為n個m維向量，並找出一k維次空間以極大化各行矩陣在此k維空間上的投射 ( 以卡方距離計之 ) 的變異量。由於候選人及地域皆可標示為此空間中之向量，我們可以描述候選人間的卡方距離，及候選人與地域的卡方距離。本文運用對應分析法來描述台灣立委選舉中候選人與各地域的卡方距離，並判斷每一候選人之主要地盤。我們並以對應分析空間之慣量 ( inertia ) 及維度 ( dimension ) 來解釋政黨的選舉表現。結果發現，在得票率和維度不變的情況下，執政黨每超額提名一個候選人就可能喪失11.7%的席次。

It is well recognized that intra-party competition marks a major consequence of the single nontransferable vote under multi-member district system(SNTV-MMD), currently used in Taiwan's legislature elections. A key issue in Taiwan's electoral study is thus how candi-dates, especially those from the same party, divide their votes a-mong various categories that differentiate the voters. Since the data set revealing this information is usually a large matrix of categorical variables, existing multivariate statistical methods are either inap-plicable or inadequate. This paper demonstrates that correspond-dence analysis is an appropriate statistical tool to solve this prob-lem. This method treats an n × m matrix as n vectors in an m-di-mensional space, and finds a k-dimensional subspace that maxi-mizes the variance of the projections of the n vectors. Utilizing this method, this paper first illustrates the Chi-square distance between candidates and the sub-district regions. It then designs indexes to describe the degree of vote concentration for each candidate, fac-tion, and party, and correlate it with the electoral performances of these actors. Statistical analysis indicates that, given the same vote share, inertia, and dimensionality, overnominating one candidate decreases the seat share of the KMT by 11.73%.