本文從測量理論與潛在變數模型的角度出發，說明測量誤差的不確 定性及其在迴歸分析中可能產生的影響，並且以廣義潛在變數模型與貝 氏統計將各種不確定性的來源納入模型的估計中。本文應用上述模型分 析2013 年「台灣選舉與民主化調查」的面訪資料，同時呈現統獨立場 的測量結果以及統獨立場對政黨偏好的影響。分析結果顯示台灣選民的 統獨立場大致上可以區分為「強烈傾向獨立」、「有條件接受獨立」以 及「有條件接受統一」三種類型。至於在統獨立場對政黨偏好的影響方 面，本文的分析結果與目前大多數的研究結果一致，即越傾向獨立（統 一）的選民有越高的可能性偏向泛綠（藍）政黨。本文認為雖然測量誤 差可能因為經驗資料的性質而不必然會在每次的分析中造成嚴重的影 響，但在研究者無法確定量表的測量誤差或變異程度是否足以改變統計 分析結果的情況下，最謹慎的作法是在評估這些潛在變數的影響效果 時，將此量表的不確定性一併考慮在內。

In this article, we focus on the measurement of Taiwanese attitudes on the independence-unification issue, and argue that, when analysts are not certain about how measurement errors influence the results of analysis, the best way is to take measurement errors into account in their analyses. Based on the methodology of generalized latent variable modeling, we treat Taiwanese attitudes on the independence-unification issue as unobserved, latent variables, which are measured by several manifest indicators, and evaluate the effect of these attitudes on party identification. Analyzing survey data from the Taiwan’s Election and Democratization Study (TEDS) project conducted in 2013, we show that, first, Taiwanese attitudes on the independence-unification issue includes three types: principled supporters of independence, conditional supporters of independence, and conditional supporters of unification. Second, as shown in most of the studies on Taiwan politics, the stronger the voters prefer unification, the more likely they lean toward pan-blue parties, and vice versa.