制衡觀之所以重要，不僅因為制衡觀是民主社會的重要價值，也是 解釋分裂投票的重要因素。2016 年總統與立委選舉合併進行，隨著小 黨林立、候選人參選爆炸，媒體認為這次是中央層級選舉分裂投票最激 烈的一次。在討論制衡觀與分裂投票的關係時，不能忽略政黨偏好同時 影響這兩個變數所帶來的內生性問題。因此，本研究藉由TEDS2016 的 定群追蹤資料，重新分類民眾的制衡觀，並應用推廣之結構方程式模型 （GSEM）方法應對內生性問題，驗證制衡觀與分裂投票的關係。 分析結果顯示，民眾的制衡觀的確顯著受到本身政黨偏好所影響， 民進黨偏好者在過去比較支持制衡，但隨著民進黨執政，他們轉而不支 持制衡；國民黨偏好者則剛好相反，在過去不支持制衡，但是隨國民黨 敗選，他們轉而支持制衡。在投票決定方面，制衡觀雖然顯著影響民眾 採一致投票或分裂投票，但是有較高機率採取民進黨一致投票的卻是轉 向不支持制衡觀與穩定支持制衡觀的民眾，而這兩群民眾都是民進黨偏 好者為主的族群。這意味著民眾制衡觀對投票行為的影響，反映出來的 仍是他們本身的政黨偏好。上述結果顯示以制衡觀解釋分裂投票行為的 確存在內生性問題，凸顯GSEM 方法的必要性。本研究認為未來研究 在驗證制衡觀與分裂投票之間的關係時，仍需注意政黨偏好所產生的影 響。

Cognitive madisoniansim is crucial in political situations. It is not only an important value of democratic societies, but also a factor in explaining split-ticket voting. With the increase of minor parties and candidates, the media believe that Taiwan’s 2016 general elections have shown the most fierce split-ticket voting. It is worth mentioning that we shall not ignore the issue of endogeneity caused by partisanship when discussing the relationship between cognitive madisoniansim and split-ticket voting. Based on the panel data of TEDS2016, this study aims to recategorize the cognitive madisoniansim of the respondents and resolve the issue of endogeneity by applying a generalized structural equation model (GSEM). By doing so, we aim to examine the relationship between cognitive madisoniansim and splitticket voting. The findings show that the public’s cognitive madisoniansim was indeed affected by party preference. DPP supporters have tended to support cognitive madisoniansim in the past. However, they stopped supporting it once the DDP took over the government. The KMT showed the opposite situation. They had been against cognitive madisoniansim in the past. When they began losing elections, they started to support it. Regarding voting decisions, cognitive madsoniansim has positive effects on people’s decisions about straight-ticket voting or split-ticket voting. Nevertheless, most voters who cast straight-ticket voting for the DPP are those who stopped supporting or constantly supported cognitive madisoniansim. These two groups of voters both prefer the DDP. This result indicates that the effect of voters’ cognitive madisoniansim on their voting behaviors still reflects their party preference. The above-mentioned issues present the endogeneity issue derived by explaining the split-ticket voting behaviors by cognitive madisoniansim and the inevitability of GSEM methods. We suggest that researchers not ignore the effect of party preference as they examine the relationship between cognitive madisoniansim and split-ticket voting.