過去對於政策與投票的研究偏重個別縣市的影響因素，並未考慮縣市之間的差異，推論範圍有限。我們利用台灣選舉與民主化 (TEDS 2010C) 的調查資料，合併來自台北市、台中縣、台中市、高雄縣、高雄市等五個縣市的受訪者，而且運用五都改制前的決算資料，以多層次貝氏統計模型估計縣市之間的差異以及個人的政策滿意度影響現任者支持的程度。結果發現在考慮政黨認同之後，社福、交通、環保等三項政策的施政滿意度顯著地影響投票行為，而且縣市之間的確存在差異，但是這三項政策的預算並未影響縣市之間的差異。我們也比較多層次模型的估計結果與勝算對數迴歸模型以確認結果的可信度。

Democratic governments should seek for the best interest for citizens. In theory, incumbents should persuade voters to support them again with their performance, including budget distribution. It is important to evaluate the extent to which voters respond to government performance and re-elect the incumbents. This paper uses data from Taiwan’s Election and Democratization Studies (TEDS 2010C) and set up five clusters of respondents according to their residence in five counties or cities. Utilizing multi-level Bayesian analysis, I combine aggregate-level budget data of social welfare, transportation, and environmental protection, and corresponding survey data to test the hypothesis of individual and geographical influence on voting behavior. The results show that individual’s policy evaluation significantly influences voting behavior, while partisanship is taken into account. But government spending seems to have no impact on the probability of supporting the incumbents. A validation check of the estimates is provided.