預測選舉是個普遍存在的興趣，不論基於好奇心的滿足或實用的考慮，選舉預測都是件極具挑戰性的工作。本文以Kelley及Mirer（1974） 所發展的選民決策模式作基礎，略作修訂，以政大選舉研究中心在不同選舉前所做的調查來預測台北市長、台灣省長、及總統選舉的結果，並與選後的資料對照比較。預測的結果有些與真實投票結果有相當的差距，有些則極接近投票的結果，效果並不穩定。改良Kelley與Mirer理論並將之應用於台灣選舉的模型並不成功。在各種嘗試中，有時是最接近Kelley與Mirer原來設計的「選人模型」有比較好的預測力。但「選黨模型」也有表現比較好的時候。這樣的結果也許在告訴我們：預測選舉結果，模式不只一個。而從事不同的選舉之預測，要先從各該選舉的外部環境條件來判斷與使用適當的模式，較能奏功。

Accurate forecast of elections is an exciting challenge for every student of electoral behavior. The aim of this article is to apply a model developed by Kelley and Mirer (1974) to the forecast of elections in Taiwan. Three models -- a "revised Kelley & Mirer model," "Candidate-oriented model," and "party-oriented model" are tested by applying data collected by the Election Study Center at the Chengchi University in three different elections. The tests indicate that the revised model is not as powerful as the author expected. Performance of other two models are not stable either. These results imply that in forecasting election we need not only individual level data, we also need to examine the macro environmental factors before we pick an appropriate model for the job.