由於大多數的調查研究始終無法知道目標母體的參數值，也因此缺少一個作爲比較衡量的基準值。然而透過民意調查的選舉得票率預測則不然，可以透過選前民調與選舉結果的比較來檢驗調查方法的良窳，並作爲改進調查方法的依據。本研究以2005年台北縣長選舉爲例，透過計劃性固定樣本與「選前民調、選舉結果、選後民調」的研究設計，進行剖析研究，實證探討民調中的非抽樣誤差問題，包括未表態、拒訪、困難聯繫等三類型受訪者對選舉預測的干擾與影響。此外，本研究也針對加權變數的使用對民調結果的影響進行分析，同時也探討選前未表態受訪者的投票行爲，以及在選後民調中當選者的得票率會被高估的現象。

In most survey research, it is difficult to evaluate the survey accuracy. The pre-election poll (survey) is unique, since the election result can be used as its target and counting its performance. Indeed, there are hundreds of research improve the survey methodologies through the experimental study from election polls. This study will conduct a panel pre-election poll and post-election poll for Governor of Taipei County 2005. Through comparison of the differences among pie-election poll, election result, and post-election in the designed panel study, three special categories of respondents: undecided, hard-to reach, and refusals and their corresponding election behaviors are investigated. Besides, this paper is focused on the following two issues: boa non-sampling errors might affect the poll accuracy, and how the use of weighting variables might affect the poll results.