選舉謂查(預測得票率)是一種很特別的調查，因為可以有最終 的選舉結果作為計算調查準確度的目標值，而藉自評價、比較謂查機 構在歷屆選舉預湖中的表現，得以便調查技術有持續的進步，不僅有 助能提高選舉預瀾的品質水準，改進後的調查技術更可廣為其它領域 所應用。然而由於缺乏一套公正、客觀、精確的評量方法，圍內民輯 工作者與學者都很少在選後針對如何評價選舉預測表現的問題進行探 討。 近年來在總統與北高市長陸續改由人民直選後，民眾對選舉的關 心與參與已明顯提高，學術界、媒體、調查專業機構更是紛輯在選舉 期間投入選舉頭測，因此我們有足詞的民謂樣本來探討準確度評量的 問題。然而由於在台灣2000 年的總統選舉中主要候選人有三位，且大 多數民調中未決定選民的比例都超過25%' 因此Mitofsky (1 998) 用 以評量1996 美國總統大選民詞，以及Traugott (2001) 用以評量2000美團總統大選民謂的方法並不適用評量台灣2000 年選舉的民調，因此 本研究將提出修正的方法。此外，本研究也將比較不同民調準確度評 量法間的一致性與相關性，以及探討影響調查準確度的七個因子(調 查頻率、未決定受訪者比例、民謂出資機構、民調委託與否、調查作 業時閻長度、樣本數、與選舉目距離)對台灣2000 年總統選舉民調的 「平均誤差值J 與「對個別候選人高估(或低估)值」的影響。

The majority of surveys often cannot be evaluated due to a lack of actual feedback. However, pre-election polls offer a unique advantage that it can be examined by the final outcome of the election. Through measuring the margin of error from the final vote tally, we can improve the survey methodologies in a wide area of application. N evertheless, few research has focused on assessing Taiwan's pre-election poll performance in the past. In Taiwan, people are becoming more and more interested in and concerned with election campaigns. Before every important election, journalists, pollsters, and political commissions perform dozens of pre-election polls. Therefore, we can have sample size large enough to make quantity analysis now. In this paper, we collected poll results for the 2000 election campaign of Taiwan, R.O.C, where the data was restricted to those groups that have polls published in public. After modified Mitofsky's (1996) and Traugott's (2001) polling accuracy methods and reallocated the “ undecided" in polls, the modified method was adopted for evaluating Taiwan's 2000 poll performances. Furthermore, some statistical techniques were used for analyzing seven influential factors, including survey frequency, undecided rate, recommenda tion organization, conducting status, field days, effective sample size, and days to election in this paper.