本文以「台灣選舉與民主化調查」（Taiwan’s Election and Democ-ratization Study；以下簡稱TEDS）2001、2003及2004P的再測信度訪問為分析對象，運用相關統計方法來探討TEDS問卷各題組間的信度為何？歷年的訪問是否有所變化？並進一步探討影響信度檢測的因素。研究結果顯示，TEDS問卷中的不同題組其再測信度雖有高低之別，但各題組的信度除了少數題目外，皆能保持一定的水準，其中，民眾對事實題型的「投票意向」其前後兩次回答的一致性程度最高；其次是屬於民眾心理依附感的「政黨認同」；再者則是台灣主要政治競爭議題的「統獨立場」及「族群認同」；民眾對於「總統候選人形象」的認知穩定性也不錯；至於前後兩次回答穩定性相對較低的，則屬民眾對民主政治的內心態度及評價，其中包含「台灣民主實行滿意度」、「施政比較」與「民主治理」。而各題組間的再測信度結果並不會隨著年度的不同而有顯著差異。 另外，在影響再測信度的因素方面，受訪者個人特質中的「教育程度」其影響力相對最顯著，教育程度愈高的民眾其兩次回答的穩定性愈高，反之，教育程度較低的民眾，則容易在兩次訪問中改變其態度；此外，在2001及2004年的結果顯示，男性民眾比女性民眾在兩次訪問回答的態度較不穩定。其次，在兩次訪問的間隔天數上，2001及2003年皆出現顯著的影響力，間隔天數愈短的受訪者其回答穩定性愈高，不過，歷年TEDS對於再測信度訪問所間隔的時間差距皆相當大，故對於執行再測信度訪問的「集中抽樣」方式應可重新考量。最後，兩次訪問是否為同一位訪員，並不會影響再測信度的結果，顯見TEDS訪員皆能謹守「標準化」的訪問方式，既使前後兩次由不同訪員來進行訪問，也不會影響受訪者的回答態度。

The major purpose of this article is to analyze the test-retest reliability in the series of Taiwan's Election and Democratization Study (2001, 2003 and 2004P), and explore what factors influencing the questionnaire reliabi-lity. This research finds that TEDS has consistent questionnaire reliability although the degree of reliability varies with subjects. The "voting intention" has the highest reliability, followed by "party identification" . Atti-tude for the "position of independence/unification" is the third, ahead of the "Taiwanese/Chinese identity" and the "presidential candidate im-age" . Attitudes about democratic evaluation, including the "degree of sat-isfaction about the practice of democracy in Taiwan" and "comparing the DPP vs. KMT government" and "democratic governing" are the least re-liable measurements. It is also found that there is no significant difference in reliability regarding the same subject across different datasets of TEDS. The respondent's education is the important factor of questionnaire re-liability. The high-educated respondent has more consistent response for the same question in different interview than others. Besides, female respon-dents has more stable political attitude than male in TEDS 2001 and TEDS 2004P. The shorter the period between the first interview to the second one, the more stable attitude respondents show, especially in TEDS 2001 and TEDS 2003. Last, no matter who the interviewer is, test-retest reliability is the same. In other words, the TEDS interviewers obey "standard interview principle" during the surveys.