Exploring the Impacts of Interview Intervals, Process, and Respondent’s Traits on Survey Reliability
受訪者的回答內容是民意調查的核心價值，當受訪者的回答出現前後不一致的情形時，易使研究者感到困惑，甚至影響訪問計畫的資料品質。本文藉由TEDS 歷次全國性面訪計畫中的「再測信度」，探尋受訪者在前後兩次訪問回答的一致性分布，並依據學理檢證各項可能的影響因素。研究發現，平均而言，有76.8% 的受訪者在前後兩次訪問中的回答明確且一致，7.0% 在兩次訪問中皆一致無反應；另外，一次訪問明確回答，另一次卻無反應的比例占7.6%，至於兩次訪問皆明確回答，但卻前後矛盾的比例則占8.5%。此外，分析結果顯示，立法委員選舉相對於總統選舉、兩次訪問間隔時間愈久、訪問當時距離選舉投票日愈遠、受訪者的教育程度愈低、被訪員評斷為對問卷題目敏感或回答內容不可信的受訪者，皆有顯著偏高的相對機率在其中一次或兩次訪問皆無反應，甚至出現前後回答矛盾的結果。這證實了在民意調查的執行層面上，減低受訪者對調查計畫的疑慮，尤其是針對教育程度偏低的受訪民眾而言，是提升整體訪問品質的重要關鍵。
Acquiring respondent’s true voting choice is the core request in electoral surveys. Unfortunately, it is not rare to see respondent’s different, even conflicting l, answer to the question of voting choice in a reliability test after the general survey is completed. The existence of respondent’s inconsistent answer certainly undermines the quality of surveys and deserves closer academic examination. This paper uses the TEDS’s survey data to retest the consistency of respondent’s answer to the question of voting choice in surveys and explores the possible causes of the inconsistency. It is found that, on average, there are 76.8% of respondents have maintained consistent of whom they vote for in preelection and post-election interviews. By contrast, there are 8.5% of respondents reply their voting choices differently and 7.0% of respondents are consistently no response in the two interviews. In between, there are 7.6% of respondents are either no response in the first interview but have a definite voting choice in the second interview and vice versa. In addition, this paper suggests that the longer time of interviewing after election day, the longer time between the two interviews, the lower level of respondent’s education, the higher level of respondent’s sensitiveness to the question, and the more distrustful respondent evaluated by interviewer tend to contribute to respondent’s different answers or no responses in two the interviews. These results also confirm that, particularly for those less-educated, eliminating respondent’s suspicion of survey is a crucial factor to the improvement of survey quality.
list=SELECT top 5 iJournalItem,sJournalItem1 FROM ViewJournalItem WHERE iJournalItem in(select top 15 iJournalItem from DCSearch7.dbo.SearchImp where (SearchDate > '20200921') group by iJournalItem order by count(iJournalItem) desc) and (iShelf = 1) and iJournal = '758'