原始樣本、替代樣本、與追蹤樣本的比較：「2001年台灣選舉與民主化調查研究」訪問失敗問題的探討

Abstract

本文檢視「2001年台灣選舉與民主化調查研究（TEDS）」抽樣與面訪資料蒐集過程中可能因為訪問失敗而造成推論偏誤的困素，資料分析包括原始中選樣本的人口特徵及其代表性，訪問失敗的原因及其影響，預備樣本的替代效果，訪問失敗樣本的追蹤，原始樣本、替代樣本、與追蹤樣本三項資料中各種認知、態度、與行為的差異，結果證實成功樣本和失敗樣本的差異主要是在年齡，以及因為年齡衍生而來的相關人口特徵，在政治認知態度與行為方面差異並不大。訪問失敗的最大宗是「有接觸但沒找到人」的情形，年齡越低，這種類型的比例就越高，男女皆然；拒訪和受訪者的性別或年紀大小並無關係，而是和居住或生活作息型態有關；「無接觸」的訪問失敗是隨機的，沒有什麼特別的影響因素。替代樣本的人口特徵因TEDS執行定義而和訪問失敗者一致，也比較接近母體的人口特徵，但是在政治認知態度與行為方面的表現並沒有比較好，有比較接近原始成功樣本、偏離追蹤成功樣本的現象。本文建議日後的調查規劃以預先估計訪問成功率膨脹樣本為上策，不要使用替代樣本，也不要使用預備樣本，如有資源，應規劃追蹤訪問失敗樣本。以成功樣本推論母體，應該在人口變數方面加權處理，以使推論偏差降至最低。

The study looks into the nonresponse problems of the 2001 TEDS survey, particularly in factors related to the failure of data collection and the effect of substitutes. Data analyses include the demographic characteristics of the sample originally selected, nonresponse statistics during different waves of interview attempts, the effect of substitutes, the follow-up interviews, and the differences of survey results from those samples. The results indicate that the main difference between complete interviews and those of unit-nonresponses lay on respondents’age and their related characteristics. There is no significant difference between these two groups on political attitudes. The most frequent cause for unit-nonresponse is failure to locate the selected respondent after reaching their dwelling houses, especially for those of younger respondents. Refusals are not correlated with respondents’ gender or age, but do correlate with areas and respondents’ life style. Substitutes are similar to those of complete interviews, hence fail to fulfill their original goals, i.e., replacing those hard to reach or those sampled respondents with high risk of nonresponse. The study suggests that for the future survey it is better to inflate sample size at the beginning, and try hard on follow-up interviews. Using substitutes will not correct the bias caused by unit-nonresponses.