在問卷調查中若有涉及個人隱私或非法行為的敏感性問題時，問卷的設計以直接詢問（Direct Inquiry）經常會造成拒答，或即使願意回答也難以確保內容的真實性。因此，為了減少受訪者拒絕回答、逃避回答、或不實回答等現象，可利用隨機作答方式（Randomized Response Technique，簡稱 RRT）蒐集正確的問卷資料，降低統計推論的誤差。RRT可以讓受訪者在不洩露自己的身分且不知道問題的真正意圖情況下，誠實地回答敏感性問題，但目前國內學界鮮少於實地調查中使用 RRT來蒐集敏感性問題，僅有楊文山（1994）於大型面訪調查中應用不相關問題模型的隨機作答方式，估算台灣地區選舉的賄選比例。本文利用 Hsieh et al.（2013）在 2012年至 2013年參與的 「台灣社會變遷基本調查計畫」中所提出的多元類別隨機作答（Randomized Response Technique in MultiCategory），蒐集 2012年總統大選投票行為的資料，估計各候選人的得票率，並與直接詢問法及中央選舉委員會公布的總統候選人之得票率進行有效性比較。

Direct inquiry in social surveys of sensitive questions such as personal privacy or illegal behaviors often encounters refusal or untrue response. In order to collect reliable sensitive data, reduce the unwillingness of respondents, and reduce bias in estimation, Randomized Response Technique (RRT) is one of the important methods. RRT has been commonly used to avoid biased answers in surveys on sensitive issues by ensuring respondents’ privacy, but are rarely adopted in faceto-face interview surveys in Taiwan, except for Yang (1994) in his study estimating election bribery by using the unrelated question RRT of Greenberg et al. (1969). We use Randomized Response Technique in Multiple Categories (RRTMC) developed by Hsieh et al. (2013) in the Taiwan Social Change Survey (TSCS) conducted in 2012 and 2013 to collect two waves of 2012 presidential election data to validate RRTMC based by comparing the estimates with the responses to direct inquiry collected in the same survey and official voting records.