Information Theoretic Approach on Randomized Response Models in Surveys

Survey biases can appear in different forms, particularly as response or non-response biases. Such biases may occur when interviewees are asked questions on sensitive topics to which they may self-censor their response or hesitate to answer altogether to avoid social undesirability and feeling of self-embarrassment, or simply to preserve their privacy. Several studies suggest that misreporting is widespread when the survey includes sensitive topics including race, gender, sexuality, political affiliation, etc. A randomized response (RR) method has been proposed by Warner Ill to eliminate response and non-response biases by increasing cooperation and trust between interviewees and interviewers. In this paper, an information theoretic approach is taken on RR methods for surveys requiring YES or NO responses, and the tradeoff between randomization and survey accuracy is investigated.