On the Confidentiality Guaranteed under Randomized Response Sampling: A Comparison with Several New Techniques Munir Mahmood Department of Econometrics and Business Statistics Monash University Clayton Australia Sarjinder Singh and Stephan Horn Methodology Division The Australian Bureau of Statistics Belconnen Australia Summary In this paper, three simple alternative randomized response models are proposed to be used in the situation where Moors' (1971) model fails as shown by Mangat et al. (1997). The proposed estima- tors are shown to be more efficient than the estimator proposed by Greenberg et al. (1969). Key words: Randomized response technique; Respondents privacy; Unrelated question model. 1. Introduction It is widely understood that direct questions on sensitive issues often result in deliber- ate incorrect answers and/or refusal to respond from the respondents. The real essence of Warner's (1965) randomized response technique is to reduce the response error of potentially ªsensitive questionsº. Warner, while estimating p, the proportion of popu- lation possessing stigmatized character (say) A, uses two randomized questions in- quiring about the respondents status in relation to the sensitive character only. Green- berg et al. (1969) in their theoretical development, dealt with two situations of p y being known and unknown. Here p y is defined to be the proportion of neutral charac- ter (say) Y in the population. Greenberg et al.'s (1969) estimator for p is given by ^ p G  1  P 2  ^ q 1 1  P 1  ^ q 2 P 1  P 2 ; 1:1 Biometrical Journal 40 (1998) 2, 237±242