Unbiased Estimation of a Sensitive Proportion in General Sampling by Three Nonrandomized Response Schemes

Abstract

Contrasting with well-publicized randomized response (RR) techniques (RRT) claimed to be useful in estimating the proportion of people bearing a sensitive characteristic in a given community, recently nonrandomized response (NRR) techniques (NRRT) are emerging. As with most early RRTs, the NRRTs to date are applied exclusively to samples selected by simple random sampling (SRS) with replacement (SRSWR) scheme alone. This article shows how three NRR schemes in vogue may be used in unbiased estimation even when a sample may be selected following a general scheme. The current literature stresses maximum likelihood estimation (MLE) rather than unbiased estimation (UE) when these NRR schemes are employed.