A Two-Stage Design for Superior Efficiency in Estimating Sensitive Attributes

Abstract

Examining sensitive characteristics or data that individuals are hesitant to disclose in surveys poses a challenge due to the ethical duty to protect respondent privacy. Warner’s randomized response (RR) technique, while enabling confidential estimation of such attributes’ prevalence in populations, suffers from increased variance as the likelihood of directly probing sensitive questions rises. To address this limitation, we propose an innovative two-stage RR framework designed to enhance practicality and statistical efficiency compared to Mangat’s model, while improving trustworthiness in real-world applications. Privacy protection metrics were computed for the proposed models, with efficiency analyses consistently showing that the new model surpasses Mangat’s model in efficiency.