A Ratio Estimator for the of Mean of a Sensitive Variable in the Presence of Lack of Trust in RRT and Measurement Errors

Abstract

RRT Models, both binary and quantitative, have been used extensively in surveys that include sensitive questions. These models allow respondents to provide randomized or scrambled responses which can later be unscrambled at an aggregate level but not at individual level. This feature protects the respondent privacy during face-to-face surveys which sometimes become necessary to decrease non-response.In this presentation we will focus on improving respondents’ trust in RRT methodology. We do so by allowing more scrambling options while maintaining the overall model quality. We will also discuss how we can take advantage of auxiliary information and account for measurement errors.