Enhancing the applicability of randomized response techniques

Abstract

Surveys addressing sensitive research topics such as domestic violence or sexist attitudes are subject to self-protecting response biases. Randomized response techniques (RRTs) have been proposed to encourage honest responses to sensitive questions by guaranteeing privacy protection of survey respondents through randomization in the questioning design. Thereby, they aim to increase the validity of estimates of prevalences of sensitive attributes. However, the applicability of RRTs is impaired by a still less than ideal validity of prevalence estimates and high sample size requirements. In this dissertation, I propose two approaches to enhance the applicability of RRTs. First, I present a testable model that incorporates a parameter measuring non-adherence to instructions in a common variant of the RRT. The results of an empirical study on intimate partner violence indicate that applying this extension enables a more valid description of the mechanisms underlying responses. Second, I propose incorporating RRTs into a sequential hypothesis testing framework using a curtailed sampling plan. Theoretical considerations and first empirical results show that following this approach the sample size requirements of RRTs can be substantially diminished while preserving an easy-to-conduct sampling procedure. In summary, the proposed procedures can render applications of RRTs more feasible and, thereby, enable insightful future investigations of sensitive research questions.