Two new nonrandomized response models for surveys on sensitive topics

Abstract

If the direct questioning in surveys on sensitive variables leads to non-ignorable nonresponse and untruthful answers, a considerably biased estimator might be the consequence. For such cases, the higher complexity of the indirect questioning designs such as the nonrandomized response models may pay off in terms of estimation accuracy assuming an increase in respondents’ cooperation. In this methodological article, two new nonrandomized response questioning designs are proposed that are simple to implement in several survey modes and easy to understand. These models, the double triangular model and the flat parallel model, are nonrandomized versions of two well-established and practically applied methods, the forced and the contamination method, to overcome the limitations of these randomized response techniques. Additionally, the theory of all nonrandomized response models is developed for general probability sampling under one theoretical roof. Moreover, all these models are compared from the point of view of their respective levels of privacy protection. The flat parallel model combines the advantages of the other models and can thus have a positive effect on both, cooperation willingness and estimation accuracy.