Abstract:
This dissertation addresses the questions of whether or not Randomized Response Models (RRMs) are suitable for use in criminological research and how RRM applications can be improved in general. RRM is the umbrella term for certain survey methods which were developed to minimize socially desirable response behavior. As socially desirable responses may especially occur when asking questions on sensitive topics, RRMs are — at least in theory — especially well-suited for such questions. This method could be particularly promising in the field of quantitative criminology, as topics such as crime, criminal prosecution, and victimization provide for a large number of sensitive topics. To answer these research questions, a form of case study was conducted initially, and evidence for the existence of social desirability bias in a criminological survey was collected. Additionally, the Poisson model, a new method that enables particularly efficient measurement of a behavior’s prevalence and rate of occurrence, was introduced to improve RRM applications. In an online survey on the prevalence of drinking and driving, the Poisson model was combined with one specific RRM, the Unrelated Question Model (UQM). This survey study revealed problems regarding the functionality of the UQM, a problem not uncommon for RRM applications. Furthermore, the extent to which comprehension aids during participant instruction can contribute to improving the validity of RRM applications was investigated. However, it remained unclear how important the comprehension of instructions actually is for the functionality of RRMs and whether comprehension aids do, in fact, improve the validity of such applications. Overall, this dissertation illustrated that while RRMs provide for a promising methodological approach, the conditions for their optimal use have yet to be identified completely. Consequently, RRMs cannot yet be fully recommended for use in criminological research.
Randomized Response Models