Relevance & Research Question: Asking sensitive questions in online surveys is a difficult task as respondents might not always tell the truth and answer in a socially desirable manner. As a consequence, the prevalence of sensitive behaviors is likely to be underestimated and correlations between individual characteristics and sensitive behaviors might be biased if subgroups of the surveyed population differ in their probability to answer truthfully. The Randomized Response Technique (RRT; Warner 1965), a method proposed to face this challenge, achieved mixed results so far in providing more valid estimates than direct questioning. A related approach called the Crosswise Model (CM; Yu, Tian and Tang 2008) seems more promising. However, empirical evidence on the performance of these methods is still sparse and inconclusive, especially in the case of online surveys.
Methods & Data: In the context of an online survey on plagiarism and cheating on exams among students of two Swiss universities (N = 6494), we tested different implementations of the RRT and the CM and compared them to direct questioning using a randomized experimental design. To evaluate the different methods, we analyzed differences in prevalence estimates, breakoff rates and respondents’ impression of the techniques.
Results: Results reveal a poor performance of the RRT, which failed to elicit higher prevalence estimates than direct questioning. Using the CM, however, significantly higher prevalence estimates could be achieved.
Added Value: Our study provides a thorough experimental comparison of the RRT and the CM, which to date is missing in the literature. Furthermore, we present different implementations of the RRT and the CM, specifically tailored for use in online surveys. Implications of our findings for the future use of sensitive questions techniques such as the RRT and the CM are discussed.