PUB - Publikationen an der Universität Bielefeld

When it comes to sensitive questions, data is often affected by bias due to non-response or effects of social desirability. Several methods have been introduced to eliminate answer bias by using randomization processes and probabilistic theory to obscure the respondent's answer and create anonymity, thus facilitating honest answers. The probably most traditional method is the Randomized Response Technique by Warner (1965). However, this method is loaded with certain disadvantages. Therefore, in the last decade, newer methods were introduced that aim at balancing the disadvantages and weaknesses of previous methods, for instance, the non-randomized models Crosswise Model and Triangular Model (Yu et al. 2008) as well as the Parallel Model (Tian 2014). Although especially the Triangular Model is easy to implement in a study, there is only little empirical evidence on its application in different survey modes and populations. Further, it is to assume that certain questions are not equally sensitive for everybody due to specific personal characteristics. Thus, indirect questioning might not be effective in general but only for certain populations. The present study extends prior work on the Triangular Model by evaluating it for different subgroups. The conducted experiment asks for sensitive characteristics in the context of mental stress among students. The Triangular Model achieves significantly higher percentages than conventional direct questioning for illegal drug use among persons that answer socially desirable according to the characteristic of Self-Deception. For the other analyzed subgroups (Impression Management, gender, and depressiveness), the Triangular Model could not achieve higher prevalence rates compared to direct questioning on a sufficient probability level. But still, hard evidence on the effectiveness of indirect questioning models is thin and further critical discussion is needed.