Experimental investigations on the validity of randomized response surveys

• Applicant: Dr. Adrian Hoffmann
• Subject Area: Personality Psychology, Clinical and Medical Psychology, Methodology ; Empirical Social Research
• Term: from 2017 to 2020
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 393108549

Social desirability bias threatens the validity of prevalence estimates in surveys on sensitive personal attributes. Dishonest answers can result in underestimates of the prevalence of socially undesirable, or overestimates of the prevalence of socially desirable attributes. To make more valid prevalence estimates feasible, indirect questioning techniques such as the Randomized Response Technique (Warner, 1965) aim at motivating respondents to provide truthful answers. To this end, they grant full confidentiality by encrypting individual answers via a randomization procedure. In an effort to advance previous approaches, we present the new Cheating Detection Triangular Model (CDTRM) that combines the easy-to-understand instructions of established Nonrandomized Response Models with a mechanism for the detection of noncompliant responses that is inherent in some Randomized Response Models. In so-called strong validation studies, we want to compare the CDTRM with competing approaches such as the Crosswise Model (CWM), the Cheating Detection Model (CDM), and the Stochastic Lie Detector (SLD). We intend to employ a paradigm specifically designed to experimentally induce a sensitive attribute, which will then serve as an external validation criterion with known prevalence (Hoffmann, Diedenhofen, Verschuere, & Musch, 2015).In a first experiment, we want to investigate whether the validity of prevalence estimates can be improved by employing mechanisms for the detection of instruction nonadherence. In a second experiment, we want to evaluate whether models with symmetrical answer options, none of which eliminates the possibility of being a carrier of the sensitive attribute, increase participants' willingness to cooperate. In a third experiment, we want to examine the models' susceptibility to deliberate faking, the comprehensibility of instructions, and the subjectively perceived confidentiality of individual answers. In a concluding fourth experiment, we want to further evaluate the best-performing model in an empirical investigation of group-focused enmity. To this end, we intend to use a representative sample to obtain undistorted prevalence estimates for hostile attitudes including xenophobia, islamophobia, and sexism. By conducting these experiments, we want to contribute to the development of state-of-the-art indirect survey techniques. We also want to obtain unbiased prevalence estimates for the German population regarding attitudes of high sociopolitical relevance, and derive recommendations on how to conduct future surveys on sensitive topics.

• DFG Programme: Research Grants
• Co-Investigator: Professor Dr. Jochen Musch