The Crosswise Model (CM) has been proposed as a method to reduce effects of social desirability in sensitive questions. In contrast with former variants of Randomized Response Techniques (RRTs), the crosswise model neither offers a self-protective response strategy, nor does it require a random device. For these reasons, the crosswise model has received a lot of positive attention in the scientific community. However, previous validation studies have mostly analysed negatively connoted behaviour and thus draw on the principle of “more is better”. Higher prevalence rates of socially undesirable behaviour in the crosswise model cannot be attributed unambiguously to a reduction in social desirability bias, since random ticking resulting from respondent confusion about the question format cannot be ruled out as an alternative explanation. Unlike most research on crosswise models and randomized response techniques, we conduct an experiment in a general population survey that does not assess negatively connoted but socially desirable behaviour (namely, whether respondents had donated blood within the last twelve months). This design allows us to empirically disentangle the reduction of social desirability bias from random responses. We find signifcantly higher prevalence rates in the crosswise condition than in the direct question. What is more, we could not identify any subgroup of respondents, in which the CM successfully reduced social desirability bias. These results cast doubts on the validity of cosswise models. They suggest that a considerable number of respondents do not comply with the intended procedure.

Probability

Self-administered questionnaire: Web-based; Self-administered questionnaire: Paper