Compared to laboratory studies, participants much more likely drop out in Web studies, particularly if sensitive questions are asked. The high hurdle technique is thought to control drop out by maximizing de-motivating factors at the very beginning of a study or even earlier to reduce subsequent drop out, but previous research is inconclusive as to its impact and process (Göritz & Stieger, 2007; Reips, 2002). To investigate the technique’s dependence on other factors and its impact on drop out behavior, we conducted a Web experiment in which we (1) measured pre-experimental intended seriousness and (2) manipulated the high hurdle and the placement of sensitive items. Twelve different conditions resulted from the 2 x(seriousness) x 2 (high hurdle) x 3 (placement of sensitive items) between-subjects design.

Dependent measures were drop out, data quality and reaction time. After the seriousness check item, 396 Students answered a questionnaire about sport and nutrition. The high hurdle was placed in the introduction and was created by saying that the questionnaire contains sensitive items, which might make the person feel uncomfortable. In the control condition, there was no such high hurdle. The sensitive items were set either at the beginning, in the middle, or at the end of the experiment.

There was a significantly larger drop out directly after the high hurdle introduction, compared to the control condition (p<.001), i.e. the high hurdle served the first of its purposes. Also, there was an interaction with intended seriousness.

Further analyses showed that satisficer (people who secede to deliver accurate data in favor of roundings or estimates, likely due to decreasing motivation) had more missings than nonsatisficers, in contradiction to widespread intuitions. The assumption that participants take a longer time to answer sensitive items was confirmed (p=.005). Also, at sensitive items the frequency of chosen „don’t-want-to-answer“ options exceeded the amount of missings, whereas at non-sensitive items more missings than „don’t-want-to-answer“ options were counted. Implications of these results for research and questionnaire design will be discussed in the presentation.