Conferences

Survey literature has documented that filter questions reduced the percentage of people who gave opinions and increased “Don’t know” responses when compared to questions directly asking people to give opinion on issues (the so-called corresponding “direct questions”). Various rationales have been proposed to account for this phenomenon. One recent explanation is that direct questions carry a presupposition that can influence respondents’ subsequent responses. By contrast, filter questions cancel the presupposition behind the corresponding direct questions and allow for people the choice to claim that they don’t know about the topic of interest or the topic doesn’t apply to them. This paper extends the previous work on presupposition in survey by examining the effects of presupposition in web surveys, a self-administered mode. Two experiments were embedded in web surveys. The first experiment tested the effects of filters on questions asking people about the perceived importance of various issues. For a random half of the sample, respondents were asked directly how important they consider an issue is (the “direct question” condition) and the other random half received a filter first asking whether they consider the issue important or unimportant (the “filter question” condition), followed by a second question asking them how important (unimportant) they think the issue is. The second experiment studied the effect of filter questions on behavioral/factual questions. Again a random half of the respondents were asked directly how many times they’ve witnessed a crime (or have been victimized) in the “direct questions” condition and the other half were given a filter asking whether they have witnessed a crime (the “filter question” condition). Both experiments showed that filter questions change the response distributions by cancelling out the presupposition carried by the corresponding direct questions. Furthermore, some respondents are more prone to the filter effect than others. I will examine how respondent characteristics interact with the presence/absence of filters in survey responses.