Abstract: | Extending previous work, the authors find that the wording ofa filter question can make a significant difference in the percentageof "don't know" (DK) responses elicited by an item, especiallywith topics that are more abstract or less familiar to surveyrespondents. They also find, however, that the content of anitem can have a substantial, independent effect on DK or "noopinion" responses, regardless of how the filter question isworded. In general, it appears that the less familiar the issueor topic, the greater the increase in DK responses producedby adding a filter. Even more important, the analysis showsthat filtering can in some instances dramatically affect theconclusions a pollster would draw about the distribution ofpublic opinion on an issue. Indeed, such effects may occur moreoften than has previously been suspected, though the circumstancesunder which they emerge remain elusive. The authors suggestthat such effects may become amenable to analysis by probingrespondents about "what they had in mind" as they answered thequestion. |