When survey respondents are faced with long, cognitively demanding interviews, they may take measures to reduce their cognitive burden. In particular, when questions are presented in a predictable pattern, it is relatively easy for respondents to learn that certain behaviors will help them get to the end of the interview more quickly. One frequently-used pattern is to ask questions in an interleafed format, in which each question in a series of filter items is succeeded by follow-up questions only if the individual responds affirmatively to the filter. In such cases, respondents may learn that negative responses to filter questions help them end the interview more quickly. Such underreporting can lead to biased survey estimates.

This presentation will report on the results of two recent surveys of members of a probability-based web panel in which respondents were asked a series of advertising recall questions in an interleafed format. The order of the ads was randomly assigned for each respondent and recorded. Preliminary results of this study show that later placement in the order of presentation has a significant, negative impact on reported ad recall. This presentation will also discuss the effect of respondent demographics and interest in the survey topic on the extent of underreporting, as well as the impact of length of time spent in the panel. Implications for existing survey practice and directions for future research will be discussed.