Effect of Questionnaire Structure on Nonresponse and Measurement Error:Sequential vs. Grouped Placement of Filter Questions
Carley-Baxter, L. R., Peytchev, A. A., & Black, M. L. (2010, May). Effect of Questionnaire Structure on Nonresponse and Measurement Error: Sequential vs. Grouped Placement of Filter Questions. Presented at AAPOR 2010, .
Based on a cognitive perspective, survey practitioners aim to structure questionnaires in the way that respondents’ memories are organized. If details about multiple behaviors are requested, asking the respondent to report a type of behavior and then following up with more detailed questions is assumed to help recall, reduce measurement error and possibly decrease item nonresponse. When surveys ask about multiple behaviors, a fear arises that asking many detailed questions after such a filter question will deter some respondents from reporting the next type of behavior on a filter question, inducing measurement error bias in survey estimates. This may also lead to higher breakoff rates prior to the answering of all key filter questions and higher item nonresponse rates to the follow-up questions. There is some evidence from self-administered surveys that when respondents recognize such a skip pattern, they become more likely to provide responses that avoid subsequent questions. The lack of interviewer presence may encourage this behavior. There is also similar, yet limited evidence on grouping filter questions from in-person interviewing in a study with a relatively long interview. Furthermore, the high interviewer rapport in face-to-face interviews may mask potential effects on breakoff and item nonresponse. We randomly assigned respondents in a dual-frame landline-cell phone study on intimate partner violence to either a sequential or grouped filter question design. In both versions, respondents were asked to identify the number of perpetrators for different types of victimizations. In the sequential design, multiple in-depth follow-up questions about each reported person were asked immediately after each filter question. We evaluate whether key survey estimates, item nonresponse, and breakoff rates differ when multiple in-depth follow-up questions are asked about each reported perpetrator for different types of victimization (sequential) versus asking all filter questions before asking for additional detailed information about the perpetrators (grouped).