The Use of Responsive Split Questionnaires in a Panel Survey

Abstract

Lengthy surveys may be associated with high respondent burden, low data quality, and high unit nonresponse. To address these concerns, survey designers may reduce the length of a survey by eliminating questions from the original questionnaire, but this means that some information would never get collected. An alternative may be to divide a lengthy questionnaire into subsets of survey items and then administer each subset to distinct subsamples of the full sample. This is referred to as a split questionnaire design and has the benefit of collecting all of the original survey information. We identify a significant deficiency in the current set of split questionnaire methods, namely, the incomplete use of prior information about the sample unit in the design. In most contemporary applications of split questionnaires, generally only characteristics of the survey items (e.g., content, cognitive burden) are used to inform the design; however, if joint consideration is given to characteristics on the survey items as well as the sample unit when designing a split questionnaire, then there may be the potential to improve the split questionnaire's utility. In this dissertation, we explore the extent to which, if any, jointly considering both types of information at the design stage will yield more efficient split questionnaires. We propose various methods for incorporating prior information about the sample unit into the split questionnaire using features of responsive design. We highlight how this specific application of a responsive split questionnaire can be used to address the concerns present in a major federal survey. Finally, we draw from the literature pertaining to survey design, experimental design, and epidemiology to develop and implement a framework for evaluating the proposed new elements of our split questionnaire design.