Toward a Structural Approach to Questionnaire Design

• Activity Number: 28
• Type: Contributed
• Date/Time: Sunday, July 31, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Government Statistics
• Author(s): Ioana Schiopu Kratina*+ and Christina Maria Zamfirescu
• Companies: Statistics Canada and The City University of New York
• Keywords: questionnaire design ; node ; path ; coverage of a question

Abstract

Statistical agencies rely on well-designed questionnaires to collect good data, minimize costs and control response burden. We view a questionnaire as a directed graph, with questions as nodes and flows as paths. In this set-up, we would like to evaluate the structure of a questionnaire and improve on its design. The coverage of a question is a verbal description of the sub-population that is asked that question, and is very important to the analyst. The graph structure should point to questions where the coverage is hard to verify. A node with a high in-degree (several paths converge into it) corresponds to a question asked of several subpopulations following different paths, which eventually converge into this node. As it is hard to test the completeness of the coverage of this question, we should attempt to place it earlier in the questionnaire to minimize the problem. The link between information theory and questionnaires viewed as graphs (Picard 1980) would give us insight into the ordering and formulation of questions so as to maximize the amount of information collected and minimize the response burden. We illustrate these ideas with examples from a complex survey.