Does balancing survey response reduce nonresponse bias?

Barry Schouten
18. April 2016 bei GESIS in Mannheim (Konferenzraum B2,1) 13:45 Uhr

Abstract Nonresponse, or more generally, missing data, occurs in virtually all surveys. It is known that the response rate limits the maximal impact of nonresponse on bias in survey statistics. The actual bias itself is, however, also determined by the associations between survey variables and nonresponse. Literature has put forward a number of indicators that attempt to measure such associations. These indicators all share a reliance on available auxiliary variables, either from frame/administrative data or from survey paradata. A legitimate question, therefore, is whether any detectable bias through auxiliary variables can be adjusted just as well in the estimation afterwards. In the presentation, I will focus o

n one such type of indicators, the so-called R-indicators, and will link these indicators to the bias of well-known nonresponse adjustment estimators. Next, I will give theoretical and empirical motives for improved accuracy of survey statistics in case of a more balanced survey response.

Barry Schouten is senior methodologist at Statistics Netherlands and associate research professor by special appointment at Utrecht University. He graduated in mathematics at the Technical University Delft in 1995 and finished a Ph D in mathematical statistics at VU University Amsterdam in 2000. In 2002, he joined the Methodology department at Statistics Netherlands where he was involved in research on nonresponse reduction and adjustment, mixed-mode survey design, mode effects and adaptive survey design. He was coordinator for EU FP7 project RISQ.