Nonresponse and Measurement Error in Mixed-Mode Designs.

Survey Nonresponse;Data Collection;Mode of Administration;Sequential Mixed-mode Design;Measurement Bias;Social Sciences (General);Social Sciences;Survey Methodology

Levenstein, Rachel M.Kreuter, Frauke ;

University of Michigan

关键词: Survey Nonresponse;    Data Collection;    Mode of Administration;    Sequential Mixed-mode Design;    Measurement Bias;    Social Sciences (General);    Social Sciences;    Survey Methodology;   

Others  :  https://deepblue.lib.umich.edu/bitstream/handle/2027.42/78764/rmlev_1.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Because of declining response rates in surveys, survey researchers are increasingly using data collection methods that can increase response rates while containing costs.One such method is the use of sequential mixed-mode designs, in which the mode of administration is switched to a second mode within the survey field period.This dissertation examines the use of sequential designs in two panel surveys: The Relationship Dynamics and Social Life survey (RDSL) and the Panel Arbeitsmarkt und Soziale Sicherung (PASS) in terms of nonresponse behaviors, nonresponse bias, and measurement bias.We find that past behavior—the likelihood of response, the likelihood of responding late enough to warrant a mode switch, and the likelihood of responding in a particular mode—is strongly related to current behaviors in RDSL. The likelihood of response is related to the frequency and timing of response in the past; the likelihood of responding late in past waves is positively related to the likelihood of responding late at a current wave; and the likelihood of using a mode is positively related to the number of times the individual responded via that mode in the past. Nonresponse in the panel survey is not a significant problem after responses are combined from both modes. Finally, the sequential design either does not affect or it reduces the nonresponse bias.In Wave 2 of PASS, nonresponse bias, measurement bias, and the relationship between response propensity and measurement bias is examined. As above, the sequential design either does not affect or it reduces the bias. Although measurement bias is a significant problem, it is unaffected by the mode of response. No relationship between response propensity and measurement bias is found within or between modes.