Assessing the Relationship Between Proxy Burden Measures and Survey Response in a Longitudinal Household Survey Using Regression Trees

• Activity Number: 453 - Recursive Partitioning for Modeling Survey Data and Randomized Trials
• Type: Topic Contributed
• Date/Time: Thursday, August 6, 2020 : 10:00 AM to 11:50 AM
• Sponsor: Survey Research Methods Section
• Author(s): Morgan Earp* and Brandon Kopp and John Dixon
• Companies: National Center for Health Statistics and Bureau of Labor Statistics and Bureau of Labor Statistics
• Keywords: Respondent Burden; Regression Trees; Survey Response; Current Population Survey

Abstract

This paper explores the relationship between five proxy measures of burden and survey response for a longitudinal household survey using regression trees. We used several proxy measures of burden to model the propensity to respond to the Current Population Survey (CPS) over the eight month duration of data collection. Our analysis focused on all households/persons in sample between 2015 and 2017 who responded at least once to the CPS. Our proxy burden measures fit into five dimensions of burden outlined by Bradburn: 1) length (interview mode, number of contact attempts, and number of months in survey); 2) effort (number of adults, reporting for themselves, reporting for others, or both); 3) stress (refusal indicators for income, ethnicity, marital status, and education questions); 4) frequency (the number of months in the survey as well as the number of additional supplemental surveys they were sampled for); and 5) saliency (employment status, disability status, home ownership, presence of children, and education level). Our paper examines the relationship between these five burden dimensions and longitudinal survey response using the R tree modeling package rpms.