Incorporating Nonresponse Follow-Up Into a Main Survey Data Set

• Activity Number: 255
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Shin-Jung Lee*+ and James Lepkowski and Kristen Olson and Dandan Zhang
• Companies: University of Michigan and University of Michigan and University of Nebraska-Lincoln and Westat
• Keywords: Multiple imputation ; complex sample design ; weighting ; unequal probabilities ; fraction missing ; sequential regression

Abstract

The University of Michigan Dioxin Exposure Survey (UMDES) was conducted using a complex sample design to measure the association between environmental exposure and blood level of dioxin-like compounds. A main interview, blood, soil and dust data were collected, however nonresponse bias was a concern. At the conclusion of the main data collection (AAPOR RR4 rate 75%), non-complete cases were re-visited in a nonresponse follow-up study (NRFU) that 45% responded. The NRFU data were added to the main data set, and three types of missing values were imputed. A sequential regression multiple imputation leads to slight improvements in precision. The fraction of missing information varies widely across estimates. Study findings suggest that NRFU data be treated as a second phase sample of main data collection incomplete cases and combined with first phase sample data in a final data set.