Imputing Biomarker Outcome Refusals and Unasked Questions in a Nonresponse Follow-Up for Environmental Exposure Models

• Activity Number: 242
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Survey Research Methods Section
• Author(s): Brian M. Wells*+ and James Lepkowski
• Companies: University of Michigan and University of Michigan
• Keywords: biomarker outcomes ; complete case analysis ; fraction of missing information ; multiple imputation ; nonresponse follow-up ; nonresponse weighting adjustments

Abstract

The University of Michigan Dioxin Exposure Study obtained multiple biomarker outcomes (BMOs) from respondents willing to provide a blood sample. However all predictor data is available for respondents who refused to provide a biological sample in addition to a subset of predictors for respondents to a nonresponse follow-up (NRFU) who refused the main interview. BMOs were multiply imputed along with a subset of predictors from NRFU to help retain sample size. Environmental exposure models regressed the complete cases, the former plus respondents who refused to provide a blood sample, and the former plus imputed NRFU cases. Comparisons among these three data sets reveal that imputing BMO refusals to the main survey attenuates most coefficients relative to the complete case analysis (CCA) and greatly increases the instability of the estimates based on the coefficient specific fractions of missing information. Including the imputed NRFU cases does not markedly increase the attenuation or instability relative to the first comparison. These findings raise important questions about whether CCA with nonresponse weight adjustments provide a sufficient basis for inference to a population.