247 – Topics in Sample Design and Data Collection
Predicting Proxy Status in Nonresponse Follow-up Workload
Andrew Keller
U.S. Census Bureau
For the 2010 Census, the Nonresponse Followup (NRFU) operation verified the status of units that did not respond by mail in areas that received a mailback census questionnaire. The NRFU workload totaled 47 million units at a cost of $1.6 billion. The Census Bureau then classified these units into four categories: occupied, vacant, deleted, and unresolved. Among occupied, vacant, and deleted NRFU units, respectively, 24%, 98%, and 90% of interviews were completed via proxy respondents . In general, survey results often demonstrate reduced data quality when allowing proxy respondents for household interviews. By understanding the unit- and area-level characteristics that lead to proxy respondents, we can better understand where lower data quality might be present in census results. This research implements a modeling approach to predict proxy status. The approach relies on housing unit-level covariates defined by census operations prior to beginning NRFU fieldwork. In addition, the model uses tract-level characteristics known prior to the census as independent variables to predict proxy status for the NRFU universe.