A Simulation-Based Approach to Refining Estimates of Sampling Variability for the Planning Database's Low Response Score

• Activity Number: 637 - The Use of Auxiliary Data in Frame Development, Coverage Assessment, and Field Data Collection
• Type: Contributed
• Date/Time: Thursday, August 2, 2018 : 8:30 AM to 10:20 AM
• Sponsor: Survey Research Methods Section
• Author(s): Luke J Larsen*
• Companies: U.S. Census Bureau
• Keywords: Census Bureau; Planning Database; Low Response Score; sampling error; multivariate regression; simulation

Abstract

The Planning Database (PDB) contains tract- and block-group-level U.S. population and housing statistics from decennial census and American Community Survey (ACS) data. One of the chief PDB uses is field operations planning for survey projects. To that end, the PDB features a metric called the Low Response Score (LRS), which is modeled on Census 2010 mail return rates. The LRS is predicted by a linear regression model that is fit with covariates from the PDB. ACS-based statistics have estimable sampling variability; since some of these measures are model covariates, the LRS must also have sampling error. This paper extends earlier work in which replicate data were used to fit the LRS model while also accounting for sampling error, allowing for calculation of LRS margins of error. However, that attempt introduced sampling error only during model fitting and not during prediction, leading to underestimated margins of error. In this paper, a Monte Carlo simulation is used to approximate total sampling variability in the LRS. Analysis of this outcome relative to previous work will determine whether the calculated margins of error are sufficient estimators for the true margins of error.