A Model-Based Approach to Assessing and Mitigating Nonresponse Bias for the Monthly Wholesale Trade Survey

• Activity Number: 665
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Government Statistics
• Author(s): Joanna Lineback*+ and Eric B. Fink
• Companies: U.S. Census Bureau and U.S. Census Bureau
• Keywords: nonresponse bias ; auxiliary data ; response prediction ; response propensity ; design-based ; model-based

Abstract

This paper addresses the topic of nonresponse bias in business surveys. Some common techniques for investigating nonresponse bias post-data collection have been established from a design-based perspective. Many of these techniques, however, cannot account for small nonresponse adjustment cells which are common for business surveys. For the Monthly Wholesale Trade Survey nonresponse bias analysis, we tried alternative, model-based approaches. To evaluate the potential for nonresponse bias, we examined propensity and prediction response models using frame data compiled from auxiliary data sources, including the Economic Census and the American Community Survey. Additionally, we modeled response propensity and prediction given the current adjustment cells variables. To minimize variance and bias, the variables used to define nonresponse adjustment cells should be highly predictive of key survey estimates and the likelihood of responding to the survey. Evidence otherwise is evidence of nonresponse bias. Our findings are discussed.