Online Program

Nonresponse Bias Analysis for Business Surveys - Perspective from the US Census Bureau
*Katherine (Jenny) Thompson, U.S. Census Bureau 


Although nonresponse bias is only one component of total survey error, mitigating it often poses substantive challenges for business surveys. For these programs, the statistics of interest are generally totals or percentage change in totals, and biases in either direction could lead to erroneous conclusions about the state of the economy. Business populations are highly skewed, with a large proportion of the estimated totals originating from a small set of cases. Moreover, viable response data can be obtained from different units than sampled: for example, a company may provide data for more than one sampled unit. High quality administrative data may be available from alternative data sources. This lecture discusses these challenges in detail and provides a broad overview of techniques used to conduct nonresponse bias analysis with business data. Finally, I present some methods designed to avert nonresponse bias in production, such as the usage of p-charts to monitor response rates and R-indicators