Online Program

Return to main conference page

All Times ET

Program is Subject to Change

Wednesday, June 16
Wed, Jun 16, 10:30 AM - 12:00 PM
TBD
Leveraging Machine Learning to Improve Economic Surveys and Programs

A Smart Instrument NAPCS Classification Tool for the Economic Census (SIN-CT) (309694)

*A.J. Goldsman, US Census Bureau 
Javier Miranda, US Census Bureau 
Anne Sigda Russell, U.S. Census Bureau 

Every five years, the U.S. Census Bureau conducts an economic census and collects extensive statistics about businesses that are essential to understanding the American economy. We discuss the development of a smart instrument classification tool to help respondents in the Economic Census self-classify the products and services they sell. Product classification is traditionally burdensome to respondents to the Economic Census with the NAPCS classification system covering over 8,000 distinct products. It is not surprising that in 2017 over 900,000 write-in descriptions were provided to the Census Bureau by respondents. In this paper we discuss the use of Natural Language Processing tools to develop SIN-CT using a variety of public use and confidential (protected) data sources. We discuss the testing methodology, performance metrics, and results from an early live test. In addition to reducing respondent burden we expect SIN-CT will reduce the number of write-in provided by respondents, reduce analyst costs, and improve the quality of the statistics derived from these data.