Most of the news coming out of the U.S. Census Bureau lately has, understandably, focused on our preparations for the 2020 Census. Yet, important work on measuring the nation’s people, places and economy continues even amongst the flurry of activity for the once-a-decade count. This includes work to transform the Census Bureau into a 21^st-century statistical agency to meet the nation’s need for high-quality and reliable information.

The Census Bureau has a proud history of innovation including playing critical roles in developing statistical sampling and digital computing. These innovations helped transform economic and social measurement during the 20th century. As I described in a recent paper, the explosion of digital data, improved computation and modern analytical methods present us with new opportunities to transform economic and social measurement on an even greater scale. This includes more timely and more granular statistics, and as well statistics tailored for particular use cases. 

Providing more timely and granular statistics is not feasible with a 20th century measurement infrastructure that relies primarily on sample surveys for source data. Doing so requires statistical agencies like the Census Bureau to tap rich sources of administrative, transaction and other sources of digital information that are generated every day throughout the economy. On March 15, 2019, a number of economists and statisticians from U.S. and international statistical agencies, academia and the Federal Reserve will be gathering to discuss progress on harnessing such data, often referred to as Big Data, for economic measurement. The conference, titled Big Data for 21st Century Economic Statistics, is organized by the Conference on Research in Income and Wealth (CRIW) which has been involved in improving economic statistics for many decades. The conference will feature a keynote by Tjark Tjin-A-Tsoi who is director general at Statistics Netherlands. He has led Statistics Netherlands’ modernization campaign featuring an aggressive move from surveys to administrative and Big Data to support official statistics.

The papers presented at the conference explore new source data that promise to yield a much richer view of the state and evolution of the U.S. economy. An excellent example are the rich data on retail sales and consumer spending that arise from point of sale transactions and credit card spending. These offer the opportunity to track sales by product at subnational levels and to track consumer spending in real time to see the impact of shocks including natural disasters. Replicating the timeliness and granularity offered by these new data sources with surveys would be prohibitively expensive and burdensome on respondents. But these new sources lack complete coverage and are outside of the control of the statistical agencies. Thus, it is likely that we will need to explore a blended survey-big data model as we seek to incorporate these news source into official statistics.

This brings us to another a focus of the papers at the conference: new methods. The traditional toolkit of the Census Bureau and other federal statistical agencies focused on methods for sample surveys.  Many of the conference papers explore the use of machine learning, artificial intelligence and natural language processing to improve economic measurement. These papers, however, are just scratching the surface and much work needs to be done to build the methodological toolkit to optimally blend survey, administrative and big data to provide accurate and reliable statistics in a production setting.

The papers at the conference demonstrate there’s great potential for new data sources, improved computation and modern analytical methods to transform 21st century economic and social measurement. But work so far has been limited to relatively small-scale research projects. Achieving the benefits of improved measurement, lower response burden and increased efficiency requires scaling these research activities to the level of national statistical programs. The Census Bureau and other federal statistical agencies are not currently resourced or organized to do this. They need to invest in building relationships across government agencies and the private sector to secure access to high-quality source data. They need to invest in staff with the skills to acquire, process and curate large data sets, and to build reliable and privacy-protected statistical products from blended data. Information systems need to be redesigned to accommodate both survey and alternative data processing. Clearly, much work needs to done to build a statistical system capable to producing the timely and granular data that the papers in the conference hint at.