IFS Calls on U.S. Census Bureau to Retain ACS As Is

Editor’s Note: Today, the Institute for Family Studies sent a letter to the leadership of the United States Census Bureau concerning recent proposed changes to public data files for the Decennial Census and the American Community Survey. The text of the letter is below (note: all affiliations listed are for identification purposes only, and do not reflect the views or positions of the institutions themselves).  

June 4, 2021

Gina M. Raimondo
Secretary of Commerce
U.S. Department of Commerce

Ron S. Jarmin
Acting Director
U.S. Census Bureau

Dear Secretary Raimondo and Dr. Jarmin:

The Census Bureau's recent proposed changes to public data files released for the Decennial Census as well as the American Community Survey are extremely concerning, and threaten to create irremediable, negative effects on the study of American families. The general problems with these new procedures have been ably described by the professional staff of the Integrated Public Use Microdata Series (IPUMS). This letter is offered to specifically describe the likely adverse impacts of proposed privacy procedures specifically on the study of American families.

The Decennial U.S. Census and the American Community Survey are irreplaceable sources of information about American families, yet the Census Bureau proposes to practically eliminate information about families in these surveys. The proposed adoption of a differential privacy standard would replace the true microdata reflecting actual American families with statistically modeled “synthetic data.” This data is intended to emulate the patterns observed in the true data but is developed through a simulation based on a limited set of equations representing the true data in a simplified fashion. Furthermore, the Census Bureau’s current proposal would result in the loss of all reliable data on intrahousehold relationships.¹

These changes would make it effectively impossible to identify trends in the marital status and structure of families and make it difficult or impossible to match children to their parents in the data. Likewise, family characteristics of statistically less-common groups, such as adoptive parents, same-sex couples, low-income single mothers, multiracial households, and Native Americans, could not be reliably estimated using these new procedures. These losses in data quality are extremely concerning.

Until the Census Bureau can demonstrate an actual risk of re-identification of sensitive information about individuals from Census microdata that could not have been acquired by an attacker from other public or less secure sources, additional privacy protections will not create any public benefit. The Census Bureau's concern for privacy is understandable but must be balanced against the historic mission of the Census to make public things that are presently unknown. Efforts by Census Bureau professional staff to erect stumbling blocks before the use of public microdata by non-Census Bureau-affiliated personnel are hostile to the values of an open society and reflect a gross overreach by the Bureau.

Sincerely,

W. Bradford Wilcox
Founder and Senior Fellow, Institute for Family Studies
Director, National Marriage Project, University of Virginia

Lyman Stone
Research Fellow, Institute for Family Studies
Chief Information Officer, Demographic Intelligence
Adjunct Fellow, AEI

Wendy Wang
Director of Research, Institute for Family Studies
Former Senior Researcher at Pew Research Center

Nicholas Zill
Vice President, Westat (retired)
Former Member, National Committee on Vital and Health Statistics
Former President, Council of Professional Associations on Federal Statistics (COPAFS)

*Note: All affiliations are for identification purposes only, and do not reflect the views or positions of the institutions themselves.

1. Current Census Bureau procedures indicate that this simulation would be run at the individual level. The Census Bureau has not provided any information about how they would join these simulated individuals to households, their current algorithm has no household intrarelationship component, and published demonstration data identifies serious errors in the matching of households and individuals. Thus far, it is not even clear if the Census Bureau’s intended synthetic data release would have any information at all about household intrarelationships, making it almost impossible to analyze any questions related to family structure. Even if the Census Bureau does adopt a procedure for assigning household intrarelationship data to individuals, it would model household intrarelationships using equations derived from individual-level traits, essentially imputing population-wide individual-level trends onto households.