Census Bureau Lead Researcher Tells Court Importance of Protecting Data

In a filing with a U.S. District Court in Alabama that was made public this week, the Census Bureau’s Chief Scientist, John Abowd swore a declaration that amounts to a comprehensive history of the Census Bureau’s legal, statistical, and moral responsibility to keep respondent information confidential.

Abowd made the core point that every survey the government conducts relies on trust that the personal information respondents volunteer will remain confidential. “Though participation in the census is mandatory under 13 U.S. Code § 221, in practice, the Census Bureau must rely on the voluntary participation of each household in order to conduct a complete enumeration,” the chief scientist wrote. This ethic at the Bureau dates as far back as when Congress first established confidentiality protections for individual census responses in the Census Act of 1879.

The declaration amounts to an expansive history lesson on how privacy protections have evolved over the decades at the Census Bureau. It describes why privacy is vital to government surveys and censuses that support a wide array of critical government and societal functions at the federal, state, tribal, and local levels.

The declaration is part of the government’s response to a lawsuit by the State of Alabama and others seeking to block implementation of new disclosure avoidance methods that some believe will make data less accurate, especially for the upcoming process of redrawing federal, state and local political jurisdictions. Abowd describes for the court the public process Census has engaged with stakeholders over many years to balance the need for privacy against the need for accuracy.

Abowd argues that while the Census Bureau’s confidentiality methodologies for the 2000 and 2010 censuses were considered sufficient at the time, “… advances in technology in the years since have reduced the confidentiality protection provided by data swapping.” He describes in detail a simulated “attack” Census itself conducted that showed using just 6 billion of the over 150 billion statistics re-leased in 2010 would allow an attacker to accurately re-identify at least 52 million respondents and with some third party data could re-identify around 179 million Americans or around 58% of the population.

The Census Bureau is continuing with stakeholder engagement on their latest privacy protection effort, often described as “Differential Privacy.” In the coming weeks they will be releasing a new demonstration file for stakeholders to assess and comment upon.