Internal Census Emails Show Bureau Concern Over Data Fuzzing

  • Staff argued over privacy algorithm’s redistricting impact
  • Redistricting head said small counties could be the losers

The U.S. Census Bureau expert in charge of providing redistricting data to states raised alarm that a new privacy-protecting process could hurt small towns and make data too inaccurate for redrawing district maps, internal emails show.

The emails are attached to a brief filed by the Alabama attorney general in a lawsuit challenging the Census Bureau’s new “differential privacy” techniques. Those methods are meant to thwart sophisticated data miners by planting mathematical noise into the information that the bureau distributes to the states. That information is used to redraw congressional, statehouse, and local government district maps.

“The data must reflect what is seen in the real world because it is used to change how the real world interacts with itself and with its government,” James Whitehorne, chief of the bureau’s Redistricting and Voting Rights Data Office, wrote in a September email to John Abowd, the chief scientist for the bureau.

“This does not mean I do not understand our obligation to protect the public’s data, it just appears that in our zeal to protect the data we are harming the very same people we are protecting.”

The bureau is using this kind of data-fuzzing for the first time, seeking to protect privacy by making it harder to pinpoint which household gave which answers on the Census form.

Accuracy Tradeoff

Nine counties in Texas, Nebraska, Montana, and Alaska with populations smaller than 1,000 might have difficulty drawing voting districts if the bureau’s algorithm moves respondents from one census block to another, Whitehorne wrote. The “accuracy versus privacy trade-off is farther to the accuracy endpoint than you are likely comfortable in recommending,” according to the email.

In a brief filed April 13, the bureau asked the court to dismiss Alabama’s claim and preserve the tools designed to protect citizens from a “reconstruction attack” from computers that can piece together and make inferences in publicly available Census data to uncover respondent’s private information.

“If the Census Bureau were to continue doing what it did in 2010, it would be violating not only federal law, but also the confidentiality promise that it made to census respondents. And with that bond of trust broken, future census response rates would undoubtedly fall, and the accuracy of future censuses would suffer,” the bureau said.

The bureau didn’t immediately respond to a request for comment about the emails in Alabama’s brief, which was filed Tuesday.

The case is Alabama v. U.S. Dep’t of Com., C.C.M.D. Ala., No. 3:21-CV-211-RAH, Alabama reply brief filed 4/20/21.

To contact the reporter on this story: Alex Ebert in Columbus, Ohio at aebert@bloomberglaw.com

To contact the editors responsible for this story: Katherine Rizzo at krizzo@bgov.com; Anna Yukhananov at ayukhananov@bloombergindustry.com

Top