Census Privacy Technique to Face Scrutiny of Three-Judge Panel

Alabama seeks a quicker release of census data and questions the accuracy of differential privacy, the technique the Census Bureau plans to use to safeguard personal information for next 72 years.

A briefcase of a census taker is seen as she knocks on the door of a residence in Winter Park, Fla., last August. (AP Photo/John Raoux)

(CN) – After a count of more than 331 million people across 3.8 million square miles all the while dodging wildfires, hurricanes and a pandemic, the U.S. Census Bureau is months away from fully releasing the data that will parcel out political power in the states and direct the flow of government funding for years to come.

But on Monday, a panel of three federal judges in Montgomery, Alabama, will decide whether to step into the process after the state sued last month based on claims that the bureau’s rollout of the 2020 redistricting data was too slow and the steps the bureau has taken to protect respondents’ personally identifiable information has altered the data so much that it’s unusable for redrawing political maps.

Alabama said it collaborated with the agency for the last five years to get the data out on time. “Alabama upheld its end of the deal, but the bureau has decided—unilaterally—that it will instead submit data to the states by September 30, 2021,” the state’s lawsuit says.

The state is supposed to pass a redistricting plan by May 18.

Furthermore, the Census Bureau plans to use a technique called differential privacy to make sure no one in the age of powerful computing and big data can figure out what answers someone gave the agency by triangulating their answers or analyzing other datasets.

The agency’s privacy protections have seen a long evolution from the early years when all the answers were posted in public. Differential privacy is a 15-year-old method that defines privacy mathematically, analyzing the dataset and making sure any one person’s answers don’t stand out. This is done by tweaking some of the answers, a process the National Institute of Standards and Technology compared to static on a radio in a 2019 video.

Alabama says the fuzzing of the data will lead to inaccurate congressional districts, which will lead to lawsuits over how the maps were drawn. It asks the judges in the Middle District of Alabama to make the Census Bureau use the privacy techniques it used in the 2010 census, arguing that seemed to work just fine.

“There have been no reports that anyone outside the bureau has gained access to the responses of a particular identified person from the released 2010 data,” the complaint states.

Alabama asks the judges to come to the conclusion that differential privacy violates statutes and runs afoul of constitutional rights, and to prevent the Census Bureau from using the technique in the future.

The bureau said in its response brief the two questions before the federal court in Montgomery are “two large obstacles to the successful operation of the 2020 decennial census.” Thanks to delays caused by the pandemic, for instance, the slow rollout was drive by factors outside its control, the agency said.

It further argued it was important to get the count right, as it would determine how political power is allocated. “It underpins our nation’s representative democracy,” the brief states,

The state’s request to stop the Census Bureau from using differential privacy “raises significant concerns,” the agency argues, as it would still need to somehow protect respondents’ individual replies.

But the process with how it protected the responses to the 2010 census did not keep up with evolving technology, the bureau said, and it made a promise to every resident of the United States that their answers would be protected.

If it broke that promise, “future census response rates would undoubtedly fall, and the accuracy of future censuses would suffer,” the bureau said.

A 2020 census letter mailed to a U.S. resident in Detroit. (AP Photo/Paul Sancya)

The case is being heard by a three-judge panel because of the differential privacy claims, and any appeal will go directly to the U.S. Supreme Court. Chief U.S. Circuit Judge William Pryor Jr. of the 11th Circuit appointed the panel hearing the case: U.S. District Judges Austin Huffaker Jr. and Emily Marks and U.S. Circuit Judge Kevin Newsom, all appointed by Donald Trump.

Ahead of the hearing, states, lawmakers, privacy expert and law professors have filed a flurry of amicus briefs weighing in, primarily on the topic of differential privacy.

A amici brief filed on behalf of 20 experts in cryptography and data privacy said the Census Bureau is expected to keep individual responses confidential, even at a time when computing and big data is growing.

Among the experts joining the brief was Cynthia Dwork, a computer science professor whose Harvard biography notes she helped develop differential privacy.

They also noted that some terms in the court record were “used imprecisely, and at times inaccurately.”

Already, people have been able to take intelligent guesses as to how people answered surveys such as the census using only a couple hundred lines of code and a laptop. Differential privacy, the experts wrote, is the only way to keep Americans’ privacy secure in the future from the likes of companies that gather large amounts of data, like Facebook, or counties such as China.

“It means that whether the adversary is a hostile nation state or an angry teenager the confidentiality promises the Census Bureau makes will hold,” the brief states.

Advocating for differential privacy, the Electronic Privacy Information Center filed a brief of its own. The group known as EPIC battled with the Trump administration over the addition of the citizenship question in the 2020 census, for instance.

As the Census Bureau asks about homeowner status, family relationships, sex and race, “the extraordinary reach of the bureau into the private lives of Americans brings extraordinary risks to privacy,” the Washington, D.C.-based group said.

People affected, according to the brief, could include same-sex couples concerned about discrimination and people living in housing units exceeding the occupancy limits for their leases.

Differential privacy can also be implemented more quickly than other disclosure avoidance techniques used in the past, EPIC said. Furthermore, the census count has historically contained errors.

“Given this, Plaintiffs have failed to demonstrate that any errors in privacy-infused redistricting data will be meaningfully different than errors in past census data,” it said.

A group of 16 states led by Utah filed an amici brief saying a batch of 2010 census data treated with the differential privacy technique made it impossible to make accurate districts at the local level.

“The analysis showed that when differential privacy was applied to the 2010 data, there was a statewide net loss of nearly 15,000 people from Utah’s cities and towns, including two cities that lost 50% of their populations,” the brief states.

Utah said distorted numbers would affect longitudinal studies, federal dollars sent to the state, and the resources the state distributed to its communities based on census data, such as for housing, transportation and emergency management. Differential privacy, the state said, would negatively affect minority and rural populations.

Meanwhile, Jane Bambauer, a professor of law at the University of Arizona, said the Census Bureau took an extreme position with differential privacy, calculating for every type of data leak instead of hewing towards a more accurate count by judging what information needs to be protected more.

Taking this approach, Bambauer wrote in an amicus brief, affects the census data journalists, researchers and nonprofits use to do their jobs.

“Political fights are already suffering from a dearth of shared facts,” Bambauer’s brief said. “By using differential privacy, the Census Bureau is putting one of the few sources of ground truth at risk.”

Exit mobile version