New census privacy formula could affect redistricting, but opinions differ
Illustration by VlatkoRadovic | iStock/Getty Images Plus
A new effort by the U.S. Census Bureau to protect people’s privacy could make it harder for the Arizona Independent Redistricting Commission to draw districts primarily composed of people of color that will withstand litigation under the Voting Rights Act.
Federal law requires the Census Bureau to protect the identities of individual respondents to each decennial census for 72 years. In order to meet that requirement in the face of computing and other technological advances that make it easier to glean individual identities from census data, the bureau began using a system known as differential privacy for the 2020 census.
Differential privacy is a mathematical system that uses arbitrary changes in data to prevent the identification of individuals within a dataset. What that means for redistricting is that the Census Bureau will add “noise” to change and obscure some data in order to prevent people from using census data to identify individuals. For example, if all residents of a census block are white except for one who is Latino, that person would be easily identifiable using census data. Differential privacy would obscure that information so that the person couldn’t be identified individually.
A federal appellate court last month rejected a request by the state of Alabama to bar the Census Bureau from using differential privacy in census data. The appellate court also rejected the state’s request that the bureau be forced to move up the release of the census data that states need for redistricting. The bureau now plans to release that data, in an archaic but usable format, on Aug. 16.
The commission on Tuesday heard competing presentations from two redistricting efforts about the effect that differential privacy will have on its efforts.
Thomas Bryan, the founder and CEO of GeoDemographics, a demographic and analytic consulting firm, told the commission that differential privacy could cause problems for the AIRC if it faces lawsuits over majority-minority districts, or districts where the majority of people are minorities.
Differential privacy can create strange anomalies at the census block level. Bryan showed the commission data from a Census Bureau demonstration project in the Alabama litigation that showed blocks in Arizona where all residents were listed as being under 18 years or age, blocks with more than 15 people per household, or blocks with no population at all.
But the biggest problem for redistricting is how differential privacy changes population data at the block level on racial and ethnic demographics, which Bryan said disproportionately affected minority populations. Of nearly 242,000 Arizona census blocks in the demonstration project, Bryan said differential privacy changed about 6,000 from being majority white to majority non-white.
Section 2 of the Voting Rights Act requires districts where minority voters have the ability to elect the candidates of their choice. States and other jurisdictions often accomplish this through the creation of districts where minorities make up the majority of a district’s voters.
Those population disparities are concerning, Bryan said, because there’s no way to know exactly how or how much the population numbers in any given block have changed.
“We’re not going to be sure when you craft a district that it is what you say it is, because we know that the data for those building blocks is wrong,” Bryan told the commissioners.
Moon Duchin found differential privacy to be much less concerning from a redistricting standpoint.
Duchin, a researcher at the MGGG Redistricting Lab at Tufts University, took a broader view of differential privacy’s effects and urged the AIRC to look at it in a way that was much more reassuring. While differential privacy changes the data at the census block level, she said, the population statistics are accurate when aggregated into larger units.
“The whole design of differential privacy is many small changes to the smallest units so that, by the time you aggregate it up, things look really good,” Duchin said.
That held true even with much smaller districts than the commission will draw, Duchin said. Arizona’s congressional districts will each have about 795,000 residents, while legislative districts will average nearly 239,000 people.
Duchin presented the commission with data from a series of theoretical county supervisor districts in Navajo County with populations of 20,000. After adding “noise” from differential privacy, those districts saw only a half-percent error rate in the Native American populations of those districts. Based on the amount of change the Census Bureau is applying to blocks using differential privacy, Duchin said they produced “nearly invisible levels of discrepancy” in the Navajo County districts.
“The big take-home message is that, for every geography we considered … we concluded that the census data are clearly adequate for every redistricting application that we considered,” said Duchin, who added that her organization found no threat to Voting Rights Act enforcement from differential privacy.
Bryan countered that the aggregate numbers Duchin touted wouldn’t matter much in Voting Rights Act litigation.
For plaintiffs to prevail in Voting Rights Act challenges to districts, they must show that it was possible to create majority-minority districts that weren’t ultimately drawn. That makes the precise location of minority populations important, Bryan said. In districts with very small majorities of minority populations, he explained that differential privacy can be the difference in whether a district qualifies a majority-minority.
Bryan predicted that the advent of differential privacy in the census will end the practice of drawing such districts and force mapmakers to use a “higher bar” to ensure their districts can withstand litigation.
“I understand that in aggregate, if you aggregate a thousand blocks, it’s going to look great. But in litigating these cases, we don’t have the luxury of just saying we want everything in aggregate. We have to get down to the scalpel, look at small, individual pieces of geography and say we want this one in, we want this one out,” he said.
Commission Chairwoman Erika Neuberg noted that differential privacy could be a bigger problem for congressional than legislative districts. While the courts allow some deviations in the populations of a state’s legislative districts, within about 5% of the average, no such aberrations are permitted in congressional districts, which must have almost the same exact populations within a state.
“It sounds like the smaller, narrower group is going to be more at risk from the noise,” Neuberg said.
But even if differential privacy raises problems for the AIRC, Commissioner David Mehl questioned whether there was other data it could rely on instead.
“Is there any reason we should hesitate to use the census data, and if so, what would be the alternative?”
The only possibility either of the AIRC’s experts raised was the Census Bureau’s American Community Survey, which collects more limited amounts of population data used to craft regular estimates between census years. Duchin noted that some states have already begun redistricting work using such data, while Bryan emphasized that the data are only estimates based on 10-year-old information, and there are sometimes notable differences between official data and the survey’s data.
Using the American Community Survey’s estimates for Arizona could be especially tricky, given how significantly they differed from the official census numbers. The ACS estimated that Arizona’s population had reached 7.4 million in 2020, but the census pegged the state’s population at less than 7.2 million. The difference cost Arizona a widely expected 10th congressional seat.
The differential privacy issues may matter less for Arizona than they would have in previous redistricting cycles. For decades, Arizona was subject to a provision of the Voting Rights Act known as “preclearance,” which required the U.S. Department of Justice to approve all changes to election and voting laws, including new districts. But the U.S. Supreme Court in 2013 struck down the Voting Rights Act formula that determined which jurisdictions were subject to preclearance, lifting the requirement from Arizona and other states.
The commission’s Tuesday meeting was cut short due to a power outage that hit several buildings in the state Capitol complex, including the Arizona Department of Environmental Quality building, where the AIRC is headquartered. The remainder of the meeting will take place at 8 a.m. Monday, followed by the commission’s regularly scheduled meeting on Tuesday.
Our stories may be republished online or in print under Creative Commons license CC BY-NC-ND 4.0. We ask that you edit only for style or to shorten, provide proper attribution and link to our web site. Please see our republishing guidelines for use of photos and graphics.