Expert: Census Bureau’s actions to protect individual privacy could have big impact on redistricting

Research suggests that technology moves will cause NC communities that are more integrated and diverse to be undercounted

In a nod to privacy concerns, technology experts at the U.S. Census Bureau are using newly developed algorithms to inject statistical “noise” into this year’s census that will make it harder to identify individuals.

Independent researchers, however, are worried that the privacy protection mechanism — known as “differential privacy” — may cause significant overcounts and undercounts for purposes of legislative redistricting.

Researchers voiced their concerns after the Census Bureau published the last testing data in late April ahead of the redistricting data release. Last week in an apparent response to those concerns, Census Bureau announced that it will reduce but not eliminate such noise in the data.

However, the data compiled with the new method won’t be released until September — after the expected date the information was to be used for redistricting.

Concerns over inaccuracy of data to be used in redistricting

Harvard researchers analyzed the April data for four states. In North Carolina, researchers found more racially diverse precincts lose 20 to 30 people on average, while less diverse precincts gain 10 to 15.

The lead author of the Harvard study, Christopher Kenny, a doctoral candidate at the Department of Government told Policy Watch that the difference between the population counts in North Carolina precincts in the proposal compared with the actual 2010 census count is “very stark.”

The imbalanced skew of population counts for communities with different racial compositions means that some voters’ power will be diluted while others will be amplified.

“Minority voters in those districts that have large mixed minority precincts and areas will have a less powerful vote because there will be more people voting than in the other areas,” Kenny explained.

There are more than 2,600 precincts in North Carolina. Each averages over 3,900 residents, based on the census count for congressional apportionment in April, which didn’t use the privacy protection algorithms.

The 2010 census used a different mechanism for privacy protection. At the time, population in a census block — the smallest geographic unit of the census tally — was not altered. The differential privacy proposal for the 2020 census though, sometimes changes the total counts of population in each census block.

The impact of this deviation, the analysis suggests, is that map drawers may not be able to reconfigure districts that reflect the true populations, as mandated by the “One Person, One Vote” principle affirmed by the U.S. Supreme Court in the 1960s.

In addition, the privacy protection mechanism will result in white people being better counted than other groups that aren’t as large, because it counts the largest racial and ethnic groups more accurately, Kenny said.

“Once they release the 2020 data, they’re not going to release other accounts for you to compare it to,” Kenny explained  — noting that researchers can examine the accuracy of the algorithms only in September using the 2010 census.

Meanwhile, in Alabama a lawsuit is challenging the release of the 2020 census in the format in which differential privacy is applied.

Census Bureau takes steps to address concerns

The Census Bureau said in a Wednesday press release that the final redistricting data products will reduce the statistical noise from the April testing data, after accounting for feedback from groups like Kenny’s team. However, the agency said the population count accuracy won’t increase.

The release also says census experts are addressing concerns about the racial bias and the geographic bias that gives smaller, low-turnout and mixed-party geographies a disadvantage. Yet the results can only be examined after the September quality measurement data release.

“The decisions strike the best balance between the need to release detailed, usable statistics from the 2020 Census with our statutory responsibility to protect the privacy of individuals’ data,” said Ron Jarmin, acting director of the U.S. Census Bureau in the press release. “They were made after many years of research and candid feedback from data users and outside experts – whom we thank for their invaluable input.”

Kenny said he remains skeptical that the Census experts can resolve all the concerns with the short time it has to prepare for the newer data release. “I am hopeful that their changes will help fix some problems, but I can’t be confident that it will fix the scale of the problems that many users identified,” Kenny said in an email Friday. “It remains possible that their fixes to these few problems have introduced new errors on other dimensions.” He explained that if researchers identify these problems in September, many states will have already drawn their maps based on the problematic counts.

Kenny’s team issued a statement of its own on Wednesday in response to the Census Bureau announcement asking for greater transparency. “Although scholarly communities have not fully resolved the issue of incorporating additional noise into redistricting simulation analysis, the availability of such information should facilitate future methodological development,” the statement read. “In particular, it is of interest to examine whether or not the additional noise makes it more difficult to detect partisan and racial gerrymandering.”