HELENA — As the 2020 census gears up to produce the official demographic figures used for political redistricting, demographic research, and the allocation of billions of public dollars, state and federal officials are engaged in a full-court press to make sure the count reaches as many Americans as possible.
However, a “differential privacy” measure designed to ensure that individual information reported to the census remains private could muddle the figures ultimately reported for rural Montana communities. In some sparsely populated Montana counties, the official population counts could be skewed by 10%.
While the adjustment proposed by the U.S. Census Bureau, which would essentially inject small errors, or statistical “noise,” into published counts, wouldn’t have a substantial effect on overall statistics for Montana’s larger municipalities, it could leave residents of the state’s smaller towns lacking confidence in their official population counts.
A Montana Free Press analysis of comparative data released by the Census Bureau and compiled by the University of Minnesota’s IPUMS program also indicates the measure could affect how accurately Montana’s official census stats describe American Indian populations, which are a particular focus of get-out-the-count efforts.
In Vermont, a low-population rural state like Montana, nonprofit news outlet VTDigger reported Feb. 23 that local census leaders are pushing back on the Census Bureau privacy proposal, expressing concern that it will compromise the utility of data from the 2020 count.
“From what I’ve seen, it could drastically change the characteristics of our geographies,” Vermont State Data Center coordinator Michael Moser told VTDigger.
The Montana Department of Commerce, which includes the Montana Census and Economic Information Center, referred an inquiry about the issue to federal census officials.
“This issue is on our radar, but we are laser focused right now on making sure Montanans are educated about why responding to the Census is so important,” said commerce department spokeswoman Emilie Saunders.
“It is possible — even likely — that scientists, planners, and the public will lose the free access we have enjoyed for six decades to reliable public Census Bureau data describing American social and economic change.”
IPUMS researchers, in a December 2018 white paper.
In a statement, Census Bureau Chief Scientist John Abowd said the comparative data reviewed by MTFP was produced by an “interim version” of the DAS, or Disclosure Avoidance System.
“In its current iteration, the DAS does very well at ensuring the data’s fitness for use for some important use cases, and falls short in others,” Abowd said. “But we know that much more work needs to be done to identify and improve those aspects of the data that are not yet of sufficient quality.
“Releasing these demonstration products allowed us to crowdsource that process,” he continued.
By law, the Census Bureau has to keep individual census responses private, a mandate intended to encourage accurate counts by reassuring Americans who may be hesitant to trust the federal government with private information. Census workers can face up to five years in prison and a $250,000 fine for violating confidentiality protections.
Historically, the Census Bureau has released a wealth of public information following decennial censuses, including overall population counts as well as detailed information on race, age, sex, housing, and other data to help Americans understand their communities.
But with increasingly sophisticated computing technology widely available, census officials have become worried that it’s possible to take detailed information published for areas with relatively low population counts and reverse-engineer it to determine the age, race, and other information reported by specific individuals — a process one Census Bureau presentation likens to solving a sudoku puzzle. That information, including potentially sensitive responses about race or ethnicity, can then be linked to commercial databases, the Census Bureau says.
To prevent that, the bureau wants to add statistical noise to the census results it reports publicly, fuzzing the figures to frustrate marketing firms and others who might otherwise be able to crack the identity puzzle. Doing so, however, means making publicly available figures less reliable.
In larger cities, where the number of people means there’s less risk of identity-cracking, the noising algorithm won’t have as much impact on overall census results. However, it could affect reported counts in subsets of the data: for example, Native American women over age 65 in Rosebud County. And in low-population rural areas, particularly towns and counties with a few hundred people or less, the noising method produces substantial shifts in the official population counts.
In order to give a sense of the impact its noising algorithm would have, the Census Bureau produced figures showing what 2010 census data could have looked like with the algorithm applied. The University of Minnesota IPUMS researchers, who have expressed concern about the method, compiled that sample alongside the actual 2010 census figures to allow for easy comparison.
A Montana Free Press analysis of the comparison data indicates the noising method would have produced substantially different published census figures for portions of rural Montana had it been in place in 2010.
Golden Valley County’s official population count, for example, was 884 in 2010, but was pegged at 965 in the noised data, a 9.2% difference. Petroleum County’s official 2010 count was 494 residents, but it was listed as 544 people in the noised data, a 10.1% difference.
Small municipalities would have been subject to population uncertainty, too. For example, the town of Winnett, the seat of Petroleum County, was reported at 182 people in 2010, but would have been reported with the sample noising at 242, 33.0% higher. The city of Harlowton, the seat of Wheatland County, had an official 2010 population of 997 and a noised figure of 894, a 10.3% decrease.
The noising method would also make census counts unreliable for some American Indian populations. In Fergus County, including Lewistown, which the newly recognized Little Shell Tribe of Chippewa Indians describes as one of its population centers, the 2010 census counted 143 residents with an American Indian/Alaska Native racial affiliation. The noised figure, in comparison, indicates a Native population of 169, an 18.2% difference.
Published counts of Native residents could be noticeably affected by the noising method even in higher-population counties. Gallatin County, with the state’s fourth-largest city, Bozeman, and Montana State University, had an official count of 781 Native residents in 2010. The noised version of the count puts that figure at 821, a 5.1% difference.
In addition to helping researchers, policymakers, and entrepreneurs understand the state’s population trends, Montana data from the 2020 census will be used to redraw state legislative district boundaries and allocate nearly $2 billion a year in federal funds, excluding Medicare and Medicaid spending.
The proposed noising method is designed in a way that it won’t affect the overall state population count, meaning federal spending allocated to state government by population won’t be affected, nor will Montana’s chance of earning back a second seat in the U.S. House. Imprecise population data at the county or city level could, however, affect funds allocated to local governments and shift spending by programs that rely on population-derived statistics like per capita income.
The Census Bureau estimates that its data is used to distribute funds from 132 separate federal programs, including highway construction, unemployment insurance, the National School Lunch Program, Section 8 housing vouchers, crime victim assistance, and grants used by rural communities to help pay for water and sewer system upgrades.
“This issue is on our radar, but we are laser focused right now on making sure Montanans are educated about why responding to the Census is so important.”
Commerce department spokeswoman Emilie Saunders
Additionally, inaccurate counts would affect the work of Montana’s 2020 redistricting commission, whose work is based on census data and is required by the Voting Rights Act to avoid producing legislative maps that dilute American Indian voting strength.
At a national scale, demographic researchers have also raised alarm about the impact the proposed privacy measure would have on the usefulness of the 2020 counts and another widely used census program, the American Community Survey.
“Adoption of differential privacy will have far-reaching consequences for research,” the IPUMS researchers wrote in a December 2018 white paper. “It is possible — even likely — that scientists, planners, and the public will lose the free access we have enjoyed for six decades to reliable public Census Bureau data describing American social and economic change.”
Montana lawmakers allocated $100,000 in 2019 to fund census outreach efforts in the state. The Census Bureau also planned to hire hundreds of census takers in Montana and spend $240 million nationally on media buys promoting census participation.
The official 2020 Census Day is April 1, and the Census Bureau says American households will begin receiving mail with information about how to respond in mid March. Census takers will follow up with people who haven’t yet responded to the census in early summer, and official counts are set to be delivered to Congress in December.
This story is published by Montana Free Press as part of the Long Streets Project, which explores Montana’s economy with in-depth reporting. This work is supported in part by a grant from the Greater Montana Foundation, which encourages communication on issues, trends, and values of importance to Montanans. Discuss MTFP’s Long Streets work with Lead Reporter Eric Dietrich at edietrich@montanafreepress.org.