New 2020 census privacy method makes it impossible to find people and homes.

Jessica Stephenson, a three-bedroom colonial-style home in Milwaukee, has been bustling with activity for six years. It is filled with children's chatter at the daycare center she runs from her home.

New 2020 census privacy method makes it impossible to find people and homes.

According to the U.S. Census Bureau, no one lives in that area.

Stephenson stated, "They should go and see it for their own eyes."

The Census Bureau used a new method to protect confidentiality in 2020 census. It made people and occupied houses disappear -- at least on paper -- in order to make her neighborhood, which is majority Black in Wisconsin, accessible to Hasidic Jews in New York's Catskill Mountains, accessible to a park in Tampa, Florida.

This is not a magic trick, but a new statistical technique the bureau uses called differential privacy. It involves adding errors to data to hide the identity of any participant.

Officials at the Bureau say privacy is essential in an age of sophisticated data mining. Technological innovations increase the risk of people being "re-identified". This can be done by using powerful computers to match census data with other public databases. Census answers must be kept confidential by law.

However, some demographers and city officials believe it is too far from reality. This could lead to errors in data that are used to draw political districts or distribute federal funds.

One analysis shows that differential privacy could lead to minority communities being penalized by undercounting areas which are racially or ethnically mixed. Harvard University researchers discovered that this method makes it harder to create equal-sized political districts and could lead to fewer districts with majority-minority populations.

For its part, the Census Bureau argues that the data are just as accurate as previous censuses, and that low-level inaccuracies do not pose a major problem.

One thing is certain: the method can produce strange, contradictory, and false results at the smallest geographical levels such as blocks in a neighborhood.

According to official census data, 54 people reside in Stephenson's Census Block in Milwaukee. However, there are no occupied houses. The streets are lined with cars, and some of them date back over a century. However, almost two dozen houses live on the streets. According to census data, 48 of the residents of the block are Black. However, it is difficult to determine if this is true due to differential privacy.

Another case is that the census does not list any people as living in Flatwoods Conservation Park, outside Tampa. However, it states there is a home that is occupied. Todd Pratt, a Hillsborough County spokesperson, said that two county employees reside there while providing security for the park.

In a small group of Hasidic Jews living in Kiamesha Lake in New York, 81 residents are listed, although the census says that there are no occupied houses. Sullivan County property records reveal almost a dozen homes whose owners have ties with the Vizhnitzer Hazidic community.

Planners in small communities and city managers have been plagued by unreliable data, which has caused problems. They fear that the data may not be reliable enough to make informed decisions. Eric Guthrie is a senior demographer with the Minnesota State Demographic Center. He said that he was contacted by half-dozen state managers who were concerned about the potential impact on federal and state funding.

Guthrie stated, "I explain to the them that there is no way to correct it, and that it's a mistake in the traditional sense." "The bug is there by design."

A wider lens helps to see the scale of these changes clearly. Florida is the third-most populous state in the country, with over 21 million residents. The 2020 census identified 15,000 blocks of neighborhoods as having 200,000 residents. However, there were no occupied houses. Rich Doty, a researcher demographer and coordinator of geographic information systems at the University of Florida's Bureau of Economic and Business Research, stated that 1,200 of 484,000 blocks in the state had occupied homes, but no residents.

Doty stated that "we expected these anomalies" as we had been warned by the Census Bureau, and other states. "We didn't anticipate this many."

Ron Jarmin, the acting director of the Census Bureau, warned that the application could produce "fuzzy" figures at neighborhood block level. He urged data users not to combine blocks in order to obtain accurate results. The bureau claims that the quality of 2020 data is comparable to previous censuses, which were based on data quality measurements.

This claim is difficult to assess since the raw data are not being made public without differential privacy, according to Stefan Rayer, a University of Florida Demographer.

Rayer stated, "We have to believe their word for it."

The Harvard researchers used test data to determine that differential privacy was more likely than mixed-race or mixed-partisan precincts to undercount, "yielding unpredictable and racial biases" because it prioritises the accurate population count for the largest race group within a given area.

They stated that their findings "underline the difficulty of balancing accuracy with respondent privacy in Census," in a report.

The Census Bureau disagrees and the courts have not found any reason to stop it.

The state of Alabama unsuccessfully challenged differential privacy earlier in the year. John Abowd (Census Bureau chief scientist) declared that the data were "extremely accurate" and stated that the use of differential privacy did not show bias in regard to racial minorities.

Abowd stated that redistricters could be confident in the accuracy and demographic characteristics of the voting areas they draw, despite noise from individual building blocks.

Some people don't believe the technique is right to protect confidentiality.

In a paper published by the University of Minnesota, two researchers claimed that the Census Bureau experiment did not show any genuine threats to confidentiality. They also stated that any risk of re-identification was similar to random guessing of household characteristics.

Steven Ruggles, a demographer, stated this month during a presentation that the Census Bureau's fear about re-identification, and the resulting justification to use differential privacy, could undermine confidence in census data.

Ruggles stated that it should not be used to justify the destruction of the country's statistical infrastructure. "All of it is likely to backfire."

You need to login to comment.

Please register or login.