Privacy algorithms used for cybersecurity at the heart of latest criticism of 2020 census data

[ad_1]

The head of the US Census Bureau has defended a new tool intended to protect the privacy of people taking part in the statistical agency’s questionnaires against calls to abandon it by leading researchers who claim it jeopardizes the utility figures that are the basis of the country’s data infrastructure.

The tool known as Differential Privacy “has been selected as the best available solution” against efforts by outside groups or individuals to piece together the identity of participants in the bureau’s censuses and surveys using third-party data and powerful computers, US Census Bureau Director Robert Santos said in a letter last week. Privacy concerns have increased in recent years as cyberattacks and threats of misusing personal data have become more common.

Several prominent demographers and academic researchers in the state had asked the statistical agency in August to drop the use of differential confidentiality on future annual population estimates, which are used in the distribution of 1.5 trillion dollars in federal funding each year, and future releases of American Community Survey data, which provide the most comprehensive information on how people live in the United States

Demographers and researchers said applying the secrecy method for the first time to 2020 census data delayed their release and created inaccuracies in the numbers used to determine political power and distribute federal funds. The researchers said in their letter that there are thousands of smaller jurisdictions across the United States that will not get usable data due to algorithms applied to the numbers to protect privacy.

By continuing to use differential privacy algorithms, “the Census Bureau risks failing in its responsibilities as the federal statistical agency to provide relevant, accurate, timely, and credible information for the public good,” the researchers and researchers said. demographers. “In fact, the experience of the past few years has undermined user confidence in the Census Bureau.”

Differential privacy algorithms add intentional errors to data to mask the identity of a given participant and are more noticeable in smaller geographic areas, such as census blocks. Data used to determine how many congressional seats each state gets and to redraw political districts was released last year, but more detailed 2020 census numbers won’t be released until next year, nearly three years after their collection.

Some biases using the privacy tool “were unavoidable from a purely mathematical point of view”, but the bureau’s statisticians worked to minimize them, and delays were caused by the pandemic, which pushed back a series of publications from the 2020 census data, Santos said.

Meanwhile, the bureau’s watchdog said in a report last week that the statistical agency had failed to stop simulated cyberattacks it had carried out as part of a covert operation to test the desktop cybersecurity vulnerabilities. The U.S. Commerce Department’s Office of Inspector General said its team gained unauthorized access to a domain administrator account, obtained personally identifiable information about office employees, and used insecure programs to send fake emails.

The Census Bureau said in a response to the report that the exercise helped it improve its cyber defenses.

[ad_2]