Differential Privacy: What Connecticut Data Users Should Watch
If you use the American Community Survey (ACS) to write a grant, conduct research, report to a funder or your stakeholders, or conduct planning, changes are coming that may impact the availability of the data you have grown accustomed to accessing.
According to a June 4 rule published by the Office of Privacy and Open Government at the Commerce Department, key statistical products from the Department will be returning to out-of-date statistical methods to protect privacy, which is likely to result in much less data being available for the public to utilize.
As a data user in Connecticut, here's what we know and what you can look out for with upcoming data releases.
Differential Privacy
First up, the most complex of these recent items: the June 4 ban on the application of differential privacy by the Census Bureau and Bureau of Economic Analysis. According to an article in Science, a recent directive bans these agencies "from using time-tested methods designed to allow the maximum amount of data to be released from national surveys while protecting the privacy of respondents."
Instead, these agencies must use out-of-date methods of coarsening, which "rounds or aggregates data, such as by using 5-year intervals rather than annual increments" (Science) or suppression, which means completely withholding data.
What is differential privacy?
Put simply, differential privacy is a method used to protect our privacy when data about us is released to the public. Because of the powerful computing power now available to so many people, without privacy protections, in theory, someone (a bad actor in particular) could compile information from various sources and identify individuals in the data. However, that has not happened yet.
Put slightly more technically, differential privacy is a type of statistical disclosure avoidance technique, which includes "data swapping and noise injection to safeguard against linking the data back to any specific individual" (from our piece on Census Privacy Protections).
This helpful video explains more.
Differential privacy seeks to balance our individual privacy with the utility of the data being released. If we tip too far into the individual privacy priority, then our data becomes completely inaccurate and useless. If we tip too far into the data utility priority, then it is easy for bad actors to compile that information and re-identify individuals.
Differential privacy has been a controversial subject in the data world, as it reduces the accuracy of data for small populations (especially as we get into disaggregations). Differential privacy is a compromise: by applying this method, we are all protected, and we get data that is fairly useful.
However, the application of these older methods is likely to result in much less data available to the public for use in decision-making, planning, and holding our elected officials accountable.
In our 2019 post on this topic, we shared information from the Census Bureau about the kinds of information that could be compiled on residents with data if it were not properly protected, and why moving to a more modern protection methodology was so important. An experiment by the Census Bureau found that they could correctly reconstruct data about individuals in all census blocks in the US.
As we learn more about the power of AI, and the lack of regulations and oversight into its development and deployment, the urgency of proper privacy protection is clear. However, the proposed changes do not increase protection but simply reduce the availability of data.
What products will be affected?
While we don't yet know the scale of the data that will no longer be available, John Abowd, the former Associate Director for Research and Methodology and Chief Scientist of the US Census Bureau, provided a list of the data sets that had previously had disclosure avoidance applied.
Most importantly, Abowd points out that "All tables and PUMS [Public Use Microdata Sample] from the American Community Survey" (ACS) will be affected. The ACS is the most commonly used Census Bureau product since it is updated annually. Again, we don't know how it will be affected, but we will keep you updated as we learn more.
What we do know — there will be a delay in releasing the ACS. Typically, by now, the Bureau would have posted the public release schedule, and it has not been published yet.
Other products that will be affected, according to Abowd, are:
Many tables from the Economic Censuses
All tables from County Business Patterns and Zipcode Business Patterns
All tables and microdata from the 2020 Census of Population and Housing, including redistricting data
Other Noteworthy Data News
Several years ago, CTData convened several groups to provide input into Statistical Policy Directive No. 15, the rule that determines the race and ethnicity questions and categories available on all census forms. We and others recommended changes that better reflect the way views of racial and ethnic identity have changed since 1997, the last update. The Census Bureau will be scrapping the implementation that has been underway for several years and will return to the 1997 standards. Census Bureau Quietly Scraps Plan for Improved Data Collection on Race and Ethnicity (TPM)
Recently the Equal Employment Opportunity Commission determined that demographic characteristics of employees, such as race and gender, are required to know whether there has been discrimination by employers. Employers will no longer be required to provide this information. Federal civil rights agency moves to stop requiring demographic data collection after 60 years (PBS)
Medicaid officials improperly shared data with Immigrations and Customs Enforcement (ICE) officials in January and then shared them with private surveillance company Palantir. While a US District Judge had ruled that some details could be shared, federal officials admitted to the judge in May that shared Medicaid data with ICE "that went beyond what the court order allowed." ICE shared Medicaid data it wasn't supposed to have with Palantir (NPR)
The Consumer Product Safety Commission, tasked with "tracking and issuing recalls of dangerous products sold in the US," has been pressuring hospitals to share "personally identifiable health data with a private contractor." The agency is seeking medical records from "emergency room visits for most injuries," as well as names, addresses, diagnoses, and other personal details" to private contractor, Konza Health. Trump administration demands hospitals share emergency room records (CNN)
How to Get Updates
As the Connecticut Census State Data Center, we will continue to stay on top of census news and share it as it becomes available. You can bookmark our blog or get updates by subscribing to our newsletter and following us on Facebook, and LinkedIn, and Instagram. To learn more about the Census and resources provided by CTData, head to our Census Data Portal. Explore other data sets and analysis at data by topic and data projects. Join us for our conference to connect with other data users, or learn some new data skills at our workshops.