Social Explorer is calling on its users – and anyone else who needs accurate and timely demographic data – to sign an appeal asking the U.S. Census Bureau to stop the use of its destructive differential privacy method for public data.  The Census 2020 redistricting file, the only file so far released has built-in errors, that the Bureau itself cautions against using the Census for block data and other small counts and small area data.  Consequential biases have been found in the redistricting file, as well.  The rest of the data has been massively delayed (so far, almost two years late compared to 2010) and the data will have much less detail and many fewer tables than released from the 2000 and 2010 Census, and nonetheless still not be as collected, but rather will still be inaccurate in many ways.  Social Explorer depends on these data, as do literally thousands of users and millions of citizens in the United States.

The problems and potential issues for the future of demographic data from the Census Bureau have become so profound that one of the main official bodies of Census users, the steering committee of the Federal-State Cooperative for Population Estimates (FSCPE), has organized the letter to the Census Bureau Director Robert Santos, expressing concerns and outlining a set of recommendations to mitigate some of the damage and guard against it happening again to other data collections and the 2030 Census.  The letter will be available to sign until August 22nd.

Find Information on the letter and how to sign it here. 

The new method massively changed the traditional Disclosure Avoidance System, which was used in 1990, 2000 and 2010.  The new method operates by changing the locations of many individuals, as well as their characteristics (sex, age, race, Hispanic status).  This noise injection makes the published census results substantially different than the data as collected.  Thus, it modifies the results purportedly to avoid any possibility of disclosing easily discovered information.  Significantly, there are no examples of users attacking the Census data to uncover such information, and as a recent Proceedings of the National Academy Sciences publication shows that:

1. The purported protection is no better than that provided by the old method in 2010 and earlier.
2. That the new method gives little or no protection in any event, and was based upon a still unpublished analysis of the risks.  

The Census Bureau has damaged the integrity of the decennial headcount. Ramifications could include the creation of unrepresentative political boundaries; government programs and policies designed for people who do not need them while ignoring the intended audiences; and flawed research outcomes for business, professional, academic, and non-profit users.  Even state and local policymakers also stand to be adversely affected by the ill-considered policy.  As does any institution that uses Census data for comparative purposes, such as the health system, the educational system, the criminal justice system, as well as survey researchers in the Bureau and elsewhere, who rely upon the Census results to design samples, and adjust results to the Census.

There are similar disclosure avoidance plans for the Census Estimates Program and the American Community Survey, the main survey that is done by Census Bureau, which has information on many important topics.  

The problems and potential issues for the future of demographic data from the Census Bureau have become so profound that one of the main official bodies of Census users, the steering committee of the Federal-State Cooperative for Population Estimates (FSCPE), has organized the letter to the Census Bureau Director Robert Santos, expressing concerns and outlining a set of recommendations to mitigate some of the damage and guard against it happening again to other data collections and the 2030 Census.

As noted, that the differential privacy policy seeks to address a problem that does not exist and has not existed for more than eight decades.  The Census Bureau is legally bound to keep individual-level data secure and with government agencies, as well as others.  If the Census Bureau fears disclosure, it has an information technology problem that should not be covered up with a demographic methodology solution.

The 2020 Census – theoretically, a snapshot of the population taken on April 1, 2020, that asked the gender, age, ethnicity, race, household relationships, and housing tenure of every American – is, of course, a crucial tool for people who use data.  But it is an extremely limited set of data, and it suffered from more limitations in 2020 than previous efforts.  Besides the use of differential privacy to compromise its results, the 2020 Census endured:

• A hostile president and administration who understood very little about the Census, other than the potential to use it for political advantage. 

• Revisions to racial and ethnic classifications that rendered comparisons to previous years nearly impossible; and

• A global pandemic.

The results of its mismanagement are still playing out and may not be known for several more years.  A number of headlines dominated coverage of the decennial headcount, but many poor decisions were made without public knowledge or discussion.

If the goal of the previous presidential administration were to use changes in methodology to damage the integrity of the decennial headcount, it may well have succeeded.  A post-Census analysis found the Latino, Black, and Native American populations were undercounted, while the white and Asian populations were overcounted.  Obviously, this had ramifications that may have altered the allocation of congressional seats and hence, the balance of political power in the U.S.

The result is a bleak picture for demographers and anyone who uses the results of the decennial Census.  Already, private companies are jockeying to provide their own estimates as a substitute for unreliable government data; the results are likely to confuse an already muddy situation.  Fortunately, the Census Bureau is working to provide its own alternatives, based on its annual series of population estimates.  (However, there seem to be plans to do similar damage to them, as well).

The population estimates are likely to also improve the American Community Survey, the Census Bureau’s flagship program that survived the Trump administration and a major database available in Social Explorer, an award-winning website that allows users to create customized reports and visualizations based on Census data, as well as other sources. 

Abandoning its traditional methods to appease political opportunists with completely unnecessary and useless changes under the banners of privacy protection and policies that seek to maximize the power of individual racial or ethnic groups, will leave the field open to private data brokers, whose interests will clash with the public’s ability to find and access free, unique, accurate, and actionable information about their communities.

Slides from the Recent Joint Statistical Meeting are here.

The Associated Press Coverage of these issues is here.

Andrew Beveridge is Professor Emeritus of Sociology at Queens College and Graduate Center, The City University of New York.  Since 1993 he and now the team at Social Explorer have assisted the New York Times in its coverage using the Census and other demographic and social science data.

Author: Andrew A. Beveridge, Co-founder, President of Social Explorer