The U.S. Census Bureau announced March 27, 2023 that it will release files that the Election Law Clinic at Harvard Law School had sought through litigation in order to evaluate possible bias in the 2020 Census. 

The Election Law Clinic and Selendy Gay Elsberg PLLC represent Dr. Justin Phillips, a political scientist at Columbia University, in litigation to enforce a Freedom of Information Act (FOIA) request for the files. These data may reveal unintended distortions in 2020 Census data, which are used to distribute political power and resources and to conduct research across disciplines.   

Scholars have contended that the privacy protections adopted by the Census Bureau for the 2020 Census may have systematically skewed the data used for redistricting in a way that underrepresents people of color. In order to determine whether that skew exists, and if so, how large that bias is, Dr. Phillips had filed a FOIA request with the Bureau on July 7, 2022. 

Dr. Phillips’s FOIA request sought two files (called the “noisy measurements files”) that would enable outside observers to determine whether the published data from the 2020 Census are systematically skewed. The first is a demonstration file based on 2010 Census data. The second is the actual noisy measurement file used to generate the final 2020 Census data. 

Dr. Phillips sued the Census Bureau to enforce his FOIA request on October 31, 2022. On December 1, 2022, the Bureau denied Dr. Phillips’ request for the files, contending that the 2010 file had been deleted, and the 2020 file was exempt from release under FOIA. 

Reversing course, in January 2023, the Bureau announced it would recreate and publish the 2010 noisy measurement file. The 2010 file is due to be released on April 3. The Bureau further announced today that it will release the 2020 file, with a schedule for release still to be set. View the full statement here.

BACKGROUND

The decennial census is the backbone of our nation’s statistical infrastructure. Among its many uses, the census determines the legality of state and local redistricting plans, the distribution of public funding at all levels of government, and the results of important research projects conducted by academic researchers and government agencies. It is crucial that the census not only report the total population of each state as accurately as possible but also present a realistic picture of how a population is distributed at lower levels of geography within the states and across demographic categories. 

The census faces competing interests of accuracy and privacy. To maintain privacy, the Census Bureau uses a “disclosure avoidance system,” or “DAS”. This system masks identifying information within the data before releasing it to the public. In 2020, the Bureau adopted a new DAS in response to concerns about the privacy of the data. This new DAS added a pre-determined amount of random noise to the dataset. While this DAS may provide improved privacy compared to 2010, some experts have voiced concerns that the DAS amplifies existing disparities within the census, which persistently undercounts specific groups including the Asian American, Black, Latino, American Indian, and Alaska Native populations. The 2020 Census DAS may be biased in the direction of reducing the populations of racially and politically diverse neighborhoods and increasing the counts of homogenous neighborhoods.

To analyze the potential impact of the 2020 DAS on accurate population demographics, including the effect of post-processing, Dr. Phillips requested access to the noisy measurements file underlying the published 2020 Census data and the 2010 Census demonstration data product. The files requested through the FOIA are important because biases in the data can harm communities’ ability to obtain their fair share of government funding and their ability to enforce their civil rights, including the right to equal political opportunity under the Voting Rights Act. Bias in census data also interferes with academic research in health, public opinion, and many other fields.  Like much of political science research, Dr. Phillip's public opinion research relies on census data as a key input to accurately characterize population-level estimates. With the planning of the 2030 Census underway, understanding the biases accompanying the Census’s current methods is pivotal to improving the accuracy of its data, ensuring privacy, and promoting civil rights under the Voting Rights Act.

The complaint details that Dr. Phillips is concerned that his research has been harmed because the post-processing phase may have systematically inflated the census-reported populations of sparsely populated and homogeneous areas while shrinking those with greater population density and diversity. This distortion in the 2020 Census would make the data less fit for use in Dr. Phillip’s research while also resulting in an inequitable distribution of political power and resources, likely harming racial minority groups.