We use cookies to understand how the website is being used and to ensure you get the best possible experience.
By continuing to use this site, you consent to this policy. About cookies

Controversial Census Bureau Plan That Makes Data Less Accurate Goes to Court

SUNDAY, MAY 02, 2021

As some states celebrate and others reel from the release of the state population figures, a storm continues around the Census Bureau’s unprecedented and massively controversial plan that makes the Census data inaccurate, purportedly to make it less vulnerable to privacy attacks.  A hearing will take place Monday in an Alabama federal courtroom before a three-judge panel (with the possibility for a direct appeal to the U.S. Supreme Court). It could determine how useful the data will be when it is released and could therefore have an impact on the future of redistricting, federal funding, public policy decisions and more. (Filings for the case are available here.) 

Redistricters across the political spectrum, voting rights advocates, legislators, and legislative staff members, as well as mayors, town and village supervisors and other parties are very concerned about the accuracy of the soon-to-be-released Census data.  Letters expressing these concerns have been sent to the Census Bureau by many state demographers, a wide variety of researchers, redistricting professionals, civil rights organizations, and others.  Aside from the idea that the data may not be usable for its main intended purpose of redistricting, the Alabama case argues that the Census ignored the requirement to consult with the states and obtain agreement on the data to be released.  This consultation never happened, and the Census Bureau is relying upon its own assessment to decide what is released.  Disagreement over the new data plan exists within the Bureau as well — many internal Census documents submitted in the case show that Census staff were also quite concerned about this new approach to “protecting” census data.

The Census response, so far, has been a series of “demonstration products” based upon the 2010 Census, which researchers must assess to see if the latest version of these is “fit for use” as the Census describes it.  The most recent set of these was released on April 28, 2021. 

The first wave of detailed Census data is scheduled to be released in mid- to late- August. (This release is colloquially called the PL 94-171 files, named after the 1975 statute that set up the standards for constructing the files after each census.) The main official use for those data is to draw legislative districts from the congressional level down to the city and village council.  Such work requires adherence to population equality among districts, especially for Congress, where absolute equality of population is almost always required.  It must also be precise enough so that the districts comply with the Voting Rights Act, most particularly when localities may be required to draw so-called majority-minority districts, and also accurate enough so that one can determine whether voting in certain jurisdictions is racially polarized.

Having worked in a number of redistricting cases at the level of Congress down to the Village level and having used the Census data in many research projects and in over 100 court cases of various kinds, I decided to do an informal assessment of the new Census data release’s “fitness for redistricting.” I examined the data from Alabama (which brought the case against the inaccurate data) from the latest demonstration product and compared it with the 2010 Census. The results are not encouraging and raise serious questions about the accuracy of the data for redistricting, as well as how the data will be viewed when released to communities across the United States for redistricting and other purposes.

First, at the block level, the important level for redistricting that is used to draw lines and assess plans, the data seems to have been massively changed. In effect, massive numbers of people have been moved from block to block. The table below compares the 2010 data with the demonstration data, and as one can quickly see, there are substantial changes at the block level:

The total state population was the same for both the demo product and the 2010 Census and the other characteristics were virtually identical at the statewide level. But when one examines the blocks, we can see how much change was induced.  For instance, looking at the approximately 1.2 million non-Hispanic Black population in Alabama, about 69,000 were added to some blocks and a similar number were subtracted from other blocks, making for a total of about 11 percent of the Black population having in effect been moved around in Alabama. 

Obviously, such moves within the whole state cancel each other out, but that does not answer the question how this would have affected redistricting if these had been the data available in 2010.  This comparison reveals the important block level changes that occurred when the Census Bureau used its new technique. For 2020, of course, if the data are released with these block-level inaccuracies, we would not know how any plan or district, not to mention blocks, places or other geographies, would be affected, because the manipulated data would be all that was available.

For this reason, I decided to examine the situation of African-Americans in Alabama based upon these data.  Map 1 shows the distribution of the non-Hispanic Black population from the 2010 Census in the Montgomery, Ala., area.  This is an area with a relatively high Black population concentration.  The next two maps show the same area using the demonstration product data, displaying those blocks that have an increase in their non-Hispanic Black population and those blocks that have a decrease (See Maps 2 and 3).  In this area (as in all of Alabama) and presumably all the United States, much rearranging of the African-American population occurred.  How exactly this would affect redistricting, especially in terms of efforts to protect the voting rights of African-Americans, could not be determined in the time available since the data have only been released for a few days.  However, the potential for serious effects is obvious.  If the Black population were shifted one way, it might increase their likelihood of being able to argue for an African-American district; if the Black population were shifted another way, it might undermine their claim.  But if the 2020 data are released using the new technique, there would be no way to determine which groups were actually eligible for majority-minority districts. 

It seems likely that the various minority groups would be spread out more and might make it harder for them to garner a so-called majority-minority seat.  This could easily affect redistricting at all levels.

Map 1. Non-Hispanic Black Population at the Block Level from 2010 Census

Map 2. Increases in Non-Hispanic Black Population by Block from Demo Product Compared to 2010 Census as released.

Map 3. Decreases in Non-Hispanic Black Population by Block from Demo Product Compared to 2010 Census as released

Sparking added concerns is the fact that more 19,000 of the 74,000 blocks with no non-Hispanic Black population in 2010 had Black population added in the demonstration product (See Map 4).  For Hispanics, some 29,200 of the 107,000 blocks with no Hispanics but with other population had Hispanics added to them by the Census process. In short, the new Census method of manipulating data in the name of privacy protection generally seems to run a high risk of complicating the difficult task of redistricting, and, perhaps, makes it more difficult for minorities to achieve the goal of a seat that can “elect a candidate of choice,” a major provision of the Voting Rights Act. 

Map 4.  Blocks in Alabama with No Non-Hispanic Blacks in 2010 Census with One or More Non-Hispanic Blacks in Demo Product

Map 5. Blocks in Alabama with No Hispanics in 2010 Census that Had One or More Hispanics in Demo Product


The latest demonstration product plainly shows the difficulties using it for redistricting.  The problems from this admittedly very quick analysis are many, and include the following:

  1. The likely difficulty in being able to know where the percent of a minority group is very close to achieving a majority in a district.  It appears that the methods have the effect of moving population to less-concentrated areas, which would mean that some districts that seem to not have achieved minority-majority may — with the accurate data — have done so. 
  2. Complications in analyzing racially polarized voting.  The same issues of making the data inaccurate at the block level mean that the denominator used for the two methods of demonstrating racially polarized voting (homogenous precinct analysis and the various regression or regression-like methods) are seriously compromised.  In short, if a precinct denominator is not accurate, then the racially polarized voting analysis will be compromised.  The degree of the compromise will not be possible to glean from the released data.
  3. Because of the two issues enumerated above and others, it will be harder to choose exactly how to draw lines and exactly which areas should be included or excluded in each district.  It is also possible that in some cases (one will not know which ones), the total population of a district or a plan will meet or fail to meet various population equality thresholds.  It will also be difficult to trade off the various criteria for acceptable plans since all the typical numeric thresholds will not be computed accurately.
  4. It will make sharing plans difficult for those overseeing their drawing, as well as with various community stakeholders.  Displaying maps using blocks that do not accurately portray the ground truth can only lead to confusion and, perhaps, controversy and distrust.
  5. Finally, the new Census Bureau method seems to have other problems, including an inability to create tables that include both persons and households within them (e.g., number and age of persons 17 or younger in a household, tables that are scheduled for the release after Pl-94-171). 

Given these inaccuracies and limitations, whether this planned method is “ready for prime time” seems very questionable. What is not in question is the large number of states and stakeholders that have tried to alert Census Bureau leadership about the extreme risk that applying this untried method holds for redistricting and other purposes that depends upon the Census for accurate and reliable data. 

(The opinions expressed in this article are those of Andrew Beveridge and may not represent the views of Social Explorer. His bio is available here.)

Author: Andy Beveridge

Data insights are waiting to be uncovered
Get Started

Already using Social Explorer? Log in.