Policy Issues / Election Integrity

Policy Brief: The Census Bureau Is Defrauding American Voters

The elimination of differential privacy is of paramount importance for restoring political power to citizens and ensuring that the will of voters is upheld. It is the main cause of the error-riddled 2020 Census and operates as the “ace up the sleeve” for those who want to mask the citizenship status of individuals who are counted.

Synopsis

The Census Bureau is in dire need of immediate reform. Two significant problems must be addressed, both of which undermine both the purpose of the census and the very essence of self-governance within our republic. The first is that the most recent census, in 2020, made significant counting mistakes that marred the electoral and congressional apportionment process for numerous states, thereby erroneously distributing political power. Second, the bureau perpetuated a unique data falsification process known as differential privacy, ostensibly to protect people’s identities, that distorted the census process and tilted the political scales favorably toward Democrats for redistricting purposes. The implementation of differential privacy has made it difficult to ascertain the citizenship status of those counted, guaranteeing that noncitizens will continue to be included in the decennial census so long as this murky algorithm remains in use. 

The bureau has followed a methodological approach that justifies intentional deception to manipulate data, defraud voters, and favor progressives and reward noncitizens at the expense of Americans. Commerce Secretary Howard Lutnick, who oversees the Census Bureau, has a once-in-a-lifetime opportunity to reform the census process to not only ensure that all Americans are accurately counted and properly represented but also guarantee that the census serves the interests of citizens and citizens alone.

Overview: How the Census Operates

The census is a constitutional mandate that requires a population count every decade to facilitate representation in the House of Representatives and apportionment within the Electoral College. The authority to conduct the census is found in Article I, Section 2, Clause 3 of the Constitution—often referred to as the Enumeration Clause—and empowers Congress to direct the process through statutory directives.1 The relevant portion reads as follows:

Representatives and direct Taxes shall be apportioned among the several States which may be included within this Union, according to their respective Numbers, which shall be determined by adding to the whole Number of free Persons, including those bound to Service for a Term of Years, and excluding Indians not taxed, three fifths of all other Persons. The actual Enumeration shall be made within three Years after the first Meeting of the Congress of the United States, and within every subsequent Term of ten Years, in such Manner as they shall by Law direct.2

The ratification of the Fourteenth Amendment superseded the three-fifths compromise language in Article I, Section 2, providing full representation to newly liberated slaves.3 Following the Constitution’s requirement for Congress to direct the census process by law, the statutory underpinnings for the census are currently found within Title 13 of the U.S. Code.4 

Importantly, every census begins with creating a map of the entire country. This map is then divided into what is known as census geography or a census geographic hierarchy. First, a map is made of each state with its existing counties and political lines. Next, the counties are divided into what are known as census tracts. The census tracts are further divided into block groups, and lastly, the block groups are divided into blocks. This geography will always nest. This means that the process can be reversed: The blocks can be aggregated into block groups, and the block groups can be aggregated into tracts, and so on. This map is known as the TIGER file within the Census Bureau.5

The Census Bureau reports data on political units that are not part of the census geography. These include municipalities, metropolitan statistical areas, census places, and, perhaps most importantly, voting districts (VTDs). VTDs represent precincts and wards where people cast their ballots and election results are reported. This is called the political geography or political geographic hierarchy. The political geography does not nest. Political geography is only symmetrical with the smallest element of census geography: the census block. 

“Symmetrical” means that the blocks can be aggregated so that they fill in the boundaries of the political geography in question without leaving gaps of uncovered territory or overlapping the boundaries of that political geography. Blocks are the only piece of census geography that can do this. Block groups and tracts will both overlap the political boundaries and potentially leave gaps. This is particularly so for VTDs, which are the smallest unit of political geography and the one most important to political analysis and reapportionment, including redistricting. The only way to obtain an accurate census population for VTDs is to have the most accurate block-level data possible.

After the census geography has been updated, a list of every housing unit in the country must be constructed, and each housing unit must be accurately placed in its appropriate census block. This list includes group quarters institutions (college campuses, military installations, nursing homes, homeless shelters, prisons, etc.). The bureau acquires this data through administrative records. It combines lists from the U.S. Postal Service, local and state agencies, Native American tribes, and other entities. Ideally, this list would be the final and complete list of housing units in the nation. But errors have been discovered during the post-census review called the Count Question Resolution (CQR).6

When it comes time to count people, the bureau gets reports from the group quarters institutions (GQRs).7 The officials of each GQR institution report their counts with identifying information. This information is critical for deduplication. For example, if a student’s family claims that the student lives at home, but the school also reports him or her as living in a dormitory, the Census Bureau will use the identifying information from the household return and the GQR, as well as administrative records, to verify that this is the same person. The bureau will then use its household resolution rules to delete one of the responses and place the respondent either in the household or the dormitory. Once a housing unit has what the bureau believes to be the proper number of people in that housing unit with the respondents’ characteristics, it considers the housing unit “resolved.” 

Individual households largely self-report either online or by returning a mail-out/mail-back questionnaire. In 2020, about two-thirds of households self-reported by going online or mailing back a questionnaire. The questionnaire includes the total number of residents and characteristic information for each resident. To count the remaining third of nonresponding housing units requires the bureau to hire tens of thousands of census workers to go to the nonresponding housing units to get the information in person. These are the non-response follow-up teams. 

In most cases, nonresponding housing units are counted by talking to someone who has lived in the household. Occasionally, the information comes from a neighbor. Finally, some information is gathered through the use of administrative records, i.e., data that exists in a governmental or other official database.

After the data has been collected, the bureau must deduplicate responses and resolve households. As described above, this process relies heavily on identifying data that is in the questionnaire and on administrative records. The bureau tabulates this data for the various levels of census and political geography. It is required to report the state-level population data by December 31 of the decennial year. This report is then used to calculate the number of congressmen each state shall receive based on the mathematical method of equal proportions. 

Next, the clerk of the U.S. House of Representatives sends a certificate to each state informing them of the number of congressmen who will be seated in the next Congress. Normally, after the state totals are released on December 31, the detailed data (sometimes called the PL 94 data)8 for the lower levels of census and political geography will be released on a state-by-state basis. This information is usually released beginning the first week of February and ending before April 1 of the first year after the census year. By statute, the state-by-state release must be completed by the first day of April.9

Once the PL 94 data is released, state and local jurisdictions review it, including both the population counts and the housing unit counts, along with their locations. If the state and local jurisdictions detect what they believe to be errors, the state and local jurisdictions report those errors to the bureau using a CQR report. The bureau then reviews that report and determines whether the state or locality has identified an error. If it is determined that there is an error in the Decennial Census, the bureau releases an errata sheet correcting the error. This process is normally completed by the third year of the decade. 

Background: The Error-Riddled 2020 Census

In 2022, the Census Bureau admitted that its population count was erroneous in fourteen states—an astonishingly high error rate.10 In eight states, the bureau overcounted the population. This occurred in Delaware, Hawaii, Massachusetts, Minnesota, New York, Ohio, Rhode Island, and Utah, with the error size ranging from 1.5 percent (Ohio) to 6.8 percent (Hawaii).11 In six states, the bureau undercounted the population. This occurred in Arkansas, Florida, Illinois, Mississippi, Tennessee, and Texas, with the error size ranging from −1.92 percent (Texas) to −5.04 percent (Arkansas).12

The result of these mistakes was a net benefit of at least six congressional seats for Democrats. Florida lost out on two additional seats that it should have gained. Texas was deprived of an additional seat that it should have gained. Colorado gained a seat that it did not deserve. And Minnesota and Rhode Island each maintained a congressional seat that should have been lost.13 

The Post-Enumeration Survey (PES), which is conducted after the decennial census to assess the overall accuracy of the census, claims that the errors were partially due to an overcounting of homeowners (0.43 percent) and an undercounting of renters (−1.48 percent), as well as significant mistakes regarding group quarters populations.14 The Census Bureau based on the 2020 PES that the unique circumstances of the COVID-19 pandemic presented substantial challenges and disruptions for counting individuals in group quarters, such as college dorms, nursing homes, and correctional facilities.15 Specifically, nursing homes were closed beginning in March 2020, and many colleges and universities shuttered dorms and sent students off campus.

It remains unclear whether the bureau counted empty dorms as full or whether another illegal methodology was implemented, such as adjustment,16 which could have significantly altered the numerical head counts. There has yet to be a full-scale investigation into these errors.

When the bureau attempted to close gaps in the group quarters assessment at the end of 2020 and well into 2021, many problems were uncovered, including missing facilities, inaccurate resident counts, and incorrect geographical placement.17 Inaccuracies were so significant that the bureau released a onetime Post-Census Group Quarters Review (PCGQR) in June 2022 to correct its mistakes, but it revealed that the corrections would not be used to alter the 2020 Census for purposes of reapportionment and redistricting.18

It is wildly improbable that the bureau’s combination of overcounts and undercounts, whether a result of group quarters inaccuracies or other anodyne mistakes, would so significantly benefit a single political party. And yet, that is precisely what occurred. A letter from Rep. James Comer (R-KY), the chairman of the House Oversight Committee, highlighted the fact that the preceding census, in 2010, did not have statistically significant miscounts.19 No other census in modern history comes close to the statistical errors found in the 2020 Census when factoring in its impacts on apportionment and redistricting. Yet the bureau reviewed its work and thought it did a great job: “We assessed that the 2020 Census was fit for the purposes of apportionment and redistricting upon their release and after rigorous quality review protocols, and that assessment remains valid.”20

It is noteworthy that around the same time as the 2020 Census came the Supreme Court’s opinion in Trump v. New York (2020).21 While the opinion was narrow in its scope, it temporarily scuttled efforts to ensure that illegal immigrants were not counted for purposes of reapportionment and redistricting. The outcome of this opinion did not mitigate the validity of any future effort to keep noncitizens out of the census population counts, but rather focused on the Supreme Court’s opinion that the first Trump administration did not properly adhere to the Administrative Procedures Act (APA) as part of its directives to the Department of Commerce regarding the 2020 Census.22

This means that the 2020 Census definitively counted illegal immigrant populations as part of its congressional reapportionment duties, further skewing the redistricting process and stripping political power away from citizens in favor of noncitizens.

In that light, the PES likely only tells a portion of the story. In 2010, the bureau reported an error rate of just 0.01 percent.23 This was equivalent to about 36,000 people nationwide. For perspective, the bureau undercounted more than 560,000 people in Texas alone for the 2020 Census. While some observers and the PES chalked the mistakes up to the unique challenges of the COVID-19 pandemic, the reality is that the bureau also implemented a new methodology that intentionally distorted the data and removed the longstanding backstop of the CQR process. Understanding this methodology is critical for policymakers to grapple with an uncomfortable reality: the Census Bureau is defrauding American voters to protect noncitizens and tilting the balance of power toward progressives.

Analysis: Differential Privacy and Why It Matters

Among the many concerns with the 2020 census is the bureau’s adoption of a new methodology for conducting its constitutional duty to count the population accurately: differential privacy.24 This term refers to the bureau’s process of randomly altering correct data into false data within defined parameters by applying an algorithm to the dataset. The bureau implemented differential privacy for the express purpose of fulfilling its statutory disclosure avoidance requirements under 13 U.S.C. § 9. Unlike past methodologies, however, differential privacy is uniquely problematic and opaque. It essentially justifies widespread data manipulation and falsification under the pretense of protecting individuals from having their characteristic data and information exposed. While this may sound like a prospectively reasonable and perhaps even good approach in theory, the problem is that it intentionally imposes widespread instability in the numerical accuracy of block-level data that both damages the reapportionment process and masks the citizenship status of those being counted. Regardless of the stated intent behind this methodology, its effects are erroneous reapportionment within the states, skewed federal programs that rely on population-relevant formulas for the distribution of tax dollars, and an inability to assess the citizenship status of a counted individual.

Here is how differential privacy works. Say a census unit in Wisconsin contains a known population of 500 people, with 240 men and 260 women. Three hundred are at least 18 years old, with racial data showing that there are 400 white individuals and 100 black individuals. The bureau implements a differential privacy process that transforms this unit and randomly changes the numbers. Now the census unit contains 484 people, with 225 men and 259 women. Two hundred and fifty-six are listed as at least 18 years old, with racial data showing 376 white individuals and 108 black individuals.

It’s important to note that differential privacy could also go in the opposite direction, adding a total of 18 people to this hypothetical Wisconsin census unit (i.e., town) so that there are 518 total people, with 245 men and 273 women. The methodology could reduce the total number of voters in the unit to 294 people who are at least 18 years old, with 422 white individuals and 96 black individuals. 

When this intentionally opaque methodology is applied within every census block, which in this example would include every census block in Wisconsin, those inaccuracies scale upwards and inevitably become a statewide issue. This effect is not mere speculation but is confirmed by outside evaluations.

An analysis by Harvard University in 2021 attempted to simulate the bureau’s use of differential privacy within the Disclosure Avoidance System to determine potential impacts. The results of Harvard’s simulations, which used a publicly released Demonstration Data sample that provided the contours of the differential privacy algorithm without knowledge of the actual inputs, found that differential privacy causes two significant problems.25 First, it “has a tendency to transfer population across geographies in ways that artificially reduce racial and partisan heterogeneity.”26 And second, it “makes it impossible to accurately comply with the One Person, One Vote principle as currently interpreted and implemented.”27 

Differential privacy accomplishes its goal of making individual data impossible to determine by intentionally manipulating all of the relevant block-level data, injecting noise into individual census blocks, and shifting characteristic data to different blocks altogether. This algorithm makes it much more difficult for local governments, data analysts, and other entities associated with checking administrative records to identify people and source mistakes, including the total number of people in a particular census block. Differential privacy was implemented nationwide with the 2020 Census, distorting block-level data on demographics, languages, and immigration status. The Harvard simulation found that symmetry misalignment among census blocks and VTDs, combined with the scrambled data at the block level, “could still affect redistricting analyses.”28  

According to the bureau, differential privacy is intended to enhance disclosure avoidance amid technological advancements that may make it easier to guess the identity of an individual through characteristic data.29 In defense of its new differential privacy approach, the bureau cited a mathematical model out of the University of Washington that shows the relative ease to “reveal the identity of people who are transgender in previously published tables.”30 The Bureau also conducted an experiment following the 2010 Census that reconstructed the data to identify individuals of Hispanic origin in more than six million census blocks and stated that the results were “concerning” and, if republished, would have violated confidentiality protections.31

The bureau ostensibly confined itself to the broader contours of the Supreme Court opinion in Department of Commerce v. House of Representatives (1999) to avoid the use of statistical sampling when conducting the census,32 a proposed methodology that was struck down on statutory grounds. But the implementation of differential privacy adds some murkiness to that question. The bureau is legally required to ensure that statewide population counts remain invariant within the disclosure avoidance process. That is to say that the bureau has to ensure that the total population of states is accurate to the best of its knowledge, regardless of disclosure avoidance techniques. But that view does not appear to extend to the detailed data provided to the states for redistricting purposes.33 Numerical accuracy at the VTD level should be of paramount importance, but differential privacy inherently abandons that approach. This creates significant mathematical problems and is devastating when it comes to redistricting. Because the bureau is legally required to ensure that total state population numbers are invariant, differential privacy confines its distortionary impacts to population data within the state, and block-level changes regarding geographic accuracy are drastically altered. These altered datasets are detrimental to redistricting efforts and severely damage the legal requirements for the reapportionment process, of which redistricting is a part.

Critically, the accuracy of voting district data, comprising the individual precincts and wards for voters, requires reliable and accurate block-level data of the census geography. Because VTDs overlap with blocks, block groups, and census tracts, accurate block-level data conveys how many eligible voters are within a VTD. Differential privacy intentionally scrambles the block-level data, affecting its accuracy. This means that even if the citizenship question were to be included in the next census, the differential privacy algorithm would allow the bureau to continue to mask the identity of noncitizens. In many respects, it serves as the “ace up the sleeve” to continue hiding such data. Unless this methodology is systematically eliminated, efforts to effectively reform the bureau in time for the next census will not succeed.

John Abowd, an economics professor at Cornell University, is one of the primary architects of differential privacy’s use in the 2020 Census.34 He served as the associate director for research and methodology as well as chief scientist at the bureau from 2016 to 2022, overseeing the implementation of differential privacy and engineering the most inaccurate census in modern history. Abowd began work at the bureau in 1998 and was required to submit an affidavit in Alabama v. Department of Commerce defending the bureau’s practice of including all residents, including noncitizens, in the decennial census.35 

In that affidavit, Abowd unequivocally defended differential privacy’s application for disclosure avoidance: “If the block population is reported with some random fluctuation around the confidential value, then only by chance will the block identifier be correct in any potential reconstructed microdata. Compound this effect over 8,000,000 blocks and the number of feasible reconstructions explodes exponentially. This is what provides the protection against re-identification from the reconstructed data.”36 

Aside from the implication that differential privacy is designed as a buffer against revealing citizenship status, Abowd’s logic also suggests that if this compound effect nullifies the ability to accurately reengineer the characteristic microdata, it also nullifies the ability for local governments and other administrative data entities to check for errors. Some studies have touted the benefits of differential privacy and vouched for its alleged statistical accuracy,37 but it is important to note that the algorithm is not open-source, as some proponents claim.

The bureau implemented differential privacy with the explicitly stated purpose of muddying the ability of outside entities to reengineer data to identify specific individuals. The argument that the methodology is open-source is predicated solely on the fact that the formula for how differential privacy works is readily available through the use of the bureau-selected Demonstration Data, which does not provide accurate inputs for the algorithmic framework. The values that go into the formula remain masked by the bureau. And it is those precise inputs that dictate outcomes. 

The fact that the CQR process, wherein state and local jurisdictions report inaccuracies to the bureau, could not accurately be used for the 2020 Census serves as additional evidence that the values within differential privacy remain inaccessible outside the confines of the bureau itself.38 Under differential privacy, CQR is rendered moot. This is because localities cannot determine whether errors are actual count or data errors or whether they are due to the intentional falsification process of characteristic microdata.

The dubious methods the bureau used in the 2020 Census are not the first time it has attempted to operate outside its stated principle of “counting every person in the 2020 Census once, only once, and in the right place.”39 Before the 2000 Census, the bureau attempted to statistically adjust the decennial census to address undercount issues within minority populations, but was stopped by the United States Supreme Court in Department of Commerce v. House of Representatives when it found the bureau violated numerically accurate statutory obligations within the census process.

Differential privacy’s intentionally distorted data collection methodology presents significant problems for policymakers and everyday citizens. The three primary problems are as follows:

1. Structural Inaccuracy: Because differential privacy’s algorithmic processes intentionally distort characteristic data down to the block level, it is extremely difficult to determine the extent of damage inflicted on both geographic and numeric accuracy. Differential privacy skews reapportionment processes and scrambles accuracy within the VTDs, potentially contributing to the very population overcounts and undercounts that marred the 2020 Census.

2. Intentional Opacity: Count Question Resolution had previously allowed local governments to analyze census data for errors, inform the bureau of those errors, and obtain corrections. Because of differential privacy, local governments lack clarity on whether the errors are actual errors or are caused by the differential privacy algorithm. Further, the bureau can’t provide detailed answers to any inquiries about errors because to do so could reveal how the algorithm works, which could potentially lead to reverse engineering of the algorithm and an actual revealing of individual answers. This has rendered the 2020 Census the least transparent in American history.

3. Citizenship Nullification: Even if the citizenship question is added to the census, it will be impossible to ascertain the status of individuals so long as differential privacy remains in effect. The algorithm will be able to mask characteristic data, including citizenship status, preventing the ability to determine total voter eligibility at the block level due to numeric and geographic falsification.

Policy Approach: Making the Census Great Again

Time is of the essence for this administration to correct the errors and eliminate the manipulations within the Census Bureau’s operations. Below, policy recommendations for addressing these problems are broken down into two sections: correcting the overcount/undercount issue and eliminating the use of differential privacy. 

Addressing the 2020 Census

There are ongoing discussions in both the executive and legislative branches about fixing the errors within the 2020 Census. This is a good and necessary effort to ensure that Americans are aware of the extent to which apportionment and political power were improperly distributed. The 2020 Census must be corrected and republished. Below are recommendations for the best way to move forward.

  • Appoint the Right Personnel: Appoint trusted personnel to run the Census Bureau as soon as possible. This should include quickly nominating and confirming a director of the Census Bureau. Even before that individual is confirmed, the president should appoint subordinate staff immediately (including a general counsel), one of whom should be the interim director to begin the hard work of correcting the 2020 Census and laying the groundwork for reforming the bureau before the 2030 Census begins.
    • The interim director should immediately order the bureau to compare the enumerated population numbers for each level of census geography from the enumerated dataset with the population numbers for each level of census geography from the published dataset that had differential privacy applied to it. 
    • This effort will help determine the best path forward regarding both future litigation and necessary statutory changes. Importantly, it will provide insight into the best way to republish the information that illuminates the problems within the 2020 Census. 

  • Restart the Administrative Records Project: The first Trump administration started a process for using administrative records to ascertain citizenship data among 2020 Census respondents. This project should be resuscitated.
    • Immediately reinstate Executive Order (EO) 13880 from the first Trump administration. This EO mandated the creation of a block-level citizenship database.
    • Expand EO 13880 to include a provision to require state and local agencies and even private entities receiving federal funds to provide administrative records to the Census Bureau that are of assistance in identifying the citizenship status of residents of the United States.
    • Authorize the bureau to request such records, and if such records are not provided to the bureau, the bureau should inform any appropriate federal agency that is providing funds to the entity from which the bureau is requesting records of that entity’s failure to cooperate. Federal funds should be embargoed if these entities do not comply.
    • Congress should simultaneously codify the language from EO 13380, which mandated the creation of a block-level citizenship database. This should optimally be incorporated into an appropriation of funds to the Census Bureau. Further, appropriations should be conditioned on the compliance of state and local agencies to provide microdata on citizenship, residence, and other characteristic data.

  • Investigate the Group Quarters Errors: The Post-Census Group Quarters Review suggests that there were significant mistakes during the 2020 Census regarding dormitories, nursing homes, and other group facilities.
    • Personnel should be appointed to the director’s office to begin an investigation into the legality of the bureau’s statistically adjusting, imputing, or resolving household status to report in the 2020 Census empty college dormitories as fully occupied. 
    • An office of general counsel in the director’s office should be created, staffed, and tasked with this investigation. The general counsel should be allowed to detail additional staff to his office to complete this investigation. Importantly, all personnel involved in the investigation must be Title 13 sworn so that no document or data can be withheld from this internal investigation. 
    • The general counsel should be tasked with producing a report to the director, the Under Secretary of Commerce for economic affairs, the general counsel of the Commerce Department, and Secretary Lutnick as to how and why these dormitories were reported as fully occupied. 
    • If the discrepancies in the group quarters data were accomplished through the use of illegal means, such as statistical adjustment, then a republication of the 2020 census data will be required. Specifically, republication should redistribute the student population to their permanent addresses where they resided on the census day, April 1, 2020.

  • Implement Statutory Requirements for Permanent Addresses: In order to avoid future errors regarding group quarters datasets, Congress should implement statutory changes with new requirements.
    • Congress should pass a new law that requires counting all members of the military, dormitory, and prison populations at their permanent addresses. At a minimum, this should be a policy rider in the relevant appropriations bill.
    • The permanent addresses are included in the GQRs; therefore, this provision can and should be applied to the 2020 Census results as well as all censuses going forward. 
    • An APA notice and comment period should also be considered to change the current census guidance and instructions on institutional population. An EO may also be issued that would require the dataset to be constructed immediately, but the dataset could not be released as a republication, errata, or amendment of the official PL 94 redistricting data. Regardless, the release of such data, even if merely for public consumption, would allow the dataset to be constructed quickly. It could be made official once there is a provision approved in an appropriations bill or a change to the guidance and instructions through the APA.

  • Release the Non-Differential Privacy Data: The 2020 Census should be republished, taking into account what the results would have been if differential privacy had not been applied.
    • This should be done by placing language in an appropriation for the Census Bureau banning differential privacy and stating that differential privacy violates prior law. Specifically, differential privacy violates 111 STAT. 2482 PUBLIC LAW 105–119—NOV. 26, 1997 § 209 and 13 U.S.C. § 195 as interpreted in Department of Commerce v. House of Representatives, 525 U.S. 316 (1999).

Permanently Removing Differential Privacy

The damage that this opaque algorithm has inflicted on accurate numerical data and characteristic information, including the ability to acquire citizenship information, is difficult to fully grasp. If the bureau is to be truly reformed and Americans are to have a census that accurately reflects data necessary for apportionment and redistricting, differential privacy must be eliminated. If it is not, the inclusion of a citizenship question will not be enough to accomplish the objectives and purpose of the census. Below are recommendations for the best way to achieve this goal.

  • Create the “Undetermined” Category: A new category should be created to pierce through the fog of the bureau’s differential privacy scheme.
    • Specifically, an EO should be issued by using a rule proposal under the APA and placing language in an early appropriation for the Census Bureau. This new category should be called “undetermined.” 
    • The “undetermined” category should require that all characteristic categories for which the bureau cannot obtain direct or circumstantial evidence shall be reported as “undetermined” for that respondent. “Undetermined” should also be used where the bureau, to comply with Title 13 § 9, needs to obscure characteristic data in small or homogeneous blocks. 
    • If the bureau needs to obscure characteristic data to avoid revealing the actual answers on a census return or other privileged document, then one household within the census block may have all of its characteristic data reported as “undetermined.” The actual characteristic data for this obscured housing unit should be included in any aggregated higher level of geography unless to do so would, to a mathematical certainty, reveal the actual answers on a census return or other privileged document.
    • The characteristic data for this obscured housing unit should be included for any voting district or higher level of political geography. Further, language should be included in any rule or EO that states that the confidentiality of personal data is never compromised by publishing the accurate number of people residing at any level of census geography, including but not limited to census blocks. 
    • Any requirement for confidentiality of individual characteristic data at small levels of geography with a low population can be accomplished by changing that characteristic data for one housing unit to the designation “undetermined.” Numerical accuracy at any level of geography is never to be compromised in any census taken for apportionment, including redistricting purposes. 

  • Eliminate Differential Privacy: Issue an EO stating that differential privacy is both illegal and unconstitutional, and direct the Census Bureau to immediately begin eliminating the false data injected by the differential privacy process and publish as soon as possible the census counts as enumerated in the 2020 Census.
    • The privacy of individual responses should be protected through the use of the new “undetermined” response to the characteristic questions on the census questionnaire.

  • Restore Count Question Resolution: This should occur automatically with the removal of differential privacy. The rules and regulations for CQR still exist, but were not accurately applied or transparently made available to the public in the 2020 Census because it is impossible to conduct the program using a differential privacy census dataset intended to distort information. The restoration of this program would be a tremendous step forward in restoring transparency to the decennial census.

  • Seek a Declaratory Judgment: Simultaneously with the signing of the appropriate EOs and the publishing of the notices in the congressional record, the United States should immediately seek a declaratory judgment before a three-judge federal district court.
    • As provided in 111 STAT. 2482 PUBLIC LAW 105–119—NOV. 26, 1997 § 209 (f), the goal of such a judgment would be to validate that the application of differential privacy to the 2020 Decennial Census was illegal and unconstitutional. 

  • Issue Census-Related APA Guidance: To avoid a similar outcome to Trump v. New York concerning the citizenship question, and to further insulate future census efforts from data falsification efforts like differential privacy, the administration should begin the process of issuing census-related guidance to both correct the 2020 Census and pave the way for an accurate 2030 Census.
    • Publish a notice in the congressional record that the bureau is reinstituting the administrative records citizenship data project, and begin an APA notice and comment period. 
    • Publish a notice in the congressional record that the bureau is eliminating the false data injected by the differential privacy process and will publish as soon as possible the census counts as enumerated in the 2020 Census.
    • Publish a notice in the congressional record about the new “undetermined” published response to the census characteristic questions. 
    • Publish a notice in the congressional record that the bureau will include a citizenship question in the 2030 Census based on the citizenship question currently in the American Community Survey (ACS). Begin an APA notice and comment period.

Conclusion

The elimination of differential privacy is of paramount importance for restoring political power to citizens and ensuring that the will of voters is upheld. It is the main cause of the error-riddled 2020 Census and operates as the “ace up the sleeve” for those who want to mask the citizenship status of individuals who are counted. The intentionally distortionary algorithm implemented by the bureau defrauds Americans of their rights and diminishes the efficacy of our republican form of government. As such, the Bureau must be overhauled and its approach to the 2030 Census must be fundamentally altered to ensure that apportionment is accurate and the will of the voters is respected.

Adding the citizenship question to the census is an important and necessary action. But unless the 2020 Census is corrected and differential privacy is eliminated from the bureau’s methodology, that addition alone will be insufficient to accomplish the broader objective of restoring trust in the apportionment and redistricting process. The executive branch and Congress should waste no time in taking the steps necessary to return the bureau to its constitutional mission.

We consulted Dale Oldham on some of the technical mechanics of the census process.

Endnotes

1.  The Constitution of the United States (1787), Article I, Section 2, Clause 3. The National Archives. https://www.archives.gov/founding-docs/constitution-transcript 

2.  Ibid. 

3.  The Constitution of the United States (1787). The Fourteenth Amendment (ratified 1868). The National Archives. https://www.archives.gov/founding-docs/amendments-11-27#toc-amendment-xiv

4.  Title 13, U.S. Code. Legal Information Institute. https://www.law.cornell.edu/uscode/text/13 

5.  United States Census Bureau (Accessed July 29, 2025). “TIGER Data Products Guide,” United States Census Bureau. https://www.census.gov/programs-surveys/geography/guidance/tiger-data-products-guide.html 

6.  United States Census Bureau (Accessed July 29, 2025). “2020 Census Count Question Resolution Operation FAQs,” United States Census Bureau. https://www.census.gov/programs-surveys/decennial-census/decade/2020/planning-management/evaluate/cqr-faqs.html 

7.  United States Census Bureau (Accessed July 29, 2025). “Census Glossary,” United States Census Bureau. https://www.census.gov/glossary/ 

8.  United States Census Bureau (Accessed July 29, 2025). “Decennial Census P.L. 94-171 Redistricting Data,” United States Census Bureau. https://www.census.gov/programs-surveys/decennial-census/about/rdo/summary-files.html

9.  13 U.S. Code § 141. Legal Information Institute. https://www.law.cornell.edu/uscode/text/13/141

10.  America Counts Staff (May 19, 2022). “2020 Census Undercounts in Six States, Overcounts in Eight States,” United States Census Bureau. https://www.census.gov/library/stories/2022/05/2020-census-undercount-overcount-rates-by-state.html 

11.  Ibid. 

12.  Ibid. 

13.  Von Spakovsky, H. (August 2, 2022). “Census Bureau Errors Distort Congressional Representation for the States,” Heritage Foundation. https://www.heritage.org/election-integrity/report/census-bureau-errors-distort-congressional-representation-the-states 

14.  Press Release (March 10, 2022). “Census Bureau Releases Estimates of Undercount and Overcount in the 2020 Census,” United States Census Bureau. https://www.census.gov/newsroom/press-releases/2022/2020-census-estimates-of-undercount-and-overcount.html 

15.  Ibid. 

16.  Department of Commerce v. House of Representatives (1999), 525 U.S. 316  

17.  Jacobsen, L. (February 10, 2023). “How Accurate Was the 2020 Census–And Why Should You Care?,” Population Reference Bureau. https://www.prb.org/resources/how-accurate-was-the-2020-census-and-why-should-you-care/ 

18.  Ibid

19.  House Oversight Letter, Rep. James Comer (September 18, 2024). “Letter to the United States Census Bureau from the Committee on Oversight and Accountability,” United States House of Representatives. https://oversight.house.gov/wp-content/uploads/2024/09/Letter-to-Census-091824.pdf

20.  America Counts Staff (May 19, 2022). “2020 Census Undercounts in Six States, Overcounts in Eight States,” United States Census Bureau. https://www.census.gov/library/stories/2022/05/2020-census-undercount-overcount-rates-by-state.html  

21.  Trump v. New York (2020), 592 U.S. 125 

22.  Ibid

23.  Groves, R. (May 30, 2012). “How Good Was the 2010 Census? A View from the Post-Enumeration Survey,” United States Census Bureau. https://www.census.gov/newsroom/blogs/director/2012/05/how-good-was-the-2010-census-a-view-from-the-post-enumeration-survey.html 

24.  2020 Census Brief (March 2023). “Why the Census Bureau Chose Differential Privacy,” United States Census Bureau. https://www2.census.gov/library/publications/decennial/2020/census-briefs/c2020br-03.pdf 

25.  Kenny, C., Kuriwaki, S., McCartan, C., Rosenman, E., Simko, T., and Imai, K. (August 23, 2021). “The Impact of the U.S. Census Disclosure Avoidance System on Redistricting and Voting Rights Analysis,” Harvard University. https://arxiv.org/pdf/2105.14197 

26.  Ibid., p. 2

27.  Ibid., p. 2

28.  Ibid., p. 23 

29.  2020 Census Brief (March 2023). “Why the Census Bureau Chose Differential Privacy,” United States Census Bureau. https://www2.census.gov/library/publications/decennial/2020/census-briefs/c2020br-03.pdf  

30.  Ibid. 

31.  Ibid. 

32.  Department of Commerce v. House of Representatives (1999), 525 U.S. 316 

33.  See also Oldham, D., “The Census, the Court, and Redistricting,” in Census 2000: Considerations and Strategies for State and Local Government (Benjamin E. Griffith, ed., American Bar Association 158, 160–67 [2000]). The cited section conclusively shows that this limited reading of the holding in Department of Commerce v. House of Representatives is completely rejected by the plain text of the decision. 

34.  Abowd, J. and Hawes, M. (December 27, 2022). “Confidentiality Protection in the 2020 Census of U.S. Population and Housing,” United States Census Bureau. https://arxiv.org/pdf/2206.03524 

35.  Alabama v. Department of Commerce; Affidavit: John Abowd. https://perma.cc/E6TJ-SSTU 

36.  Ibid., p. 33 

37.  Cohen, A., Duchin, M., Matthews, J.N., and Suwal, B. (May 31, 2021). “Census TopDown: The Impacts of Differential Privacy on Redistricting,” Leibniz International Proceedings in Informatics, Volume 192, pp. 5:1–5:22. https://mggg.org/uploads/DP.pdf 

38.  Census Memorandum (October 19, 2021). “Changes to the Count Question Resolution Operation,” United States Census Bureau. https://www2.census.gov/programs-surveys/decennial/2020/program-management/planning-docs/CQR-detailed-operational-plan.pdf  

39.  2020 Census. “Residence Criteria and Residence Situations for the 2020 Census of the United States,” p. G-1, United States Census Bureau. https://www.census.gov/content/dam/Census/programs-surveys/decennial/2020-census/2020-Census-Residence-Criteria.pdf