Salvo 05.14.2021 10 minutes

Fictive Counting

Bar Chart

Serious irregularities in the 2020 Census process must be addressed now.

On April 26, 2021, the U.S. Census Bureau announced the results of the 2020 Decennial Census. The Census, which is Constitutionally mandated, determines “Apportionment,” or the number of representatives each state will have until the next Census in 2030. 

The states of California, Illinois, Michigan, New York, Ohio, Pennsylvania, and West Virginia each lost a seat. Texas picked up two seats, and Colorado, Florida, Montana, North Carolina, and Oregon each gained one seat. Many expected a different result. Texas was expecting to pick up three, while Florida was supposed to gain two, and Arizona one. Alabama, Minnesota, and Rhode Island were each expected to lose a seat. Others were expecting the population shift within that same number of seats to be different as well. These expectations were based on estimates, but the final numbers were significantly off from projections.

In the end, the Census added 2.5 million persons to blue states above the December population estimate. Many state estimates were entire percentage points off from the Decennial Census, which is an unusual margin of error. Estimates can be wrong, of course, but a peek under the hood raises important questions about the probity of the 2020 Census.

I worked at the U.S. Census Bureau in the 2020 Decennial Census in Queens, New York. I then received an appointment to the U.S. Department of Commerce in April 2020, to work on Census issues, as part of the wave of technical experts and problem-solvers that John McEntee focused on hiring during his tenure at the Presidential Personnel Office. I spent long hours on the Census and spoke with the Secretary and the White House on technical details of the Census. The president tasked us with accomplishing the Decennial before the statutory deadline.

Contrary to its image in the mainstream media, the Census does not behave like a studiously nonpartisan agency. Many of those in leadership made their negative opinions of former president Trump clear. Historically, Census Apportionment numbers appear to favor liberal, blue states over conservative, red states in the final rounds of representative allocations. This likely stems from the use of “hot-deck imputation,” a controversial method where the Census will use a nearest, most similar neighbor to impute the characteristic, population, etc., of a household who does not fill out the Census or respond to enumeration attempts. A more radical method called “count imputation” creates a whole person at an unresponsive household. These are persons without available Federal or state records. In Utah v. Evans, 536 U.S. 452 (2002), the Supreme Court left imputation to the Secretary of Commerce’s discretion because, in the majority view, it does not constitute statistical sampling, prohibited in 13 U.S.C. §195. It can be argued that hot-deck imputation favors more liberal states over more conservative ones, more urban states over more rural states. Specifically, it favors states that have areas of low response rates to the Census. 

The Census justifies count imputation methodology through comparison to existing, known datasets. But this is an information asymmetry. Epistemologically speaking, you cannot compare an unknown unknown to a known unknown variable by simply randomizing a sample of known quantities. The quality of these households overwhelmingly self-responding to the Census, or responding to Census Enumerators (workers), makes these households sufficiently distinct from nonresponsive households with zero records. While imputing missing demographic characteristics makes sense, adding entire people does not. In 2000, the Census Bureau imputed 1,172,144 people and in 2010, 1,163,462.

After field data collection, the Census begins data processing. This phase showed a discrepancy of tens of thousands of persons in particular Northeastern states, primarily centered around “Group Quarters,” a special kind of residence that the occupants don’t normally rent or own, and where they usually receive a service, such as a nursing home or university. Normally, these are enumerated separately by specialists trained to handle their unique circumstances. But these facilities were difficult to access in 2020 because of the pandemic-related lockdowns. 

Crossing the Line

Given the complications, the Census decided to apply “Group Quarters Imputation,” a process that was never brought to the states to give input on, nor practiced in the 2020 Decennial Testing Phases. This process created hypothetical persons missing in the data, but it is not an imputation method by definition. This “Group Quarters Imputation” used a linear regression analysis based off estimates from the Group Quarters themselves, yielding a ratio by which Census analysts would impute the population of each facility. The use of the term “imputation” in this case is egregious. It constitutes statistical sampling, which is forbidden by law for Census enumeration. This method also heavily favors the Northeast because of the density of college campuses in that region and the fact that the lockdown was the most severe there during the Decennial Census. 

To understand the full extent of injury or discrepancy, Texas, Florida, and Arizona would need to acquire the record of state-by-state imputations from the Census Bureau, broken down by the types of imputation and whether the imputations occurred in Group Quarters or Households. To my knowledge, the Census does not normally produce such documentation and it takes years for the Census to publish studies on itself. These aforementioned states would also need records pertaining to the decision-making processes around these novel data calculation processes. Obtaining these records would enable these states to determine whether it has grounds to challenge the Census outcomes. This is in addition to any actions taken by Census staff at the behest of the Biden Administration since taking office on January 20.

A Needed Correction

Successfully challenging the Census results would affect appropriation and could affect apportionment. Given the change in the Supreme Court balance, it is possible for the court to limit imputation precisely because of its disparate impact. Currently, the Census does not make further efforts after the Decennial to enumerate imputed households, nor does it conduct any formal analysis of imputation outside of refining the method itself. When I asked if there is a post imputation survey like there is a Post-Enumeration Survey (PES), I received no answer. The Census typically takes years to officially release information on the Decennial, making it impossible for states to seek redress if action is not taken quickly.

The 2020 Census was possibly the most difficult in American history because of the lockdowns and media hysteria . Census Bureau employees performed admirably, despite adversity. However, the imputation issue is not the only one, but the most grievous one that the states and courts must address immediately. Other issues still exist, such as the proposed use of Differential Privacy, a method that could render population counts inaccurate for redrawing legislative districts, and the questions surrounding Citizen Voting Age Population districts. Democrats seek to muddy the enumerative waters by counting illegal aliens and imputing populations that don’t exist, but the Census must be protected from partisan corruption. The assault on the Census cannot stand and the states must take legal action to secure their appropriate representation.

The American Mind presents a range of perspectives. Views are writers’ own and do not necessarily represent those of The Claremont Institute.

The American Mind is a publication of the Claremont Institute, a non-profit 501(c)(3) organization, dedicated to restoring the principles of the American Founding to their rightful, preeminent authority in our national life. Interested in supporting our work? Gifts to the Claremont Institute are tax-deductible.

Suggested reading from the editors

to the newsletter