The Urban Institute recently published projected 2020 Census population miscounts by state and selected demographic groups. These miscounts, and the fairness and accuracy of the census, are governed by four simple numbers:
- correct listings
- inclusion mistakes
- the uncounted
We use these terms for simplicity; the US Census Bureau refers (PDF) to these as correct enumerations, erroneous enumerations, whole-person census imputations, and omissions, respectively.
These four numbers affect the ability of a neighborhood, city, county, and state to get its fair share of congressional representation and funding resources. In this post, we use estimates from the 2010 Census Coverage Measurement (PDF) program to explain how these numbers determine the official census count.
In a decennial census, the Census Bureau seeks to count every person living in the US, count them once, and count them in the right place (PDF). Correct listings represent the collection of just those people who are correctly counted. This happens when a correct, completed census form is returned or completed online or when an enumerator secures the correct information in the field from households who fail to send in a form. The 2010 Census had about 285 million correct listings (PDF) out of an official US population count of just under 301 million.
People mistakenly appear in every official decennial census count. These inclusion mistakes typically occur in three ways:
- Someone may be counted more than once at different locations, such as college students counted in a dorm and counted back home by their parents.
- A self-responder may include someone who died before Census Day or a child born after Census Day.
- Someone may be counted only once but in the wrong location, be it neighborhood, city, county, or state (e.g., when people move just before or after Census Day and use an incorrect address).
Just over 10 million inclusion mistakes (PDF) were estimated in the 2010 Census, of which 8.5 million were undetected duplicates (the first example above).
When the Census Bureau fails to receive or retrieve completed forms from occupied households, it uses two predictive approaches to fill them in. For 2020, some households will be imputed using high-quality administrative records, such as IRS or Social Security data. And when such sources prove insufficient, they’ll use statistical predictive approaches that also take households’ neighborhood characteristics into account. Just less than 6 million people were imputed (PDF) in 2010, and that number is expected to increase in 2020 because of the use of administrative records to reduce enumerators’ workloads.
There are many reasons why some people who should be counted aren’t included in the official census count. Some forms are completed but don’t include everyone (e.g., a resident may only report their own family members in homes with more than two families rather than list everyone in the house).
People can also be overlooked by the system. People in homes without postal addresses (e.g., in colonias or tribal lands) may never receive a census form. The same is true for the homeless population living outside of shelters. And some people choose not to participate in the census. In 2010, an estimated 16 million people went uncounted (PDF).
Why do these numbers matter?
Consider how the 2020 Census counts are developed in your neighborhood, city, or county. The correct listings and inclusion mistakes are added to the predictions. The result is the official census total population count. Right or wrong, we must live with this count until 2030.
But what about all the uncounted people in your community? The “true” population count is simply the people who are correctly listed plus the uncounted—this is what the official census count should be. The true count is never known with certainty, but it is estimated by the Census Bureau via a rigorous statistical assessment (PDF) conducted just after each decennial census.
When the Census Bureau’s official count is smaller than the true count, we have a net population undercount; when it’s larger, there is a net overcount. Net population over- or undercounts are determined by how well the uncounted stack up against the sum of inclusion mistakes and predictions. In the 2010 Decennial Census, the uncounted number was nearly equal to the sum of inclusion mistakes and predictions (with a net overcount of 0.01 percent), and the census was hailed as highly accurate.
But overall accuracy masked unfairness. White, non-Hispanic people were overcounted, while other demographic subgroups (including African Americans, Hispanics, and young children) were undercounted, thus creating an unfair counting situation. Undercounted communities receive neither the representation nor the federal funding they deserve for the decade following a census.
Our research suggests the 2020 Census could see up to a 1.2 percent net undercount of the US population, which means a net undercount of more than 4 million people (largely from communities of color and young children). Moreover, far more than the 2010 estimate of 16 million people would be uncounted to lead to the 4 million net undercount we projected for 2020.
Despite data quality checks, decennial censuses always experience inclusion mistakes. The fairness and accuracy of the 2020 Census may hinge on how well the Census Bureau can predict the household compositions of the uncounted using administrative records and improved statistical prediction methods.
More importantly, there is another way to strengthen the census’s fairness and accuracy: by making sure that as many people participate as possible, thus decreasing the uncounted and the influence of predictions on the final official count.
That’s why returning your completed form is not enough. We shouldn’t skip questions. We should encourage our friends, family, neighbors, and community to participate on April 1, 2020. And we should help Census Bureau enumerators if they ask for neighbors’ information. Fairness and accuracy depend on it, and we the people deserve our fair share.