The U.S. Census Bureau today released more results from its 2020 Post-Enumeration Survey (PES), showing population undercounts and overcounts by state and the District of Columbia.

The PES estimated that counts for 36 states and the District of Columbia did not have a statistically significant undercount or overcount in the 2020 Census and that there were undercounts in six states and overcounts in eight.

Previously released PES results for the nation showed no statistically significant net coverage error for the total population, although undercounts and overcounts by race, Hispanic origin, age, sex and home ownership varied.

The PES measures the coverage of the household population (excluding Remote Alaska areas). It doesn’t measure the coverage of people living in group quarters, such as college dormitories, nursing homes, prisons, etc.

A net coverage error rate is the difference between the census count and the PES estimate of the number of people in the United States expressed as a percentage of the PES estimate. Note that neither represent a “true count” because, of course, it is impossible to achieve a perfect count.

If an estimated net coverage error is negative, it means the census counts may have been too low and the census missed some people. We call this an undercount. If it is positive, it means the census counts may have been too high, indicating some people may have been counted in error or more than once. We call this an overcount.

While the 2020 Post-Enumeration Survey can estimate undercounts and overcounts in the census, PES data cannot answer why a particular state may have experienced one.

But keep in mind that the net coverage error is based on survey estimates derived from 114,000 household responses. As such, it is subject to statistical uncertainty as with all survey-based estimates.

Here, we answer a series of questions about the state and national PES results, including insight on specific census operations.

What is the PES?

The PES estimates show how well the 2020 Census counted everyone in the nation by creating an independent estimate of the number of people living in the United States on April 1, 2020 (excluding people in group quarters, such as nursing homes or college dorms, and people in Remote Alaska areas), surveying a sample of people in households in the United States and matching those responses to their records in the 2020 Census.

What is its intended use?

Together, the PES, Demographic Analysis, and 2020 Census quality indicators offer insights into how well a decennial census counted the population.

These statistical products cannot be used to change the final census count but are useful in assessing the current census, determining how best to estimate the population between now and 2030, and helping to improve future censuses.

What is the methodology?

The PES surveys a sample of people in the United States and matches their records to the 2020 Census. Out of a sample size of 161,000 housing units for the 2020 PES, the Census Bureau completed 114,000 interviews.

Today you released PES results by region and state. What did coverage look like from a regional perspective?

The Census Bureau breaks the nation into four geographic regions: Northeast, South, Midwest, and West. See the PES report for more info. We estimated an undercount in the South region (- 1.85%) and estimated an overcount in the Northeast region (+1.71%).

According to the PES, which states had undercounts?

Arkansas (-5.04),
Florida (-3.48),
Illinois (-1.97),
Mississippi (-4.11),
Tennessee (-4.78), and
Texas (-1.92).

And overcounts?

Delaware (+5.45),
Hawaii (+6.79),
Massachusetts (+2.24),
Minnesota (+3.84),
New York (+3.44),
Ohio (+1.49),
Rhode Island (+5.05), and
Utah (+2.59)

Why was there an undercount or overcount in my state?

While the 2020 Post-Enumeration Survey can estimate undercounts and overcounts in the census, PES data cannot answer why a particular state may have experienced one.

Based on the PES results, can states change their 2020 Census counts?

The quick answer is “No.”

Post-Enumeration Survey estimates are created to help data users better understand the quality of the census counts. The estimates demonstrate the Census Bureau’s commitment to transparency and are a key tool for the Census Bureau to build and apply lessons from the 2020 Census while planning for the 2030 Census.

The Census Bureau offers the 2020 Census Count Question Resolution (CQR) operation to give states (and tribal and local governments) the opportunity to request a review of their boundaries and housing counts to identify errors that may have occurred while processing their 2020 Census counts.

CQR does not alter redistricting data, apportionment results or other 2020 Census data products. If changes are made, corrected 2020 Census counts will be issued to governmental units, and these corrected counts will also be included in the population estimates base.

The Census Bureau is also proposing a new program called the 2020 Post-Census Group Quarters Review that, upon approval, will allow governmental units the opportunity to request the Census Bureau to review their population counts for group quarter facilities. The PES results are not factored into either of these reviews.

The Census Bureau has begun researching the feasibility of taking coverage measures from 2020 Demographic Analysis and the 2020 Post-Enumeration Survey to help inform the development of the annual population estimates.

Why do some states’ undercounts and overcounts have such high standard errors?

Standard errors quantify the amount of uncertainty in the estimates because they are based on a sample of people rather than a census of everyone. The amount of sample in each state depends on the number of people in the state and other more technical considerations.

The PES results are derived from 161,000 housing units in a random sample of 10,000 blocks across the United States and Washington, D.C. States with smaller sample sizes generally have larger standard errors.

The standard errors are also impacted by the number of completed interviews, so they vary among the states because they had different numbers of respondents.

How did these state-level undercounts and overcounts compare with those in the 2010 Census?

See Appendix Table 3 on page 16 in the National Census Coverage Estimates for People in the United States by Demographic Characteristics Report for a list of the PES results from 2010 and 2020.

The PES confirms that while there were overcounts and undercounts in the 2020 Census, they were not outside the range of variability we have come to expect from prior decades.

The 2010 PES estimated that no states had statistically significant undercounts or overcounts in the 2010 Census. In the 2000 Census, the PES estimated 22 states and the District of Columbia had statistically significant undercounts or overcounts.

However, it’s important to note that we have improved the methods we use to estimate 2020 state-level net coverage errors and their sampling error. These improvements were designed to reduce the bias of the state coverage estimates and more accurately measure the sampling error of the estimates.

That means any such comparison of estimates is subject to different sources of error and uncertainty. As a consequence, the statistical uncertainty associated with 2020-to-2010 state level comparisons will be greater than that indicated by the published standard errors, but how much greater is unknown.

So, while informative, the 2020-to-2010 state level comparisons cannot be considered definitive.

We discuss the methodology in more detail in the Source and Accuracy Statement.

Do you know what groups were undercounted or overcounted in each state? And where coverage error happened in a state?

No, because sample sizes within most states do not support such estimates. That is why today’s results are not broken down by demographic characteristics or geographic areas within the states. The Census Bureau did not release state estimates of coverage by demographic groups following the 2010 PES, either.

Given the sample size for the 2020 PES and the assumptions required to make substate geographic estimates, we cannot include county or place estimates in the 2020 PES reports.

The Post-Enumeration Survey: Measuring Coverage Error provides more information on why county and place estimates are not included in the PES reports.

Today, you also released PES results by census operations — or how you counted the population. What do they show?

The Census Coverage Estimates for People in the United States by State and Census Operations Report shows components of census coverage for a variety of operational outcomes, including response mode, type of enumeration area, level of self-response rate, type of respondent, and the month the Nonresponse Followup (NRFU) interview was conducted.

The results show different rates of correct enumerations, erroneous enumerations due to duplication, erroneous enumerations for other reasons, and whole-person census imputations for various operational outcomes.

What do we mean by components of coverage?

Correct enumerations refer to people counted in the census who were living in the U.S. on April 1, 2020 (the reference day for the census). According to the PES, the individuals should have been and were counted in the census.

Erroneous enumerations include duplicate records of people who were correctly counted in the census as well as people who were counted but should not have been according to the PES. For example, they may have been born after April 1, 2020, or were just visiting the country.

A whole-person census imputation is a census enumeration for which all the person characteristics are imputed. This occurs when very little information about the household or the people in the household was obtained from direct data collection efforts. (Note that whole-person census imputation methods were only used on a small fraction of people: 3.4%).

For example, a household or proxy respondent might have reported the number of people in the household but no other information about the people.

In such cases, the Census Bureau used administrative records to fill person characteristics when the census population count matched the administrative records population count associated with that address.

For households in which there were no administrative records available, or the census population count did not match the administrative records population count, the Census Bureau used imputation methods to fill person characteristics based on people in a similar nearby household.

See the National Census Coverage Estimates for People in the United States by Demographic Characteristics Report for more information on whole-person imputations.

Omissions are people who were in the population but not correctly counted in the census according to the PES. Omissions include people who were missed by the census as well as people who were included in the census without enough information to meet the PES strict definition of being a correct enumeration.

For example, if a neighbor accurately reported that three people were living in a house but didn’t provide a name or any other characteristics about the people in the house, the three people would have been included in the census as whole-person census imputations, but they would have been classified as omissions in the PES.

What else do we know about the quality of enumerations from the NRFU operation?

The PES results also provide insight into the quality of the NRFU enumerations broken down by how we completed the enumeration — through an interview with the household, a proxy respondent (such as a neighbor or landlord) or through administrative records.

The 2020 Census was the first to use administrative records to count people who had not responded. What does the PES tell us about those counts?

The correct enumeration rate for administrative record enumerations was 94.5%.

What did the PES show about undercounts and overcounts by operation?

It is not possible to say. The PES doesn’t have a way of estimating undercounts and overcounts by census operations.

We can only estimate whether the census records enumerated in that operation were counted correctly (correct enumerations), counted in error (erroneous enumerations) or whether we had to fill in all the characteristics of the individuals (whole-person imputations).

The PES is used to estimate the population size. However, it is not possible to estimate the population size for specific operational variables since we must be able to measure the characteristic the same way between the census and the PES.

For example, the PES cannot estimate the number of people in the population who should have been counted by proxy respondents or administrative records.

What do today’s results tell us about the quality of the 2020 Census?

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.

As with previous decennial censuses, the PES informs the strengths and limitations of the 2020 census and provides a foundation for the Census Bureau's work to improve the next census. It has never been intended to change a decennial Census count.

Secondarily, we believe the data are fit for a wide range of other purposes — understanding our population, planning for services, making decisions about funding, etc.

The Census Bureau’s primary responsibility is providing census data that are fit for use (meaning they are of suitable quality) for apportionment and redistricting. Prior to releasing the apportionment and redistricting results, we thoroughly analyzed a host of operational quality metrics, and compared the results to our independent Demographic Analysis as we have done in prior censuses.

In determining that the data were fit for release we did not yet have access to the Post-Enumeration Survey. This was true not only in 2020, but in all prior censuses.

Although we strive to minimize undercounts and overcounts, we experienced them to varying degrees in previous censuses. However, we defer to policy experts and decision makers on whether the data meet their specific needs. Based on our statistical expertise, the Census Bureau has produced counts that provide a vivid portrait of our nation’s people.

Additionally, the components of coverage by census operation also give us valuable insight as we plan operations for the 2030 Census.

How should we treat the counts in these 14 states that had a significant undercount or overcount? Can we legitimately use them?

The PES state and national level estimates indicate that the 2020 Census data are fit for use in apportionment, redistricting, and a wide range of other purposes.

The PES provides estimates that are subject to various errors. Many of these errors are described in the Source and Accuracy Statement and other methodological documents.

We encourage a review of the strengths and limitations of the 2020 Census and our quality assessments to better inform how consumers can best use the 2020 Census data for their specific purposes.

For instance, national, state and local 2020 Census data can provide useful portraits of the population as far as rental housing and family composition and other characteristics gathered in the census.

The Census Bureau is committed to providing information about the quality of the 2020 Census so that people know the strengths and limitations of the data.

How will the Census Bureau use the PES results?

The quality assessment of the 2020 Census provides a foundation for the Census Bureau's work to improve the next census — as opposed to changing the 2020 Census count.

Together, the PES, Demographic Analysis, and 2020 Census quality indicators offer insights into how well a decennial census counted the population.

Over the decades, each census has expanded the use of outreach, advertising and partnerships with key stakeholders and trusted community voices with the goal of increasing awareness and participation.

We’ve also worked to improve our methods, activities, and other operational components to improve the coverage of the various groups that have experienced an undercount in previous censuses. Our goal for the census is always to get a complete and accurate count.

Each decade there is room for improvement, and this one — especially with the unprecedented challenges we faced in the wake of the COVID-19 pandemic — is no exception.

There is still work to be done, and these coverage measures will help shape what that work looks like. As we plan for the 2030 Census, we’re committed to learning from the 2020 Census and continuing to collaborate with people with diverse perspectives from across the country.