Tennessee was undercounted by -4.78 percent according to a new census report.
Additional information about the quality of the 2020 Census in Tennessee was released on Thursday, May 19, and it showed the state’s population was undercounted by an estimated -4.78 percent.
That error rate equates to an additional 322,000 people living in the state’s households that may not have been counted in the decennial census. The new numbers are a dramatic shift from the 7,100 person overcount that was announced following the 2010 Census.
Only Arkansas’s -5.04 percent error rate was lower this decade.
The U.S. Census Bureau’s new reporting comes from the 2020 Post Enumeration Survey (PES) and is based on a nationwide a sample of 161,000 housing units in 10,000 census tabulation blocks across the country. Unlike 2010, the new state-level quality measures were not accompanied by undercount and overcount data for cities and counties with more than 100,000 residents. That change makes it more difficult to geographically isolate the sources of the error.
Tennessee was one of six states in the country with a statistically significant undercount in 2020, with five of those states—Arkansas, Florida, Mississippi and Texas in addition to Tennessee—falling in the south region of the US.
Eight states had statistically significant overcounts this decade. Hawaii’s 6.79 percent net coverage error ranked highest. The balance of the 36 states in the survey couldn’t definitely be classified as undercounts or overcounts because the confidence interval for the survey included values both above and below zero.
Tennessee’s -4.78 percent undercount rate compares to a -1.85 percent undercount in the Southern U.S. and a -0.24 percent net undercount for the U.S. as whole. Nationally, the 2020 Census was found to have had a net undercount totaling -782,000 people.
The magnitude of the coverage error rates are substantially larger than some pre-census forecasts, but there are no provisions to revisit the decade’s congressional apportionment counts. Those figures, which are used allocate seats to the U.S. House of Representatives on the basis of population, are fixed. However, the Census Bureau has convened an internal working group to assess the feasibility of incorporating the PES findings into its annual population estimates.
Omission Rates Climbed in 2020, But There is Little Data to Understand Why
For Tennessee, the PES showed that 94.2 percent of people were correctly enumerated in the county where they resided on Census Day—April 1, 2020. Another 0.2 percent of residents had correct census information reported but were counted in a different county within the state. These numbers were comparable to 2010 figures for the state and similar the national levels in the new 2020 study.
The balance of the state’s population (5.6 percent) had an error associated with their enumeration. These errors are measured in three separate components:
- Duplications: 1.7 percent of people were counted at more than one location
- Erroneous Enumerations: 0.9 percent of people were fictitious persons, children born after April 1, people who died after Census day or non-residents of the state.
- Imputed Characteristics or Counts: 3.0 percent of people had incomplete responses or no response, and were tallied using statistical techniques called imputation; for most of the households in the U.S., the basic count of residents were known but characteristics were missing.
The big news for Tennessee was its 10.1 percent omission rate. Omissions include people that the PES was unable to determine were accurately counted.
In many cases, people who were omitted from the 2020 Census were either erroneously enumerated or had all or some part of their response imputed. For example, consider the case of a college student counted at a parent’s home rather than at an off-campus college apartment serving as the usual place of residence on Census day. That student would have been both erroneously enumerated at their parent’s dwelling and omitted from the census because they were not counted at their off-campus residence. In this simple example, the net coverage error is zero, but localized population counts or other Census data products tabulating race, age or household characteristics could still be affected.
Most census errors are not so straight-forward and, in the case of omissions, have been shown to vary by a number of important characteristics such as age, housing tenure and race/Hispanic ethnicity.
Omissions also include people that who were out-right missed in the 2020 Census counts.
The state’s omission rate grew from 5.8 percent in 2010 to 10.1 percent this decade and this increase is the source of the state’s undercount. At present, the Census Bureau has no plans to release additional data to help states understand the source of increase other than it national-level characteristics.
Nationally, Rates of Undercounts and Overcounts Increased This Decade
The national levels of coverage error released in March 2022 showed a number of statistically significant differences in rates among racial and ethnic groups, renters and homeowners, and age groups.
Unfortunately, in many cases these errors rates widened between 2010 and 2020.
In the lead up to the 2020 Census, an often-cited statistic was the -4.6 percent undercount rate among young children in the 2010 Census. An estimated 1 million children ages 0 to 4 were missed in last decade’s count according a second census evaluation, the Demographic Analysis. In 2020, the number grew to -5.4 percent. The PES estimate for that same group grew from -0.72 percent in 2010 to -2.79 percent in 2020. The new numbers equate to a “larger undercount of young children than every other census since 1970.”
The new PES report also showed that the undercount of people who rent their dwelling declined to -1.5 percent. The trend of overcounts of homeowners and undercounts of renters has persisted for decades and a mix reasons are behind the undercounts. Among renters, the more transitory nature of the group, lower incomes and lower homeownership rates among people of color are among a number of reasons cited.
There were also statistically significant differences of coverage errors among several racial and ethnic groups. People who reported their race as non-Hispanic White alone were overcounted by 1.64 percent, up from 0.83 percent last decade. Several of the fastest-growing racial and ethnic groups last decade also had some of the highest rates of undercount. This includes the Hispanic or Latinos group which saw their error rate fall from -1.02 to -5.0 percent.
|Race or Hispanic Origin||2010||2020|
|Race alone or in combination|
|Non-Hispanic White alone||0.83*||1.64*|
|Black or African American||-2.06*||-3.30*|
|American Indian or Alaska Native||-0.15*||-0.91*|
|American Indian Areas Off Reservation||3.86||3.06|
|Balance of the United States||0.05||-0.86*|
|Native Hawaiian or Other Pacific Islander||-1.02*||1.28*|
|Some Other Race||-1.63*||-4.34*|
|Hispanic or Latino||-1.54*||-4.99*|
*Net coverage error is statistically significantly different from zero
Source: 2020 Post Enumeration Survey, US Census Bureau, 2022
For more information about the Post Enumeration Survey, visit the Census Bureau’s 2020 PES website. The site includes recorded webinars and additional detail and data from previous decades.