The 2020 Census: Counting Under Adversity

By Constance F. Citro

With official Census Day, April 1, 2020, now behind us, the vast majority of households in the United States should have received a letter from the Census Bureau. The letter invites response via the internet using a code, but people can also respond by telephone and online without the code. As of April 2, 43% of households had responded. The Census Bureau hopes that at least 60% will respond before it has to send field workers out to follow up. A small percentage of households—in remote Alaska and northern Maine, and on American Indian reservations that requested in-person enumeration without receipt of a letter—have already been or will be visited by census field workers to collect their information in person.

Not surprisingly, however, the COVID-19 pandemic has disrupted census-taking. The Census Bureau announced on March 18 that it was protecting its temporary work force by delaying all field operations until April 1, and then again on March 28 that it would delay all field operations until April 15. Operations to visit remote areas and deliver letters for self-response to households in areas with spotty mail delivery began March 16, but were halted. The operation to follow up with visits to the estimated 40% of households that do not self-respond is now scheduled to begin in late May rather than early May. Operations to enumerate college students and others in group quarters and the homeless are delayed. And the spread of the coronavirus will likely necessitate further delays in field operations. There is a real possibility that the Census Bureau will not be able to meet its legal deadlines to deliver state counts for reapportionment of the US House of Representatives by December 31, 2020, and to provide block-level data on the voting-age population to each state for redrawing legislative seats by March 31, 2021.

Census-taking has never been easy

The 2000 census featured a partisan debate about methodology. The Democratic administration wanted to use statistical techniques as an integral part of the census, and the Republican-controlled Congress insisted on a full enumeration. The census design was not fully determined until the US Supreme Court in January 1999 ruled that a traditional enumeration was required by law for reapportionment. The court did not rule out use of statistical methods to improve census accuracy for other purposes, including legislative redistricting and allocating billions of dollars of federal funds to states. The Census Bureau scrambled to finalize computer systems and get operations specified and in the field on a timely basis for the enumeration. In the end, the 2000 census was carried out on schedule and achieved more complete coverage of the population than previous censuses.

Other censuses have also encountered difficulties in completing an accurate enumeration. The 1920 census required an additional five months to complete due to the Spanish flu, and many parts of the country questioned the results, which showed the lowest population growth rate in US history—no doubt caused by the combined effects of the flu and World War I. The 1980, 1990, and 2000 censuses faced numerous lawsuits demanding that their counts be statistically adjusted to correct differences in net undercount—the difference between omissions of people who should be counted and erroneous enumerations of people who should not be counted. (The latter would include people who died or moved out of the country before Census Day, and duplicates, such as “snowbirds” counted at their summer and winter residences.) The reason to consider making such adjustments is that certain groups, notably minorities, young children, and renters, tend to have higher net undercount rates than other groups, such as whites, older people, and homeowners. These differential net undercounts can reduce the accuracy of census data for reapportionment, redistricting, and other critical uses. As it turned out, the inability of the Census Bureau to produce sufficiently accurate net undercount estimates on a timely basis led to the abandonment of statistical adjustment for redistricting in 2000. Statistical methods were limited in 2010 to evaluating census completeness.

Is the 2020 Census in jeopardy?

I fear the 2020 census will encounter greater challenges in completing an accurate enumeration than any census in US history. The effects of the coronavirus—delaying operations, making it harder to enumerate special groups such as college students and the homeless, and hampering the ability of census workers (once they return to the field) to obtain responses from households that are still wary of being infected—will undoubtedly have negative impacts on census quality. But the difficulties facing the 2020 count have been long brewing. Even though the Census Bureau rigorously protects data confidentiality under Title 13 of the US Code, immigrant communities are nervous about whether their information might be used against them. Major computer systems used by the Bureau have not been adequately tested. Amid public distrust of government, “fake news” purveyors may seek to undermine the census, and overt cyberattacks may also suppress the count. The potential magnitude of these effects is highly uncertain.

At the core of several of these concerns is a question of trust. The effort of the administration to include a question on citizenship on the census questionnaire, in combination with the current low regard in which the public holds the federal government, led advocacy groups and others to fear that many immigrants, citizens or not, would not respond to the census. And that fear remains, even though there is no citizenship item on the 2020 questionnaire due to rulings by three US District Courts and the US Supreme Court. What the Census Bureau has been directed to do instead is produce block-level estimates of the citizen voting-age population from Social Security and other records.

The Bureau had already planned to use Social Security and other administrative records to improve estimation for households with no response after repeated follow-ups. Typically, such households amount to no more than one or two percent of the population. Whereas the Bureau would ordinarily follow up with all nonresponding households before any estimation, the coronavirus may necessitate its having to estimate the number of people in many more households than in past censuses. Moreover, households needing estimation will likely be concentrated among minorities, younger people, and renters, as these groups are historically least likely to self-respond. And the quality of responses—in terms of the correct number or characteristics of household members—obtained from follow-up or estimation is never as good as responses from households themselves near Census Day. (For example, field workers may have to ask neighbors to answer for nonresponding households, and neighbors may not be able to do so accurately. Also, the administrative records that would be used for estimation tend to disproportionately miss Hispanics.) Because of lower self-response rates and the deficiencies of follow-up and estimation, there may well be an increase in the differential undercount between minorities, younger people, renters, and others. In addition, Hispanics may be especially undercounted because Hispanic households that do respond may still omit family members who were not born in the United States, regardless of their legal status.

Steps toward quality

One method of estimating the completeness of the count for the population, which is not affected by the coronavirus, is termed “demographic analysis” (DA). DA updates the previous census (already corrected for net undercount) with administrative records (such as births and deaths) to estimate the population in the current census year. But DA is reliable only at the national level and for a limited number of demographic groups. Another method for evaluating the count is a Post-Enumeration Survey (PES), in which households in a large nationwide sample of blocks are separately surveyed after most of the census enumeration has been completed. The results are then matched to the census enumerations, and net undercount estimates are derived. PES-based estimates can be derived for states and smaller geographic areas than possible with DA, and for more demographic groups as well. But the PES survey may itself be compromised due to the coronavirus—not only because its start will inevitably be delayed but also because of reluctance of households to be interviewed.

For the constitutionally required complete enumeration, the Census Bureau may need to request an extension of the statutory deadlines for providing data for reapportionment and redistricting, so that it can try to complete the count during a period when the coronavirus is temporarily in abeyance. Although such a step may be necessary, I have already noted that data quality deteriorates the farther the enumeration is from Census Day—for example, people may forget to include a family member who moved away since Census Day or count a college student at home who should have been counted at their college.

For the PES, the Census Bureau could, and should, plan to conduct the survey by internet, mail, and telephone to the greatest extent possible, instead of sending interviewers into the field. The Bureau will also need time to produce net undercount estimates should the courts require statistical adjustment of the counts for redistricting. I fully expect lawsuits, as in 1980, 1990, and 2000, to require statistical adjustment using the PES estimates even if this means delaying the availability of the census results. Given the level of partisanship in the nation today, the census is likely to be litigated in any case, but the threat posed by the pandemic to census quality is all too real. The census is a critical tool of our democracy and for helping us understand who we are. A successful census will depend on the efforts of dedicated, principled civil servants who are working hard, under conditions of adversity, with state and local governments and civic groups to achieve as accurate an enumeration as possible in 2020. You can support their efforts by promptly filling out your census questionnaire and urging everyone you know to follow suit.