Abstract
This paper analyzes and provides estimates of the undercount of the foreign-born in the US Census Bureau’s 2020 American Community Survey (ACS). It confirms that a differential undercount occurred in the 2020 ACS. In particular, noncitizens that arrived from Central American countries after 1981 had undercount rates of 15–25 percent, but undercount of noncitizens that arrived from European countries in the same period was not detectable by the methods described in this paper. The Center for Migration Studies of New York (CMS) and others use ACS data to derive annual estimates of the US undocumented population. The Census Bureau recently reported that the total population count for the 2020 Census was consistent with the count for recent censuses, despite the Covid-19 pandemic and the Trump administration’s interference in the 2020 Census. Nonetheless, the accuracy of 2020 ACS data for the noncitizen population that arrived after 1981 remains a major concern given the fear generated by the Trump administration’s abusive rhetoric and anti-immigrant policies. The estimates set forth in this paper were derived by analyzing trends in annual ACS data for 2016–2020 compiled from the IPUMS website (Ruggles et al. 2021). Decennial census data cannot be used for this purpose because data on country of birth, citizenship, and year of immigration are not collected in the census. However, it is reasonable to believe that the 2020 census and the 2020 ACS experienced similar challenges because they were conducted under comparable conditions. The patterns of undercount of noncitizens described here for the 2020 ACS are likely mirrored in the 2020 census and will reduce federal funding and representation to affected cities and states for the next decade.
Introduction
On March 10, 2022, the US Census Bureau released estimates of undercount and overcount in the 2020 Census. Director Robert L. Santos said, “Today’s results show statistical evidence that the quality of the 2020 Census total population count is consistent with that of recent censuses. This is notable, given the unprecedented challenges of 2020” (US Census Bureau 2022a). The Covid-19 pandemic and the administration’s unprecedented interference in the count were two of the major challenges to completing a timely and accurate 2020 census.
The finding that the general level of population coverage in 2020 was consistent with recent censuses is a relief to those who feared the 2020 census would be a disaster. However, all users of 2020 census data, as well as users of the 2020 ACS, must be alert to shortcomings and inconsistencies in their areas of interest. For example, the accuracy of 2020 ACS data for the noncitizen population that arrived after 1981 will be a major concern for organizations that derive estimates of the undocumented population.
The Covid-19 pandemic and the administration’s interference in the count were not the only major challenges to the Census Bureau in 2020. The Trump administration’s abusive rhetoric and anti-immigrant policies also presented a serious challenge to completing an accurate count of the foreign-born population. An especially harmful aspect of undercount based on fear of the government is that it can have a large impact on some groups but have little or no effect on others. The analysis presented here confirms that such a differential undercount occurred in the 2020 ACS. Noncitizens that arrived from Central American countries after 1981 had undercount rates of 15–25 percent, but undercount of noncitizens that arrived from European countries in the same period was not detectable by the methods described in this paper.
These estimates were derived by analyzing trends in annual ACS data for 2016–2020 compiled from the IPUMS website (Ruggles et al. 2021). Decennial census data cannot be used for this purpose because data on country of birth, citizenship, and year of immigration are not collected in the census. However, it is reasonable to believe that the 2020 census and the 2020 ACS experienced similar challenges because they were conducted under comparable conditions. The patterns of undercount of noncitizens described here for the 2020 ACS likely are mirrored in the 2020 census.
Findings (All Data are for Noncitizens That Arrived After 1981)
About 1 million noncitizens that arrived after 1981 were omitted from the 2020 ACS (Table 1).
The 2020 ACS omitted about 400,000 noncitizens from four Central American countries – El Salvador, Guatemala, Honduras, and Nicaragua (Table 1).
About one-fourth of noncitizens from Guatemala, Cuba, and Iraq were omitted from the 2020 ACS (Table 1).
For the six countries with the largest number of omissions, the undercount rate was about 25 percent for recent arrivals, and the rate declined by half with each additional 10 years of residence (Table 2).
The ACS count for 135 countries in 2020 was about the level expected based on trends from 2016 to 2019, indicating Covid-19 was not a major contributor to undercount of noncitizens in the 2020 ACS.
The following findings are for El Salvador, Guatemala, and Honduras combined.
Undercount rates for males and females were the same (18 percent), and undercount declined by age, dropping from 22 percent for those under age 18, to 11 percent for ages 50 and over (Table 3). Eighteen states had undercount rates of 10 percent or higher in the 2020 ACS. Rates were relatively lower for states with large noncitizen populations (Table 4). More than 50 percent of noncitizens in Arizona, Alabama, Ohio, Kansas, and Arkansas were omitted in 2020 (Table 4). Twelve metro areas had undercount rates of 10 percent or higher. Rates were 10 percent in Los Angeles and Houston, 54 percent in Austin, and 61 percent in Phoenix (Table 5). The undercount in these states and metropolitan areas will cause a reduction in federal funding and representation for the next decade.
Difference Between the Actual and Expected Count in 2020 for Selected Countries of Origin.
Note: Excludes arrivals in the year of the survey. See text.
Source: Columns 1, 2 and 4 from ACS. Column 3 = (Column2-Column 1)/3 + Column 2.
Difference Between the Actual and Expected Count in 2020 for Mexico, El Salvador, Guatemala, Honduras, Cuba, and India Combined, by Period of Entry.
Note: Excludes arrivals in the year of the survey. See text.
Source: Columns 1, 2 and 4 from ACS. Column 3 = (Column2-Column 1)/3 + Column 2.
Difference Between the Actual and Expected Count in 2020 for El Salvador, Guatemala, and Honduras Combined, by Age and Sex.
Note: Excludes arrivals in the year of the survey. See text.
Source: Columns 1, 2 and 4 from ACS. Column 3 = (Column2-Column 1)/3 + Column 2.
Difference Between the Actual and Expected Count in 2020 for El Salvador, Guatemala, and Honduras Combined, by State of Residence.
Note: Excludes arrivals in the year of the survey. See text.
Source: Columns 1, 2 and 4 from ACS. Column 3 = (Column2-Column 1)/3 + Column 2.
Difference Between the Actual and Expected Count in 2020 for El Salvador, Guatemala, and Honduras Combined, by Metropolitan Area of Residence.
Note: Excludes arrivals in the year of the survey. See text.
Source: Columns 1, 2 and 4 from ACS. Column 3 = (Column2-Column 1)/3 + Column 2.
Methods and Sources of Data
The Census Bureau uses two primary methods to evaluate the completeness of coverage in censuses and surveys: The Post-Enumeration Survey 1 and the Demographic Analysis Method. 2 The Census Bureau evaluations are done at the national level. Organizations that estimate the size of the undocumented population base their estimates on a much smaller segment of the total US population – the noncitizen population that arrived after 1981. In the text that follows, the terms “the noncitizen population,” or “the population,” refer to the noncitizen population that arrived after 1981 unless stated otherwise.
Detailed information required to make precise estimates of undercount for noncitizens is not available. For this report, we identified likely undercount in the 2020 ACS by analyzing recent trends in the population. First, we derived an “expected” population in 2020 by adding the average change for 2016–2019 to the population in 2019:
The Demographic Analysis Method cannot be used to derive undercount estimates for this population because data are not available to measure all the components of change. We use the term undercount here, but it might properly be called “likely undercount.” Caution should be used in interpreting the results because of uncertainty in the estimates of the expected population in 2020 and because of sampling variability.
In deciding which estimates to show in the tables, we inspected each trend line to determine whether factors other than undercount, such as increased emigration or random fluctuations, could explain the difference between the count and the expected population in 2020. To avoid overestimating undercount, we display rates for only those countries that had undercount rates of 10 percent or more and had 20,000 or more noncitizens in 2019.
The data for Guatemala shown in Figure 1 illustrate the method used to estimate likely undercount in the 2020 ACS. The black line represents the actual count in each ACS, and the red line indicates the expected population for 2020 based on the trend from 2016 to 2019. The estimated undercount for Guatemala in the 2020 ACS was 192,000, or 24 percent (based on unrounded numbers). The same type of analysis of 2020 ACS data was done for 149 countries or areas, by age, sex, state, and metropolitan area of residence.
Noncitizens from Guatemala Arrived After 1981.
The estimated undercount rates shown in this report are in addition to the rates that would occur in a “normal” year. Census Bureau estimates of undercount in the 2010 ACS are described in Appendix (Jensen 2015). Based on those estimates, the annual ACS data from 2010 to 2019 likely had undercount of about 5 percent, plus or minus a few percentage points.
Before proceeding, it is important to note an aspect of the data used in this analysis. As noted, the annual data shown in Figure 1 and in Tables 1–5 below are for the noncitizen population that arrived after 1981. However, in every case, one relatively small segment of that population was omitted from the analysis: noncitizens that arrived in the year the ACS was conducted. It was necessary to omit new arrivals from the analysis because arrivals in 2020 were reduced more by restrictions on entry than by undercount. 3 Including new arrivals in the analysis would have overestimated undercount. Appendix provides additional information.
Undercount by Country of Origin
The first step in the analysis was to determine which countries had undercount large enough to be detected using trend analysis. We analyzed ACS data for 2016–2019 for 149 countries or areas included in the ACS data. 4 Table 1 shows the countries most likely to have undercount in the 2020 ACS. The undercount rates for 2020 shown in Column 6 of each table in this report are in addition to undercount inherent in the annual ACS data (see Appendix B).
Undercount in the 2020 ACS was not confined to just the 14 countries listed in Table 1. Undercount rates are shown for only those countries with rates of 10 percent or more and 20,000 or more noncitizens in 2019. The procedure used to estimate undercount was not precise enough to derive estimates for smaller populations and those with relatively lower undercount rates.
Total undercount was almost 1 million for the 14 countries shown in Table 1. Four Central American countries – El Salvador, Guatemala, Honduras, and Nicaragua – had relatively large undercount rates in 2020 (Table 1). Of those, Guatemala had the highest rate, 24 percent, followed by El Salvador, Honduras, and Nicaragua at 14 percent each. Between a quarter and a third of noncitizens from Saudi Arabia, Iraq, and Cuba were omitted from the 2020 ACS (Table 1).
About 222,000 noncitizens from Mexico were omitted in 2020 but that constitutes only 4 percent of the expected population from Mexico in 2020. The relatively small percent of undercount for Mexico could be explained by the fact that so many undocumented residents voluntarily returned to Mexico from 2017 to 2019. About 865,000, or 290,000 per year, emigrated to Mexico voluntarily in the 2016–2019 period compared to 120,000 each year from 2010 to 2016 (Warren 2021). Also, undercount rates drop by the length of time in the country, and 80 percent of the noncitizen population from Mexico has resided in the United States for 10 years or more.
The analysis of data for 149 countries or areas yielded only fourteen countries that had undercount large enough to be detected by trend analysis and included in Table 1. Many of those countries have substantial undocumented populations in the United States. In fact, 74 percent of undocumented residents counted in the 2019 ACS were from the 14 countries listed in Table 1. 5 The estimates for Europe are included in Table 1 to illustrate the apparent lack of undercount for countries that have few undocumented residents.
If the Covid-19 pandemic had been the primary cause of undercount in the 2020 ACS, undercount rates would have been closer to uniform across countries. As the estimates in Table 1 show, that was not the case. Undercount was about a million in the 2020 ACS for the 14 countries listed in Table 3. About three-quarters of the undocumented population are from those countries. Europe, the home of relatively few undocumented residents, had no detectable undercount. In fact, the 2020 ACS count was very close to the expected population for all 40 of the European countries or areas that we examined. These findings indicate that the undercount in 2020 shown in Table 1 was not due to Covid-19 but to the Trump administration’s interference in the 2020 census, its abusive rhetoric, and its anti-immigrant policies (Kerwin and Warren 2019).
Undercount by Period of Entry
Table 2 shows estimates of undercount in the 2020 ACS for the six countries (combined) that had the largest estimated undercount in Table 1. The undercount rates in Column 6 of Table 2 are smaller than the rates shown in other tables because noncitizens from Mexico make up a large proportion of the data, and the undercount rate for Mexico was only 4 percent, as shown in Table 1.
As expected, the undercount rates in Table 2 are highest for more recent arrivals. Unfortunately, the analysis was limited to noncitizens that arrived before 2016. 6 Undercount rates for noncitizens that arrived after 2015 would be considerably higher than the 12 percent shown for arrivals in the 2006–2015 period (Table 2). Based on the progression of rates in Column 6, with the rates doubling in each 10-year period, undercount rates for noncitizens that arrived in the Trump era would likely be 25 percent or higher.
Noncitizens that arrived more recently would be expected to have higher undercount rates than long-term residents because they have had little experience with US censuses and surveys and because of increased official hostility toward immigrants since 2016. Organizations that use ACS data to estimate undocumented immigration should adjust the 2020 ACS data to take account of differences in undercount by period of entry.
Fully addressing the consequences of the undercount described here is beyond the scope of this report. Note, however, that the undercount in the ACS described here very likely occurred in the 2020 census as well. That means undercount will be statistically “imbedded” in ACS data throughout the next decade. 7
Analysis of Data for El Salvador, Guatemala, and Honduras Combined
The next step planned for the analysis was to compare the 2020 ACS count with the expected population for each country by age and sex and for states and metropolitan areas to identify additional differential undercount in all segments of the population. For example, was undercount larger for children or adults? By gender? In certain states? In large or small metro areas? The numbers involved were too small in many cases to carry out such detailed analyses. Instead of examining these questions for each of the 14 countries shown in Table 1, we focused the remainder of the analysis on the combined populations of three Central American countries that had large undercount rates in the 2020 ACS and also have large numbers of recent arrivals – El Salvador, Guatemala, and Honduras.
Undercount by Age and Sex for El Salvador, Guatemala, and Honduras Combined
Table 3 shows the same type of analysis described for each country of origin, but the data are for El Salvador, Guatemala, and Honduras combined (hereafter “the three countries”).
The estimated undercount rate was 18 percent for the three countries combined (Table 3). Males and females from these countries had the same undercounted rate, 18 percent. The undercount rate tended to decline with age, but undercount was spread among the age groups, with undercount for most of the age and sex groups in the 10–20 percent range. The similarity of the rates shown in Column 6 suggests that undercount was not confined to specific age or sex groups but was evenly distributed over the entire population (Table 3). This information should be useful for devising procedures for adjusting the data for undercount.
Undercount by State for the Three Countries
The estimates by State of residence in Table 4 range from 10 percent for California to 77 percent for Kansas. These results should be interpreted with caution because of the uncertainty inherent in the trend analysis method and because of sampling variability. However, the figures in Table 4 suggest that undercount from the three countries was larger in 2020 for states with relatively smaller noncitizen populations. It is possible that concentrations of noncitizens in the large metro areas lessened some of the fear of the census. Another possible reason for relatively high undercount in smaller areas is that the population in those areas has arrived more recently, and recent arrivals have larger undercount rates.
Undercount averaged 16 percent for the six states with the largest populations in 2019 (Table 4), and 52 percent for the six states with the smallest populations. An interesting feature of the estimates for the last six states shown in Table 4 is that, except for Illinois, approximately 50 percent of the population in 2019 arrived from 2016 to 2019. 8 In general, recent arrivals have higher undercount rates than long-term residents.
Undercount Rates by Metro Area for the Three Countries
The estimated undercount for the three countries was lowest in the Los Angeles and Houston metro areas (10 percent) followed by San Francisco and Boston at 12 percent, and Miami at 13 percent (Table 5). These are five of the eight areas with the largest population in 2019. As was the case with states, metropolitan areas with the smallest populations analyzed in this paper had the highest undercount rates: Phoenix (61 percent), Austin (54 percent), and Chicago (32 percent). The noncitizen populations in Dallas, Austin, and Phoenix grew rapidly from 2016 to 2019.
Discussion
The list of countries in Table 1 – rather, the brevity of the list – provides important information about undercount in the 2020 ACS, and likely in the 2020 Census. This analysis was done for 149 separate countries or areas, but only 14 countries had sufficient evidence of undercount to be included in Table 1. This finding indicates that Covid-19 was not responsible for the large undercount rates for the countries shown in Table 1. That is, Covid-19 could not have caused the undercount in those 14 countries while leaving the remaining 135 countries unaffected.
The 14 countries shown in Table 1 include populations (mostly Latin American and Caribbean) with large numbers of more recent arrivals. Most of the countries also include large numbers of undocumented residents. In fact, these countries comprised 74 percent of the undocumented population counted in the 2019 ACS.
The large undercount rates shown in Table 1 occurred in 2020 but not in 2019 or earlier years of the Trump administration. That means the abrasive rhetoric and anti-immigrant policies of the Trump administration, which spanned five years, was not the principal cause of the striking increase in undercount in the 2020 ACS. That leaves the adverse publicity associated with administration’s unprecedented interference in the 2020 count as the likely proximate cause of the substantial undercount rates in 2020.
The estimate of 23 percent undercount for noncitizens from Cuba is puzzling. Based on the previous discussion, an undercount that large suggests the presence of large numbers of undocumented residents. Before 2017, undocumented migration from Cuba was very low because any Cuban remaining in the United States more than one year could adjust to legal permanent resident (LPR) status under the Cuban Adjustment Act of 1966. In 2017, President Obama limited LPR status to migrants from Cuba that arrive legally. Although it is possible that the undocumented population from Cuba has increased since that policy change, the ACS data do not support such growth. The number of arrivals from Cuba counted in the ACS each year from 2017 to 2019 is about the same as the number of arriving LPRs. No data are available to estimate the growth of the undocumented population from Cuba after 2019.
The estimates of undercount in Table 3 indicate that undercount of noncitizens in 2020 was evenly distributed by age and sex and was not confined to a particular age or sex category. This information should be useful for devising procedures for adjusting the data for undercount.
Even though undocumented residents were affected the most, family members and others residing in their households, including LPRs and US-born children, were also missed. Undercount rates were low for naturalized citizens regardless of the country of origin. The 2020 ACS count of naturalized citizens from El Salvador, Guatemala, and Honduras (combined) was within 3 percent of the expected number in 2020.
The estimates of undercount by state and metro area in Tables 4 and 5 indicate that undercount from the three Central American countries was larger in 2020 for states with relatively smaller numbers of noncitizens. In general, undercount rates in smaller metro areas are higher for recent arrivals than for long-term residents and higher in smaller areas than in large ones.
Conclusion
The estimates presented here demonstrate that the 2020 ACS, and likely the census, omitted at least 1 million noncitizens that arrived after 1981. Undercount for Mexico, India, El Salvador, Guatemala, Honduras, and Cuba combined was about 835,000, or 84 percent of the undercount identified in this report. For the Central American countries highlighted here, undercount was relatively higher in states and metro areas with smaller populations and where immigration increased rapidly from 2016 to 2019.
Most of the undercount of noncitizens in the 2020 ACS occurred among countries with relatively large undocumented populations. Organizations that derive estimates of the undocumented population should take account of these estimates of undercount to avoid skewed results.
This analysis indicates that Covid-19 was not responsible for the large undercount rates shown here. It could not have caused high undercount rates for 14 countries while leaving the remaining 135 countries unaffected. Likewise, the anti-immigrant rhetoric and policies of the Trump administration spanned five years, so that would not explain the upsurge in undercount in 2020 relative to earlier years. That leaves the unwarranted inference in the 2020 census, especially efforts to keep undocumented residents from being included in the count, along with the attendant negative publicity surrounding those efforts, as the most likely reason more than 1 million noncitizens, mostly undocumented residents, were omitted from the 2020 ACS, and, most likely, from the 2020 census.
These estimates of undercount demonstrate the destructive power of virulent anti-immigrant speech, xenophobic policies, and political interference in nonpartisan activities such as the census. The huge undercount of noncitizens in 2020 quantifies the damage such actions can do to just one of our essential institutions – the decennial census – and its offshoot, the American Community Survey.
The omission of 1 million or more from the 2020 ACS and the decennial census has reverberations that cannot be measured precisely. However, this report provides a roadmap to which jurisdictions and residents have been most damaged by the Trump administration’s interference in the 2020 census. The effects are real, lasting, and damaging to the lives of millions of people. The post-census population estimates used to derive ACS estimates for the next 10 years will inherit the defects described here. The distribution of population, political power, and financial resources will be skewed by the flawed numbers for an entire decade.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Disclosures
This article did not involve human or animal subjects, which obviates the need for informed consent.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.