July 24, 2020 – The Trump administration announced Tuesday that it will exclude unauthorized immigrants from the 2020 Census data used to reapportion the 435 seats in the U.S. House of Representatives among the 50 states, by matching Census data with administrative records. Without commenting on its constitutionality or legality, this plan seems likely to exclude millions of U.S. citizens from the once-a-decade exercise to decide political representation. The reason is that there is no foolproof method of knowing which Census respondents are U.S. citizens, which are legal immigrants, and which are unauthorized immigrants. In our estimation, up to 20 million U.S. citizens could incorrectly be lumped together with unauthorized immigrants due to matching errors. Such errors are likely to be greater in low-income urban and rural communities, thereby exacerbating any undercount that may occur in those communities and reducing their voting power relative to more affluent communities.
Since passage of the 14th Amendment, House seats have been allocated based on states’ total populations. This is the first time that the count would be limited to U.S. citizens and lawfully present noncitizens. In President Trump’s July 21 memorandum, he argued that he has the discretion to define “inhabitants”—the people living in each state—as those residing legally in them, and that counting unauthorized immigrants for apportionment purposes would provide incentives for states to enact policies that thwart federal immigration enforcement.
The stakes are high, as states with smaller numbers of unauthorized immigrants could gain seats in Congress at the expense of states with larger such populations. Moreover, states could follow the administration’s lead and exclude noncitizens or unauthorized immigrants from the data when drawing congressional and state legislative districts, believing they are tilting the allocation of seats toward suburban and rural areas at the expense of the large cities where most unauthorized immigrants reside.
Matching Census and Administrative Data: An Exercise Fraught with Errors
To implement its plan to exclude unauthorized immigrants, the federal government must identify the citizenship and immigration status of Census respondents. While demographers have produced credible state and national estimates of the unauthorized immigrant population, the Supreme Court ruled in 1999 that “actual enumerations” are required for apportionment purposes. After the Supreme Court in 2019 invalidated the administration’s attempt to add a citizenship question to the 2020 Census, the president issued an executive order requiring the Census Bureau to match Census respondents with administrative data from various federal agencies to “help establish citizenship for 100 percent of the population.” This week’s proclamation goes further, requiring the Census Bureau to determine both citizenship and immigration status.
But is it really possible to use administrative data to determine citizenship and immigration status for the entire U.S. population? Prior Census Bureau experience with data matching suggests this will be difficult if not impossible, and risks wrongly identifying millions of U.S. citizens as unauthorized immigrants and removing them from reapportionment and redistricting data.
The Census collects names, birthdates, and addresses, and this information can be used to match people to administrative records maintained by the Social Security Administration, Internal Revenue Service, Department of Homeland Security (e.g., naturalizations and green cards), State Department (visas and passports), and other federal agencies (e.g., regarding the use of Medicare or Medicaid). The Census Bureau is also requesting data from states. The administration’s plan is to use these administrative data to identify which individuals in the Census are U.S. citizens or lawfully present noncitizens. Those not identified as citizens or lawfully present could be classified as unauthorized immigrants. The problem is that millions of citizens and legal immigrants cannot be matched to administrative records.
Earlier Evidence of Data-Matching Problems
In 2018, the Census Bureau attempted to match 2010 Census responses to administrative data. In 9 percent of cases (28 million out of 309 million people counted in 2010), the matches failed or the Census Bureau collected insufficient information to even attempt a match. With a 9 percent failure rate and an estimated current population of 330 million, this year’s Census could fail to match 30 million people and risk improperly determining their citizenship and immigration status.
The most credible estimates of the unauthorized population fall in the 10 million to 12 million range. Thus, if there were 30 million unmatched people, subtracting unauthorized immigrants would leave between 18-20 million U.S. citizens and lawfully present noncitizens who might be mistakenly excluded.
A separate, unpublished analysis by one of the authors bolsters the idea that administrative data matching could mislabel tens of millions of U.S. citizens. When the Census Bureau’s 2017 American Community Survey—which asks people about their citizenship—was compared to a Census administrative data file, about 30 million people could not be matched (similar to the number in the 2010 Census), with two-thirds (20 million) of those non-matches U.S. citizens.
The Difficulties Inherent in Matching Data
Why do so many matches fail? Some individuals, including many unauthorized immigrants, are not present in administrative data. But other individuals, many of whom are U.S. citizens, cannot be matched because name, birthdate, and address information in the Census is inaccurate or incomplete.
To begin with, many people do not respond to the Census initially, and the Census Bureau cannot locate them to conduct a follow-up in-person interview; in such cases, people are counted based on the size of the housing unit, but their names and birthdates are not collected. Some people may respond to the Census but fail to provide their name or birthdate. Others may use a different version of their name on the Census form than used in government records (e.g. a nickname or a different representation of multiple last names). Still others have moved between the time of the administrative data collection and the Census, and if they have missing names or birthdates, or a common name, they are also unlikely to be matched. The quality of the matching is only as good as the quality of Census data collection, no matter how many different sources of administrative data are employed.
It is quite likely that the Census data quality will be lower in 2020 than it was in 2010, making it even more difficult to match to administrative data. The 2020 Census was scheduled to start in March, even as severe COVID-19 outbreaks were occurring in several states. While the pandemic did not affect online, telephone, or mail-in responses, it delayed in-person follow-up and made it much more difficult amid fears Census takers and respondents could be exposed to the virus. Thus, it is highly likely that the 2020 Census will collect incomplete data from or miss more people altogether than the 16 million omitted from the 2010 Census.
The administration is still working on its plans for the administrative data-matching process. Prior Census Bureau reports suggest that the government would use statistical estimation to fill any gaps in data matching between the Census and administrative data. But using estimates instead of actual counts of people seems likely to run afoul of the Supreme Court’s 1999 ruling that “actual enumerations” must be used.
Who Will Be Most Affected under This Plan?
Who is most likely to be missed or unmatched in the 2020 Census? Data collection has historically been more difficult in densely populated urban areas where more people crowd in each housing unit, and in rural agricultural areas where migrant farmworkers live for part of the year. It is also difficult to collect data from younger people, especially those who move back and forth from home during their college years. During the pandemic, it may be more difficult for Census workers to follow up with nonresponders in hotspots—many of which are also in major cities or agricultural areas. Further, large numbers of people who had the option fled urban COVID19 epicenters such as New York City for other parts of the country and may have responded to the Census in that alternate location. In addition to an undercount, the Census is likely to produce an undermatch in such areas, leading to higher rates of exclusion of misidentified “unauthorized immigrants.” Given that many coronavirus hotspots are occurring in lower-income neighborhoods where essential workers live, the undercount combined with the undermatch could further reduce the voting power of U.S. citizens living in these areas.
The process of matching the decennial Census to administrative data is too error-prone to rely on for an important constitutional responsibility such as reapportionment. Prior experience with matching suggests up to 20 million U.S. citizens could be excluded from this crucial exercise of American democracy. Those not matched are likely to live in low-income urban or agricultural rural areas. To date, the Trump administration and the Census Bureau have offered only limited information on how the matching will happen or how to address difficulties in the process. This lack of transparency could further reduce trust not only in the Census, but also in the fairness of our political system.
Randy Capps is Director of Research for U.S. Programs.
Jennifer Van Hook is an MPI Nonresident Fellow and demographer at the Population Research Institute at The Pennsylvania State University.
Julia Gelatt is a Senior Policy Analyst with the U.S. Immigration Policy Program.
Founded in 2001, the Migration Policy Institute (MPI) has established itself as a leading institution in the field of migration policy and as a source of authoritative research, learning opportunities, and new policy ideas.