Does enumerating undocumented residents in the US census affect congressional apportionment?

Abstract

In recent years, American political figures, media pundits, and others have argued that undocumented residents in the United States should not be included in census data used for congressional apportionment. They argue that including them unfairly benefits some states—especially politically Democratic-leaning (or “blue”) states, according to most arguments—at the expense of others. Some people have speculated that many more House seats and Electoral College votes would have been apportioned to politically Republican-leaning (or “red”) states and that many fewer would have been apportioned to politically “blue” states if undocumented residents were excluded from apportionment data. The only systematic empirical examinations of this question occurred prior to the 2020 census and projected—mostly inaccurately—how apportionment would differ after the 2020 census under this hypothetical. Does including undocumented residents in US census data used for congressional apportionment alter (ⅰ) political party representation in the House of Representatives or (ⅱ) presidential Electoral College votes? We use annual state-level estimates of the undocumented resident population at each census date to examine the consequences of including them in official apportionment data since 1980. If undocumented residents had been entirely excluded from census data used for apportionment, no more than two House seats and three Electoral College votes would have shifted between political parties in any year since 1980; this would have had no bearing on party control of the House or the outcome of presidential elections.

Significance Statement.

The main constitutional purpose of the decennial US census is to allocate seats in the House of Representatives—and thus Electoral College votes—to states. In practice, the census enumerates all residents of each state—regardless of their citizenship status or legal right to live in the country. There have recently been high-profile claims that including undocumented residents in census data used for apportionment unfairly inflates the number of House seats and Electoral College votes for politically “blue” states at the expense of politically “red” states. We find that this is largely not the case; including undocumented residents in census data used for apportionment has had a negligible impact on apportionment since at least 1980.

Introduction

The primary constitutional purpose of the US census is to apportion seats in the House of Representatives—and thus Electoral College votes—to states. Section 2 of the 14th Amendment states that “Representatives shall be apportioned among the several states according to their respective numbers, counting the whole number of persons in each state.” As a result, residents are enumerated in the decennial US census without regard to their citizenship or legal status (1, 2).

In recent years, there have been high-profile claims that including undocumented residents in census data used for apportionment inflates House seats and electoral votes for some states at the expense of others; most often, the claim is that “blue” or Democratic-leaning states benefit at the expense of “red” or Republican-leaning states (2, 3). For example, in a February 2024 commentary the Heritage Foundation argued that “Democratic-controlled states have gained congressional seats by welcoming and harboring illegal aliens” (4). Likewise, in a March 2024 interview with journalist Don Lemon, entrepreneur Elon Musk argued that “…the Democrats would lose approximately 20 seats in the House if illegals were not counted in the census and that's also 20 less electoral votes for president.”

Is it true that including undocumented residents in census data used for apportionment has had these consequences in recent decades? Have any such effects meaningfully altered the balance of power in the US House of Representatives or the outcome of presidential elections?

In this article, we address these questions using high-quality state-level estimates of the size of the undocumented resident population at each census date. We begin by describing how many House seats—and thus how many Electoral College votes—individual states would have gained or lost under the counterfactual situation in which no undocumented residents were included in data used for apportionment after the 1980 through 2020 censuses. We then examine whether party control of the House and/or the outcome of Electoral College votes would have changed under this counterfactual situation.

These questions have never previously been empirically addressed for censuses prior to 2020. Three studies conducted just prior to the 2020 census sought to project the number of House seats that would change if undocumented residents were removed from the 2020 population used for apportionment (2, 5, 6). All three projected the population to be used for apportionment after the 2020 census—as opposed to using actual census apportionment data—and all three projected the number of undocumented residents that would be counted in each state in 2020.

None of the sets of projections turned out to be very accurate. Camarota and Zeigler (6) concluded that Alabama, Minnesota, and Rhode Island would each lose a seat in 2020, and that Arizona, Florida, and Texas would each gain a seat. In fact, none of these states gained or lost a seat in 2020. Passel and Cohn's (5) projections for 2020 were incorrect for four states. They projected gains of three seats for Texas and two seats for Florida; the actual gains were two seats for Texas and one seat for Florida. Their projected losses of one seat each for Minnesota and Rhode Island did not occur. Baumle and Poston's (2) projections were equally off the mark.

These three studies followed the general methodological approach that we use here. That is, they distributed House seats (and thus electoral votes) to each state based on population data that included or excluded undocumented residents. The shortcomings of those three studies are not in their methodology but in the data they used. In all three cases, the projected apportionment population was based on 2010 census data “carried forward” to 2017 or 2018 and then projected forward to 2020. Their projections of the size of states’ undocumented resident population to 2020 had the same limitations. In contrast, the estimates we provide are based on actual apportionment populations for each census and estimates of the number of undocumented residents in each state derived for the same year as each census.

In short, our research questions have never been addressed for censuses prior to 2020, and for 2020 they have never been addressed using actual (as opposed to projected) populations.

Results

State-level estimates of the undocumented resident population, 1980–2020

Table S1 reports (a) the total population count used for apportionment for each state in each decennial census between 1980 and 2020 and (b) estimates of the number of people who were undocumented residents in each state as enumerated in each census year. The former determines the number of House seats apportioned to each state after each census.

One notable trend in the state-specific estimates of the numbers of undocumented residents: Whereas the percentage of all undocumented residents living in Democratic-leaning California fell from 50% (1.024 million/2.028 million) to 22% (2.072 million/9.470 million), the percentage living in Republican-leaning Texas rose from 9% (186,000/2.028 million) to 17% (1.602 million/9.470 million).

Effects of enumerating undocumented residents on apportionment, 1980–2020

Figure 1 depicts for each census the number of House seats each state would have gained or lost under the hypothetical situation in which undocumented residents were removed from census data used for apportionment. The data underlying these figures appear in Table S2. In general, California and Texas would lose seats under this hypothetical situation, and a variety of states would gain seats. In 1980 and 2020, only two seats would have switched states; in 2010, five seats would have switched states. These results are in stark contrast to recent claims that more than 20 seats would change hands if undocumented residents were removed from the data used for apportionment.

Fig. 1.

Change in apportionment if undocumented residents were removed from census data used for apportionment, 1980–2020.

Consequences for party control of the US House of Representatives, 1980–2024

What impact would the above results have had on political party representation in the US House of Representatives? Figure 2 depicts—for each congress elected between 1982 and 2024—(a) the actual advantage in seats for the party in the majority and (b) the advantage in seats for the party in the majority that would have been observed had undocumented residents been excluded from the data used for apportionment. Table S3 reports the data underlying Fig. 2.

Fig. 2.

Actual party control (bars) of the US House of Representatives vs. hypothesized party control (dots) if undocumented residents were removed from census data used for apportionment, 1980 to 2024.

As described above, we cannot know with certainty which party would have controlled a hypothetically newly added House seat in any state; nor can we know with certainty which party would have suffered from a lost seat. However, it seems reasonable to assume that the political party that controlled redistricting after the census would maximize their gains and minimize their losses. Under this assumption, Fig. 2 shows that removing undocumented residents from the data used for apportionment would make very little difference for party control of the House: In no congress would more than two seats have switched from Democratic to Republican control. Even if we were to make the most extreme possible assumption—that every “lost” seat would have cost Democrats and every “gained” seat would have benefitted Republicans—the number of seats switching parties could never have exceeded five (because, as shown in Fig. 1, no more than five seats switch states).

Consequences for presidential elections, 1980–2024

What impact would excluding undocumented residents from census data used for apportionment have had on Electoral College votes in the 1984 through 2024 presidential elections? Figure 3 shows both the actual margin of victory for Democrats and Republicans in each election, as well as the hypothetical margin of victory that would have been observed had undocumented residents been excluded from apportionment data. The data underlying Fig. 3 appear in Table S4.

Fig. 3.

Margins of victory in electoral college if undocumented residents were removed from census data used for apportionment, 1984 to 2024.

Excluding undocumented residents from the data used for apportionment would have had almost no bearing on Electoral College totals for presidential elections since 1980; at most, the margin of victory would have changed by three votes. Partly this is because—as shown in Fig. 1—no more than five electoral votes switch states in any election. But it is also partly because “lost” and “gained” seats do not always favor one political party over the other. While it is true that excluding undocumented residents from the data used for apportionment would have cost traditionally Democratic-leaning California electoral votes, it also typically would have cost traditionally Republican-leaning Texas electoral votes. The states that would have gained votes from California and Texas are also diverse with respect to political party control.

Discussion

In recent years, political figures, media pundits, and others have argued that undocumented residents should not be included in census data used for apportionment. They argue that including them unfairly benefits some states—mainly Democratic-leaning states, according to most arguments—at the expense of others. Some people making these arguments have speculated that many more House seats and Electoral College votes would have been apportioned to “red states” and that many fewer would have been apportioned to “blue states” if undocumented residents were excluded from apportionment data. The only systematic empirical examinations of this question occurred prior to the 2020 Census and projected—mostly inaccurately—how apportionment would differ after the 2020 Census under this hypothetical.

In recent decades, how much of a difference has it made to apportionment to include undocumented residents in apportionment data? Have any such effects meaningfully altered the balance of power in the US House of Representatives or the outcome of presidential elections? We find that since 1980, including undocumented residents in census data used for apportionment has (ⅰ) made no more than two seats difference for party control of the House and (ⅱ) made no more than three votes difference in the Electoral College.

Our basic finding—that including unauthorized residents in census apportionment data has had little impact on House control or presidential elections—is all about the past. What might we say about the future? We would note two trends that suggest that including unauthorized residents in census apportionment data will continue to have minimal effects on these outcomes. First, the undocumented population in California dropped considerably from 2010 to 2020, declining from 2.7 to 2.1 million in this period (Table S1). The relatively large proportion of undocumented immigrants in California in the previous censuses was responsible for the results for California in 2000 and 2010 (Fig. 1). If undocumented residents were removed from its apportionment population, California would have lost three seats after those censuses (Fig. 1). The lower number counted in California in 2020 reduced that number to a single seat.

The second reason that continuing to include unauthorized residents in the apportionment population will likely have minimal consequences for these outcomes is that the statistical effect of counting them is limited. In 2020, Texas (or California) would have gained 11 seats in the extreme case that every undocumented resident in the entire country was counted in just that state. As we have shown, in the past five censuses the largest number of seats that changed because of undocumented residents was California with three seats. At the other end of the spectrum, no state would gain or lose a seat in the House if the count of undocumented residents mirrored the distribution of legal residents in every state, regardless of the number of undocumented residents counted.

Finally, the question might be asked: What would be the effects on the estimates shown here if the undocumented resident population really were as large as 20 or 30 million, as some nondemographers have asserted. If that much higher estimate—which has no basis in empirical fact—were true, it would have to mean that (contrary to prior evidence) most undocumented residents are not enumerated in the census. Thus, the answer to the above question would be: “No effect.” This is because the undocumented resident population relevant to the apportionment process is the estimated population counted in the census rather than the total undocumented population.

Limitations

We perceive three possible limitations of our research. First, we rely on estimates of the sizes of states’ undocumented resident populations in each census. Although the methods used to derive these estimates are well-documented, well-established, and validated, they are still estimates. Second, our conclusions depend on a hypothetical: If undocumented residents were not counted in census data, what would apportionment look like? As with any unobserved counterfactual, we do not know what else about the world would change if we changed methods for apportionment in this way. Finally, we have not said anything about the consequences for federal appropriations to the states or other potential consequences of including undocumented residents in census data used for apportionment.

Conclusions

Since at least 1980, including undocumented residents in census data used for apportionment has had minimal impact on party representation in the US House of Representatives and on the outcome of presidential Electoral College votes. Arguments to the contrary rely on (ⅰ) projected apportionment populations and dubious estimates of the sizes of states’ undocumented resident populations and/or (ⅱ) unrealistic assumptions about how much the major political parties benefit or suffer from including undocumented residents in apportionment data.

Materials and methods

For each census from 1980 to 2020, we calculate the number of House seats that would have been apportioned to each state if undocumented residents had been excluded from the census count. The calculations require two sets of data and a formula for apportioning the 435 seats to states. That formula and one of the two sets of data—the total population count used for apportionment for each state after each census—are immediately available from the US Census Bureau (7). The second set of data required is the estimated number of undocumented residents, by state, counted in each of the five censuses from 1980 to 2020.

Estimating the number of undocumented residents

Our estimates were derived by analyzing data from censuses and from the American Community Survey (ACS). In most cases, the estimates were derived by comparing the number of foreign-born persons enumerated with estimates of the legally resident foreign-born population. The difference is the estimated number of undocumented residents counted in the census or ACS. The methodology used to produce these estimates is described more completely and validated elsewhere (8–10).

Using these methods, estimates of undocumented residents in each state have been published for each of these census dates. State-specific numbers of undocumented residents counted in the 1980 census were estimated by Passel and Woodrow (9). Estimates for each state in 1990 and 2000 are available from Warren and Warren (10). The Center for Migration Studies of New York (CMS) derived estimates of undocumented residents counted in each state in 2010 and 2020 (8). The CMS methodology used to compile all of estimates used here is described in detail elsewhere (11).

The total numbers of undocumented residents shown in Table S1 for each census date range from 2.0 million in 1980 to 9.5 million in 2020. Much higher estimates have been reported in the media and by nondemographers. So far, those estimates have been based on conjecture or unsupported assumptions; see Warren (11) for a critique of these dubious estimates and a defense of the basic methodology we deploy here.

Recalculating apportionment after removing undocumented residents

We began by applying the apportionment formula (7) to the total population count used for apportionment for each state after each census between 1980 and 2020; following that formula, we then allocated the 435 seats for each of the five censuses. In so doing, we first verified that our allocation procedure and the total population counts yielded the actual number of House seats for each state after each census.

Next, for each state and census, we subtracted the estimates of undocumented residents from the apportionment population and used those smaller revised populations to hypothetically reallocate the 435 seats. The differences between the state-by-state numbers of seats allocated using the official apportionment population and the numbers allocated based on the population without undocumented residents demonstrate the consequences of including undocumented residents in apportionment data after each census. Results reported below can be immediately replicated by applying the apportionment formula to the data provided in Table S1.

Hypothetical outcomes of house and presidential elections

Using the numbers of House seats (and thus Electoral College votes) that hypothetically would have been apportioned to each state after each census if undocumented residents had been removed from census data used for apportionment, we then estimated (ⅰ) how many House seats the two major US political parties would have gained or lost in each Congress between 1980 and 2024 and (ⅱ) how many electoral votes the two major party candidates would have gained or lost in each presidential election between 1980 and 2024.

For House seats, we cannot know with certainty which party would have won seats newly added to a state after our hypothetical reapportionment—or which party would have lost seats. Instead, we used information about political party control of states’ redistricting efforts after each reapportionment (12, 13). We assumed that the party that controlled state redistricting would benefit from any gained seats and not suffer from any lost seats. In cases in which states’ redistricting efforts were governed by politically split state governments, by independent commissions, or by nonpartisan processes, we alternated assigning seats to Democrats and Republicans. Based on these assumptions and using the hypothetical numbers of House seats for each state, we calculated the number of House seats that Democrats and Republicans would have won in each election from 1980 through 2024 had undocumented residents not been included in the census data used for apportionment after the most recent census.

For Electoral College votes, we assumed that the party winning each state in each election would not have changed; all that would change would be the number of electoral votes awarded to the party winning that state. Using the hypothetical numbers of Electoral College votes for each state, we calculated the number of electoral votes each major party candidate would have received had undocumented residents not been included in the census data used for apportionment after the most recent census.

Supplementary Material

Supplementary material is available at PNAS Nexus online.

Funding

We also appreciate center grant support provided to the Minnesota Population Center by the National Institute for Child Health and Human Development (grant no. P2CHD041023) and to the Life Course Center by the National Institute on Aging (grant no. P30AG066613).

Author Contributions

Conceptualization, methodology investigation: R.E.W.; visualization and writing—review & editing: R.E.W. and J.R.W.; writing—original draft: J.R.W.

Data Availability

All data are included in the manuscript and/or Supplementary material.