Who counts in 2030

On September 10, 2026, the Census Bureau proposed changing who gets counted for congressional apportionment. Under proposed section 60.4, foreign citizens are counted only if they are “either citizens or lawful permanent residents of the United States.” Everyone else — people on student and work visas, people awaiting asylum decisions, people here illegally — would be, in the rule’s words, “[n]ot counted for apportionment.”

The comment period, first set to close October 13, has been extended to November 2, 2026.¹

Who stops being counted

The rule draws its line at lawful permanent residence, not citizenship. Citizens count. Green-card holders count. Everyone else living here does not.

That removes about 11 million people from the apportionment base — out of roughly 23 million non-citizens in the country. Close to half the non-citizen population goes on being counted, which is why the effect below is about half the size of what a citizen-only count would produce.

projected population in 2030, fifty states

352.4 million

not counted under the rule (3.2%)

11.4 million

counted for apportionment

340.9 million

people per seat, down from 810,000

784,000

Where those 11 million live is what decides the rest, and they are nowhere near evenly spread. The rule takes 5.9% of Texas’s projected population and 4.7% of California’s, against 1.5% of Wisconsin’s and 1.3% of Michigan’s — roughly four times as much, proportionally, from the first pair as the second. Apportionment divides a fixed 435 seats by relative population, so a state loses ground whenever it is cut harder than the rest. No state has to lose a single resident for its delegation to shrink.

Three seats move

The middle column is the story most people already know: on current trends, twelve seats move between sixteen states between 2020 and 2030, with Texas up four and Florida up three. The right-hand column is what the proposed rule would do on top of that — and it is a much smaller change. Three seats, six states.

Texas loses

The obvious reading of this rule is that it helps Republicans. The arithmetic is not that clean.

Texas is projected to gain four seats by 2030 under current rules — the largest gain in the country. Under the proposed rule it gains three. Texas grew the way it did partly through immigration, and a rule that stops counting part of that growth takes back some of what it gave.

The seats do land in states Trump carried in 2024 — Michigan, Tennessee, and Wisconsin — so on the 2024 map the net is two seats toward Trump-won states and two away from Harris-won ones. That overstates it. Michigan and Wisconsin are the closest states in presidential politics: Biden carried both in 2020 and Trump both in 2024, each time by less than three points, and both have Democratic governors. Only Tennessee’s new seat lands in a reliably Republican state.

Who draws the lines after 2030 cuts the same way. Republicans would draw one more district in Tennessee and one fewer in Texas, which nets to zero; Democrats in Illinois would draw one fewer. California’s and Michigan’s districts are drawn by independent citizens’ commissions — California’s temporary legislative map runs only through 2030 — and Wisconsin’s needs the governor’s signature as well as the legislature’s. Counting only districts one party draws alone, Democrats lose one and Republicans come out even. For Republicans, that is a thin return on a change this disruptive.

Inside the states

Apportionment decides how many seats a state gets. Districting decides where the lines fall, and there the rule bites differently — and more widely, because it reaches every state with more than one district, not just the six whose seat counts change. Districts must hold equal population, so a district that loses part of its count has to reach further out to get back to the line, absorbing neighboring territory and, with it, neighboring voters.

Running the districts actually in force for the 2026 election — the ten states that redrew mid-decade included² — against a citizens-and-green-card-holders base:

Lean is the two-party margin in the 2024 presidential race, except * New Jersey, where 2020 is the most recent result published at block level.³

Across all 429 districts in multi-district states the tilt is consistent: Democratic- leaning districts sit 0.7% below their state’s new ideal on average, Republican-leaning ones 0.6% above. Of the fifty most underpopulated districts, thirty-eight lean Democratic. Map-drawers in most states would be moving people out of Republican districts and into Democratic ones — the raw material for a favorable redraw.

Not all of the hardest-hit districts lean Democratic. Three of the ten voted for Trump: Texas’s 28th and 15th, along the border, and Florida’s 27th in Miami. They are heavily Hispanic districts that moved toward Trump in 2024, and they are part of why the partisan effect below is smaller than the raw pattern suggests.

Wherever the lines are drawn

Those are today’s districts. The maps drawn after 2030 will have different lines, so it is worth asking the question without any map at all: within each state, how much of the population lives in places that vote Democratic, counting everyone, and how much counting only the people the rule keeps? Using each census tract’s estimated exclusion (mapped under “Where the people who keep counting live”) and its most recent presidential result:³

The shift is the change in share times the state’s seats: the number of districts’ worth of population that moves out of Harris-won tracts. * New Jersey is measured on 2020, so its tracts are Biden-won.

Every one of the 42 states with more than one district moves the same way. Added up, about 2.3 districts’ worth of population leaves Democratic-voting neighborhoods nationwide. Many of the hardest-hit neighborhoods are only narrowly Democratic, though, so weighting each person by how their tract actually voted cuts the net to about 0.9 of a district.

This is suggestive, not a forecast. It says which way equal-population lines would have to move, not who will draw them or how: a map-drawer can concentrate the change in one district or spread it thin, and in a close seat a few points of new territory can decide the outcome. But the direction holds in every state, and it falls hardest in the same places the apportionment does — Texas and California above all.

State by state

Six states change hands, and they divide into two kinds. Three are cut deeply enough to lose a seat. Three are cut so lightly that they gain one without anything happening to them at all.

Texas gains three seats instead of four

Texas has the largest projected gain in the country — four seats, on the strength of a decade of growth. The rule removes 5.9% of its projected population, a bigger share than any other state, and takes one of those four seats back. Much of what Texas grew by is what the rule stops counting.

California loses a fifth seat

California is already projected to fall from 52 seats to 48 on population trend alone. The rule removes 4.7% of its count and takes a fifth. It sits about 228,000 people short of holding that seat — the largest single loss on the board, and the one least sensitive to how the rule is read: California loses at least one seat under every version of it.

Illinois loses a second seat

Illinois drops from 17 to 16 on trend, and the rule takes it to 15. It misses keeping that sixteenth seat by about 48,000 people, one of the closest calls in the country: a modestly smaller estimate of its excluded population would be enough to save it.

Michigan keeps the seat it was going to lose

Michigan is projected to fall from 13 seats to 12. Under the rule it stays at 13 — not because anything changed in Michigan, but because 1.3% of its population is excluded against 4.7% in California and 5.9% in Texas. Nothing had to happen here for the delegation to survive.

Wisconsin also keeps one it was losing

The same story as Michigan, and one of the narrowest margins in the set. Wisconsin goes from 8 seats to 7 on trend and back to 8 under the rule, holding that eighth seat by roughly 51,000 people. It is among the first seats that would move if the excluded population were estimated slightly differently.

Tennessee gains outright

Tennessee holds at 9 seats under current projections and reaches 10 under the rule. It is the only state in the set that gains a seat it was not otherwise in line for. Its 10th is held by about 61,000 people.

Where the people who keep counting live

Half the non-citizen population goes on being counted, so where that half lives matters as much as where non-citizens live — and the two maps are not the same.

DHS publishes green-card counts by county for the fifty largest metro areas, but only for people who became green-card holders in the past ten years.⁴ That is not a count of everyone who holds one, so it is used here only to divide each state’s total among its metro counties. Set against the ACS non-citizen count for the same counties,⁵ the share still counted runs from 20% to 82%:

The widest gaps are inside states, not between them. California’s metros all draw on the same statewide total, yet Sacramento keeps 82% of its non-citizens in the count and Bakersfield 33%; Los Angeles, at 52%, would stop counting about 900,000 people. In Texas, Houston and Dallas keep about two in five and McAllen one in five. A count of non-citizens alone would treat these places alike. Under this rule they are not.

Two caveats matter for reading these, and they pull in opposite directions. DHS records a green-card holder at the place they obtained that status, not where they live now, so long-settled gateways — New York above all — carry people who left long ago, which inflates the share still counted. And because the county counts cover only the last ten years of admissions, those same gateways, whose green-card holders mostly arrived earlier, get less than their share when a state’s total is divided, which deflates it. The size of neither error is known.

Mapped, the difference is stark. This is not a shaded-county map: each tract contributes its own value⁶ and the space between is a distance-weighted blend, contoured like a weather chart. It shows the share of everyone — not of non-citizens — who would stop being counted, because that is the quantity that moves representation.

The country is mostly pale. The rule reaches 10% or more of the population in a handful of places: the Rio Grande Valley, Dallas and Houston, the Central Valley, northern New Jersey across the Hudson from Manhattan, and the meatpacking towns of the High Plains around the Oklahoma Panhandle. Los Angeles, Miami, Orlando and Atlanta sit in the band just below. Half of all tracts sit below 1.5%.

Two things about how to read it. The surface stops where its subject does — where almost nobody lives, nothing is drawn, which is why the interior is full of holes.

The hatching marks where the map is less certain. Every tract’s figure comes from the same calculation: its non-citizens, minus the share of them estimated to hold green cards. What differs is where that green-card share comes from. DHS counts recent green-card holders county by county only in the fifty largest metro areas, and there each county gets its part of the state’s total in proportion to that count, as described above. Everywhere else, the hatched areas, there is no county count, so each county is assumed to have the same green-card share as its state as a whole, and only the split among its tracts is estimated.

That assumption is not harmless. In the counties where both figures exist, using the state share instead of the county count changes a tract’s value by a little under one percentage point on average — enough to move 18% of tracts into a different color band on this scale. About a quarter of the non-citizen population lives in hatched areas.

How solid is this

Reasonable people will estimate the excluded population differently, so it is worth asking how much the answer depends on that choice. Bounding it from both sides:

The first two rows come from genuinely independent sources — one from DHS estimates of the lawful permanent resident population, the other from Pew’s estimates of the unauthorized population — and they land within 1.4 million of each other and give the same answer. The third row is the citizen-only count that is not what was proposed; it is here as the ceiling, and note that even at that ceiling only six seats move.

Across all three scenarios the same states move the same way: Illinois and Texas lose, Michigan, Tennessee, and Wisconsin gain. That much is solid.

How solid is a harder question than it looks, because the last few seats are always decided by very little. A seat is worth about 784,000 people in 2030. Under the proposed rule, Wisconsin holds its eighth seat by roughly 51,000 of them and Tennessee its tenth by 61,000 — under a tenth of a seat either way. Illinois, one of the three states that loses, misses keeping its sixteenth by about 48,000.

And the closest call on the whole map belongs to a state that does not appear in the table at all: South Carolina finishes some 22,000 people short of an eighth seat, less than half of one percent of its population. Any of these could land the other way on a different set of assumptions, or on the actual count.

South Carolina’s is not a hypothetical. The largest assumption behind the rule’s column is how the non-citizen population is carried forward to 2030. The ACS count rose from 21.2 million in 2021 to 24.3 million in 2024, and the trend used here treats most of that rise as temporary, projecting 23.4 million for 2030. Weight the last few years heavily enough that the rise keeps going, and a fourth seat moves: Florida loses one and South Carolina gains its eighth. The other six states move exactly as above. Which is closer to right depends on whether the 2022–24 increase lasts, and no trend line can say. The green-card share matters less: holding each state’s share at its 2024 level, rather than projecting it, moves the same three seats.

What this does not capture

Two other parts of the proposal are harder to model and could matter as much as the population base. Proposed section 60.2 would require that a person’s residence be where they “lawfully spent the greatest number of days,” evidenced by tax records, and would compress the enumeration period to January 3 through April 1. Both narrow the count further in ways no population estimate here accounts for.

Nor does any of this address the constitutional question. The Fourteenth Amendment directs that representatives be apportioned “counting the whole number of persons in each State.” In Department of Commerce v. New York (2019) the Court rejected the last administration’s stated rationale for a citizenship question as “contrived,” and in Evenwel v. Abbott (2016) it upheld total population as a districting base without requiring it. Whether “persons” can be read to exclude lawfully present people on visas is a question for the courts, and it will get there.

Method

Apportionment is calculated by the method of equal proportions (Huntington–Hill), 435 seats, one to each state first — the method in use since 1941.

State population comes from Census Bureau vintage 2025 estimates, projected to 2030.⁷ Who makes up that population is projected separately. Each state’s total, citizen, and non-citizen counts come from the American Community Survey’s one-year estimates for 2005 through 2024,⁸ and each series is carried forward to 2030 on its own trend, so the non-citizen population in these figures is a 2030 projection, not a 2024 count. Lawful permanent residents come from the Department of Homeland Security’s annual estimates for January 1 of each year from 2015 through 2024.⁹ Each year’s figure is set against that year’s ACS non-citizen count, and each state’s green-card share of its non-citizens is carried forward to 2030 on its own trend, then applied to the projected non-citizen population. Nationally that share rose from 58% in 2015 to 62% in 2019–21 and has fallen since, to 52% in 2024, as the non-citizen population grew faster than the green-card population; the projection puts it at 51% in 2030. DHS names only the twenty or so largest states each year, so the rest carry a common share of its pooled total. The non-citizens left over are the people the rule stops counting, and their share of each state’s projected ACS population is applied to the Bureau’s 2030 total. Unauthorized estimates from the Pew Research Center¹⁰ are used only to bound the result. Each source is used for what it measures well: the Bureau’s estimates for how many people, the survey for who they are.

Every projection here is made the same way. It is not one year’s change carried forward, or an average of recent changes: a straight line is fitted through every year of the series by weighted least squares and extended to 2030, with each year weighted by e^{−0.15 × (years before the latest)}. For the ACS series that means 2024 counts in full, 2023 at 86%, 2021 at 64%, 2019 at 47%, 2010 at 12%, and 2005 at 6%; the four years since the pandemic carry about half the total weight, and 2005 through 2014 about a fifth. There is no 2020 one-year ACS file, so that year is simply absent. The Bureau’s state totals are fitted the same way over 2020 through 2025, and the green-card share over 2015 through 2024.

Recent years get more weight because the 2020–21 population base is distorted by the pandemic and tells us less about 2030 than 2023–24 does. That choice matters, and it is why the middle column here moves twelve seats where the 2020-to-2030 map on this site moves thirteen: that map is fitted without the weighting. For that column the rate of down-weighting changes nothing at all — every decay constant from 0.15 to 1.0 moves twelve seats.

The rule’s column is less settled, because the non-citizen series is not a straight line. Fitted across all nineteen years of data at the decay used here, 0.15, the jump of the last three years mostly washes out, and the projection for 2030 comes in below the 2024 count. Any decay constant up to 0.3 moves the same three seats. From 0.34 up, the recent years dominate: the non-citizen projection climbs past 24.5 million, the green-card share keeps falling, and a fourth seat moves, as set out under “How solid is this.”

The district figures are present-day, not projected: 2020 census population by block with citizenship from the ACS five-year estimates,³ joined to the block assignment file for the plan each state will actually use in 2026.² Blocks nest inside districts exactly, so nothing is interpolated — every state’s districts sum to its census population to the person. Two assumptions remain. Block data carries citizenship only for the voting-age population, so each district’s non-citizen count is scaled to all ages by its state’s ratio from the ACS. And the share of those non-citizens who hold green cards comes from the state estimate, divided among a state’s big-metro counties in proportion to DHS’s county counts of recent green-card holders where those exist,⁴ and weighted by how much of the district sits in each county.

A projection is not a census. Small differences in state population can decide the last few seats, and the 2020 census itself missed people in fourteen states. The direction and rough size of these shifts are more trustworthy than any single state’s seat count.

Code and data: github.com/jcervas.