That reported multi-million decline in US immigrants just doesn’t add up

The recent Bureau of Labor Statistics (BLS) data showing that there were 2.2 million fewer immigrants in the US in July 2025 than in January is almost certainly wrong.

Yes, that huge drop in the foreign-born population comes from a respected government survey, the Current Population Survey (CPS), or the “household survey,” conducted monthly by the US Census Bureau for the BLS. And the decline was reported by multiple think tanks and media outlets.[1] But it is implausibly large, uncorroborated, and implies absurd gyrations in the labor market.

Two think tanks pointed to anecdotal news reports as evidence of slower immigration or increasing deportations, not solid corroborating evidence from other data sources. The Census Bureau warns against using the CPS to measure levels or changes of immigration. And another think tank released research by a team of highly credible researchers who projected net immigration for all of 2025 would be between a gain of 115,000 people and a loss of 525,000—far less than the CPS estimated decline of 2.2 million over six months.

To be sure, the rate of immigration in 2025 has dropped sharply from a net annual increase of almost 3 million in mid-2024, leading to slower population growth and slower employment growth in immigrant-reliant industries. But the drop is not as steep as reported in the CPS. Some or most of the decline in the reported foreign-born population is probably because immigrants are increasingly wary of responding to government surveys, particularly as the administration is trying to use IRS taxpayer data and insurance claims data for immigration enforcement.

The reported decline in the foreign-born population in the CPS—if true—would lead to absurd implications about the labor market. Had the foreign-born population really fallen by 2.2 million people since January, then nonfarm payroll employment would need to be revised downward by millions—or the true unemployment rate would now be at a 72-year low of 2.6 percent and the economy would be dangerously overheating. The numbers just don’t add up.

If that many immigrants are leaving, then the adult US population is shrinking

A key implication of that reported decline in the foreign-born population is that if it were true, then the overall US population would also be falling sharply.

Never mind that the CPS reported that the native-born population leapt over the same period: That native-born jump is wrong, as I explain here. Any reported change in the foreign-born population will be mechanically offset by an opposite reported change in the native-born population, in order to sum to the predetermined CPS "population controls" that the Census Bureau calculated in late 2024.[2] According to the CPS, the foreign-born population declined 2.2 million between January and July; the adult (16+) foreign-born population fell 1.9 million. The predetermined population controls baked in a total population increase of 1.1 million adults from January to July. Accordingly, by construction, the adult native-born population grew 3 million in the CPS. But that’s impossible. Native-born population trends depend on fertility and mortality rates, not on immigration swings. Outside of extreme events like a pandemic, native-born trends are stable and predictable.

Here’s a better calculation of how the total adult population would have changed if the foreign-born adult population really fell by 1.9 million. Census population estimates consist of net immigration plus “natural increase,” which means births minus deaths. I estimate the change in the adult population that’s not due to immigration similarly, by looking at the number of people entering adulthood in 2025 minus the number of people exiting adulthood by dying. During 2025, 4.3 million people who were 15 years old in December 2024 will turn 16 and therefore enter adulthood; 2.9 million adults are projected to die based on mortality rates.[3] That is a net 1.4 million increase for the year, half of which I assume to occur in the six months from January to July. Thus, I estimate that the “natural increase” in the adult population for these six months is almost 700,000 people.

Combining the 700,000 natural increase number in the adult population with the CPS-reported decline in the adult foreign-born population of 1.9 million implies that the total US adult population declined by 1.2 million people from January to July.

That’s an extreme drop! The Census Bureau projects the adult population to continue growing until 2085 and then start declining by only 100,000 or 200,000 people per year. A six-month 1.2 million drop in the adult population is far outside of anything in modern US history. Even in the deadliest months of the COVID-19 pandemic, in late 2020 and early 2021, the US adult population continued to grow.

The implausibility of the CPS estimate is also apparent when compared with recent labor market data.

Absurd implication #1: Nonfarm payroll employment would have to be revised downward—by millions

Labor market measures like the US unemployment rate, the employment-population ratio, and labor force participation rate are all based on responses to the CPS survey. While the number of people who are employed, unemployed, and out of the labor force must sum to the predetermined population controls, the ratios and rates are largely unaffected by population changes. So, let’s assume the rates and ratios from the household survey are correct, and estimate what the population decline implies for nonfarm payrolls.

The employment-population ratio for all adults fell from 60.1 percent in January to 59.6 percent in July, according to the CPS. Some of this decline was due to the population aging into retirement; the rest reflects a weakening labor market. Combining that decline in the employment-population ratio with the 1.2 million drop in the adult population implies a drop of 2.0 million nonfarm payroll jobs—more than 300,000 jobs lost per month![4]

The establishment survey—a separate monthly survey of individual worksites used to estimate the number of employees on US payrolls—reported that employers added 486,000 jobs between January and July. If there are truly 1.9 million fewer adult immigrants in the US this year and therefore a 2.0 million drop in nonfarm payroll jobs, then the establishment survey has overstated job growth over the past six months by 2.5 million jobs. That’s ten times the revision to the July jobs report that led to President Donald Trump firing the BLS commissioner and three times the preliminary benchmark revision last year that caused Trump to accuse the Biden administration of manipulating jobs data.

In reality, nonfarm payroll employment is probably overstated and likely to be revised downward, but not by anything near the magnitudes that these immigration numbers imply. Guy Berger at the Burning Glass Institute estimates that the preliminary benchmark revision coming on September 9 could reduce the level of nonfarm payrolls by almost 800,000 jobs, but much bigger revisions would be needed if the immigration decline were truly so large and therefore recent payroll numbers overstated job growth by millions.

Absurd implication #2: True unemployment is at historical lows and the economy is overheating

Suppose instead that the establishment survey estimates of nonfarm payrolls are correct, but the rates and ratios from the household survey are suspect.

Nonfarm payrolls rose 486,000 between January and July. Combining that increase in employed adults with a 1.2 million drop in the adult population would boost the employment-population ratio to 60.6 percent.[5] That’s a full percentage point higher than the published figure for July of 59.6 percent and a half-point increase from the published figure for January of 60.1 percent. It would be extraordinary for the employment-population ratio to rise half a point in six months, which typically happens only during a post-recession rebound. That would imply that the labor market has tightened dramatically in the past six months and could be at risk of overheating—enough that the Federal Reserve would be considering raising interest rates, not lowering them.

Suppose further that the labor force fell proportionately with the 1.2 million drop in the adult population, in line with the labor force participation rate as reported in the CPS. If the establishment survey is correct, and employment increased while the labor force shrank, then the implied unemployment rate in July would be 2.6 percent. That would be the lowest level since 1953, when the US economy was booming from post-World War II consumption and Korean War defense spending.

This is absurd, of course. There’s no evidence that the labor market is rapidly tightening. Private sector wages are growing more slowly than a year ago, and anecdotally the balance of power in the labor market is shifting from workers to employers.

Accurate immigration numbers might take a while

The CPS’s reported decline in immigration is almost certainly overstated. As noted above, there’s little corroborating evidence yet from other data sources on immigration or deportation trends, and such a huge decline in immigration implies nonsensical labor market shifts. Unfortunately, it might take years before current immigration trends become clear in official data.[6]

In the meantime, immigration and jobs data need to be interpreted carefully, with skepticism especially for numbers that seem to tell a dramatic story. Beware partisans and policy advocates who latch on to data that prove their point. All data are imperfect. Immigration is notoriously hard to estimate; payroll employment changes get revised, sometimes significantly; and the unemployment rate is based on a small household survey with declining response rates, plausibly declining more among foreign-born residents wary of current immigration enforcement methods. Numbers from different sources and methods will never line up perfectly. But they do tend to move together coherently enough over time so that a shock to one measure should reverberate in others. When the numbers don’t add up, dig deeper—or sound the alarm.

Notes

1. The Center for Immigration Studies reported a 2.2 million drop from January to July. Pew Research reported a 1.5 million drop from January to June, as covered in the New York Times and elsewhere. The Wall Street Journal was appropriately more skeptical.

2. Census population estimates serve as the “population controls” for the CPS. That means that the total population reported each month in 2025 in the CPS equals the monthly population projections that Census calculated in late 2024. The population totals in the CPS are not affected by survey responses. These population controls predetermine not just the total population but also the population by age, sex, and race and ethnicity. In contrast, the native-born and foreign-born populations are not predetermined by Census estimates; they are not part of the population controls. Instead, nativity (that is, whether someone was born in the US or elsewhere) is based on survey responses. But the native-born and foreign-born populations must add up to those predetermined population controls for the total population and for demographic groups.

3. Population by age comes from the Census Bureau. Estimated mortality rates by age come from the Congressional Budget Office. These and other calculations are shown in the accompanying spreadsheets.

4 The household survey estimates employed people, while the establishment survey reports nonfarm payroll jobs. Employed people can have multiple nonfarm payroll jobs. Jobs in agriculture or private households, unpaid jobs, and self-employment are not counted as nonfarm payroll jobs. CPS data shows there are roughly 96 nonfarm payroll jobs for every 100 employed people; I use this adjustment to convert a decline of 2.1 million employed people to a drop in 2.0 million nonfarm payroll jobs.

5. That corresponds to an increase of 505,000 employed adults, using the ratio described above of 96 nonfarm payroll jobs for every 100 employed people.

6. Census will release its next immigration and population estimates at the end of 2025, but these will be based in part on lagged survey data from 2024 and therefore won’t show a complete picture of 2025 immigration trends. Survey data from 2025 will inform Census immigration and population estimates produced in late 2026. Furthermore, the final benchmark revision to nonfarm payrolls is due in February 2026, but that revision will fully align payroll data to complete administrative records only up to March 2025. Estimates from this summer’s jobs reports will be adjusted again as part of the following benchmark revision, due in February 2027.