The wage floor for American farmwork fell by a third. Two different things could have caused that, one defensible and one not, and no one has separated them.
In late summer 2025 the USDA discontinued the Farm Labor Survey, the instrument that measured what American farmworkers earn and set the wage floor for the H-2A guest-worker program. On October 2 the Labor Department rebuilt that floor on a different dataset, the Bureau of Labor Statistics’ Occupational Employment and Wage Statistics survey, and made two additional changes: it added a lower entry-level wage tier, and it began deducting the value of employer-provided housing from the hourly rate.
The floor fell. Agricultural economists writing in Choices calculated the drop from Farm Labor Survey data and Labor Department figures: entry-level rates going from $15–20 an hour in 2025 to $8–14 in 2026. The Economic Policy Institute put the annual loss at $1.8 billion, or 26.8 percent, if state minimum wages are fully enforced. The Labor Department’s own estimate is a transfer of roughly $1.7 billion in the first year and $24 billion over ten. Employer-side and worker-side analysts disagree about almost everything else and do not disagree about this.
So the wage cut is not in dispute, and the essay you might expect — did wages fall? — has no work left to do.
Here is the one that does. Three separate changes happened at once, and only one of them was forced. Losing the Farm Labor Survey compelled the Labor Department to find another dataset. It did not compel a second, lower wage tier pegged near the bottom of the occupation. It did not compel subtracting housing a worker cannot decline, cannot shop for, and cannot leave. Those were choices made alongside the substitution, in the same rule, and they hit the same number.
Which means the drop has at least two possible parents. It could be the sampling frame — the new survey counting a different and cheaper population than the old one did. It could be the two discretionary parameters, which would have cut the floor no matter which dataset they were applied to. Almost certainly it is some of each.
Nobody has published the split. Not the agency, not the litigants, not the economists who ran the gross comparison. And the split is the whole argument, because the two causes have opposite implications for whether this rule is lawful.
Why the split decides it
The Immigration and Nationality Act permits H-2A workers only on terms that do not adversely affect the wages of similarly employed Americans. Every dispute about this rule is a dispute about that clause.
If the drop is mostly the parameters, the Labor Department has a real defense. It can argue the old rates were inflated — that the Farm Labor Survey included overtime and production-based pay that the new base-hourly measure properly strips out, that regional aggregation had distorted rates in states with agricultural overtime laws, that a skill tier merely brings farm methodology into line with how every other occupation is measured. Reasonable people run wage surveys this way. The rule would be a defensible policy choice that someone disagrees with, which is not the same as a violation.
If the drop is mostly the sampling frame, the defense collapses, because the frame is not a policy choice about how to weigh wages. It is a question of whether the number describes farmwork at all.
And here the record is worse than the agency’s framing suggests. OEWS does not survey farm establishments — BLS’s own documentation says it excludes the majority of the agricultural sector, and it publishes no wage estimates for crop production or animal production. What it captures within agriculture is support activities: farm labor contractors, the intermediaries who supply crews by the day, and only those filing unemployment insurance. The Center for Immigration Studies — an organization that argues for less immigration, not more, and has no stake in defending farmworker wages — ran the occupational composition and found that two-thirds of the workers whose wages set the 2026 farm rates fall in the hand packers and packagers classification, most of them employed in nonfarm warehouses such as Amazon rather than packing apples. The Economic Policy Institute, which agrees with that organization about nothing else, describes the same defect from the other side: OEWS surveys nonfarm subcontractors who pay farmworkers substantially less on average.
That is the decomposition question in one sentence. How much of an American farmworker’s wage floor is now set by warehouse labor, and how much by a percentile choice? Both would be lawful to ask about. Only one is plausibly lawful to do.
What is known, and what the agency has conceded
The Labor Department has not disputed that OEWS misses farms. It has scheduled a fix, and the schedule is the concession: OEWS will begin surveying farm employers with the May 2026 collection, with the first data appearing in the May 2027 release. Because OEWS pools three years of data, adequate farm coverage arrives no earlier than 2029.
Read that plainly. The agency’s own remediation timeline concedes that the instrument setting farm wages will not adequately see farms for roughly four seasons. It does not concede that the gap biases the number downward — that is precisely the undecomposed question — but it concedes the gap.
There is a second, quieter backstop that turns out to matter. Because H-2A employers must pay the highest of the federal rate, the state minimum, or a prevailing wage, the state minimum catches the fall wherever it is high enough. The American Farm Bureau Federation’s own analysis found that as of January 1, 2026, domestic workers would receive the state minimum rather than the federal rate in eight states — California, Connecticut, Florida, Hawaii, Maine, Maryland, Rhode Island, and Washington — plus urban Oregon and New York. Farm Credit East put New York numbers on it: $15.68 for domestic workers and $13.28 for H-2A workers, both under the $16 upstate minimum.
This is why EPI’s loss estimate carries the qualifier if state minimum wages are fully enforced. In roughly ten states the federal floor has stopped functioning as a floor at all — it now sits under the state’s. Everywhere else, the full cut lands.
That geography is also a natural experiment nobody has run. If the wage effect works through the collapsed federal floor, outcomes should break sharply at the borders between states where the minimum binds and states where it does not. A general labor-market shock would produce no such break. Kill condition on my own reading: if H-2A offered wages in the 2027 disclosure data show no discontinuity across those borders, the federal-floor mechanism is not doing the work I attribute to it.
What still cannot be seen at all
The decomposition is unpublished but computable — the inputs exist and someone could run it this year. Two other things about this program are not merely unpublished. They have never been produced.
Whether the visa’s tie binds. An H-2A worker’s lawful presence is attached to one certified employer; quitting means losing status rather than changing jobs. Whether that constraint holds in practice is answerable with a single rate — how often workers leave certified employment before contract end. The Labor Department publishes selected H-2A statistics: applications, certifications, positions. It does not publish this. Neither does Homeland Security. Both hold the inputs. Two research passes could not locate the series, which is a fact about the searches as much as about the world — but the absence of any such number in either agency’s published statistics is itself checkable, and neither publishes one.
This matters because the grower’s defense of the tie and the worker’s account of it both depend on it. He says the tie is the only bond that lets him recover recruitment costs he sinks before the first berry is picked, and he will point out that H-1B engineers have portability and employers still pay to fly them in, because those wages cover the flight and his do not. That is an honest description of his constraint. It is also the point at which his account and hers become the same unanswered question: if the tie is what prevents workers from bidding their wages up, the tie is producing the low wage rather than merely coping with it. If wages are low because produce margins are genuinely thin against import competition and buyer concentration, the tie is collateral rather than cause. I think the first. Kill condition: if offered wages track crop margins and buyer concentration, and comparable domestic-only operations show the same compression, I am wrong.
Whether any of this was coordinated. The rulemaking dockets and inter-agency correspondence would answer whether the survey cancellation and the wage rewrite were connected. That record exists and has not been examined — not by the litigants, not by any published analysis, and not by me. I decline to assert coordination on no evidence, and I decline to assert its absence for the same reason.
Where the fight actually is
The rule is in litigation. The United Farm Workers sued the Labor Department in the Eastern District of California, challenging the interim final rule under the Administrative Procedure Act — arbitrary and capricious, contrary to the INA’s adverse-effect mandate, and issued without notice and comment. The complaint names the sampling frame among its grounds. Reporting indicates the court denied a preliminary injunction in May 2026, leaving the rule in effect while the case proceeds. (Case posture is flagged for verification below; the denial is load-bearing and I have not read the order.)
An arbitrary-and-capricious challenge asks whether the agency considered an important aspect of the problem. The decomposition is that aspect, stated as a legal question: did the Department establish that the wage drop it projected came from methodological correction rather than from measuring the wrong population? If the administrative record contains that analysis, the rule is on solid ground. If it does not, the agency changed the instrument and the parameters simultaneously, projected a $24 billion transfer, and did not separate the causes.
That is a question the court will reach and the public record has not.
What to do
Run the decomposition. Apply the new rule’s parameters — the entry tier, the housing adjustment — to the old survey’s wage data, and the old parameters to the new survey’s. Four numbers, one table. It separates the defensible cut from the indefensible one and it can be done from published data. GAO can run it; so can any agricultural economics department; so can the plaintiffs. This is the single most useful unwritten page in this dispute.
Fund the farm sample now rather than waiting until 2029. The last request to add farm and ranch collection to OEWS was $1,137,000 and went unfunded. Against a rule the Department projects will move $24 billion, the instrument costs a rounding error. The agency should want this: an accurate measure is what vindicates a clean modernization. So should growers: if there is no adverse effect, an honest survey proves it and the litigation ends.
Publish the separation rate. No new authority, no new collection, one decision. It would settle whether the tie binds, which is the load-bearing fact under both the grower’s defense and the worker’s complaint.
Request the record. A committee letter or a FOIA would answer the coordination question this essay declines to answer.
And if the decomposition shows the frame is doing the work: let workers move between certified employers in the sector without losing status, with the receiving farm repaying a pro-rata share of documented recruitment costs. That answers the grower’s actual objection in his own terms. I state it conditionally and as a value I hold rather than an entailment — it follows only if the decomposition lands where I expect, and it asks you to accept that a person’s legal presence should not be held by the person setting their pay.
A wage floor for American farmwork fell by roughly a third in a single rule. Everyone agrees it fell. The Department says the drop is a better measurement. Its critics say it is the wrong population. Both could be partly right, and the ratio between them decides whether this was housekeeping or a violation of a statute that is still on the books.
The number that settles it does not require a new survey, a new law, or anyone’s cooperation. It requires one table that nobody has made.
cafebedouin.org 