No federal document names a share of unemployed workers a US state unemployment program must reach, and Kentucky counted 8.9 percent of its unemployed workers as insured in 2024

What is happening?

In 2024 the United States counted 26.8 percent of its unemployed workers as insured unemployed under the regular state unemployment insurance programs. The same federal series put Kentucky at 8.9 percent and Minnesota at 56.3 percent — a 6.3-fold spread between two states running the same federal-state program, as of the chartbook values retrieved from the Department of Labor on 2026-08-08.

The spread is neither new nor narrowing. The national ratio has stayed inside a band of roughly 25 to 29 percent in every year since 2011, apart from the pandemic year of 2020 when it reached 74.8 percent. In the most recent federal quarter, calendar quarter 1 of 2026, dividing the two columns the Department of Labor publishes by state gives a national figure of 26.9 percent, a low of 4.8 percent in Florida and a high of 52.3 percent in North Dakota — an 11-fold spread, with winter layoffs lifting the top.

None of these numbers breaks a rule. The Department of Labor writes in its own May 2024 account of the federal-state partnership that there are no federal standards for benefits in terms of monetary qualifying requirements, benefit amounts, or duration of regular benefits. Kentucky put a variable benefit duration into statute in 2022 and its recipiency rate fell from 20.5 percent in 2019 to 8.9 percent in 2024. Florida revised its available weeks in 2011 and now shares the shortest statutory band in the country with Arkansas, 9 to 12 weeks as of 2026-01-01.

Whose problem is this?

The body that would have to set a floor is Congress, and the bodies that set the actual thresholds are fifty-two separate legislatures. Nobody in that arrangement is failing at an assigned task.

Where does this problem end?

The boundary matters because the program already exists in every state and money is not the binding constraint. What differs across state lines is who is allowed through the door.

What is the state now, and what should it be?

Now

Needs a new measurementthe target state: no federal document opened here names a share of unemployed workers a state program must reach, nor a maximum acceptable spread between jurisdictions. The only number in this entire material is an advocacy proposal from June 2023 asking for a standard of 50 percent of unemployed workers receiving benefits, paired with a denied-claim accuracy standard below 10 percent. That proposal is absent from the federal core measures as of 2026-08-08, and the federal government has published this statistic annually for half a century without ever stating what a good value would be.

How big is it?

Between 1.96 million and 2.4 million people in an average week of 2024. The chain is short. The Bureau of Labor Statistics puts the 2024 annual average at 6,761,000 unemployed persons. The gap between the highest jurisdiction and the national rate is 56.293 minus 27.335 percentage points on one federal measure and 56.293 minus 26.751 on the other, which multiplied out gives 1,958,000 and 1,997,000. The upper bound of 2.4 million is the figure the National Employment Law Project published on 2025-07-15 for the same counterfactual, computed against a Minnesota value the federal series no longer carries.

A second chain built from a different federal report, in different units, lands nearby. Applying the 2024 Minnesota conversion from initial claim to first payment, 0.871, to the 8,130,922 new intrastate initial claims filed nationally in 2024 yields about 7.08 million first payments against 5,333,058 actually made, or roughly 1.75 million additional benefit spells started. Neither chain is published by anyone.

Read what these numbers are. The recipiency rate is a ratio of average weekly stocks, not a headcount of distinct persons over a year, so the first chain says how many more people would have been on the rolls in a typical week rather than how many would have been paid during the year. The second counts spells, and one worker can start more than one. No federal source publishes a state-level count of distinct people who experienced unemployment in a year, so a true person-level recipiency rate cannot be built by state at all.

Under what conditions does it arise?

1. The federal conditions govern method, never reach. The conformity list a state law must satisfy for its employers to receive the federal tax credit runs to roughly eighteen items. Every one concerns how compensation is administered or which grounds of denial are forbidden — payment through public employment offices, deposit in the federal trust fund, no denial for refusing work vacant because of a labor dispute, no denial solely because of pregnancy, no denial for taking approved training. Not one item sets a floor on who qualifies or how long benefits last.

2. The only numeric federal levers sit on the financing side. Relief from the automatic credit reduction is conditioned on a state not having taken action that would reduce the solvency of its fund or decrease its unemployment tax effort. An expansion of eligibility is a solvency-reducing action with a named federal consequence. A contraction is a solvency-improving action with none. The ratchet turns one way and the direction is written into statute.

3. The federal scorecard measures speed and fraud. Twelve core measures carry acceptable levels of performance — first payment promptness at 87 percent or more, determination time lapse at 80 percent, appeals aged 30 and 40 days, improper payments under 10 percent, overpayment detection between 50 and 95 percent, overpayment recovery at 68 percent. Three separate measures police overpayment. None polices non-payment to eligible workers. First payment promptness governs how fast a payment goes out and is indifferent to how many go out, so a state that admits almost nobody can score near the top of it.

4. Duration is indexed to the state jobless rate. The sliding-scale states tie available weeks to the state average unemployment rate, which shortens benefits exactly when the labor market is tight and does not lengthen them past the statutory cap.

5. Generosity and reach come apart. Kentucky pays a higher share of prior wages than the national average and carries one of the highest weekly caps in the table at USD 720, while admitting the smallest share of its jobless workers in the country. Across the 52 jurisdictions the correlation between the 2024 recipiency rate and the 2024 replacement rate is 0.415 — real, and far from determinative.

What has been tried?

Two of these attempts move the number and both move it downward. The one direction that would move it upward across all states at once has been proposed and never enacted.

What was found?

Why is it still unsolved?

Institutional gap — the rule that would close it has never been written, and every actor in the system is behaving lawfully.

Start with what a state that pays 9 percent of its jobless workers is violating. Nothing. The federal conditions attached to the unemployment tax credit are a list of methods and of forbidden reasons for denial, and a state can satisfy every clause on that list while admitting almost nobody. The Department of Labor puts it in its own words, in a document written by the office that would enforce such a standard if one existed, saying that there are no federal standards for benefits in terms of monetary qualifying requirements, benefit amounts, or duration of regular benefits. A shortened duration band is not an evasion of the rule. There is no rule to evade.

The second movement is that the one place federal law does carry hard numbers, it carries them on the other side of the ledger. Credit-reduction relief turns on whether a state took action that would reduce the solvency of its fund or decrease its tax effort. A legislature that widens eligibility triggers a named federal consequence. A legislature that narrows it triggers none, and improves the only federal financial benchmark that applies to it. The 2024 data is what that incentive looks like from outside: four of the ten lowest-recipiency states clear the solvency floor while only one of the ten highest does, and among the highest, five sit at or near zero. That is a correlation of minus 0.306, not a law of nature, but it runs in the direction the statute would predict.

The third movement is the one that keeps the pattern stable. The federal government scores state agencies on twelve measures with published acceptable levels, three of which police overpayment, and none of which asks how many unemployed workers were reached. The rate itself is measured, published free of charge, and broken out by state every year since before most of the current statutes existed. What is missing is not data and not capacity. It is a number that says what good looks like, and because no such number exists, a state that halves its reach produces no failing score, no conformity finding and no headline. The enforcement layer that does exist is a second and separate weakness — the two available sanctions are total loss of federal administrative funding and a tenfold jump in the per-employee tax, both so large that neither has ever been used — but a perfectly calibrated enforcement regime would still have nothing to enforce against a duration band that breaks no rule.

What observation would mean it is solved?

Candidates — (a) a federal standard naming a recipiency floor takes effect and states are scored against it (b) the spread between the highest and lowest jurisdiction narrows for several consecutive years (c) the national rate rises toward the share of unemployed workers estimated to be eligible.

(a) alone is weaker than it looks. A floor set at a level most states already clear changes nothing, and the shape of the measure matters more than its existence. A standard applied to recipiency among eligible workers rather than among all unemployed workers would reward a state for tightening eligibility, which is the exact behaviour the standard is meant to stop. The proposal that exists names this trap itself. A standard also has to survive the enforcement problem described above, and the only sanctions in the system have never once been imposed.

(b) alone is worse, because the spread narrows when the top falls. Between 2024 and 2025 Minnesota moved from 56.3 percent to 49.9 percent while Kentucky rose from 8.9 to 12.4. That is convergence, and one of those two movements is not good news. The ratio also swings with the business cycle and with the quarter, since winter construction layoffs lifted North Dakota and Montana to the top of the quarter 1 2026 table without any statute changing.

(c) alone counts claims rather than payments. The published rate is built from weeks claimed, and in 2024 the weeks actually compensated were 15.6 percent fewer, which puts the same year at 23.2 percent on a paid basis. A rise in the published figure is fully consistent with no additional dollar reaching anyone. The three would have to be read together, with (b) read against the year it came from and (c) read against the paid-basis series that nobody currently publishes.

What is it connected to?

Fills with researchthe classification of gig and platform work for social insurance purposes, the design of automatic stabilizers in a recession, state trust-fund financing and employer experience rating, and the administrative-burden literature on why eligible people do not claim benefits they qualify for. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• No source states a target. The federal government publishes this rate by state every year and attaches an acceptable level of performance to twelve other things — 87 percent, 80 percent, 75 percent, under 10 percent, 30 days, 40 days, 70 percent — and to this one, nothing. The only numeric recipiency target anywhere in this material is a 2023 advocacy proposal. A statistic published annually for half a century, against which no one has ever said what good looks like.
• The series has no vintage and no revision history. Two publications two years apart, both naming the same federal chartbook as their source, both land 4 to 5 points above where that series now sits for the high-recipiency states while the low-recipiency states barely move. The chartbook page footer reads as updated on 2026-08-07. There is no revision note, no archived edition, no per-observation date, and the download endpoint returns only what is current. The discrepancy is therefore permanently unresolvable from the materials of the publisher.
• The same agency gives two national figures for 2024 and never mentions it. Item 12 says 27.335 percent and the national row of item 13 says 26.751 percent. Neither page defines its aggregation, and the item 13 file ships rates without denominators, so a reader cannot reconstruct either one. One agency, one product, one year, two numbers, and no sentence anywhere acknowledging that they differ.
• The measure is a stock ratio counting weeks claimed, and every source narrates it as people paid. The federal definition is the insured unemployed in regular programs as a percentage of the total unemployed, which is an average-weekly ratio built from continued weeks claimed. Sentences of the form 27 percent of unemployed workers received benefits do two kinds of work the number does not support at once: they convert a stock ratio into a person count, and they convert a claim into a payment. No source opened here flags either conversion.
• Nobody divides claims filed by claims paid. Initial claims and first payments are published side by side, monthly, by state, in the same federal report going back to 1971. The division is never performed. Done here it gives 0.327 for Kentucky against 0.871 for Minnesota in 2024 — a measure of the funnel rather than the stock, and arguably the more legible number for a reader asking whether filing leads to payment.
• The federal running record of state law changes stops in 2019. The state-law page says the comparison is published annually, yet the most recent comparison it links is the edition as of 2023-01-01 and the most recent report on state legislation it links is the 2019 cumulative report. The 2024 and 2025 editions were probed and both return 404. So the federal record of what states changed ends before the 2022 Kentucky statute and the duration changes that followed it. The current-state table exists; the change log does not.
• Nobody publishes who does not file. The denominator is everyone the household survey counts as unemployed and the numerator is everyone certifying for a week of state benefits. The people in between — those who never applied because they expected to fail, or could not navigate the process, or did not know the program existed — are the substance of the gap and appear in no federal series. No federal source publishes an eligibility rate by state, so the split between ineligible under state law and eligible but unpaid cannot be made state by state, which is exactly the split a legislature would need.
• The asymmetry between overpayment and wrongful denial is never named as a choice. Three core measures with numeric targets police money going out in error. Improper denials are measured through the same federal accuracy program and published, and carry no acceptable level at all. No federal document opened here explains, or acknowledges, why one direction of error has three enforceable targets and the other has none. Alongside it, the blank cells go unexplained too — Puerto Rico shows no insured-unemployment figure in the current quarterly summary while its exhaustion rate is printed, and the household survey returns a hole for October 2025 because of a lapse in appropriations.

See the evidence

Every figure in this document comes from a federal statistical product or a state legislative record, and the two computations that carry the most weight were performed here rather than read. The claims-to-payment conversion and the paid-basis recipiency rate are divisions of columns that the Department of Labor publishes side by side and does not divide; the handbook row above is included precisely because those two results depend entirely on reading the column definitions correctly. The recipiency figures themselves were pulled from the chartbook form target rather than the landing page, which is the retraceable citation given here. Where sources disagree the disagreement is left standing rather than resolved. The same agency publishes 26.751 percent and 27.335 percent for the same year and reconciles neither, so both appear above. The July 2025 anchor publication reports Minnesota at nearly 59 percent while the same series retrieved on 2026-08-08 gives 56.293, and because that publication names the chartbook as its source and the chartbook keeps no vintage, the difference is recorded as an unresolvable revision rather than assigned to either party — which is why this document uses 56.3 percent with its retrieval date and attributes the 2.4 million upper bound to its publisher and its publication date. Three sources refused automated retrieval and their rows are kept with empty URLs, and the direct consequence is that no federal bill and no North Carolina statute is named anywhere here. This is a Path A output, so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range 1,958,000–2,400,000 As of 2024

Sensitivity The width of this interval is not sampling error. It is the distance between two national figures the same federal agency publishes for the same year, 26.751 percent and 27.335 percent, plus one downward revision of the Minnesota value that the publisher does not record anywhere, since the chartbook keeps no vintage, no revision history and no archived edition. The width is therefore a measure of how unstable the published inputs are rather than of how uncertain the underlying phenomenon is. What this number is not is a headcount of distinct people. The recipiency rate is a ratio of average weekly stocks, so 1.96 to 2.4 million describes how many more people would have been counted among the insured unemployed in a typical week of 2024, not how many people would have received a payment at some point during the year. A second chain built from a different federal report and in different units lands nearby and is worth stating: applying the 2024 Minnesota conversion from new initial claim to first payment, 0.871, to the 8,130,922 new intrastate initial claims filed nationally in 2024 gives about 7.08 million first payments against 5,333,058 actually made, or roughly 1.75 million additional benefit spells started. That chain counts spells rather than persons, and one worker can start more than one in a year. Neither chain is published by anyone. The range is also too generous in one direction and too narrow in another at the same time. Too generous, because the counterfactual of every state matching Minnesota silently includes independent contractors, gig-platform workers, the self-employed and most new entrants and re-entrants to the labor force, who sit in the unemployment denominator but whom no change to a state duration band would reach. Too narrow, because it starts from either the insured-unemployed stock or from claims actually filed, so everyone who never applied is invisible to both chains, and that population is unmeasured in every federal series and is plausibly the largest single component of the gap. A third limit applies to the numerator itself: the published rate counts weeks claimed, and in 2024 the weeks actually compensated were 15.6 percent fewer, which would drag the national figure from 26.8 to 23.2 percent on a paid basis and widen the gap the range is trying to measure.

Regional breakdown The figure is a counterfactual difference rather than a measured population, and it cannot be split by state from the data this chain rests on. The federal file carrying the 2024 annual recipiency rates by jurisdiction ships rates without denominators, so the number of unemployed persons standing behind each state rate is not recoverable from it. One federal product does publish total unemployment and insured unemployment by state side by side, the quarterly UI Data Summary, and dividing those two columns is how the state figures for calendar quarter 1 of 2026 were obtained. It covers a different reference period from the 2024 annual series this chain uses, and a single winter quarter is seasonally atypical at the top of the distribution, so it cannot be substituted for the annual denominator without changing what the number means. Beyond that, no federal source publishes a state-level count of distinct persons who experienced unemployment in a year. Splitting the national gap by state population would be proportional allocation and would be wrong by construction, because the entire content of this problem is that the rates differ by a factor of six across state lines.