The FMLA defines who it does not reach — for a worker failing the tenure, hours or worksite-size test the unpaid entitlement simply does not exist, so no denial is ever recorded

What is happening?

The Family and Medical Leave Act gives 12 weeks of unpaid, job-protected leave, but only to employees who clear three tests at the same time — 12 months with the employer, 1,250 hours of work in the preceding year, and a worksite where the employer has at least 50 employees within 75 miles. A Department of Labor survey fielded in 2018 and reported in July 2020 found that 56 percent of employees cleared all three. The other 44 percent were outside the law entirely, and that is still the figure carried by analyses current as of January and February 2026.

The exclusion is not mostly about employer size. The 44 percent breaks down further, as shares of all employees — 15 percent fail only the worksite-size test, 21 percent fail only the hours or tenure test, and 7 percent fail both.

Nothing paid sits on top of this. Federal law does not require any private employer to provide paid leave of any kind, and as of March 2023 only 27 percent of private-industry employees had access to paid family leave through an employer. Fourteen states and the District of Columbia have built mandatory paid family leave insurance of their own as of April 2026 — but a state benefit pays wages; it does not extend the federal right to get the job back.

Whose problem is this?

The party that sets the thresholds never meets the person a threshold excludes, and the party the thresholds were written to shield — the small employer — is the only one of the four that is counted, organized and named in the statute.

Where does this problem end?

The boundary matters because two different gaps get reported as one thing. The federal entitlement is unpaid, and a large share of workers cannot reach even that.

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

Now

Needs a new measurementthe target state. No federal target exists for the share of the workforce that should be covered. Congress set the thresholds in 1993 and the sources opened here state the current share without stating any goal against which it is short. What the advocacy publications ask for is paid leave, which is a different instrument, so even they do not name a coverage number the unpaid entitlement should reach.

How big is it?

The affected population is the employees who fail at least one test. Applying the 44 percent share to the June 2026 workforce gives 67.2 million to 71.4 million people.

The interval is not a confidence interval. It is the distance between two denominators: 152,801,000 wage and salary workers on the low side and 162,264,000 employed civilians on the high side, both seasonally adjusted for June 2026. The lower bound is the better founded of the two. Self-employed people were never inside the statute, so counting them raises the number without adding anyone the law could have covered.

Under what conditions does it arise?

1. The thresholds are statutory, not regulatory. No agency can widen them, so there is no rulemaking to petition and no interpretation to litigate. 2. Nothing is refused. A worker below the thresholds is not told no — the entitlement simply does not exist for that person, so no denial, no complaint and no case is ever recorded. 3. The exempted side is a constituency and the excluded side is not. Employers with fewer than 50 employees share an interest and a lobby. The excluded 44 percent have no employer, no industry and no place in common, and a person crosses in and out of eligibility by changing jobs or hours. 4. State programs drain the pressure unevenly. Where a state pays benefits the loudest part of the complaint moves away from Congress, and where it does not the complaint stays with a body that has not amended the thresholds since 1993.

What has been tried?

Every attempt in this table works on the money rather than on the thresholds. None of them changes who the federal statute reaches.

What was found?

Why is it still unsolved?

Institutional exemption — the gap is not a failure of the rule, it is the rule. The statute defines who it does not reach, so there is nothing to enforce and no one to appeal to.

An employee below the thresholds has not been denied anything. The law does not apply, which means there is no violation to report, no agency with jurisdiction and no case to bring. Failures shaped this way leave no trace in complaint counts or enforcement statistics. The only instrument that can see them is a survey that asks the workforce directly, and that instrument has been fielded once, in 2018.

The asymmetry of organization does the rest. Small employers are a named class with a shared interest in the threshold staying where it is. The 44 percent are not a class at all — the three tests cut across every industry and every firm size, and a person moves in and out of eligibility by taking a new job or dropping below 1,250 hours. There is no membership to organize and no moment at which the loss is announced.

The measurement problem then feeds back into the politics. A share that is restated rather than remeasured for eight years is hard to argue about, because both sides can say the world has changed since. Meanwhile the visible reform energy has gone into paid leave, which is a different instrument aimed at a different gap, and progress there can be reported as progress on leave generally while the eligibility line stays exactly where 1993 put it.

What observation would mean it is solved?

Candidates — (a) the eligible share rising above 56 percent in a repeated federal survey (b) the number of jurisdictions with mandatory paid family leave insurance rising (c) a fall in the number of workers who report needing leave and not taking it.

(a) can move without any rule changing. If employment shifts toward larger employers or job tenure lengthens, the share rises on its own. It also cannot move at all unless somebody fields the survey again, and no repeat was found in this round.

(b) counts jurisdictions, not people, and it measures the wrong instrument. A state program pays wages; the FMLA holds the job. A worker can draw a state benefit and still be dismissed.

(c) blends the two gaps. Two thirds of the workers who needed leave in 2025 and did not take it said the reason was that they could not afford it, which is an answer about pay, not about eligibility. Read alone it would improve as paid leave spreads even if not one person gained the federal right.

What is it connected to?

Fills with researchthe federal paid-leave gap, state job-protection statutes that use lower size thresholds, the small-business exemption pattern in other federal employment law, and the same threshold logic in health-coverage mandates. Relation type and evidence grade were not established in this round.

What these sources do not say

• Whether the 44 percent has moved since 2018. No federal re-measurement was found. The 2026 publications restate the 2018 figure rather than replacing it, so the age of the number is carried forward silently.
• Why the three published reasons sum to 43 percent against a stated 44 percent. Neither the survey report nor the publications restating it explain the one-point difference in the material opened here.
• How many of the excluded workers live in a state with a mandatory paid leave program. The eligibility share is national and the state programs are listed separately. Nothing opened here joins the two, so the overlap is unknown in both directions.
• What share would gain coverage if the worksite-size test alone were removed. The 15 percent excluded on size alone is not the answer, because the hours and tenure tests would still apply to those same people afterwards.
• A BLS paid family leave access figure with a reference month later than 2023-03. The March 2025 benefits release and its published leave table cover sick leave, vacation and holidays only.
• Whether any employer manages headcount or worksite structure to stay under 50. The sources describe the threshold; none of them tests for behavior clustered around it.

See the evidence

Three of these items were confirmed by opening the document directly. The rest sit behind publishers that refuse automated requests, and those were read through a text extraction proxy that returns the page text unchanged — the canonical address is listed rather than the proxy address. The 2018 survey report itself was read that way, so the 56 percent comes from the report and not only from the publications restating it. One divergence is left standing rather than reconciled. The state counts do not agree: CRS records 10 active programs including the District of Columbia as of 2025-03, the Center for Economic and Policy Research says 13 states plus the District of Columbia in early 2026, and the Bipartisan Policy Center says 14 states and the District of Columbia mandatory as of 2026-04 inside a wider total of 24 that also counts 9 voluntary state laws — so the wider total mixes mandatory and voluntary and must not be read as programs paying benefits. Carrying the CRS count forward does not close it either: adding the four programs CRS lists as scheduled brings its 10 to 14 including the District of Columbia, which is one jurisdiction short of 14 states and the District of Columbia, so the gap is left visible rather than argued away. The Section 45S difference did close and is recorded here as resolved: CRS states the credit expires at the end of 2025 because that document was written in 2025-03, and the Bipartisan Policy Center guide of 2026-08-06 records the One Big Beautiful Bill Act making the credit permanent from 2026, so the earlier statement is superseded by date rather than contradicted. No publication date is printed on the Center for Economic and Policy Research piece at all. The February 5 recorded for it here is taken from its opening sentence, which marks that day as the 33rd anniversary of the statute, and the year is fixed at 2026 by that framing together with its statement that Minnesota and Maine launched programs this year. This is a Path A output, so observation_refs is empty and provenance_mode is press-derived.

People affected

Estimated range 67,232,440–71,396,160 As of 2026-06

Sensitivity The interval is not a confidence interval. It is the spread between two workforce denominators — wage and salary workers only against all employed civilians — applied to a single point share, so 0.44 x 152,801,000 = 67,232,440 and 0.44 x 162,264,000 = 71,396,160. The lower bound is the better founded one, because unincorporated self-employed workers are not covered by the statute under any reading and their presence in the upper denominator inflates it without adding anyone the law could reach. The dominant uncertainty is outside this interval altogether: the 44 percent comes from a survey fielded in 2018 and reported in 2020, and eight years of change in employer size distribution, job tenure and weekly hours are unmeasured here. A second unquantified item is that the three published components of ineligibility — 15 percent on worksite size only, 21 percent on hours or tenure only, 7 percent on both — sum to 43 rather than 44, and no source opened in this round explains the difference.

Regional breakdown The federal survey reports national shares and the sources opened in this round carry no state-level eligibility rates from the same table as the national total. Splitting the national figure across states by population would be proportional allocation, and it would be wrong in an unknown direction because employer size distribution and industry mix vary sharply by state — the share of employment sitting in worksites below 50 people is exactly what differs, and it is the variable being estimated.