No federal limit on restraint and seclusion in schools — 105,700 students were restrained or secluded in one school year

What is happening?

Approximately 105,700 public school students in the United States were physically restrained, mechanically restrained, or placed in seclusion during school year 2021-22. That is the figure the Office for Civil Rights published in January 2025, drawn from 17,704 public school districts and 98,010 schools, with 99.9 percent of required submitters certifying their data. As of 2026-08-08 it is still the most recent federal count.

There is no federal law that limits the practice. GAO told a House committee on 2009-05-19 that it found no federal laws restricting the use of restraints and seclusion in public or private schools, and that it could not locate any website, federal agency or other entity collecting information on how often the methods were used. The Department of Education published fifteen principles on 2012-05-15, and the footnote on the first page of that document states that it sets out no new requirements and confers no rights on anyone.

A bill that would set a rule has been filed in every Congress since 2009. The House of Representatives passed H.R. 4247 on 2010-03-03. The Senate never took it up. Sixteen years later the current version, H.R. 6617, was introduced on 2025-12-11 with 32 cosponsors and referred to committee, where it sits as of 2026-08-08.

Inside the school year 2021-22 total: about 68,800 students physically restrained, approximately 8,200 mechanically restrained, and almost 28,700 placed in seclusion. Students served under IDEA were 14 percent of K-12 enrollment that year and 76 percent of those physically restrained.

Whose problem is this?

The group carrying the exposure is spread one district at a time, and the group that would carry the compliance duty is organized into permanent national associations.

Where does this problem end?

The boundary matters here because the remedy has already been drafted and has already passed one chamber. What is missing is not a diagnosis and not a text but a floor vote and an effective date.

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

Now

Needs a new measurementthe target state: no source opened here names a number that restraint or seclusion should fall to, or a date by which. H.R. 6617 would set a rule and require annual state reporting of incidents disaggregated by injury and death, and the federal collection sets a count. Neither sets a goal, so nothing on this record can be behind schedule.

How big is it?

Between about 68,800 and 105,700 students in school year 2021-22. The published federal total is approximately 105,700 and it is the exact arithmetic sum of its three components — 68,800 plus 8,200 plus 28,700. The label on that total says physically restrained, mechanically restrained, and/or placed in seclusion, which reads as a count of distinct students, and a student both restrained and secluded should appear once. No endnote or analytical note in the report says whether the components overlap. If the three groups are disjoint the distinct total is 105,700. If the two smaller groups fall entirely inside the largest, the distinct total is 68,800.

That is the width in one direction. In the other direction the whole range is a floor. GAO reported that 70 percent of districts recorded zero incidents in school year 2015-16, that the federal quality check reached fewer than 100 large districts, and that district and state officials did not share a common understanding of the definitions. No source publishes a correction factor, so how far above this range the real figure sits cannot be stated from anything opened here.

The count also measures students rather than events. One state record shows how far apart those units are. More than 20,000 seclusion incidents were logged across more than 100 Illinois districts between school year 2017-18 and early December 2018, against a national figure of 28,700 students secluded in a full school year.

Applying the published IDEA shares to each component gives roughly 74,100 students served under IDEA in school year 2021-22, about 70 percent of that year total drawn from 14 percent of enrollment. That figure is derived here rather than published by the department, and it inherits the same overlap question as the total.

Under what conditions does it arise?

1. The binding federal instruments are all general. IDEA guarantees a least restrictive environment and an individualized education program. Section 504 and Title II of the ADA bar disability discrimination, Title VI bars race discrimination, and constitutional claims reach excessive force through civil rights suits. None of them names restraint or seclusion and none sets a threshold, so the practice is never unlawful in itself. It becomes unlawful only when a particular use is proved discriminatory or excessive, one student at a time and after the fact.

2. The executive branch declined to close the gap by rule. In 2012 the department published fifteen principles and stated in its own footnote that they impose no requirement and confer no right. No source opened here records any later rulemaking that would set a threshold. What remains is the statute route.

3. The statute route has one veto point and it has held since 2010. The House passed the bill on 2010-03-03. The Senate companion arrived on 2010-09-29, near the end of that Congress, and died. Every version since has died in committee without a floor vote in either chamber, and no source opened here states a reason for that.

4. The exposed group is concentrated and the compliance burden is dispersed. Students served under IDEA were 14 percent of enrollment and about 70 percent of the counted exposure in school year 2021-22, and their families act one district at a time. The reporting and training duties would fall on 17,704 districts represented by permanent associations. About 60 organizations endorsed the 2025 bill and it has collected no floor time.

5. The federal act that does exist is a count, and the count cannot see itself fail. Districts report their own numbers against definitions that officials read differently — GAO found disagreement over whether confinement counts as seclusion when an adult stays in the room. In school year 2015-16, 70 percent of districts reported zero.

What has been tried?

Two directions were tried and only one of them moved. The statute never passed, and the text of the statute nevertheless rewrote state law in nineteen additional jurisdictions between 2009 and 2019 by being copied.

What was found?

Why is it still unsolved?

Institutional gap — the rule that would close it was drafted in 2009, passed one chamber in 2010, and has never been scheduled since.

A general instrument can punish a bad outcome after the fact, but it cannot tell anyone what to do beforehand. That is the whole difference between the instruments in force and the instrument in draft. IDEA, Section 504, Title II, Title VI and constitutional claims all reach restraint and seclusion, and none of them names the practice or sets a threshold, so a family seeking a limit has to win a case rather than read a rule. The department had the option of writing that threshold by rule and instead published fifteen principles that disclaim, in their own footnote, any requirement at all.

The second part is that the remaining route runs through a chamber that has never scheduled it. The House passed the bill on 2010-03-03 with a large majority. The Senate companion was introduced nearly seven months later, with almost no time left in that Congress, and died. Every version since has died in committee. Nothing opened here states a reason — no committee report, no cost estimate, no recorded opposition testimony. The bill is described everywhere in the passive voice, and a failure with no named cause leaves no one to answer for it.

The third part is the loop that holds the first two in place. With no rule, the only federal act is a count. The count is self-reported by districts against definitions they read differently, and in school year 2015-16 seventy percent of districts reported zero while the federal quality check reached fewer than 100 large districts. A number assembled that way is one any district can set aside as not describing itself, so it never becomes the pressure that would force a rule. The gap produces the weak count and the weak count protects the gap.

There is a fourth movement, and it is the uncomfortable one. The bill has been governing anyway. States copied the text, and emergency-only restraint limits for all children went from 3 jurisdictions in 2009 to 22 by 2019-07-01 without a single enactment in Washington. That is a real gain and it is distributed exactly backwards. In the first half of 2019 protective bills died in eight states with weak or no existing law while two states that already had statutes strengthened them. Which rules protect a student depends on the state line, and the states with no rules are the states where new rules keep failing.

What observation would mean it is solved?

Candidates — (a) a federal statute or rule takes effect that sets a threshold on restraint and bars seclusion (b) the federal count of students restrained or secluded falls across several consecutive collections (c) every jurisdiction carries a binding law with emergency-only restraint limits and mandatory reporting.

(a) alone is weaker than it looks. A statute can arrive with exemptions wide enough or an effective date distant enough that nothing changes in a classroom. The current bill routes reporting through the states without stating what follows for a state that reports nothing, which is the position the federal count is already in.

(b) alone is worse. The count cannot detect a district that reports zero incorrectly, so a fall could equally mean fewer children restrained or fewer districts reporting. The collection is biennial and the most recent published year ended in 2022, so several consecutive collections is close to a decade of waiting. The unit is also wrong for the question, because it counts students rather than events and a child restrained forty times counts once.

(c) alone counts paper. Twenty-two jurisdictions already required an emergency before restraint as of 2019-07-01, and the practice continued at the scale the federal count records. Coverage in a statute book is not the same as a classroom where it does not happen. The three have to be read together, and (b) cannot be read at all until someone publishes what the zeroes are worth.

What is it connected to?

Fills with researchrestraint and seclusion in hospitals, residential treatment and juvenile facilities · school policing and the presence of law enforcement in discipline, which the current bill reaches explicitly · special education funding and staffing shortages · the same practice and the same regulatory question in other countries. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• Whether the headline number counts each student once. The published total of approximately 105,700 is the exact arithmetic sum of its three components, while its label reads and/or, which reads as a count of distinct students. No endnote, analytical note or methodology statement in the report addresses the difference. The most cited number in this field has an unexplained construction.
• How wrong the zeroes are. GAO named false zero reporting in April 2020 for school year 2015-16 data, and the department closed all six recommendations by August 2024, but no source opened here publishes a correction factor. Nothing states what the count would be if the zeroes were right, or what share of districts reporting zero actually had incidents.
• What left the paperwork. For the same collection and the same 17,717 annual responses, the estimated annual burden moved 2,290,195 hours on 2024-10-17, to 2,378,410 on 2025-01-16, to 1,762,790 on 2025-08-07 — a drop of about 26 percent in seven months. None of the three notices contains the word restraint or the word seclusion, and none names a single removed data element.
• When the next count arrives. The department overview for the 2023-24 collection lists the restraint and seclusion items in full, including instance counts, and states no publication date. Both department data portals returned only a loading state across four attempts on two hostnames, so a reader cannot determine from the source whether a newer federal count exists.
• Whether anyone counts harm. The federal collection counts exposure, not injury, hospitalization or death. GAO reported in 2009 that it could find no entity collecting such information, and nothing opened here shows that one now exists. Section 202 of the current bill would require injury and death to be reported, which means the bill itself treats this as a known hole.
• What a limit would cost. No cost estimate, no committee report and no opposition statement was reachable in any source opened here, in sixteen years of filings. The one quantified compliance figure that exists anywhere in this file is a paperwork burden estimate that does not mention the practice.
• How many jurisdictions have no binding protection today. The most detailed comparison across 51 jurisdictions still serves its July 2019 edition at its canonical address as of 2026-08-08. The 2024 peer-reviewed review states that most states have legislation or guidance and publishes no counts.
• Who was cut. The March 2025 announcement reports the halving of the department and lists the statutory programs that continue. It does not state whether the Office for Civil Rights, which administers the count and enforces Section 504 in schools, was among the offices reduced.

See the evidence

Four primary federal documents were read end to end. The 2021-22 First Look report, the 2009 GAO testimony, the two congressional bill texts and the 2023-24 overview were opened as PDFs or plain text and extracted locally rather than summarized, and every figure in this dossier that carries a number came from one of them or from a GAO page read verbatim. The three Federal Register notices were fetched twice each so the burden-hour figures could be compared character by character; the burden drop is a comparison performed here and is not asserted by any single notice, and two of the three rows carry no address because only the document numbers survived into the research record and a plausible address is not assembled to fill the gap. Three disagreements are left visible rather than resolved. The introduction date of the current bill is 2025-12-11 in the primary text and 2025-12-12 in the one trade outlet that reported it, and no source explains the one-day difference. The same federal report uses 14 percent of enrollment for students with disabilities in the restraint section and 17 percent in the discipline section, under two different definitions that a reader comparing sections would not see. And the total of 105,700 is at once described as a count of distinct students and published as the exact sum of three possibly overlapping components, which is why the population range in this dossier has two ends rather than one. One methodological warning belongs on the record: during this research a GAO PDF fetch returned an unrelated defense infrastructure report, and the summarizer produced a confident fabricated quotation matching the subject requested. Every non-primary figure here was checked against raw document text for that reason. This is a Path A output, so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range 68,800–105,700 As of school year 2021-22

Sensitivity The width of this range is an overlap the source never resolves. The published total of approximately 105,700 is the exact arithmetic sum of the three components, while its label reads physically restrained, mechanically restrained, and/or placed in seclusion, which reads as a count of distinct students in which a student both restrained and secluded appears once. No endnote, analytical note or methodology statement in the report says which reading applies, so the number of distinct students lies between the largest component and the sum. What the range fails to count: incidents rather than students, since a child restrained forty times counts once and the report publishes only student counts even though instance elements exist in the collection; students in placements outside school districts, where elements to capture them were only proposed for the 2025-26 collection and advocacy comment says rates run highest; injury, hospitalization and death, none of which any federal collection records; and any school year after 2021-22, four of which have ended with no published federal figure as of 2026-08-08. The limit in the opposite direction is that the whole range is a floor, not a band around the truth. GAO reported that 70 percent of districts recorded zero incidents in school year 2015-16, that the federal quality check reached fewer than 100 large districts, and that officials read the definitions inconsistently; no source publishes a correction factor, so the real figure sits above this range by an amount nobody has quantified. Two smaller effects run inward: counts of 1,000 or greater are rounded to the nearest hundred, and public-use counts carry small random adjustments for privacy. For scale rather than for precision, 105,700 against public elementary and secondary enrollment of 49.4 million in fall 2021 is roughly 0.21 percent, about one student in 470 per school year; that ratio is approximate because the enrollment denominator includes prekindergarten while the shares in the report are K-12.

Regional breakdown The report publishes national totals and demographic shares only. No state, district or regional breakdown of the restraint and seclusion counts appears in any source opened in this round, and the two Department of Education data portals that would carry district-level figures returned only a JavaScript loading state on four attempts across two hostnames. The one sub-national figure found anywhere is an Illinois incident count assembled by journalists from district records, which uses a different unit and a different period and therefore cannot be placed inside this range.