One in Five: IEPs and 504 Plans in America

Bottom line

A mixed verdict — with a tilt toward real need, counted through an elastic system

The assessment

The recent rise is driven mainly by expanded identification of real need — autism, ADHD, developmental delay, and Section 504 plans. But who gets identified, where, and under which label is shaped powerfully by the system itself: state rules, local practice, family resources, and classroom context.

The strongest single piece of evidence is the category pattern. Specific learning disability — the one category the “poor instruction” theory predicts should be swelling — has been shrinking for two decades, down roughly 450,000 students between 2002 and 2023 even as total IDEA enrollment grew by about 1.1 million. Growth is concentrated where diagnostic criteria broadened, screening improved, and the law widened: autism (up more than eightfold in count), other health impairment (the ADHD category, more than tripled), and 504-only plans (nearly quadrupled since 2009–10).

That does not acquit the system. A child is roughly 1.8 times as likely to have an IEP in Pennsylvania, New York, or Maine as in Hawaii, Idaho, or Texas — a map no distribution of child biology explains. Texas suppressed its rate for a decade with an administrative benchmark, then surged 67% in eight years once the benchmark was banned. Instructional casualties are real and numerous among struggling readers — but they are a large standing problem, not the engine of the current national increase.

Explanation A — Real need

The country is counting need it used to miss

CDC counts autism in 1 in 31 eight-year-olds (2022), up from 1 in 150 in 2000, with earlier identification a large part of the rise.
11.3% of children ages 5–17 had ever been diagnosed with ADHD in 2020–22, up from about 9% in 2007–09.
40% of high-school students reported persistent sadness or hopelessness in 2023.
Adjusted for achievement, Black, Hispanic, ELL, and female students are under-identified for learning disabilities in replicated national studies — unmet need remains.

Explanation B — System-produced

The system shapes who gets a label

State IDEA rates run from ~11–12% to ~21% under the same federal law.
The youngest children in a grade are ~34% more likely to be diagnosed with ADHD than the oldest.
Tennessee's statewide RTI rollout coincided with a 61% drop in the odds of first-time SLD identification by grade 3.
504 plans skew affluent and White: White students are 44% of enrollment but 60% of 504-only students.

Figures above are detailed with sources in the sections that follow.

The numbers

Half a century of growth — a plateau, then a new climb

IDEA was signed in 1975, when 8.3% of public-school students were served under it. The rate roughly doubled by the early 2000s, sat nearly flat for a decade, and has climbed again since about 2014 — with a clear post-COVID acceleration: the Section 618 child count grew about 3% in 2022, 3% in 2023, and 3.8% in 2024.

Students served under IDEA, as a share of public-school enrollment

Sources: NCES Digest of Education Statistics (1976–77 through 2010–11; the 2000–04 plateau ranged 13.3–13.7%); The Advocacy Institute's analysis of IDEA Section 618 data (14.1% in 2019, 15.2% in 2023); NCES Condition of Education 2024 (15.2% in 2022–23). Part of the recent rise in the percentage is a shrinking denominator: total K–12 enrollment fell about 2% from FY16 to FY24 while special-education enrollment rose about 17% (Bellwether, FY24 analysis).

Three federal counts, three different totals — all correct

News coverage of this story cites IDEA totals of 7.5, 7.6, 7.8, and 8.2 million, sometimes in the same week. They are not contradictions; they count different universes. OCR itself warns that its figures “may differ from those reported by other offices within ED due to differences in methodology, data sources, and survey population.”

Source	Universe	IDEA total	Rate
CRDC 2023–24 (OCR)	Preschool–grade 12 enrollment, 49.7M students	~7.6M	15%
Section 618, 2024 (OSEP)	All eligible students ages 3–21	8,194,424	—
NCES Condition of Education 2024 (2022–23 data)	Ages 3–21 against public enrollment	7.53M	15.2%
Bellwether FY24 analysis of 618 data	K–12 only	~7.8M	nearly 16%

Section 618 figures via The Advocacy Institute; Bellwether via its FY24 funding analysis. Section 618 also counts 458,920 infants and toddlers under Part C (birth–2) in 2024, down 1% from 2023.

One more caution on the headline: the jump from 17% (2021–22 CRDC) to 19% (2023–24 CRDC) sits almost entirely on the IDEA side. The 504-only share was already 3.9% in 2021–22 and rounds to 4% in 2023–24. No source decomposes that two-point jump into real growth versus reporting change — anyone who attributes it to a single cause is guessing.

The discriminating evidence

The category that should be growing isn't. The ones that are, tell a different story.

If weak core instruction were manufacturing disability at scale, the fingerprint would show up first in specific learning disability (SLD) — the reading-and-math category. Instead, SLD peaked at about 6.1% of all students around 2000–04 and has fallen ever since: from 45% of all IDEA students in 2000–01 to 32% in 2021–22, and roughly 452,696 fewer students in 2023 than in 2002. The growth is in autism, other health impairment (OHI — where ADHD sits), developmental delay, and 504-only plans.

Share of all IDEA students, by category: 2000–01 vs. 2021–22

2000–01 2021–22

Source: Pew Research Center analysis of federal data (2021–22), July 2023; NCES. By 2024 (school-age, Section 618), autism had risen further to nearly 15% of IDEA students and accounted for 40% of the entire 2023→2024 increase. “Other” combines the remaining categories (multiple disabilities, orthopedic/visual/hearing impairments, traumatic brain injury, deaf-blindness).

+866,089

Autism, 2002 → 2023

More than an eightfold increase in students served under the autism category — the largest absolute gain of any category. CDC's surveillance network counted autism in 1 in 31 eight-year-olds in 2022 (1 in 36 in 2020; 1 in 150 in 2000), and children born in 2018 were 1.7× as likely as those born in 2014 to be identified by 48 months — earlier identification explains a large share of the school-age rise.

−452,696

Specific learning disability, 2002 → 2023

The largest absolute decline of any category, over the same two decades in which total IDEA enrollment grew by roughly 1.1 million. OHI more than tripled in the same window. In 1976, one in four IDEA students had SLD, autism, or OHI; by 2023–24, more than two in three did (CRPE, Outmatched, 2025).

Labels migrate: diagnostic substitution

Part of the autism rise is children who would once have carried a different label. Researcher Paul Shattuck documented that as autism's administrative prevalence rose from 0.6 to 3.1 per 1,000 students between 1994 and 2003, the intellectual-disability (“mental retardation”) label fell 2.8 per 1,000 and the learning-disability label fell 8.3 per 1,000. Intellectual disability has since fallen from 2.2% of all enrollment in 1976–77 to 0.9% in 2010–11. Substitution reshuffles the composition of special education more than it inflates the total — but broader criteria (the autism spectrum, ADHD presentations in girls, anxiety under 504) also genuinely enlarge the eligible pool.

One category conspicuously not rising: emotional disturbance, IDEA's mental-health category, flat at roughly 4–5% of IDEA students even as adolescent distress indicators spiked. Where that need is being routed — OHI, 504 plans, services outside school, or nowhere — is one of the open questions at the end of this page.

Category counts: Education Next summary of NCES Digest tables; CRPE, Outmatched (Oct. 2025); CDC ADDM Network, 2022 surveillance (released April 2025); Shattuck, Pediatrics (2006).

One law, fifty rates

Where a child lives predicts an IEP almost as much as anything about the child

IDEA is one federal law, but identification rates run from about 11% to about 21% depending on the state — and the pattern repeats across every recent snapshot, whatever the year or denominator. A child in the Northeast is roughly 1.8 times as likely to have an IEP as a child in Hawaii, Idaho, or Texas. No plausible map of underlying child biology looks like that.

IDEA identification rate by state, highest and lowest (2021–22)

Source: Pew Research Center / NCES, 2021–22. The 2023 snapshot (Section 618, via The Advocacy Institute) shows the same spread: Pennsylvania highest at 21.1%, Hawaii lowest at 11.7%, national rate 15.2%.

Why the gaps? What is actually known

Eligibility rules genuinely differ. A 2020 Congressional Research Service comparison of state regulations found variability in “specificity, severity, method of identification, and timeline” — least for low-incidence disabilities, most for high-incidence ones, “most notably” SLD. A child eligible in one state can be ineligible in another (GAO, via ASHA).

State law moves counts mechanically. States may use the “developmental delay” category up to age 6, 8, or 10 — the choice alone shifts totals. Dyslexia laws and screening mandates add students without any change in children: Texas's 2024 jump of 77,404 school-age students — 27% of the entire national increase that year — is attributed by The Advocacy Institute to new Texas requirements for serving students with dyslexia. K–12 Dive notes Texas now also pays districts an additional $1,000 per initial special-education evaluation.

Funding incentives are weaker than the folklore. States mix weighted formulas (dollars per identified student, which can reward identification), census-based formulas (dollars on total enrollment — designed, like California's AB 602 and the federal formula, to remove the identification incentive), reimbursement, and high-cost pools. But federal IDEA funding was about $14.2 billion in FY24 — roughly 11% of average per-pupil expenditure, against the 40% Congress authorized — and real funding per IEP student has fallen since FY16. A district that identifies another student usually takes on a legal entitlement costing far more than the marginal dollars that follow. The crude claim that “schools label kids for the money” is not supported nationally.

The Texas natural experiment

In 2004 Texas set an 8.5% special-education benchmark. Exposed by the Houston Chronicle in 2016 and investigated by the U.S. Department of Education, it had functioned as a cap: federal monitors found children “shifted to response-to-intervention programs… but, once there, some languished for years,” and districts sidestepping IDEA by serving students through Section 504 instead, which carries fewer rights. After the benchmark was banned, Texas IDEA enrollment grew 51.8% from FY16 to FY23 and 67% through FY24 — the largest growth in the nation — and Texas still sat near the bottom on rate in 2021–22. For over a decade, in the second-largest state, an administrative rule — not child need — set the rate.

Sources: Education Week (Jan. 2018) on the OSEP monitoring report; Bellwether FY24 analysis.

The instruction question

Are untaught children being mistaken for disabled children?

This is the sharpest version of Explanation B: children who were never properly taught to read hit a wall, get referred, and get labeled — “instructional casualties.” The honest answer from the research is: the casualties are real, the mechanism is documented, and it still cannot explain the current national rise — because the category it acts on is shrinking.

Evidence for the hypothesis

Prevention works in principle. In the classic studies (Vellutino 1996; Foorman 1998; Torgesen), daily one-to-one tutoring for a semester brought the majority of first-grade poor readers into the average range — though at a dosage almost no general-education system provides at scale.
The field's own standard presupposes good teaching. Miciak & Fletcher (2020): a valid SLD determination requires word-reading deficits “combined with inadequate response to evidence-based instruction.” Where instruction wasn't evidence-based, the determination isn't valid — the data cannot distinguish casualty from disability.
Tennessee. Statewide RTI adoption was associated with a 61% average decrease in the odds of first-time SLD identification by third grade (Gilmour et al., 2024), with larger declines for Black and economically disadvantaged students. The state education department said SLD counts fell by a third and suggested many struggling students had been “misidentified… because their instructional needs simply weren't being met.”
Instruction moves achievement at state scale. Mississippi's 2013 literacy law (screening, coaches, phonics-based training) took the state from 49th in grade-4 NAEP reading in 2013 toward the top 10 by 2024, with estimated gains of +0.14 SD in reading for exposed students.

Evidence against — and the complications

The category trend. SLD is falling in share (45% → 32% of IDEA) and in count (−452,696, 2002–2023) while totals rise. Weak instruction may sustain a large stock of struggling readers, but the flow into SLD IEPs is shrinking.
Under-identification is what adjusted studies find. Controlling for achievement and other factors, Black students' adjusted odds of LD identification run 0.50–0.69 versus White students; Hispanic 0.45–0.83; English learners 0.32–0.59; girls 0.64–0.90 (Morgan et al.). “We observed no evidence of LD over-identification of students of color.” (Raw, unadjusted rates do show Black and Native American students identified at higher rates — both facts are true at once.)
RTI as actually practiced didn't prevent much. The IES national evaluation (Balu et al., 2015; 146 schools, 13 states) found first-graders assigned to RTI reading intervention scored lower in spring than near-identical peers not assigned, with no significant effects in grades 2–3.
Dyslexia laws alone don't move identification. Across 47 states with dyslexia laws (Hu et al., 2025), 55% showed no significant change in SLD identification after enactment, 28% increased, and 17% decreased.
Special education isn't obviously the stronger intervention. In one quasi-experiment (Burns et al.), students with severe reading deficits in Tier 2 intervention grew more than similar students receiving special education for SLD.

Handle with care: the “95%” figure

The most-quoted number in this debate comes from G. Reid Lyon, former chief of child development at NICHD, in an interview for Children of the Code: “Ninety-five percent of the kids hitting the wall in learning to read are what we call NBT: Never Been Taught… About five to six percent of those kids have what we call dyslexia or learning disabilities in reading.” That is an expert's assertion about struggling readers — not a peer-reviewed estimate of false-positive IEPs, though it is routinely misquoted as one. In the same project, Yale's Sally Shaywitz, asked how many children have inherent dyslexia versus a mix, answered in substance that it is “an empiric question. We don't know the answer.” No peer-reviewed national estimate of the share of IEPs preventable with better core instruction exists. The defensible bounds are the intensive-tutoring experiments above (an upper bound, under ideal dosage) and Tennessee's 61% (an identification effect whose composition — prevention versus delayed or denied services — its own authors leave open).

The system critique, in its strongest current form, comes from CRPE's Outmatched (2025). Its argument is not that children lack disabilities; it is that the delivery system manufactures demand for the label: one teacher delivering one grade-level lesson to 20–30 children “regardless of their readiness to learn” makes special education “the tantalizing prospect that reliable help is within reach,” while eligibility rests on subjective determinations that cannot reliably separate disability from “poverty, language, or poor instruction.” Two third-graders who both cannot read grade-level text can get opposite determinations — one a legal right to specially designed instruction, the other “whatever the school chooses to make available.” CRPE quotes Laurence Lieberman (1980): “Special education has been confused with being an answer to the problems of regular education.”

Sources: Miciak & Fletcher, Journal of Learning Disabilities (2020); Gilmour et al. (2024); Balu et al., IES/NCEE (2015), via Education Week; Hu et al., Annals of Dyslexia (2025); Morgan et al., Educational Researcher (2017) and replicated NAEP analysis; CRPE, Outmatched (2025). Whether Mississippi's literacy law reduced SLD identification is unverified — achievement effects are documented; no source reports the identification effect.

Natural experiments

Four moments when the system revealed itself

The cleanest evidence that identification is elastic — responsive to rules and context, not just to children — comes from accidents of policy and calendar.

1. The birthday effect

In states with a September 1 kindergarten cutoff, August-born children — the youngest in their class — had a 34% higher ADHD diagnosis rate than September-born children (85 vs. 64 per 10,000). In states without that cutoff, the gap vanished (Layton et al., New England Journal of Medicine, 2018). Some share of ADHD identification is immaturity read as disorder against older classmates. How large that share is nationally, the study cannot say.

2. Tennessee's RTI switch

When Tennessee rolled out RTI² statewide, the odds of a child being identified with SLD for the first time by third grade fell 61% on average (Gilmour et al., 2024). Either better early intervention prevented disabilities from taking hold — or the new gate kept children out, or delayed them. Texas showed the dark version of the same mechanism: RTI used as a waiting room where children “languished for years.” The Tennessee authors explicitly leave the composition unresolved.

3. The testing-accountability era

Under the old high-stakes testing rules, researchers (Cullen & Reback, 2006; Figlio & Getzler, 2006) found schools reclassifying low-performing students as disabled in tested grades in ways that could exempt their scores — movement concentrated in the subjective categories; severe disability categories did not rise. ESSA's participation and subgroup rules have since narrowed that payoff, and no recent replication exists.

4. The College Board change

In 2017 the College Board began automatically honoring school-based accommodations on the SAT. A Princeton study (Tiffany Liu, 2026) found 504 enrollments rose afterward — with the sharpest increase in wealthier schools. A testing-policy change, aimed at fairness, measurably moved who seeks a disability plan.

Also relevant: teacher referral is a biased gate in both directions. In an NYU vignette experiment (Fish), teachers were more likely to refer White boys than Black or Latino boys for academic problems, and more likely to refer Black and Latino boys for behavioral ones.

The quieter expansion

Section 504: the plan that nearly quadrupled

Section 504 plans — accommodations without special-education services — are the fastest-growing major plan type in American education, and the least examined. Education-law scholar Perry Zirkel (Lehigh) tracks the rate at 1.1% of students in 2009–10, 3.9% in 2021–22, and 4% (1.9 million students) in 2023–24. The starting point is not a coincidence: 2009–10 is the first school year after Congress, in the ADA Amendments Act of 2008, broadened what counts as “substantially limits a major life activity.” Growth continued well beyond the amendment's direct effects.

Students with a Section 504 plan only, share of enrollment

Sources: Zirkel, “Concerns About Misidentification of ‘504-Only’ Students” (2026), analyzing U.S. Dept. of Education data; OCR CRDC 2023–24 First Look. ADHD has historically been the most commonly listed impairment on 504 plans, with diabetes next in mid-2000s survey work; anxiety and physical-health conditions (asthma, food allergies) are also common. CRDC does not publish a national condition-by-condition breakdown — any precise national percentage by condition is unverified.

Two systems, two income gradients (Oregon, 2008–09 to 2018–19)

Lowest income Highest income

Source: Ainsworth et al. (Jan. 2026), Oregon student records linked to IRS tax data, via The 74 / Disability Scoop. In high school the 504 gap widens: 4.5% near the top of the income distribution vs. 2.0% near the bottom. CRDC 2023–24 shows the same racial pattern nationally: White students are 44% of enrollment but 60% of 504-only students.

The two plans run in opposite directions along the income ladder — and that single fact captures the whole 504 debate. One reading: affluent families, who can hire evaluators and advocates, obtain plans their children don't strictly need (the College Board finding above shows the incentive is real). A second: affluent children genuinely have more of the anxiety and mood conditions that 504 — not IDEA — is built to accommodate. A third: low-income children with identical needs can't navigate an advocacy-driven system. The Oregon study cannot distinguish among the three, and no audit of 504 eligibility accuracy at scale exists.

The spread within 504 is wider than within IDEA. Texas averaged 7.6% 504-only in 2021–22 — nearly double the national rate — with New Hampshire also around double, while New Mexico and Mississippi ran at less than half. Texas accounted for 23 of the 25 highest-rate districts in the country (14.4%–17.4%); individual Texas middle schools reported 23.1%–30.1%, while other Texas schools reported zero. Zirkel's inference: “The disparity between these extremes likely cannot be attributed simply to differences in local incidence rates of physical and mental impairments.”

Why is 504 so elastic? Its eligibility test is broad — an impairment that substantially limits a major life activity — with no requirement of need for special education, no federal funding attached, and no evaluation apparatus comparable to IDEA's. Zirkel's conclusion is deliberately two-sided: loose identification practices produce both over-identification (diverting time and resources from “true positives”) and under-identification (Child Find liability at the other end), and both warrant administrative attention.

Sources: Zirkel, ATPE News (Mar. 2026); Ainsworth et al. via Disability Scoop (June 2026); OCR CRDC 2023–24.

The other headline in the same report

Restraint, seclusion, and discipline fall hardest on the students the system labels

Whatever explains the identification numbers, the CRDC's discipline data describe what school looks like for many identified students. About 129,700 K–12 students were physically restrained, mechanically restrained, and/or secluded in 2023–24.

84,300 students physically restrained

32,300 students secluded

13,100 students mechanically restrained

27% of students receiving out-of-school suspensions were IDEA students (15% of enrollment)

IDEA students were 73% of students physically restrained and 68% of students secluded. Boys were 51% of enrollment but 79% of students physically restrained; Black students were 15% of enrollment but 29% of students physically restrained and 46% of those mechanically restrained. IDEA students also accounted for 24% of in-school suspensions and 23% of expulsions.

Read the numbers carefully

Some advocacy coverage of this same release reports higher figures — around 83% of restraint and 78% of seclusion. Those are shares of incidents, not of students; one child can be restrained many times. The 73% and 68% figures above count students, as OCR reports them on page 9 of the First Look.

Source: OCR, A First Look: CRDC 2023–24, pp. 9 and 15.

On the record

Expert voices, in their own words

Exact published quotes only. Where a position is summarized rather than quoted, it is labeled as a position.

“Students with disabilities and students who are black are still subjected to disproportionately high exclusionary discipline, restraint and seclusion. Every time schools resort to suspending, restraining or secluding, the child is not learning. The program has failed them.”

Denise Marshall, CEO, Council of Parent Attorneys and Advocates (COPAA) — on the CRDC discipline data, Disability Scoop, Oct. 2026

“Many of the children now being counted in the autism category would probably have been counted in the mental retardation or learning disabilities categories if they were being labeled 10 years ago instead of today.”

Paul T. Shattuck, then University of Wisconsin–Madison — on diagnostic substitution, Pediatrics research (2006)

“These are benefits that don't come with a lot of costs. Your child is typically not leaving the classroom. They might be seen as beneficial without much downside in terms of tradeoffs.”

Paul L. Morgan, University at Albany — on why families seek 504 plans; in the same article he added: “I imagine there is abuse or manipulation of the system, including by parents who view it as a way for their child to get additional support.”

“Only additional research that audits 504 plans and investigates the evidence of disability for each student can really determine the degree to which these three factors explain the disparities.”

Ben Lovett, Columbia Teachers College — on the 504 income gradient: unnecessary plans, genuinely higher affluent anxiety, or navigation barriers for low-income families

“When kids are pulling As and Bs, school staff generally aren't referring them to assessments, whether for 504s or IEPs. So it really has to come from the family — and that's where you need to have the ability to educate yourself, or hire someone to help you with it.”

Miriam Nunberg, former OCR attorney, now education-rights consultant

“We had some hypotheses about what that would look like with respect to 504 plans, but we did not expect to see those differences favoring high-income students.”

Nick Ainsworth, UC Irvine — lead author of the Oregon income study

“Currently, research does not show that living in certain communities puts children at greater risk for developing ASD.”

CDC, ADDM Network — attributing autism's geographic variation (1 in 103 to 1 in 19 across surveillance sites) to identification practice, and framing higher identification as ascertainment to be copied

“There's been a very interesting, barely acknowledged and not understood shifting in the demographic profile of autism. When we started… autism was over-represented among affluent communities and white communities compared to minority and low-income communities.”

Walter Zahorodny, Rutgers New Jersey Medical School, ADDM co-author — in 2022 ADDM data, prevalence was higher among Asian/Pacific Islander, Black, Hispanic, and Native children than White children: catch-up ascertainment

Two further positions, paraphrased because the sources state them as analysis rather than quotable lines: Perry Zirkel (Lehigh) holds that 504 growth reflects both rising incidence of ADHD, dyslexia, anxiety, and health conditions and loose identification producing over- and under-identification; he has said the resulting accommodation gap “unfortunately increases the gap between the haves and the have-nots in our schools.” Ashley Jochim & Alexander Kurz (CRPE) argue special education has become the stopgap for a general-education delivery system that cannot differentiate — a system problem, not a claim that children's disabilities aren't real.

The scorecard

Strength of evidence, hypothesis by hypothesis

Hypothesis	Evidence	Key supporting facts
A1. Autism, ADHD, and mental-health need is genuinely more prevalent or more impairing than in the past	Moderate	ADHD ever-diagnosis rose ~9% → 11.3%; adolescent distress rose 2013–2021, then partially receded; pandemic-born cohort studies show developmental effects. But no study separates true incidence from ascertainment for ADHD/anxiety, and IDEA's emotional-disturbance category is flat — need and labels are poorly coupled.
A2. Better identification of existing need — Child Find, screening, earlier autism ID, catch-up in missed groups, suppression regimes ending	Strong	Autism's IDEA share rose 1.5% → ~15% with earlier median identification and demographic catch-up (CDC); diagnostic substitution documented; Texas +67% after an illegal suppression regime ended; preschool Part B surges (+13.6% in 2022, +9.8% in 2023); adjusted studies still find under-identification of students of color, implying unmet need remains.
A3. Expanded legal eligibility — the 2008 ADA Amendments Act, ADD's addition to OHI (1990), developmental delay, dyslexia laws	Strong	504-only rose 1.1% → ~4% after the 2008 broadening; OHI identifications rose over 400% in the decade after ADD was added to eligibility; Texas's 2024 jump is tied to dyslexia-serving rules. Legislatures chose wider nets; wider nets catch more fish — real ones included.
B1. Weak core instruction produces “instructional casualties” who get labeled, especially in reading	Moderate for the stock; weak for the rise	Prevention experiments, the Miciak–Fletcher validity standard, Tennessee's RTI association, and Mississippi's achievement gains all show instruction matters enormously for reading. But SLD identification is falling in share and count while total IDEA rises — B1 cannot explain a rise driven by autism, OHI, and 504. The “95%” figure is not evidence.
B2. RTI/MTSS as implemented fails — referral becomes the intervention, or RTI becomes a delay mechanism	Moderate	The IES national evaluation found null-to-negative early-reading effects for RTI as practiced; federal monitors found Texas using RTI to park children for years; Tennessee's RTI adoption changed identification odds by 61% with the prevention-vs.-denial composition unresolved; Tier 2 outperformed SLD special education for severe readers in one study.
B3. Incentives — funding, accountability, test accommodations, parental advocacy advantage	Moderate at the margins; weak as a funding story	Accountability-era reclassification studies; the College Board policy → 504 bump concentrated in wealthy schools; the 504 affluence gradient; Texas's per-evaluation payment. Against: funding is largely census-based, real per-IEP funding is falling, and identification imposes net costs and legal obligations on districts.
B4. Diagnostic substitution and category migration inflate headline growth	Strong for composition	Shattuck's substitution accounting (autism up as ID and LD fell, 1994–2003); CRPE's composition analysis; SLD↓ and ID↓ offsetting part of autism↑. Substitution reshuffles labels more than it creates students — though broadening criteria also genuinely enlarge the eligible pool.

What the data cannot yet answer

Eight honest open questions

What fraction of current IEPs and 504s would fail an independent eligibility audit? No national audit exists for either program. Without one, every “over-identification” percentage, from any side, is rhetoric.
Did Tennessee-style RTI declines reflect children learning to read — or children denied or delayed services? The study's authors flag it; long-run achievement and exit follow-up by subgroup is the missing research.
Have post-2019 science-of-reading laws changed SLD identification anywhere? No published study was found. Mississippi's SLD series under its literacy law is an obvious, unexploited test case.
How much of the autism rise is true incidence? Surveillance can document identification; by design it cannot settle causes. Claims beyond “broadening, ascertainment, and substitution explain a large share” exceed the evidence.
Why doesn't adolescent mental-health need show up in IDEA's emotional-disturbance category? Is it routed to 504 and OHI, served outside school, or unserved — and in what proportions, by income and race?
What explains the CRDC's 17% → 19% jump specifically? Real acceleration versus reporting and method changes has not been decomposed in any source found.
Does identification help? Outcome research by category is thin and mixed — one quasi-experiment found Tier 2 intervention beating SLD services for reading growth. There is no consensus causal estimate of IEP effects by category to weigh against the costs of labeling.
What is actually in the nation's 504 plans? CRDC does not publish the condition mix. The fastest-growing major plan type in American education is, at the federal level, essentially uncounted by condition, by race-within-condition, or by outcome.

What could not be verified

Four claims circulate in this debate that this investigation could not tie to a primary or reputable source, and they are not used as evidence anywhere on this page: (1) any national percentage breakdown of 504 plans by condition — CRDC does not publish one; (2) any claim that Mississippi's or another state's reading law reduced SLD identification — achievement effects are documented, identification effects are not; (3) a peer-reviewed national estimate of the share of IEPs that are “preventable instructional casualties” — it does not appear to exist; (4) a decomposition of the CRDC 17% → 19% increase into real growth versus reporting change.