Education Freedom Accounts:
the published record, drawn

How to read this page

Every figure carries one of five badges. The badge is the point: it says what kind of thing you are looking at before you read the number.

Every number on this page is built from the source tables by a script and served from a single data file; none is typed into a chart. The tables themselves are published below, in full, as CSV and as one workbook. Method, sources and known anomalies →

What this page found

Every figure below carries a badge saying what kind of claim it is. The tables behind every chart are published at the foot of the page as CSV.

The program today

2023-24 and 2024-25 from the DESE annual reports. 2025-26 approvals from the 2024-25 report's forward-looking Outlook section. 2026-27 approvals from records the Department released under FOIA, published by the Arkansas Advocate on 1 July 2026. Each tile carries its own badge.

Applications, approvals, active accounts

The gap between approved and active is small; the gap between applied and approved is not. In 2023-24, 795 of 6,495 applications did not result in an approval. In 2024-25, 4,464 of 18,921 did not — 12% of applicants, then 24%.

The series stops there. The state has published no applications figure for 2025-26 or 2026-27, so it cannot be said whether the rate kept climbing.

Six cells read not published — and of the four approved figures, one is a projection rather than a count. The last column says which each is. Only the two closed years have an active-participant figure at all.

What it was budgeted, and what it becamepublished

Each dot is one figure from one primary document — an appropriation act, a Legislative Council transfer, the Governor's budget letter, or an annual report. For 2025-26 there are five separate figures across four documents — a base appropriation, a reserve authorization, the approval to transfer it, a supplemental, and the matching transfer for the supplemental. They measure different things and are not in conflict; what no single document does is total them.

Every point is tied to a named document; hover for the citation. A widely quoted 2025-26 award total of $326.1M is deliberately absent — it traces only to a September 2025 internal report shared with reporters and legislators, never published, so it cannot be sourced here.

The sector flipreported

Homeschooling was not an allowable use of the funds in year one. Two years after it became one, homeschool families are nearly half the program. This is the largest single change in its composition and it has gone almost entirely unremarked.

Two years are drawn, not three: 2023-24 is absent because homeschooling was not an eligible use of the funds that year. 2024-25 comes from the annual report. The 2026-27 split is reported — not published, and not obtained: an Arkansas Department of Education spokesperson gave the approval counts to the Arkansas Advocate, which printed them on 1 July 2026.

Where participants came frompublished

The program's own reports answer the central policy question directly, and both years answer it the same way: the share of participants who had attended an Arkansas public school the year before fell from 18% to 12%. That one comparison is like-for-like. Nothing else about the two years is.

Beneath it, the 2024-25 breakdown — the only year that has one. A third of that file has no prior-school entry at all, shown as its own category rather than redistributed, because redistributing it would be a guess.

The public-school share is stated outright by both reports and is compared directly. The remainder is not: in 2023-24 it is that report's own single “not previously public” category, and in 2024-25 it is five listed categories plus a third of the file with no entry — so the two remainders are drawn as one neutral block each and are never set against one another.

Who qualified, and under which doorpublished

Disability qualification fell from 57% of participants to 36% — while the count of them rose, from roughly 3,160 students to 5,130. The share fell because far more of everyone else joined, not because fewer disabled students did.

The doors widen because the statute widened them, on a schedule fixed before the program opened. Ark. Code Ann. § 6-18-2506(a)(3), added by the LEARNS Act (Act 237 of 2023, § 42), sets eligibility in three steps: the named categories in 2023-24; an expansion in 2024-25 to students from D-rated as well as F-rated schools and to the children of veterans, reservists, first responders and law enforcement officers; and then, “[f]or the 2025-2026 school year and each year thereafter, any resident of this state who is eligible to enroll in a public elementary or secondary school shall be eligible.” Participation was capped over the same three steps — 1.5% of 2022-23 public school enrollment, then 3%, then, from 2025-26, “no limitation”. The two columns above are the first two steps. The third has no column because no 2025-26 annual report exists, and it is the last: the schedule ends at universal eligibility.

The percentages do not sum to 100: a family may qualify under more than one door, and each report counts every door separately. The two columns are not a series — three categories appear only in 2024-25, and the F-rated-school door widened to D-or-F between the years, so it is printed as two rows rather than joined into one. Homeless is reproduced exactly as printed in 2024-25: “<1%”, never estimated into a number.

What the money boughtpublished

Tuition is 81% of the dollars but 46% of the transactions. Supplies are the mirror image: 22% of all transactions, 8% of the money. Dollars and transactions tell different stories about what this program is used for, so both are drawn — on their own scales, side by side.

2024-25 report Table 3. The fourteen category rows sum to $84,506,226 and 180,602 transactions — both reconcile exactly to the report's stated totals. reconciles

Two programs wearing one namepublished

Split the same spending by who did it and the program stops being one thing. Private school families spend 97% of their money on tuition. Homeschool families spend 39% on supplies, 21% on curriculum, and 21% on tuition to somebody else. Almost nobody reports this, because it sits in two tables the summary table hides.

2024-25 report Tables 4 and 5. Each reconciles exactly to its own stated totals, in dollars and in transaction counts. reconciles

Home schooling, 1997 to nowpublished

The longest continuous series the state publishes on any of this. Home schooling grew steadily for twenty-five years, jumped with the pandemic, and then jumped again — this time with public money attached.

DESE annual home school reports, 1997-98 to 2025-26, as compiled in this project. Shaded band marks the EFA era. These are all home-schooled students, not only those with voucher accounts.

Four shops, then forty-nine pagespublished

In 2023-24 every non-tuition purchase in the program went to four retailers — Staples for supplies, and The Toggery, Especially For You Designs and Khakis for uniforms. $1.5m between them, against $32.6m of tuition. A year later the recipient appendix ran to forty-nine pages and 1,971 names.

Most of that is the second program arriving. Homeschooling became an allowable use in 2024-25, and homeschool families buy from everywhere.

2023-24 report Table 3 (six line items, of which one is “private school tuition” in aggregate and one is the processing fee). 2024-25 report Appendix B, spring transactions only. The split between purchases, tuition and fee follows the source table's own expense-type column.

Every recipient, rankedpublished

1,971 named recipients sharing $43.2m. The median one received $483. The top 25 took half of it. Best Buy is the fourth-largest recipient of Education Freedom Account money in Arkansas, ahead of every private school but three.

2024-25 report Appendix B, pages 39–87, extracted programmatically. The 1,971 rows sum to $43,185,261 against the appendix's own printed total of $43,185,234 — a $27 gap consistent with whole-dollar rounding of 1,971 cent-level values, not a missed row. source rounding

The twenty largest, by name

The head of that curve, written out. These are the names the chart above can only place as points.

2024-25 report Appendix B, spring transactions only. Names appear exactly as printed in the data; display capitalization only is normalized. No attempt has been made to join these to the participating-schools table — several names differ between the two (“Baptist Prep”, “Central Arkansas Christian Schools, Inc.”) and a fuzzy join is known to mis-match Arkansas school names.

How much of a school is voucher-fundedpublished

At the average participating school, voucher students were 48% of enrollment in 2024-25. At some schools they are all of it.

2023-24 report Table 1 (97 schools); 2024-25 report Table 1 (126 schools). The mean of the 2024-25 column is 47.8%, matching the report's stated 48%. reconciles The 2023-24 equivalent does not reproduce: averaging that report's own 97 percentages gives 37.9% against its stated 39%. See reconciliation.

Growth, school by schoolpublished

The 68 of 126 schools whose names match exactly across both years, ranked by the number of voucher students they gained. Little Rock Christian Academy went from 360 to 637; Shiloh Christian from 233 to 488. The other 58 cannot be tracked at all.

Participant counts from Table 1 of each report. Only schools matching on an exact normalized name are shown — see the identifier problem below for why the others cannot be tracked.

The same schools, in dollars

Per-school dollars do not exist in the published record for 2023-24 at all, so growth in money received cannot be read off the reports. It can be computed: participants × that year's account value. Nothing here is a model — it is one published number multiplied by another, and the counts it multiplies are in the two columns beside it.

Assumes every participant carries the standard award. Former Succeed Scholarship participants carry 100% of the prior year's foundation funding rather than 90%, so these are slight under-estimates at schools serving them. The award is one statewide figure per year — $6,672, then $6,856 — so the dollar ranking is the student ranking; the quantity is here because the record does not hold it, not because it reorders anything. The cross-check that makes the arithmetic usable →

The school-identifier problemabsent

Schools appear by name only. No code, no ID, no address. That sounds like a technicality until you try to follow one school across two years of its own program.

Matching is exact on a normalized name (case, punctuation and spacing removed; the parenthesized city stripped, which is what makes campuses collapse). No fuzzy matching was attempted anywhere in this project: it is known to mis-match Arkansas school names, and an earlier trial produced a documented false match (“Saline County Christian School” to “Clover Community School”).

Assessment, school by schoolpublished

Average percentile on a nationally norm-referenced test, by school. Not the state test that public schools sit, so this cannot be compared with a public school's results — only with the national norm of 50.

2024-25 report Appendix C. 128 rows, of which 119 report a math percentile; the rest are printed N/A. Circle size is the school's voucher enrollment.

Retention, school by schoolpublished

Arkansas publishes per-school retention, which most voucher states do not. And this table is the only school-level 2025-26 data in the entire published record — it measures the 2024-25 → 2025-26 transition, not the year the report otherwise covers.

2024-25 report Table 2, pages 17–21. The overview above the table states “Entering its third year (i.e., 2025-26), the EFA program retained 91% of participating students from the previous year (i.e., 2024-25)”. The table carries 127 rows against 126 participating schools because the last row is a Homeschool aggregate, not a school — 126 + 1 = 127, exactly, and the aggregate is excluded from the chart above. The 2023-24 report carries its own Table 2 for the previous transition (2023-24 into 2024-25): the two are consecutive and non-overlapping, and must never be differenced against each other.

The 90% rule, and the gap it leaves familiesderived

An account is worth 90% of what the state spent per public school student the year before. That holds to the cent in all four years. A payment-platform fee is then withheld before the money reaches the family — the dashed line.

A statutory formula reproducing to the cent, four years running. The four bars are not transcribed from the reports — they are computed as 90% of the prior year's published foundation rate, and all four land exactly on the amounts the state published. That is the strongest evidence on this page that its arithmetic can be trusted: the chart demonstrates the statute rather than restating it. reconciles

What a voucher actually coversobtained

The state collects tuition on every school's application form and publishes none of it. The Arkansas Advocate requested those applications under freedom-of-information law and got them. At the average participating school, tuition runs about $9,800 against a 2026-27 account worth $7,346 — so the account covers roughly three-quarters of a year, and the family covers the rest.

Arkansas Advocate, 1 July 2026 (Antoinette Grajeda), from 170 application packets released by the Department under a state Freedom of Information Act request: “Data was collected from Educational Freedom Account program applications provided by the Arkansas Department of Education through a state Freedom of Information Act request.” This is the request that sentence describes — it does not cover the 2026-27 approval counts elsewhere on this page, which are reported, not obtained. The tuition figure is stated by the source as “about $9,800” and is drawn as an approximate band for that reason. The award amounts it is compared against are the state's own. See the absence register: tuition is a “collected, not published” row, and this chart is that row seen from the other side.

Which schools wanted inobtained

The same records show how the state handled the schools themselves. 176 private schools applied to take part in 2026-27; 164 were approved, 2 were denied and 10 were still under review when the records were released.

Same source and same request as above. No state publication carries these counts for 2026-27 — the coverage matrix marks the row obtained for that reason. As with student applications, nothing published explains either denial.

Seventeen different testspublished

Participating schools choose their own nationally norm-referenced test. Seventeen were used. The result published as the program's headline is an average across all of them.

2024-25 report Table 8. Completions sum to 12,218, matching the stated total. reconciles Names are shortened on the axis only; hover for the instrument exactly as printed.

The state test, at four schoolspublished

ATLAS is the assessment every Arkansas public school sits. Four participating private schools administered it to about 300 voucher students — the only directly comparable outcome data the program has produced in three years. They chose to administer it; the other 122 did not. Four self-selected schools are not a sample of 126, and nothing published says why these four opted in.

2024-25 report Table 11 and Appendix D. About 300 students at 4 of 126 participating schools. Clear Spring School did not administer the science assessment. Six of the fourteen published rows sum to 99 or 101, because the source rounds each of the four levels independently — the bars are drawn exactly as printed and are deliberately not normalized to 100, which would invent precision the source does not have. Rows that do not sum to 100 say so.

What exists, by yearabsent

The single most disorienting thing about this record is that it is not the same record from year to year. Twenty-six measures against four years: filled where published, half-filled where partial, violet where obtained under FOIA rather than published, empty where absent.

A measure counts as published only if it appears at the stated granularity in a report or appropriation act for that year. 2025-26 and 2026-27 are thin because no annual report exists for either; everything shown for them comes from appropriation acts, the 2024-25 report's forward-looking Outlook, its retention table — which reaches into 2025-26 — and, for 2026-27, records released under FOIA.

The absence registerabsent

Probes performed against the two annual reports, the DESE program pages, the Arkansas General Assembly's document server, and this project's own earlier audits. Every row carries the evidence that establishes it, of one of two kinds. 14 are searched — someone looked and did not find it, which is only ever true as of a moment, so those rows carry the date the probe was run. 3 are documentary — the report itself establishes them, by saying so or by demonstrably not containing the thing, so those rows carry a citation instead. A published PDF does not change, and dating a re-reading of one would record a search that adds nothing. “Obtained by” records a published instance of someone getting the data from the state anyway; it is evidence of releasability, not a claim about what should be released.

Completeness auditabsent

Every numbered table, figure and appendix in both reports, and whether this page uses it. This is how the claim “nothing has been missed” is made checkable rather than asserted.

Enumerated programmatically from the documents themselves, not from their contents pages. Objects marked “found in sweep” were missed by the first pass and located by a later completeness audit — including both sector spending tables and the 2023-24 report's own retention table, which turns out to cover the transition before the one the 2024-25 report measures.

Reconciliation

Every figure on this page that has a stated total in its source has been checked against it. The checks that fail are printed with the ones that pass.

The data

Arkansas publishes all of this as PDF tables and nothing else. A page that says so has an obligation to publish it properly, so every table behind every chart above is here, as CSV and as one workbook. They are generated by the same build that generates the page, so they cannot drift from what you see.

© 2026 Joel Rush · all rights reserved — permissions in the methodology. Read it first — in particular the three warnings about Appendix B covering one semester, the two retention tables measuring different transitions, and the absence of any persistent school identifier.