What we got wrong, and when we fixed it

We retract the “$1.6 billion of debt with no debt levy” story. It was our error, not theirs.

This page carried an open question we thought might be the best story on the site: 38 Missouri districts holding $1.6 billion of long-term debt while levying nothing to repay it — 21 of them wealthy suburbs, the list running through Parkway, Ladue, Lindbergh, Hazelwood and Webster Groves. We said we could not explain it. We went looking for the mechanism.

There was no mechanism. The data was wrong, and it was our data. Those 21 St. Louis County districts levy a debt-service tax every year, exactly as anyone would expect. Ladue levies 86¢. Hazelwood levies $1.24. Normandy levies $1.80. The State Auditor’s machine-readable export — the file we built from — records $0.0000 for St. Louis County districts in most years. The Auditor’s own printed report prints the true rate on the page. We had been reading a gap in a spreadsheet as a fact about school finance, and we built a story on it.

What should have given it away. The pattern was too clean. All 21 districts appeared to stop levying in the same year and resume in the same year — Normandy, one of the poorest districts in Missouri, moving in perfect lockstep with Ladue, one of the richest. Twenty-one school boards do not coordinate like that. We wrote it up as a mystery when we should have read it as a symptom. Real behaviour is messy. Only bad data is tidy.

Every affected figure is now taken from the Auditor’s printed report and reconciled against DESE’s independent record of the same levies: 20 of the 21 agree to the cent (Maplewood-Richmond Heights cut its rate from $1.35 to $0.85 between the two years, which both sources confirm).

Districts carrying a debt-service levy: 306 → 327 Median total school levy: $4.08 → $4.12 Districts with debt but no debt levy: 38 → 17 Debt they hold: $1.6 billion → $10.6 million Suburbs among them: 21 → 0

What is actually there, now the error is gone, is small and dull. Seventeen districts — sixteen rural, one a town, not one a suburb — carry long-term debt with no debt-service levy. They hold $10.6 million between them: about one-tenth of one percent of Missouri’s school debt, where we had claimed 17% of it. The largest owes $2.7 million and nine owe under $200,000. They are almost certainly small lease-purchase and equipment obligations paid out of operating funds because they are too small to warrant a dedicated tax. That is an unremarkable answer and it is the true one.

This one stings, because the false version was the more interesting version. That is exactly why it is here.

A sweep of every published figure, after the levy rebuild — eight more errors

Rebuilding the operating levy (below) changed a file that a lot of other numbers quietly depended on. Rather than assume the rest of the site was fine, we re-derived every published figure from the source data. Eight were wrong. Some were casualties of the rebuild; several had been wrong for weeks and had nothing to do with it.

Home page, what a dime raises: $420 vs $67 → $716 vs $39 Median assessed value per student: $130,711 → $130,885 Districts levying exactly $2.75: 65 → 64 Districts that have never won a levy election: 500 → 503 Gain per student, richest quartile, if all reached $3.43: $1,058 → $813 Poorest quartile below the performance levy: 53% → 54% Richest quartile below it: 38% → 35% Correlation, property wealth and local funding share: 0.73 → 0.71

The one that mattered most was an explanation, not a number. The Methods page said we had once used a DESE tax-rate field that was “the ceiling rather than the levied rate,” and that the two differ in 72% of districts “because most districts do not levy their full ceiling.” That reasoning was wrong, and it contradicted our own headline finding — we say elsewhere, correctly, that the overwhelming majority of districts are at their ceiling. Re-examined: 89% levy their ceiling to the cent. The real fault in that DESE field is vintage — DESE labels its files by school year, so its “2025” figure is tax year 2024. It matches the Auditor’s 2024 ceiling for 99% of districts, and differs from the 2025 levied rate for 76% of them because 2025 was a reassessment year and Hancock rollbacks moved nearly every rate. The Methods page now says so.

We are listing the small ones alongside the large. A site that only corrects its embarrassing errors is not keeping a record; it is curating one.

We used the wrong tax rate for 21 St. Louis County districts — and it moved a headline number

Twenty-one districts in St. Louis County do something unusual: they levy a different rate on each class of property — one rate on homes, another on farmland, another on business property, another on cars and equipment. Everywhere else in Missouri a district has a single operating rate.

We published the residential rate as those districts’ operating levy. For a map of what a homeowner pays, that is the right number, and it is the number we will keep showing homeowners. But it is the wrong number for the $3.43 performance-levy test. RSMo 163.011(12) defines a district’s operating levy as the levy of the district — not the rate borne by any one class of property. The district’s levy is its subclass rates weighted by the assessed value in each class, and for these 21 districts it is materially higher than the residential rate. Parkway’s operating levy is $3.14, not the $2.67 we published.

How we caught it: a stray flag in DESE’s levy-by-fund file did not agree with our count of districts at their tax rate ceiling. Chasing it turned up the real problem. We rebuilt every district’s operating levy from the Auditor’s own subclass rates and assessed values, then checked the result against DESE’s independently published operating rate: it now reproduces DESE exactly for all 516 districts in tax year 2024, and for 6,183 of 6,192 district-years (99.85%) across 2013–2024. That check also surfaced two things our earlier parse had missed — voter-approved temporary operating levies (33 districts) and Kansas City’s Article X levy, without which Kansas City read $2.07 instead of $4.86.

Districts below the $3.43 performance levy: 216 of 513 → 211 of 513 Children in them: ~347,000 → ~319,000 Of those, already at their ceiling: 188 (87%) → 194 (92%) Median operating levy: $3.57 → $3.60

What it did not touch: 495 of Missouri’s 516 districts were unaffected. Five districts leave the below-$3.43 group — Ferguson-Florissant, Hancock Place, Mehlville, Pattonville and University City. The central finding got stronger, not weaker (92% of these districts are already taxing at their legal maximum), which is not why we changed it. We also had to retire the Parkway/Wheaton comparison that ran on the old figures; it has been replaced by a cleaner one that holds the tax rate perfectly still — 64 districts levy the identical $2.75, and among them a dime raises $57 per student in Thayer R-II and $702 in Pettis Co. R-XII.

A dead end we also chased: the statute’s wording (“teachers’ and incidental funds”) reads as though the capital-projects fund should be excluded from the operating levy, which would have pushed the count to 239. It should not be excluded — five districts land on exactly $2.75, the state-aid floor, only when capital projects is counted. We tested the narrower reading and rejected it. The full reasoning is on the Methods page.

Debt-service levies were wrong for 74 districts

Our first pass at the debt-service levy was read out of the State Auditor’s printed PDF, and the parser carried a fault. In the Auditor’s layout a taxing subdivision’s name is printed only on its first row; continuation rows are blank. Our parser only updated the “current subdivision” on rows whose purpose it recognized as school-like — so a blank-named debt-service row belonging to a neighboring city or fire district could be silently added to the last school district it had seen. The City of Savannah’s 9¢ debt levy, for instance, was landing on Adair County R-II.

How we caught it: we pulled the Auditor’s own machine-readable tax-rate export and found it disagreed with us for 74 districts. A corrected re-parse of the PDF then matched the export on 420 of 420 districts we could check — confirming the export was right and we had been wrong.

Districts carrying a debt-service levy: 348 → 306 Median total school levy: $4.17 → $4.07

What it did not touch: operating levies were unaffected — they matched the Auditor’s export for all 516 districts — so the performance-levy analysis and the levy-election findings stand unchanged. Files corrected: the total-levy map, the district lookup cards, and the Funding page.

A statistic on the Methods page was computed on corrupted data

While building the multi-year levy history we discovered that the State Auditor’s 2024 export had been downloaded twice and both copies ingested — the export filenames do not contain the year, so the duplicate went unnoticed. Every 2024 levy, ceiling and debt figure in our working file was therefore exactly double its true value.

Almost nothing published rested on the 2024 figures, so the site was unaffected — with one exception. The Methods page cited the number of districts that a naive “a rise in the tax-rate ceiling means the district won an election” rule would wrongly flag. That figure was computed on the doubled data.

Districts wrongly flagged by the ceiling rule, 2024–25: 88 → 130

The underlying point is unchanged and, if anything, stronger: the rule flags 130 districts as having won a levy election in 2024–25, when the true number is zero. It is not used anywhere on this site.

A test we ran, lost, and are publishing anyway

We say plainly on this site that the free/reduced-lunch rate is no longer a usable poverty measure. So the obvious next step was to rebuild the Beating the Odds regression on Census child-poverty data instead. We tried it, and it made the map worse.

We pulled SAIPE child poverty for every one of the eleven years the achievement data covers (2009–2019), so the two measures describe the same era. Free/reduced lunch explains 42% of the variance in district achievement; child poverty explains 21%. And the weaker control does real damage: with child poverty alone, Ladue and Clayton climb into the state’s top over-performers — not because they are doing more with less, but because a model that under-controls for advantage hands the advantaged the credit.

Outcome: no change to the map. It keeps the free/reduced-lunch model, using the 2009–2019 average — the window before community eligibility spread widely enough to break the measure. We are recording the failed test rather than burying it, because the intuition behind it was right and someone else will have it too. The reasoning is on the Methods page.