How we built the school lead testing database

This is the methodology behind a survey of all 51 US jurisdictions — the 50 states and the District of Columbia — on a single question: what does each of them publish about lead in school drinking water? Every number on this page is computed from that survey when the page is built. If the data moves and this page does not, that is a bug, not an editorial decision.

The most likely way a project like this fails is not missing data. It is publishing a comparison that does not hold. Every rule below came out of something we hit while doing the work — a state whose action level is a laboratory limit rather than a health threshold, a dataset that is comprehensive and seven years old, a portal that answers but holds nothing — and not out of a checklist written beforehand.

The survey, in numbers

Jurisdictions surveyed: 51 — the 50 states and the District of Columbia, each checked individually rather than compiled from a single list.
Publish results: 43 of 51, up from 26 in the 2021 baseline, where 9 were recorded as not publishing and 16 were unresolved.
Publish nothing we could find: 7 — 4 of them confirmed at source (Georgia, Illinois, North Dakota and South Dakota), and 3 resting on second-hand evidence only (Alaska, Connecticut and Missouri).
Publishing status unresolved: 1 — Kentucky, left open deliberately rather than guessed. It is counted neither among the jurisdictions that publish nor among those that do not.
Evidence: 36 rows confirmed at source, 15 resting on a credible third-party report we have not yet confirmed. The second group is never presented as verified.
Changed since 2021: 28 entries moved against the 2021 baseline: 23 on whether results are published, 11 on the action level, and 4 on whether testing is required. A value that did not move is a finding too.
Last re-verified: 51 of 51 rows carry the date they were last checked; the most recent is 2026-08-09. We re-verify the survey quarterly, and each jurisdiction’s page shows its own stamp.

The governing rule

Jurisdictions are not comparable by default. Every cross-state number must name the axis it compares on, and every axis has at least one jurisdiction that breaks it.

We publish jurisdiction-level facts freely. We publish a cross-state comparison only where all four of the axes below are equal, or explicitly normalised and stated as such.

The four axes every figure carries

A figure about school lead testing means nothing without four things travelling beside it: the threshold it was counted against, the unit it counts, how much of the jurisdiction it covers, and when it was collected. Drop any one and the number is not wrong so much as unreadable.

1. Action level — the threshold is not a constant

Jurisdictions publish against thresholds from 1 ppb to 20 ppb — a 20-fold spread. A share of results “above the action level” is therefore a measure of the threshold as much as of the water, and ranking jurisdictions on it would mostly rank their thresholds.

Published action levels, and the jurisdictions using each
Action level	Count	Jurisdictions
1 and 15 ppb	1	Rhode Island
4 ppb	2	Maine, Vermont
5 ppb	9	District of Columbia, Illinois, Maryland, Michigan, Minnesota, Montana, New Hampshire, New York, Washington
10 ppb	1	North Carolina
15 ppb	25	Alabama, Arizona, Arkansas, California, Florida, Hawaii, Indiana, Iowa, Kansas, Louisiana, Nebraska, Nevada, New Jersey, New Mexico, North Dakota, Oregon, Pennsylvania, South Carolina, South Dakota, Texas, Utah, Virginia, West Virginia, Wisconsin, Wyoming
20 ppb	2	Idaho, Tennessee
No level of its own	2	Georgia, Ohio
Not recorded by this survey	9	Alaska, Colorado, Connecticut, Delaware, Kentucky, Massachusetts, Mississippi, Missouri, Oklahoma

2 of the rows with no level in the table above — Massachusetts and Oklahoma — do have one recorded in their research note, which we have not yet promoted into the structured column. That is a gap in our encoding, not a statement that those jurisdictions publish no threshold.

Store each jurisdiction’s own action level alongside every result, and never quietly coerce results to a common threshold. Where a comparable figure is genuinely needed, recompute it from raw ppb at a stated common threshold and say which one.

Two jurisdictions can also share a number for different reasons. Montana’s 5 ppb is the practical quantitation level — the lowest concentration a laboratory can reliably measure — while New York’s 5 ppb is statutory. They are not the same policy and we do not describe them as one. The widely-quoted Michigan figure of 12 ppb is a third trap: it is the community water supply action level, not a school standard, and EGLE’s school recommendation is 5 ppb.

2. Unit of analysis — fixtures, schools, districts

The same dataset yields wildly different headlines depending on what is being counted. Our Rhode Island row records the illustration that produced this rule, and it is published here as we wrote it, with the source’s own figures rather than any rate of ours:

nearly 80% of FIXTURES detected no lead and under 4% exceeded 10 ppb, but about 70% of SCHOOLS had lead detected in at least one fixture and 21% had high levels in at least one. Same data, wildly different headline depending on the unit of analysis. Our methodology page must state which unit every figure uses.

Every published figure states its unit — fixtures, outlets, buildings, schools, districts or LEAs — and no comparison mixes two of them. Where both are available we publish the fixture rate and the share of schools with at least one exceedance, because the gap between them is itself informative.

Denominators hide the same trap. Pennsylvania has 500 local school districts but 795 LEAs once charter LEAs, intermediate units and state agencies are counted (NCES Common Core of Data, 2023) — 295 entities of difference, and any share computed from the wrong one moves accordingly. We match reporting entities to the NCES universe and state which agency types are included.

3. Coverage — “publishes” is not a quality bar

Of the 43 jurisdictions that publish results, 8 are comprehensive, 26 are partial, 2 are negligible — so thin that the disclosure tells a parent almost nothing — and 7 have not yet been rated for breadth by this survey. A yes-or-no column would file all 43 of them identically.

Carry a coverage rating on every jurisdiction, and never render a publishing status without it. There is no component in this section that can print “publishes results” without also printing the coverage, the vintage and the evidence grade beside it.

Most jurisdiction programmes are voluntary, and a voluntary dataset is a floor, never an estimate of how common lead is: schools that test are self-selected, and schools that both test and report are doubly so. Nebraska’s own page says plainly that a facility missing from its list may simply never have tested. Every figure we derive from a voluntary programme carries that caveat inside the same block as the number, never as a footnote.

4. Vintage — currency is independent of breadth

A jurisdiction can publish comprehensively and still tell a parent nothing about today. Of the 43 publishers, 18 are running programmes or actively refreshed datasets, 3 are closed, 1 is a snapshot the state itself flags as outdated, and for 21 we have not established currency at all — a gap we set out in full further down this page.

Arizona — programme closed. The survey records its vintage as “2017 one-time screening; ADEQ results URL now 404”.
California — programme closed. The survey records its vintage as “2018-2019 mandate; program CLOSED 2019-07-01”.
Maine — programme closed. The survey records its vintage as “2021-2023 cycle, closed”.
Oregon — flagged outdated. The survey records its vintage as “2016 only — state flags it as outdated”.

Carry the vintage on every jurisdiction, and never describe one as monitoring on the strength of a programme that has ended. On any map, table or ranking, the vintage renders next to the value.

The evidence standard

Every row carries an evidence grade. Primary means we confirmed it against the jurisdiction’s own page or dataset. Secondary means a credible third party reported it and we have not yet confirmed it at source. The current split is 36 primary and 15 secondary, and a secondary row is never presented as verified — its badge reads “reported, not verified at source” wherever it appears.

A negative claim requires primary evidence. “This jurisdiction does not publish results” is a claim about absence, and a search that found nothing is not evidence of one.

That rule is not theoretical; it caught us twice. Ohio was recorded as not publishing after a search of its environmental agency’s drinking-water pages — Ohio runs a public results dashboard that simply is not linked from them. Texas was recorded as publishing a participant roster only, and publishes full results at a separate address. Both errors had the same shape: treating absence in the place we looked as absence everywhere.

So before any “does not publish” claim we check the jurisdiction’s environmental agency, its health agency, its education agency, and EPA’s index of state grant-programme contacts, which is the authoritative index of state results portals. Results often live under a different agency than expected — Arizona’s under the Department of Health Services rather than the environmental agency, Arkansas’s under the Department of Education.

Of the 7 jurisdictions we record as publishing nothing, 4 are confirmed at source — Georgia, Illinois, North Dakota and South Dakota — and 3 — Alaska, Connecticut and Missouri — rest on EPA’s “coming soon” marker, which means the jurisdiction has no grant-funded results portal rather than that it publishes nothing anywhere. We publish the two groups as two groups, never as one number.

A portal that responds is not a portal that holds anything. Confirmed at source and filed as publishing nothing: Georgia, whose results dashboard is live and lists facilities, not one of which carries a result. Checking that a URL returns a page is not checking that the page contains data, and we grade on the second.

The transparency tier rubric

A reader collapses a 51-row table to one verdict per jurisdiction whether we offer one or not. The tier is us doing that collapse by a published rule instead of leaving it to be done badly from a yes-or-no column. The rules below are the code: the page renders the same text the function that assigns tiers is written against, and the counts beside them come from running that function over all 51 rows.

A tier never travels alone. Every surface that shows one shows the coverage, vintage and evidence it was derived from in the same block.

Comprehensive and current

6 of 51 jurisdictions

The jurisdiction publishes results covering essentially all of its schools, and the programme producing them is still running. This is the only tier where a parent checking today learns something about today.

Comprehensive but closed

2 of 51 jurisdictions

Coverage is essentially statewide, but the programme that produced the data has ended or the state itself flags the published dataset as outdated. The numbers are real and they describe the year they were collected — not now. California's mandate expired on 1 July 2019 and its database is closed to new submissions; Maine's sampling cycle ended in 2023.

Comprehensive, currency not established

0 of 51 jurisdictions

The jurisdiction publishes results covering essentially every school, but we have not established how current that dataset is. The open question is in our research, not in the state's disclosure — which is why these rows are not filed as unresolved. No jurisdiction sits in this tier today; it exists so that rating a state's breadth before establishing its currency cannot file a confirmed publisher as unresolved.

Partial

26 of 51 jurisdictions

Results are published for some schools only — a voluntary programme, a grant round, a rolling testing cycle, or a selected set of facilities. Most state programmes are voluntary, so a school's absence from the data means nobody reported a test for it, never that its water is safe.

Negligible

2 of 51 jurisdictions

Results are published, but for so few schools that the disclosure tells a parent almost nothing. Iowa's report covers four organisations and 28 buildings in a state with roughly 1,300 public schools; counting that as 'publishes results' without saying so would flatter it against New York's 4,489-school dataset.

Publishes results, breadth not yet rated

7 of 51 jurisdictions

The jurisdiction publishes results, but our survey has not yet rated how much of the state that publication covers. The open question is in our research, not in the state's disclosure, which is why these rows are not filed as unresolved: that the jurisdiction publishes is settled, only the breadth is not. New York sits in this tier with a 4,489-school dataset — an unrated breadth must never be read as a thin one.

Publishes nothing (confirmed at source)

4 of 51 jurisdictions

We looked at the jurisdiction's own pages or data platform and found no published results. A claim that a state publishes nothing is a claim about absence, so we make it only on primary evidence. A portal that responds but contains no results counts as publishing nothing: Georgia's dashboard is live and lists 42 facilities, none of which carries any result.

Publishes nothing (not confirmed at source)

3 of 51 jurisdictions

The only evidence that the jurisdiction publishes nothing is second-hand — typically EPA's 'coming soon' marker, which means the state has no WIIN-grant results portal, not that it publishes nothing anywhere. We report it as unconfirmed rather than proven, because searching and not finding is not evidence: that error put Ohio and Texas in this column once, and both publish.

Unresolved

1 of 51 jurisdictions

Whether the jurisdiction publishes results at all is unresolved, and we declined to guess. Kentucky is the only one: EPA marks its programme 'coming soon', while an ArcGIS layer exists that we could not confirm carries any results. A jurisdiction we know publishes is never filed here no matter how little we know about its breadth or its currency — each of those gaps has a tier of its own.

Non-detects and the denominator

A non-detect — a result published as “<5”, “<0.5” or “ND” — is a censored observation: the laboratory could not measure a concentration below its reporting limit. It is not a zero and it is not the limit. Recording it as either would not change any single result much; it would change the denominator, and the denominator is what decides the headline.

Here is the same count of exceedances divided two defensible ways, in two of the jurisdictions whose published rows we hold in full:

California

41,628 of 43,978 samples came back below the reporting limit, leaving 2,350 with a measured value. 517 of them were at or above 15 ppb.

1.2% (517 of 43,978 samples collected) — fixtures at or above 15 ppb, the state's action level
22.0% (517 of 2,350 measured samples) — fixtures at or above 15 ppb, the state's action level

Same numerator, two denominators: the measured-only figure is 18.7 times the all-samples figure. Both are defensible. Publishing either without saying which is not.

Washington

7,673 of 20,366 samples came back below the reporting limit, leaving 12,693 with a measured value. 4,531 of them were at or above 5 ppb.

22.2% (4,531 of 20,366 samples collected) — fixtures at or above 5 ppb, the state's action level
35.7% (4,531 of 12,693 measured samples) — fixtures at or above 5 ppb, the state's action level

Same numerator, two denominators: the measured-only figure is 1.6 times the all-samples figure. Both are defensible. Publishing either without saying which is not.

Every rate states its denominator, and we default to all samples collected — excluding non-detects inflates the rate and flatters the jurisdictions honest enough to report them. Non-detects are stored as their own flag and never written into the result column.

Reporting floors and schema breaks

Where a jurisdiction publishes results only for the taps that exceeded a threshold, a low count can mean clean water or it can mean a high reporting floor, and the number of tests performed is invisible. The denominator a rate would need does not exist at all, so there is no rate to compute — only a count of exceedances, which is a different thing and has to be labelled as one.

Record whether a jurisdiction publishes all results or exceedances only, and never compute a share above a threshold from an exceedances-only dataset. Never compute a trend across a known schema break or a change in a reporting threshold either.

Pennsylvania is the live example of both at once — its reporting threshold and its file schema changed in the same year. Our note on it, verbatim:

CHANGED 2026: July 2026 budget raised the REPORTING threshold from >0 ppb (Act 39 of 2018) to >=5 ppb. The xlsx schema also changed in May 2026, breaking the time series. Act 39 does not require testing - schools may opt out by discussing lead at an annual public meeting.

The mirror-image trap is a falling count read as improvement. Wisconsin’s statewide count of lead service lines fell from 256,000 to 180,000 largely because unknown pipes were identified, not replaced. Any year-over-year change is checked for reclassification before it is described as progress.

What we have not done yet

Two gaps in this survey are ours rather than the jurisdictions’. Neither changes a rating today, and both would be easy to leave in a drawer — which is the reason to put them here.

Data currency, unestablished for 29 rows

29 of the 51 rows record the currency of the published data as unestablished. 8 of those have nothing to date: the 7 jurisdictions that publish no results, plus Kentucky, whose publishing status is itself unresolved. The remaining 21 do publish. Our survey located and graded a results surface for 21 of them, so this is not data we could not find — it is data we did not date.

It moves no rating today. Currency only changes a tier for a jurisdiction already rated comprehensive, and of the 8 rows rated comprehensive, 0 carry an unestablished vintage. Were there one, the rubric would file it in its own tier — comprehensive breadth, currency not established — rather than as current, and never as unresolved, which would report a confirmed publisher as though we did not know whether it publishes at all. But this page commits us to re-verifying quarterly, and a commitment made with 21 publishers’ currency unestablished is worth stating plainly rather than leaving for a reader to find. Closing them is the first job of the next pass.

Breadth, unrated for 7 publishers

Indiana, Michigan, Minnesota, New York, Pennsylvania, Vermont and Washington all publish results that this survey has not rated for breadth. They sit in their own tier above, and the open question there is in our research rather than in their disclosure — which is why they are not filed as unresolved alongside the jurisdiction whose publishing status we could not settle.

New York is the clearest reason the distinction matters: we hold 4,489 of its school reports row by row, and an unrated breadth read as a thin one would get that exactly backwards. Rating these 7 needs fresh confirmation at source for most of them, so they stay unrated and say so, rather than taking a rating we have not earned.

What we will not do

Each of these is a comparison a reader could reasonably want, and each is one the data cannot support. They are listed so that we can be held to them.

Rank jurisdictions on the share of results above the action level, when those action levels differ.
Present a publishes-yes-or-no scorecard without the coverage and the vintage beside it.
Describe the results of a voluntary programme as an estimate of how common lead is.
Compute a trend across a known schema break or a change in a reporting threshold.
Publish a claim that a jurisdiction does not publish results on second-hand evidence alone.
Name a school without the state's remediation note where one exists.

Cite this page

TapWaterData Editorial. "Methodology: national school drinking water lead testing survey." TapWaterData national school drinking-water lead testing database. Last updated 2026-08-09. Licensed under CC BY 4.0. https://www.tapwaterdata.com/school-lead-testing/methodology

@misc{tapwaterdata_school_lead_2026,
  author       = {TapWaterData Editorial},
  title        = {Methodology: national school drinking water lead testing survey},
  year         = {2026},
  howpublished = {\url{https://www.tapwaterdata.com/school-lead-testing/methodology}},
  note         = {Last updated 2026-08-09. Licensed under CC BY 4.0}
}

Published under a Creative Commons Attribution 4.0 (CC BY 4.0) licence — free to republish, chart or build on with attribution. Last updated 2026-08-09.

All 51 jurisdictions — the scorecard, the tier map and the per-jurisdiction pages every rule on this page is applied to.