The short version

We publish a rating only when we have real data for at least two of three components. Where we don't, we say so and leave the gap visible. A large share of the profiles in this directory carry no rating at all, and that is the methodology working as intended rather than failing.

The three components

Component 1Academicstate assessment proficiency, percentile within state
Component 2Resourcingstudent:teacher ratio, percentile within state
Component 3Climatereported incidents per 1,000 students

The rating is the mean of whichever components have real data, expressed on a 1-10 scale. Every rating we show is labelled with how many components it rests on and the vintage of each.

Why percentiles are always within state

States use different assessments with different cut scores. A 60% proficiency rate in one state and a 60% rate in another are not the same measurement, and comparing them directly produces a number that looks precise and means nothing. Every percentile in this directory is calculated against other schools in the same state.

When we withhold a rating

  • Fewer than two components have data. One component is not a rating, however good that component looks.
  • Private schools. They do not participate in state assessment reporting or federal civil-rights data collection, so two of three components are structurally unavailable. We publish directory data and state the absence.
  • Sources contradict each other beyond reconciliation. Where two providers assess the same place and reach opposite conclusions, averaging them would manufacture false precision.
  • The underlying data is internally incoherent. We found published figures including a "1 in 0.0 chance nationally", a neighborhood rate computed over a 12-resident denominator, and a table showing a city with zero property crime and violent crime equal to its total. We do not build ratings on those.

Our sourcing standard

  • Every figure is traced to a named source with a stated vintage. Where enrollment is from one year and safety data from another, both years are shown.
  • Where sources disagree, we show the disagreement rather than picking the more flattering or more alarming number.
  • Where a figure can be checked against independent sources, we check it and say which one is wrong. Where it cannot, we present the conflict as the finding.
  • We never apply one place's data to a different place. Data for East Providence does not describe the East Side of Providence; Charleston does not describe South Charleston; Seattle's West Precinct does not describe West Seattle.
  • Suppressed data stays suppressed. Federal datasets mark figures that fail quality standards, are missing, or do not apply. None of those becomes a zero.
  • Gaps are stated, not filled. "Not available" is a finding.
On neighborhood crime indices. Several widely-cited commercial providers publish neighborhood crime grades. In the course of building this directory we documented roughly forty defects in them, including letter grades contradicting the same page's own index, two providers reporting a neighborhood's population differing by 67%, national-average baselines differing between providers by 45%, and a template error producing an impossible statistic across three different states.

They also measure the wrong thing. A crime index describes a ZIP code or neighborhood, not a school. Where we cite this material we name the provider and flag the defect; we do not build ratings on it.

On incident data

Where we report school-level incident figures, they carry these caveats, and they matter:

  • Federal collection is biennial and the most recent release covers a pandemic year with widespread remote instruction, so counts are likely depressed and not comparable to a normal year.
  • Figures are self-reported by school districts.
  • Small counts are suppressed to protect student privacy. An absent figure does not mean zero.
  • A school that records and reports diligently will show higher counts than one that under-reports. Higher numbers can indicate better reporting rather than a less safe school. For this reason we do not rank schools by raw incident count.

Coverage and its limits

This directory does not yet cover every school in the United States. There are roughly 130,000 K-12 schools nationally; the profiles here are a sample built to a documented standard rather than a complete listing. Each state page states how many sourced records it holds. Where a state shows one record, that is one school researched, not one school in the state.

Corrections

If a figure here is wrong, we would rather know. Every profile names its sources so any claim can be checked against them directly. Where a school's own records differ from the federal directory, the school's records are likely the more current, and federal directory data can lag by one or more collection years.