Safety analysis: methodology and limitations

Sources

Per-accident records come from the NTSB Aviation Accident Database. Fleet and exposure inputs come from the FAA Aircraft Registry. Both get pulled on a biweekly cadence, slowly enough not to strain either source.

Matching NTSB accidents to FAA models

Each NTSB airframe record carries a free-text make/model and usually a tail number. Reconciling those against the FAA registry takes work, especially where typos or shorthand crept into the original record. We use both name matching and tail-number cross-reference to settle each match, and we flag the row as provisional where the two signals don't both agree. Amateur-built airframes get rolled up to the manufacturer rather than to a model (see Exclusions below).

Reporting-lag correction on trailing years

The trend chart's most recent years come in under-reported, because the NTSB takes anywhere from months to a couple of years to finalise an event. We apply a per-year correction derived from older, stable years and cap it so a near-empty recent year doesn't blow up into a hockey stick. Where there isn't enough stable history to anchor the correction, we drop it entirely rather than guess.

Observed counts are always available in the chart's accessibility description, even when the visual shows the adjusted value. The hatched overlay is the correction. The solid bars are the underlying data.

Exposure (the rate denominator)

Comparative accident rates are expressed per 1,000 aircraft-years of FAA registration within the 2008–present window. Note that this isn't flight hours; we have no per-tail flight-hours feed we trust as a primary source. A model that is flown more per airframe than its peers will look like it has a higher accident rate even if its per-hour safety is average.

Registry rows with missing entry dates or missing deregistration exits are ambiguous. We publish a confidence band alongside the central estimate, and where that band is wide enough to matter a "± X yrs" note shows up next to the exposure-years figure on the model page. Modern, well-documented types are usually fine. Vintage and niche models are often where the uncertainty shows up.

Peer-cohort comparison

Each model is grouped into a peer cohort sharing aircraft category, propulsion class, and capacity range. The cohort-share comparison only renders when the cohort is large enough to be meaningful; very small cohorts get dominated by single outliers, so they don't get a comparison at all.

The comparison isn't flight-hours adjusted (see the disclaimer on the anchor itself). A model flown more per airframe than its peers will show an accident share above its exposure share even if its per-hour safety is average.

Exclusions

Amateur-built and kit aircraft generally have no FAA type-certificate model code, so there's no canonical per-model page to publish their accidents on. Where we can, we roll the accidents up to the manufacturer instead so the data still gets disclosed somewhere. We just don't publish a separate page per kit model.

One-off builds whose manufacturer name doesn't canonicalise stay unmatched. They get disclosed here but never get a page.

Things this page doesn't do

It doesn't issue a safety verdict, an overall rating, or a peer rank, and the bars aren't coloured red and green to imply one. The page shows raw counts and structured aggregates without interpreting them. NTSB probable-cause and narrative text isn't ingested at all, so we don't surface it anywhere either.

Last updated

Dataset as of May 1, 2026. The pipeline re-runs against the latest NTSB and FAA exports every two weeks. Numbers on this page are computed live from the read-model tables and reflect the most recent run.