What the numbers mean, and where they mislead

Counts are reports, not rates

This is the single most important thing on the site. A complaint count is how many people filed a complaint. It is not divided by how many cars were sold, so a popular model collects more complaints than a rare one that fails more often. A Camry will out-count a car that is ten times worse and sold a hundredth as many.

The comparison that survives this problem is between model years of the same car. Sales volumes are broadly similar year to year, so a year that stands out against its neighbours is telling you something real. That is what the model page is for.

Not all records carry the same weight

An owner complaint is unverified: NHTSA records what was submitted and does not investigate it. A technical service bulletin is repair guidance a manufacturer sent its dealers, and a high count often means a manufacturer documents thoroughly rather than that the car is bad. A defect investigation means a federal regulator decided a pattern was worth examining. A recall means the manufacturer conducted a remedy campaign. The last two are acts, not reports, and they carry the most weight.

Counting a complaint once

NHTSA publishes one row per complaint per component, so an owner who reported three problems in a single filing appears three times. We count each complaint identifier once, which is why our totals match NHTSA's own figures rather than exceeding them.

The component breakdown is the deliberate exception. It answers “how many complaints mention the brakes”, so a complaint about three components belongs in three rows there. Those rows will sum to more than the complaint total, and that is correct.

Matching records to a vehicle

Each source names vehicles in its own way, and none of them share an identifier. We match on make, model and model year after normalising the text, and we do not guess: a record we cannot place with confidence is kept in the raw data and left off the page rather than attached to a car it might not belong to. This is conservative by design, and it means a page can under-report rather than mislead.

Absence is not evidence

A model year with no crash rating usually means NHTSA never tested it, not that it did badly. No investigations is the normal case, not a clean bill of health. No fuel-economy figures before 1984 means the data does not exist, not that the car was unrated. Where we have nothing, the page says so rather than showing a zero that reads like a finding.

Freshness

Every source is re-checked daily against the publisher. Pages show what was in the most recent file we successfully loaded. NHTSA and the EPA each have their own publication lag, so a complaint filed this week will not appear here until it appears in their release.

Corrections

If a figure here disagrees with the federal record, the federal record is right and we want to know. Tell us which page and what you expected, and we will trace it back to the source file.