Methodology
A short, checkable method. Every number on the model pages can be rebuilt from the raw JSON saved with this prototype.
What we pull
For each model year from 2010 through 2024 we asked NHTSA for complaints and recalls by make, model, and year. We also paged the products catalog to find official model-name variants (CR-V HYBRID, F-150 SUPER CREW, ROGUE SELECT, and so on) and queried each of those names too.
The complaints API is not a clean exact-match. Querying F-150 and querying a cab style can return disjoint sets, and some cab-style names return identical sets. In at least one year (2021 Honda CR-V), the hybrid query returned a superset of the base-name results. We therefore merge by unique ODI number for complaints and by NHTSA campaign number for recalls. Identical records are counted once.
What stays out of the main score
Held out as related models, with their own counts listed on the page: Toyota RAV4 EV; Ford F-150 Lightning (BEV); Chevrolet Equinox EV; Nissan Rogue Sport. Hybrids and plug-in hybrids that are sold as the same nameplate (CR-V Hybrid, RAV4 Hybrid, RAV4 Prime, F-150 HEV) are included in the main union. Rogue Select, a prior-generation carryover, is included with the Rogue; in the years it appears, its records were a subset of the base Rogue query after deduping.
The score
For each model year, after the union step:
- 1 point per unique complaint
- 5 points per complaint with the crash flag set
- 5 points per complaint with the fire flag set
- 10 points per person listed in
numberOfInjuries - 50 points per person listed in
numberOfDeaths - 3 points per unique recall campaign
- 15 extra points per campaign flagged
parkItorparkOutSide
Age-adjusted score = severity score ÷ years on the market, where years on the market = max(1, 2026 − model year + 1). A 2010 has 17 years, a 2024 has 3. This only accounts for time to accumulate filings. It is not a rate per vehicle.
Years to avoid
Within one model, find the median age-adjusted score across 2010–2024. Flag a year Avoid if its age-adjusted score is at least 1.5 times that median and the year is one of the three highest age-adjusted scores. Flag Caution if the age-adjusted score is above the median but the year is not Avoid. Otherwise Typical. Years with fewer than 20 complaints get a limited-data mark in the table; none of the five prototype models had a year that thin in the core set.
The rule is deliberately strict, so a model with a fairly even record may have only one Avoid year. The raw complaint bar chart is there so a high-count older year (for example the 2013 F-150) is still obvious even when age adjustment ranks a newer year higher.
Limitations
- No sales-volume adjustment. We do not have a sourced figure for how many of each year were sold. A best-selling truck will usually have more complaints than a slower-selling car even if the rate is lower. We would only add a per-vehicle rate if we obtained a cited sales series.
- Time on the market is a calendar proxy. Model years often go on sale in the previous calendar year, and some years were sold longer than others. We did not invent a more precise on-sale window.
- Complaints are unverified consumer filings. NHTSA uses them as a signal, not as a verdict. Death and injury counts are as reported on the complaint, not a confirmed census.
- Newer years are still accumulating filings. A 2024 can look quieter in raw counts and louder after age adjustment. Both views are shown.
- Model-name quirks in the API can hide or duplicate records if you query only the base name. This prototype unions variants, but the catalog is not perfect, and recalls for an EV variant may be filed under the base name.
- No VIN-level repair status. The public vehicle API does not reliably say whether a particular VIN still has an open, unrepaired recall. That check belongs on NHTSA’s own VIN tool.
Source
NHTSA Datasets and APIs. Endpoints used are documented in data-notes.md in the project folder. Bulk flat files at static.nhtsa.gov exist and were consulted for page-count estimates; the prototype scores themselves come from the live APIs.