Everything here can be recomputed from public files. If a number cannot be reproduced, it should not be trusted, including ours.
The three corrections
Raw complaint counts mislead in three ways, and all three are fixed here.
Age
A 2012 car has had thirteen years to collect complaints and a 2023
car has had two, so raw totals rank by age. Only complaints filed in each model
year’s first three years on the road are counted, which makes model
years comparable and is immune to later recall-driven bursts.
Volume
Counts scale with how many were sold. The CR-V sold 66,752 in
1997 and 403,768 in 2025. Complaints are divided by US sales, using the mean of
the model year’s own calendar year and the one before, because a model year
sells across two. Launch years use their own calendar year alone.
Model names
NHTSA’s model string is not stable. The 2014 Camry
files 466 complaints under “CAMRY HV” and 14 under “CAMRY”;
2023 Civics file under “CIVIC SEDAN”. Every string a make used in a
year is gathered and deduplicated by complaint ID, because complaint IDs repeat
once per affected component and counting rows overstates complaints by about 39%.
What is not done
No single composite score. Two cars can rank in opposite orders depending on
whether you weight severity, so one number would hide the choice rather than
inform it.
No ranking on a handful of complaints. Rankings use a 95% lower confidence
bound, so a model year cannot top a list on nine reports.
No claim that a vehicle is defective. These are filings by owners.
What this cannot tell you
It sees safety complaints only. Alternators, air conditioning and
infotainment are under-represented unless they became a safety issue. For repair
frequency across everything, owner-survey sources such as TrueDelta and Consumer
Reports measure something this cannot.
Filing propensity is mixed into every rate. An engaged owner community files
more per failure, and that cannot be separated from the failure rate itself.
Mileage data thins out badly after 2021, when NHTSA largely stopped recording
it: about 82% of 2020 filings carry a mileage reading against roughly 19% since
2022.
Only a minority of denominators have been checked against an external
published figure. Each page says which kind it rests on.
Sources
NHTSA complaints (FLAT_CMPL), investigations (FLAT_INV) and the
recalls API, all public and refreshed daily. US sales from Wikipedia tables
citing manufacturers’ own releases. Vehicle details decoded from the VIN
through NHTSA’s free vPIC service.