Guides
How to read the federal vehicle data behind this site — including the places it is easy to draw a confident conclusion that the numbers do not support.
Reading owner complaint data without fooling yourself
NHTSA publishes millions of complaints filed by ordinary drivers. Counted properly they tell you what tends to go wrong with a particular car. Counted carelessly they tell you which cars sold well.
6 minute read
What a car costs to run, and what the sticker leaves out
The EPA's annual fuel figure is a good number built on assumptions that are probably not yours. What it assumes, how far real drivers land from it, and the costs it was never trying to measure.
6 minute read
MPGe is not MPG, and other ways electric car data misleads
A 2021 Model 3 reads 134 in the same database field where a Camry reads 32. They are not the same unit, they are not comparable, and neither is the number an EV buyer actually needs.
5 minute read
A recall and a complaint are not the same thing
One is a manufacturer admitting a defect and owing you a free repair. The other is a stranger filling in a form. They live in the same database and get quoted interchangeably, which serves nobody.
6 minute read
What a VIN actually tells you
Seventeen characters that identify one specific car rather than a class of them. One of those characters exists purely to catch you mistyping the other sixteen, and you can check it by hand.
6 minute read
How to read NHTSA crash ratings
Five stars is not five stars. The rating compares a vehicle against others of similar weight, so a five-star small car and a five-star large SUV did not survive the same test.
5 minute read
Every figure quoted in these guides was measured against the live federal APIs rather than taken from secondary sources, and the dates say when. Where a measurement contradicted something we had assumed, the guide says so — those are usually the useful parts.