A confidence interval shows how much a figure could vary by chance: with 95% confidence the true value lies inside it. Caropen uses the Wilson interval, which behaves well for small samples and for values near 0 or 100%. Charts show it as a shaded band or as whiskers.
Why do some models have a wide interval?
The fewer inspections (n), the wider the interval. A cohort of 100 inspections can be ±10 percentage points, one of 10,000 less than ±1.
How do I compare two models?
If their intervals overlap a lot, the difference may be chance. That is why the rankings always show the interval and n next to the rank, and model pages – for example BMW 3 Series – show an interval for every year of manufacture.
Where is the formula?
In the methodology. The figure itself is explained under first-time pass rate.