Comment by criddell

> 15% of people are ALWAYS speeding, by definition

Not necessarily. If the data is heavily clustered around a single value, the percentile collapses to that number.