I also don't have a reference right now, but I would assume they've taken a thousand of random people and ran some statistical analysis on that sample.

Chances of someone from this thousand people being among the highest end are indeed pretty small.