multiple factors based on data from many sources; essentially a Monte Carol analysis. It takes into account the uncertainties involved to try to determine a reality. This kind of statistical analysis is run by lots of businesses and data scientists so it isn't unknown, though I'm sure they account for different factors in different ways. 538 has been remarkably accurate actually and like good statisticians they caveat their analysis by pointing out that they aren't saying an underdog can't win. They are merely saying after running thousands of simulations what the approximate probability of the underdog winning is. Of course, like good modelers they take into account what really happened (like 2016) and where their model was off.

I think this is a pretty good write up. Despite all their efforts it is still just a forecast. When the chips fall they will calibrate again for the next cycle.