http://fivethirtyeight.com/features/a-users-guide-to-fivethi... is a decent overview of the most popular forecast. There's plenty of other relevant content on that site, too.
I didn't read what they said about their process but they were completely off too.
No, they predicted about a 70%:30% chance of Clinton vs. Trump win, not a 70% to 20% win by Clinton.
> I didn't read what they said about their process but they were completely off too.
It wasn't so much that they were completely off as that some people mistake a claim of probability > 50% with a claim of certainty.
Sure, the most-likely winner map was wrong in a number of states, but the reason 538 had a 30% chance of Trump winning rather than much lower chances given by other models was that 538 takes into account the fact that deviations in results from polling in different states tend to be correlated rather than independent.