Well, the problem is, as 538 will tell you, is that all the polls have been wrong in the same direction before. Repeatedly. (The happy side of this is that it's also possible there's still the Clinton landslide we need.)
Say I told you exactly how each ethnic group (perhaps further subdivided by age & gender) will vote by state - whites will be 56% Trump, blacks will be 97% Clinton, Hispanics will be 70% Clinton, whatever. You still have to guess the turnout-by-group, which is notoriously way harder to poll than how any particular group feels. That's a case where response bias is huge and crippling because the kind of people who respond to polls and the kind of people who are marginal voters who might show up, might not don't have much overlap.
That said, the interesting thing for me is...
http://fivethirtyeight.com/features/final-election-update-theres-a-wide-range-of-outcomes-and-most-of-them-come-up-clinton/FINAL NATIONAL POLLING AVERAGE
YEAR DEM CANDIDATE GOP CANDIDATE UNDECIDED/OTHER
2000 Gore 43.6% Bush 46.8% Undecided/other 9.6%
2004 Kerry 47.4 Bush 48.9 Undecided/other 3.7
2008 Obama 52.1 McCain 44.5 Undecided/other 3.4
2012 Obama 48.8 Romney 48.1 Undecided/other 3.1
2016 Clinton 45.7 Trump 41.8 Undecided/other
12.5WTF at 12.5% of voters being undecided (which is really why 538 threw their hands up in the air and said "I dunno"). How can you possibly still be undecided this late? My personal *hope* is that "undecided" is code for "I don't like either candidate, and I have a preferred candidate, but I don't want to say I'm a supporter of them." Or alternatively "I'm not really voting." If that's the case, it's unlikely undecideds will hugely break for Trump, which is part of why the model is unsure about what is going on.
