The difference in the election comes down to turnout and margins amongst non-whites.

When you dig around in the numbers behind the polls, the key variables consistently center on the racial composition of the electorate and margins amongst non-whites.  A great number of people were surprised to see Hillary Clinton spending time in Arizona, and two recent public polls from Arizona each show Donald Trump leading by 5 points there.  However, the key thing here is margin with Latino voters.   The NBC/Marist Poll put the Clinton lead among that group at 40%, whereas a Univision poll just of Hispanics in the state found a larger lead of 49%.  That difference alone is worth 2% overall, potentially taking a 5% lead down to a 3% lead.  If the Latino percentage of the electorate also increased slightly beyond the public poll projections then already that would make the race still a bit closer.  Add in a marginally better Native American vote (which is pretty much all anti-Trump), and Arizona is every bit the battleground as anywhere.  

 In Nevada the effect of Hispanic voters upon poll accuracy can be even bigger. CNN’s poll of Nevada which had Trump up 6 points, had him winning 32% of all non-white voters. In contrast, the Univision sample put Trump’s share with Hispanics at a meager 19%.  Given how poorly Trump is likely to perform with the other major non-white group in the Sliver State (African-Americans), CNN’s prediction is giving Trump a Hispanic share at close to 40%.  The gap between 19% and 40% clearly holds the potential to determine the outcome. 

In poll after poll, nationally and at the state level the difference in anticipated margins almost invariably comes down to how non-white do you anticipate the electorate to be and how do you think the non-white electorate will vote.  When we look, for example, at the Republican polling firm Remington, they find in Pennsylvania a somewhat unprecedented 18% of the African America vote going to Trump.  That would be unusual for a Republican.  Their poll in Florida has 22% of African Americans for Trump and 42% of Hispanics.  And so on and so forth.  Pollsters attempting to survey just African Americans or just Hispanics have produced much different results. The two most robust polls of Hispanics put Clinton’s leads at 60 and 40 points respectively. The only poll that exists of just African Americans finds Clinton leading 90% to 2%. These are big gaps.  There also compositional questions. In Pennsylvania we have seen polls with the electorate being as small as 9% African American when exit polls from the last four elections have put that number at 13%.  A dip might make sense this year, but dropping from 13% to 9% is a lot.   

This is not to say that one answer is automatically right and other is wrong.  But we can safely say that this question does play a haunting role over all of the public polling in this race and may also explain some of the 2012 polling failure, when Obama led by only .7 nationally but ended up winning by 3.9%.  Tomorrow we will tackle the early vote and maybe crank out one more state, before also doing our final predictions.