As of the present, the latest polls, in terms of their overall assessment, are showing a completely disparate set of numbers as to who is where in the presidential race. On one end, a McClatchy-Marist poll claim that Clinton leads Trump by 15% nationally, while, on the other end, a LA Times-USC claims that the Clinton lead is just 1%. Note that the polls were taken roughly at the same time.
Yet, once examined under the hood, the story told by all these surveys is not only the same, but an extremely familiar one: Trump is doing quite well among the non-college educated whites, rather poorly among college-educated whites, and not well at all among non-whites. McClatchy-Marist poll’s numbers, broken down by demographics, with possible adjustments for nonrespondents/don’t knows, could have come from anyone: Trump leads 46-31 among non-college whites, while Clinton leads 48-36 among college whites, while more than 90% of African-Americans would vote for Clinton while 55% of Latinos would (only 26% of Latinos favor Trump, however). Only two numbers really stand out as somewhat unexpected: Clinton’s 12% lead among college-educated whites is rather than other polls, although not necessarily that shocking, while the proportion of Hispanics who support her seems rather smaller than I’d have expected, although this too is not really that shocking. Electorally, the first number is the pertinent one: no Republican would have been expected to “win” the Latino vote, anyways, and limiting the shortfall, as Marco Rubio might conceivably have pulled off, would not have given him the winning edge in the Midwest, where states remain large and competitive. However, no Republican can expect to be even competitive while losing the college white votes–which they carried quite consistently for decades–by that kind of margin. And this is a story that repeats itself across every poll.
What is likely is that the gap is showing up, for the most part, in course of post stratification, of weighing the data in the form of expected turnout. Or, in other words, pollsters are all over the place in anticipating who will be showing up at the polls. But, given reasonable variability in turnout (up among non-college whites, down among African Americans, maybe up or not among Latinos, and decidedly down among college whites–many Republicans, I expect, would abstain rather than vote Clinton, which might lead to an odd outcome where college grads are no more likely to vote than among non college grads), all these point to a relatively narrow (perhaps 2-3% in national vote totals) victory for the Dems, I would guestimate. The truth, however, is that the actual vote %’s are probably impossible to predict meaningfully–the turnout numbers will be all over the place. There is a small chance that, with oddly shaped turnouts, Trump might pull off a win even while losing college white votes, for example. It is far more instructive to track the demographics rather than try to predict the overall outcome!