There are basically three ways that the presidential polls can be wrong. The first is that they have the wrong models for who is actually going to vote. So, for example, they may be underestimating Latino or black or Millennial turnout. Conversely, they could be overestimating those groups and not giving Trump enough credit for mobilizing white working class voters who never warmed to Mitt Romney. In these cases, the pollsters are accurately collecting people’s voting preferences but failing to weigh them correctly. This happened in 2012, as analysts did not anticipate that blacks would respond to suppression efforts by having, for the first time ever, a higher turnout percentage than whites.
The second way the polls could be wrong is actually a combination of two related things. One is that there’s some kind of disparity among Trump and Clinton voters in terms of who will respond to pollsters. It could be that Trump voters are just less likely to answer the phone and take the time to go through a litany of questions. Or, it could be that its the Millennials who are disproportionately uncooperative with survey-takers.
There could also be a lot of people who aren’t honest about who they’ll vote for. They are women who aren’t telling their husbands that they’ll be voting for Clinton. Are they going to be honest with a pollster? What if their husband is sitting next to them on the couch? At the same time, there are Trump voters who don’t want to admit that they’re willing to overlook his racism and misogyny. If these people basically balance out, the polls will be accurate, but what if they don’t?
The final way the polls could be wrong is if they simply cannot account for the difference in quality between the Democratic and Republican Get Out the Vote operations. There’s some reason to believe that Obama’s turnout operation was so superior to Romney’s that it added a few points to his margin and largely explains why Obama outperformed the polls.
Most of these possibilities are more likely to favor Clinton. She has the better turnout operation. She has the benefit of an aroused and angry Latino population that wants to punish Trump. She doesn’t have to do much to exceed expectations for black and Millennial turnout. There’s a big universe of people out there who might not want to admit that they’re supporting her, either because their husbands don’t approve or because they’ve always voted Republican and live in conservative communities.
But there are also Trump voters who are ashamed to admit that they support him. There are a lot of people who never voter before but got involved because of Trump’s campaign.
On the whole, though, especially when considering Trump’s non-existent ground game, I have to say that the odds are better that the polls are underestimating Clinton’s support rather than the reverse.
My main caveat for this is that Trump is coming out of a very low point for his campaign, and the polls may be a kind of nadir for him. It’s a problem because a lot of voting has already happened, but I wouldn’t be shocked if he’s recovered a bit from his low by Election Day. In that case, the polls won’t be wrong because they’ll continue to tighten.
It’s still hard for me to even make a worst case scenario where Trump is actually going to do far better than the polls predict.