Let me do a quick knowledge dump on the various potential sources of error in polling
1. Errors in weighting. (GENDER) If a pollster underestimates the share of electorate of males voting, they're definitely underestimating Trump. (EDUCATION) If a pollster underestimates the share of the electorate of non-college people voting, they're probably underestimating Trump, particularly if they're underestimating the amount of non-college whites as a share of the electorate. (GEOGRAPHY) If a pollster underestimates the share of the electorate coming from rural areas and the like. (RACE) If a pollster underestimates the share of whites voting, they're certainly underestimating Trump. Remember that pollsters don't have anyone holding them accountable if they're wrong, so oftentimes they won't care if they're being too urban/nonwhite/female/educated in who they poll.
And let's assume they're weighting the electorate properly, there's still the potential for error because of....
2. Non-Response Bias. The fact that 98% of people aren't even bothering to pick up the phone is going to create issues with polling that even the most honest pollsters will have trouble correcting. Even if a pollster tries to call rural areas, there is the potential that the people in those rural areas more likely to be comfortable talking to pollsters will be the left leaning ones, while the right-leaning ones are more distrustful and hang-up immediately (or hang up before the poll is over). This could lead to pollsters thinking that, say, Biden is doing better in rural areas than Hillary was because the people who actually care enough to respond to pollsters are more left leaning and more media-trusting. Meanwhile it could be that Biden is actually doing worse because all the right-leaning people are not even responding to begin with. And of course, this applies to more than just rural areas - I mentioned rurals simply because I want to emphasize that *even if the polls are being correct in terms of how they weight the different types of voters* - this could still lead to error.
Decades ago, non-response bias was *NOWHERE* near as bad because pretty much everyone picked up the landline phone and candidates like Gerald Ford vs Jimmy Carter weren't controversial to discuss. Now? Not only are people not bothering picking up the phone if a stranger calls because of an abundance of telemarketers/robocalls/general-distrust-of-strangers, but we're in the most polarized era since the Civil War.
now, let's say that a poll weights correctly and that people are actually responding in a way that aligns with the real electorate...there's still a third way that error can manifest....
3. Deceptiveness in those who actually bother responding. There are three main ways this could manifest in a way that underestimates Trump.
First: Trump voters claiming they're "undecided" when they really are decided - on voting for Trump. If you look at some of the 2016 polls in states like Michigan, you can see polls that had results like Hillary: 46% Trump: 40% Third Party:8% Undecideds: 6% (horey shit a 6 point lead, drumpf is finished!). Only for the final result to be Trump: 47.something% Hillary 47.something%. Well, looks like pretty much all the undecideds went to Trump - thing is, were they really 'undecided' to begin with if it was that disproportionate?
Second: Trump voters claiming they're voting for a non-major party candidate. If you look at many 2016 polls, they consistently overestimated the amount of voters that Gary Johnson and Jill Stein would get. Now, some of that was third-party voters genuinely changing their minds at the last minute and going for a candidate "with an actual chance to win." But some of that literally was shy Trump voters claiming that they were voting for Gary Johnson because it's a less controversial thing to admit, socially speaking. This time around, Jo Jorgensen sometimes gets 3-5% in the polls...which is total *bullshit.* She's not even as prominent as Gary Johnson was in 2012, when he got 1% of the vote. Gary Johnson was a well known governor of New Mexico. Jo Jorgensen is more of a nobody, like the candidates the Lolbertarians put up in 2008 and 2004 (can you name them off the top of your head?). I guarantee you that some of those 'Jorgensen supporters' are shy Trump voters.
Third: Trump voters claiming they're voting for Biden. This would have a more significant impact on the margins than the previous two types of deceptiveness.
Let's say that there are 12 voters for Trump and 11 for Biden and 1 for Jorgensen. The real final margin would be Trump+1 over Biden.
EXAMPLE of how the (First) kind of error can manifest: Now let's say that in a poll, it's 8 for Trump, 10 for Biden, 1 for Jorgensen, 5 Undecided. Biden+2 over Trump
4 of those 5 Undecideds could be shy Trump voters, 1 could be a genuine undecided who goes to Biden. --> Actual final result: 12 Trump, 11 Biden, 1 Jorgensen. Poll off in the margin by 3.
Example of how the (Second) kind of error can manifest: Poll says 8 for Trump, 10 for Biden, 5 for Jorgensen (Biden+2) --> 4 Jorgensen 'supporters' go for Trump, 1 Jorgensen supporter decides to vote Biden instead. Actual final result: 12 Trump, 11 Biden, 1 Jorgensen. (Trump+1). Poll off in the margin by 3
Example of how the (Third) kind of error can manifest: Poll says 8 for Trump, 15 for Biden, 1 for Jorgensen (Biden+7 hory shit blomf is finished). --> 4 Biden 'supporters' turn out to be bullshitting and vote for Trump. Actual final result: 12 Trump, 11 Biden, 1 Jorgensen (Trump+1). Poll off in the margin by 8!!!!
So as you can see, this kind of polling error can cause even bigger misses than the first two forms of deceptiveness. Question is - how many people in the polls are Trump supporters claiming to be Biden supporters to fuck with the polls? Well, we're going to have to wait until the election is over to have an idea. But given that many Trump supporters will be hesitant to be honest to a stranger from a media/university polling organization, and may want to fuck with said pollster for amusement, there could be more fake D supporters than in 2016.
Note that in 2016, it seems that the type of error most common was the (First) and (Second). However, given that Jo Jorgensen and Howie Hawkins are much less prominent, I speculate that the same type of Trump voters who lied and said they were voting for Gary Johnson, are more likely to lie and say they're voting for Biden this time around - but they also could be claiming Jorgensen or Undecided or another third party candidate.
Of course, there are ways in which error can lead to Trump being overestimated, but intuitively they're less likely to happen. Biden supporters tend to trust the media and respect the polls a lot more, which means they'd lie to pollsters much less. And there'd be less non-response bias because they'd be more likely to talk to the pollsters for extended periods of time. And some pollsters won't care if they have too many urbans/educateds/nonwhites/women in their polls. So overall, all these sources of error point to Trump outperforming the polling. Question is...by how much?
Well, Trafalgar and Richard Baris both seem to have worked hard about counteracting (1-Weighting) and (2Non-ResponseBias). Their polls point to a close race where Trump is either ahead in the key battlegrounds, or just *very slightly* behind. However, they cannot really fix (3-DeceptivenessInResponse) because after all, they're still strangers to the people they're polling. I've seen too many Jorgensen supporters and undecideds in their polls which leads me to think that even *they* are underestimating Trump and that Trump probably will win the states where he is slightly behind in their polls,
Now, when it comes to pollsters like CNN/NYTIMES-SIENA/CHANGE RESEARCH/QUINNIPIAC/MONMOUTH/etc ? I think they have (1 -Weighting) (2-NonResponseBias) and (3-DeceptivenessInResponse) manifesting in spades. So could they miss the real final result by as much as double digits, such that Trump wins despite being behind by a lot? Absolutely!
---
edit: oh and there's a 4th way in which polls can miss the mark. It's voters who are being honest but genuinely change their mind before voting. They can be *genuinely undecided* but come around to supporting Trump. They can also be *genuinely planning on voting for Biden* but change their mind and vote Trump.
I can write more about this later, but this is another thing to keep in mind. I speculate that the way the race is going, this kind of error will benefit Trump. I don't see large numbers of Trump supporters changing their minds at this juncture.
I do see *genuine undecideds* being more likely to vote Trump, because if they're still *genuinely undecided* after nuclear warheads of media negativity, I don't think they'll decide to vote against Trump.
And are there still some soft-Biden supporters who are wavering because they're wondering if he's an old weak silent gen dementia patient who will be controlled by the left and other malignant interests, who may not be able to be Moderate Friendly Uniting (tm) Joe like he was advertised to be? I'm sure there are.
