For president. But feel free to comment on House and Senate polling as well. For simplicity I’ll stick with 538’s final eve of numbers as the aggregator archetype, but feel free to comment on how you think the different source may diverge.
My guess is an underestimate of Clinton’s margin by 4ish.
I think that the internet world has changed the game on polls. As a sort of momentary quick check on sensibilities, they may still have some use, but the internet’s speed seems to have resulted in much larger masses of people changing their opinions very rapidly.
That, plus the number of people conducting what they clam are real polls, but most of which are more like heavily biased snark-fests, or carefully crafted attempts to shape voting by playing the “join the in crowd” trick on people.
I think many of the polls are now more like people taking selfies. Two minutes after they answer the poll, they are on to some other outrage or fad, and changing their “planned vote” again.
Mostly, I think more and more people are answering polls the same way that you hear people say things like “If Trump wins, I’m leaving the country!” (Or "if Hillary wins, for that group). In the event, they will do something else.
My short answer is, that I don’t expect the polls taken NOW, to have any relation to the final votes.
Supposedly Clinton has a much better GOTV effort than Trump, which could translate into a bigger margin than the polls are showing.
I think maybe the aggregate polls will end up being pretty accurate – not because they’ll actually BE very accurate, but rather because the various ways they get things somewhat wrong will happen to cancel each other out, in this particular election. I mean things like “Wilder effect* pro-Trump,” “Wilder effect* pro-Clinton,” how the Johnson and Stein supporters break in certain states, etc.
*This refers to folks who, in the secrecy of the booth, vote for someone unpopular in their social circle. Also called the “Bradley effect.”
There is no option for a miss between 1 and 2, which is what I wanted to vote.
Clinton will win in a massive landslide. Massive.
Quite possibly… but the OP’s question is, will that be reflected in the 538 prediction around November 7?
Round.
The accuracy or inaccuracy of your initial portion aside, your short answer has nothing to do with the question asked.
The op is not about “polls taken NOW” but final poll aggregation on the eve before.
Will the polls overall be “skewed”? Will they, overall, in aggregate, accurately get a bead on who turns out? Are they overestimating or underestimating in one direction or the other who is a likely voter? By how much and why do you think so?
The answers have ranges. 
I would guess between 1 and 2 percent. That is consistent with the margin by which the poll aggregators underestimated Obama’s margin in 2012. Which Clinton has an even bigger advantage in polling day mechanics, she’s a less enthusiastically supported candidate who won’t draw as much of the African American vote as Obama did and I think those balance out.
I went with 2 percent.
The RCP avg in 2012 underestimated Obama’s final margin by more than 2 less than 3 % points. But there’s no way to know, or I have no idea anyway, what changes pollsters have made in their assumptions due to that miss, among the various other adjustments they’d make for that they think are good reasons. And of course depends where the individual polling outfit came out, which is then more influenced by pure statistical noise.
I think no error is the best estimate, unless I believe polls are deliberately biased or I think know better than the avg professional pollster what’s going on with turnout. For example Clinton has a better GOTV operation…but everyone knows that. Strikes me as kind of like predicting the stock market will go up because the economy is basically stable…but that’s already known by participants setting the price where it is. OK the polling process isn’t as ‘crowd sourced’ as that, but still. Such predictions can be fun or lead to interesting discussions but I don’t see a reason why the probability distribution of the error wouldn’t be centered pretty close to zero.
1936 and 1948 and 1992 were some big differences: Election Year Presidential Preference Polling and Voting Outcome | The American Presidency Project
If Hillary stays double digits then there has never been an election like this one, which means there is no way Trump can overcome double digits without a bombshell miracle.
This national election has been so crazy anyway it misewell end crazy.
I think that Clinton’s superior ground game will result in an extra 2-3% going her way. I’d be surprised if it’s more than about 4%, or less than 1%.
My suspicion is that many people are embarrassed to tell pollsters they’ll vote for Trump, but will end up voting against Hillary anyway. So I suspect Trump is doing better than the polls indicate.
Every Trump supporter I’ve had the pleasure of interacting with are loud and proud about it.
As one of the few choosing the Clinton +4 to 6 (4ish) my reasons are both the GOTV machine (especially in swing states) but also that I understand LV screens to use reported intent to vote and previously voted as strong factors and I those may overestimate who will actually come out for Trump this time.
Some previous GOP voters are barely hanging on with Trump and may end up staying home and some saying they will vote for him are people who when push comes to shove won’t be bothered to come out.
The NBC/SurveyMonkey explanation of their LV screen choices, rationa as they may be, illustrate the truth of their initial statement:
This is known as the “Shy Tory” phenomenon.
There’s very little evidence to support that this is a substantial thing. Generally speaking, people vote for who they tell the pollsters they’re going to vote for. As Chisquirrel points out, Trump voters seem proud of it, and the great majority of voters (on either side) do not interact very much with people with opposing viewpoints; they watch news outlets that reinforce their beliefs, read articles that reinforce their beliefs, and on and on. That is probably more true now than it’s ever been. They’ve no reason to be shy.
Put me down as well for “Clinton’s superior GOTV machine is unreflected in the polling,” to the tune of 2-3 percent. It could be more, but I refrain from mistaking a hope for a belief.
I don’t offer 538 as infallible oracle as some might tend to but here’s an interesting discussion of possible poll error among their principals from fairly recently.
In this poll thread, the ‘mode’ of answers is that Clinton’s assumed GOTV advantage will result in her overperforming the final poll average by something like what Obama did in 2012, 2+% points. The 538 discussion tends to agree more than not though not quantifying it and presenting the opposite view, which is that while literal GOTV on election day wouldn’t be expected to have shown up in polls, a ground game in general is about motivating people before election day and that should show up in likely voter screens at least in theory. And pollsters can change LV screens if they are wrong, as they were in 2012. Silver himself seems in that discussion to say the historical basis of LV screens is a reason they might not capture Clinton’s advantage, but that seems to me potentially the wrong way around in that the last major (presidential election) historical data point, 2012, would have spurred pollsters to increase their estimate of turnout among groups a superior Democratic GOTV was aimed at.
The 538’ers don’t give much credence to the ‘shy Trump voter’ hypothesis because it didn’t show up basically at all in the primaries, net. Nor do they think Trump actually turned out that many previous non-voters (likely to also refuse to answer polls, a more likely mechanism for such hidden support than lying to pollsters) in the primaries, making the possibility of a surge of previous (white non-college educated) non voters in the general for Trump speculative.
I would say though that some of the discussion in the article like some posts here seems to mix together outcome v what polls say now with the actual question of outcome v what polls will say on election eve. In that respect also I don’t see any basis to think Clinton would outperform what polls will say on election eve by anything like 6% points. That possibility IMO does exist (though I’m not predicting it) for Trump based on general ‘revolt against the establishment’ seen in politics in a number of countries recently; like the extreme example of the Colombian peace accord vote, where well respected international polling org missed by ~15%-points (it wasn’t just some problem with local orgs not tested elsewhere, they missed by a little less). Ideological people especially on the left don’t see it, but to a lot of people Clinton is yes to the establishment and status quo, Trump is ‘f-you’ to it, and the possibility exists IMO of a shocker in that direction, not so much in the other direction. But I’d still guess small error, 1% or less either way, is most likely to be right.
Isn’t this thread really asking “How biased are you? Unskew the final results!” 