Well I will say my doubting highly is no absolute certainty. It is not impossible that instead Trump supporters actually vote, reluctant Republicans come out and hold their noses, and younger voters and Black voters just stay home.
If the news cycle bounce hits for Trump at just the right point and it is within 3, let alone 2, on election day I can’t say I won’t have any worries.
The early vote analysis seems to be showing just that. The only assumedly pro-Clinton group turning out in higher numbers is Latinos, and they just don’t matter much at this point, although they might turn a close race. But if Clinton loses NC, PA, OH, and FL because black voters returned to their old voting patterns, higher Latino turnout won’t save her. It especially doesn’t save her if too many of them vote for Gary Johnson.
That post is a bit of a headscratcher, and not just because of the inexplicable lack of capitalization of Nate Silver’s name. You cite Silver, say he looks accurate, but precede all that with an “Actually” in response to my having said Hillary’s been over 49% for three weeks. Well, actually, that number comes from…Nate Silver. Go to that 538 forecast tool you just cited, scroll down to “How the forecast has changed”, switch from “chance of winning” to “popular vote”, and voila: the slider will give you the very same numbers I provided. 
That’s his forecast, I was citing polls. Silver, probably correctly, assumes most Stein voters will end up Clinton voters, plus gives her some undecideds to boot, to get her to 49%.
The problem is that there are no guarantees with undecided voters, and since they exceed her margin of advantage by 2-1, Trump still has a very real chance to win.
But his forecast is based on polls (other than “polls plus”, which doesn’t differ much at all). And he was closest to accurate in 2012. So the point stands: if she’s had the kind of polling over the past three weeks that makes the computer model of the most accurate prognosticator of our time keep her in the “forty-nine point something” lane for over three weeks now, that’s got to change or be wrong to give Trump a chance of passing her–because Johnson, Stein, McMullen, and everyone else are not going to collapse to less than one point combined nationally.
You do seem to be going between things. Silver does not do separate 2 and 4 ways. His NowCast (not projection but model of if the election today, polls) has it C+4.4.
I think you are doing a chimera of RCP’s 2 and 4-way nationals and then inadvertently implying that it is something that Silver is saying based on his polling analysis (which includes national and state polling data to come with an aggregate that is historically more on the money).
Obviously. Is there any contrary evidence out there–i.e., showing that Trump does better in subjective LV screens? I’m not aware of any.
No, but he does do better in automated polls than live telephone polls.
Can you share a link? I recall reading a comparison that didn’t show a big difference, but can’t remember where.
Sure, this was Nate Silver himself pointing this out:
This is truly a remarkable post. You are taking your act to a new level with this one.
I don’t think this is correct. I believe the Democrats have always had a far better GOTV operation. (This is due in large part to having a more concentrated and easily identifiable base of support.)
The LA Times poll, as noted previously.
And this is my commentary, what happens if the automated telephone polls are measuring Trump’s true level of support, not the live telephone polls? That’s enough of a difference to give Trump the win.
I’m comfortable ignoring that poll, since it has a ton of methodological differences from other polls. It’s a cool idea. Hopefully more firms will experiment with big panels. But having one black guy in Iowa sway your poll by so much is not a great result.
Panels are a terrible idea. Human psychology of consistency means that once people have stated a preference to a pollster, they are much less likely to change their minds than if they have never yet been asked. It’s almost like Schrodinger’s polling question.
Has there been and recent studies into the demographic of those “most likely to answer polls”?
The angry rural white voter with a land line who is at home every night who wants their opinion heard seems much more likely to submit to polling that someone with a cell phone in a suburb who checks the caller id and ignores it if they don’t know who it is.
Don’t pollsters make adjustments based on that? They don’t just submit the raw data, do they?
That’s all very nice for the right argument. If someone was claiming that the LA Times poll was conclusive, you have a rejoinder. Or if you have conclusive counter-evidence of your own.
But the context here is that you pointed to some evidence - which is perfectly fine and valid - but when someone pointed out issues with your own evidence your response was that well, you’re not aware of any contrary evidence. In that context, pointing out that the LA Times methodology is also inconclusive doesn’t get you there. Bottom line is that there’s inconclusive evidence on both sides of the question.
Well, they try to, at least. How good a job they do at it depends on how well they understand the reasons why people don’t answer their phones, and how well they can pin down the relevant demographics. And it might even tie in with the likely-voter problem, if there’s a correlation (positive or negative) between those who answer polls and those who get out and vote.
And cell phones have another problem, besides being legally required to be called by humans. Cell phones move around. I currently live in Ohio, one of the most important (and thus most-polled) of swing states, but at the time I first got cell service, I lived in Montana, and hence my area code is 406, not 216. I’ve only once ever been called by a political pollster, and wasn’t able to provide any data for them, because they were polling Montana (yes, even Montana gets polled occasionally). I’m not likely to ever get called by an Ohio pollster, unless someone is keeping a very careful eye on Big Data to see that that particular phone number now lives here.
And what’s more, this effect probably varies by demographic, too. Someone who’s been retired since the time cell phones took off probably still lives in the same state as their area code. Someone who was in college at that time likely doesn’t. Again, it’s possible to try to apply demographic weighting to counter this, but the less data you have to base your weighting on, the more noise you’ll get. And who knows? Maybe having lived in more than one state is itself a politically-significant demographic, in which case the weighting is doomed to fail.
I would quibble with whether the LA Times poll is indeed evidence of the proposition for which you cite it, but that strikes me a boring argument. So I’ll just amend my statement to “I’m not aware of any evidence that people should find persuasive…”
If you didn’t read the UpShot’s deep dive into the LA Times poll’s data, you should. It was fascinating.
They try to but the demographics change so fast it’s hard to know how accurate the adjustments are. If you google “Gallup 2012 prediction” there’s a lot of articles trying to explain why Gallup had Romney as the winner. Most of it pointing to their adjustments just not keeping up with real world changes.