And if you have time, you can also canvass, which I am doing tomorrow morning and afternoon.
Yeah, well, it’s not Clinton +4 right now. It’s a polling average that favors Clinton +1.7, which is well within the MOE. It’s not even clear if Nate Silver’s judgment that Clinton’s numbers have stabilized is accurate. They’re moving as we speak, and it’s not clear in which direction.
It’s only 1.7 on RCP because they include that LA tracking poll that’s way outside at Trump +5. HuffPollster has it at Clinton +4.2
Indeed, if the LA Times tracker is included that warps the numbers pretty badly. Literally no one, including, I suspect, the people running it, does not think the LA Times tracker has a huge Republican house effect. That warps the RCP average by about a point.
Pardon my ignorance, but why can’t the LA Times run a decent poll? Aren’t they something like the 2nd larges paper in the country?
They just chose to run an experimental setup this cycle, that’s all. They did some innovative stuff, and next time they’ll tweak it and make it better. It might be the future of polling–it just wasn’t ready for prime time this cycle.
If you’re saying not enough data, why are you worried? Why worry about the data showing a Trump comeback these last two weeks?
Why it’s important to examine the Nevada voting information is that it’s NOT a poll. It’s an actual election. It would be awfully hard for a late surge to beat Clinton because the election in Nevada is mostly over; most people in the state vote before Election Day, and I would guess over half the votes are cast already. According to witnesses, Latino voters in Clark County (Las Vegas) were lined up by the score to vote.
Even 538 gets really bearish on Trump if he loses Nevada; according to their November 1 story on it, it cuts his chances massively.
The thing is, though, that this can’t be worked into their model. For all you or I know Silver has already talked to Jon Ralston on the phone and looked at this numbers and is thinking “well, Trump’s doomed.” But he lets his model do the talking, and Ralston’s analysis (which has a solid history, btw) doesn’t fit into it.
As Nate Silver has pointed out, you can’t necessarily toss out polls you don’t like, even if they’re outliers. They capture different parts of the electorate. Further, even if you accept that the LA Times poll is off, the average is actually consistent with recent polls such as McClatchy/Marist, which gives Clinton a 1 point lead. Fox News gives her a two point lead – these are the latest polls.
Oh, OK. Thanks.
It’s not an election, it’s part of an election that is far from finished – that’s why.
No question, and I mean this with as little snark as possible, that you are far on the end that believes that the electorate is extremely volatile and who have reacted to every jerk of the needle as a sign of potential total collapse one way or the other.
Wang’s extremely accurate predictions of how the polls would behave were however based on a very different perception of the electorate, a perception based on analysis of how the numbers had behaved in past elections, were behaving in the early days of this cycle, and have continued to behave, of a highly polarized hardened electorate.
Heck, back in May he was writing about how the February poll spread was a reasonable starting point and that one should pretty much ignore the ongoing polling as pretty much of little predictive value until sometime in August anyway.
At that point his median utilizing states-based meta-margin was actually just under 4.0. Since then it moved just over two up and two down and now is 2.7%. End of May he was predicting Clinton 336, Trump 202 EV; now Clinton 312, Trump 226 EV … his model is fairly stable.
Funny enough 538 floated the Wang view for a bit as Clinton was at her pneumonia and deplorables lows:
The polls then behaved as that hypothesis said they would, consistent with the polling shifts from the long term mean being short term illusory swings, i.e. noise.
But as they did Silver returned to beating the uncertainty drums and having his model swing much wider than anyone else’s.
There is something wrong with a model that swings as much as Silver’s has and flipping in PollsOnly from near 90% to 56% in roughly a month is basically an argument that polls are of little predictive value.
Doesn’t Wang’s model assume the state polls are independent from each other? Isn’t that obviously a false assumption?
There was a point this week when his model showed 99% Clinton victory (as it has for some time regardless of events) when Trump only had to exceed his public polls in a single state by a percent or two. That just can’t be right.
There is no MOE for a polling aggregate let alone for RCP’s rolling average of certain polls. At best what you can say is how much the aggregators have been off on election eve before. Silver has not been off by 3 or even close. RCP in particular was off last time, but the true aggregators not so much.
And here’s the essential divergence in views.
In your view (and, possibly with some disingenuousness, Silver’s) polls are equally likely to move 2 points Trumpward from the current C+2 RCP four-way (or 3 from his C+3 popular vote in PollsOnly) as move the same Clintonward. Each news cycle is an independent additional impact. the same view would have said at C+7 that moving two more up was as likely as dropping back down two points.
The more Wangian perspective is that thinking of movements as have degrees of elasticity and of inelasticity from the setpoint. Easy for a news cycle to move it the first two from the setpoint, from +4 to +2 say, or +4 to +6. The moving another point away from the setpoint hits lots more resistance and another point from there is a very hard stretch indeed requiring significant and constant force. Let up on that force with a news cycle getting old and it moves back towards the set point.
Reality has comported better with the Wangian perspective to date.
No it does not. Or rather it does only for the snapshot, which is similar to NowCast.
Also here:
The difference is that Silver uses a much more complex covariance model based on which states are more like each other demographically. It is unclear how much that adds and when added complexity is not proven to add value some us would prefer the simpler approach.
So he randomly covaries all the states? Why is that more simple or better than assuming that states that have been closely correlated in the past will be more closely correlated in the future, all else being equal?
But we’ll see. We’ll have a few tests, since it looks like Silver and Wang are going to disagree on at least a few states/districts. I’d put my money on Wang being wrong about, for example, Clinton winning ME-2.
For Nevada “far from finished” is an overstatement.
In case that is not clear, it is likely that two thirds of the votes to be cast have already been cast. On election night a clear lead with 66% of the vote in may be called. It is not over, and we do not know exactly what those counts really are, but “far from finished”? Nah.
No he covaries them all at 100% with possible national shifts: “an adequate assumption is total covariation”
“In this case a simple and effective way to vary long-term change is to covary all states together by a random amount. This is at the core of the prediction, a feature that was introduced starting in 2012.”
RCP average was off by 2.7 points, which means that if the polls are underestimating Trump, we wake up on November 9th to President-elect Trump. In fact, they don’t even have to be off by that much. In fact, the polls could be right and Clinton could win the popular vote by 1.7 percent – and we could still wake up on November 9th to President-elect Trump.
I don’t know if they’re equally likely. But I reject the idea that because numbers have been stable throughout the race that it has predictive validity at the end of the race, regardless of what events occur in the news cycle. I might be mischaracterizing Wang or over-simplifying his position in this regard – I admit that I’m probably not as familiar with him as I could be, but that’s how I interpret his take on the race. I believe that I’m more in line with Nate Silver’s analysis, which is that Hillary has been the stronger candidate throughout the race, but that there has nevertheless been an opportunity for Trump to remain competitive, and at times, due to various events in the campaign, he has been been surprisingly competitive, which is something that has to be factored into the race as it closes. That’s especially true if polls indicate that Donald Trump is now actually increasing his republican base of support.
Is Hillary still the favorite to win? Absolutely, and for a number of reasons. Could she win 320 or 330 EVs? Yes, she could. But right now, we don’t have enough good data to tell us one way or the other. I tend to believe that her floor is most likely at 278, and her ceiling is probably 322. I suspect she will be somewhere in the middle (maybe 307 or low 300s). We shall see.
Again, wait until the votes are finished.