Thanks! That’s really helpful.
Some great news from A- rated pollster Ipsos - Clinton at +7 nationally, with no adjustment, in a poll taken from 10/28-11/1.
Also, I’m starting to wonder if Nate Silver is taking his whole “correlation between polls in different states” thing a bit to far. For example, in his latest update Florida went from lightest blue to pink for seemingly no other reason than because DFM Research, a B- rated pollster, just put Trump at +9 (adjusted to +11) in Missouri. I get that shifts in individual states can often be part of a nationwide trend, however HRC operatives have basically invaded Florida in the past week or so to GOTV, whereas my guess is that she is merely keeping up appearances in bright red Missouri, so I think that in this point of the election this correlation starts to weaken as each candidate starts to further fine tune their strategies for individual states.
Very welcome. The only problem is they’re boring. They don’t change the prediction every news cycle. 
So those five A-minus Ipsos polls show Clinton steadily climbing between Oct 24 and Nov 1 nationally, and Silver raises her chances of winning a measly 0.2 percentage points. Then a poll from a B-minus-rated pollster shows Trump up by 9 in Missouri and Silver drops Hills’ chances of winning by a whopping 1.4 percentage points. All that from Missouri?
Come on Nate. I know better, but I keep hitting refresh anyway.
In his User’s Guide, Nate Silver says the model is conservative early and aggressive late. In other words, when the election is soon it has a tendency to consider fluctuations in the polls as meaningful when they may be merely noise.
Hadn’t checked out the user’s guide yet. Thx!
My name is skdo23, and I’m a 538 addict.
You could also be a doper. You have the 1666th post.
Anyways, the nowcast shows Nevada flipped towards Trump. Polls Plus should soon follow. Might be a useful filler state if Utah is won by Evan McMullin. The 538 shows that his chances are also slipping in Utah, so now Trump needs to win Colorado or Pennsylvania. If that’s the case, Trump just needs to win the equivalent of DC (3 points) to tie this ballgame up. In this perfect map I generated, Clinton would still win, but not by much. And any deviation in the map (Pennsylvania, Colorado, District of Columbia losing their minds (lol)), could point to Trump relocating from Manhattan.
EDIT: Any state that flips will result in Trump becoming the President- or at least tying up the election.
We also have a poll from Colorado showing the race tied:
http://www.realclearpolitics.com/epolls/latest_polls/
And since that same poll shows the Senate race at D+8, that rules out oversampling of Republicans.
For those of you who enjoy trying to make heads or tales of early voting numbers, here are the latest I could find:
I’d also recommend reading this.
It’s starting to look like Democrats got excited early in the process and now have slacked off.
Although this morning has started off with some decent polls for Clinton, 538 drops her to about a 65% chance of winning.
The Nevada poll is troubling but the polls have been all over the place in that state, and early voting seems to be favoring Clinton. However, Clinton and Trump supporters both have a sense of desperation in Nevada, so I think this will go down to the wire. Nevada is probably ground zero for white push-back against the wave of Hispanic immigration. And of course it is also where Clinton’s Hispanic support is strongest. It’s most likely a toss-up state, as it has been all along. My gut tells me Clinton will keep it blue by the thinnest of margins - margins so thin that it could be challenged after the election.
Jon Ralston says yesterday was a big win for Dems in NV: x.com
The most accurate poll of all: your interlocutor’s wife, who leans strongly Democratic (my influence, though she doesn’t like to admit it) renewed her voter registration for this election and is finally returning her Florida mail-in ballot today. She’s never voted before, though she planned to vote (for GWB) in 2004. She was originally pro-Hillary, but now she’s just anti-Trump. Her mother, who grew up as a New York Republican but listens to a lot of Limbaugh, switched from Trump to Gary Johnson just before early voting (and has now voted).
Final Prediction:
Clinton 323
Trump 215
CO-PA-NH-NV-VA all hold blue
Trump flips IA and (probably) OH from 2012, that’s it
Clinton flips NC and ekes out FL
At noon ET on Jan. 20, I’ll be in the air en route to Vegas.
Good luck to all, and sayonara.
This is almost where I have it. As of now, I give NC to Trump, which puts my map at 308 for Hillary.
Nate Silver’s model might not be correct, but it follows the data. The state-to-state correlations were empirically determined. Of course the correlations vary from election to election, and this election might well be an outlier anyway. But I can’t fault him for being fact-based rather than injecting guesses, no matter how reassuring those guesses might be.
When the results are all know how should we score the aggregators?
How much should confidence play in?
Can we agree ahead of time that the Brier score is the best metric to use?
Point of the Brier score is that it rates based not just on being right or wrong but the confidence of the predictions. Saying that X will happen with 51% confidence and being wrong less bad than saying X will happen and being wrong with 75% probability. But saying that X will happen with 51% confidence and being right less good than saying X will happen and being right with 75% probability.
Here’s is a discussion about the use of that score to judge the different aggregators in 2012 - which found the best was Lizner:
But there is also this critique of using the Brier score and suggesting an alternative:
Personally I’m not sure I follow the discussion that continues.
But that first link does rank the scores in 2012 by that metric as well (scroll down) - Wang and Lizner both did better than Silver by that metric as well.
The models going from the extreme confidence of Wang to the extreme uncertainty of Silver should give us a good chance to evaluate the models when all is said and done if we, preferably in advance, decide to judge based on some metric that evaluates justified vs unjustified confidence along with accuracy.
I give Florida to Trump (I just can’t trust Florida, ever. Sorry), but North Carolina BARELY to Hillary. Otherwise, I agree – hope – that this map is what happens.