He went where the data took him based on a very particular set of assumptions. A set of assumptions that made his models’ forecasts flop around a lot, as soon as news cycles moved one way or the other. The model that was anchored to reduce flopping around to transient news cycle blips was anchored on an assumption that the proper Bayesian prior to use was econometric data that predicted a virtual tie.
Wang also went where the data took him, based on particular assumptions. His assumptions created a long range forecast that remained fairly accurate at predicting where the polls would end up throughout the news cycles blips. It included being anchored to a Bayesian prior of long term polling averages, an assumption based on the data. He had forecast about how wide the range polling would mostly vary within and predicting that most likely after all was said and done it would return to near that Bayesian prior of about C+4.
Which model was providing the more accurate forecast of where the polls would end up on average over most of the cycle?
Wang is no guru but his fundamental hypothesis this time around has been that Trump and Clinton were not per se all that special, that this election’s final polling would be determined more upon the hardened divide of a polarized electorate than anything else. Interesting that Obama won by 3.9 in this polarized electorate four years ago as a very different Democratic candidate than Clinton, against Romney, a very different GOP candidate than Trump (understatement of the year) and now polls are finishing with 4 as the number on Silver’s blog.
It seems to me that the polarized electorate hypothesis is well supported.
It’s possible that the election results end up being a +4 election night victory for Clinton, and it’s also possible that Obama outperforms her in the electoral college, which is really what Nate Silver has been trying to point out all along. If Clinton runs up the score in starts that are naturally inclined to lean in her direction but comes closer to losing or actually does lose in states that Obama won, then that would probably support Nate Silver more than Sam Wang.
And in any case, you really cannot ignore the qualitative aspects of the campaigns that have moved polls. The races of 2012 and 2016 could end up with similar point spread in terms of the final percentages, but they will have arrived at these numbers differently, with Clinton having possibly outperformed Obama with some demographics and worse with others. Blue collar workers, for instance, appear to have liked Obama more than Clinton. There was no talk of Clinton losing PA a few years ago, but now in 2016 it’s a real possibility that Clinton could – odds are still in her favor but it’s much shakier for her. Obama won Ohio and Iowa. Those states are probably out of reach for Clinton. So they’re not really the same, even if she, like Obama, ends up winning the popular vote by 3 to 4 percentage points.
The electoral college map is one small part of what Silver has been saying. But both Silver and Wang agreed that this election cycle did not result in a major realignment of states.
Yes individual swing states can swing one way or the other.
Yes some zero sum game in demographics but the bigger point is that for right now it almost has to be. Pander to the resentful blue collar White group and lose in other demographics …
Not so positive that Ohio and Iowa are completely out of reach btw. Likely so. But if Clinton outperforms by 3, something that I do not think is impossible, then one or both can still flip.
Yeah, I agree that I should probably walk back the idea that Ohio and Iowa are ‘out of reach’ - they’re not. But those seem to be long shots. She has about as much chance to win Ohio as Trump does PA. But this is a state where a GOTV effort could be the difference maker. I suspect that she closed strong there, especially in Cleveland.
Because IF you believe Nate’s probability-percentages express something real, THEN in that whole article he explains what they mean in an especially entertaining and helpful style.
If you DON’T, well, okay, carry on with the debating. But my post was meant in the spirit of “don’t fight the hypothetical.”
Nationally it looks like this, and yet the electorate in Iowa has swung several points red and the electorate in Arizona has swung several points blue. It is certainly not inconceivable that with different candidates we could have seen the Republican pickups in non-college educated whites without the losses in Hispanics, or alternatively the Democratic gains among Hispanics without the losses in non-college educated whites.
It seems to me like it is something of a coincidence that the electoral swings this cycle look set to cancel each other out. I’m not sure I’d credit Wang for successfully predicting “no change in the electorate” when there have been two counterbalancing changes in the electorate.
Of course, some of this remains to be seen in the results.
Weight of money seems just as confused as the statistical predictors:
In betting markets, Trump was at 21% chance early Sunday, then when the FBI announced termination of the investigation he dropped to 16% Sunday night. Since then, Trump has steadily rallied, and this morning is back at an implied 21% chance.
In the financial markets, stocks $ & peso all rallied Sunday night on the FBI announcement, followed through Monday and continued with those rallies, and now seem to be almost fully discounting a Hillary win.
That is not actually what Nate Silver has been saying all along at all. He has said - in probably a dozen or more discussions, blog posts, and the like - that his uncertainty is based on what he perceives as being a high degree of volatility in the race, primarily because there are many third party/undecided voters. The uncertainty in his model does not come from an unusually high EV-popular vote mismatch.
[QUOTE=Nate Silver]
the number I have on my mind today is “4.” That’s because it kept coming up over and over as national polls were released today: It seemed like every pollster had Clinton leading by 4 percentage points.
[/QUOTE]
Maybe Data is trying to send him a message from the future?
Statistics professor from the Univewrsity of Illinois has an article talking about the problems he sees in Silver’s model. He claims that it’s too baroque and not sufficiently transparent.
Rosenthal says that he was motivated to write this because he was distressed that a respected statistician would respond to criticism with invective on Twitter rather than a useful explanation. Silver missed a teachable moment, and probably in part because of his model’s over complexity.
It was an interesting read. Silver didn’t respond for comment in the article. This is probably not the day for it. Hopefully, he’ll be open to more nuanced discussion in the coming weeks.
Actually it looks like Silver is being vindicated. There appears to be a big polling error and Trump is far outperforming recent Republicans in rural areas making the election much closer than expected. Josh Marshall from TPM:
“Appreciate the congrats”. You know you want to say it.
Personally my dominant emotion is now terror for the fate of the global liberal democratic order that has kept the peace and allowed billions to prosper over the last 70 years.:eek:
No question that Silver’s uncertainty was well justified.
You, asahi, called it every which way.
Meanwhile yes my reaction is deep horror. If this turns out as it is looking now that it might then I believe our country is turned back many decades and the world is in deep shit.
