No he did not make that claim. He stated that a number of outcomes had a particular probability of occurring - which, based on the plain facts, was the only thing logical to say, and I said as much before the election. Anyone who said they KNEW what would happen was a fool.
Silver is very smart, and as a baseball fan knows not to say something WILL happen for sure (unlike a neuroscientist I can think of who had to eat a bug.) He said that there was a probability Clinton would get 270 electoral votes. Not that it was certain, or even overwhelmingly probable.
Moreover, on the day of the election polls-only model being used was giving a Trump victory in over one run out of four, and that was largely from polls taken prior to the public announcement of the FBI reopening an investigation into Clinton’s emails. That may sound like poor odds to someone unversed in statistics but it actually quite high, indicating considerable variability (either low sample sizes or a high number of undecided voters). The standard uncertainty, by the way, assumes that the errors in polling sample are randomly distributed; a systemic bias–which clearly occurred here–may give a much larger shift and is why Silver’s model accounts for an unexpected flip in one state having similar effects on culturally similar states.
This was an election in which voter turnout was lower than it has been in over a decade, in which both major party candidates were disliked to an extent unseen since the beginning of modern polling, and discussion and ultimately voter decisions were based more on character than actual policy, and most of that character was pretty poor on both parts. This is the first time in living memory that a major party candidate literally implored his adherents to be angry and violent, stoking a certain kind of emotional response that is really outside experience in presidential elections since George Wallace and the historic shifts of both the Democrat and Republican parties.
The two most surprising results are that the antipathy toward Clinton as a person was so great that demographics that would normally vote in solid blocks for a Democratic presidential candidate were split, and that Trump introduced the novel tactic of being as deliberately outrageous as possible on a daily basis which apparently appealed to as many or more people than it incised and certainly got him a lot of free coverage. Although the latter has worked well in other countries to the point of becoming a standard tactic in elections of nations we tend to view as being politically unstable, it has never been a feature in US presidential elections to this extent, where even when candidates attacked each others’ character they did so from the angle of performance in office rather than the sort of vicious personal statements calling one another or their adherents “nasty”, “crooked”, “deplorable”, et cetera. It is noteworthy that both candidates engaged in this behavior to differing degrees of effectiveness, which may go to inform how candidates are selected in the future (and not for the better).
George E.P. Box famously said, “All models are wrong, but some are useful.” In this case, the model and method of polling that has been pretty useful and validated by data from the beginning of modern polling was pushed to the edge of predictive ability by the volatile circumstances of the election and dislike of the candidates, which was clearly indicated by the degree of uncertainty in the predictions. That there was a bias in interpreting and reporting the results clearly exists, in large measure because a major party candidate as baldly hateful as Trump has not existed in living memory, and we’ve never seen two candidates so widely disliked with net negative approval ratings throughout the entire campaign. The assumption that the less hated candidate would receive more undecided votes in each demographic in support is probably questionable (although Clinton did win more votes overall). There may have also been systemic bias in how polls are conducted (either underreporting by Trump voters for various reasons or just a larger degree of undecided or switch voters), or just net shifts in local demographics due to various dynamics that were not captured by polling.
In short, you can’t blame the model for being fed data with a systematic bias, and you can’t blame the sample data for being more volatile than historical analysis would indicate. You can blame the pundits for relying far too much on the data than actually understanding the collective mood of the key demographics, and the people who concluded that poll results showing a difference that was within the margins of error was definitive, but the honest truth is that Donald Trump was unimaginable to most people as a serious candidate even four years ago, and that expectation informed the opinions of people who were not driven by his rhetoric. The really ugly, unfortunate conclusion is that hatred and fear are much more motivating factors for voters now than they have been since at least the Great Depression, and the public response of a candidate deliberately acting badly to tap into that has become far more acceptable than it has been since the Civil War.
Stranger
So, if the weather forecast says there is a 30% chance of rain and it rains, are they wrong?
Yes…Silver took considerable care to emphasize the uncertainty. And, he also arguing strongly against the conventional wisdom of some Clinton backers who felt that even if she lost the popular vote, she still had a firewall in the electoral college that would prevent her from losing the election. He pointed out how this logic was essentially imagining a national vote shift without a corresponding vote shift in the states and said that the firewall was simply a byproduct of having the lead in the national polls.
Furthermore, he emphasized that, in fact, the 538 model showed that Donald Trump losing the popular vote but winning the electoral college was a much more likely scenario than the other way around.
Interesting fact. The polling/data guru for the Trump campain was on TV last night.
He said he predicted 305 (or 303 or something very similiar) for Trump on the Friday before the election (and I think he said he was predicting a good chance of a win well before that).
IIRC I think he missed one upset (Minnesota?) that he didn’t think Trump would get and one he he would get that he did not. The two roughly balanced out.
That dude should be damn busy the next election cycle or two.
He was wrong about this year’s election cycle – it happens. If I hire a stock picker and he gets it wrong once in a while, I can live with that if his longitudinal success speaks for itself, which it does. I know you’re amped up about your Trumpster winning and shocking the rest of the country, which is something I totally understand. But it doesn’t invalidate Nate Silver’s success over the course of a decade. Save your contempt for the guy who promised to eat a bug if Trump got over 240 EVs. LOL
Unlike Hannity’s self-waterboarding promise, Sam Wang actually did eat a bug on TV.