Y’know, I read the first three books twice, most recently forty years ago, and found them so utterly tedious that I gave up on Asimov’s fiction, except for robots. Obama is definitely R Daneel Olivaw.
If you were a fan of R Daneel Olivaw then you might have enjoyed the way they ended it.
That article is worth reading, as it delves into some of the problems when you take forecasting seriously. As Nate does.
In some ways he sells himself short. There were ~17 candidates last summer. That gives the random candidate 6% odds. Nate assigned Trump a probability of 2%. That doesn’t seem off base to me, given Trump’s complete absence of political experience. Over the Fall Nate assigned Trump higher odds , but not fast enough. That’s where the errors occurred.
There were 2 key developments. The first was that Trump received a ton of free media via his antics. That was outside the historical record. The second was that the donor class never organized an ad-bomb against Trump. Nate and I assumed that they would: overestimating their competence or rationality was a mistake. The free media factored might have been noticed in early Fall. But hints of the incompetence of the donor class might only have become discernible just before Thanksgiving. Because it was clear that mobilization was in their general interest, but that it wasn’t happening.
Billionaire donor behavior needs to be modeled better. We have less experience with that, since funding sources shifted away from the affluent and towards the mega-rich after Citizens United et al.
All that said, it made sense to place nontrivial skepticism on the odds of an outsider before the first ballot. Because such clowns tended to flame out in the past.
Polling on the eve of each individual primary worked well, with the exception of Michigan. And we should expect such a failure around 5% of the time or so. I argue that scientific punditry is vastly superior to the jackass variety. Science permits us to fall forwards.
I liked the first one, thought the second was ok if a little stilted, didn’t bother with the third. Read them last year I think. Once.
This thread has done wonders to remind me exactly how very, VERY much I hate the names Asimov gave his characters. They’re like nails on a chalkboard that are somehow simultaneously rubbing a cheese grater on my gums.
Hari Seldon? Daneel Olivaw? Sorry, forgot the pretentious initial at the beginning of the last one, thus differentiating him from all the OTHER Daneel Olivaws.
Like a chainsaw earwax-cleaning, I’m telling ya.
Kind of apropos, as that’s how I feel about Trump as well.
“R” was a designation all robots were required to use because of earth’s paranoia about robots at that time.
The ‘R.’ Is for Robot.
Trotsky’s on Fourth.
I agree. The odd names had a bit of futuristic quality to them, but apparently a future where people wanted their names to give everyone a throbbing headache.
But Trosky’s on first.
I saw right from the start that Silver’s predictions about Trump were flawed. All he really did was to slice up the process of winning the primary into a bunch of small discrete events, and then arbitrarily assigned a probability to each slice. It’s really no different from, and no more useful than, the Drake equation. If he had instead applied his standard models that he uses for everything else to the actual data (most especially, the polling), he would have gotten a much better result.
In other words, he missed not because he underestimated how different Trump is, but because he overestimated. In a normal cycle, the guy who’s leading the polls is probably going to win. In this cycle, that also happened. In that regard, this cycle was a lot more normal than it appears.
This is the Dope. We have all long since been seduced by the Dork Side.
Brilliant OP, Salvor. 
No. Polls in the summer and early Fall before the primaries are useless.
Well, maybe not useless - they might be noisy indicators.
Your analogy to the Drake equation is spot-on though. It’s dubious as a forecasting method, but a defensible way of organizing your thoughts. It did allow Nate to ratchet up his forecast probabilities, just not fast enough. More polling machinery and harder thinking might have helped.
Recall though in 2012 that all manner of candidates received their bump in the polls. Didn’t mean much. I still say Trump’s odds were very low in August. The lesson I draw is, “If you must forecast, forecast often.” Except it’s worse. You have to know when to shift your paradigm. Stating a framework has the advantage, usually overriding, of permitting you to ignore predictably noisy information. I say that if you want to be serious about these things, you have anticipate circumstances where you think you can say your model is broken. Would 60 consecutive days of strong Trump polling be sufficient for that? 90 days? Not sure.
ETA: Hm. Conditional probabilities. Perhaps the early hurdles were 50-50, but he later ones favored Trump more heavily conditional on clearing the earlier hurdles. I’m still surprised there was no ad-bomb though: clearly my implicit model of mega-donor behavior was amiss.
How would you predict Donald Trump? Something like him just hasn’t happened in modern presidential politics. You could call him a mule and say that he’s an exceptional individual, but I think mostly we’re just running into the limits of the small sample size that we have on presidential election. Nate Silver wasn’t wrong when he wrote that candidates like Trump tended to not do well, he was just overconfident in predicting that that trend would continue to hold, without a large enough sample to say it with the amount of confidence that he did.
Don’t remember 'cuz I was asleep. Those guys who say that sleep reading increases your retention? They’re wrong.
I skimmed a Foundation novel involving the Mule for about 10 minutes in Marshall Field’s while waiting for friends and I got most of the OP. 
Billionaire donor behavior needs to be modeled better. We have less experience with that, since funding sources shifted away from the affluent and towards the mega-rich after Citizens United et al
(Quoting Measure for Measure, phone messed up quote tags)
But the ability to do that is very limited. Both in SF and real life, social science can only accurately predict the behavior of large groups of people. The smaller the group of decision makers, the less powerful tbese methods are, as Silver has said about bis failure to come up with a good model for predicting Supreme Court decisions.