"Sen. Bernie Sanders has the next-best shot, with a 22 percent chance at a majority, followed by Sen. Elizabeth Warren at 12 percent and former Mayor Pete Buttigieg at 10 percent. There’s also a 14 percent chance — about 1 in 7 — that no one will win a majority of pledged delegates by June 6, which could lead to a contested convention. The model works by simulating the nomination race thousands of times, accounting for the bounces that candidates may receive by winning or losing states, along with other contingencies — such as candidates dropping out and polls moving in response to debates and news events. Like all of our models, it’s empirically driven, built using data from the 15 competitive nomination races since 1980. Since the primaries themselves are fairly complex process, the model is fairly complex also — which we mean as a warning as much as a brag. Models with more complexity are easier to screw up and can be more sensitive to initial assumptions — so we’d encourage you to read more about how our model works."
Nate Silver explains, using his amazing science.
ADDED: I had to publish this post to click on my "over-complication" tag, which I probably could have used a few more times if I'd kept it in mind. It's the kind of tag I love, specific but abstract, so it collects things from scattered topics that resonate. Today's post is only the 6th time I've used it since I created it in 2009 to observe that I'd "crossed the over-complication line" with a post that had a strange set of tags ("abortion, Althouse + Meade, Beccah Beushausen, beer, blogging, dolls, fake, James Frey, Meade, Oprah, Orson Welles, prayer, writing"). It took me a year to use it again, with this great quote from Gertrude Stein: "She always says she dislikes the abnormal, it is so obvious. She says the normal is so much more simply complicated and interesting." Didn't use it again until 2011 — "A Very Simple Venn Diagram of Where the Tea Party and Occupy Wall Street Agree" — and then once in 2017 (a labyrinthine sentence about feminism) and once in 2019 (a New Yorker critic bothered by the complexity of the movie "Joker").
January 11, 2020
Subscribe to:
Post Comments (Atom)
104 comments:
I got to use my "over-complication" tag again. Wonder what else is in there...
He's saying who knows, anything can happen. But with percentages!
You pays your money and you takes your chances.
It's going to be Buddha gig. (And that's how voice text interpreted his name.) Buttigieg. The whole field is terrible and there's no one that can Unite all the tribes of the Democratic Party but at least he's a gay so white liberals can feel good about voting for him
I hold the position that Biden - flaws and all -- is only one of these clowns who can beat Trump.
Models with more complexity are easier to screw up and can be more sensitive to initial assumptions
That is an understatement.
Silver is somewhat good at what he does, but it sounds like more of a hobby than a serious endeavor. If he was really good at developing predictive models, he would become a billionaire hedge fund manager.
it’s empirically driven, built using data from the 15 competitive nomination races since 1980. Since the primaries themselves are fairly complex process, the model is fairly complex also
A very small sample size to be drawing any conclusions about anything, yet he shows the percentages down to single digits. Kind of irresponsible really...
Models with more complexity are easier to screw up and can be more sensitive to initial assumptions
It would be nice if the idiots who push climate models would take this to heart.
I hold the position that Biden - flaws and all -- is only one of these clowns who can beat Trump.
Hunter is looking to cash in BIG TIME!
No more million here and million there. We are talking Clinton-type dollars.
Unlike last time Nate's got top men working on 2020 projections.
Top men.
A very small sample size to be drawing any conclusions about anything, yet he shows the percentages down to single digits. Kind of irresponsible really...
Yes, his modeling and projection thing was more useful with baseball because a baseball player bats 600+ times a year. With a sample size of 15, what reliable conclusions can you really draw?
I had a dream last night that Biden secretly spoke with a stuffy English accent that he inexplicably forgot to disguise while giving an interview. I think this premonition is as helpful as most polls.
I shall state that with the current people (clowns) running for the Democrat nomination, not one could beat Trump.
Biden is right twice a day.
Your accuracy gets better as the probability approaches zero. (Or, what is the same thing, one.)
It's nomen's land.
I thought in 2016 that a Biden-Warren ticket would have won an easy victory over Trump. They’d have done well in the rust belt and Warren would have kept some of the Bernie energy as a proxy. I don’t think that’s true in 2020 because Trump has vast economic success in his column, and Biden is way past his sell-by date.
You'd have to separate out nomination campaigns run for smart people and campaigns run for stupid people. They're different, and you have to decide which you're going to do.
"over-complication"
Which of parts of the model he described should be removed or simplified?
This is the same "amazing science" that had Hillary at a 81% chance of winning in '16.
The only question that needs to be answered at this time, is just how bigly will Trump win?
"Top men."
Family guy reference. Ha.
Ryan:
"Top men."
Family guy reference. Ha.
You know you're getting old when even the pop culture references you recognize have been recycled and reconstituted.
I further hold the position that Grampa Biden is a doddering old fool and his adult son, Hunter, is a parasitic drifter.
But.
66 Million folks voted for Hillary. Lotta sweet, but confused, old timers out there will vote for Biden. (Exhibit A, my 85 year old father in law)
And, these fuckers cheat. Lotta illegal voting in the blue inner cities.
Warren & Bernie lose the general - too liberal, too statist.
If Biden gets the nomination, the Obama machine revs up and reluctantly supports him. He'll have a shot.
Lichtman’s 13 keys to the whitehouse are about the most promising predictor I have seen. It predicts basically whether or not the incumbent will win. I saw him in Chicago at the American Statistical Society conf. It does not have good measures of uncertainty but back casts well. Now everyone wants to stuff everything into a neural network and believe the output even though you have no idea what the model is doing. Right now the 13 keys are clearly in Trump’s favor, which is probably why no one is talking about it.
I thought from the beginning of this election cycle that the most formidable opponent for Trump would be Mayor Pete.
Whether he can get the nomination is another matter, but I have a bad feeling that he might.
I have little doubt the Democrats will leave no (grave) stone unturned in harvesting votes in November, 2020. To them an undocumented alien is just a potential vote to be garnered. Yes, garnered.
And they'll feel totally justified in doing so. After to all, Trump!
@Ryan:This is the same "amazing science" that had Hillary at a 81% chance of winning in '16.
That's why no one ever rolls 7 in Monopoly or craps or other dice games. There's an 83% chance that it won't happen, so you can just assume it won't.
Things with 81% chance of happening, don't happen 19% of the time, which is way more often than "never". Not sure why this is hard for people.
"Models with more complexity are easier to screw up and can be more sensitive to initial assumptions"
Now do global warming.
If he was really good at developing predictive models, he would become a billionaire
Silver/Yang in '24?
I agree with Ryan. When a guy calls his website "538" and so badly predicts the electoral college; he shouldn't be called amazing. There's a 19% chance snake oil might actually have some benefit to a person.
The correct phrasing is:
"Top. Men."
@Leland:There's a 19% chance snake oil might actually have some benefit to a person.
Sometimes drawing to an inside straight works. It's not the way to bet, but that doesn't mean it never happens, and it doesn't mean that the people who say you shouldn't are wrong when some dufus pulls it off.
For some time I’ve thought 2020 will be a redo of 2016 with Joe as “the guy who should have been the nominee in 2016”. If he becomes the nominee I believe Trump will win. However, this time around it won’t be the unspoken reality that “Hillary just isn’t likable” but more of a gradual sad realization that Joe is just too old, too frail, too “not quite there” to get the votes.
(The only wild card in the above conjecture is: “Will the Black vote come out to vote for “Barack’s VP” and/or “the white guy who can WIN!”?)
Althouse was being sarcastic re:amazing science.
It is impossible for Silver's predictions to be proven wrong. It's more like a horoscope than science.
@Lewis Wetzel:It is impossible for Silver's predictions to be proven wrong. It's more like a horoscope than science.
That's not true. See how many elections he calls correctly and compare that with chance. Elections are things that happen in exactly one outcome, they can't be interpreted away.
"It's poor strategy to draw to an inside straight" is correct even though sometimes you fill an inside straight. It's not a horoscope, and filling an inside straight does not prove the rule to be incorrect.
It's not science.
Silver is no Harry Seldon.
Re: Top. Men...
So I see the Raiders of the Lost Ark Scene. Is this where it came from?
Peter Griffin does it with a wink and nod so you know it's a reference to gay sex.
@Birkel:It's not science.
Not a bit. It's math, a very simple application of probability. He's probably doing it in R... Neither is poker a science, but nonetheless some bets are poorer bets than others and that is math too. Casinos can count on it sufficiently to make money if the mob doesn't siphon off too much.
Craps and poker odds can be accurately computed.
@Ryan:Craps and poker odds can be accurately computed.
Ah, you're moving the goalposts now. You were saying before that because Trump was elected, Silver could not have predicted the odds correctly. You seem to be retreating from that position.
Nate Silver hands you two dice and says your chance of not rolling 7 is 83%. You roll 7. Is Silver wrong? When I replace "dice" with "Trump" you say yes, but on dice you say no. Based on what?
Thats not what I said at all. I agree that Silver very well could have predicted the odds correctly.
81% was the probability that Clinton would win the popular vote, which she did. The probability of winning the electoral college was 71%, and he made clear that Trump still had a chance, that Clinton was much weaker in swing States than Obama, and if Trump overperformed nationally then Clinton's lead in Michigan and Pennsylvania would be severely reduced.
@Ryan:I agree that Silver very well could have predicted the odds correctly.
Then why the sneer quotes on "Amazing science" and why did you reference the outcome of the 2016 election? That's a very funny way of saying Silver could have got the odds right.
@Mr Wibble: The probability of winning the electoral college was 71%, and he made clear that Trump still had a chance, that Clinton was much weaker in swing States than Obama, and if Trump overperformed nationally then Clinton's lead in Michigan and Pennsylvania would be severely reduced.
That said there were definitely things that Nate Silver was doing that he later said were not the right things to do.
The thing he is known for is modeling and percentages. What he admits later that he did, was quoting percentages when they were NOT based on modeling, instead they were based on the same sort of verbal arguments that other commentators do. But they're not Nate Silver, not known for modeling, and so he (illegitimately in my view) let opinions pass as math when he should have clarified it was opinion.
But modeling elections, even the Electoral College, is a perfectly sane and simple thing to do, and no wronger than it is for poker, even when the less probable thing happens.
Gabriel, running his model is a trivial exercise in math. Building his model, though, is completely a matter of non-mathematical value judgements. It is not at all comparable to printing a deck of cards or stamping out dice.
@Tom T:Building his model, though, is completely a matter of non-mathematical value judgements. It is not at all comparable to printing a deck of cards or stamping out dice.
Certainly the model does not spring from the election returns like Athena from the brow of Zeus.
The test of the model is not one election, however high profile or surprising. If you didn't believe poker odds you could track the outcomes of hands and compare to the odds. The same is true for election models. SIlver predicts lots of elections. If he predicts the winner of 80-20 elections 80%, then he's doing it right for all the elections, not just the ones where the favored winner won.
It's no different from weather. If there's a 20% chance of rain tomorrow you would be foolish to assume it won't rain, and extremely foolish if you engaged in some costly activity that rain would absolutely ruin. And you would be foolish if even in the end it didn't rain.
Gabriel, elections are neither identical in the factors producing the outcome, not are they independent events. And I suspect you're mistaken in suggesting that Silver does not adjust his model for different elections.
What do you mean by an 80-20 election?
Bernie can't be nominated. 78 y/o with a bad heart. He needs to drop out and endorse Warren. Buttigig is a loser. Only two real choices, Biden or Warren.
The D's win when they fool people into thinking they are "Moderate" - Obama, Carter, Clinton and LBJ (1964).
That's why no one ever rolls 7 in Monopoly or craps or other dice games. There's an 83% chance that it won't happen, so you can just assume it won't.
The difference is that you can throw dice over and over again to validate your probabilistic model, whereas you can't with Presidential elections because each one is unique and only happens once.
Essentially, Silver was saying was that Hillary would probably win, but might not. Which is a lot more accurate than a lot of "analysts" out there who said that Hillary would definitely, absolutely, no doubt win - Megan Mcardle, looking at you. Still, Silver's phony mathematical precision has always bugged me. There is zero value in coming up with numbers like 81% in a one off event like a Presidential election.
"Acknowledged" is wrong in Althouse's first post in same way that "True" in the Bored Panda headline is wrong, and in the same way that it is wrong that Nate Silver's predictions are amazing science.
Gabriel might know math (unproven) but he is definitely not betraying much knowledge of poker. Explain implied odds, table image, and the perception of the player against whom you're calling to hit an inside straight.
And now it's not even math.
Nate Silver's "amazing science" gave an 85% probability to a HRC victory in 2016.
Scott Adams predicted Trump's victory.
Guess who I'm more likely to listen to.
Gabriel,
No. We could track the odds of Silver if he managed to predict the very close races with some high degree of certainty. But then he would have a tendency to rank less competitive races as very close to increase his own "accuracy", increase his public reputation, and increase his fees for appearances.
How does Nate’s formula weigh Hunter Biden? I still think there’s a very good chance the Trump Impeachment sinks Biden’s candidacy. With Iowa coming up, the Dems need this to happen sooner rather than later.
The problem with falling back to poker is this:
Poker has a scoreboard. We can tell definitively who can win most often. Money is a really good measuring stick.
Does the behavior Nate Silver involves himself in have a scoreboard? Not if you let him get away with "but there was a XX% that the other guy might have won. What were his confidence levels? What was his standard deviation?
Not science. Barely math.
I don't know who will win in 2020. No one does. Trump has peace and prosperity on his side (unless things change). On their side, the Dems have lots of TDS sufferers who beleive that Trump is literally worse than Hitler. These folks will beleive it is their moral duty to vote for whoever the Dem candidate is. That means voting and engaging in voter fraud. Plus, they will continue to have the media running interference for them.
So my advice to Trump supporters is: Don't get cocky. Even if you think it will be a landslide, act like it will be a nail biter. Because what if you're wrong?
Biden is this year's Jeb Bush. I think the narrative will be a lot different when votes start being counted.
Still, Silver's phony mathematical precision has always bugged me. There is zero value in coming up with numbers like 81% in a one off event like a Presidential election.
My biggest peeve with what Silver does. Representing greater accuracy with more significant digits than you should is the first sin in my eighth grade science handout on scientific method and representation.
Probabilistic prediction models mean never having to say you're wrong.
Nate wasn't wrong you see, his model gave Trump a 28.6 percent chance of winning!
"It's poor strategy to draw to an inside straight"
This isn't a real rule- there is more to it than that. You have to take into account the pot size, the size of the bet, and then also consider what might happen in the betting if you hit the straight. Good poker players draw to inside straights quite often.
@Birkel:Gabriel,
No. We could track the odds of Silver if he managed to predict the very close races with some high degree of certainty. But then he would have a tendency to rank less competitive races as very close to increase his own "accuracy", increase his public reputation, and increase his fees for appearances.
Sounds like you think he's acting in bad faith, whatever. That said, his "accuracy" (your sneer quotes) is not based on who won a particular election, because there's probabilities involved, but on do his percentages come out right.
If he started calling all races 60/40 then his percentages would not come out right. The guy he calls 60 wouldn't win 60% of the time. Others would quickly notice they could do as well as he by flipping coins... it would be very easy to demonstrate.
Matt may well be right. Biden's polling support may well be completely illusionary, like Jeb!'s was. Primaries and caucuses are quite different animals than general elections- more fervent voters show up, and significantly more partisan. I could easily imagine Biden 25% polling support in Iowa resulting in actually only 10% of the caucus vote, 25% polling support in New Hampshire resulting in only 12%, and 35% polling support in South Carolina only showing in at 20% of the vote. In other words, the polling support relies almost entirely on name recognition, and people who use that metric to answer a poll question are less likely to actually be active primary voters.
Gabriel,
You are wrong. No shame in that.
We could measure Silver's accuracy if he told us what the vote share would be for candidates and then gave us his standard deviations. Try to understand that.
Also, Gabriel, you are wrong about me.
I do not think people acting in their own self interest means they are acting in bad faith. I am saying people respond to incentives. Feel free to argue that point. I also believe most of the incentives are offered by the Left to those on the Left. Feel free to disagree.
Good point Yancey.
Silver's model spits out a number, just like polls spit out numbers. If it was ~81% would everyone feel better?
Generally, the reason that you build probabilistic models is that they help you make better decisions about recurring situations. Such as, in blackjack: Should I stand or hit on 17? Or in baseball: Should I intentionally walk Anthony Rizzo to face Kris Bryant with the score tied and a man on 2nd in the bottom of the 8th? These types of decisions about situations that happen over and over again.
Whereas Silver's model is just lending an air of scientific legitimacy to an exercise that is really pure political punditry. That is, Hillary Clinton has an 81% chance of winning the election and Hillary Clinton will most likely win the election are actually the same statement, but former has a fake mathematical precision to it.
AW: "Whereas Silver's model is just lending an air of scientific legitimacy to an exercise that is really pure political punditry. That is, Hillary Clinton has an 81% chance of winning the election and Hillary Clinton will most likely win the election are actually the same statement, but former has a fake mathematical precision to it."
It's even worse than that.
There is nothing "repeatable" about national elections. Each election contains its own unique parameters which are not effectively modeled.
Even things we believe we understand well (party registration, # of new voters, etc) are not well understood.
It's all one big guess and Trump blew out the millionaire pollsters whose livelihood depends on "amazing" and "wowing" the candidates, big donors and then, eventually, business/corporate interests.
Probability does not tell you who will win, just how to bet.
There is no one way a gay mayor is going to be elected President.
Ed Koch has been dead for years.
Bloomberg is going to buy the nomination. Many if not all of the super delegates took his money in their local elections. He will hire all the various political consultants from all of the other campaigns. Money for everyone!
Money talks and bullshit walks.
Bloomberg May buy the Democrat nomination, but he will be butt-whipped in the general. There are an estimated 100 million legal gun owners in the United States, ignoring the number who possess their guns illegally (e.g., inner city gang-bangers). After seeing what is transpiring in Virginia right now, there is no way any of legal gun owner will vote for him.
"Probabilistic prediction models mean never having to say you're wrong."
Weather forecasters have played this game for years.
Trump must be beat so it doesn’t matter who it is.
After seeing what is transpiring in Virginia right now, there is no way any of legal gun owner will vote for him.
I think this will be a burden for any Democrat. They have forgotten what happened in the 90s.
"Not a bit. It's math, a very simple application of probability. He's probably doing it in R... Neither is poker a science, but nonetheless some bets are poorer bets than others and that is math too. Casinos can count on it sufficiently to make money if the mob doesn't siphon off too much."
It's not probability because the sample size is one. Whatever the result, you cannot show that Silver was correct or incorrect.
I read somewhere Soleimani was polling at 19 percent.
Have you got tags like, “yeah but you were completely wrong last time” or “you had one job, just the one”?
Every election season, I need to remind myself that polls measure the estimated percent of the vote a candidate would receive, but Silver, and many betting markets estimate the probability that a candidate will win (in the case of this article "will win more than 50% of the first ballot delegates at the convention"). Both measures are numbers between 0 and 1, but they measure completely different things. And this year there is an additional complication because it is possible for Biden to have a 100% chance of winning every single primary and caucus and still have a 0% chance of winning more than 50% of first ballot delegates. (Note this is a mathematical possibility and not at all what Silver projects.)
the derp state covering the bases
https://www.bloomberg.com/news/articles/2020-01-10/u-s-probes-if-russia-targeting-biden-in-2020-election-meddling
There was one key action that Trump took, though, that I believed changed the outcome:
He made a lot of noise about the potential for voter fraud and not conceding if he appeared to have lost till his people took a look at it. This made the people in Miami, Detroit and Milwaukee report their votes as they came in with a much more limited ability to discover new votes once the rest of the state returns were in. It was a necessary and unremarked antidote to George Soros' Secretary of State project. If Trump wants to get reelected, he needs to not only again win Michigan, Wisconsin and Florida. He needs to push the voter roll clean up hard so that the counters in blue counties are nervous about playing games.
In theory there's no difference between theory and practice.
>> It's not probability because the sample size is one. <<
That's what priors are for. Welcome to Bayesian statistics :)
Amazing science is at least better than settled science.
"So while Biden’s in a reasonably strong and perhaps even slightly underrated position, it’s slightly more likely than not that Biden won’t be the nominee."
Maybe.....if we can re-arrange the deck chairs, then we can keep the ship from listing!!
Earnest Prole said...
In theory there's no difference between theory and practice.
But, in Practice, there Is!
Fritz,
Silver may not be Harry Selden but last year it occurred to me that PDJT may be the mule.
It's been 50 years since I read the trilogy so I had to do some research.
I came to the conclusion that PDJT really is the mule.
I also found that I am not the only one to have had this thought.
John Henry
As usual, the comments here demonstrate that most people don't understand probability. Silver's predictions are completely testable. Take all the times he predicted something with 70% probability, and see if 70% of them happened. Etc.
Apparently he does very well; that's how he revolutionized baseball with PECOTA.
>> see if 70% of them happened. Etc. <<
Problem is, you need a lot of elections to make that check. And candidates have personal qualities and strategies that are hard to quantify. It's as if each candidate has their own model and you need to model the models. My own impression is that Trump squeaked out a win by personally campaigning in selected states that he needed win, holding rallies right up to the end. Not everyone could pull that off, simple physical endurance was a necessity. Hillary's model, OTOH, told her that she didn't need to worry about some of the states that she lost. Like all contests, some elections are probably easy to predict, others can be full of surprises, accidents, and unknown unknowns.
>> I also found that I am not the only one to have had this thought. <<
I also posted that thought somewhere.
I think its closer to the atreides clan on harkonnen territory.
Ladbroke's has skin in the game.TRUST THAT THEIR ODDS REFLECT THEIR ACTUAL BELIEF.
I think you can roughly model human behavior in the aggregate, but you can’t model individual results, and if there is anything at stake, your model will be gamed.
"Apparently he does very well; that's how he revolutionized baseball with PECOTA.”
Baseball, you get 160 games a year, hundreds of at bats, there is no way that anybody makes an important decision in baseball after looking at just 15 games. It’s a different beast entirely. The error spread is too wide because the data is too thin. Besides, Silver has a point of view in politics he avoided in sports. In politics he thinks like a fan with a rooting interest. And you know what they say: “Think like a fan, sit with the fans."
"That's what priors are for. Welcome to Bayesian statistics :)”
The formulation of statistical models using Bayesian statistics has the identifying feature of requiring the specification of prior distributions for any unknown parameters. Indeed, parameters of prior distributions may themselves have prior distributions, leading to Bayesian hierarchical modeling[9], or may be interrelated, leading to Bayesian networks.. - Wikipedia
Even bayesian statistics can’t bootstrap experimental data, it’s not magic. It can just characterize the limited data that we have. It depends on assumptions, which are a huge vulnerability when there is a rooting interest in the outcome. If you are trying to figure out where a lost submarine may have finally landed on the bottom of the ocean, the process of making assumptions is a lot different than trying to dispassionately assess data about the chances of a man you have been programmed to hate by your whole social circle.
The reason that the climate models have performed so poorly since their inception thirty years ago is that the modelers see so much at stake that they are unable to formulate reasonable assumptions. It’s like on D Day when the bombers didn’t hit the Germans because they feared hitting our own troops so much.
Based on 3 years of data, I predict a 95% chance that Trump will prove experts wrong,
Monkeys might fly out of our butts.
Post a Comment