A Bet is a Tax on Bullshit

Nate Silver, whose models give Obama a high probability of winning reelection, has offered one of his critics a bet. “Putting your money where your mouth is,” is a time-honored principle of integrity in my view but the NYTimes Public Editor is very upset. Margaret Sullivan, however, never offers an argument against betting instead treating it as unseemly.

[Betting is] inappropriate for a Times journalist, which is how Mr. Silver is seen by the public even though he’s not a regular staff member.

“I wouldn’t want to see it become newsroom practice,” said the associate managing editor for standards, Philip B. Corbett. He described Mr. Silver’s status as a blogger — something like a columnist — as a mitigating factor…

…When he came to work at The Times, Mr. Silver gained a lot more visibility and the credibility associated with a prominent institution. But he lost something, too: the right to act like a free agent with responsibilities to nobody’s standards but his own.

The closest to an argument against betting is this:

…whatever the motivation behind it, the wager offer is a bad idea – giving ammunition to the critics who want to paint Mr. Silver as a partisan who is trying to sway the outcome.

My best parse of the argument is that by betting Silver has given himself an interest in the election and this hurts his credibility. Nothing, however, could be further from the truth.

A properly structured bet is the most credible guarantor of rigorous disinterest. In order to prove his point, Silver is not required to take the Obama side of the bet! At the odds implied by his model (currently between 3 and 4 to 1) Silver should be willing to take either side of a modest bet. Indeed, we could hold a coin toss, heads Silver takes the Obama side, tails he takes Romney.

In fact, the NYTimes should require that Silver, and other pundits, bet their beliefs. Furthermore, to remove any possibility of manipulation, the NYTimes should escrow a portion of Silver’s salary in a blind trust bet. In other words, the NYTimes should bet a portion of Silver’s salary, at the odds implied by Silver’s model, randomly choosing which side of the bet to take, only revealing to Silver the bet and its outcome after the election is over. A blind trust bet creates incentives for Silver to be disinterested in the outcome but very interested in the accuracy of the forecast.

Overall, I am for betting because I am against bullshit. Bullshit is polluting our discourse and drowning the facts. A bet costs the bullshitter more than the non-bullshitter so the willingness to bet signals honest belief. A bet is a tax on bullshit; and it is a just tax, tribute paid by the bullshitters to those with genuine knowledge.


'offers an argument against betting instead treating it as unseemly'

Some people consider that an argument - in other words, 'unseemly' means something not to be done.

Sort of like saying that spitting in public is 'unseemly' - the point is considered adequately covered right there, without exploring any of the other reasons that people do not spit in public.

Or why journalists should not be taking or placing bets on what they report - those who understand why something is 'unseemly' may not believe that there people who don't understand why.

Calling something unseemly isn't an argument - it's a conclusion.


Saying something isn't an argument isn't an argument - it's a conclusion.

Saying something isn't an argument could be an observation rather than a conclusion.

Sullivan appears to be implying that a portion of the public will find it distasteful or inappropriate for a Times reporter. That is a meaningful argument from the Times's point of view--their reputation is crucial to their business. If lots of people will object to something, that is a legitimate reason not to do it. It doesn't matter whether the public's opposition is well-founded or not (at least, not when the issue is something of peripheral concern, like their employee making a bet).

I agree fully with Alex's take on the underlying merit of the bet. It would have been nice if Sullivan had made this case as well. Nevertheless, her argument that the bet is unseemly is a real argument.

Unfortunately, it is much too late for the NYT to worry about their reputation.

The Times is America's most important newspaper, whether we like it or not. I strongly disagree with the paper's ideological slant, but I still find it essential reading.

You are correct, using a word like "unseemly" is taken, by the user, to be the full argument. This is what makes such words, especially when used as the sole premise for a claim, a "begging the question" type of fallacy of the "loaded language" sub type. An argument is not presented with specific reasons and explanations, e.g., what makes it unseemly? To whom? Why is that sufficient for prohibition of a given behavior? Etc. Instead, the build-in evaluative freight of the term is used to short-circuit such discourse.

I was following you until this "At the odds implied by his model (currently about 3 to 1) Silver should be willing to take either side of a modest bet."

This makes no sense. Why would you bet against the odds favorite? Also, why would you have him take a bet based on a random 50-50 coin toss. How would that tax bullshit at all? Do you understand the first thing about statistics? This might actually explain a lot about economics.

@Benny Lava
In betting, the phrase "at the odds" implies that Silver would be willing to stake $1 for a $3 payoff on a Romney win. That's how betting odds work - the payoffs are weighted by the probabilities to make expected values equal ($1 * 75% = $3 * 25%). If the odds match your personal estimate, you would be indifferent between either side of the bet.

Sure, you'd be indifferent to which side you'd take if you'd bet often enough that the payoff would average out.

But this isn't a casino. This bet only occurs once every 4 years--the average payoff isn't meaningful.

The bet would need to be handicapped. If Silver's opponent is confident that Obama may win but not by a 70-electoral seat margin then they could set the handicap at say a 20-seat margin in favour of Obama and Silver could take Obama to beat the spread. If Obama wins by 70 electoral votes as predicted, Silver would be vindicated in his prediction. If Obama only wins by 10 seats, Silver's opponent would be vindicated because the race was a "toss-up" and not as stacked for Obama as Silver suggested (and implied because of his partisan position).

Silver's predictions could be broken down into a huge quantity of bets: absolute winner of each state, margin of victory for each state, as well as predictions for each congressional race.

An accurate model would pay out $1, but systematic partisanship (and other failures) would result in a substantial haircut.

What you're talking about is a strategy for the more risk averse ( http://en.wikipedia.org/wiki/Risk_aversion ), which has less to do with likely outcomes and more to do with the evolution of human psychology (and constrained resources), I think.

I think this comment explains a lot about you


If you are going to be smugly condescending, it would be well to make sure you know the first thing about gambling and odds.

Yikes, somebody can't read can he? Nate offered a wager with terms of 1:1 payout. Changing the payout would make sense only if you thought Silver's model was accurate. Since Scarborough thinks it is 1:1 odds the wager had a 1:1 payout. But then reading has never been your strong suit has it.

I read perfectly fine. You didn't write what you claim to have written in the 1st comment, and this second comment of yours reinforces the notion you have no clue what you are writing about, or even understood Tabarrok's post.

All Tabarrok wrote is that Silver himself should be willing to take the other side of the bet by offering the odds he himself calculates. If Scarborough thinks the odds are 1:1, he will be more than willing to take the modified offer, not less willing, even if it means he ends up betting on an Obama victory.

Lol. Hahaha.

You are so incredibly dumb. Before calling out a econ professor (I presume) on maths, don't you think you should have put some thought into the maths first?

I am with Ward. You are so stupid, you don't even get you are stupid.

Benny's comments may be deserving of criticism, but..."LOL"? Who's lowering the discourse now? Also, you are imputing statements to Ward that he did not actually make.

“At the odds implied by his model (currently about 3 to 1) Silver should be willing to take either side of a modest bet.”

Alex's argument is less than clear here. What I understand him to be saying is that Silver should be willing to take either side, given 3-1 odds - i.e. $100 if Obama wins, $300 if Romney does. This is confusing since Silver offered the bet at 1-1 odds. The modest aspect has to do with risk neutrality.

It took me a couple of passes through the article to come to this conclusion as well.

@finley joe Scarborough put the odds at 1-1 so silver challenged him on that. I think Alex was recommending to silver a less controversial but equally effective way of making his point.

Silver accepted the bet at 1:1 odds because he believes the true odds are 3:1. He isn't indifferent at 1:1, he wants Obama for sure. At 5:1 he wants Romney for sure.

I have several bets with friends at 1:1. They take this to mean I wish Obama to win; however this isn't necessarily the case. I offered to take Romney at 4:1 odds but there were no takers. I'd even bet Obama at 2:1 but my partisan friends seem to think taking odds demonstrates a reduced commitment to their guy.

Same, I have a bet on the election and got better than 1-1 odds on Obama since I bet a richer friend of mine a "nice dinner". Hopefully Obama wins and I get a really nice dinner out of it. If Romney wins I have to buy and we probably don't eat as expensive a dinner and the wine won't be as pricy. Pro election betting tip, find rich people that watch too much Fox News to bet with and you can get some really good odds.

This is probably one of the greatest Marginal Revolution posts of all time.

(I am willing to bet that belief. Contact my email for details.)

I agree completely.

Intellectual jujitsu at its finest

You miss a point here. If your odds are 3 to 1, and someone is claiming the odds of 1 to 1, then you will try to bet based on his odds if you are sincere in your estimation. You suggest that Mr. Silver should bet based on his odds. That would be just stupid. You have the potential to make money when the estimates made by the others are completely off.

I don't read that at all. The two arguments are not tied.

I guess one could read "Silver is not required to take the Obama side of the bet!" to mean this specific, 1:1-odds bet. The definite article is misleading: perhaps "Silver is not required to take the Obama side of [a] bet [given different odds]!"

He should be willing to take the Obama side all the way to where he's modeled the odds - call it 3:1. Then willing to take Romney all the way up past that.

Note that the expected value of Alex's "blind trust bet" is zero over time; I'd toss in risk aversion as a reason this won't come to pass: most people won't take even-money bets.

The whole point is that the long run EV of the trust is zero. It's not supposed to be a money-maker, it's supposed to signal just that, zero-bias plus skill at setting the odds. Of course, it will be non-zero based on either the odds-maker BSing or the odds-maker being bad at his job. But that's a feature, not a bug.

Put Romney down for $10K of action here.

Of course, the secular economist is essentially suggesting that the New York Times and other news outlets institute a religious test for employment. Some Christians regard betting as immoral and tantamount to covetousness. Some argue this extends *even* to betting in which all winnings go to charity as in this instance because it nevertheless encourages the continuation of an immoral practice. You could argue that the type of structured betting being contemplated here, like prediction markets and ideas futures, is not like normal gambling. Alternatively, it might become standard practice for journalists and commentators, in which case you'd make such betting a bona fide occupational qualification.

Some Christians = very few

The funny thing is that Silver offered the bet with the winnings going to the Red Cross for either wagerer. So in that sense the immorality of the wager is reduced to almost nothing.

I don't know. Look at how difficult it's been to get regulated internet poker back. A lot of the pushback is from folks who may not be pushing an explicit Christian argument, but are definitely pushing a social conservative argument against money betting in things with an element of chance. There are a lot of these folks. I'd have guessed 10 percent of the country. How different is this from the criticisms you see of hedge fund managers and bankers which imply it's evil to be a speculator?

The other opponent of internet poker is the hardcore liberals who want a monopoly for their state lottery systems so they can extract more from the population. They want casino games because they're a better deal for the state, and prohibitions against competitors. It's a funny world.

A significant portion of Christians do oppose gambling for personal gain. However, I think that very few would when a charity is the recipient of the winnings.

Hence the ubiquity of church raffles and suchlike.

So recruiting for their presses ( which run on the Sabbath ) is a religious test too?

...and covering religious issues without being allowed to claim '[my side!] is correct, of course!', or .....

Prediction markets have even less to do with "gambling" than stock exchanges. Do these same christians oppose existence of Wall Street and retirement funds as well?

Buying stock is taking part ownership in a company that is producing some good or service. Even when doing something like trading commodities, you are contributing to something being produced, though more indirectly.

The stock market may look similar to gambling, but the difference is that there is something "real" going on underneath the trades. Gaining through participation in a productive activity is morally different than gaining through a pure chance event, such as a bet on the future.

You may argue that some activity on Wall Street is just manipulation, without anything real being produced. But lots of Christians would oppose that kind of activity as well.

How about derivatives base themselves solely on the change in value of stocks, are these out then?

I suppose that would depend on whether one thought derivatives were providing a useful service. I'm not knowledgeable enough to answer that.

I agree that everyone should back up their beliefs by wagering on them, but I think its far from certain that Silver would back his own model against the line at pinnacle or betfair, though maybe he would against Intrade.

It wouldnt really make sense for him to take odds worse than what is publicly available anyway. Even if he truly believes in his model he should just kelly criteria the hell out of the best price available.

but yea lol NYTimes.

The irony being that he explicitly said he doesn't bet on InTrade in his book.

I think he's just pissed that people from Politico (Tiger Beat on the Potomac) and Morning Joe are bitching.

"I think he’s just pissed that people from Politico (Tiger Beat on the Potomac) and Morning Joe are bitching."

He's probably pissed that those people can't raise a legitimate complaint to save their lives; their complaints come down to 'I don't understand stats 101' or 'his conclusions are unpleasant' or 'he's showing that we're just BS artists'.

'A properly structured bet is the most credible guarantor of rigorous disinterest.'

Some people have noted the point, but let us be more explicit. A casino does a fine job of making a properly structured bet, and though I wouldn't personally describe a casino as being an institution of rigorous disinterest, it does attempt to be a credible guarantor. Of ensuring that all properly structured bets favor the hourse, of course.

As some have explained, Silver would only be smart to be betting if he was setting odds to ensure his risk balanced his chance of profit.

Anyone want to guess how it just might be unseemly for the person who is setting the odds and taking bets to be the one also reporting on the situation, thus influencing the very environment in which the odds are set?

Or to put it differently - someone reporting, setting odds, and taking bets is not what one thinks as an example of 'rigorous disinterest.'

Alex addresses the conflict of interest concern by suggesting a blind trust. Forward to futarchy!

I think we could agree on "accurate" if he is taking bets on both sides, and for someone in the prediction business that is what you want.

People taking bets on both sides aren't testing their beliefs, but arbitraging other people's beliefs. What's important is that the person with the most extreme belief puts his money on the line to force other people to deal with his beliefs.

Hmmnn, I don't think that's quite the point. Consider the situation where you are to split some amount of stuff with another party, fairly. A way of approaching this without having to appeal to a third party is for you to make a proposed cut in the assets (making a Pile A and a Pile B) and then allow the other party to pick either pile. You are, as the "cutter", doing something equivalent to setting odds on the election (or any other event.) However, by virtue of letting the other party pick either pile you are signalling that you have made as fair a cut as possible (that is, reflective of reality as you see it) and that you are now indifferent to which pile you end up with.

By analogy, Silver is effectively setting odds via his style of reporting/model building. The best way to ensure that he is actually setting "fair" odds that properly reflect reality, that he is actually reporting/model building as an honest disinterested party is to be willing to take either side of those odds. Those expecting him to do the "smart" thing are also missing the point. The bet isn't supposed to be a money-maker, it's a way of proving disinterest by saying that the expected value is equivalent on either side. He always has the option to put his thumb on the scales, just as the person above can try and make either pile of "stuff" more valuable, but then he would run the risk of being gamed by the people he's trying to (falsely) prove his honesty towards. i.e. If I have a super-duper model that indicates that Silver is full of crap and the real odds are 1:1, he'd be a fool to offer 3:1 in either candidate's favor because I'll take every bit of that action I can get.

This is not about Nate Silver. This is about the NYT. NYT has more to lose than Nate Silver does.

I think Alex would be correct if Nate played an acknowledged partisan role, the way say CNN would have a Democratic or Republican partisan on TV. Carville and Noonan should take bets, not Silver.

This seems more to me like a variation of a principle-agent problem.

He plays an acknowledged prediction role.

"This seems more to me like a variation of a principle-agent problem."

Alex is off base here, because he's ignoring the rules that the NYT requires of it's contributors. It's highly likely Nate Silver was breaking rules he agreed to abide by and Joe Scarborough would be breaking MSNBC rules by accepting.

You can argue that the rules are wrong, but that doesn't make Mr. Silver's actions right. If you agree to the rules you follow them, if you don't want to follow them, go get a job somewhere else.

But the NYT needs Silver more than Silver needs the NYT -- which puts them in a difficult position to enforce silly ad hoc rules. He should place a bigger bet and tell the NYT to go screw themselves.

So if Obama's odds of winning were at 80% yet he still manages to lose, how is Silver wrong? Obama's odds of losing were at 20%.

I don't get the Nate Silver thing. Of course there's a lot I don't get because I'm kind of obtuse.

Completely agree. Most people who see a 20% chance of rain in the forecast will pack an umbrella for the day. We certainly wouldn't say the meteorologist was "wrong" when the rains do come.

That's not the way you look at it for multiple trials of the same system, or even multiple trials of similar systems. The meteorologist isn't wrong if he predicts a 20% chance of rain and it does happen to rain, he's wrong if the "bucket" of days he predicts a 20% chance of rain for (say, 30 each year) actually have rain 50% of the time. Likewise, Silver will not be "wrong" if Romney wins, he'll be wrong over the long-haul if he regularly gives better odds of winning to losing candidates and vice versa.

He isn't necessarily. You would need multiple trials to say anything meaningful. But anyone placing a large bet at least believes what he is saying.

Silver's authority comes from his track record of correctly predicted results in multiple elections. The way to evaluate his record after this one is not to look at the winner of a single race (the national presidential election), but to look at the winner of each state, their margin, and the winner of each Senate race and their margins, (all of which Nate has predicted) then compare this to all the other predictors and see where Nate stands. Historically, he has been at the top. Even without betting, his reputation, which earns him his livelihood, is on the line to a significant degree. He makes hard predictions (who will win, by how much, and how confident he is of that result) that are easily testable after the fact.

But he assigns probabilities. He doesn't just pick a winner and a loser. This is claptrap to me.

I know of, although I'm unfamiliar with, the element of probability theory that deals with the accuracy of statistical predictions. But the way I see it, in order to test the accuracy of "Obama is 80% likely to win this election" we would have to get a hold of a bunch of clone universes and repeat this same election--with every last one of the innumerable factors at play unchanged-- over and over and over until we have a large enough sample of Nov 6, 2012 general-election results to determine if 80% is an accurate figure. Clearly this is impossible.

Comparing the results of different races assigned with different odds tells you nothing about the odds of each race.

It would impress me more if he simply picked a winner. No hedging with probabilities.

Also, his track record, although not nearly as solid as people seem to think (see Scott Brown race two years ago), means nothing to me. He's like 13 years old, for one. So hasn't been analyzing too many presidential elections.

He's not hedging. He is saying his model suggests that Obama will win and then assigns a confidence to that prediction: 80%. If the confidence level was 51%, he would be saying that the race is actually a toss-up, but as it stands, he's pretty confident (or rather his model is) that Obama will win. In other words, it's better to think of the 80% number as expressing confidence rather than probability. Note that Nate's proposed bet with Joe S. contained no probabilities; it was for an Obama win, and he's confident enough to back it with $1000.

Note that confidence also helps understand how to evaluate his model. If his model works, you'd expect it to be better at predicting races where it has a higher level of confidence. If it consistently gets high confidence predictions wrong, then it is flawed.

I think you need to study more probability before calling Nate's analyses 'claptrap.' Just sayin'.

"Chances of winning: 80.9% Obama -- 19.1% Romney"

That's what it says at his blog. Not, "Obama will win, my confidence level is 80.9%."

It's the difference between saying, "This coin will come up heads when I toss it. I'm 80.9% sure." and "This coin has an 80.9% chance of coming up heads when I toss it."

I see a difference. But hey, what do I know? I need to study more probability.

Just what is the difference? In Nate's simulations (which he runs 10,000 times via the wonders of computers), Obama wins 80% of the time. The coin comes up heads for Obama 80% of the time, that is.

I am also more impressed with Divination than Analysis. Analysis has the better track-record though...

Your criticism is showing a conflation of the difference between a probability and a confidence interval. When Nate says "Obama is 80% likely to win this election" this is NOT a probability, it's a confidence level in his projections. The random event is not the outcome of the election itself, which is not going to be happening repeatedly, but only a single time, so probability doesn't apply. The random events are the *sampling* of the voting populations by the various polls. The confidence level indicates how likely those polls are to represent the true voting preferences of the populations from which the random samples are drawn. What he's saying is that if all these polls that he's basing his projection on were repeatedly done in an infinite number of universes, they would produces results leading to him making a correct projection 80% of the time.

He doesn't make a prediction with a CI attached to it. You can keep reading into it people, it doesn't change anything.

He says "chance of winning." Like the chance of my winning Powerball is 1 in 175M. As in, I should expect to win 1 out of 175M times. This doesn't work for a non-repeatable event. The Romney v Obama 2012 election is a non-repeatable event. You can't simulate hundreds or thousands of 2012 elections in a computer model. Elections aren't coin tosses or lottery tickets. They are all unique.

I may be able to divine something about how Silver predicts generic elections by looking at a large set of his predictions, but it won't shed any light on the accuracy of his prediction regarding this particular, unique, election.

If I knew nothing about the probabilities associated with a coin toss, I could guess the chances of it landing heads up arbitrarily. I could say there is X or Y chance it will land heads up. And being aware or, perhaps, unaware of my own ignorance, I could assign a CI of 50% or 100% or 1%. Regardless, it doesn't affect the true probability of the outcome. It remains 50/50, despite my CI.

Silver offers his prediction as a probability. 80.9% Obama victory, as in Obama wins 80.9% out of 100 elections and Romney wins 19.1%. Not, Obama is going to win and my CI is 80.9%. There is a difference people.

If Obama wins, there is no way of proving he had an 80.1% chance of winning vs a 50.000001% chance. And Silver can get every other prediction correct, it won't shed any light on how likely it was that Obama would win, only how likely the guy who is favored by Silver's prediction wins.

That's why I say pick a winner. Don't bullshit me with fake-ass statistics.

The key is the odds. If the odds are fairly 3:1, Silver wins $100 if he bets on Barry and Barry wins while his opponent wins _$300_. He has no reason to overstate Barry's chances because he's taking on additional uncompensated risk. Imagine if the "real" odds are 1:1, each candidate has a 50% chance of winning. If Silver offers 3:1 odds he is now taking on a 50% chance of winning $100 and a 50% chance of losing $300. The expected value (EV) is against him due to his own false reporting of the odds. Even better if the opponent can choose either side of the bet, as it basically puts a cap on Silver's willingness to lie in either direction for any reason. If he consistent lies (or is just not good at setting the odds), he will get gamed and will lose a bundle.

My comment was not about the bet. It is about the Nate Silver racket in general.

I actually like the bet. It eliminated the hedge. 2K from JS if Obama wins, 2K from Silver if he loses. There are no odds at play in this bet. No tiered payout schedule.
Where are people getting this idea this this is a day-at-the-dogtrack type bet? Sincere question, I could have missed something.

Similarly, trading profits are the tribute uniformed market participants pay to the informed. Or more nobly, the justifiable price of keeping markets efficient. Here, betting profits would be the price of keeping public discourse honest. In both cases it's about the cost of supplying high quality information in an unbiased fashion. In the case of newspapers, however, the fix has been in for centuries. There is no reasonable doubt that they are pipe organs for their wealthy owners, is there?

All you need to know is that Margaret Sullivan is the Times' "public editor." At any other newspaper, that position is called ombudsman. This is about brand management, not the validity of betting.

And Pete Rose was also just about 'brand management.' It isn't just the NYT that has an institutional aversion to gambling - and in sports, bullshit is not answered by betting on an outcome, it is answered by creating that outcome.

Now you're not making any sense

Nate Silver could sway this election one way or the other? Dude is way more awesome than I originally thought!

It's possible via the following mechanism: Nate predicts a Romney win. D turnout is depressed. Romney wins. It's also possible to go the other way, but I think there's a lot of noise on the Romney side
that's drowning out Silver.

Allegedly, this is why some countries have laws about reporting results prior to polling being closed.

Bravo, Alex!

Here's a little wager. Every day a maximum temperature record is broken, anywhere, you pay me $x. Every day a minimum temperature record is broken, I pay you $x. We keep going until one of us admits defeat. I invite any takers to suggest a value for x.

I love it! It reminds me how Lloyd's of London, the oldest insurance market in the world, has already started "betting on" global warming. http://www.guardian.co.uk/money/2006/jun/06/business.environment

x = -100

How's that for a start?

I still wish I could have made that temperature bet with Silver, would have made a bunch of money.

Believe in taking bets is one of the most important things I've learned from MR and Bryan Caplan.

There's just one problem: betting requires courage, and courage is in very short supply in this day and age, especially when it comes to things like ideology and public image.

I'll wager that is not true. And I have the courage to set the odds before I will let some coward influence my noble stand to take your money.

And don't make me double dog dare you to show everyone just how serious this entire discussion really is.

I don't even know what you're talking about now.

I also have the courage to wager that your lack of understanding about someone having the courage to take your bet about people lacking the courage to bet is true.

Even without having to double dog dare you to show how such 'betting' is more of a joke (and a distraction from real politics) at this point than anything else.

He didn't offer you a bet

Sorry, p_a. I'm still confused. You help me by answering the following:

1 - What is the meaning of the phrase "real politics?"
2 - What bet are you offering me, specificially?
3 - Based on what you said, your bet seems that your bet is, "RPLong's lack of understanding... is true." What is the meaning of this phrase?
4 - How will you know who has won the bet?
5 - Why do you find what I've said so objectionable?

Silver is already betting - what's at stake is his reputation. By having odds that diverge so significantly from In-Trade he is already staking his hard-earned (and valuable) reputation on one election. If he is proved wrong - It will be difficult to come back from this easily.

The 1,000$ is just distasteful and also considerably lower than the costs associated with a loss in reputation. Perhaps some of us are just old fashioned (I'm in my 30s !).

Silver's actual bet is ill-defined and serves him poorly, in that case. Even if he were predicting a 90% chance of an Obama victory, that would mean there's a 10% chance of it coming out the other way *if he's correct in his analysis*. The outcome of the single election therefore proves nothing, and if he has wagered his reputation on it, he's made a mistake. 0.71 =/= 1.00. The point of the system Alex describes is, in fact, a viable way of actually staking one's reputation, by generating an EV=0 (that is, a fair) portfolio of bets.

Anyone who thinks the race is a tossup will be indifferent to Silver's bet. He could offer a bet with odds of say 1.5 to 1 and Scarborough would no longer be indifferent to the bet. A 1-1 bet offered to someone who thinks it's a coinflip is bullshit.

Everyone's made great points, e.g. why he could accurately create odds and take either side, but believe that rather than a coin toss he should randomly be assigned to one side based on probability (e.g. if he says the odds are 50/50, flip a coin. If he says they're 200/1, make 200/1 odds that he takes the favorite). Otherwise, the incentive is to create outsized odds to maximize your winnings. Assuming the bet is $1, and he assigns a probability of 200-1 that Obama wins. If you flip a coin his expected winnings would be .5(1x200+1x.005)=$100.0025 right?

Oh, this is the answer to Granite26 below I think.

UIVMM, The numbers don't even come close for random betting.

Take a case of an assumed probability split of 25A/75B This would imply a 3x payout on A 50% of the time, and a x.33 payout on B 50% of the time.

If the actual probability was 75A/25B, your expected payout would be 3 * .75 *.5 + .33 * .25 * .5

That adds up to an expected payout of 1.17x your bet.

Considering that the bet is designed to be neutral if the prediction is accurate, you don't want to reward people with more (expected) return if the prediction is inaccurate.

Good point. What about for more realistic numbers? Maybe require him to accept bets from others on either side.

Why not buy contracts at Intrade? I would if the US government did ban me from using a credit card.

the point is Nate should be giving odds - that's his CLAIM.

No. Scarborough's claim is that odds are equal. Silver does not believe they are equal, so he offers the bet at equall odds to call Joe on his claim the odss are 1:1. If Joe thinks the odds are equal he would be smart to counter Nate with an offer of a bet at 2:1 odds. If Nate believes the real odds are 3:1 and Joe believes the odds are 1:1 each should readily make a bet at 2:1

This, a thousand times this. Tabarrok's article completely misses this point.

Thank you! Alex has clearly misstated the proposal. At 3:1 odds Nate is clearly disinterested in either side of the wager. His willingness to accept Joe's 1:1 shows that Nate stands by his prediction. I think the partisan BS is easy to spot.

No, Alex has offered a new proposal that puts silver's money on the line as a matter of routine for his continued employment.

I agree on the value of betting.

If Silver's model is unbiased, and we run the 'experiment' many times, he
would break even if he bet the odds according to his model.

Two small comments.

1. Although Silver should be willing to take either side, he should not be forced to bet
the fair value odds implied by his model. After all, if his model is unbiased, he would on average break even if he did so. What he should be allowed to do is bet with odds that deviate from his model's. That way he would win on average, assuming his model were superior to his opponent's.

2. It would take a number of elections for Silver's winnings to be convincingly positive even if his model were correct. If he were, say, betting against a (biased) model that said 50/50 Obama/Romney while his (unbiased) model said 75/25 Obama/Romney, and he agreed, say, to pay two dollars if Romney wins and receive one dollar if he loses (63/37), then on average he would win a quarter, but the standard deviation would be roughly five times that. You'd need to run the 'experiment' ~ 25 times to get even a rather low sharpe ratio of one.

It's won't really be difficult to find out if there is bias in Nate Silver's model after the election. All you need to do is look at his predictions for each state. He has 50 Obama/Romney state predictions. If his model is non-biased, then he should have overstated Obama's percentage of the vote 25 times and understated it 25 times.

If after the election his model has produced a result that significantly deviates from 25 overstate vs 25 understate then the model is biased. A standard deviation for n=50 would be roughly 3.5. So if Nate's model is off by 4 or more it's over 1 standard deviation from the expected norm. If it's off by 7 it's over 2 standard deviations from the norm.

This won't be a hard check to perform after the election and it won't matter who won the election. 50 samples is enough to get a good idea if there was any bias.

How does one take his confidence intervals into account? I mean I could construct a model that gives you a 25/25 split right now: Obama gets 0% of the vote in the first 25 states alphabetically, and 100% of the vote in the second 25. Done, and my model is inherently better than Nate's right?

You are correct that there is additional information in the individual state forecasts and this could be used to try to asses biases in Silver's model.

However the outcome of state elections are certainly not independent events; deviations from expectations are presumably highly correlated (imagine for example that Obama were discovered before the election to be having an affair or if Romney were discovered to have lied about his income taxes).

So the additional information in the outcome of state elections is (presumably much) less than it would be if state election results were independent.

That said, even with the highly unrealistic assumption that the 'noise' in state by state election events is uncorrelated, and Silver bet in each of the markets according to the scheme I outlined above, his Sharpe ratio would still only be ~1.4 which means there would be a non-negligible chance that he would lose money overall in these bets. Given that noise is correlated, we it is even more likely that Silver could lose, even if his models were unbiased.

My take away is that betting is good, but one needs to make many forecasts and many bets to get reasonable stats.

Your approach won't give adequate statistics because the results are already known for about 40 of the 50 states and if unweighted will bias the results. The difficulty is that the sample size for the Presidential election alone is too small. If you take the 10 tossup states (probably less but 10 is a good number), Silver could miss 2-4 of them and still have called the election correctly because of the differential in electoral vote. So you then have to ask yourself whether you need to weight individual states differently which complicates the analysis even more.

Silver offers a predicted vote share for each state, which could be used to test his predictions after the election, regardless of how well known the winner of each state is, or its weight in the electoral college.

"If his model is non-biased, then he should have overstated Obama’s percentage of the vote 25 times and understated it 25 times."

The problem is that there is correlation involved... whether in events that happen between the polls and election day, or in bias of the polls, or whatever. Therefore, if Nate's model predicts that Obama will win 60% of the vote in Vermont, but he only wins 55%, then one would expect that the model will have also overpredicted Obama's vote share in New Hampshire and Massachusetts. Nate's model takes into account this correlation: if you take his state by state odds and use them blindly to predict a national winner, it would actually have Obama significantly higher than Nate's more complex model does.

So you really do need actual independent elections to test the model.

(think of predicting 50 numbers. Each number is made up of the sum of two random variables: the first random variable is independent for each, but the 2nd is common across all 50. A model should be judged by how well it represents the random variables. An unbiased model will, on average, guess high on each variable as often as it guess low. The problem is that one cannot observe directly the half of the sum that comes from the independent variable, so if the model guesses too low on, say, 40 of the variables you can't tell if that's because the model was biased low, or because the 2nd, common variable happened to have a high outcome. Think of this second variable as representing the "national mood"... but add in other variables which might be "the weather on the East Coast the day of the election" or "the latest job report which came out after the polls that went into the model").

"The problem is that there is correlation involved… whether in events that happen between the polls and election day, or in bias of the polls, or whatever. "

That's a valid point, but only to a limited degree. And frankly, that's my whole issue with the cult of 538. It's an attempt to model an essentially chaotic system. The prediction (at least as presented by the column itself) implies a level of confidence above what is justified by the underlying data.

This is the most retarded thing I've ever heard.

We don't need to incentivize Nate Silver into producing a more accurate prediction model because he has all the incentive he needs riding on this election right now, credibility.

Plus, The point of FiveThirtyEight isn't to pick a stone-cold lock for the election, its simply to use the data available to construct a model that would help see what chances a certain candidate has of winning an election. For instance, if the Packers and Bears were playing right now and the Packers had a 7 point lead with 4 minutes left in the 4th Q and the ball, then you could say that the packers are a huge favorite to win the game at that point in the game. That's not to say that they are guaranteed to win the game as the bears could still score and tie the game, etc.. but the odds are certainly in the Packs favor.

So regardless of the outcome of the election, Nate Silver can STILL say his models were correct. If Romney wins its because model had built in a 20% chance of doing so, not a 0% chance.

Credibility is an imperfect incentive,else we wouldn't still hear the prognostications of the Dick Morris's and Bill Kristol's. Those are rather extreme examples, but the point is that news media place very little emphasis on the correctness of pundit predictions.

A model cannot be correct regardless of outcome. An unverifiable model serves no predictive purpose. Indeed, most useful models fail in a number of ways.

One difference between Silver and most other election pundits is that Silver probably thinks his job is to be right, whereas most political pundits' jobs are much more about being entertaining to their readership. Political pundits are routinely wrong about everything, and often (especially in interviews) turn out not to know really basic stuff about the subject they're discussing. This has almost no impact on their careers, because their job is not really prediction, it's filling airtime with something that won't make the viewers change the channel.

I think Nate's credibility is the perfect incentive for him considering his platform. Most people familiar with 538 don't see him as political entertainment (as I view pundits), they view him as a math wizard taking a statistical, scientific approach to the election. If his math or assumptions in his models were flawed, then nobody would pay attention to his analysis; Whereas pundits get ratings from making outlandish (note: not always correct) statements.

The model could certainly be correct regardless of outcome of one election. Anybody who has taken a stats class knows that only one sample is not enough to prove (or disprove) his predictions. Now if we could hold the election about 100 times, then we could start to make assertions about the models accuracy

I have much respect for Nate, but to most readers his methods and conclusions are opaque - which means that there is a perverse incentive to skew toward audience expectations, rather than deliver well-studied predictions. The scent of science is easily used to cover bullshit.

You are confusing the map with the territory. A probability is not a measure of frequency of occurrence, it is a measure of belief. If we could rewind time and hold the election 100 times, the outcome would still be the same. If you flip a coin, it is common to predict heads with a probability of 50%. But of course, if one had more information (the amount of energy stored in the thumb prior to release, the air density, the mass of the coin, etc.) one could know the outcome with much greater certainty.

When one takes a series of measurements of a field, there will be errors due to limitations in our measurement methods. That doesn't mean that the field is randomly changing in size. The measurement has uncertainty due to limitations in our information gathering tools. But the measurement can still be called inaccurate or accurate when verified against methods with higher certainty.

Nate's model (the map) is a prediction of electoral outcome (the territory). If the model has high uncertainty, then it is a useless model - one might as well hire a monkey. If it has high certainty and is wrong, then there were methodological errors.

I'm pretty bad at this; you're better off reading Eliezer Yudkowsky's older stuff at lesswrong.com or overcomingbias.com

Hear, fucking, hear!

Am currently reading Nate Silver's The Signal and the Noise and Tetlock's book on political prediction, and I would note two things regarding political predictions:

1. Silver took political predictions made by TV pundits and found that you would be better off flipping a coin;

2. Tetlock demonstrated that ideologues did worse in predicting than foxes--non-ideologues who drew information from many sources.

As to betting, I agree with the commenters above who say Silver's reputation is more valuable than his bet. I would note though--and, I don't think he does this, because his model is dynamic and would prevent him from doing it if the polling data changed--there is a risk that he may be human and ignore data that is inconsistent with his beliefs, and adding a bet too it is not a good strategy. At the same time, let's not forget the origin of the bet__it was someone else challenging him to a bet, not him going out seeking bets. Had he said no, he would have been disbelieving what his poll is telling him.

I'm betting the South with military bases and Farm States which have large farm populations and which receive large federal agricultural price supports, quotas and guarantees will continue to vote Republican.

The bet is not to incentivize Silver. He has enough incentive. It is to incentivize Scarborough who has attacked Silver's model to stop his bullshit.

I personally don't think the money is the real point of betting. The two important things are getting people to define exactly what the mean in their prediction and assign probabilities. And second to hold people accontable on the public record for cheap talk.

Moreover, betting by journalists has a long legacy in American History. As a historian of American politics, my college students are always shocked when I show them evidence of published bets on elections in every election cycle from 1790 until 1840. These bets were a means for publicly expressing your support (basically, they were an effort by "experts" to rally readers by saying "I'm willing to bet so many hundreds or thousands of dollars, I'm obviously confident that my guy is the right guy, so you should be, too." Now, on the surface, your proposal is different, in that you're seeing the wager as an objective measurement of conviction in the prediction, rather than a subjective expression of partisan support. But, in practice, I'm not so sure you can separate the two. Silver's weighting methods do favor more liberal polls. And Silver writes for the Times. It's not that either of these facts should preclude the wager you're talking about, but it does mean that backing his prediction is, in reality, less about backing the BEST scientific aggregation of polling data than it is about backing HIS aggregation (with all the preferences he has built into it). In sum, my point is that the line between objective fact and subjective politics is fuzzy and virtually inseparable. No matter how many numbers you're crunching, your science is influenced by your politics.

I did the same thing with my boss. I firmly believe Obama was going to win (this is back in July), but I told my boss I would take four to one odds the bet on Romney, which he accepted. It was a good bet, though as we get closer the the election date and there is almost no chance for a shock to the election I am almost certain to lose...

No, you're only 70% likely to lose, if we believe Silver; I'd hardly call that certain. Your expected value is positive until the book odds are 80%.

I'd take Romney at 4:1 today.

Right. If you already have a bet on Romney at 4:1 you can hedge by going to any of the online markets and taking the Obama side of the contract, because they're all running at 2:1. You're in good shape. (This is another reason why Silver's bet was a poke in Scarborough's eye -- he could easily have hedged it to make it a winner no matter how things turn out.)

I am interested in the idea that betting eliminates bullshit. Does not this principle fail in many circumstances because there is often an asymmetry in information and resources? If I am playing poker against someone with a much larger purse, am I not vulnerable to being bullshitted (bluffed) by the guy with more chips? Of course, if I am a brilliant player, I might be able to account for this, but I am not that smart in real life.
Other examples surely exist.

So you are saying we should expect more bullshit from the Kochs and Adelsons :)

What seems to be missed is that in defending Nate Silver's hypothetical bets, Alex has actually provided a strong argument that Silver's actual bet was inappropriate.

If part of his salary is based on the performance of his model, by taking modest bets randomly selected on either side--so he'd offer $1 in return for $3 if Romney should win, then he'd have no incentive but to make his model as accurate as possible. In particular, he has no financial incentive concerning the outcome of the election. He comes out ok if his model is accurate, regardless of who wins.

In actuality, Silver took a bet at 1-1 odds, which are out of line with his model. His model is not really being tested, except in the crude sense that he thinks Obama's chances are greater than 50-50, and he now has an interest in the result of the election.

Of course this is a small incentive and one that probably won't affect his decision making, but you can say the same about a lobbyist taking a politician or journalist to lunch--it's improper even if it has a low probability of swaying their decision making.

This is an argument that Silver should bet MORE, to bring his expected odds across all bets in line with his expectations.

That's fine. But it's still an argument that his current bet, as handled, is inappropriate.

No, this is Silver calling Scarborough's bluff!

Silly Justin, no pundits' salaries are based on the performance of their "model". That's why Scarborough has a job in the first place.

There is also a long history of sports betting between mayors and governors when their respective teams are in championship competition (World Series, Super Bowl, NBA) and I don't hear anyone calling this unethical or voting against the candidate in a subsequent election because he/she made such a bet. This was just a dumb column by Ms. Sulliavan

One thing missing in this discussion is that his prediction affects the outcome.

It's not missing, read the post again.

Agree. Undecideds may be followers and if it is perceived a close race you bring out more voters because, in a marginal state, the marginal vote counts, whether or not informed.

Wouldn't this in fact lead to systematically conservative odds, assuming Silver were risk-averse? He'd tweak the odds to the other side which, while lowering his payout, would increase his chance of winning.

Ah: blind trust bet. He has to not know which side of the bet he has until the outcome is revealed.

There is a problem, even assuming you take the economist's favorite trick of assuming away the moral/ethical issues.

That is, Silver, because of his position at the NYT (and his reputation, which he earned on his own), has an outsized effect on the popular conception of what the odds "truly" are. If Silver starts taking bets on the side, this gives him an incentive to feed incorrect information to the public, keeping the best numbers for himself, and then bet the difference between his stated odds and his actual belief. Even if he doesn't do it, the possibility is there. This is a serious chink in Silver's, and by extension the NYT's, reputation for disinteredness and neutrality.

Now you might say, well this is only a $1,000 bet, and the money doesn't even go to Silver. True enough, and I think we can all agree that Silver values his reputation more than he values this $1K going to the Red Cross. But I can see why the NYT would prefer to draw a bright line rule rather than have to decide on a case-by-case basis whether this particular bet is problematic.

read the post again, you have missed the point which is exactly to deal with the issue you raise.

Which is a counterfactual. Alex is saying "can you believe the NYT is bashing Silver? It should be applauding Silver, assuming that instead of the thing he did, he did some completely different thing that I just made up!"

So, in other words, Alex is saying the NYT was right.

There are more colors than black and white.

I think part of "unseemly" conclusion of the public editor is that it is viewed as Silver betting on an Obama win rather than Silver betting on the accuracy of his model- in other words, it appears to be a partisan act to the critic. I think Silver is a partisan (he pretty much admits this), but Tabarrok is simply pointing out that Silver can bet on his model by placing a weighted bet on either candidate, thus removing this argument against his betting. Given the Intrade odds, and Silver's conclusion, he can make his point by placing his bets there, too.

The NYT is shocked, shocked I tell you, that one of its employees is behaving in a partisan manner.

But even a weighted bet doesn't address the fundamental intellectual disconnect over probabilities. Silver has already quantified his odds; he was asking Scarborough if he was prepared to quantify his.

They could even place their bets with Intrade rather than with each other. Intrade, at the moment, has odds for Obama victory lower than Silver believes, and odds of Romney winning lower than Scarborough believes.

Her point that this makes him seem more partisan is terrible, for obvious reasons already highlighted by others.

That's not the main problem I have with this though. If you regard him as a Times reporter, you should be defending him against his critics, not publicly scolding him in a way that he cannot respond. Apparently, this is nothing new. Publicly throwing her colleagues under the bus through a one sided story is what she likes to do (http://www.poynter.org/latest-news/mediawire/191284/margaret-sullivan-reminds-rowdy-nyt-freelancer-hes-highly-replaceable/).

Lastly, he wrote this on his twitter feed, not in a Times column. She's the one who's bringing it up and drawing more attention to the bet, and now associating the NYT with the bet, even though they weren't before. She's only fueled the negative stereotype of the Times being for pompous, self-important, high brow bitches who condescendingly look down upon others with this holier than thou attitude that makes you want to puke in your mouth.

Obviously both men should jump at the chance to bet at 2:1 odds if each actually believes what they have said.

Growing up, I remember my friends and I would say "Wanna bet?" when we couldn't agree on an argument. It's actually good way of shutting an idiot up.

Re: the incentives of Alex's proposal

It's not clear that Alex's proposal of a blind trust would properly incentivize Mr. Silver. Suppose Mr. Silver believed that Obama has an 80% of winning. Under Alex's proposed blind trust, would Mr. Silver be better off claiming that the odds are 80% in favor of Obama, or 99.9999% in favor of Romney?

I think that Silver Betting publicly with a (or another if you take that position) pundit, for charity is not really a conflict at all. I think the conflict, and a significant one at that, would be if he was betting PRIVATELY without disclosure for his own profit. In which case he would have significant incentive to misrepresent the public betting line, as it is clear he is followed with significant credibility. Then use the public perception of those lines to bet in a way to be highly profitable.

In terms of this? Sure I guess I can see the NYT wanting to draw a line in the sand somewhere, but face it, Nate Silver is putting a price on an event, in this case an election. The ONLY practical way to call BS on it is to wager in some fashion. This is not an endorsement which does not say who you think will win, but merely who you want to win. To me an endorsement is a FAR FAR more partisan statement then a backed prediction.

This is basically the premise of intrade, that "put your money where you mouth is," generally leads to more accurate predictions.

As to putting his money into a blind trust and forcing him to take action on his predictions is, uh, a little odd. In theory if he is really creating what he believes to be accurate odds, he would not actually want to take either side at that price. Since I imagine he would want an edge of some type, on an amount of money that could be significant. Perhaps anyone can take any side at a price 10% worse that published (or publish a spread like a casino), then in the long run if his predictions are acurate he would make money, as opposed to breaking even but suffering huge swings (which could theoretically bankrupt him in the interim).

Before people say that would not be fair. I ask how many of you would be willing to flip a fair coin for $1? Ok, now for $10,000? Now for $1,000,000 Assuming you have it? Not many I would imagine.

Change this to heads you lose 9000, but tails you win 10,000 I would imagine more would fade that risk. Or 10,000 flips for $1 each at even money v. 10,000 flips where you win $1 for heads, and only lose 90 cents for tails. I would imagine most people would take the 10,000 flips with the edge over the other 2. You know you have an edge, and can exploit it over many flips. Who really would want the risk of ruin on a single flip? Normally even less so with no edge at all. That would just be silly.

"Bullshit is polluting our discourse and drowning the facts"

Pot, meet kettle. Kettle, meet pot.

There's no logical necessity in Alex's argument. He wants badly to see a bet made, but the cold hard fact of the matter is that one need not place a bet to have a credible model, as I said, there is no necessary connection. Credibility comes from reasoned arguments and sound methodology. And that's all there is to it folks, the man can have his own reasons for not betting.

You do not have knowledge of his U function, and even if you did you need a model for expected U. Good luck.

Clever as Hell.

I'm not an economist or statistician, but the aspect that seems to be lost in all the discussion about Nate Silver's predictions is time to election. To me, the predictions are only as good as the moment they are made. Then they have some value until they are eroded by events. So, for instance, if there was a prediction made in the morning, but one of the candidates was found in the afternoon to be engaged in an extramarital affair, the prediction would be all but worthless. So Nate is essentially betting on two things--that his prediction model is accurate, and that no events will occur to affect the election.
This is also, of course, the reason why people are obsessed with checking his latest predictions--they want to know if anything has shifted the election outcome.

This point is brought up a lot actually. "Stuff happens," as David Brooks says.

And this is exactly what makes Silver's model superior and the most useful - it deals with the unexpected far more precisely and accurately than other methods of analysis. Indeed, he is so precise, ppl often mistake this for certainty. Silver is not a prophet, and frequently stresses this. He is simply giving a probability at a particular moment in time. Those probabilities can change based on events, but that doesn't mean the earlier probability was incorrect. He can use past data to have an idea of how likely it is that such a random event that fundamentally shifts the dynamics of the race will occur, and put it into the model. As we get closer to the election the probabilities are affected, and they get shifted up or down accordingly. He does various other things with his model that depend on the time until the election.

Plus, every other person must also deal with these uncertainties. What's the other option - throw your hands in the air and say "it's a toss up" (50/50)? Or "this candidate sure has a lot of momentum right now, which should continue"? We get these types of statements all the time from pundits, and they are far less precise and accurate than Silver's methodology. The pundits make probabilistic assessments/predictions all the time, but people don't realize b/c it's not expressed quantitatively (even though it inevitably is quantitative, albeit less accurate).

Thanks. Great points. For the record, I did mean specifically the coverage that occurred regarding Nate making the bet--it shifted from ordinary news reporting to the more sensationalized reaction-driven coverage that tends to leave these important aspects behind.

"And this is exactly what makes Silver’s model superior and the most useful – it deals with the unexpected far more precisely and accurately than other methods of analysis."

I haven't seen any evidence of this at all. What evidence do you have that his model deals with the "unexpected far more precisely and accurately than other methods of analysis."?

I meant compared to the pundits, not against other models. That's self-evident.

There may be other election forecast models that are better. I'm not aware of any though, but open to suggestions.

Not so. Silver has both a "forecast" and a "now-cast". The now-cast is the outcome probabilities if the election were held right now, the forecast is the outcome probabilities for Tuesday. The difference is, effectively, the size of the uncertainty bars going into the simulation. Andrew Sullivan confused the heck out of me a couple weeks ago talking about 538 having predicted a 93% chance of Obama victory before the debates -- that was the now-cast at that time, not the forecast, and so didn't take into account things like the possibility of Obama flubbing the first debate (an unlikely, but not impossible, event).

Wow. You are absolutely correct. Would love to see that model, even I wouldn't understand half of it.

Tyler and Alex ought to announce more bets on MR.

They announce all the betting websites people set up to try to formalize this.

The problem I have with the "betting your true beliefs" argument is Nate Silver is getting a lot of pub from this. And if he's right it's really smart because you might get a few Romney supporters to pony up out of bravado or PR.

Even if betting were anonymous it wouldn't be perfect. Are people buying Treasurys because they think they are an awesome bet? Some are, but even those mostly think there are no alternatives.

New York City is one of the relatively few places in the United States that permits legal betting on the state's horse races; the NYT, however, seems to regard gambling for money as unseemly, outside the genteel domain of the rotisserie league (see: Daniel Okrent, former public editor) and the NCAA basketball pool.

I would have taken this bet with Nate Silver in a heartbeat. But two issues
1) As discussed, Nate must not believe his model b/c he's only offering even odds
2) His models depends on input data (polls) which are making assumptions about voter turnout about which there is valid concerns

If you're smarter than your average Krugman, you are looking at the data behind the polls and realizing that for Obama to win he has to significantly outperform his 2008 turnout. Unless you're living under a rock, it should be clear to any sane person that not only will Obama not turnout Democrats liked he did in 2008, he's also lost swing (independent) voters in record numbers.

And yes, I would have bet on this on Intrade except for I just looked into and it requires a statement from your minister, a vial of blood, and your first lost baby tooth to get an account as an American.


> Nate must not believe his model b/c he’s only offering even odds

Why does even odds imply that he doesn't believe his model? If he offered to take the Romney side of a 1:1 bet then yes, that would be betting counter to his model. With his current prediction of 80%, however, any bet in Obama's favor at 4:1 or lower *as well as* any bet in Romney's favor at 4:1 or higher indicates trust of his model.

In a sense, it does, in that they way the bet is structured most other rational bettors wouldn't take it unless they differ with him radically. If I thought the odds were, say, 60/40, I still would think Barry likelier to win, and I can't take the bet to express my difference with Silver.

Here's the chain of logic. The fear is that Silver might skew his models in some way because he either A) lets his beliefs (of one kind or another) influence the design of the models or B) skews them on purpose to actually alter the course of the campaign (cheering up the troops, making Barry look inevitable, making a donation to Mitt look like a waste, etc)

A bet is a way to alleviate those fears, because if Nate regularly skews his reporting away from reality, whether by design or accident, his expected value on betting over the long haul will be negative and he'll pay for it.

BUT. Even odds bets don't serve the purpose. Say Silver publishes model results that say Barry has a 90% chance of beating Mitt, and offers even odds. I have a better model that says the real probabilities are more like 55% Barry / 45% Mitt. I'm *still* not going to call Silver on it, because taking the the Romney side on the even odds bet is a losing proposition at any level beyond a 50/50 election.

So the bet offer becomes much less informative, because only folks that actually model Mitt as being *more* likely than Barry to win would take it. You can still get a "battle of the psychos", with Silver saying 90% Obama and somebody who is apparently in an entirely different universe saying, say, 65% Mitt, but neither model is really being challenged and most people will avoid betting. Offer odds, however, and the EV for taking the opposite side to Silver (or even the same side to Silver, as Alex designed it) becomes zero and even small differences in opinion/model can be arbitraged, so Silver will get rational people who differ from him betting. Of course, this also means Silver in effect becomes the bookie for a prediction market, assuming he'll take bets on either side. The Times' next big move? InNate!

No, no, no!!! Nate is not offering the even odds, he is taking them. If I think the Falcons are 3:1 favorites to beat the Cowboys and you offer even money (1:1) I will take that bet. That is what is going on here.

My apologies to Professor Hanson, but money is not the best predictor. It fails in the extremes and it's even more susceptible to emotion. At the time of this post InTrade shows democratic president win at 68%, democratic senate control at 75%, and republican house control at 95%. You can also bet on the triplet at 63%. If those numbers were realistic the triplet number should be 48%. Changing the democratic president likelihood to 88% makes the numbers line up. Maybe InTrade is closer to Nate's prediction than you realize.

Finally, I am not smarter than the average Krugman and I agree with you that it is hard to believe that Obama will turn out numbers like he did in 2008. But the polls (experiments) say he still has enough support to win and I try to take data over over beliefs.

Is Alex unaware that basically all people are risk-averse? The NYT would have to offer Silver a higher expected value for tying his forecast's performance to the outcome in order for this idea not to effectively be a salary cut.

Risk aversion is why Tabarrok says "modest bet."

Betting when the given odds are the odds that one believes to be true is strictly evidence that one derives utility from the act of gambling itself. Silver shouldn't necessarily take 3:1. He should take anything (<3):1 .

Alex Tabarrok has a mistake in his argument. He suggests we flip a coin to decide which side the pundit (Nate Silver, in this case) has to bet on (at the odds that he's forecasting). That wouldn't incentivize Nate Silver to make an accurate forecast at all.

It's like the trick where the two of us have to split a piece of cake, and the solution is that I cut the piece in half, and then you get to pick which half is yours. If instead, after I cut the cake in half, we flip a coin to decide which piece I get, then I'm indifferent between all possible cake slices -- my expected value is half the cake, no matter how I cut it. (Please grant me the ludicrous assumption of linear utility of cake for this analogy.)

To make this scheme work, you have to give Nate Silver an adversary that gets to pick "over" or "under."


Actually, that's exactly what Alex Tabarrok is proposing, that is, an odds-weighted bet and the ability for the bettor to take either side of the EV=0 bet.

Even if Nate's odds differ from his actual beliefs, randomly assigning his side of the bet still results in an EV=0. Thus, no incentive to be accurate is created. To work, the blind trust would need to take bets on the over/unders of the published odds, as Anonymous lays out below.

I'll agree that random assignment is a problem, but I think Anonymous' proposal gives Silver a way to game the system. OTOH, letting a bettor *select* either side properly incentivizes Silver to set odds to the best of his ability.

Perhaps I'm misunderstanding the proposal here, but if someone told me they were placing a $100 from my salary to place on either side of a bet, I'd assume I was risking $100 on an unknown side. With such a bet, I would maximize my return by making the actual favorite a heavy underdog rather than by being accurate.

The idea of a 'blind trust bet' is wonderful.

But I think more accurately, you'd want the journalistic prognosticators to have their compensation inside a 'blind trust *book*' that takes other bets whenever they are consistent with the public predictions, *plus* a vig.

Otherwise, this trust is only a negative motivator -- its expected return, taking bets on either side, offering exactly the right odds, is zero... and at a cost of variability. Add the vig and let the trust seek out offsetting bets in other markets, and then participation is both a signal of the writer's confidence, and a positive-expected-return activity.

Managing such blind trust pools for professional prognosticators could become a third-party service.

Perhaps Silver putting his silver where his mouth is, and Tabarrok adding the 'blind trust' angle, signals the dawn of 'Delphic Journalism'.

Silver's model currently (his "now cast" at 2:35 Pacific time) makes two intertwined predictions:

1. Obama currently has an 82.9% probability of reelection with
2. 303+ Electoral votes

he would find a lot of action if he offered 4:1 odds that Obama tops 300 EVs. Such a bet strips the partisanship out of the betting and places it on confidence in the accuracy of Silver's model.

here's another: one of Silver's claims to fame is that he called 49 of 50 states in 2008 (although, just about everyone called the first 47, so he really got 2 of 3).

Let's say the over/under on correctly called states this time is 48.5 . . . anyone taking the over? how about at 47.5?

Let's just look at the supposed 8 battleground states of Nevada, Colorado, Florida, North Carolina, Virginia, Iowa, Ohio and New Hampshire. Over/under on correctly called is 6.5. He currently has all but Florida and North Carolina going to the President. Anyone taking the over?

It's strange to propose the blind trust to incentivise him to care about the outcome. His expected profit/loss doesn't change as he varies his prediction probability. This is trivial to show mathematically, but the intution is that even if he picks a ridiculous probability (say, Obama 1% to win), then one side of the bet is very good and the other very bad, but since he'll randomly get one of those two sides, he's indifferent.

It DOES affect the variance of his profit/loss. The more accurately his prediction resembles the true outcome, the less volatility there will be in this bank account over time.

Don't assume he'll find takers on both sides, just that the blind trust will try around the predicted odds. So a ridiculous probability choice will be a sure loser: it'll only find counterparties who "take him to the cleaners".

I agree that would be make more sense, but I was responding to Alex's model: "randomly choosing which side of the bet to take"

I think that most of the people who are up in arms about Silver's prediction are honestly confused about the math. We have a long journalistic tradition of reporting poll results, which are two percentages that sum to 100%. We have almost no journalists tradition of publishing win/loose probabilities, which are also two percentages that sum to 100%, but answer a different question that the one answered by the poll results. Silver is reporting ~70%/30% as the answer to the second question, when almost no one has ever written about that question before. There are lots of reports reporting ~55%/45% as the answer to the first question, which is the question that people are used to having answered. People who are not statistically savy are going to assume that Silver is answering the same question as everybody else and his answer is going to look suspect to them because it is so far off the other numbers they are hearing.

I suspect Sullivan's objection to the bet originates more in conservative anti-gambling norms than conflict of interest concerns. She would have been just as upset if Silver bet on the accuracy of his model's electoral college outcomes or Senate race outcomes.

There were probably once good reasons for those norms - wealth used to be quite communal, and groups required norms to enforce communal risk aversion.

Bush vs. Obama


You guys, Tyler in particular, should start making your various claims in a form of firm predictions and offer bets on them. Call it bullshit reduction program.

"Furthermore, to remove any possibility of manipulation, the NYTimes should escrow a portion of Silver’s salary in a blind trust bet. In other words, the NYTimes should bet a portion of Silver’s salary, at the odds implied by Silver’s model, randomly choosing which side of the bet to take, only revealing to Silver the bet and its outcome after the election is over. A blind trust bet creates incentives for Silver to be disinterested in the outcome but very interested in the accuracy of the forecast."

You don't understand betting. If my "model" predicts a 66.6% chance of heads on a coin toss. that means odds of 1/2 on heads and 2/1 on tails (UK odds). If you then RANDOMLY assign me heads or tails i'll still be tending towards a long run win/loss of ZERO.

Do the math.

At every stop, Clinton reminds voters of his own days in office. "I am the only living former president that ever gave you a budget surplus," he said in Palm Bay. And only the brainless sheep bleat “Yes!” There was this little thing called the INTERNET that Gore invented. It’s global and explosive growth was a once-in-history event that entrepreneurs, not you and congress, made happen.

I see you also tweaked Obama’s slogan to suit you: FOREPLAY!

This betting system makes absolutely no sense. The expected value for Silver is 0 *regardless* of what he sets the odds at.
Since the article proposes that "we could hold a coin toss, heads Silver takes the Obama side, tails he takes Romney.", his expected value will always be 0. For example, let's say Silver sets the odds at 10:1 in favor of Romney (and for the sake of example, let's assume the real probability is 80/20 Obama, though this number could be anything and the argument would still hold true):
If heads, and he is randomly assigned Obama, then for each dollar bet, his expected return is (10*.8) - (1*.2) = 7.8
If tails, and he is randomly assigned Romney, the for each dollar bet, his expected return (1*.2) - (10*.8) = -7.8
But since he's equally likely to be assigned either of those scenarios, his overall return, even setting the odds at this outrageous line, is 0.

A better proposal, it seems to me, would be to set aside a portion of his salary, and bet on any lines that differ from it according to the model. For example, if Silver says 80/20 Obama (implying 4:1 odds), and there's a bet offered at 2:1, the NY Times should bet on Silver's behalf for Obama at 2:1 odds. Then, Silver would actually have an incentive to make accurate predictions.

The new proposal does incentivise him correctly, but reintroduces the conflict of interest that the blind trust was set up to avoid.

"Silver should be willing to take either side of a modest bet."

If he were risk neutral and there were no transaction costs, this would be true, but neither of those are true of the real world. Smart bettors only place wagers when they believe the odds are substantially in their favor.

The bigger problem here is that people have a poor understanding of how probability works, so they think that since Silver says Obama has an 80% chance of winning, if Romney wins, then Silver's prediction is a failure. But Silver's been quite clear that even an 80% chance of winning isn't a sure-thing prediction. The 20% chance of winning definitely makes Romney the underdog, but it's still more likely than the World Series being a sweep.

To me, Sullivan's criticism is simple: the New York Times should not insult its readers. (It should *entice* the public, not taunt it.)

Will Silver's public bet earn the New York Times more readers, in the long run? That's hard to determine. But even harder to determine are the potential nasty effects (i.e., long-term reader disenchantment).

So Silver would only bet 1:1? Seems like he's admitting his "Model" is bullshit. Not that we didn't already know that.

Isn't the bet itself BS? An admission that the Time's and Silver's reputation were in themselves not enough to ensure the veracity of the prognostication? By making this BS bet silver seems to be putting his money behind his calculations. The editor then comes to the rescue and claims that Silver is now linked to an organization that is above this sort of thing. Silver now has the benefit of seeming to go all in with out putting up anything.

The truth is that there is huge wager.already placed on the outcome of this election. Can the media br trusted at all in regards to their desired outcome? If Romney wins this election there will be a flurry of excuses and claims of voter suppression/fraud to explain the lie of the present.

I predict a tie: http://andreasmoser.wordpress.com/2012/11/06/my-predictions-for-the-us-presidential-election-2012/

Comments for this post are closed