[PassedOut]

As usual, I follow Nate Silver closely during election seasons. (These days the New York Times publishes his blog at 538: Nate Silver's Political Calculus.)

Today he calculates Obama's chances of winning at over 80%. Although I'd like to believe it, that number seems to me to be overly optimistic. I'd appreciate hearing the assessments of posters here who have strong backgrounds in statistics on the reliability of Nate's methods.

[hrothgar]

 PassedOut, on 2012-September-10, 14:07, said:

As usual, I follow Nate Silver closely during election seasons. (These days the New York Times publishes his blog at 538: Nate Silver's Political Calculus.)

Today he calculates Obama's chances of winning at over 80%. Although I'd like to believe it, that number seems to me to be overly optimistic. I'd appreciate hearing the assessments of posters here who have strong backgrounds in statistics on the reliability of Nate's methods.

I think that Silver's methods are quite sound. (As I recall, I was the first one promoting him on this board, though folks like Justin already knew him from sports betting)

Its important to note that Silver himself states that his estimates are likely to be biased towards Obama at this point in time.

[barmar]

[quote]
Rather, it was what the model did in September of that year, when it detected very, very quickly after the collapse of Lehman Brothers that John McCain’s goose was cooked,
[quote]
While he's trying to point out how well his model worked, this also points out a shortcoming of any forecasting model: it can't account for surprises.

This is why forecasting tends to be more accurate as you get closer to the event in question: there's less time for surprises, and predicting that the status quo will stay the same is generally pretty safe.

[Phil]

270towin has it at 85%.

Somehow a 3-0 split seems a LOT more likely than Romney winning.

[jonottawa]

Intrade has it a little less than 60-40.

Wow, I typed that before I checked (it's been there for a long time), but as I type this it's just over 60-40.

Barring a huge game-changer Romney can't win unless voter suppression efforts by the Repugs reach new levels in Ohio and Florida.

I think the 80-85% numbers ignore the SuperPAC $ and underestimate the gullibility (and amnesia) of the American people.

FDR weighs in on Mitt Romney (this is classic.)

[PassedOut]

 jonottawa, on 2012-September-10, 15:11, said:

I think the 80-85% numbers ignore the SuperPAC $ and underestimate the gullibility (and amnesia) of the American people.

Here in Upper Michigan I'm no longer seeing Romney ads, so perhaps his campaign has thrown in the towel here. Still we see some rabidly anti-Obama bumper stickers and signs here.

On the other hand we are getting lots of mail and robo-calls supporting our "tea-party" backed representative, freshman Dan Benishek, elected in 2010 after Bart Stupak retired. The main thrust of the ads for Benishek is his -- and the Republican Party's -- strong commitment to preserving Medicare. I suspect that this strategy is working for him here. We'll see.

[kenberg]

An 80% probability is not the kind of thing that allows for verification, but here is how I would assess it: If someone put a gun to my head and said that I must either put up two grand against eight grand and take Romney, or put up eight grand against two grand and take Obama, I would put up the two grand on Romney. Probably I would lose it, but that's the way it would go. Six against four would be tougher. I would really want to see if the gun is loaded.

I heard on NPR that Romney was on Meet The Press and said he would keep parts of the ACA. I got the idea that he said he would keep all the good things that give people nice things, and would get rid of all the bad things that cost people money. Now why didn't Obama think of that?

[JLOGIC]

Nate>Intrade.

Seriously, it is shocking that the market does not just end up at whatever probability he has, but w/e more free money for me!. Needless to say I am significantly on Obama on intrade. Last election was even better because fivethirtyeight was largely unknown.

[JLOGIC]

Also, it is interesting that last year just before (or just after?) Palin was announced as the VP nominee, intrade had Obama at ~53 and Nate had him at ~80. Nate was also the only one predicting a big Obama win. Sure enough he was right.

I remember the senate elections the put him on the map, he got 49 of the 50 states right. The one he got wrong he had at 52/48 and the 48 came in.

Basically nate just interprets data far better than others, and the other polls are still respected enough that the intrade market remains inefficient.

@Richard: I don't know nate from sports betting, but from poker. Back then, he was well known as some super baseball stats guru, apparently he changed the game completely in terms of predictive models for baseball, and changed fantasy baseball with his methods. He says that politics is just like baseball, there is a ton of data (in the case of politics, polls), people just interpret the data poorly. He has far and away been the best predictor in politics, just as he was in baseball, ever since he emerged with 538 as poblano.

Trustin one source over the collective wisdom of people (as I am here, trusting 538/nate over intrade) is very dangerous, but nate has earned it. Dude is legit. Even with 1 election, he always simulates how the various states will go so you get a big sample and he is always very accurate. Dude is legit, and if you believe you can make a lot of money on intrade via him (for now, I assume he will eventually just set the market).

[JLOGIC]

 kenberg, on 2012-September-10, 18:08, said:

An 80% probability is not the kind of thing that allows for verification, but here is how I would assess it: If someone put a gun to my head and said that I must either put up two grand against eight grand and take Romney, or put up eight grand against two grand and take Obama, I would put up the two grand on Romney. Probably I would lose it, but that's the way it would go. Six against four would be tougher. I would really want to see if the gun is loaded.

That's great but what do you base this on? Likely natural bias, biased sample (since no one has an unbiased sample of people), etc etc. If someone had a gun to your head that's all you could go on, but having someone who is actually analyzing a ton of evidence and is a genius in that area, is paid to do it for a living, etc etc is going to be more accurate. It's like someone who is not a professional sports bettor (even if they're smart, and a big sports buff, whatever), trying to tell you which side of a point spread they think is right. It's silly to think they will have any reason to pick a winner (and I like to bet sports for fun, but since I do no hard analysis it's not like I'd tell you I am going to be a winner long term, in fact I will certainly be a loser long term going on my own intuition and feel). So I'm not trying to insult you or anything.

[semeai]

 kenberg, on 2012-September-10, 18:08, said:

An 80% probability is not the kind of thing that allows for verification, but here is how I would assess it: If someone put a gun to my head and said that I must either put up two grand against eight grand and take Romney, or put up eight grand against two grand and take Obama, I would put up the two grand on Romney. Probably I would lose it, but that's the way it would go. Six against four would be tougher. I would really want to see if the gun is loaded.

I heard on NPR that Romney was on Meet The Press and said he would keep parts of the ACA. I got the idea that he said he would keep all the good things that give people nice things, and would get rid of all the bad things that cost people money. Now why didn't Obama think of that?

 JLOGIC, on 2012-September-10, 21:42, said:

That's great but what do you base this on? Likely natural bias, biased sample (since no one has an unbiased sample of people), etc etc. If someone had a gun to your head that's all you could go on, but having someone who is actually analyzing a ton of evidence and is a genius in that area, is paid to do it for a living, etc etc is going to be more accurate. It's like someone who is not a professional sports bettor (even if they're smart, and a big sports buff, whatever), trying to tell you which side of a point spread they think is right. It's silly to think they will have any reason to pick a winner (and I like to bet sports for fun, but since I do no hard analysis it's not like I'd tell you I am going to be a winner long term, in fact I will certainly be a loser long term going on my own intuition and feel). So I'm not trying to insult you or anything.

Another point here is that you shouldn't imagine yourself choosing between 2000 against 8000 or 8000 against 2000. You should at least instead imagine yourself choosing between 2000 against 8000 for Romney or 2000 against 500 for Obama. Your aversion to losing 6000 or 8000 is going to come into it otherwise. Not that I'm claiming he's necessarily 80% to win, just that in imagining this scenario, you've set yourself up to be irrational (besides the "gun to the head" bit etc).

[jonottawa]

I spent a little time looking at the individual swing states on Intrade, too.

The 5 closest states (with Obama win%) are Colorado (64.8), Florida (47), Iowa (59.7), Ohio (61.9) and Virginia (58.9). If everything else breaks as it's supposed to, Romney has to win Florida, Ohio and Virginia as well as either Colorado or Iowa. (If Obama wins only Virginia of these, he wins 270-268.)

I'm surprised that Ohio has a higher win% than Obama. It's a virtual lock that if Dems win Ohio, Obama wins. And he might even win without it. (Though that would break Ohio's streak of getting it right. The last time they picked a loser was 1960.)

[semeai]

 jonottawa, on 2012-September-10, 23:58, said:

I spent a little time looking at the individual swing states on Intrade, too.

The 5 closest states (with Obama win%) are Colorado (64.8), Florida (47), Iowa (59.7), Ohio (61.9) and Virginia (58.9). If everything else breaks as it's supposed to, Romney has to win Florida, Ohio and Virginia as well as either Colorado or Iowa. (If Obama wins only Virginia of these, he wins 270-268.)

I'm surprised that Ohio has a higher win% than Obama. It's a virtual lock that if Dems win Ohio, Obama wins. And he might even win without it.

Nate Silver has all these at 74 or 75 except Florida, at 65. Two comments: 1) They'll move together to some extent, so the win percentage for a likely tipping point state should probably be close to the overall win percentage. 2) Intrade often has little inefficiencies/irrationalities (at least when I've looked closely at things in the past; maybe it's gotten better, though by what you say, I guess not). I suppose the market lacks some liquidity due to their cut, small volume, and the wait for the payoff.

[helene_t]

Intrade changes surprisingly little but maybe they will change now in response to Nate and/or just because the odds will generally diverge when election day comes closer. RCP fluctuates quite wildly and should probably be smoothed more, especially long time ahead of the election.

I was in baseball prediction in my previous job and it was tremendous fun. Boy how can the Americans produce amazing sports statistics, and downloadable for free! American sports is yummi for every forecaster, even someone like me who has no interest in sports at all other than bridge.

[kenberg]

 JLOGIC, on 2012-September-10, 21:42, said:

That's great but what do you base this on? Likely natural bias, biased sample (since no one has an unbiased sample of people), etc etc. If someone had a gun to your head that's all you could go on, but having someone who is actually analyzing a ton of evidence and is a genius in that area, is paid to do it for a living, etc etc is going to be more accurate. It's like someone who is not a professional sports bettor (even if they're smart, and a big sports buff, whatever), trying to tell you which side of a point spread they think is right. It's silly to think they will have any reason to pick a winner (and I like to bet sports for fun, but since I do no hard analysis it's not like I'd tell you I am going to be a winner long term, in fact I will certainly be a loser long term going on my own intuition and feel). So I'm not trying to insult you or anything.

Absolutely no insult taken. This was my intent to get at "what does it mean to say that the probability is 80% for Obama?" We cannot run the experiment a large number of times, but still it should mean something. As a first try, I look at "would or would not take the bet". Presumably, if Silver believes his own statement, he would be delighted to place a large bet on Obama at 70-30 odds. If not, then I have a hard time understanding what to make of his 80% evaluation.

I would not place a large bet at all, at least not at any odds anyone would be willing to give me. But this involves personality and circumstance. I am retired, I have enough money to live comfortably if I don't do anything stupid with it. Winning a large bet would change my life little, losing a large bet would be a disaster. So no real bet, but I can fantasize about how I would go about betting. Certainly going on intuition is a loser in the long run, although I have a fair amount of faith in my gut. But it is not so easy to pick an expert either. If 32 monkeys bet on five elections, there is a fair chance one of them would get all five right. So I would have to examine the model and see if I agree. Uh oh. This is also a judgment call and probably, at least partly, would be based on my intuition. Successful betting, no doubt, requires a lot of hard work. I have known many people who gamble on many things, and most of them tell me how much money they make. 'Tain't necessarily so, but no doubt it can be done if you choose your bet with care.

Bottom line: Whatever the 80% means, I don't believe it. Yes, it's my intuition talking.

[JLOGIC]

You are probably correct given that intrade is low 60s and betfair (so I heard from mickyb) is high 60s or low 70s. I would guess based on all of that and how highly I weight Nate that Obama is something like low to mid 70s.

Nothing wrong with speculating/using your intuition if you're not betting (or just betting for fun) obv, like I said, sadly I do it with sports

[nigel_k]

 kenberg, on 2012-September-11, 07:30, said:

If 32 monkeys bet on five elections, there is a fair chance one of them would get all five right.

This is the main point here.

It is a binary outcome once every four years, and there are tons of people making predictions, which means there are bound to be people with excellent track records regardless of how accurate they really are.

Conversely, Intrade is a combination of the forecasts of many people - and they only get paid if they get it right.

[jonottawa]

If the conventional wisdom is 80% but the people putting their money on the line have it at 60%, I'd tend to think the true number is actually less than 60%. Look at it this way: You've got lots of people like Justin who are betting on Nate being right and yet the Intrade number is barely moving, even though it's WAY off from Nate's prediction. That's a lot of people betting on 'Buster' Douglas, maybe they know something I don't.

This reminds me of:

"When I was a young man about to go out into the world, my father says to me a very valuable thing. He says to me like this... "Son," the old guy says, "I am sorry that I am not able to bank roll you to a very large start, but not having any potatoes which to give you, I am now going to stake you to some very valuable advice. One of these days in your travels, a guy is going to come to you and show you a nice, brand new deck of cards on which the seal has not yet been broken. This man is going to offer to bet you that he can make the jack of spades jump out of that deck and squirt cider in your ear. Now son, you do not take this bet, for as sure as you stand there, you are going to wind up with an earful of cider."

[MickyB]

That doesn't explain why Betfair has Obama at 68%. Whatever the "true odds", the 7% gap is bizarre.

[semeai]

 jonottawa, on 2012-September-11, 15:02, said:

If the conventional wisdom is 80% but the people putting their money on the line have it at 60%, I'd tend to think the true number is actually less than 60%. Look at it this way: You've got lots of people like Justin who are betting on Nate being right and yet the Intrade number is barely moving, even though it's WAY off from Nate's prediction. That's a lot of people betting on 'Buster' Douglas, maybe they know something I don't.

Perhaps. I'm certainly not one of the people betting. I'm not sure how much stock [har har] to put in Intrade. The lifetime volume for Obama to win is just under $8 million (just under 800K shares). Is this big enough that it should be really reliable or not?

There's also the problem that it's presumably good for a candidate to be doing well on Intrade (even independent of how well the candidate is really doing). If some wealthy would-be-donors just decide to go shift the market at Intrade, how much would that be worth compared to extra advertising?