Imagine that you laid all the states out on a 538-unit-long number line, from the most certain Romney state at one end, to the most certain Obama state at the other end, with each state taking up as many units along the line as it has electoral votes.
You don’t really know what the right order is, but Nate’s saying that Virginia has a better chance than all but three other states of straddling the 269-unit mark.
That doesn’t mean it’s a tossup. If Obama is expected to win by 330-208, for instance, the tossup states will be close to the 330-unit mark, counting from Obama’s end towards Romney’s.
538 absolutely does reflect possible voter suppression in 2012. Nate has written extensively about how he has figured the various voter ID laws, and their effect on turnout, into his model. The bottom line is that he predicts a 1-2% advantage for Romney in certain states based on voter suppression, and has included that information in his predictions for those states.
Yes, I’m confusing ordinal and cardinal. Ignorance fought.
In House races, “Lean” means 86% chance of victory, while Likely implies 95% chance of victory. Tossup implies 50-50 in aggregate, though if you think about it that last percentage could disguise a lot of intra-group variation. http://cookpolitical.com/about/accuracy
FWIW, (86+50)/2 = 68%. I’d say 64% is closer to the “Tossup” category than the “Lean” category.
SpoilerVirgin: Didn’t know that. I see now that Nate started incorporating voter suppression for selected states in mid July.
I guess that still leaves Citizens United and Super PACs unaccounted for as well as other wild cards if you believe those aspects favor Romney on average.
The OP: Here’s what Charlie Cook thinks about the race: http://cookpolitical.com/story/4754 It is becoming clear that if President Obama is reelected, it will be despite the economy and because of his campaign; if Mitt Romney wins, it will be because of the economy and despite his campaign. …
This is a very close race and one that still could go either way. But the odds of Romney capitalizing on this economy, and the opportunity it affords, seem lower than they were before the conventions. If Republicans and Romney supporters are growing nervous, they should be. This perspective can be compared to Nate’s after the election.
Q1: Internal consistency. Does Nate’s popular vote forecast track with his electoral vote forecast? Because while both suggest “Close race”, the odds forecast does not.
Q2: Does the Fair Model, once all the data are in (i.e. in Q2 2013), predict an Obama victory or a Romney Victory? If Romney, what percentage point advantage would Obama need? What percentage point advantage did he secure?
Incidentally, while he is numerically literate, Charlie Cook has an old school perspective: he seems to think that decomposing the polling data by state is a near-pointless exercise. The national popular vote is a sufficient summary. I’m unconvinced of that, but it’s an interesting hypothesis. Charlie Cook is one of the very few horse race commentators that are worth listening to IMHO. Nate Silver is another.
How in the world did he get reliable data on voter suppression? His figures of a 1-2% loss intuitively seem about right, and that was the target of the voter suppression exercise, to trim off enough to win but leave some plausible deniability. But I’m thinking that isn’t really sufficient grounds for what he’s doing, so…where? How he know, huh?
I respect his work and am inclined to take him on say-so, but I would still like to know so that the brain itch stops.
The reason Silver is respected is because he was absolutely bang on in 2008. His predictions were BY FAR the most accurate of any mainstream prediction. He nailed pretty much every state within a tiny range of error and predicted every Senate race, and that was after utterly demolishing the established talking heads in predicting primary results. He is respected because he is correct and nobody’s ever shown why he isn’t correct now.
First of all, this isn’t really a field where you can use terms of art in a way that are at odds with how most people understand the words being used, because this sort of thing gets a wide audience, rather than being published only in scholarly journals.
If a horse goes off as a 1-2 favorite, it’s because most bettors would say it’s likely that the horse would win. If it went off as a 1-5 favorite, people would say it was extremely likely that the horse would win. That’s a 67% chance and an 83% chance of winning, respectively. That isn’t tossup territory. And a horse that went off as a 1-20 favorite would be regarded as all but certain to win, not just ‘likely.’
Second, even as terms of art completely divorced from normal usage, the ranges seem badly designed. Nomenclature that lumps everything from a 15% chance to win and an 85% chance to win into the same bucket is failing to make distinctions between things that are very, very different. And then it goes through two categories between 85% and 96%? Sheesh.
I mean, here I am, trying to decide which House races to toss some money at, but now that I know what the terms mean, I find that they’re of little use to guide me. OK, it’s good to know that ‘likely’ means ‘extremely, extremely likely’ so don’t waste your money on a race where either candidate is even ‘likely’ to win, and only toss money into a race where it ‘leans’ one way or the other if that race is quite important to you for other reasons.
But then pretty much all my money is going to ‘tossups’ that are undifferentiated by this nomenclature, but in order to contribute intelligently, I really need to understand which ‘tossups’ are really tossups, which ‘tossups’ lean a bit one way or the other, and which ‘tossups’ lean pretty heavily one way or the other. But all I know is that they’re all ‘tossups,’ without knowing whether or not they’re tossups or anything close.
This rant isn’t aimed at you, MfM, you’re just the bearer of the news. But jeez, all these years I’ve been thinking that these words meant something like what they usually mean. And now I find out that they don’t, and in a way that drastically reduces their utility.
I don’t get the use of the 50 in ‘(86+50)/2’ but maybe you can fight my ignorance there. I’m thinking (86-50)*2 gives one a width, in probability terms, to the tossup zone, but the other formula, I don’t get its meaning.
At any rate, ‘tossup’ to me implies ‘your guess is as good as mine about how this will turn out,’ and even 60-40 either way seems clearly out of that zone. So I’m glad that Nate gives probabilities that I can understand.
Wish Nate had a staff that was doing House race forecasts; I really want to spread some money around, but not at random.
Silver’s model is pretty solidly predicting an Obama win. This election is definitely going to be a proving ground for his forecasting, because I don’t think he disagreed with the other guys so much in 2008 or 2010.
I’d take a more charitable view of Charlie Cook. Pundits have blovated for decades, but he was one of the first who was a) numerate and b) had the cojones to actually make some predictions. It was only years after he assembled his methodology that somebody put numbers to his predictions. It turns out that “Lean” means “Likely”, that “Likely” means “95% Safe”, and “Safe” means “Barring a dead body or a live boy”. But his framework was constructed during the 1980s for gosh sakes. It’s not surprising that the calibration was off: what is impressive is that it worked as well as it did.
Nate Silver has him beat. But the more qualitative old timers have value too. And not to put a fine point on it but the Washington Press Corps is just pathetic – they are value subtractors. As an example I might take George Will who said with regards to the economy: “*f the Republican Party cannot win in this environment, it has to get out of politics and find another business.” Snappy line. Wholly ignorant: election year economic growth seems to matter more than the level of the unemployment rate. And the current trend isn’t too bad. In other words Charlie Cook at least has some familiarity with the basic literature. George Will was just picking superficialities out of the air.
2nd Cite: Ray Fair’s “Fundamentals only” model gives Obama 49.5% as of July, which is in “Too close to call” territory, not “Terrible environment”. Vote-Share Equations: November 2010 Update
‘(86+50)/2’ : Back of the envelope calculation: average of the two figures. Very arguable - can be improved on.
adaher - Actually a lot of people are making probabilistic forecasts now. And most of the models currently favor Obama. ultrafilter has pointed out though that scoring these sorts of exercises is a scholarly topic all its own. Imagine running Romney v. Obama in 1000 universes. In ~200 of them Romney wins (if you believe Nate’s model). That’s really not too bad. Romney’s chances of being President are a lot higher than mine or yours.
This of course is the key point. The house edge on a single baccarat bet is only 50.5 to 49.5; but the chance a given table is profitable at the end of the day is much more than 50.5%
Wow! I clicked Intrade now and see that Democratic-Winner has indeed soared from 57% to 64% in a single week. We really need to know which of these gambler’s markets handles the most money – that’ll be the one with the most trustworthy predictions. I wonder if it’s Betfair, mentioned in a recent thread. Clicking it I see Romney has fallen almost 2% there just in two days.
If anything, you understated how he did in 2008. He picked 49 of the 50 states in the electoral college and went 35 for 35 in Senate races. Read bias into his predictions at your own peril, people.
What I find more interesting is that intrade lagged behind Nate Silver’s big move by more than a couple of days. If they are both shown to predict the electoral outcome pretty well and Nate Silver precedes intrade, then that implies that his views are superior to the conventional betting wisdom.* Or that some of Romney’s minions have their thumb on the scale and that the saavy can take advantage of that. Or both.
What’s even more impressive is that intrade did move contemporaneously with 538, albeit to a less than one-to-one extent. In other words, intrade was aware of 538, but has weighted it less than optimally. So they had to play catch up.
Silver has Obama ahead in Virginia by just under two points. If Silver says one thing and NBC says another, NBC is almost certainly wrong. It’s a close race in Virginia.
It’s too much to say that this poll is wrong. The poll is accurate to within its margin of error if it shows a 5 point lead and the ‘real’ spread is 2 points. There’s no use overstating the case.
