So … I’ve noticed a lot of folks, particularly liberals, seem to place a lot of stock in Nate Silver’s probabilities. He’s got an interesting model, but his probabibilites definitely seem to be on the high side for Obama. Take, for example, Virginia:
RCP Average = R +0.6
Cook Political Report = Toss up
Sabato’s Crystal Ball = Toss up
FOX News = Toss up
CNN = Toss up
NBC News = Toss up
NY Times = Toss up
vs
The reason Nate Silver is so well respected is he’s accurate, he just doesn’t look a polls but he also looks at trends, tipping points, outliers, and just today we see this
What’s the difference between toss up and 63%?
13%
I’m not sure how one compares “Toss up” to a percentage. I also don’t know what methodology a lot of those sites use to determine what’s a “toss up”. That’s one thing that’s nice about 538; Silver is very detailed in describing his methodology.
In any case, suppose the polling average has Obama ahead in Virginia by between 1 and 2 %. Someone might look at that and say, “that’s a small lead; it’s a toss up”. But if your margin of error on that 1 to 2 % is small enough, that could be much more than a 50% chance of victory, especially if your model also predicts that the lead is not likely to diminish between now and election day.
It’s the same with North Carolina, which at least CNN and Fox have listed as a tossup. Silver has it as a 67% chance of a Romney victory. He’s consistent.
Probably you’re being facetious, but it’s worth pointing out that those saying “It’s a tossup” are probably not trying to say “Obama and Romney have equal chances of victory”. I think it more likely that their meaning is “Obama and Romney at least have enough chance of victory that I’m not willing to pick a winner yet.”
And again, what’s their methodology? Are they just saying “Obama is ahead by X% of the vote, and this doesn’t meet the cutoff where I’m willing to make a prediction?” Or are they actually calculating a percentage for how confident they are that Obama will win, as Silver is doing. I suspect in most cases they aren’t, or else they’d say “Obama 53% chance” or whatever, instead of “toss up.”
and this is older but it contains a little more detail
http://www.fivethirtyeight.com/2008/03/frequently-asked-questions-last-revised.html
Therefore his numbers won’t look the same as you see on other sites.
Silver is a Democrat, but more to the point, and the reason he gets so much respect here, he’s a big honking math nerd. He made his money working with Sabermetrics. It doesn’t get nerdier than that.
And aside from his nerd cred, Silver’s track record is impressive. It wasn’t partisan wishful thinking that caused Silver to predict 36 of 37 gubernatorial races in 2010, for example.
It varies from site to site, but generally speaking when the average of recent poll results are within 4 percentage points of 50% they indicate a toss-up and 4 to 8 points indicate leaning. Safe states are more than an 8 percentage point lead. The actual numbers may vary a few points from this but you’d be safe, pun intended, to assume something like this for just about any site.
Silver is one among a number of prognosticators who put poll numbers into historical context. A 63% chance indicates that given these numbers at this time before the election, a computer forecast running, say, 10000 models of the outcome gives it to one side 6300 times.
Such models are inherently volatile. If you followed Silver during the primaries you’d have seen that his state-by-state primary results prediction wobbled all over the place until right before the election. The difference between the primaries and the campaign is that there are far fewer undecided voters today and fewer people whose votes are swingable. That makes predictions more reliable farther out.
Basically, the comparisons that the OP are making are meaningless. You can’t compare a today result - tossup - with a forecast result - 63%. They’re apples and oranges.
I should correct myself slightly. Really what he’s doing is using his model to simulate the results of the election. When he says “Virginia: 64.3% chance of Obama win”, that means in the simulation Obama won 643 out of 1000 times. But essentially this is how confident the model is in an Obama victory, based on the polling numbers and other factors.
But my point remains, I’m not sure if these other sites have any method of determining likelyhood of victory from the polling numbers, or if they’re just grouping states into buckets based on how strong the polls lean one way or the other.
More than that i think the main reason that he gets so much respect is because his model has been proven extremely accurate in predicting elections already.
Romney has fallen off a cliff on Intrade. Anybody who doubts 538 could make a boatload of cash betting on the Mittster.
Nitpick - Nate currently has Obama’s chance of winning at 80.8%.
Thanks, that confirms what I suspected. Silver currently gives Obama an “adjusted polling average” of +1.4, with a “projected vote share” of +1.8, so this seems consistent with what other sites would call a toss up.
I.e., it’s not that he thinks Obama’s lead is greater than the other sites do. But as you point out, he’s using this and other factors to extrapolate that Obama has a 64% chance of winning the state.
The OP was using Virginia as one particular example.
Also if you look at Nate’s site, he forecasts just 50.9% of the vote going to Obama in Virginia. But don’t conflate probabilities with vote distribution.
I’ll give an example say there is a poll that shows Obama with 51% and Romney 49% with and a 3 point margin for error. To make things simple lets say that there is a 50% chance that Obama gets 51%, 15% he gets 50 or 52%, 8% he gets 49 or 53% and 2% he gets 48 or 54% (You could do a a normal distributons using the Standard deviation and do it mathmagically, but this is a close enough approximation to make the point). In this illustration Obama gets greater than 50% of the vote 75% of the time, Romney gets it 10% of the time with 15% of the time it is a dead heat. That is how a close race can still have a seemingly large probability for one side winning.
Saying Nate Silver has an “interesting model” is a bit of an understatement. His model blows all other public models out of the water - I can’t imagine getting that type of analysis would cost less than $500k - $1m if done privately. I’d be willing to bet money that his blog is the first thing read in the morning (or when updated) by the chief pollsters for both Obama and Romney. Yes they do their own polling, but Silver is fantastic in his analysis.
He explains his model in great detail - and I’ve never detected any type of bias in his reasoning.
I too am surprised that Virginia is doing as well as it is for Obama, but it’s not like a landslide - and he explains how he got there. Note that intrade (where people put their money down) - also suggests an Obama win(but not as large a margin):
http://www.intrade.com/v4/markets/contract/?contractId=745847
Nothing. Go to the 538 site and look at the Tipping Point States: “The probability that a state provides the decisive electoral vote.” Virginia is ranked #4. It’s a tossup.
I agree though that 538’s odds are a bit on the high side, as they are based on historical data and therefore do not reflect Citizens United, 2012 Voter Suppression, or the tsunami of Super Pac cash backed by special interests. I take the average of 538 and intrade, each which have their own problems.
ETA: The odds of rolling a doubles with 2 dice are 16.7%. 13%<16.7%. 13% is not a large margin. And Romney’s 20% odds as determined by 538 isn’t awful anyway.
I’d think that’s what “leans Bedfellow” and “likely Bedfellow” are for. (Substitute whoever you like for ‘Bedfellow.’ (“How’d you launder the Libyan kickback money, Senator Bedfellow?” - Milo Bloom) ‘Tossup’ should mean ‘each candidate’s chances of winning are pretty damned close to 50%.’
