I'm sure it would have predicted a Dukakis win as well. Look, things change.

Some days you lament the uselessness of polls and other days you salivate over them. Please let us know in advance which day we are facing early. You could post it at the top daily.

Do you have some type of profanity filter that translates Wang into Wennington?

I'm honestly curious at how you arrived at "Wennington".

-----

I remember well Wang's attempt to analyze the '04 election. He had the election as a sure thing for Kerry, and didn't have much of a concept about what his data actually meant.

He does seems to have dialed back his rhetoric a notch this time around - it really was incredibly fantastical last time around. But he's still missing some pretty basic elements of what he's trying to measure.

One of the most interesting features of Silver's methodology is the way it reveals the "swing value" quotient of each state.

The takeaway is that for all we talk about Colorado and Virginia, this is still an Ohio takes all election.

Silver has a useful number up about Obama's chances of winning when he loses OH, and it's consistently ranged between 15% - 25%.

This is lame to say, but the reality is somewhere in between.

There's something flawed about the way Silver does it that gives outsized chance to low-probability outcomes. For instance his model still thinks McCain has 2% chance of winning California, when in truth it's almost certainly 0%. And so it's much smoother. At the moment the election is pretty closed to locked down; Obama has a 2% or greater lead in all the Kerry states plus Iowa, New Mexico, Ohio, and Virginia. Now, things could change, but that's where we are today.

So why is it in between; well, because Silver is correct that polls tighten over time. But it is closer to Wang than Silver.

The takeaway is that for all we talk about Colorado and Virginia, this is still an Ohio takes all election.

Petey, this just isn't right. Allocating the Kerry states plus Iowa and New Mexico to Obama, it's not an "whoever wins Ohio is the president" election. It's an election where McCain has to win all of Virginia, Colorado and Ohio. And it's very easy to conceive of a situation where he wins Ohio but loses one of the other two states. For example, if he picks Ridge as his VP he probably increases his chances in Ohio but decreases them in Colorado and Virginia. Or if Obama picks Kaine -- he could easily carry Virginia without Ohio. Or if McCain picks Romney, he easily could lose Ohio AND Virginia but win Colorado and flip Michigan, which I think would win him the election (I haven't counted the EVs).

Let's play some catch-up on Wang:

- On the eve of the 2004 election when the polls showed something very close to a tie, Wang projected a 98% chance of a Kerry win.

- Under Wang's "methodology", if the last three polls in a state show a candidate up 44% to 41% with a MOE of 2.5%, Wang would assert that candidate's chances of winning the state are 100%. Think about that one for a bit. There are plenty of equally large howlers present in his "methodology".

- Wang was heavily pushing various exit poll conspiracy charges after the 2004 elections.

Dude is a charlatan, straight up.

The best thing about Nate Silver's emergence is that Wang will get less publicity this time around. Silver isn't right about everything, but at least he's got a clue, is making a real attempt to cover this stuff, and isn't conjuring his math out of thin air.

Why folks like Ezra and Kevin choose to give this fraud oxygen after he got pretty thoroughly discredited in 2004 is beyond me.

Travel throughout Ohio -- I dare you. Tell me that you don't think all the racists attacks that are coming aren't going to keep that state red. Sad, but true.

"For instance his model still thinks McCain has 2% chance of winning California, when in truth it's almost certainly 0%."

Well, no.

It's somewhere around 98% certain that McCain has a 0% chance of winning California.

McCain could still win this election nationally by 55% - 45%, and that would put California into play. Just because the odds of McCain doing that are 1 in 50 doesn't mean that they're zero.

"Now, things could change, but that's where we are today."

One of the interesting things about Silver's approach is that it isn't just about where we are today. It's also an attempt to see where we'll be in November by factoring in a state's historical partisan index.

But Wang is not generating a particularly useful measure of where we are today because his methodology is so fucked up. For example, he gets you the 100% for Obama result you seek in CA, but he also gets your an 84% for Obama result in VA. Is 84% what you see when you look?

I'm highly skeptical about all statistical and historical models in predicting this election.

Obama has brought too many people into the process for the first time. These young and minority voters are notoriously hard for pollsters to sample (cell phones, no longtime fixed addresses). And contra past models, they are more likely to vote in November, because they voted in the primaries. It wouldn't surprise me if these people turn out to account for 5-10 percent of the November electorate.

McCain, too, can count on an invisible electorate: unpolled or mispolled elders, evangelicals and Bradley/Wilder effect racists, possibly another 5-10 percent.

If I were betting (I'm not), I would say the outcome this year will be as close as in 2004, but that turnout will be 15-20 percent higher.

"There's something flawed about the way Silver does it that gives outsized chance to low-probability outcomes. For instance his model still thinks McCain has 2% chance of winning California, when in truth it's almost certainly 0%."

To go further, Wang's "methodology" produces only 4 states out of 50 where the chances of Obama winning are between 20% and 80%.

Would you say he might be going a smidge too far in avoiding giving an outsized chance to low-probability outcomes?

Would you say that, just judged on where we are today, assigning Obama a 93% chance of winning MI might be a little loony?

I agree with Petey's comments, and I would add that even if the methodology were more sound, the big difference between this analysis and what Silver is doing is the fact that the election isn't being held today, so this kind exercise is useless. If it were going to be held today, then the last month of campaigning would have been extremely different, both because the campaign strategies would be different and the public would be more engaged.

Even if the analysis were sound, the numbers of today are grounded in the fact that the election isn't today.

"I would add that even if the methodology were more sound, the big difference between this analysis and what Silver is doing is the fact that the election isn't being held today, so this kind exercise is useless."

I think there is a place for both a methodology like Silver's that incorporates historical voting patterns, and a methodology that only incorporates current polling.

Neither is "correct". Both are interesting and useful.

Unfortunately, Wang's methodology is so utterly brain-dead that no one is producing a decent "where we are now" set of numbers.

Petey's view that it comes down to Ohio isn't what Nate says. He says there are several scenarios for Obama, and less so for McCain. McCain will need a perfect storm to win. Obama doesn't. This is the first electoral map in a quite along time where the fundamentals favor the Democratic candidate. To say that it comes down to Ohio implies that there is a race too close to call and would suggest a landscape favoring at least historically the GOP. Does anyone here buy that as the fundamentals of this race?

I agree with Petey's comments...

Holy crap, so do I! What's going on here?

"Petey's view that it comes down to Ohio isn't what Nate says."

What Nate says tends to be far less interesting than Nate's numbers.

As of the moment, Nate's numbers have the chances of Obama winning the election if he loses OH at 17%.

17% isn't zero, but it's only about one in six.

reading numbers without obtaining context of what they mean through the analysis of the person giving them to you is ignoring fundamentals. for example, understanding that obama has the strongest GOTV in history and McCain doesn't have this level of support can not be understood simply by reading 17 percent since that 17 percent doesn't include the unique aspects of the present race. By analogy, your argument is the equivalent of ignoring Rove's evangelical outreach in 2004. it also ignores or under represents the changed map with CO for example in the mix. your argument depends on Mccain winning states like Michigan.

As of the moment, Nate's numbers have the chances of Obama winning the election if he loses OH at 17%.

Yeah, but isn't that because Obama losing Ohio would be likely to happen along with a general decline in his popularity nationwide? If that's the case, then it's not really right to say the election depends on Ohio. Losing Ohio won't cause Obama to lose Virginia, Colorado, Iowa, or New Mexico.

"it also ignores or under represents the changed map with CO for example in the mix. your argument depends on Mccain winning states like Michigan."

Go look at Nate's numbers. They're free.

At the same time he has Obama only at a 17% chance of getting to 270 EV's without OH, he's got Obama with a 58% chance of winning CO and a 67% chance of winning MI.

Look at all the numbers and also understand them in context. He's giving the chance of an event happening based against historical context/modelling plus polling data. He's not however discussing things like the economy, GOTV, the nature of the GOP brand, etc. The later are the fundamentals that tell you whether the probable outcomes are presently more or less likely to happen. You said you just check out the numbers, and I can see why- they favor your spin, but they don't reflect the entire reality- thats my point. Its a little like reading the ramussen numbers recently without reading the pollsters own critique of his likely voter model. and again, it's like not understanding the evangelical vote of 2004. I note you ignored all of that to just focus on the numbers, but that's not a response to what I am writing.

"Yeah, but isn't that because Obama losing Ohio would be likely to happen along with a general decline in his popularity nationwide? If that's the case, then it's not really right to say the election depends on Ohio."

Semantics, to a large degree. To be precise, one would say that OH is likely going to be an excellent proxy for the entire election.

"You said you just check out the numbers, and I can see why- they favor your spin"

What on earth is my spin here?

Do you think I own TV stations in Dayton and Columbus that need more ad revenue?

I told you to check out the numbers because you claimed the importance of OH in his results was based on the assumption of Obama losing CO and MI, and that's not the case.

let's play an exercise petey- what were kerry's numbers in MT, IN and CO in 2004? when you can answer that one honestly and compare and contrast with those of obama, you will understand the fundamentals argument better. for that matter, what was kerry's GOTV at this point in the race. as i said I can see why y ou don't want to read. it destroyes your thesis. you aren't reading me here either. and what i said was that you can not simply reduce it to the numbers- that was my thesis. i used the states as an example of the thesis about fundamentals. not about the numbers. ie, the chances of him losing CO and MI aren't just a matter of historical understanding, the same is true of other places.

I see that there's a lively discussion of my site. Petey seems to be everywhere.

The methods for my Meta-Analysis are transparent and available to all. It presents a high-precision snapshot of recent polls. It is not a prediction - contrary to the description above.

The Meta-Analysis is automated, making it free of unwarranted assumptions. I check it myself to see what is happening. Because it is high precision, with an effective Margin of Error of less than one percentage point, it detects real swings and rejects noise. I find it a useful corrective for the daily buzz of polling commentary.

In 2004, I presented analysis using the same snapshot. As the race tightened in late October, I added an additional calculation that made an assumption about what is called the "incumbent rule," in which undecideds tend to vote for the challenger. I presented both the decided-voter calculation and the plus-undecided "correction." The correction was wrong. However, the decided-voter calculation was dead-on: Bush 286 EV, Kerry 252 EV. This is documented at election.princeton.edu.

In my view, the additional assumptions made by Nate Silver are not unreasonable, but they add uncertainty and speculation to what would otherwise be a more useful metric.

Sam Wang
Princeton University

"Petey seems to be everywhere."

I am the institutional memory of the blogosphere.

"It presents a high-precision snapshot of recent polls."

I'm not sure what makes it "high-precision", but it certainly is an snapshot of recent polls.

The problems come from the indiscriminate way you assemble the recent polls, the lunatic way you characterize your results, and your ignorance of simple polling math concepts.

"with an effective Margin of Error of less than one percentage point"

Should be a pretty good tip-off that we're not dealing with the sharpest knife in the drawer.

Referring to MoE in this context is utterly without meaning.

"it detects real swings and rejects noise."

It's actually incredibly noisy. Several of Nate's decisions in his methodology serve to dramatically reduce noise compared to Wang's methodology.

"This is documented at election.princeton.edu."

Worth noting that you've chosen not to document most of what you wrote in 2004, which is somewhat understandable given how much more bizarre your characterizations were in 2004.

Ahhhh, see then clearly the Kerry landslide in 2004 actually did happen!!

Did they say how the undecideds voted? Because they sure would like to know....

repeat: was kerry competitive in the states in which obama is now competive and did he have safe states like obama does?

So you think Obama depicting Asians in sweat shops overseas stealing Americans jobs was a bit of fear mongering and racism???

Or were we just supposed to assume he was warning us that he planned to sleep with all Asian women as President?

oh- and as FL Dem at open left pointed out- if this is another wave year after 2006- its difficult to model this if not impossible- how is this factored in. what about the record level of dems registration even outside of OBama's efforts?

Petey,

You are reading Nate Silver's stats wrong. He says there is a 17% chance of Obama losing Ohio and winning the election, not that if Obama loses Ohio he has a 17% chance of winning the election. In fact, Obama has approximately a 40% chance of winning the election if he loses Ohio according to Silver's model. Here's how it works: according to Silver's model, if you ran the election 100 times Obama would win Ohio 57 times to McCain's 43. Out of 57 times that Obama won Ohio, Obama would win the election approximately 54 times to McCain's 3. Out of the 43 times that McCain won Ohio, he would win the election approximately 26 times to Obama's 17. So who do you think needs to win Ohio more?

"You are reading Nate Silver's stats wrong. He says there is a 17% chance of Obama losing Ohio and winning the election, not that if Obama loses Ohio he has a 17% chance of winning the election. In fact, Obama has approximately a 40% chance of winning the election if he loses Ohio according to Silver's model."

Dead wrong.

According to Silver's numbers today:

17.4% chance of Obama getting to 270 if he loses OH.

7.4% chance of Obama getting to 270 and he loses OH.

-----

There are a couple of ways to show why you're dead wrong here, but one way is to look at your results. You write:

if you ran the election 100 times Obama would win Ohio 57 times to McCain's 43. Out of 57 times that Obama won Ohio, Obama would win the election approximately 54 times to McCain's 3. Out of the 43 times that McCain won Ohio, he would win the election approximately 26 times to Obama's 17.

This would mean that Silver's numbers today would have Obama winning the election 54 + 17 or 71 times out of 100. But Silver's numbers today have Obama with a win percentage of 63%.

All right Petey. You win that round. As much as it pains me to say it. But I still think your conclusions are off. According to Silver's site there is a 40% chance of Ohio being the "tipping" point state, 33% chance of Michigan, 24% Colorado and 21% Virginia. Nate Silver doesn't seem to think his numbers give the sort of out-size importance to Ohio that you do. It's number one on his list of tipping points, but not overwhelmingly so.

If I were a betting man, Nate Silver's model would probably be better... in all other cases of trying to understand what's going on in the race? Princeton times infinity.

I just want a better picture of what poll numbers are doing... I don't need Nate to weight those numbers based on some bizzaro demographic model for me. I can read.

"It's number one on his list of tipping points, but not overwhelmingly so."

No disagreement there.

But his numbers still have OH going with the winner over 90% of the time, which is pretty damn interesting.

-----

I'll add that I wish I could figure out what the percentages on the "Tipping Point States" actually are percentages of.

Those are some of the most interesting numbers he breaks out of his model, but Nate has never updated his FAQ to explain exactly what the percentages stand for.

And also, it'd be interesting to see the results if Nate broke out individual "candidate X loses state Y, wins election" numbers for MI, CO, and VA as he currently does for OH.

Petey: I'm surprised, as I would have expected a smart guy like you would prefer unadulterated data... I didn't expect that you wanted Nate Silver to chew your food for you first. Fascinating.

Petey likes data that confirms Petey's predetermined views. that's why Petey can't respond to my issue about the fact that we've had a wave election in 2006, and 2008 is expected to be the same.

"If I were a betting man, Nate Silver's model would probably be better... in all other cases of trying to understand what's going on in the race? Princeton times infinity."

Now, y'see, it's stuff like this that really confuses me. You say a compliment like it's an insult.

The model that is better "if you were a betting man" is the model more useful in getting an understanding of what's going on in the race, since it's the model more likely to correlate well with actual outcomes.

Isn't that the point?

If you just want to see who won last week's news cycle, check out the Gallup and Rasmussen trackers, since at least they're current.

-----

"I just want a better picture of what poll numbers are doing... I don't need Nate to weight those numbers based on some bizzaro demographic model for me."

I fully agree it would be useful for someone to apply sensible methods to current state polling data alone. But that's not what Wang is doing. The characterizations he arrives at from his data are lunatic, and his data selection is lousy.

If you just want a crude picture of the state of the map, Electoral-Vote.com isn't any worse than Wang, and at least they doesn't mischaracterize their results the way Wang does.

Isn't that the point?
No Petey, because I don't expect Silver's model to be an accurate predictor... at least no better than your average demographic model. If I had to bet and Silver vs. Wang was all I had, I'd probably go Silver. Which reminds me... has Silver ever thrown white noise in his model to see what it does? He should put in white noise for poll results and report how his model performs if he wants to be all pro... my guess? No different, because I don't think polls are really driving his results.

"Petey likes data that confirms Petey's predetermined views. that's why Petey can't respond to my issue about the fact that we've had a wave election in 2006, and 2008 is expected to be the same."

Y'know, prior to the Berlin stuff showing up in the polls and pushing Obama's state polling numbers down a bit, Silver's numbers had a 25% chance of Obama winning 375+ EV's, which would indicate to me that his model is not fundamentally hostile to '08 being a wave election for Dems at the Presidential level.

"I don't think polls are really driving his results."

Of course they're driving the results. That's why his numbers move daily. The non-poll stuff he's mixing in doesn't move.

And if you read the FAQ, you'll see that his weighting of non-poll data is pretty light.

The bulk of the results he's deriving actually do come from polls, and as polling frequency increases, the weight given to the non-poll stuff goes even lower.

I understand that you'd prefer a model based solely on poll data, but that doesn't mean Silver's methods fit the exaggerated view of them you seem have. His methods are open to examination, and consist of a large number of very sane decisions.

"I don't expect Silver's model to be an accurate predictor"

Neither do I. But I am interested in what Silver titles his page: electoral projections done right. And he basically lives up to that title - certainly better than any publicly available effort ever done.

except that most polls say that most people aren't paying heavy attention right now. so the trip abroad may have moved political junkies, but doesn't explain the poll fluctuation. I think a lot of this is again about the fundamentals of what this year is. this is why i compare the the states in which obama shouldn't even be close to see that he is in fact close, especially compared to Kerry. This to me along with the 4 year shift in partisanship that's be moving away from the GOP tells me a lot. This was reported today in the NY Times. As has been the change on the ground game of the two parties with the Democrats now leading. Remember it requires in some of these states very little to win the state. I think we have only a rough idea of the fact Obama is probably ahead, but after that, this is such a different animal because this is a continuation of a potential reallignment going on here. this is difficult, if impossible to model. We probably will not know until after the election exactly what happened and why. This is why I keep returning to the fundamentals argument. It's because its about the best we can hope to gauge.

Petey gets results!

Nate updated his FAQ this morning so I can now understand precisely what the percentage in the "Tipping Point States" is referring to.

Petey: I don't mean to suggest that what Nate Silver is doing is wrong... indeed, I said that his model would be more valuable to me than Wang's if I was trying to make money off of this election. However, my personal opinion is that Nate Silver is half stats and half pundit... and I like the pundit part, as he is a good writer, but his model conforms to his own world view and his own conception of the important factors. Now, he may be right, but I don't like the idea of assuming interdependence between states via demographics... certainly that is reasonable, but I see no reason to do it unless you are trying to grab every last scrap of power our of the model to make a prediction... i.e. if you are trying to make money. Why not just let the data tell you what's dependent on what?

Personally I just want a solid and statistically rigorous poll aggregate. YMMV and that's OK.

Sam Wang would be more impressive if he defined "useful metric" more clearly. What specific "metric"? And how is it "useful"? As far as I can see, he's just as much in the oracle business as Nate Silver, but with a rather more pompous tone. Justify the rhetoric, Sam!

A metric is a quantity that describes some important aspect of a phenomenon. In this case the metrics of interest are the expetcted number of electoral votes for each candidate and a "win probability." Both can be calculated for today's conditions (my site) or in some complex model of the future (Silver's site).

Also: pompous? Moi?

And the use of these "metrics"? Electoral probabilities aren't facts, and as such are of little use, especially daily snapshots that go out of date, like, tomorrow, dude. You can calculate a thousand possibilities, depending on "metric" you prefer - but they remain just possibilities. Now a tire gauge, that's useful!

And yes, you do come across as pompous. Sorry to tell you the bad news, but hey, better now than never, eh? But there's still time. Just take a deep breath, look courageously in the mirror, embrace your inner higher angel and yell: "Yes, we can!" Repeat ad lib, or until the neighbors intervene...

"Also: pompous? Moi?"

FWIW, Sam, I don't think you're pompous.

However, I do think you are Humpty Dumpty.

"When I use a word,' Humpty Dumpty said in rather a scornful tone, 'it means just what I choose it to mean — neither more nor less."

Your model is very crude, but reasonably unobjectionable. However, your characterization of your results is lunatic. You use words to characterize the results your model produces that are well outside of the common usage of those words.

You're either intentionally fraudulent or kinda stupid. Which of the two is not something I can venture.

The important difference between you and Nate Silver ends up not being the model. The difference ends up being intellectual honesty.

Petey, have you forgotten your own long and shameful record of slandering everyone who disagrees with you? Sam Wang may be mistaken about things, but I see no sign that he doesn't believe in what he says. You should take some time, reflect on your own problems, and then leave the internet alone. Society would be better for it.

Petey, Petey, Petey... first, you need to calm the fuck down, 'cuz calling an academic "intellectually dishonest" is about as nasty as you can be. I hope you didn't mean it, but you are full of surprises, so we'll see. Second, Professor Wang has MATLAB scripts of his analysis available to the public... does Nate Silver open up his black box for the masses?

In my book, intellectual honesty is publishing you methods so that any rube can check them.

Sometimes I suspect that Petey is just Rick Santorum in drag. They even sound the same. What's wrong, Petey? Did Larry Craig promise you forever before the bathroom incident? Still bitter that Edwards dumped you for Rielle?

"Petey, Petey, Petey... first, you need to calm the fuck down, 'cuz calling an academic "intellectually dishonest" is about as nasty as you can be. I hope you didn't mean it,"

I'm perfectly calm. I did mean it. And I'm correct.

Dig into the weeds here, if you have the time and inclination, J.W. Hamner, and I think you'll find I have truth on my side, just as I had truth on my side when I made a strident assertion upthread to contradict Sean on a different matter.

"Second, Professor Wang has MATLAB scripts of his analysis available to the public."

As I've written here and elsewhere, I've got no war with Wang's formula. It's a very crude measurement of state poll data, but it does what it does.

The problem comes when Wang explains what his formula produces, and consistently says things which aren't so, according to the commonly accepted meanings of his words.

Then the question becomes one of fraudulence or stupidity, and while I certainly can't determine which for sure, folks have been pointing out the problems to him for a while, yet he keeps repeating the same misinformation. That seems to qualify for intellectual dishonesty, no matter what the source of the mistakes.

-----

Is this a Bio-Bros before Truth kinda issue for you, by any chance?

" does Nate Silver open up his black box for the masses?"

He does. Exquisitely. In a rather simple FAQ page that can be understood by someone with reasonably basic HS math.

But whether Nate is playing on the up and up or not has absolutely nothing to do with the snake oil that Sam is selling.

Wang is out of his discipline, and he either doesn't recognize the difference here, or has intentionally chosen to ignore them.

""Second, Professor Wang..."

And finally, in you assertion of authority on the part of "Professor Wang", you do understand that Wang is a Professor in a very different discipline than this one, right?

(I've got no problem with folks coming in from a different field, bringing in their knowledge, learning the local language, and becoming an expert. That's bascially what Silver did when he came in from his baseball stats background. But that's not what Wang has done here in any way shape or form.)