Hurricane October Surprise will carry the ball for Obama, all Romney can do in the interim is look stupid.

It's over.

Well, aside the normal old guard/new guard war going on, a 77% chance is nothing close to a lock. In an all or nothing game with a single roll of the dice, a 1-in-4 chance of losing is a huge risk. It's nowhere near pack up and cede the race to Obama territory.

I think that part of the problem here is that any non-statistical pundit will look at any race where both sides have a legitimate shot at winning and call it close to 50/50, because anything more than that implies that the underdog has no shot (which isn't the case in a 77/33 race).

I think that part of the problem here is that any non-statistical pundit will look at any race where both sides have a legitimate shot at winning and call it close to 50/50, because anything more than that implies that the underdog has no shot (which isn't the case in a 77/33 race).

bhaakon loves the ladies

Hurricane October Surprise will carry the ball for Obama, all Romney can do in the interim is look stupid.

It's over.

Does low voter turnout in the states that are most affected by the Hurricane , and also happen to be all solidy Obama have any impact on Swing states like Ohio and Virginia ?

If Voter turnout is low in Ohio and Virginia, does that favor Romney ?

Honest questions.

Also, I picked up on this dynamic the author is talking about right after the first debate when Silver went on Piers Morgan, and Morgan pretty much sneered at him derisively as the interview ended and said "Nate Silver ladies and gentlemen, not nearly as confident as he was a few weeks ago". It was pretty weird and the attitude made me immediately think of the anti saber backlash.

Hurricane October Surprise will carry the ball for Obama . . .

I'm hard pressed to see how approving the states disaster area designations and helicoptering over NJ has any impact.

Speaking of New Jersey, I'm sure people of all persuasions will condemn Senator Robert Menendez (D-NJ) for stiffing a couple of poor working girls.

Speaking of New Jersey, I'm sure people of all persuasions will condemn Senator Robert Menendez (D-NJ) for stiffing a couple of poor working girls.

As long as it's adult women instead of young boys or farm animals. It's getting hard to work up a moral outrage these days.

As long as it's adult women instead of young boys or farm animals. It's getting hard to work up a moral outrage these days.

What about Menendez offshoring jobs that could have gone to Americans?

Seriously though, a Senator that can't get laid in Washington? Without the risk of blackmail?

I think part of the problem, in addition to the left versus right and John Henry versus machine stuff, is that most people really don't understand what a 75% probability means. I often see arguments along the lines "you don't know A will win, and you don't know B will win, so you don't know anything." Only probabilities of 0%, 50%, or 100% are allowed in the discussion.

There’s a slight lead for Mitt Romney in national polls and slight leads for Barack Obama in swing-state polls, and no good way of predicting next Tuesday’s outcome beyond flipping a coin. ...

Way to announce you have no grasp of the subject.

I think that part of the problem here is that any non-statistical pundit will look at any race where both sides have a legitimate shot at winning and call it close to 50/50, because anything more than that implies that the underdog has no shot (which isn't the case in a 77/33 race).

Good to know someone's giving 110%!

I think part of the problem, in addition to the left versus right and John Henry versus machine stuff, is that most people really don't understand what a 75% probability means. I often see arguments along the lines "you don't know A will win, and you don't know B will win, so you don't know anything." Only probabilities of 0%, 50%, or 100% are allowed in the discussion.

There's plenty of this kind of thing even on the site for the Thinking Fan. Silver's 'out on a limb' for having the audacity to say that Obama is 70% to win the election. It's pretty funny. Or sad, depending.

As a resident of Nj I have no idea of menedez's qualifications or accomplishments. I did see him quietly standing in the background while gov Christie and president Obama did their thing in jersey. Seems as a senator he could get laid if he wanted to, but as someone has said, "I don't pay them for sex, in pay them to leave afterwards."

Good to know someone's giving 110%!

That's what the PEDs are for. If you're not cheating, you're not trying.

As a resident of Nj I have no idea of menedez's qualifications or accomplishments. I did see him quietly standing in the background while gov Christie and president Obama did their thing in jersey. Seems as a senator he could get laid if he wanted to, but as someone has said, "I don't pay them for sex, in pay them to leave afterwards."

It not like a notable figure like a senator can just go cruising and expect to remain anonymous, and he's probably too busy to put in the leg work anyway. If he wants to be a world class scuzzball, high priced escorts may be his only realistic choice. The key, as you and others have said/implied, is not to leave them with a motive to talk come forward down the line.

I figure a senator can make an "arrangement" whether or not cash is exchanged is obviously an issue. I happen to know his opponent's name but I would guess I would be in the very small minority. Obviously this late into the election season it would be difficult to "prove" whether the allegations are true or not, but paying for sex isn't an automatic forced resignation (i dont think this changes whether or not he gets reelected).I also have no idea whether or not the governor or the state legislature appoints a new senator in the event that menendez is elected and then resigns, but I guess that could potentially be significant.

It's a crapshoot. What we need is a full campaign season of between 130 and 154 contests between a field of candidates, followed by best-of-7 elimination rounds.

I wonder how sophisticated our Founding Fathers were in setting up the Electoral College? That turns marginal outcomes into all-or-nothing results.

If Voter turnout is low in Ohio and Virginia, does that favor Romney ?

Historically turnout helps Democrats, probably moreso in Ohio than Virginia I would guess. Turnout tends to vary more among the young and the urban, both historically stronger democratic bases. I guesstimate a bigger effect in Ohio because it has two large cities (plus Akron and Toledo aren't small). Virginia is more suburban DC and rural areas. 2004 was a bit of an anomaly in Ohio -- the Dems got great turnout but, somewhat unexpectedly, the Republicans matched them.

But as noted, NY, NJ, etc. are all pretty heavily Obama territory so while low turnout in those states might hurt his total vote count it's unlikely to impact that much on his electoral vote count. Obviously it could be hugely problematic if the feds aren't seen as doing enough to help people.

I knew Bob Menendez early on from some youth/HS-team bowling leagues we were in...and I'm still shocked he ever got his factory-crisp white shirt and eternally pressed black slacks off long enough to have children.

Purdy sucky bowler at the time who supposedly got a lot better by practicing all the time in the single laner under the White House.

Nate Silver going after Joe Scarborough on Twitter right now...

Silver: @JoeNBC: Every bookmaker from Las Vegas to London stands with our assessment of the odds.

Silver: @JoeNBC: If you think it's a toss-up, let's bet. If Obama wins, you donate $1,000 to the American Red Cross. If Romney wins, I do. Deal?

As a longtime Primate, I obviously like to look at things from the saber perspective. But I'm not convinced Nate Silver's model is the right way to do that. As I recall, it did pretty poorly in 2010 as well, in much the same way that it's behaving this year. All year long, national polls indicated a groundswell toward the Republicans, but Silver's model didn't pick it up.

To keep the issue technical rather than partisan, I think Silver's approach is too heavily weighted toward state polls, which show a big discrepancy with the national polls this year (see Sean Trende on the subject).

Conventionally, people think of local elections as being heavily correlated with national elections. A particular state or district will tend to have a consistent offset from the national generic vote, so that if the Democratic ticket gets 50%, then say 52% of the vote in that district will go for a generic Democrat -- plus a little for good candidates, minus a little for bad.

If you take that perspective, a big shift in the national polls should dramatically affect all of the state races, just like a storm surge affects the water level in every block of a city at once rather than each block independently. Going from 2 or 3 up to 2 or 3 down should completely upend the race -- Obama should go from 80% to 20-30%, not 70%.

From what I understand of Silver's model, it is heavily weighted toward the state polls. That's consistent with having Obama as the popular vote leader, as Trende's article suggests, and it's certainly a reasonable approach. But it also leaves you vulnerable to bad data -- polls are taken less often, have smaller samples, and may have weird samples. It also leaves you vulnerable to the possibility that the entire national vote shifts in unison -- first, because the infrequent state polls won't catch it, second because 50 polls moving in unison is very unlikely if you assume that each state is independent of the others.

If you want to incorporate both state and national polls into a consistent model, I think the best way would be to view the state polls as essentially a measurement of the offset from the national polls -- Instead of measuring Obama at 52% and a generic Democrat at 55% when the national average has the generic Democrat at 50%, the poll would really be measuring Obama as three points behind the generic
Democrat in a particular district, plus the generic Democrat as doing 5 points better in that district than in the nation at large. With this method, a three point national shift away from the generic Democrat puts Obama at jeopardy in the district, even if it's been a while since the district was last polled.

If you're going to incorporate large numbers of polls, I think it's very important that either the numbers you're compiling really are independent, or else your method has a way of teasing out the correlations (like, say, a covariance matrix). Treating the states as independent of each other is in my view wrong, and wrong in a way which will likely lead to the kinds of effects Silver is seeing.

From what I understand of Silver's model, it is heavily weighted toward the state polls. That's consistent with having Obama as the popular vote leader, as Trende's article suggests, and it's certainly a reasonable approach. But it also leaves you vulnerable to bad data -- polls are taken less often, have smaller samples, and may have weird samples. It also leaves you vulnerable to the possibility that the entire national vote shifts in unison -- first, because the infrequent state polls won't catch it, second because 50 polls moving in unison is very unlikely if you assume that each state is independent of the others.

1. Nate's model doesn't assume that each state is independent of the others. That's closer to Sam Wang's model. That's why Romney has as large as a 27% or so chance to win on Tuesday.

2. Nate's recent post, discussed in the other thread, points out that state polls have historically been more accurate than national polls, which is why Nate weights them heavily.

I'm hard pressed to see how approving the states disaster area designations and helicoptering over NJ has any impact.

It reminds voters that Romney wanted to drastically scale back federal responsibility for disaster relief, and that the last time his party held the presidency we had Katrina and Brownie and got a preview of this policy in action.

Hurricane October Surprise will carry the ball for Obama . . .

I'm hard pressed to see how approving the states disaster area designations and helicoptering over NJ has any impact.

It may or may not have any direct political impact, but when you've got Obama's handling of the situation rated at 78% positive vs 8% negative (WaPo poll), that's certainly not likely to hurt him with only 5 days to go before the election.

-----------------------------------------------

But as noted, NY, NJ, etc. are all pretty heavily Obama territory so while low turnout in those states might hurt his total vote count it's unlikely to impact that much on his electoral vote count. Obviously it could be hugely problematic if the feds aren't seen as doing enough to help people.

In that same WaPo poll, the federal government's response got a 73% positive - 8% negative rating, with only 1% calling it poor. But gee, I thought "the government" was supposed to be the enemy!

As I recall, it did pretty poorly in 2010 as well, in much the same way that it's behaving this year. All year long, national polls indicated a groundswell toward the Republicans, but Silver's model didn't pick it up.

Nate's model missed the House by a bunch of seats in 2010 - though I believe he beat RCP and Pollster - but he predicted a large Republican wave in the House, just not quite as large as the wave that happened. He was pretty much spot-on in the Senate. "Didn't pick it up" is not a good description.

From what I understand of Silver's model, it is heavily weighted toward the state polls. That's consistent with having Obama as the popular vote leader, as Trende's article suggests, and it's certainly a reasonable approach. But it also leaves you vulnerable to bad data -- polls are taken less often, have smaller samples, and may have weird samples. It also leaves you vulnerable to the possibility that the entire national vote shifts in unison -- first, because the infrequent state polls won't catch it, second because 50 polls moving in unison is very unlikely if you assume that each state is independent of the others.

The reason that Nate's model weights the state polls more heavily is not that he believes the state polls are more accurate. It's that there are more of them. The samples are larger. We have had far more people polled in the last month for state polls than for national polls.

Your argument seems based on the presumption that the national polls constitute a larger n, when in fact they constitute a much smaller sample size.

EDIT: And as has been pointed out, Nate model explicitly treats the states as interdependent based on a presumption of relatively uniform swing.

Nate's recent post, discussed in the other thread, points out that state polls have historically been more accurate than national polls, which is why Nate weights them heavily.

He didn't say that, actually. His model gives equal weight (adjusting for pollster and such) to state and national polls. The state polls only get more weight in the aggregate because there are so many more of them.

(He did say that the state polls have usually been more accurate as predictors of the national popular vote, but he didn't say that's incorporated into his model.)

It reminds voters that Romney wanted to drastically scale back federal responsibility for disaster relief, and that the last time his party held the presidency we had Katrina and Brownie and got a preview of this policy in action.

The beauty is that this particularly loopy part of Romney's agenda was almost certainly forced on him by the Tea Party crazies. It's hard to believe that anyone with Romney's undoubted overall intelligence would ever come to such a position independently, but this is what you get when the most vocal part of your party's base is absolutely deranged.

He didn't say that, actually. His model gives equal weight (adjusting for pollster and such) to state and national polls. The state polls only get more weight in the aggregate because there are so many more of them.

I didn't say he weights them more heavily than national polls.

Speaking of Brownie, he seem to be a legitimate political idiot for having the balls to come out on even remark on the subject, let along taking shots at the current administration while doing so.

The national polls no longer say Romney is leading so there is no longer as big of a disconnect between state and national polls. That is an old news narrative.

As for turnout I doubt it effects Ohio but if it did it would probably favor Obama this time around. The early votes have already been cast and they are solidly for Obama. The day of votes in Ohio went to McCain solidly in 2008 so if there is low voter turnout on the day of the election it would probably mean more to Romney than Obama. But again I doubt the storm will cause lower turnout in any state where turnout would be important to the race for either candidate.

I didn't say he weights them more heavily than national polls.

True, my bad.

To quibble, you did that the historically greater accuracy of state polls is why he weights them heavily, and he didn't say that. He weights all the polls evenly as a methodological thing, as a poll aggregator.

I should say, in response to Zach, that 2010 shows at a minimum the extent to which it's possible for a model/aggregator like 538 (or Pollster or RCP) to be wrong, and if the polls and the aggregation of the polls are off this year by as much as they were in 2010, that could easily lead to a solid Romney victory. Silver's model is giving Romney something like a 3-to-1 or 4-to-1 shot, and those are things that happen all the time. That's the Royals beating the Yankees, that's Omar Infante getting a hit off of Jake Westbrook, that's any number of things that wouldn't surprise us terribly at all if they came to pass in the baseball season.

Obama up 6 in nbc iowa poll...

I know Romney supporters herr are confident, but at some pojnt dont you want to see Romney up in some battleground polls somewhere outside of Florida and Nirth Carolina?

I'm on Team Gay Stats Wizard, but shouldn't Nate be giving Scarborough odds on a Romney victory? Making a straight up bet reinforces the false impression that his spreadsheet thinks Obama is a sure thing.

EDIT: the more I think about it, I think offering this bet was a bad idea by Nate. It's unprofessional, it doesn't reflect the current impression of the 538 formula, and it dilutes the perception that Nate is objective about the election's ultimate result.

Winner: doesn't matter
Losers: American people, and lots of people in the Middle East

To quibble, you did that the historically greater accuracy of state polls is why he weights them heavily, and he didn't say that. He weights all the polls evenly as a methodological thing, as a poll aggregator.

Right. He weights them heavily. 50% is a heavy weight. He also doesn't weight all the polls evenly -- those with a better track record get greater weight. He also has variables reflecting things like economic fundamentals. Silver is more than simply a poll aggregator.

1. Nate's model doesn't assume that each state is independent of the others. That's closer to Sam Wang's model. That's why Romney has as large as a 27% or so chance to win on Tuesday.

Could you repost it here? I couldn't find a good explanation on his site.

2. Nate's recent post, discussed in the other thread, points out that state polls have historically been more accurate than national polls, which is why Nate weights them heavily.

Historical accuracy is an unsophisticated way to think about this issue. There are only a few points to look at, and the methods and circumstances of polling change regularly.

The general problem of consolidating several different polls is tricky. Technically, you shouldn't even try to "weight polls" -- you should create a new "poll" that combines the different samples. In principle, if you're consolidating the state polls correctly, you should agree with the national result. If you want to do it right, you're probably going to have to spend some time talking to real statisticians, or else work it out as a pure math problem. Trying to weight different polls or to apply fudge factors based on limited histories is asking for trouble.

I don't think the storm matter much either way for the election. Especially once Dick Morris weighed in.

Regarding polls and Nate's method, I suspect that his model works better for Presidential elections than House elections, largely because there is much more data (many house races are only sparsly polled.

I really like that Nate is going in on his model. Good for him to stick up for it, I think you should be willing to risk a bit of pride if you believe in something.

Every bookmaker from Las Vegas to London stands with our assessment of the odds.

There are no bookmakers in Las Vegas giving odds on the presidential election. Well, at least no legal ones. Taking bets on the outcome of elections is illegal in all of the United States.

In principle, if you're consolidating the state polls correctly, you should agree with the national result.

No. In principle, you should treat state and national polls as evidence of the state of the national race and project the result based on the aggregate. There is no good reason to dismiss the evidence of the state polls if the state and national polling diverge.

Could you repost it here? I couldn't find a good explanation on his site.

The archiving at the NYTimes site really sucks. This is from the FAQ on Nate's old site:

Fourthly, we simulate the election 10,000 times for each site update in order to provide a probabilistic assessment of electoral outcomes based on a historical analysis of polling data since 1952. The simulation further accounts for the fact that similar states are likely to move together, e.g. future polling movement in states like Michigan and Ohio, or North and South Carolina, is likely to be in the same direction.

at some point dont you want to see Romney up in some battleground polls somewhere outside of Florida and North Carolina?

He's not up in North Carolina by much, if at all; the recent polls have been showing the race as a statistical tie, more or less. Obama's team has been working the state really, really hard the past couple of weeks. Obama's actually being helped, in a somewhat perverse way, by the fact that Pat McCrory has a huge lead in the governor's race; that's likely to depress Republican turnout.

Early voting always favors the Democrats, because (a) they push their base to vote early and (b) a larger portion of their base falls into the category of people likely to be working on Election Day and thus with a larger interest in voting early. This year - at least in NC - their advantage is smaller than it was four years ago.

We are seeing two things happening, neither of which I like: (1) more aggressive campaigning at the polls, including an increase in voter challenges and (2) more misinformation being spread about the voting process. Apart from the usual tactics (Republicans vote on Tuesday, Democrats on Wednesday), we're starting to see a trend of people being told that they have to have ID if they want to vote early (only true if you are not already registered), that you can register and vote on Election Day (incorrect, that's true only during early voting), that they can vote anywhere in the state on Election Day (true, but they'll have to cast a provisional ballot if they are outside of their home precinct - a process which tends to irritate registered voters when they've been told they can just go ahead and vote anywhere), and so forth.

As an election judge, my advice to registered voters is that if you aren't sure what to do, ask one of us. That's why we are there.

-- MWE

I'm on Team Gay Stats Wizard...

Winner!

I am surprised that this article doesn't mention the Rasmussen Polls. They had been nearly all in the Romney column. Rasmussen called 2004 & 2008 extremely well. I would say there's an element of zero-sum between Silver & Rasmussen in this election as well.

According to Rasmussen 16% of white democrats now support Romney.

Rasmussen called 2004 & 2008 extremely well.

Rasmussen did well both years, but it was only after 2008 that Rasmussen's results started to diverge from the national consensus. In 2010, their divergence led them to be among the least accurate pollsters in the country. They could be right this year, of course, but in the only recent election where Rasmussen diverged from the national consensus, he called things quite wrong.

According to Rasmussen 16% of white democrats now support Romney.

That sounds perfectly in the range of correct to me. There are always some crossover voters, and on top of that, lots of Southern whites still identify as "Democrats" but were never going to vote for Obama.

Rasmussen tends to overstate Republican support, just as PPP tends to lean the other way.

-- MWE

No. In principle, you should treat state and national polls as evidence of the state of the national race and project the result based on the aggregate. There is no good reason to dismiss the evidence of the state polls if the state and national polling diverge.

If your only interest is in the bottom line result, you can certainly do this. But you will find yourself exposed to all sorts of nasty surprises when your assumptions turn out to be overly optimistic.

Comparing the aggregate state results to the national results is a way of testing your aggregation process. It helps you make sure that you're doing the aggregation right. Then, once you're sure you're doing it right, you find the aggregate.

It's like debugging a computer program. If you have two measurements of the same quantity, you test to make sure they agree with each other. If they don't agree, that helps you figure out how to do it better next time.

Speaking of New Jersey, I'm sure people of all persuasions will condemn Senator Robert Menendez (D-NJ) for stiffing a couple of poor working girls.

Sounds like the guy who was hosting and/or touting for Melendez might have been the one who actually slipped out from under the bill. It amounts to the same thing for the ladies, of course. Cheating hookers is bad karma, and really stupid to boot if you're a public figure.

I don't really care about the prostitution itself because a) it's legal in the DR as long as you don't work through a pimp and b) Melendez is divorced and not really a "family values" candidate, so there isn't an obvious hypocrisy angle. I think it's skeevy and sort of off-putting to think about someone like him having sex with anyone under any circumstances, but that's not really his fault.

I guess he might be in trouble for accepting improper benefits. I'm not sure what the law is on that, re: personal services.

In a world where David Vitter remains a heartthrob among "family values" voters and taxpayers foot the bill for Callista Gingrich's healthcare, who cares about Melendez?

EDIT: the more I think about it, I think offering this bet was a bad idea by Nate. It's unprofessional, it doesn't reflect the current impression of the 538 formula, and it dilutes the perception that Nate is objective about the election's ultimate result.

Agreed on this and your odds comment (if he 'must' do it).

As an election judge, my advice to registered voters is that if you aren't sure what to do, ask one of us.

What does that entail? I've thought about volunteering to help w/ my local elections board but have yet to do so.
Anyway, I'd be very surprised if Romney didn't carry NC.

Rassmussen is historically and consistently right-leaning in its polling, and has been all year.

There are no bookmakers in Las Vegas giving odds on the presidential election. Well, at least no legal ones. Taking bets on the outcome of elections is illegal in all of the United States.

I guess that's why the law abiding Romney is always putting his money offshore.

It's like debugging a computer program. If you have two measurements of the same quantity, you test to make sure they agree with each other. If they don't agree, that helps you figure out how to do it better next time.

You're treating a social science problem as a math problem, I think. Measuring public opinion is fraught with errors. Measuring public opinion and then weighting this opinion based on likelihood of voting is even harder. There's no reason to expect in public opinion polling, even with extremely large n's, that two measurements of the same quantity will perfectly converge. It's not necessarily a bug in the program if you get divergence. That's the nature of complex human events.

It could simply be that the two measurements aren't the same. There's no way to know, in advance of the election, which measure is right. (In the past, the state measurement has been somewhat more accurate, but not by a large amount and we're only looking at a small number of very different and differently polled elections.) Given that we don't and can't know which measurement is correct, the best choice is to aggregate the two via the best available methods.

In a world where David Vitter remains a heartthrob among "family values" voters and taxpayers foot the bill for Callista Gingrich's healthcare, who cares about Melendez?

Real Americans. And they are appalled!

It was pretty weird and the attitude made me immediately think of the anti saber backlash.

And for exactly the same reasons. Folks in a job where you can just make #### up like sportswriting and political punditry don't take kindly to their gut feelings, hunches, and insider access playing second banana to analysis from someone not of the guild. They feel threatened, and in a better world, they actually would be.

Folks in a job where you can just make #### up like sportswriting and political punditry don't take kindly to their gut feelings, hunches, and insider access playing second banana to analysis from someone not of the guild. They feel threatened, and in a better world, they actually would be.

The smart pundits will learn to combine the new models with their old school shoe leather reporting, and get the best of both worlds. The ones who don't do that will still have their columns in the Washington Post to console them.

Pretty cool look at party trends in the Senate

dilutes the perception that Nate is objective about the election's ultimate result

Nate isn't objective about the election's ultimate result, and that's the problem.

Only fools suggest that Nate is cooking the books. The issue is more subtle; whether someone who is obviously very, very biased, even if that person is striving to be as objective as possible, generate unbiased outcomes?

Generally, the answer to this is "no". It's why scientists are so scrupulous about blind studies; not to prevent outright fraud (which is not particularly hindered by blind/double blind studies, b/c its so easy to cook the books post facto). Like locks or low fences, blind data keeps honest folks honest, so to speak.

It's pretty much impossible to have blind data for an election model, or to find a truly unbiased person. But given the number of subjective judgments that go into the construction of Nate's model and his acknowledged strong pro-Dem bias, the objectivity of his model is inherently questionable. To argue otherwise is to question an enormous amount of received wisdom about scientific bias and conflicts.

What about Menendez offshoring jobs that could have gone to Americans?

Heh.

Seems to me there's an "invisible hand" joke to be made in here someplace.

It's pretty much impossible to have blind data for an election model, or to find a truly unbiased person. But given the number of subjective judgments that go into the construction of Nate's model and his acknowledged strong pro-Dem bias, the objectivity of his model is inherently questionable. To argue otherwise is to question an enormous amount of received wisdom about scientific bias and conflicts.

But partisan bias isn't the only or the primary conflict Silver has. The biggest one is self-differentiation in a clogged and competitive market. You can easily see the effects of this - Silver's model is almost certainly over-complicated, and he publishes results with somewhat comically specific numbers down to the decimal place. What he's selling with his model is complication and accuracy, and he's built a model that serves those purposes.

I'm not saying that there could be no partisan bias, but I think that in considering the things that could screw up Silver's model, it's well down the list and likely swamped by other problems. Of which the biggest, again, are the pressures of the marketplace.

We can see the effects of market pressures on a number of the peculiarities of his model. We don't need to appeal to ecological fallacies about the tendencies of bias to creep in to make this judgment. If you believe that Silver's partisan bias is affecting his model - which is of course very possible - then you should be able to hypothesize where and how, and test that. (We even have several elections against which to test it.)

According to Rasmussen 16% of white democrats now support Romney.

That sounds perfectly in the range of correct to me. There are always some crossover voters, and on top of that, lots of Southern whites still identify as "Democrats" but were never going to vote for Obama.

Agreed--but I was gonna add, I wonder how many of those folks are under the age of 55.

But partisan bias isn't the only or the primary conflict Silver has. The biggest one is self-differentiation in a clogged and competitive market. You can easily see the effects of this - Silver's model is almost certainly over-complicated, and he publishes results with somewhat comically specific numbers down to the decimal place. What he's selling with his model is complication and accuracy, and he's built a model that serves those purposes.

I'm not saying that there could be no partisan bias, but I think that in considering the things that could screw up Silver's model, it's well down the list and likely swamped by other problems. Of which the biggest, again, are the pressures of the marketplace.

Agreed. And the other problem with throwing all-in with Nate's model is that PECOTA has many of the same issues with over-precision and NIMBY that I think exist with Nate's model. People act like Nate "sticking up for his model" is a good thing; it's not. If your model gets reviewed by the whole country and you insist that every critique is incorrect, then you're almost certainly ignoring good critiques.

That being said, I'm quite certain 538 has the best model out there, and that it's correctly reporting that Obama's victory chance is much higher than the "pundits" suggest; but I think that says more about the crappiness of the competitors than the quality at 538; Nate's stuff has never been very good, even BITD as a baseball stats guy.

Only fools suggest that Nate is cooking the books. The issue is more subtle; whether someone who is obviously very, very biased, even if that person is striving to be as objective as possible, generate unbiased outcomes?

That's a reasonable point, as long as you'd also apply it to the "Christian conservative" co-founder of RCP and the Republican pollster Rasmussen.

But of course Matt's got the overriding point. The marketplace forces these guys to (a) differentiate their methods in order to stand out from the crowd, and (b) strive for as great an accuracy as is humanly possible.

That's a reasonable point, as long as you'd also apply it to the "Christian conservative" co-founder of RCP and the Republican pollster Rasmussen.

To put it simply: no ####. You guys get sloppy because you argue with the laundry-rooters like Joe K.; do you really think anyone capable of independent thought thinks that Rasmussen is any less biased?

I just get exasperated when people argue that Nate is honest and trying to be objective, so, OBVIOUSLY his model is unbiased. Next time I get conflicted out of legal work because my dad is representing the other side, I'll let the client know that I'm honest and I try to be objective. You get my drift.

People act like Nate "sticking up for his model" is a good thing; it's not.

I disagree. If you do somethign and believe in it you should stick up for it. Of course if your model fails (or performs worse than you want) and you are still sticking up for it (making excuses) then there is a problem.

Right now is a terrible time to be tinkering/re-evaluating though. Why on earth would anyone listen to criticism now? The pressure is higher, deadlines loom, and there really isn't any more data (a perfect recipe for making stupid mistakes) so why fool with the model?

Wait a week see what happens and go from there with more data. In the mean time I think he should feel free to stick up for his model as much as he wants.

What does that entail? I've thought about volunteering to help w/ my local elections board but have yet to do so.

You are trained in how elections should be conducted in the state of North Carolina, then you spend Monday night setting up the polling place and all day Tuesday (from 1/2 hour before the polls open until about an hour or so after they close) managing the process in accordance with NC election law and ensuring that everyone votes appropriately. For about 99% of the people who come in, that involves looking them up in the poll book, having them verify their information is correct and signing the authorization to vote, handing them a ballot, and making sure that they put the ballot into the tabulator when they've marked it. Generally, about 1% of people have problems - you have a separate process for them. Wake County is very thorough - we have detailed procedures for just about everything that can happen and a dedicated help line for situations where the procedures don't cover it. We have 6000 registered voters in our precinct, we usually get about 1000-2000 on Election Day, and usually no more than about 20 or 30 have issues. We had about 10 provisionals last time, mostly from people who come in at the last minute and are in the wrong place. I expect more this time; we're likely to get a lot of people who haven't voted since 2008.

If you have any interest at all in the political process, I'd encourage you to volunteer. You do get paid, too.

Anyway, I'd be very surprised if Romney didn't carry NC.

He's still the favorite, but there are a lot of Obama workers on the ground right now.

-- MWE

People act like Nate "sticking up for his model" is a good thing; it's not. If your model gets reviewed by the whole country and you insist that every critique is incorrect, then you're almost certainly ignoring good critiques.

I certainly agree with this in theory, but I haven't seen Nate absolutely rejecting non-dumb critiques. If he has rejected non-dumb critiques in favor of "sticking up for his model", that's a bad thing - do you have an example?

Right now is a terrible time to be tinkering/re-evaluating though. Why on earth would anyone listen to criticism now? The pressure is higher, deadlines loom, and there really isn't any more data (a perfect recipe for making stupid mistakes) so why fool with the model?

Wait a week see what happens and go from there with more data. In the mean time I think he should feel free to stick up for his model as much as he wants.

If it's a model, it speaks for itself. It's not advocacy. It reminds me of Mann "sticking up" for his climate hockey stick. He ended up looking very silly EVEN THOUGH HE WAS MOSTLY RIGHT, because some of the critiques were correct.

MCoA, my argument would be that he should say nothing. If he's a modeler, why is he advocating in the first place? Its the conflation of model-maker and advocate that makes me suspicious that he can be as clinical as advertised; he's got a lot invested in winning in the market. As you noted, this has nothing to do with a partisanship.

Right, the proper thing to do is to acknowledge plausible or even convincing critiques, but to leave the existing model in place for now and re-evaluate after the election.

I just get exasperated when people argue that Nate is honest and trying to be objective, so, OBVIOUSLY his model is unbiased.

I think Nate is trying to ensure that his model fairly reflects what is going on, without model bias in one direction or the other. That's not quite the same thing as being objective in his writings about it.

I do think that if the model doesn't accurately predict this year's race that Nate will spend some time figuring out what happened and make adjustments to account for it.

This is a good discussion of some of the reasons that political journalists are reacting the way that they are - and indeed there are some parallels here between the way that old-school writers react to sabermetrics.

-- MWE

I know Romney supporters here are confident, but at some point dont you want to see Romney up in some battleground polls somewhere outside of Florida and Nirth Carolina?

Sure, but even the polls that show Obama ahead have him losing Independents, often by double-digits. No one has been elected while losing Independents by more than a point or two. Not every poll publicizes the partisan breakdown of its sample, and I don't claim to have checked every one that does, but it looks like the polls that have Obama winning battleground states have ~D+7 (or greater) samples. Not buying that. I know something about Virginia politics, it just isn't a D+7 state. Not saying there is deliberate bias, but the declining response rate for polls may not break evenly on partisan lines.

Agreed--but I was gonna add, I wonder how many of those folks are under the age of 55.

Very few - most of them probably started voting before the Civil Rights act.

Well, that and the less you have to worry about the future, the more sense it makes to vote for Romney. Young people still have a lot of future to worry about.

Don't pundits always pretend the race is 50/50? In 2004 and 2008, they were pretending that it was a toss-up not only right before election day, but into the night after the thing was all over but the shouting.

I'm on Team Gay Stats Wizard, but shouldn't Nate be giving Scarborough odds on a Romney victory? Making a straight up bet reinforces the false impression that his spreadsheet thinks Obama is a sure thing.

Nate doesn't think it's a toss-up, but Scarborough does. So, Nate is telling Scarborough to put his money with his mouth and bet with Nate based on Scarborough's belief that it's 50/50 (hence, no odds given).

If Scarborough really believes it's 50/50, he should have no problem with that bet.
(Assuming he doesn't mind the gambling part.)

If Scarborough thinks it's more in Obama's favour, then he wouldn't take that bet.

I just get exasperated when people argue that Nate is honest and trying to be objective, so, OBVIOUSLY is model is unbiased.

The search for the mythical unicorn of unbiased analysis continues I see. I guess anyone who has an opinion on the outcome of the election should be discounted completely? Dick Morris = Nate Silver because both are biased?

And yes I am being over the top. But everything and everyone is biased in some form - so what? Folks should look at the evidence, the methods, the past history and decide how credible a given source is. I find Nate pretty credible despite his bias, based on his writings and what I know about his model.

Put simply determining true bias and how it impacts results is a complete fools game. I can't look into Nate's mind (or obviously anyones, well except mine) and find out their bias and then know how it impacts their analysis. It is credibility and track record that matters. Sure bring up the various known biases of folks, but then one should move on to discussing their credibility. (IMO)

MCoA, my argument would be that he should say nothing. If he's a modeler, why is he advocating in the first place?

Well, he's first and foremost a writer and blogger. His area of greatest skill is writing about relatively complex quantitative social science problems. So he should be engaged in an ongoing dialogue about election modeling, which should include discussions of and responses to critiques of his own model.

Again, if he's been an advocate for his model even in cases where he should have ackowledged possible or probable weaknesses in his model, then that's bad analysis and he should be critiqued for that. I've seen him get things wrong in the past, he's more than fallible. But I'm having trouble recalling a specific, recent case where Silver advocated uncritically for his model against plausible or convincing arguments.

edit: anyone know if the House does a straight up majority vote if the election goes there after an EV tie, and no electors can be bought, er, persuaded to change their votes?

Think of the leverage!

Jack Carter asked this at the end of the Oct thread. This has come up before, but just in case, the newly elected House (not the current one) meets in January (the 6th I think) and the members vote within their state delegations, which then vote as a unit. You need to win 26 states. Republicans will certainly hold 27-33 states so they should win even if Dems somehow won the House (since Dems don't get credit for having so many seats from NY or IL while Wyoming gets a full vote with one member). Theoretically a couple of people could throw it to a standstill by simply voting blank ballot and keeping Romney to 25, which is no president. Then chaos ensues for two weeks while they keep voting. If the Senate elects Biden--if the Dems hold on--or Ryan--if the Reps take the Senate--then that guy becomes the acting President from inauguration day until the House finally elects someone.

The other realm of squishiness is the month between the actual vote and the Electors casting their vote. Some states are sloppy in who they pick, and it's not inconceivable a warhorse given a plum--an aging conservative Democrat or aging moderate Republican--could decide to act for his country to prevent a tie and switch his vote. That's illegal in some states (but would a court uphold an injunction on the Electoral College?) If that happened in a decisive way, then probably members of each House would submit protests to the vote count. (This only came from the House in 2000 and was dismissed since it didn't have a match in the Senate.) Then, if 1876 is a precedent, we'd need a compromise, as the Senate has the right to count the votes (but does that mean the right to determine which votes count?) but the House has the right to settle unresolved elections (but does that mean the right to determine when an election is unresolved?) This is a minor gap in the constitution when the houses are controlled by different parties. I personally would be surprised if the courts were to go to the mattresses to put injunctions on the Electoral College, so that would mean it was up to Congress to figure out some electoral commission like in 1876.

The silly part of the tie scenario is that it's entirely self-inflicted. For most of the 20th century we had an odd number of electors because the House likes to have an odd number so there is a clean Speaker if no third parties win seats. When we added the DC 3 electoral votes in the early 60s, we then ended up with an even numbered Electoral College. But avoiding ties in the EC is a bigger deal than avoiding ties in the House.

The search for the mythical unicorn of unbiased analysis continues I see. I guess anyone who has an opinion on the outcome of the election should be discounted completely? Dick Morris = Nate Silver because both are biased?

'zop already said he thinks Silver's is the best model. He just has some (entirely reasonable) critiques of Silver.

After Dick Morris predicted a landslide win for Romney, I felt much more relaxed and confident about Obama squeaking out the win.

I'm sure this has been talked about before but the shift of people identifying themselves Republicans to identifying themselves as Independents since the midterms is stark. Romney should do well with Independents--many of them used to call themselves Republicans.

MCoA, my argument would be that he should say nothing. If he's a modeler, why is he advocating in the first place? Its the conflation of model-maker and advocate that makes me suspicious that he can be as clinical as advertised; he's got a lot invested in winning in the market. As you noted, this has nothing to do with a partisanship.

Pride in authorship combined with a desire to market himself and his model. In a platonic ideal sense I guess the perfect modeler would create their model and never show it to anyone or ever talk about it? (Yes over the top again)

I don't mean to be jousting with 'zop and sorry if I am coming across harsh, but I don't think anything Nate has done is disqualifying. One the other side I have no problem with the House of Ras, RCP, or anyone else (well maybe the unskewed guy is a bit crazy). I just wish the discussion was more about the model(s) and less about the people other than a brief disclosure - perhaps - about known affiliations.

And I am not exactly a Nate fanboy and have argued (in threads past) against some of his conclusions.

The other realm of squishiness is the month between the actual vote and the Electors casting their vote. Some states are sloppy in who they pick, and it's not inconceivable a warhorse given a plum--an aging conservative Democrat or aging moderate Republican--could decide to act for his country to prevent a tie and switch his vote. That's illegal in some states (but would a court uphold an injunction on the Electoral College?) If that happened in a decisive way, then probably members of each House would submit protests to the vote count. (This only came from the House in 2000 and was dismissed since it didn't have a match in the Senate.) Then, if 1876 is a precedent, we'd need a compromise, as the Senate has the right to count the votes (but does that mean the right to determine which votes count?) but the House has the right to settle unresolved elections (but does that mean the right to determine when an election is unresolved?) This is a minor gap in the constitution when the houses are controlled by different parties. I personally would be surprised if the courts were to go to the mattresses to put injunctions on the Electoral College, so that would mean it was up to Congress to figure out some electoral commission like in 1876.

This is such a huge ####### hole in our electoral system that it scares me. I think it's exceptionally unlikely that it will affect the 2012 election, but I think it will affect an election eventually.

In a system where so much power is at stake, norms eventually fall. It took decades for the Senate filibuster to become an everyday tool. For decades, norms, not laws,** restricted Senators from filibustering bills they didn't like. Eventually the norms fell and the capacity of the filibuster for exercising power was realized in full.

The electoral system still works by norms in many states. Electors are restricted by norms, not laws, from voting for the president they are supposed to. So far the norms have held, but I don't think it's too cynical to suggest that something that has already happened in the past when great power was at stake could easily happen again. I expect that an elector controversy will occur in my lifetime.

He's still the favorite, but there are a lot of Obama workers on the ground right now.

And while I too agree that a Obama victory in NC is still a long shot, I believe the polling completely misses the effect of NC's new one-stop process where you can register and vote at the same time. It's exactly the kind of setup where having a strong ground game (which Obama most certainly does in the state) would pay higher dividends than normal, and among a voting bloc that would not make it through an LV/RV screeen.

Sigh... Partisan breakdowns are not the be-all, end-all of weighting and sampling simply because they're so fluid... but hey, I'm finally tiring of poll discussions too, so unskew yourself out...

I don't think Sandy will "decide" the election by any stretch - but movement over the past week has been towards Obama, both at the state and national level (except for the always reliable Scotty R!), so I do think it helps, not hurts, Obama because:

1) Romney's best surrogate (except for his convention speech, I guess) has been BFF'ing it with Obama all week... and the media has/can't help but report that either.

2) The media has likewise been playing Romney's response in a GOP debate about FEMA all week - and the campaign has been silent (while Romney - amazingly - hasn't taken ANY questions, even from friendly media in a month). Finally today, the Romney campaign issued a statement that was essentially "blah blah blah". Given the truly dizzying rate at which Romney has flipped every flop except this one, it's almost like karma decided to deliver a big FU , but made sure to time it so that Romney didn't have another debate to pretend the 'severe conservative' didn't exist.

3) Other than the hard right - the middle and the left don't hate government - they may not always trust it, they may sometimes deride it, but ultimately they believe and expect that government should work... and by every account I've seen, it is. Sure - that's not directly impacting voters in Ohio, Florida, Nevada, etc... but the Sandy news is so omnipresent that I can't see it flying over anyone's head. Coupled with #1 and #2 above as part and parcel of the reporting on Sandy, I have to think the 2 or 3 people who are truly undecided have something to think about here.

Update:

Joe Scarborough ?@JoeNBC

@fivethirtyeight Why don't we both agree to donate $1,000 to the Red Cross right now? Americans need our help now.

Well, he's first and foremost a writer and blogger. His area of greatest skill is writing about relatively complex quantitative social science problems. So he should be engaged in an ongoing dialogue about election modeling, which should include discussions of and responses to critiques of his own model.

Agreed. You have to understand that my view of Nate is colored by my background (BITD) in paleoclimatology; in particular, I was at Columbia and got to see guys like James Hansen in action. Hansen et al were good modelers, but when they became hard-core activists, IMO it started to get in the way of the science even though, having known them personally, I'm certain they never ever even dreamed of futzing with the data. When the denialists and others starting lobbing critiques, they got defensive and wrote off all critiques as loony, politically motivated crap; the problem was that 95% of the critiques were loony, politically motivted crap; 5% were substantive. They would've been better off (and we'd be better off) if they'd been able to listen. But human nature is to slip into advocacy, into defensiveness, into not-invented-here. It's also a personality thing; there were guys at GISS who were basically computer nerds who had a certain detachment from what they did; a guy like that would be your ideal election modeler, because they had a congential detachment from their work (amusingly, some of those guys were hard-core conservatives churning out models that said the earth was going to heat up, they appreciated the irony of their station in life). Of course, the computer nerds didn't have Nate's writing/communication skills, and as you noted, that's one of Nate's biggest assets and it's remarkable (and wonderful) how Nate has changed election discourse through his work. Having met both, Nate reminds me a LOT of Gavin Schmidt, but Gavin does a better job of balancing his public advocacy role with his "work", IMO.

Again, if he's been an advocate for his model even in cases where he should have ackowledged possible or probable weaknesses in his model, then that's bad analysis and he should be critiqued for that. I've seen him get things wrong in the past, he's more than fallible. But I'm having trouble recalling a specific, recent case where Silver advocated uncritically for his model against plausible or convincing arguments.

The problem is that there is no conservative equivalent to Nate. I stipulate to what that says about the current state of conservative punditry.

I'm sure this has been talked about before but the shift of people identifying themselves Republicans to identifying themselves as Independents since the midterms is stark. Romney should do well with Independents--many of them used to call themselves Republicans.

Right - as I said last thread, I've been polled several times this cycle and based on how the poll phrased its partisan ID questions, I can guarantee that I was counted as a Democrat, an Independent, and even a Republican (one poll's partisan ID screen asked if I voted in the primary - I did, and I took a GOP ballot). In none of the cases was I lying or otherwise trying to skew anything... I just answered the questions straight up, and being a poll watcher -- I could easily divine which questions were intended to serve which purpose. This is why other weighting -- M/F, ethnicity, etc -- just tend to work better for purposes of weighting.

Heck, in 2003/4 -- Howard Dean was cleaning up with 'liberals' and 'independents'.... why? Because a lot of liberals like me were so pissed at the Democratic party that we began calling ourselves 'independents'.

Joe Scarborough ?@JoeNBC

@fivethirtyeight Why don't we both agree to donate $1,000 to the Red Cross right now? Americans need our help now.

Yeah, that was the obvious answer in reply.

@fivethirtyeight Why don't we both agree to donate $1,000 to the Red Cross right now? Americans need our help now

Ooh, I like Nate, but Scarborough outflanked him but good.

In a system where so much power is at stake,

However, remember that when all this was drawn up, the President, and the federal government, wasn't nearly as powerful as it is now. Who got elected President was important and he was powerful but it wasn't the same system we have now.

I agree it should be amended to plug the gap. Or, better, give Puerto Rico or Guam or whoever 1 EV. Or DC 4. Consitutionally mandate that there be an odd number of EVs.

In the end, though, I think it wise that when the country is quite divided, that it not come down to a purely democratic vote - 50,000,001 to 50,000,000 is not decisive (or believable). I don't really mind at all if the leader to come out of that mess is the one that can best negotiate their way through the various chambers in DC.

@fivethirtyeight Why don't we both agree to donate $1,000 to the Red Cross right now? Americans need our help now

Ooh, I like Nate, but Scarborough outflanked him but good.

Of course Nate could have easily said "Sure, no problem, and now that we've done that, let's still make our bet." How did Nate actually reply?

Silver: @JoeNBC: If you think it's a toss-up, let's bet. If Obama wins, you donate $1,000 to the American Red Cross. If Romney wins, I do. Deal?

I don't get this from Silver's perspective. What could this prove? All it does is make him look like an ass.

It also seems to me like all this debate about models is silly. We're days away (hopefully) from finding out which models were right and which wrong. Moreover, the various uncertainties in the data, in my opinion, aren't properly weighted. Errors propagate. Given the data, if you come to the conclusion that Romney has a 1 in 4 chance, I don't see how you get that more accurate than 1 in 2.5 to 1 in 6 or 7. Very small changes in the original data - if, say, there is a systematic lean to Obama of a few tenths or a percent - could throw the model over. Do you really think the chances there is such a systemic problem in the data is very small? I don't.

Of course, the computer nerds didn't have Nate's writing/communication skills, and as you noted, that's one of Nate's biggest assets and it's remarkable (and wonderful) how Nate has changed election discourse through his work.

Watching this process, which is basically "Mr. Moneyball Goes To Washington," I can't get the Violent Femmes out of my head.

Third verse, same as the first.

In the end, though, I think it wise that when the country is quite divided, that it not come down to a purely democratic vote - 50,000,001 to 50,000,000 is not decisive (or believable). I don't really mind at all if the leader to come out of that mess is the one that can best negotiate their way through the various chambers in DC.

The way I always thought of it is that the election is really just the most accurate poll there is, but it's still imprecise, in terms of measuring the vote of people who intend to vote. Some people screw up their ballot; someone gets into an accident on the way to work; someone's boss calls them in early that day, etc. There's a margin of error in the election itself compared to some platonic ideal of the "true" vote, even among folks that try to vote (to say nothing of intentional non-voters).

At some point (e.g. 2000) an election is within that margin of error; at that point, the election is a democratic tie, and I have no problem with less democratic means being used to break the tie. It's not that the voice is being silenced; its that the voice was inconclusive.

Edit to add: that being said, Florida 2000 was arguably not a tie b/c of the Palm Beach thing, and in that case I think the voice should prevail over the non-democratic means. To be honest, I have no idea what the outcome of a fair Florida recount would've been.

At some point (e.g. 2000) an election is within that margin of error; at that point, the election is a democratic tie, and I have no problem with less democratic means being used to break the tie. It's not that the voice is being silenced; its that the voice was inconclusive.

Exactly. What I meant, better said.

'zop - I probably have somewhat more appreciation for advocates than you do, in situations where the human stakes are very high, but in general I don't disagree with much of anything in your post #82.

Also, I had a planned footnote to post #78 that I forgot to add. I wrote:

For decades, norms, not laws,** restricted Senators from filibustering bills they didn't like. Eventually the norms fell and the capacity of the filibuster for exercising power was realized in full.

That should have been footnoted:

**Obviously it was not just norms, but also the difficulties of cross-party organization in a system where many conservatives were Democrats and many liberals were Republicans. But even after the parliamentary-style re-organization of the parties, it still took quite a while for the norms of the filibuster to weaken and fall.

The way I always thought of it is that the election is really just the most accurate poll there is, but it's still imprecise, in terms of measuring the vote of people who intend to vote. Some people screw up their ballot; someone gets into an accident on the way to work; someone's boss calls them in early that day, etc. There's a margin of error in the election itself compared to some platonic ideal of the "true" vote, even among folks that try to vote (to say nothing of intentional non-voters).

At some point (e.g. 2000) an election is within that margin of error; at that point, the election is a democratic tie, and I have no problem with less democratic means being used to break the tie. It's not that the voice is being silenced; its that the voice was inconclusive.

So let Kevin Costner decide?

You, sir, are an enemy of all that is good in the world

Edit to add: that being said, Florida 2000 was arguably not a tie b/c of the Palm Beach thing, and in that case I think the voice should prevail over the non-democratic means. To be honest, I have no idea what the outcome of a fair Florida recount would've been.

Florida 2000 was a cockup. I thought we should have thrown out the Florida electors and let the House decide. In the end, though Bush didn't win the election, he won the political maneuvering to gain the presidency. That sort of maneuvering has always been used to decide who gains power.

You're treating a social science problem as a math problem, I think. Measuring public opinion is fraught with errors. Measuring public opinion and then weighting this opinion based on likelihood of voting is even harder. There's no reason to expect in public opinion polling, even with extremely large n's, that two measurements of the same quantity will perfectly converge. It's not necessarily a bug in the program if you get divergence. That's the nature of complex human events.

I've been trying to treat it strictly as a math problem in this thread. I'm saying that Silver is trying to aggregate measurements of 51 linearly dependent variables (50 states plus national, where the weighted average of the 50 states gives the national average), but doing so in a way which does not work well if the 50 states are correlated.

An example: suppose every state has a partisan voting index of 0, so that they all equal the national average. The national average is 51% D, with a 1% standard deviation. What are the odds of the Republican winning? The correct answer is 13.6% -- the chance of the national average moving one standard deviation in his favor. If you treat each state as independent, the odds of 25 or more states moving one standard deviation in his favor are 8 in a million. It dramatically underestimates the odds for the underdog.

Silver's model has a little bit more variation than that, but he also doesn't have very many swing states to deal with. Lets figure five states, where the challenger has to pick off three to win. If he's 50% in every state, he has a 22% chance of winning. If that changes to 70%, his odds are still only 27%. But if the states all move in unison, he's 70% likely to pick up all the states and win handily. So the same numbers can make him 70% likely to win or 70% likely to lose, depending on how you treat correlation between states.

If you want to set the odds correctly, you *have* to deal with correlation between the states properly.

In the end, though, I think it wise that when the country is quite divided, that it not come down to a purely democratic vote - 50,000,001 to 50,000,000 is not decisive (or believable). I don't really mind at all if the leader to come out of that mess is the one that can best negotiate their way through the various chambers in DC.

I wasn't talking about an electoral vote tie. Obviously in the case of a true tie, there's no "fair" tie-breaker. The "House votes by delegation" thing isn't the tie-breaker I'd choose (I'd choose "House votes by majority", on thirty seconds of reflection), but it's probably close enough for government work.

In #78, I was talking about the possibility of faithless electors. They could swing a vote against the popular will, and there are no national laws in place to prevent that. (There are some state laws, but tons of electoral votes are bound only by norms, not by laws).

Silver's model has a little bit more variation than that, but he also doesn't have very many swing states to deal with. Lets figure five states, where the challenger has to pick off three to win. If he's 50% in every state, he has a 22% chance of winning. If that changes to 70%, his odds are still only 27%. But if the states all move in unison, he's 70% likely to pick up all the states and win handily. So the same numbers can make him 70% likely to win or 70% likely to lose, depending on how you treat correlation between states.

Put differently: my office had backup generators in the basement; they also had backup servers in Stamford CT. The disaster planners thought - what are the odds of the power AND our backup generators AND our backup servers in Stamford ALL failing simultaneously?

And then the flood came, and now I'm working from home on personal email without access to my documents.

In #78, I was talking about the possibility of faithless electors. They could swing a vote against the popular will, and there are no national laws in place to prevent that. (There are some state laws, but tons of electoral votes are bound only by norms, not by laws).

Yes, but if the vote is close enough that faithless electors can play a role, it is virtually a tie.

If enough faithless electors exist in an election that clearly isn't a tie, then there is probably a good reason. Say Obama wins and in the next month we discover a) he actually is a Muslim plant* and b) he is personally responsbile for Benghazi and has been funnelling weapons and cash to Muslim extremists**. In that case, I would be glad for the electors to switch their vote.


* I do NOT believe this.

** This either.

In #78, I was talking about the possibility of faithless electors. They could swing a vote against the popular will, and there are no national laws in place to prevent that. (There are some state laws, but tons of electoral votes are bound only by norms, not by laws).

Yup -

In 2004, John Edwards actually got an EC vote by a 'faithless elector'... it's just one more reason why I'd prefer to just get rid of the whole thing, but the wasteland states -- who already get tons of over-representation as it is -- would never give up any power.