PointofLaw.com
 Subscribe Subscribe   Find us on Twitter Follow POL on Twitter  
   
 
   

 

 

2012 election: why Nate Silver and FiveThirtyEight.com might be wrong and Romney might be doing better than Silver thinks

| 44 Comments


I'm a fan of Nate Silver, if an envious one. I was a statistical geek with the Baseball Prospectus crew years before Silver was, before giving up baseball sabermetrics for the more lucrative and seemingly practical path of focusing on my legal career. As a kid, I knew the electoral college counts by heart dating back decades, and voraciously consumed board- and computer games simulating presidential elections. The week before I left for law school, I got a job offer at a ridiculous salary for a college graduate to do spreadsheet modeling the place where I worked that summer. There's surely an alternate universe where I passed up law school, stuck with baseball, and had the idea to turn my spreadsheets to doing what Silver does in 2000 or 2004.

Silver recently complained about the degree to which he is criticized for his support for Obama in attacking his model, comparing it to baseball, where "You weren't getting in huge personal fights like, 'Oh, you're a White Sox fan, so you're biased in how you're interpreting the data." I agree that a lot of the criticism of Silver is unfair; the Unskewed Polls website is particularly silly. But Silver is possibly being overconfident in how objective he is.

Silver admittedly massages his data. The massage in 2012 provides bonuses to Obama in the predictions. That could be a coincidence from sound modeling, or it could reflect conscious or unconscious decisions on how to model, for example, how undecided voters will break against an incumbent—where Silver, rightly or wrongly, differs substantially from the conventional wisdom that a president who is polling at 48% is going to end up at 48% because the undecideds will decide to vote for the challenger at the last minute.

There is one particular case where Silver's model ignores facts that favor Republicans, and I think it potentially makes a big difference in his results. Silver points to 2000 as a counterexample to the proposition that polls consistently overweight Democrats or that undecideds break against the incumbent. In 2000, Gore outperformed the last polls by 3.2 points. Silver averages this in, and says that there's no partisan bias in polling or no evidence that undecideds break against the incumbent. But the last polls in 2000 didn't capture the last-minute November surprise of the revelation of Bush's drunk driving charge. (We forget this, because of the much greater drama that immediately followed.)

Silver lets the fact that Gore outperformed his polls by so much influence his model of how to predict undecided predilections for the incumbent and how to calculate house effects, rather than tossing it out as a case where polls didn't capture Election Day sentiments. That's a subjective decision to choose a particular objective rule, not an inherently objective decision. Silver might be right to do so, but reasonable minds can differ. The choice whether to include 2000 as a data point, rather than a sui generis outlier has effects on his model. For example, if we average 1992, 1996, 2004, and 2008, Republicans outperform the last polls by a mean of 1.8 points with a median of 1.0, with 2008 a rare occasion where the polls were on the money. Silver instead starts with polls in 1972 (though there were fewer than ten polls a year prior to 1992), and includes 2000, and gets a Democratic bias of 0.9 points with a median of 0.3—and it's not even clear that Silver includes that 0.3 to 0.9 percent lean in his model instead of treating it as random chance. 1972 is just as arbitrary a starting point as 1992; I have arguments for excluding 2000 from the sample, and I haven't seen Silver defend keeping 2000 in.

Moreover, even if you go back to 1972, you see that polls are breaking not just against Democrats, but against incumbents. The polls get it generally right for Republican incumbents, where the poll bias for Democrats and for incumbents appears to about cancel each other out; the two worst poll performances involve Democrat incumbents, who dropped 7.2 points (1980) and 5.0 points (1996). The only time in the last 40 years that a candidate from an incumbent party outperformed his polls by more than 1 point is 2000—the year of the November surprise. Excluding 2000, we see polls break 1.4 points on average for incumbents; including 2000, we see polls break 0.9 points on average for incumbents—but 0.3 points for Republican incumbents and 3.0 points for Democratic incumbents, though of course, we're only talking three data points in the last 40 years there, so that could just be random chance. If we exclude 2000, and assume homoskedasticity, the difference between the Republican and Democratic results is statistically significant, and suggests that polls are biased for both Democrats and incumbents, and that the conventional wisdom is correct that Obama is in trouble because he's continuing to poll below 50 percent. (Certainly, Obama acted as if he thinks he's behind in the most recent, and last, debate.)

Of course, one could equally arbitrarily go the other way, and say that everything before 2008 is wrong because older polls weren't as sophisticated as modern models. The subjective choice of assumptions in both my arguments above and in Silver's model gives the veneer of objectivity, but can have dramatic effects in the results. There's a reasonable argument against treating 2000 as an outlier, because it introduces subjectivity—if we decide to exclude 10% to 20% of our data points because of exceptional circumstances, why not subjectively exclude still other elections over smaller last-minute issues? That's Silver's most likely counterargument, and he's fairly applied it as a reason to include ludicrously bad polls favoring Romney in the model rather than picking and choosing which polls make the cut. Still, as Silver's model goes, a ludicrously bad Florida state poll doesn't have a big effect in the results, there are dozens of other, better, polls to give a more complete picture; including 2000 in deciding whether to model for whether polls are systematically biased for Democrats or whether undecideds break against the incumbent at the last minute has a much bigger effect if, as I suspect, 2000 is an outlier without predictive value on those two questions.

Silver includes the "house effect" in his models in weighing polls; PPP tends to be overoptimistic about Democrats, Rasmussen about Republicans. But because of his treatment of 2000 as a typical election, Silver's model might be underestimating the house effect of polling in general and thus have its own house effect. A house effect of as little as 0.5% would be enough, even assuming Silver is right in every other way in his model, to turn Silver's 70-30 odds into Obama being barely favored; a house effect of the full 0.9% to 3% I suggest above would flip Silver's results to Romney being favored by at least as much as Obama is now. And there might be yet other judgment calls Silver is making similar to the decision to include 2000 polling in the model that favor Democrats that I haven't noticed.

We'll have a better sense in two weeks whether Silver's model has such a house effect. Silver was successful in 2008, but there were only five states with a spread of less than 2.5% in 2008, so correctly predicting the 45 states where results were pretty clear plus flipping a coin in the true swing states would give someone a 6-in-32 chance of getting at least 49 out of 50 states correct. (Still, give Silver credit for recognizing that Pennsylvania wasn't a swing state.) Silver provides much more analytical rigor than nearly all of the reporting on the subject; 538 is my go-to website for reporting on the polls. Silver could even be entirely right on the issues I discuss above; perhaps I'm guilty of unconscious data mining in favor of Republicans. But we can't yet exclude the null hypothesis that he's lucky, and that he's making mistakes that shade his results toward Democrats and/or toward incumbents.

Update: Welcome Volokh Conspiracy readers. I have one comment; Bernstein writes: "I'm inclined to think that it's a mistake to aggregate say, three polls from three different pollsters showing Romney with 1 to 3 point lead in Florida, but well within the margin of error, and conclude that Romney is leading. Silver, among others, clearly disagrees." I also disagree. The "margin of error" simply tells you if you have at least 95% (or some other threshold) confidence that the lead in the poll is statistically significant. But it's not the case that something that isn't 95% statistically significant is 0% statistically significant; it might just be 60-80% statistically significant. Silver's model accounts for this sort of probabilistic issue and, even more ingeniously, accounts for the fact that any errors are likely to correlate from state to state. That, combined with what is apparently a Monte Carlo simulation, is how Silver produces his bell-curve scatterplots of results, and calculates percentage probabilities of such scenarios as one candidate winning the popular vote, but losing the electoral college.

44 Comments

Nate's the most famous, so i'm curious of your view of Sam Wang. He currently has Obama an even bigger favorite than Nate.

I've also noticed that the polls have been fairly static over that last week - RealClearPolitics & the bloggingcaesar both have Obama ahead. I've followed the bloggingcaesar since the 2006 election & have found his final predictions to be pretty reliable. If all the aggregate polling sites continue to have Obama ahead on the November 5th, would you agree that he's leading?

Wang's results are certainly probative evidence that Silver is right and I am wrong. There is a material chance that the effect I think Silver is ignoring could be an artifact of statistical noise, though, as I've pointed out, Silver in other circumstances doesn't simply adopt a null hypothesis.

Or Wang could be making the same potential error as Silver in his assumptions; I'm not as familiar with Wang's model as Silver's, so I don't know whether Wang includes the sort of adjustments I think there is an argument Silver should be including.

It's also possible that I'm wrong because the polling of 2012 is so much superior to the polling of 15-30 years ago that the sort of 5-to-7-point polling error that overestimated the results of Democratic incumbents in 1980 and 1996 is just out of the question today. (1980 and 1996 also had prominent third-party candidates, which is another source of noise in polling.)

Again, the question is whether there is a systematic bias in the polls that favor Democrats and favors incumbents, and, if so, how big that effect is. I'm claiming that there is a reasonable chance of both possibilities; I'm further claiming that Silver has rejected each of these possibilities because of an anomalous data point in 2000 where polls underestimated a Democrat with a Democrat in the white house, and that Silver is wrong to do so because there were very good reasons for the polls to be wrong in 2000. At the same time, there are so few data points that Silver could be right by accident. So, if Silver is right that the polls are an accurate snapshot, then a consistent 1- or 2-point lead for Obama is evidence that he's leading. If Silver is wrong, then a 1-point RCP lead for Obama might mean the race is a tossup, or even that Romney is ahead, depending on the magnitude of the effect. Even if Silver is wrong, a 3- or 4-point lead for Obama is very likely larger than the effect I'm hypothesizing.

I think that a model that looks at the intersection of the error bands of the various polls might give a more accurate model than averaging, or even weighted averaging.

Then the other factor which indicates a larger break of undecideds to Romney is that his Poll numbers have a higher standard deviation than Obama's (1.99 - 1.40) if you look at Real Clear Politics current poll sample of 9 polls. If Undecided voters are trending to Romney at a higher rate, then his numbers are going to be more volatile than Obama's. Especially when the different pollsters try to pin down the undecideds using a different mix of questions and degree of push to their preferred candidate. Of course that also indicates a degree of softer support for Romney which could also be problematic. But given the differences of voter enthusiasm, I don't think that will be a problem for Romney.

Silver's already done the analysis without the 2000 results and came to the same conclusion. See http://fivethirtyeight.blogs.nytimes.com/2012/07/22/do-presidential-polls-break-toward-challengers/

In addition to doing the analysis for all presidential elections since 1968, he looks specifically at the elections without a true incumbent (which necessarily, eliminates the 2000 election).

There may be legitimate arguments about this analysis, but they don't include the an over-reliance on 2000.

Ted, what does the evidence of Senate, etc. races say about the question of undecideds breaking for the challenger? Dealing with an N=10, any one race can have an outsized impact. This old post by Silver concludes that in close House/Senate/Governor races, there is a very slight tendency of undecideds to break for the challenger. I'd be curious to find out if Silver has updated that research and/or if he is basing any of his Presidential undecideds predictions on this larger data set.

Nate was uncanny in predicting 08. My other source of info is the Bovado betting lines. These are people who can wind up in a shallow grave in the desert if they are wrong about huge sums of house money... the gap has narrowed, no doubt, but right now, Obama is at -210 (bet $210 to win $100) and Romney is at +170 (bet $100 and win $170)....

Most of the polls have been over sampling Democrats by about 6% to 10%. As the election has gotten closer in time, these same polls have been over sampling by about 3% to 6%, to bring them closer to Rasmussen and Gallup sampling methods.

Rasmussen over polls Democrats by 3% this cycle. It was a decision made to split the difference on turnout estimation between 2004 and 2008.

There is a belief by some that turnout will match 2008 or be more favorable to Obama. If that is the case, then a 6% to 8% over polling of Democrats is justified. It is more normal that Democrat/Republican turnout is within about 1%, give or take.

If we look at job approval and other internals in Gallup and Rasmussen last summer, they projected Romney 53% to Obama 47%.

I have heard Rasmussen and Gallup say that it is almost impossible for someone to get a win in the Electoral College, and lose the national vote if the total vote exceeds a 1% margin. As one candidate increases the percentage nationally, it will be reflected in all the states. In other words, a Republican doing well nationally will increase in very blue states such as California and New York. The states will remain blue, but the margin of loss will be less.

So Nate Silver has been treating all the over sampled polls as equivalent.

If a baseball player has a .320 batting average, that is a real measure. Same thing with an ERA. The number is the number.

But using phoney baloney polls is like Nate saying, "The BA is .280, but it should be adjusted up to .320." So read Nate Silver as you like.

I don't think you can combine margins of error. In other words, they aren't additive. If a poll has a margin of error of 3%, you can't combine it with another poll with a ME of 2% and say the combined poll has an ME of 2.5%.

Anyway, what is the margin of error of a poll which automatically assumes Democrats will out vote Republicans by 10%?

Zack: Look again at that page. It supports my argument. Silver's data shows that October polls overestimate the sitting incumbent's lead by 3.2 points. His model assumes that that is random noise and that the expected value of the overestimate is zero. The null hypothesis may be right. There's a substantial chance that it is not. It's a judgment call whether to risk overfitting and include this variable in the model, but that judgment call is inherently subjective. Silver includes other things in the model that have similarly weak evidence supporting it. Silver may be right on each of these judgment calls, but his analysis of why he doesn't think there's a systematic partisan bias in the polls -- counting a single 1972 poll overestimating Nixon's share of the vote with the same value as 11 1996 polls overestimating Clinton's, and including 2000 in his average -- isn't persuasive.

As Eric points out, it may be random noise: small sample sizes and all that, and maybe we should think that presidential elections aren't different than down-ticket elections (though we have plenty of evidence elsewhere that they are--witness the well-known coattail effect and the problems incumbents face in off-year elections). But Silver doesn't let small sample sizes cause him to reject the null hypothesis in other aspects of his model. Even if that 3.3-point figure is an overestimate of the Platonically correct number that should be plugged into the equation (and I would guess that it is), even if the effect is only half that size, that's enough to flip Silver's predictions in an election this close.

For the record, I don't read the NYT so I'm not sure how Silver's model is actually constructed. However, if I was building an election prediction model and I had candidate A winning with a 70% chance but then a poll comes out and has Candidate B winning by 3 overall and by 18 with independents, I would take a serious step back and wonder what was going on. If it was Pelosi's district, the number would be no big deal as Dems swamp any tide an opponent could build with independents. However, in an election that is fairly close to being evenly split, it would freak me out.

Good statistical practices include looking at data/results and asking does this make sense. If not why not or maybe it does but is just counter intuitive. Either way a pause is necessary and discrepancy needs to be reconciled. I think people who rely on the NYT are the ones on Intrade bidding up Obama.

One of the things that is confounding is how people such as Silver blithely seem to gloss over how polling (especially in Ohio) is using turnout and sampling models that are even more generous to the Democrats of 2012 than in 2008, which, as we know, was a very good year for them.

I am not implying that pollsters are deliberately oversampling to get a certain effect or influence numbers and conventional wisdoms, but it is interesting.

For the record, I don't read the NYT so I'm not sure how Silver's model is actually constructed. However, if I was building an election prediction model and I had candidate A winning with a 70% chance but then a poll comes out and has Candidate B winning by 3 overall and by 18 with independents, I would take a serious step back and wonder what was going on. If it was Pelosi's district, the number would be no big deal as Dems swamp any tide an opponent could build with independents. However, in an election that is fairly close to being evenly split, it would freak me out.

Good statistical practices include looking at data/results and asking does this make sense. If not why not or maybe it does but is just counter intuitive. Either way a pause is necessary and discrepancy needs to be reconciled. I think people who rely on the NYT are the ones on Intrade bidding up Obama.

Thank you Sir
I have been a fan of Silvers predictions for a while now and though I believe he is probably the best out there, he is of course reliant on the data he receives and no model he creates will ever be perfect.
Always good to hear an alternative view.(A Coherent Alternative View that is)

The real problem is 538 ignores the party ID problems. Gallup is seeing an R+2 electorate and so is Ras, which would mean Romney is winning by about 5.

And we now know from actual ballot counts that the Democrats have already lost 250K net votes vs. 2008 in Ohio -- nearly Obama's entire 260K 2008 margin. Silver has totally ignored this and instead is relying on polls that have twice as many people claiming to vote than have actually voted, despite poll response rates as low as 3%.

On average, Romney is leading indies by about 10 points. It is not remotely plausible that Romney could win indies by 10 in an election with the highest GOP enthusiasm ever measured and still lose the PV or EC. Just. Can't. Happen.

Keep in mind also that pollsters are really only judged by their final poll -- in October 2010 Nate was still only seeing a 50/50 chance the GOP took the House.

Even on his one salutary data point, Silver had a huge advantage in 2008 because Obama was secretly feeding him all the campaign's internal poll data. His analysis isn't terribly awful, but I wouldn't be surprised if the much-less-serious Unskewed average for October is closer to the actual vote results than Nate's October average.

Nate already did the correct analysis - you have to use elections with actual *Incumbents* and not 'Party in Power.' This eliminates 2000, 2008, 1988. You get 8 elections.

According to Nate's link above, the net gain in the polls from October is +4.8% for the Challenger, only +1.6% for the POTUS.

That's a 75% break to the Challenger!

Using Nov polls, you get 1.9 for Challenger and 1.1 for POTUS.
So, if race is 50-47 next week, you're likely to see a 51.9-48.1% final in favor of Romney.

I think Nate is applying the same general statistical model to analyze two fundamentally different problems.

If over the course of 160+ games in a baseball season, a player gets hits 28% of the time, the batting average is .280. This presumably means the player has faced very good pitching and very poor pitching, and everything in between. That means the player should bat significantly less well in the playoffs where better pitching can be expected. But we can't predict that based on the season batting average. In other words performance may be much better or worse over a short number of games. This applies to all baseball statistics. And it applies to all statistics in sports. Still, these are real numbers based on a finite number of measurable events.

As they say, that is why they play the games.

Polling samples .0001% of the relevant population, give or take. It makes all kinds of assumptions about who is a "likely voter." It may or may not over sample certain populations. This can be either intentional or not. Then we are trying to predict behavior based on a tiny fraction of the population and what that population said. I have read that Gallup and Rasmussen have a "hang up" rate of near 90%. That means nine phone calls are necessary to get one which will take part in the poll.

I get called on polls and never answer them. And I vote.

I remember 1980. Everyone said Carter would blow Reagan out of the water. Reagan had been trashed all summer. The polls indicated Carter would win, clear up until the day of the election. Of course, there wasn't as much information in those days.

One thing about the Bush Gore election. There wasn't really an incumbent in that election.

Nate Silver is dead wrong about Obama winning this election. All his fan boys will be looking for his scalp when they see that Mr. Silver failed to take into account one basic truth. America NEVER votes to re-elect a failed president. Obama is not popular anymore. He will lose in a landslide.

As I understand it, Silver attempts to work Bayesian analysis into his estimates. So the choice of his priors have a large impact. It is also my understanding that he rejects 2010 as a realistic prior.

I looked closely at the year to year voting patterns from 2004 to 2008. It was a terrible turnout year for Republicans, and it followed another terrible turnout year - 2006. By terrible, I mean that Republican turnout was down - a lot. That's because Republicans, unlike democrats, have grown to want to reject deep corruption in their party. So a lot of Republicans and conservatives (I tend to keep the two separate) sat on their hands.

That ended in 2010, when the desire to try to fix the country changed a lot of their outlooks. The feeling was that they could become empowered. That feeling hasn't gone away - there is still the feeling that this is a do or die moment for the country. Dems are more disillusioned. It may not depress the true blues, but it will depress those who aren't.

If I'm wrong about how he works this in - my apologies up front. I'm back to work and don't have time to do the research... I'm having to go off memory.

The falling response rates for polling, which were in the 70%+ area in the '80s, may account for the imbalanced party identification. Pew Research reports the average rate (most polling firms do not report their individual rate publicly, considering it proprietary info) had fallen to 37% by 1997 and 9% now.

If the more than half of Americans who have moved to refusing to participate are disproportionately Republican, conservative, or Tea Party supporters, it would naturally leave a more Democratic sample to be polled.

This would account for the skewed Democratic samples without any conspiracy or even fault on anyone's part - how could you determine the make-up of the refusing? They won't answer any questions.

I remember 2000 very clearly. The day the DUI story broke which was 5 days before the election, Bush had a 2-3 point lead and major momentum. I remember Zogby[who was a good pollster at the time] basically predict Bush would win by 5-6 points if his momentum continued. Then BANG came the DUI story. I remember focus group members saying they were going to switch their vote. Then Zogby comes out the day before the election saying it was a dead heat race and Gore had the momentum.

That DUI story was worth 4-7 points. Which totally screws with the data studying the quasi incumbent theory and D-R theories.

Its common sense folks. Thats all
stock market doubles, Bin Laden dead, wars ending
Obama 51.1% Romney 47.5% electoral 303-235
Email me on Nov 6 to tell me I was right...

TedF: To be accurate, your post title should be "why the polls and wagering sites might be wrong and Romney might be doing better than gamblers think".

In addition to Silver and Wang, Realclearpolitics "no toss up" page has had Obama ahead every single day. Per RCP, Romney's high point came on the 19th when he rose to 277 EV's, but since then he's lost CO & NH and Obama now has 290 EV's for their "no toss up" prediction.

In addition to intrade, iowa futures market had Obama at a 64.5 price last night.

Given that Wang has Obama at 97%, Silver at 74.4%, Intrade at $6.37, I cannot believe that the flaw is with Silver's model, and the only argument against Silver's conclusions would have to be polling bias pro-Dem or pro-Dem incumbent, and plenty of fools waiting to be parted from their money.

I am in full agreement that polling has improved since Literary Digest's 1936 poll or Gallup, Roper, and Crossley in 1948, but as long as 4 meta-polling sites & 2 wagering sites - all with good track records - have Obama leading, will continue to believe that he is leading.

As someone who has been built predictive models in the financial industry for two decades, I can convincingly say that too much stock is placed in models. Let's not forget that predictive models is partly what caused the financial crisis of 2008 as banks relied on sophisticated models to project real estate prices and borrower behavior, which fell apart when the meaningful variables in predicting performance changed fundamentally. Predictive modeling is inherently difficult even with an "objective" model. And I must emphasize that no model is completely objective. In deciding what variables are significant, there IS some subjectivity. Mine the data one way and you come out with one set of significant variables but mine it another way and you get a different set of variables. Change the weight of one variable or variables and you change the outcome.

With respect to Mr. Silver, the common refrain is "well, Silver nailed the 08 election". This is fundamentally flawed logic as it is the classical logical fallacy of appeal to tradition. Being right before doesn't compel the conclusion that one is right now. And you have to put this into the context of 2008 polling. How many polls in 08 in battleground states were as close as the polls we see now? Go back and check the data, not many. It becomes significantly easier to "nail" a projection if the underlying data is denominated by results which show gaps outside the margin of error.

We have a very different scenario now. We have polls that are much closer in many of these battleground states, which introduces much greater error into the equation. This is why I'm reluctant to give too much weight to Silver's analysis. It's a useful data point for sure, but in a close race, a model can be very wrong. I've seen many models in my time completely fail, even though the models were sound from a statistical perspective.

Maybe Silver is correct. We'll find out soon enough, but I suspect his performance this time around will be nowhere remotely as accurate to what it was in '08.

"America NEVER votes to re-elect a failed president."

Bush.

Rasmussen wrote on his site a few weeks ago that all the wild swings in the polls reported by media are largely untrue. The electorate is essentially stable with very mild divergences in one direction or the other. But the media has an interest in keeping people watching. So they create the false narrative that the election is something like the Super Bowl. Romney and the Giants have just scored a touchdown with 50 seconds left. Can Obama and Brady push it down the field to get the winning touchdown in Ohio? Nate Silver has bought into this narrative and accepting it as if it were true. But elections aren't anything like that.

I just went to Nate's site on the NYT. He uses memes created by the Democrats and the media and regurgitates them like they are proven facts. "The auto bailout gives Obama an insurmountable edge in Ohio"; "The safety net in Michigan makes Obama popular." Really? This is what irritates me about Silver and his whole scheme.

The election is a dichotomy. Either Obama will be elected. Or Romney will be elected. There is no other possibility. So how can Silver say there was an 80% chance Obama would win in September, then a 60% chance a few weeks ago, and now it is 70%? It is either going to happen, or it isn't.

It is like a .320 batter coming up in the world series. He is either going to get on base, or he isn't. He doesn't have a 32% chance of getting on base. He is either going to get on base or he isn't. If he comes to bat 1000 times, he will get on base 32% of the time. But that number can't tell us what will happen in each event.

I would imagine the electorate hasn't changed much since last spring. Looking at Gallup and Rasmussen then, Obama should get 47% of the vote. It remained that way up until July. Then there was a flood of polls which over sampled Democrats.

This year Rasmussen is over polling Democrats by 3%. Gallup by 1%. Both Gallup and Rasmussen say that presidential job approval tends to mirror the final vote. Both are around 48% for Obama.

The Boulder Colorado model, looking strictly at the economic condition of the individual states predicts Romney with 330+ EC votes. It predicts a 53%-47% split.

The unskewed poll site which adjusts the polls to model a typical election with 1% more Democrats than Republicans voting forecasts Romney 52%, Obama 47% with Romney getting 330 + EC.

So it is probably more about getting the base out, rather than convincing people to change sides. The Democrats understood this. They have been playing hard for the base since last winter. It hasn't been about big ideas. It has been, "free birth control, free abortions, they are going to put you back into chains, lots of smart people have been cheated, you didn't build that, the rich need to do their fair share, and Romney is an evil, murdering, hateful, nasty tax cheat felon who will end abortions and take your birth control pills away."

This will probably appeal to about 47% of the population. Oh wait a minute. Romney already said that.

After I confirmed that Palin had been selected the vice presidential nominee in 2008, I checked Intrade to see how she was doing there. She was trading at 2% that morning. Prediction markets, aside from being ludicrously thinly traded, are merely reflecting back what Wang and Silver and the media are telling them -- and despite that, Romney is doing substantially better on Intrade than Wang and Silver would predict.

When one person can no longer move Intrade percentage points with a few thousand dollars, and when one can invest enough money in Intrade to make it profitable to devise one's own sophisticated trading model instead of relying on publicly-available models, then I'll have more faith in Intrade, but I don't see any evidence that anyone sophisticated is betting real money there and that it is doing anything more than being a muddy mirror of what Silver is doing.

And again, even with all my criticism, I acknowledge there a not insubstantial probability that Silver is close to the mark. My point is simply that he's selling his model as more objective than it is.

"Gallup's LV model is 78 percent white. Was 75 in 2008. So despite more minorities in pop, Gallup projects much whiter vote this year. True?"

This is why Gallup and Rasmussen is junk. So to all you Unskewers who are drooling over Gallup and Ras saying this is a +1 R electorate you better eat crow.

There's a 50% chance Nate Silver will be hailed as a genius on Nov 7. There is a 50% chance he will be reviled as a putz.

Uh, Bush was not a "failed President".

In the economy , the stock market was humming, UE under 6%, big recovery from 9/11, big win in Afghanistan and while Iraq was messy, it was not yet the nasty goo that required the Surge. Also lots of "compassionate conservative" bi-partisan crap with the Dems: No Child LB and Medicare, Pt. B

Try to remember actual history instead of your Lib-tard bumper sticker, talking points.

Concerning Intrade - Have you looked at Obama's chart? I don't know any serious investor who would buy into a repetitive head and shoulders pattern like that. Right now it looks like a falling knife that just had a dead cat bounce. I wonder what loser bought at the high when it was at 79.5?

"The "margin of error" simply tells you if you have at least 95% (or some other threshold) confidence that the lead in the poll is statistically significant."

Maybe I'm misremembering my Stat 101 class, but I'm pretty sure this is wrong. The margin or error gives us the range of the confidence interval, so if a candidate holds a 5% lead with a margin of error of +/- 3 points, it means we have 95% confidence that the actual lead is somewhere between 2 and 8 points. So if a candidate is shown as leading in a poll, but the lead is less than the margin of error, then it's a bit of a stretch to claim the lead is "statistically significant."

That's one of the lamest comments I've read anywhere. "Stock market doubles"? "Wars ending"? What the "F" does that even matter??

Take a look at the polls, dude. Obama is NOT gaining any ground. He is, in fact, LOSING ground -- in CA, NJ, OR, etc. Not that Romney will win those particular States. But just the fact that Obama's LOSING POPULARITY is a sure-sign of where all of this is headed.

So feel free to email ME on Election Day after your hero loses BADLY.

Bush was NEVER elected--he was selected!

I'm no expert in polling, but today's announcement that Obama and Romney are suddenly tied in Ohio is suspect. There are what may only be rumors that this poll is based on Romney's standing prior to the final debate. If that is true than some people are trying to muddy things, and with a purpose no doubt.

I fear they will introduce Romney as having come even with Obama while a lot of voter fraud is happening behind the scenes or perhaps in anticipation the president would be tied up with Hurricane Sandy, and would try to discourage Obama voters. "He's lost momentum so don't even try to vote".

I never trust the GOP or put anything past them.

What's been bothering me about 538 vs. RCP is that these two measures of the polls have been diverging. I wouldn't mind it so much if one had a constant bias with respect to the other, but for a while the RCP average shows the race converging while 538 shows the race diverging.

I suspect what's going on is that 538's big, hairy model at this point boils down to the fact that Ohio is the critical swing state (according to how the model works out). Indeed, eyeballing the state polls, it seems to me that 588's winning percentage has lately been tracking the Ohio polls, not the national polls. And the Ohio polls have not been moving with the national trend.

Simply not enough data in the formula to make Mr. Silver's blog anything more than pages full of blisteringly unusable analysis. In 200 years, 50 more presidential election cycles and roughly 150 years after Mr. Silver's funeral (I wish him a long happy life), his formula may become mildly more useful.

As of 10/31/12, Nate appears to be backtracking. He is relying more on state polls than national polls. If the national polls are correct, he will be a sad person a day after the election. So he has written this morning.

I won't weigh in on probability modeling I don't know but consider these points:

A representative sample of Romney supporters are not wagering on him in Vegas. Vegas doesn't set odds; oddsmakers limit risk by truing up the payouts so the house rarely loses, determining odds. Since most of the money is going to Obama his payout is much lower. To balance the house risk, a higher payout tempts Romney betters. The payout has remained higher because there aren't enough Romney bettors.

The representative sample of Romney supporters don't participate on Intrade, don't know how to access it, and haven't heard of it except as a vague reference by "the liberal press". This is easy enough to check; the stock that rises has buyers.

Romney supporters do not participate widely or well in extended polling interviews. How many outreach contacts per 100 responses for each candidate? That "unsuccessful contact and/or completion" information is rarely in the cross tabs.

Perhaps one test would be install Romney and Obama themed slot machines at the casinos and see which pulls in more nickels.

How accurate are the predictive models for actual turnout itself for the candidates, and how does that matches polling predictions? We hear about registered voters and likely voters, but it is rare to hear how accurately the actual turnout number has been modeled.

Admittedly, none of this has spreadsheet and hard statistical validation. Are Silver's samples unable to "see" Romney supporters sufficiently in their mathematical modelling because they are off-radar?

The last President who won a second term with a smaller percentage of the vote than he won his first term was James Madison.

FDR won his third and fourth terms by decreasing margins.

All of these elections were during wartime, the War of 1812 and WWII, receptively.

Basically, from economic models to historical results, to anecdotal reports, all indications except for some polls point to a President Romney.

What about the counter argument to this? perhaps Nate/the polls are underestimating the true standing of Obama? We still have some palls that don't call cellphones or that don't offer the option of taking the poll in Spanish for Hispanic voters?

Data on NC early voting (interested on thoughts on what it means):

Total 2012 (through 11/2): 2,511,425
Total 2008 (through Friday before election): 2,391,747
Change: 119,678
Change in African Americans: 47,414
Change in Whites: 46,020
Change in "Undesignated": 22,482
Change in Asian/American Indian/Other/Two or more: 12,964

Rs who voted on election day in 2008 voting early in 2012: 110,272
Ds who voted on election day in 2008 voting early in 2012: 77,091
Unaffiliated who voted on election day in 2008 voting early in 2012: 51,965

New D voters (didn't vote in 2008): 362,470
New R voters (didn't vote in 2008): 226,073
New Una voters (didn't vote in 2008): 203,058

New African American voters (didn't vote in 2008): 243,645
New White voters (didn't vote in 2008): 482,535 (82,554 are 18-24)
All other race categories: 69,104

What's it all mean? One question I have is what happened to the roughly 1 million people who voted early in 2008 that didn't vote early in 2012? Will they show up on election day? If so, then you'll have probably 10% more voters in NC in 2012 than 2008 (no drop off in turnout there).

There are other states where early voting in 2012 has already exceeded 2008 (Montana (!), Iowa). Does it mean people are more engaged earlier?

I think it would be a mistake to assume turnout is down from 2008 based on early voting stats. But I'm interested in what others think of these stats.

One factor I haven't heard mentioned anywhere is the unique nature of the 2008 POTUS election and how that might skew any reliable data from that election regarding how the eloctorate was made up then and how it's make-up can be used to predict future elections.

First of all, there was no true incumbant. Secondly, the financial meltdown was in full swing. Thirdly, there was a sacrificial Republican candidate who most Republicans voters themselves were not excited about. And finally, you had the first ever African-American presidential candidate win his party's nomination. The other factors may or may not happen individually in other elections, but the last factor, by definition, can never happen again.

Concerning the "first ever African-American candidate" factor, there were undoubtedly, for better or worse, many people who fell into one of the following groups:

A: I don't really prefer either candidate over the other and so could go either way-- so if I couldn't care less which one wins, why not be part of history and vote for Obama?

B: I don't really pay attention to politics and I never vote. Even if I wanted to, I usually find something better to do than wait on line at the polling place. But I'm gonna vote this time because this is a historic election.

From group A, one could assume that only 50% of these voters will vote for Obama this election, while 100% voted for Obama in 2008.

From group B, one could assume... oh wait a minute, they won't be voting at all.

It will be interesting to see how many of these "historic" votes either flip or never materialize in 2012-- particularly in the key swing states.

Great critique of the process. But in the end, Nate was spot on. He really does deserve applause.

So what's your email address there Coperninewtonstein?

I miss intelligent Republicans that I could have a conversation with.

Now it's generalizations, hyperbole, and sticking feet in mouths.

Too bad.

It's interesting hearing the arguments and counter arguments that were made before the election. Much of that debate on on polls over sampling dems and models over estimating dem turnout. Silver was right on with regard to who would win each state, however his model underestimated Obama's totals in swing states by 1%. This difference is statistically significant. So the question is why the polls and models underestimated dem turnout.

Once submitted, the comment will first be reviewed by our editors and is not guaranteed to be published. Point of Law editors reserve the right to edit, delete, move, or mark as spam any and all comments. They also have the right to block access to any one or group from commenting or from the entire blog. A comment which does not add to the conversation, runs of on an inappropriate tangent, or kills the conversation may be edited, moved, or deleted.

The views and opinions of those providing comments are those of the author of the comment alone, and even if allowed onto the site do not reflect the opinions of Point of Law bloggers or the Manhattan Institute for Policy Research or any employee thereof. Comments submitted to Point of Law are the sole responsibility of their authors, and the author will take full responsibility for the comment, including any asserted liability for defamation or any other cause of action, and neither the Manhattan Institute nor its insurance carriers will assume responsibility for the comment merely because the Institute has provided the forum for its posting.

Related Entries:

 

 


Isaac Gorodetski
Project Manager,
Center for Legal Policy at the
Manhattan Institute
igorodetski@manhattan-institute.org

Katherine Lazarski
Press Officer,
Manhattan Institute
klazarski@manhattan-institute.org

 

Published by the Manhattan Institute

The Manhattan Insitute's Center for Legal Policy.