Princeton Election Consortium

Innovations in democracy since 2004

Sep 18: Biden 352 EV (D+5.3% from toss-up), Senate 52 D, 48 R (D+3.2%), House control D+2.0%
Moneyball states: President NV AZ PA, Senate MT KS AK, Legislatures KS TX NC

PEC in the news

November 4th, 2012, 10:00am by Sam Wang

Here’s some Princeton Election Consortium news (and about poll aggregation generally):

Upcoming: Monday KCRW 2:00pm ET with Mark McKinnon, Sasha Issenberg, and Ron Brownstein.Tuesday WNYC 12:30pm ET with Leonard Lopate.

Recent radio: KPCC Los Angeles, Colorado Public Radio, NPR’s Science Friday with Flora Lichtman and Nate Silver.

Articles in Deadspin and Wired about the aggregation-hating hordes (“Nerds Under Attack!,” October 29th). I’m glad aggregation is important enough to hate. Deadspin calls PEC “criminally underrated.” Cool.

And then there’s The Onion.

Update: I’ll probably be on CNN CNNI today at 4:20pm. (CNNI is not CNN!) Update 2: …and here’s the video feed.

Tags: 2012 Election · Site News

160 Comments so far ↓

  • steve in colorado

    Heard you on Colorado Matters on CPR. Very well done. Better than the Science Friday bit.

  • amk

    Hi Sam. I have been following your predictions for some time now.

    Just a few hours back, you had PBO’s chances at 98% and EV at 323.

    Now you have increased his chances to 98.2% but decreased the EV to 319.

    What gives ?

    • Garble62

      Just a guess, but I’d say it’s because we’re one day closer to the election.

    • Kenny Johnson

      You only need 270 to win. So if, for example Obama’s chances of capturing 270 grow, but also, his chance of losing Florida grow, his EV will go down, but his chances of winning the election increase.

    • Ms. Jay Sheckley

      This thing is automated, no? EV jumps are attributable to changes in states. But on the whole it’s operating according to plan: very stable, just like this election.

  • Paul

    in the interests of demonstrating your site’s accuracy and reliability, you might want to change your clock settings to Standard Time. it’s not 10 a.m. in New Jersey yet.

  • janjanjan

    I must admit that I’m one of your new secret compulsive site checkers. I love the absolute sense of reassurance it gives me. But, I do like the commentary I get at 538, even though Nate’s dispassion sometimes leaves me anxious. So, any and all commentary posts you add are greatly appreciated. Thanks for the site and thanks for getting back up so quickly to feed my addiction.

  • wheelers cat

    I liked that Deadspin article a lot. The same CLT argument is being played out in the twitterverse between Cost, Trende and Linzer.
    This from Dr. Linzer sums (lol) it up.

    “@JayCostTWS Methodological differences could be many things! Bottom line, since we don’t know truth, make minimal assumptions, just average.”

    i should make a chirp story out of it.

  • Mark Sillman

    In defense of the MSM take (” it’s too close to call!”) They are (by accident) doing their civic duty. If they instead reported “Obama is a 97% certain winner”, then people would be discouraged from voting (which would incidentally help Mitt Romney). For this reason other countries actively discourage polling within a few days of the election.

    Dr Wang: can I suggest that you emphasize that the House of Representatives is still uncertain? And may matter as much as the Presidential election? Don’t accidentally tell people they dont need to bother to vote!

    Maybe more people should also hear that your results show how badly Republicans have gerrymandered the House – si that they can still wi the House even while being outvoted 48%/52% nationally, if I have it right. That should be headline news!

    • CN

      Yes, Sam, could you do another round up of house polls and analysis? I think a few weeks ago (maybe before 1st debate?) you said there was ~25% chance of Dems retaking the house? I assume that’s come down some but could you do an update analysis, looking at how things have changed and look going into election day?

  • mediaglyphic

    Love the onion piece. I wonder if its time to re-read incredible popular delusions and the madness of crowds

  • Joey Scarborough

    Uh-oh for you, Dr Wang! Michael Barone’s gut says you’re wrong, nerd! He has the most powerful gut in all of the professional conservative pundit universe! Sure Dick Morris’ gut is bigger but Barone’s is like worm hole to the future.

    I hope you are busy preparing your mea culpa, Dr Wang. The guts have spoken! Read it and weep:

    • Reason


    • RDT

      And anything written on a site called “Hot Air” has got to be credible :)

    • DJG

      I wouldn’t worry about that, Mr. Scarborough. Professor Wang has a jar full of creepy crawlies ready to swallow on camera if need be. How is that for a mea culpa?

    • Ms. Jay Sheckley

      Joey- Hmmm…
      Alleged conservative gut vs poll aggregation: vs , no joke!
      Blowhard vs math.
      Tough choice.

      Say, have you read the algorithms? I didn’t, but I was shown this by someone who totally gets the math. Been following this for months. and learning, I think.

      Dr Wang is telling us what the polls say, clearly and simply. And he’s telling us probabilities based on that. Whatever happens, he has nothing to apologize for.

      But if, compared to all the pundits, he has again built a better mousetrap, the world will race zambonis to his door

  • Paul G

    I have a question concerning the PEC model and its differences with 538 (which I’m sure Sam has addressed many times, but maybe someone could comment further or point me to a link.)

    Nate seems to be arguing the following: there is plenty of room for pollster error aside from random variation, due to things like low response rate and assumptions about electorate demographics. Since most pollsters use similar methodologies and assumptions, such errors would be correlated across the different polls and would therefore also affect the aggregates. So there is a non-trivial chance the polls could be biased.

    Sam on the other hand if I understand correctly claims that the polling aggregates tend to be accurate, to the point that even if the aggregate margin is small, such as 1.5-2 points, there is a only a very small chance the other candidate will win. I gather this confidence is based on past performance of polls over the last couple of cycles? Is there some part of the site which explains the quantitative argument behind this? And do changes in polling operations, electorate demographics, phone use, etc, affect this confidence at all?

    • E L

      One difference between Sam and Nate is that Sam makes his assumptions about state polling accuracy based on a shorter time line of state polls. As I recall, Sam computes his assumptions of polling accuracy on 2000-2008. Nate goes all the way back to 1976 when few state polls were in existence. Sam argues and has the numbers to show that the more recent state polls are more accurate than older state polls. In other words, polling has become better. Then, Nate factors in national polling. Sam has shown that state polling is more accurate recently. Also, Nate factors into his analysis all sorts of non-polling information such as employment and the stock market. Sam has shown that using only state polling gives more accurate results. In the end, both conclude that Obama will win. Nate has a built in hedge… just in case. Over to you, real Nerds.

    • Paul G

      Yeah, thanks for the timeline info. I know about the differences with the other data Nate packs into his model, but I was really asking more just about the confidence in state polling issue.

    • Some Body

      Essentially (and this is also why the timeline is relevant), they disagree on the probability of systematic error (more precisely, systematic bias) in the state polls. Wang essentially assumes it is zero (which worked well for his model in 2004 and 2008), while Silver notes there was systematic bias in the years before and concludes that there is indeed some tangible chance of such systematic bias.

      If, with the increase in the amount of polls and pollsters, we have indeed seen a sharp and lasting increase in the quality of state polling in the recent decade or so, this would be an argument in favour of Wang’s approach. However, the essentially zero systematic bias in 2004 and 2008 may have been just accidental, which would then support Silvers approach.

      The truth may also be in the middle: it may be that systematic bias is more likely to appear in some election years and not in others (e.g. when there is a strong third party candidate on the ballot, or when the predicted result is relatively lopsided). In that case there would be ways to compute the probability of systematic bias for an election cycle based on how it is shaping (but, of course, the number of data points is so small that this would amount to more speculation than calculation at this point).

    • Meg

      I think-and someone please correct me if I’m wrong-that Sam has also said that if there are systemic errors that they would be less significant in the state polls than in the national ones.

      Now, if there is some systemic error in just the state polls, that would be a different kettle of fish. My sense is that Nate thinks that is a remote possibility, while Dr.Wang thinks the chances are essentially 0. Again, please correct me if I’m wrong.

      Also, a kind person here directed me to a very helpful article at Votomatic about whether the polls were to be trusted. Apologies for not remembering whom to credit for the share.

    • BrianTH

      I don’t think Sam is assuming strictly no chance of any systematic bias of any magnitude. However, the relevant question is what chance is there of systematic bias in Romney’s favor large enough to overcome his current polling deficit. The polling outcomes since 2000 (and in fact since 1996) don’t support there being a non-negligible chance of a systematic bias in his favor of that magnitude.

    • Ms. Jay Sheckley

      Unlike 538, PEC wont adding 16% of mere “uh-oh what-if” for Mitt.
      Sam & colleagues, would it be fair to say that, due to Sam’s experience as a neurology researcher in these matters of excitement, more than anyone he works to reduce added human error. PE’s special accuracy comes from the unusual precision of its set-up combined with the most automated, hand-off follow through.
      Now correct me, please, so I’ll know more.

    • Some Body

      @Ms. Jay – I think Sam Wang’s next post here is excellent in clarifying how the models (these two and others) differ.

      Essentially, it does seem to be also a matter of temperament, or rather of purpose. Wang and Silver disagree on the probability of systematic bias in state polls. Wang prefers to err on the side of precision and is willing to take the extra risk, while Silver prefers to err on the side of safety.

      This also reflects, I think, a difference in understanding the function of such a forecasting site. Wang looks at this from a more scientific perspective (offering a sharp measurement tool, and a precisely defined hypothesis for testing), while Silver is more mindful of his role in informing the broader public, which implies putting greater stress on uncertainty.

  • Jreal

    What do you have to say about university of Colorado model? They’ve been predicting since 1980 and they predict a Romney win? I’ll appreciate any input.

    • bks

      It was first published in August 2012, Jreal.


    • badni

      Not true. This is their model’s first prediction.

      They built it this year in a way that it would have been right on those prior elections. I, too have a model I just built this week that correctly predicts the last 10 Powerball winners. For a very fair price I’m happy to tell you its predicted outcome for next week’s Powerball.

    • Matt McIrvin

      Wow, there sure are lots of people asking about the CU model in the past few days.

    • RDT

      Note that if you’re building a model that “predicted” previous elections, you can probably build it to “predict” the outcome you want for the current one.

    • gryffydd


      The Colorado model is not a poll aggregating model but relies on state/national economic statistics to predict vote %s and then uses those to count EVs. The intro describes their data as :

      “In contrast to these other Electoral College models, our model includes measures of change in real per capita income, as well as national and state unemployment figures. Account-
      ing for both changes in personal income and unemployment provides amore robust approximation of state economicwell-
      being and, thus, serves to model the impact of retrospective evaluations of the incumbent party’s stewardship of the economy”

      It predicts the following vote %s (in the brackets) for Obama:

      Ohio (45.00), Florida (45.75), Virginia(46.05), Iowa (46.98), N.Hampshire(47.2), Minnesota (47.87), Colorado (48.19), Pennsylvania(48.24),

      It is obvious how it would then predict an EV win for Romney, but I’m sure people at PEC might take exception to some of those numbers.

      gryff :)

  • David Roher

    Glad you liked my piece! I discovered your site a few months ago and have been following it since. Though the article was mostly about Silver, I think your research is the best evidence that the media hate precision.

    • Olav Grinde

      David, that’s a fine article!

    • E L

      Absolutely. Great. Article. I’ve always loved the Chicago White Sox story. I remember Michael Wilbon, a Chicago boy, raging against the prediction and making fun of Silver. After the season, Wilbon never mentioned his rant or the prediction again. Now he’s a national sports commentator kissing up to Magic Johnson. I hope Joey Scarborough isn’t as lucky.

    • E L

      P. S. I bookmarked Dead Spin.

    • wheelers cat

      I really loved it. And I especially like that you actually believe your target audience can read.

    • Ms. Jay Sheckley

      Welcome, David Roher!
      I (and others, clearly) hope you’ll speak up more often.

      With your help it may soon be better known that an academic/scientist with two full-time jobs is, for a hobby, the finest election forecaster.

  • Neil S

    Not much has been said about the robustness of the outcome. There could be small changes and the probability of victory could dip from 98 to 90% very easily. This changes the odds from 1 in 50 to 1 in 10. I saw the opposite over the past few days when the winning probability went up from 93 to more than 98 % with what are surely minor poll variations. The reason is of course that the election is still very, very close. Thanks for a wonderful effort!

    • Ms. Jay Sheckley

      No, it’s far more stable.
      If you read more of the articles here, and even the comment threads, I think you’ll come to agree.

    • Craig

      So the probability changed 5 percentage points in a few days – quick, how many days until the election?

  • Olav Grinde

    Fine articles in Wired and Deadspin.
    David Roher is spot on: Dr Wang’s site is criminally underrated. Although I’m glad to see the steady increase in traffic, this excellent site deserves far more attention.

    • skmind

      Second that. I only found out about it from some commenter.

      The appeal to me is the fewer number of variables.

      I prefer knowing the chances of rain based on barometric pressure being as or more correctly estimated than if they additionally accounted for frequency of rains on Wednesdays in October because someone is seeding the clouds 473 miles away from where I live.

    • Ms. Jay Sheckley

      Couldn’t agree more. It’s frustrating. Attention goes to fame and silliness, not quality. Also, there a lot of GOP money in the past two weeks is promoting articles that appear in google searches headlined: “Poll exclusive!”

      Unsurprisingly, these promise landslides for the talented Father of Lies.

  • Steve Norton

    Sam, I’ve also wondered about Paul G’s (above) concerns about the differences between your model and Silver’s. Nate Silver recently discussed the uncertainty resulting from the possibility of a systematic bias in poling (e.g., resulting from the low response rate to polls and uncertainty in the demographic that is more likely to respond to polls). Nate claims that this uncertainty is reflected in his lower prediction probability. Perhaps you’ve discussed this before, but how does your model account for the possibility of systematic bias, and are past election results sufficiently reliable to discount (or account for) the possibility of this bias?

    • Olav Grinde

      A question along the same lines:
      Is there hard evidence of election polls becoming less accurate as the response rate dips ever lower?

  • 538 Refugee

    Shoot more messengers. That’ll learn um.

  • owlofminerva

    I’m not sure why we need these eggheads like Prof. Wang and Mr. Silver. Like Jim Cramer of CNBC, I prefer to go with my gut – Obama ten-point win and 440 EVs!

  • Anne

    While I find Nate and Sam incredibly reassuring, I still get upset when I see curveball polls on RCP. Like Susquehanna finding a tie in PA, and some supposedly democratic pollster finding Romney up by 1 in Michigan. What gives? Thanks so much, especially if you can give me more reassurance.

    • Froggy

      Anne, Susquehanna has been consistently on the pro-Romney end of the spectrum in PA. The last 30 polls in PA:
      O+3 or more: 24
      O+2: 3
      O+1: 1
      Tie: 1
      R+4: 1

      Susquehanna has five polls in this set, the R+4, the Tie, the O+1, and two out of the three O+2’s.

      FMWB has had a similar position in Michigan. Out of the last 30 polls in Michigan, FMWB has four. Those four include the three most favorable results in the set for Romney (O+1, Tied, and today’s R+1).

    • Anne

      Thanks so much., Froggy. I feel better already!

    • Joel

      @ Froggy: at what point does a house effect discredit a poll entirely? I think the offsets should not exceed +2 in either direction, frankly.

    • Froggy

      Joel, I see that Nate Silver says that Susquehanna has a 5 point Republican lean. I’m not sure what it means to discredit a poll entirely, but if you were going to do so, this would probably be one of the ones to start with.

    • Joel

      @ Froggy, I guess my concern is that if your bias is so strong, it can’t just be in your likely voter screening. I feel that a bias like that is indicative of methodological error.

  • S W

    Dr. Wang — I find the methodology and the results of the RAND polling fascinating, but curious. Is a repeatedly measured sample (specifically a sample that receives daily feedback about the measurement, like the one used in RAND) subject to Hawthorne effects?

    And, does it matter?

  • wheelers cat

    What I find super interesting right now is Sean Trende’s garden of gethsemane agony over what he is going to say Monday.

    Poor guy, he’s smart enough to know Linzer is correct– but he’s welded to Jay Cost and draws a paycheck from the Right.

    • E L

      @ wheelers cat: Keep us informed. After the election, I hope PEC will keep going with posts about the agony of the Right. Then, I believe the Right will be on to Impeachment fun led by Fox and Friends’ gasbags. Benghazi forever! By the way, whatever happened to that Clinton guy after his Impeachment adventure. The Right sure took care of him, eh? Great political strategy, guys.

    • wheelers cat

      poor Sean. he’s like the Token Numerate of the Right. He’s going to be crucified whatever he does.
      One thing that will change is that the Right wont be able to throw radar chaff so much. Sure, they will blame Sandy like they blamed the Econopaplypse in 2008…but its really game over for the GOP.
      By 2016 there wont be enough white folk left in the electorate for them to game the system with the southern strategy and the Sailer Strategy anymore.

      And Dr. Wang, Dr. Linzer and Nate Silver will be founders and charter members of the New Numeratocracy.

    • giggityd

      They have Marco Rubio for 2016. He’s a smooth salesman and will surely be used to attract non-white voters.

    • wheelers cat

      gig, it will not help them if the GOP base is still allwhite. I expect Julian Castro will be on the ticket as the antidote to Rubio. Compare their wives to see the strat. Mrs. Castro is a beautiful latina, and Mrs. Rubio is a corn-fed blonde gringa.

      E L its really like some tragic menage a trois, with Linzer, Trende and Cost.
      I predict it will end in tears. … tears of hilarity pour moi. ;)

  • Shawn Huckaby

    Heard a very nuanced and in depth discussion of the recent attacks on Nate Silver on this week’s Political Gabfest podcast:

    It’s in the third section, but it’s all worth a listen.In case you haven’t heard of it, one of the principals is John Dickerson, who recently was discussed (and posted directly) on the Ro-mentum topic. No direct mention of PEC, but perhaps an oblique reference?

  • Power word, Kill

    Science has a liberal bias! It’s not fair journalism unless you give equal time to a right wing tea leaf reader. Wait, what?

    • 538 Refugee

      For all the talk about class warfare from the right I haven’t forgotten Rush reading a list of professions that automatically make you a liberal. Basically, getting a real education put you on the list.

    • Power word, Kill

      Rush has an education too! He graduated kindergarten some time in his teens.

  • Osso

    Romniac and his fraud roadshow seem not believe in “intelligent design” when it comes to fact based analysis as they live in Ayn Rand’s fictional dream-tree-house….

    They will be crashing down on November 6 as gravity is a stubborn and unforgiving force.

  • Gordon Cormack

    Yet another article that speciously concludes from a new poll showing Obama +2 in Ohio [duh!] that it is a toss-up. Reason: the poll has a margin of error of +/- 2%, so it is a statistical tie [sic].

    • Michael

      A 2 point lead with a +/-2.2 MOE is not saying that the race is a tossup. It’s saying that there is about a 90% chance that Obama is leading.

    • Joel

      Yes, I hate this. As a Seattlite, I constantly harp on the stranger for reporting +2 polls with +/- 2.1 MOEs as “too close to call”. No, it means that the lead is “statistically insignificant” which is not the same thing. A statistically significant lead means that – should the poll hold – the candidate leading would not lose barring some failure in the polling hypothesis.

    • Stephen Coltrane

      I’m afraid the BBC here in the UK has also been constantly pushing the ridiculous “too close to call” line; though tonight a junior reporter finally admitted that Obama has a small lead in the battleground states (the anchor immediately cut her off and said, “so, too close to call, then”… *headdesk*).

    • Pat

      @ Stephen Coltrane
      Exactly, here in Switzerland, where I am watching French and Swiss news, it’s also the same problem! Tonight the Swiss news emphasized the latest national poll giving both candidates tied at 48%. Not only does it sound like they ignore the electoral college system, but also like they think only 1 national poll is published every day!
      In the French news, they also insisted on the 48-48 poll AND indicated how Romney had the momentum and has kept catching up on Obama…
      Sigh… (I wrote the journalists an angry email)

    • Ohio Voter

      Dispatch Poll is a mail poll. Still shows a two point lead. Not worried about Ohio at all. Median is firmly at 3%. Personally I expect Ohio to end up around +4 Obama

    • mediaglyphic

      Dr. Cormack as a computer scientist, don’t you scoff at articles like these, they don’t quote the data. they don’t even state that the of the 35 polls since OCT 1, only4 have romney leading. Its only a tossup, if the coin is rigged!!

  • Rafalca

    I want to see you eat a bug!

    • Olav Grinde

      I, on the other hand, want to see Karl Rove eat crow!

    • wheelers cat

      Rafalca! grand to see you here.
      Would you say Romney has lost his *impulsion*?

    • Ms. Jay Sheckley

      A bug with or without tabasco?

      Seriously, sportsfans, it’s not about Sam eating a bug. Though my gut rates it 96% probable that he’d go through with this, don’t forget he said it to teach US what these probabilities mean.

      I cant think of anything I’d like less than eat a bug, except eat a tabasco bug which is against doctor’s orders. But I’m so grateful to Sam for the free Princeton education he’s giving me, I say let’s pledge to eat whatever he must.


  • wheelers cat

    I think….this election cycle is so vituperative because 2012 demographic evolution is really the extinction event at the K-T boundary for the All White GOP.
    Now they have to evolve or go extinct.

  • Montykoolaid

    Sam, i’m sorry, but come election day you well be completely wrong alongside Nate.

    Not because your models or math are wrong, but think about it. Thousands of millionaires and billionaires dumping untold amounts of money…

    They WILL steal the election, I mean it. It means too much to them, they cannot lose this time.

    • mediaglyphic

      millions of thousandairs will beat thousands of mllionaires!!

    • orchidmantis

      Per TPM they are apparently using their extra money to flood Texas with anti-Obama Super PAC ads. Because they have to spend it somewhere, and all the ad space in swing states has been sold.

      That or they think Texas and Oklahoma are in play.

    • Ms. Jay Sheckley

      “they have to spend it somewhere”.
      Didnt i read that Rove plans to keep his Superpac zillions?

  • Dgarr

    Thanks for posting the upcoming programs. I look forward to each one. :-)

  • HowardA

    Mr. Wang, I was going to urge you not to go on CNN, because you don’t want to influence the outcome, you only want to report what is likely to happen. However, since you will most likely appear on CNN, I ditto Mark Sillman’s advice above, to emphasize things such as how important people’s votes are to determine disputed elections in the house and senate, and indeed in the presidential race as well. Likely voters need to actually vote!

  • Steve

    Mr. Scarborough: My last name is also a Scarborough. But, a big difference between us is that I am a practicing statistician who seeks the truth, be it Rep, Dem, Indep, or otherwise. In short, you are full of it.

    We will most certainly find out in a few days as to who is closer to the truth. My questions to you: will you eat a bug if you are on the losing end? Or perhaps some type of bovine organic matter?

    • Joey Scarborough

      My gut tells me that I won’t have to have any bugs in my gut. This is a horse race so nobody can know the outcome but I will drown you in political truisms!

    • Ms. Jay Sheckley

      Dear Joey Scarborough,
      What the point of saying on a revered, accurate poll aggregation site that no one can know the outcome? Are you saying there is NO correlation between any polls and election outcomes? Can yu point t studies? Because sincerely, it would be helpful to know more regarding your motivations and sources. Otherwise, I’ll have to respectfully ask you to [expletive deleted]

  • Dr. Xylem Galadhon

    Sam- I fully agree with all your posters at PEC about what a service you are doing by being so accurate and on target with your work — very heartening to the liberal side of the US today.

    But — i’m concerned about the voting machines.

    I’ve seen you only comment on this offhandedly somewhere in a thread below one of your posts (so i couldn’t even go back and easily find it at this point), but in none of your posts themselves. And, given the issues surrounding it — the *known* voter suppression, demonstrating the unethicalness of a fair number of those in the GOP, their casual willingness to subvert the democratic process to achieve their ends, it is no surprise that many *are* concerned that if they have the chance, they *will* manipulate voting machines as well. And here are just a few articles in the *corporate* media raising concern about it:

    Not even mentioning the blogosphere, which has a huge amount of info, some of it quite documented — a quick search on “voting machine problems” will give you a large selection of pieces.

    While it’s true that probably none of this is *proof* at the legal level that there has been foul play in the past, or is planned for Tues, one of my concerns is that we may not be able to know, or the documentation could be destroyed, given the messiness of 50 state election systems (as mandated by the Constitution, which explicitly says that states are to run all elections), and the willingness of many GOP operatives to sacrifice fair means for what they perceive to be noble ends.

    Any chance you could address this in a real post further?

    I am not any longer concerned about the Pres winning in a fair fight (and haven’t been for a long while, in fact) — i *AM* deeply and profoundly concerned about it not being fair though. And have enough experience and knowledge of the 2004 (esp.) election and subsequent ones to know my fears are not baseless theorizing, either.

    Thx much for all your great work–


  • Simon

    I really hope Obama wins, I think my fellow citizens of the rest of the world and I are taking a collective breath

    • Some Body

      Yeah. I can testify that I’m in the only country on Earth where the majority has been brainwashed into rooting for Romney (and it only has 0.1% of the world’s population).

  • Olav Grinde

    @Ohio Voter (& anyone), two questions:

    1) What is your take on your Secretary-of-state Jon Husted’s last-minute directive with new criteria on how to qualify/disqualify provisional ballotts? It’s interesting that Mr Husted waited until the courts closed before issuing his instructions…

    2) Do you feel any reason to be wary of the software patch that Secretary-of-state Jon Husted ordered installed on voting machines in 25 counties? It is so-called “experimental”, thus bypassing review by any testing authority at the state or federal level… Do you trust it?

  • Olav Grinde

    @Ohio Voter (& anyone), two questions:

    1) What is your take on your secretary-of-state Jon Husted’s last-minute directive on how to disqualify provisional ballotts? It’s interesting that Mr Husted waited until the courts closed before issuing his instructions…

    2) Do you feel any reason to be wary of the software patch that Secretary-of-state Jon Husted ordered installed on voting machines in 25 counties? It is so-called “experimental”, thus bypassing review by any testing authority at the state or federal level… Do you trust it?

  • Steve Roth

    Hey Sam, after the election I hope you’ll be giving us a fever graph comparing a bunch of the predictor people:

    X-axis: days before election

    Y axis: error compared to final results

    Maybe two graphs: one for popular vote, one for electoral college votes.


    • Andrew Ferguson

      hi Steve,

      yes, we wanted to do that after 2008, but didn’t have the data (and couldn’t get it from most other sites). this year, I’ve been collecting 538, CFP, E-V, ElecProj, HuffPo, RCP, and TPM daily since about July 16th. Votamatic hasn’t really moved over that time, but we could add Linzer’s lines to the mix, too.


    • wheelers cat

      Andrew, everytime you touch that track it turns into gold.
      Its Dr. Wang’s special blend mix tape.

  • Anne Romknee

    Mr. Wang,
    When can we expect your final prediction…which you said would be coming this weekend?

  • Olav Grinde

    “Experimental” software patches to be installed on Ohio voting machines… Here’s more on the story.

  • Pankaj


    You are about to become a celebrity. I hope you don’t get angry at Wolf Blitzer’s stupidity.

    Gosh, it is difficult to stand that turd

  • Tom Gavin

    Sam. What do you think of the following calculus should it make the cut of course. Whoever wins 2 of these 3 wins:Ohio, PA, and Virginia. Explanation: Conventional wisdom is that Romney can’t win without Ohio. But since he is consistently polling poorly there, he is making an effort in PA, which of course was always considered a longer shot. Should, (God forbid) that happen, i.e Obama wins Ohio but loses PA, he could survive that buy also winning Virginia, but not if he loses it.

  • Martin

    I see Colorado has fallen back to tossup. Is this based on a poll or what? I thought J.Messina said that Romney needs 59% of the leftover vote in order to win it there? Is Mr. Messina just FULL OF IT trying to string people along with his bravado? What’s the deal Jim?

    • ChrisD

      A tied CO looks wrong to me. There are 8 polls on from the last week (10/27-11/2). Ordered, they’re O+4, O+3, O+3, O+2, O+2, R+1, R+2, R+4. The median is O+2. I assume the 3pm update will get it correct.

    • Ohio Voter

      Doesn’t look like the 3pm update caught it

    • Ohio Voter

      Nevermind, yes it did

    • ChrisD

      Hear are all the changes from noon to 3pm:

      VA: O+2 => O+3
      PA: O+4 => O+3
      CO: tied => O+2
      FL: O+0.5 => tied
      MN: O+7.5 => n/a
      NM: n/a =>O+8

    • Joel


      tied CO and toss-up NC both look weird to me. Is pollster withholding data?

  • Ashbel Green

    Can you help me evaluate the seriousness of the unskewed polling brewhaha? As I understand it, pollsters release not only their numbers but how the people they are reaching identify themselves. Republicans have been saying that in these polls Democrats are seriously overrepresented, which skews the polls to a more favorable result for Obama. They are claiming that based on the +democrats in the samples, Obama would need a turnout advantage better than he received in 2008 to justify those poll numbers. NBC recently semi-acknowledged this complaint ( by offering both the poll results and the “unskewed” version. My questions are: is it true that pollsters are sampling Democrats art a higher percentage than in recently past elections? if so, why? and if so, why isn’t the Republican critique legitimate?

    • Tractarian

      “Pollsters are sampling more Democrats” because more people are identifying as Democrats.

    • Ashbel Green

      I guess I am still unclear how this debunks the “unskewed poll” meme. Could their be demographic reasons that pollsters are reaching more Democrats than normal? Is there evidence of a greater Democratic registration edge in 2012 than in 2008? What I am looking for a point-by-point critique of this Republican critique. Both Sam and Nate have acknowledged that the polls COULD be systematically off, however unlikely that seems given the market forces that make each pollster want to be right. But absolute faith in the market doesn’t help me sleep.

    • Ohio Voter

      Put simply, polls take random samples of voters, yes? If 35% of those voters say they are Democrat and 30% are Republicans, it’s because that’s what voters are identifying as.

      Party identification is fluid. Voters more likely to vote for Obama are more likely to say they are Democrat. If you start seeing consistent sampling that show more Democrats than Republican, it’s because there are more D than R. “Unskewing” actually does the opposite.

    • orchidmantis

      If you look at a graph of party ID, around the end of 2010 there is a huge dip in R and upswing in I. i.e. people fed up with the Republican brand–tea party types, probably a few moderates at the other end–started identifying as independents. The tea party are not swing voters, or moderates, or anything like that; they’re Romney voters who don’t like to call themselves Republicans. Thus the increased D over R identification, and the tendency for the much-enlarged Independent pool to trend right.

      The “pure” version of unskewed polls is that all pollsters first decide by party ID who will vote, and then keep calling until they fill in each group. (Virtually no one does this.) The version you hint at with actual undercounting of Rs posits that there are tons of Republicans unfindable by pollsters but ready to vote. Basically, no one has come up with a reason why Republicans would be hard to get on the phone: young people with cells (though live polls now call cells), non-English speakers, sure, but Republicans skew old and white. The only one I’ve even seen attempted is that they’re so seething with resentment they instantly hang up on pollsters, but will vote.

    • Michael Worley


      No, they’re not. Here is 2008 early voting (a high democratic year).

      Here is 2012 voting:

      Here’s 2008:

      And here is a recent Colorado poll:

      Historically Democrats tend to vote early, so PPP’s assertion of D+1 in Colorado seems unlikely, given early voting is R+2 this year and in 2008 it was D+1.8. Final 2008 turnout was R+1.

      Just one example.

    • Craig

      And of course the “seething with resentment” argument is undercut by the enthusiasm measures (which is why Gallup, which bets heavily on voter intensity, is such an outlier).

    • Ashbel Green

      Thanks. This makes sense. Romney’s support from independents, for example, may reflect support by former Republicans who switched affiliations

  • Martin

    i’m guessing the Zogby poll is the reason Florida is back to tossup right?

    • Joel

      I think Zogby polls aren’t counted by Pollster because their accuracy is so poor.

    • Ohio Voter

      Zogby polls are the JZAnalytics/Newsmax polls

    • Craig

      JZ Analytics is included in the Pollster database. Their numbers have been very middle-of-the-pack in the last month or so.

    • ChrisD

      Zogby polls are shown on as JZ Analytics/Newsmax.

    • Joel

      I stand corrected. Zogby did not cover himself in glory the last time around so I’m pretty skeptical of his work, generally.

  • E L

    @Sam Wang: If the host is Wolf Blitzer, he’ll sum up what you say as: “Dr. Sam Wang calls the presidential race a tie.”

  • Fred Smit

    IF Republicans studied science they would know Nate and PEC were accurate!

  • Child Progeny and Supergenius

    One issue that bugs me about the polls is the likely voter calculation. With the exception of the Rand polls it seems they transform a continuous variable (I might have a 3.14159% chance of voting) and converts it into a binary variable. This kind of transformation can cause all sorts of problems, and is something that is a red flag to me when I read a medical journal article. As an example, if candidate A had 200 people who prefer him and 49% likely to vote while candidate B had 100 people who prefer him but 51% likely to vote this model would predict B to beat A 100 to 0 when a better model would predict A beating B 98 to 51.I am honestly surprised such an unnecessary reduction of the information available is so commonly used.

    • Craig

      That’s become it’s not commonly used. I’m not sure what you’re trying to say, but it does not resemble polling methodology.

    • Child Progeny and Supergenius

      In reply to Craig most poll aggregators either only pay attention to, or heavily weight polls of likely voters. In those polls people who lean in one direction but are deemed unlikely to vote don’t count. Rand is the exception. I have done some reading on this, but perhaps I am missing something. If so, please tell me what pollsters do
      When they produce polls of likely voters instead of registered voters.

      Now if voting probability for each party is normally distributed with a mean near 50% I would imagine this would work quite well. If voting probability is not normally distributed this could result in big misses due to how we determine likely voters

  • Prairie Pundit

    The recent numbers for PA don’t make me want to reach for the revolver, but they do leave me wishing for a well-reasoned and -substantiated explanation.

    • SoleburyJim

      For what it’s worth, I live in central Bucks County, PA, the swingiest of areas in one of the counties touching Philadelphia, and the Obama GOTV effort appears to be really strong.

    • Craig


      Nine of the last tens polls give Obama a lead of between 3 and 10 pts. The outlier is a tie from Susquehanna, which models voter turnout on the GOP landslide in the 2010 midterms.

    • Prairie Pundit

      Thank you, Jim and Craig. I didn’t mean to give credence to the Rove-ney narrative that PA is “suddenly in play.” I just think it’s important to build a record empirically demonstrating why this – and other politically convenient narratives – are false. If we fail to build that record, we end up with four years in which the legitimacy of President Obama’s second term is undermined by assertions that “Sandy did it.”

    • Craig

      I understand the concern, but the narrative will be what it is regardless of the data. Just a few days ago Josh Marshall at TPM – a smart guy generally – front paged without comment a reader assertion that John Kerry led in Ohio going into Election Day. The final RCP average was Bush’s margin of victory to the first decimal point.

      To too many people, Data Doesn’t Matter.

  • Anita

    How often are you getting NC polls? I heard there were none over the weekend which is surprising considering how close the race is here.

    • ChrisD

      NC’s last poll was 10/29-31, a PPP-tie. Here are the NC polls, ordered to make the median easy to see: R+8, R+6, R+5, R+1, tie, tie, tie => R+1 Sam’s table has the median at R+0.5. He’s including a 10/23-24 O+3 poll by Grove Insight that I think is outside the week window.

  • gryffydd

    @ Anne and Froggy

    With regards to the FMWB poll of Michigan this is from their report:

    “Ninety thousand seven hundred and fifty-three (90,753) calls were placed, 19,003 most likely Michigan voters answered the initial filter question and initial ballot test but did not complete the full survey and 1,913 respondents fully participated in the survey. The response rate for this survey was 2.11% of the call universe and 10.07% of the screened participants.”

    Seems like a lot of calls to get 1913 LVs.

    Oh, and they are “telephone-automated polling” outfit with no apparent attempt to deal with cell phones, and in August they did a poll of Florida that gave Romney a 14 point lead (one heck of an outlier!)

    gryff :)

  • Pat

    Argh, I hate it that the 2 most likely outcomes are not 303 and 332 (depending on where Florida falls). The large difference due to the large EV weight of Florida will make it tricky to come up with a final estimate. If we guess 332 and Obama instead loses Florida, many more people might be able to get more accurate projection even through faulty arguments and gut guesses…

  • Tapen Sinha

    All this discussion about how Romney is going to win reminds me of an old Monty Python classic – the Black Knight


  • Reason

    Forget it. It is all over. The Redskins lost, so that means the present administration will lose. Of course, this so called “prediction” was wrong in 2004 so….yeah.

Leave a Comment