Princeton Election Consortium

A first draft of electoral history. Since 2004

Final predictions for 2008

November 4th, 2008, 5:06am by Sam Wang


The data are in. It’s time for predictions – with error bands.

PRESIDENT
Electoral vote: The final polling snapshot is Obama 352 EV, McCain 186 EV. The confidence bands are 68% [337,367] Obama EV, 95% [316,378] Obama EV. (Note added, Nov. 24: The snapshot was not stable. Variance minimization identifies longer periods for averaging, which give a result of 364 EV.)

http://election.princeton.edu/2008/11/04/final-predictions-for-2008/#more-2288

Bias analysis: I expect cell phone users missing from landline surveys to give Obama a 1% boost. Based on the bias adjustment I gave you earlier, this scenario brings the median to a level that is still within the 68% confidence band above. I make my own personal (not official!) prediction: Obama 364 EV, McCain 174 EV.

Popular vote: The median Obama-McCain margin is Obama +7.0+/-0.8% (n=9, 10/31-11/2). The error bar incorporates assignment of undecided voters. My final prediction is Obama 53%, McCain 46%, third-party candidates 1%.

Individual states: Since it’s directly contrary to the point of the Meta-Analysis, I consider this part to be strictly for fun. Normally, for a specific prediction I would use the mode. However, three states show an exact nominal tie over the last 3 polls plus last week of polling: Indiana (11 EV), Missouri (11 EV), and North Dakota (3 EV). These states must somehow be assigned. North Carolina (15 EV) has a median of McCain +0.5% (n=10). On average, each candidate would get two of these states. This would lead to 352-364 EV for Obama. Based on longer-timescale trends I assign MO, ND and IN to McCain, <del>MO and</del> NC to Obama. (Note: updated to reflect a better algorithm, 9:18am.) The map is given at the end of this post.

Turnout: 135 million. Curtis Gans of American University’s Center for the Study of the American Electorate (CSAE) has released a report estimating voter registration at 153.1 million, or 73.5% of the eligible population. He expects turnout in the range of 132-135 million. InTrade is high on turnout being at least 60% of the over-18 population in the US. Assuming a population of 228 million this suggests 137 million or more votes cast. I use the top of Gans’s range.

SENATE
Final breakdown: 56 Democrats, 42 Republicans, 2 independents, for a working 58-42 majority. The next most likely outcome is 59-41 if Franken wins (see below).
The probability of Democrats/Independents reaching 60 seats is 1.4%.
Democratic: AK (Begich), NC (Hagan), OR (Merkley)
Republican: GA (Chambliss), KY (McConnell), MS (Wicker).
Too close to call: MN (Franken-D vs. Coleman-R-i), with a tiny advantage to Coleman.
If Chambliss does not receive 50% of the vote, he and Martin will go to a runoff.

HOUSE
Final breakdown: 257 Democrats, 178 Republicans. The error bar is +/-3 seats, making a 68% CI of (254, 260) Democrats and a 95% CI of (251,263) Democrats. This estimate was made using aggregated House polling data from Pollster.com.

Postscript: 3BlueDudes.com has a compilation of 45 projections, one of which is my own. The other 44 projections give a median of Obama 353 EV, McCain 185 EV (SD=27 EV). The 68% range of their predictions is [337,367] EV. The Meta-Analysis encompasses them all.

Tags: 2008 Election

39 Comments so far ↓

  • Blair

    Great work, Sam.

    My guess is that, other factors equal (which they are not!), more recent polls will prove more predictive of individual states’ results. That would have Obama taking Missouri and losing Indiana and North Carolina, and North Dakota. But we’ll see: Local turn-out operations can make liars out of the polls, especially when the race is so close in those places. (!)

    So, the rest of us: Don’t let all this poll watching stop you from getting to the polls and voting, if you haven’t already voted!

  • Ed

    Thanks for the commentary, charts and analysis and let the day begin! Any predictions on what will replace all the political ads in TV land?

  • David

    Thanks for all the analysis, Sam. Very informative.

    My only quibble is that early voting may have locked in NC for Obama. Thoughts on that and the African-American turn-out impacts?

  • Jim

    I’m encouraged that both Intrade and Betfair have about a 92% win probability for Obama as of about 6:15AM EST on 11/4.

  • Mike L

    Prof. Wang,

    Muchas triple gracias for all your time, energy and sagacity this election. As in 2004, you have been my go-to-blogger throughout the ups and downs these past months.

    Please enjoy a richly deserved recharge over the weekend, and dog gone it, I deeply hope your EV estimate is close to the mark.

    As one who’s been disappointed far too many times going back to working as a volunteer for McGovern, my safe prediction is 312 . This result is predicated on the delivery of one Obama EV from Omaha—just to be a mavericky maverick!

  • Sam Wang

    Mike L – you’ve gone rogue!

    David – I think many pollsters include early voting in their totals. Anyway, after voting those people don’t shun the phone, do they?

  • David McConnaughey

    I’m v. curious about early voters too and how polling incorporates their votes.
    We’ve had massive early turnout in NC which has been heavily Dem. Our small town, Pittsboro, had over 2000 votes cast when WE voted a week ago last Thursday and the Dem poll watcher said it had been ~ 2D/1R. Even down east, in a very poor, rural county, a friend doing appraisal work was surprised to find a very long line ~ 80% African-American on the opening day of EVoting..Hoke County is ~ 40% Black so a great job vis a vis registration AND getting folks to the polls had been done down there.

  • Luis Villa’s Blog / voted!

    [...] any rate, we can be pretty sure what the outcome will be, but whoever you support (I voted both Working Families and Republican today (against a complacent [...]

  • Displaced Canuck

    My son has the same predicted EV as you so I hope you are both right. I say 356 so it nice to see some science backing up my gut feel. Great job on this site over the last few months.

  • Kay

    Just thought you might like to know: I live in Virginia and my canvassing has shown something that might be called the opposite of the Bradley effect. There are voters here, particularly older ones, who plan to secretly vote for Obama…

  • MB

    Thanks so much for your work.

    As I was futzing around with various scenarios that would lead to an Obama victory, here’s one that I haven’t seen:

    If Obama wins all the Kerry states, plus Nevada, New Mexico, and Colorado, and Virginia that gives him 270 on the nose. That means he could lose Florida, Ohio, and Pennsylvania and still win. Unlikely, but maybe it sheds some light on Obama’s red state campaigning strategy over the past week or so.

    Thanks again.

  • Mojo Mom

    Oh, yeah, baby–come on North Carolina!

    I drove around Chapel Hill this morning between 8 and 9 am, and there were cheering Obama volunteers standing out in the steady rain, waving signs celebrating and doing a little GOTV dance at at least six major intersections. I thought that showed a lot of enthusiasm and surely evidence that they had an army of volunteers, if they could post 20+ of them cheering around town. (2 at some intersections, as many as 8 at others.)

    I drove by three polling places and they appeared busy but not overflowing–I did not see any lines snaking out into the rain. At the CH Library polling place there was a well-staffed Orange County Democrats table and no evidence of Republican campaigning.

    We are the bluest part of our new-true-blue state. (Fingers crossed.)

  • Michael

    If Obama wins all the Kerry states, plus Nevada, New Mexico, and Colorado, and Virginia that gives him 270 on the nose. That means he could lose Florida, Ohio, and Pennsylvania and still win.

    I assume that you mean all the Kerry states minus PA. Kerry got 252 EV; Kerry + CO, NM, NV = 271, without VA. VA alone can’t make up for the loss of PA under this scenario, but VA + IA would get him to 270.

    By the way and for what it’s worth, I voted this morning in Albemarle County, Virginia. The wait was quite reasonable, but the really interesting thing is that there was no Republican presence whatsoever at my polling place. They’d put up a bunch of signs in the parking lot, of course, but there weren’t any people there. The Democrats were out in force. I’ve been voting at this location since 1998, and this is the first time that I haven’t been accosted by GOP boosters on my way in. Maybe there really is something to do this supposed enthusiasm gap.

  • David

    Professor,

    If I may, let me try to articulate my question in a more sophisticated manner – can you expand on how the high rates of early voting and African-American turn-out affects the likely voter models? You note the cell phone bias is a likely +1% for Obama – is the turn-out affect better built into the models? is there data that would let you estimate its size?

  • Lee

    Is this a reasonable interpretation of Jerseyvotes?: For McCain to win, the polls have to be wrong enough that CO switches into his column. Thus, if I stay up late enough to hear CO results, I should have a good idea of who won. Or if I don’t want to stay up past the close of voting on the east coast, then PA is my best indicator of how the election will go?

    Thanks!

  • Sam Wang

    David – It’s my assumption that as professionals, pollsters have been thinking about these issues. For example, some of them ask if people have voted early, then make a note of it. Since they concern themselves with turnout, I do not believe that we, as non-pollsters, are better at guessing African-American turnout than they are. After all, they have the data and we just comment.

    The one exception is the cell-phone-only population. This population is underrepresented, and the Pew Center has demonstrated quantitatively that it can affect poll results. For this reason, I believe it is justifiable to apply this correction.

    Of course, there is the possibility of other systematic errors in the data. All such errors – including any effect you may think is at work – may affect results in knife-edge races within two points: IN, MO, NC, and ND.

    I’m working on a guide to let you measure the net overall true difference between polls and outcomes. More later.

  • MTV признал, что рок умер - Брянск.ru

    [...] Final predictions for 2008 This entry was posted on Вторник, Ноябрь 4th, 2008 at 17:30 and is filed under Модная музыка. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site. [...]

  • Sean

    Can’t vote. South African, permanent resident. My wife, American, was up at 5am and ready to vote for O.
    Great site. Was indispensable when explaining the election to my friends in Europe and Africa.
    Well done.
    Go vote!

  • David Martin

    “Quantity has a quality all its own”

    I doubt that Stalin would have been impressed by today’s statistical meta-analyses. But it looks as though the baseball world is deeply impressed, so much that the NY Times ran a health care op-ed piece by Billy Beane, Newt Gingrich and John Kerry on October 24. Nate Silver’s in good company.

    This week, James Surowiecki at the New Yorker has a nice column on traders paying far too attention to information on what other traders are doing. It’s true that traders have access to far more data than they used to, so it’s plausible that most of them make poor use of what’s available. Could many of them benefit from training in statistically-based decision-making? I suspect. Perhaps benefit from learning a bit of practical mathematical modeling (Excel spreadsheets are hugely useful)? Maybe.

    Then again, Princeton trained a lot of “financial engineers” who were presumably able to apply the best and brightest intellectual tools to moneymaking. Right now, some of them must feel like salt-water lemmings. Surfing, anybody?

    Lets hope the election meta-analysis encourages readers to become better analysts.

  • Paul

    A personal prediction of the polling bias…. Despite 2004, you couldn’t resist the temptation, could you?

    Well, neither can I. The daily tracking polls have been truly weird in just the last few days — maybe likely voter models being confounded by who has already voted? This is an unprecedented election, and something is afoot. 352 feels right to me, but we may be in for many a surprise before Wednesday morning.

    Thanks for all your analysis, Sam. You’ve helped keep at least one politically engaged geek sane through this wild ride of an election!

    Now, off to vote, then report for duty at the Obama office.

  • S. DeDeo

    Sam — thank you for your work on this site. It’s been fun watching an expert run the statistics (and I’ve certainly learned a thing or two.)

    I think you also provided a terrific public service by allowing people to get beyond the silly generated “horserace” when every twenty days you get an poll outlier and the real news takes a backseat, and your debunking of the Bradley effect was excellent.

    By showing the public “how it’s done”, you helped raise the bar for the media a bit — as well as on countless bar-room discussions.

  • Dan Nexon

    You got in trouble last year for putting your thumb on the scales; now you’re doing it again. Hope history doesn’t repeat itself :-).

  • Michael Slavitch

    Sam;

    Do not discount the possibility of a massive power-law anomaly in turnout distribution. If that is the case there is no mean and all bets are off.

    Distributions with α<2 are referred to as being heavy tailed, as they have infinite variance. Distributions with α<1 do not even have a mean, rendering much of undergraduate level statistical analysis (i.e. mean ± 3 standard deviations type results) inapplicable.

  • Keith

    The Obama ground campaign is truly something to behold. Here in Boulder, CO every registered voter has had their door knocked on at least 4 times and they’ve been called many more times than that.

    There is a giant army of people who are steadfastly dedicated to making sure every single eligible voter is counted. There is talk of 95% turnout here in Boulder County. It’s a very exciting movement to be part of.

  • Frank

    Sam,

    Thank you very much for your information and analysis this year. I look forward to hearing your post-election debriefing.

    According to your state margins, McCain only has to hold on to all the states in which he’s leading (notably NC, but I also notice that GA is tightening), break all the ties (IN, MO, and ND) in his favor, and surge in FL, VA, OH, CO, and NV (PA and NM being even longer shots). Think Obama wrote a concession speech?

  • Bruce W

    I highly recommend that folks have a look at the 3BlueDudes link Sam provided. It’s fairly impressive that of all the pollsters, bloggers, etc. that they’ve compiled, only one (deeply crazy) guy has J Mac winning the election. If, by some miracle, he’s right, he’ll briefly be more famous that Joe (not his real name) the Plumber (not really a plumber).

  • Michael

    …one (deeply crazy) guy has J Mac winning the election…

    And amazingly, he gets this result without McCain flipping PA.

  • Sam Wang

    Michael – The deeply crazy guy is a poster child for the need to calculate a median rather than a mean from the 3BlueDudes data.

  • Eleição 2008: Corrida dos modelos (II) | Na Prática a Teoria é Outra

    [...] Os caras do Blog eleitoral de Princeton chutam 352, com Obama levando a Carolina do Norte, Ohio e [...]

  • Alan Cobo-Lewis

    Blair // Nov 4, 2008 at 5:37 am wrote:

    blair>Great work, Sam.

    Agreed!

    blair>My guess is that, other factors equal (which they are not!), more recent polls will prove more predictive of individual states’ results.

    Maybe. But there’s a tradeoff between sources of error. If public opinion has moved in the past few days, then confining analysis to recent polls will have less systematic error (bias). On the other hand, it’ll have more random error (noise) because it’ll be based on smaller aggregate sample size. This is one reason, I believe that Sam likes to keep things simple: even a more correct model might lead to poorer predictions than a simpler model if the more correct model requires a lot of data to estimate all its many parameters.

  • Michael

    Sam — I think he’s a poster child for a lot more than just that. When I figure out where we should donate to get him some help, I’ll post it here as a public service. ;-)

  • mg5904

    New guy here without much experience in statistical analysis and self-confessed scientifically/mathematically inept.

    Is it possible to quantify the effects of the precipitation occurring in the VA/NC area on the results. It has me concerned.

    Thanks,
    mg5904

  • John B.

    I, too, am mathematically inept. This is why I have so enjoyed this site. Thanks.

  • Paul

    I think that the most important judgment that pollsters must make in this election, and the one that has the least data to support it, is the number and preferenees of likely voters.

    Obama’s ground game is unprecedented, both in its use of technology and in its organization. Moreover, the effort is occurring in an equally unprecedented economic evironment. I think that this reduces the utility of past elections for predicting the voting behavior of registered voters.

    My sense is that the pollsters’ assumptions about the number and voting behavior of registered voters, not limited to first time voters, is leading them to underestimate the likely Obama vote by at least 2%. On the other hand, I don’t have the data to predict how this under-estimation is distributed amongst the states.

    Also, I love the site. Makes me wish I’d paid more attention to statistics is college.

  • John (Australia)

    Unprecedented interest in your election down here. MY basic theory is that George Dubya has trashed brand America and if Johnny Mac gets in, there goes the neighbourhood and there will be no coming back for the USA, goodwill wise.
    Final thought which should help the big O “heaps of swear words, American investment bankers”

    Cheers all and enjoy.

  • Mike L

    The Florida young Dems are reporting that Early/Absentee votes are over 50% higher than 2004, contrary to media reports.

    Link:http://www.futuremajority.com/node/3869

    This might be a case in point of what Sam said in comment # 17 about the underpolling of cell-phone only people, which has been demonstrated quantitatively by PEW.

  • Richard

    Sam:

    I did not see your challenge, unfortunately.

    Better late than never.

    http://www.geocities.com/electionmodel/2008ElectionModel.htm

    I got the recorded EV exactly right (well, if you round off the expected 365.3 EV). And 53.1% for the popular vote.

    Unfortunately, the Recorded Vote is never equal to the True Vote.
    Obama did better than the recorded vote.

    Here’s why:

    http://www.geocities.com/electionmodel/Final2008NEPScenarios.htm

    This is a summary of 2008 statistical anomalies

    http://www.geocities.com/electionmodel/2008StatAnomalies.htm

    TIA

Leave a Comment