Princeton Election Consortium

Innovations in democracy since 2004

Outcome: Biden 306 EV (D+1.2% from toss-up), Senate 50 D (D+1.0%)
Nov 3 polls: Biden 342 EV (D+5.3%), Senate 50-55 D (D+3.9%), House control D+4.6%
Moneyball states: President AZ NE-2 NV, Senate MT ME AK, Legislatures KS TX NC

The post-Democratic convention bounce begins

September 10th, 2012, 9:34am by Sam Wang

The state poll meta-analysis has jumped by 10 electoral votes, and by 0.7% in Popular Vote Meta-Margin. Movements of >5 EV and >0.5% usually happen only when a real shift is occurring. Already things are looking very unlike the GOP-convention anti-bounce. Note that the last time the Meta-analysis moved this fast on the first day was the post-Ryan-VP bounce, which peaked at about 40 EV and 3-4%.

Our calculations grind slow, but they grind exceedingly fine. They are based on close to 100 recent state polls from, are very precise, and take days to reach a new stable point.

Presidential prediction 10 September 2012

My November prediction (and warning) are exactly the same. President Obama’s November re-elect probability is…87%, as dictated by this year’s polls and the amount of movement in previous re-election races. The predicted two-candidate outcome is popular vote Obama 51.5 +/- 1.1%, Romney 48.5 +/- 1.1% (1 sigma error bar), and Obama 312 EV, Romney 226 EV.*

Recall that I gave the probability on August 8th as 87%; on August 15th, 89%; and on August 23rd, 88%. Other predictions vary more because they use econometric assumptions that add unnecessary noise.

In my mind, the real  suspense is in control of Congress. This, not the Presidential race, is where the uncertainty for 2013 lies. If you want to influence what happens, give. Democrats, I have listed key races at this ActBlue site. Republicans, give at the National Republican Senatorial Committee.

* For Obama electoral outcome, the 1-sigma CI is 281-353 EV. The 2-sigma CI is 248-360 EV. For a long-tailed distribution that includes “black swan” possibilities, these are 61% and 86% confidence intervals.

Tags: 2012 Election · President

29 Comments so far ↓

  • Froggy

    I see that Ohio is listed as Obama +5% after the latest PPP poll.

    Pollster lists the most recent polling for Ohio:
    PPP (D) 9/7 – 9/9 Obama +5
    Gravis Marketing 9/2 – 9/2 Romney +3
    Gravis Marketing 8/27 – 8/27 Obama +1
    Columbus Dispatch 8/15 – 8/25 Tied
    Quinnipiac 8/21 – 8/21 Obama +6
    University of Cincinnati 8/16 – 8/21 Obama +3

    The only way I see to have the PPP result be the median is to throw out some of the polls listed at Pollster. Am I doing something wrong in the calculation? If some polls are being thrown out, what are the criteria for choosing which polls count?

    • Sam Wang

      Froggy – as stated in Methods, should be last 3 polls/1 week/no repeat pollsters. We never throw out polls. A pity, too, since we hateses the Columbus Dispatch. However, there could be something funny with Pollster’s API feed. Will investigate.

  • Froggy

    I suppose if most recent poll means the poll that has a most recent initial polling date, you’d end up with the three most recent (non-repeated) polls as:

    PPP (D) 9/7 Obama +5
    Gravis Marketing 9/2 Romney +3
    Quinnipiac 8/21 Obama +6

    Is that it?

  • Sam Wang

    Wow, today is test day. Amitabh is grilling me over on the other thread. My recollection is that we use median polling date. But now I have to go into the back control room…

  • Ralph Reinhold

    @Froggy: I haven’t run the numbers and won’t be doing so until after the first of October. I can’t answer for what Dr Wang did. I don’t believe you can use the final numbers. They have been filtered.

    When I average polls, I synthesize the original numbers. If there are 1000 people in the poll and it is 51% for one candidate, then I take it as 510 people. Then I follow the same procedure for the other candidate and other polls.

    This still has some problems because some polls are over 3 days and some 5. Some polls adjust their population distribution numbers and some don’t.

    There is a problem with adjusting the distribution numbers because small numbers have high noise. A poll with 1024 participants has a margin of error of about 3%. A poll with 100 participants has a margin of error of 10%. If you adjust the data of 1024 divided into 100 participant blocks, your margin of error is not 3%. I would have thought that the errors would be root sum squared, but that gives an error greater than 10%. When I get time, I will look into that.

    This effect may be why my sociology prof said you had to have more than 10% participants for a poll to be reliable. Since I took the class over 40 years ago, I don’t have a reference. That’s another thing I will research after Oct 1.

  • Perry Dane

    Hi Sam, We know that convention bounces recede. But have you noticed any interesting historical patterns about whether certain sorts of bounces are “stickier” than others? For example, do incumbents’ bounces tend to behave differently from challengers’ bounces? Have the smaller bounces in recent years tended to recede more or less, proportionally or absolutely, than the larger bounces that were more common in the “old days”? Etc. Just wondering.

  • JaredL


    I’m a new reader, came here through a link on Andrew Sullivan’s blog, and am really impressed. I’ll definitely make this a daily stop.

    I’m curious why you think your assessment of Obama’s win probability (87%) and Nate Silver’s (80.7%) are so much higher than intrade’s (58.7% as of now). I find it interesting that intrade doesn’t seem to have responded to recent polling. At the very least polls the last few days have been ahead of where they were a couple weeks ago but the prices haven’t moved much at all.

    • Sam Wang

      JaredL – Start by reading this. The short answer is that InTrade is very often wrong in this way. For more answers, search for InTrade on this site.

  • Sam Wang

    In my view, Ralph Reinhold’s approach is conceptually wrong because it does not address the underlying issue of house biases. These contribute more error than sampling error. Therefore counting number of respondents does not help matters, and in fact is likely to hurt because pollster methods are no longer weighted equally. Eventually, with bias correction one could imagine making use of sample size.

  • Andrew Ferguson

    Hi Froggy,

    I spent some time digging into this. Pollster’s feed is missing the Gravis Marketing poll for 8/27.

    Our script uses the midpoint of the poll as the date we work with. Therefore, we view the three most recent Ohio polls as:

    PPP (D) 9/7 – 9/9 Obama +5 (Mid-point: 9/8)
    Gravis Marketing 9/2 – 9/2 Romney +3 (Mid-point: 9/2)
    Quinnipiac 8/21 – 8/21 Obama +6 (Mid-point: 8/21)

    the Columbus dispatch is the 4th most recent:

    Columbus Dispatch 8/15 – 8/25 Tied (Mid-point: 8/20)

    then, we choose the median value as the margin — in this case, the median value is Obama +5, as we report.

    let me know if you have any other questions.


  • Ralph Reinhold

    @Dr Wang: You may be right, I hadn’t thought about the house biases but some like Rasmussen and PPP would really screw up the works. But, given the house biases (deliberate and inadvertent), I’d be concerned about combining polls from different sources at all.

    Combining sources with different statistics and different periods is very iffy. I ran the higher order statistics on Gallop data the kurtosis suggest it is non-gaussian , in fact, it is close to being the opposite of a raised cosine. (0.66 O: 0.54 R rather than -0.59).

    The entropy suggested that the data may contain non-random components. If I get a chance, I’ll check the actual shape of the distribution with a frequency plot (and remember the entropy).

    The skew of O (0.066) vs R (-0.38) are opposite sign and the value difference is remarkable. I would have expected them opposite signs, but closer to the same value. This may indicate a bias or that R’s numbers are more affected by events. We have seen the latter in your data.

    • Sam Wang

      Ralph Reinhold – that is interesting.

      I agree that polls constitute a can of worms, if viewed closely. Indeed, the more I look into the guts of polls, the more I want to find a way to do a good calculation without ever seeing those guts again. Of course there are good reasons to do more than I have done.

      Given your interests, I wonder if an informative approach would be to do simple simulations of known polling methods. Those guys have conferences and journals, the whole works. Then calculate the descriptive statistics. It could be good for building intuitions.

  • Ralph Reinhold

    If I understand betting odds, the international bookies favor Obama…I think they would be more in line with being neutral but un-informed.

  • JaredL


    Ladbrokes has Obama at 75%, Romney 32.3%. Those add up to more than 100% which is how they make money. If you take out their cut it is Obama 70%, Romney 30%. So, much higher than intrade but lower than the predictions here or at 538.

  • JaredL


    Thanks for the link and search suggestion. Enlightening stuff on intrade. I noticed in 2008 that Silver’s model probabilities had much higher variance than intrade but didn’t realize how systematically off intrade appears to be. As an economist I have to say I’m puzzled by that because I don’t believe there are limits on how much one can invest.

    • Sam Wang

      I have pondered the lack of limits on InTrade. It’s amazing that people do not use it as a principal investment strategy. Surely they must have safeguards of some kind to avoid massive influxes of money.

  • Olav Grinde

    There is one thing that puzzles me: Why isn’t President Obama pointing the finger at the obstructionism of Senate and House Republicans?

    Senate Minority Leader Mitch McConnell, House Majority Leader Eric Cantor and House Speaker John Boehner have blocked him at every step of the way. Even Obama’s minor appointments have been obstructed or delayed.

    It seems to me that President Obama would be wise to underscore that he needs a Democratic Senate and a Democratic House — and to tour states and campaign with candidates who are struggling.

    As Dr Wang has pointed out, the Senate balances on a knife’s edge. And there’s even a chance the Democrats could flip the House.

    So why isn’t Obama emphasizing that battle?
    Are we seeing the ingrained habits of a President who thought he could be bi- or post-partisan? If so, then I suspect it is a severe mistake.

    And — perhaps more relevant to the key topic on this site — would such a strategy be likely to be fruitful? Would it have an impact on the coming makeup of the Senate and House of Representatives?


    • Sam Wang

      Olav Grinde: Given the unpopularity of Republican policy positions as espoused by Rep. Ryan and Rep. Akin, as well as the expressed legislative strategy of Mitch McConnell, the Senate Minority Leader, it would be a mistake for the President to avoid making party distinctions – on average, at a national level.

      However, individual districts and states vary. For example, President Obama would be more helpful to Elizabeth Warren in Massachusetts, a strongly Democratic state, than to Jon Tester in Montana, a strongly Republican state. One would need to look closely to understand whether the Obama and Romney campaigns are acting strategically.

  • Olav Grinde

    @Dr Wang: Thank you for replying. What you say makes a great deal of sense; I fully realize that such a strategy would need to be based on a careful case-by-case evaluation.

    Is there much indication that the Obama campaign is trying to help some of vulnerable Democratic candidates ride his coat tails?

    I am in Massachusetts and must confess that I am astonished that Elizabeth Waren doesn’t have a double-digit lead. People complain about bankers and unfair practices in the world of finance — and yet they are not prepared to vote for the one person who has clearly and visibly fought to protect the economic interests of common Americans.

    What’s up with that?

  • Froggy

    Andrew, thank you for your response, but it still seems like there is a problem in your algorithm.

    North Carolina is currently indicated as tied, but here is the data:
    PPP (D) 9/7 – 9/9 Obama +1
    Civitas (R) /SurveyUSA 9/4 – 9/6 Romney +10
    PPP (D) 8/31 – 9/2 Tied
    High Point University/SurveyUSA 8/26 – 8/30 Romney +3
    Elon 8/25 – 8/30 Romney +4

    It seems to me that this should be Romney +3 or Romney +4 (depending on whether the SurveyUSA polls are considered a repeat).

    Here’s the data for Iowa, which is currently indicated as Obama +2:
    PPP (D) 8/23 – 8/26 Obama +2
    Rasmussen 8/8 – 8/8 Romney +2
    PPP (D) 7/12 – 7/15 Obama +5
    We Ask America 6/18 – 6/18 Obama +1
    Rasmussen 6/11 – 6/11 Romney +1

    Unless I’m mistaken (which happens with increasing frequency at this stage in my life), this should be Obama +1.

    It seems in both states that both PPP surveys are being counted, instead of only the latest one. Or am I doing this wrong?

    • Sam Wang

      Froggy: thank you for this example. There might be an occasional issue with the specific formatting of the Pollster API feed. We’ll look into it. Our email addresses are available for further direct contact.

      Olav: I agree with Aaron. Additional notes…

      Third-party candidates tend to fade in the finish in a system with two dominant parties. If you watch a particular race, such as VA Goode v. Romney v. Obama, you will probably see Goode’s support shrink in October.

      We take whatever pollsters give us. Some of them give the third-party candidates and some don’t. This kind of detail might help, especially for analyzing local/state races where third-party local candidates run strong, such as Maine.

  • Olav Grinde

    I have one more question, and it concerns a fine point of the Meta-Analysis. As far as I know, the poll questions present voters with a choice between Barack Obama and Mitt Romney.

    But Romney and Obama are not the only candidates!

    In at least 40 states (and possibly every state of the Union), Libertarian candidate Gary Johnson will also be on the ballot.

    I understand that Mr Johnson is expected to receive on average about 1 % of the vote — primarily to the detriment of Mr Romney.

    It seems to me that if Johnson is not included in the polling, then Romney’s scores might be off by ± 1 %.

    And that 1 % is what separates Obama and Romney in quite a few swing state!

    My question is simple: Won’t an omission of Mr Johnson render a poll incomplete and unscientific? What portion of the state polls included in the Meta-Analysis include the Libertarian candidate?

  • Aaron

    Olav, in case Sam doesn’t get back to you:

    Even if Obama could bolster certain senate candidates by campaigning for them, there would of course be a trade-off in terms of time he could campaign for himself—Warren could certainly benefit if he spent time in Massachusetts, but Obama would end up accomplishing very little for himself (beyond helping keep control of the senate should he win again but such a strategy is naturally risky). (Also there is something to be said for the simple coattail strategy of securing a strong enough victory and motivating enough of your own party that down-ticket candidates will be helped.) That much might be obvious, but Obama has also been perceived by Democrats as not being as willing to help down-ticket candidates as much as other presidential candidates have been—so for Obama it may also be a personal preference to not get as involved in other races.

    I think the question of third party candidates is a little hairier. On the face of it, I essentially agree with you—in states where Johnson (or the Green Party candidate, whose name eludes me) is on the ballot, to not include him in polling seems problematic. (When he and the Green Party candidate are included, Romney’s numbers seem to drop a bit more than Obama’s, but I haven’t seen enough polling to know how significant this trend is.) But I wonder if there is a justification for leaving such candidates off of polls—I’d be curious to know how well polling reflects the ultimate share of a third-party candidate’s vote. I can easily imagine, for instance, someone saying they’d vote for a third party out of frustration with the two main choices, but once in the voting booth vote for the lesser of two evils, anyway. Even so, I’d prefer that pollsters include the lesser-known candidates.

  • Marc

    Sam or Andrew,

    Are you guys currently using variance minimization like you did at the end of the 2008 race? If not, do you plan to? If not that exact technique, something similar?



    • Sam Wang

      Not until the end, which is when the swings will have settled down. Also, there is always a possibility that the approach may change slightly.

  • xian

    looks like the DNC bounce may be topping out as slightly smaller than the Ryan bounce at peak, although when combined with the RNC unbounce seems to more than outweigh it.

  • Matt McIrvin

    @Olav Grinde: I think it’s voters voting for personalities rather than policies or issues.

    Massachusetts can be a surprisingly provincial place. I get the impression that Warren doesn’t come across as an ordinary Bay Stater, and that counts for a lot here. Scott Brown is great at playing the just-folks game, with his famous pickup truck and his talent for schmoozing. It’s the old “who would you rather have a beer with” criterion. Democratic voters are somehow able to convince themselves that he’s not a bad guy.

    Also, I think there’s still a considerable amount of purely sexist resistance to female candidates, especially among male voters.

    Nevertheless, the race is closer than a couple of polls from a few weeks ago might have implied.

  • Olav Grinde

    @Matt: I would far rather have a beer with Elizabeth. In Scott’s company I would be bored out of my skull!

    But then again, I am no doubt the deviate. ;)

    Your latter point is interesting. For a long time I have been trying to decipher the Hillary-hating, which was such a dominant factor in the last election cycle. There are surprisingly many women who have a very deep aversion to Ms Rodham Clinton.

    Yeah, I’ve been noticing the improvement in the most recent Warren/Brown poll. Perhaps she got a convention “bounce” after her speech?

  • Matt McIrvin

    It’s hard to say; the race really hasn’t been polled enough, considering how potentially important it is. There’s not enough polling of the Senate races in general.

Leave a Comment