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

When nerds attack!

November 3rd, 2008, 5:01pm by Sam Wang

Robin writes: “As the [EV] distributions get ever spikier, they start to look like NMR spectra, with the state-flipped pairs analogous to spin-spin splitting!” Wow.

I’ve had the same nerdy thought. The number of peaks does convey how many states are uncertain based on polls (4: IN, MO, NC, ND). This leads to 2^4=16 major permutations. The spacing between the peaks corresponds to the number of EV for each of those states. Actually, since IN and MO have 11 EV each, the situation is “degenerate,” meaning there are only 8 peaks.

Things are indeed degenerating around here…go get out the vote.

Tags: 2008 Election

20 Comments so far ↓

  • Aaron

    Ok, but that’s 1D NMR. Throw in a NOESY spectrum reference THEN I’ll be impressed…


  • Nathaniel Hedman

    What an appropriate thought, especially given the organic chemistry I’ve been blowing off to check your site!

  • Frank

    When are the final polls reported?

  • Brian

    after the election? =)

  • Evans

    Wow… perfect level of dialogue for the theoretical chemist :)

  • Alan

    Is this a coup in which chemists are overthrowing the bioscientist power structure? How exciting!
    And specifically, the EV distribution is an NMR spectrum of an AA’B system (where A and A’ are Virginia and Colorado and B is New Hampshire).

  • Sam Wang

    Wow, that’s great. I can’t imagine this thread on any of the other sites.

    Later on this evening I’m going to take this over to the Meta-Analysis thread. Too specialized in its interest!

  • dw

    OK, weird place to toss this, but I noticed something odd:

    538 — 353 EV (based on their map)
    electoral-vote — 353 EV

    And you now have 353 as well. Pollster this morning was at 367, but it looks like they’re not including late NC and MO polls which would affect the end.

    So, at the end of this, it looks like all three of you are effectively converging on one number.

    Every poll analysis blog that rises must converge?

  • Sam Wang

    dw – You know, we all use the same data. Unless added assumptions do something wild to the calculation, this isn’t a crazy convergence.

  • Walter

    Well, many of the sites do “projections”, whereas Dr. Wang simply does snapshots.

    As we get closer to election day, the difference between these should decrease — since a projection is just a snapshot extrapolated forward.

  • Paul

    And — I just have to say it — as the number of states in play asymptotically approaches infinity, presuming their electoral votes follow a power law distribution and there is no lower bound on the EVs of states in play, the resulting graph has a nonintegral fractal dimension.

    There. I said it.

  • Andy

    Thanks for saying it, Paul. But don’t forget that you also need to assume infinite time resolution. Some condition on the independence of states in play is also necessary; if your infinite set of states behaved as a finite set of blocs which moved together, you would still end up with fractal dimension 1.

  • JohnL

    Paul and Andy, don’t forget that the fractals have an end point, whether they’re *2 or /2…and that point’s tomorrow.

  • Paul

    Andy: You’re quite right; we need states to be in play such that we don’t get a single spike, or a uniform bell curve. I don’t think Sam’s model allow for states moving together, so I wonder if it’s sufficient simply to specify that the win probabilities are all different?

    JohnL: Never forgotton. I will be out tomorrow pouding the pavement of Minneapolis from approx 10 AM – 8 PM.

  • Andy

    Paul: I don’t think it is sufficient. Even if Sam’s model doesn’t allow for states moving together, in reality it could happen that way. Imagine an infinite set of states with a finite number of subsets, such that changes in the polling of any one state were always correlated with others in its subset. Even though his modeling wouldn’t predict it, the snapshot could still perpetually move in a nonfractaline manner, driven by actual correlations between states’ polls.

    Paul and JohnL: I wish I cold be pounding the pavement tomorrow, but my ride to NH fell through, and there’s not much point canvassing CT…

  • robin

    I’m the one cited for making the uber-geek observation that the EV distributions look like NMR spectra (for interesting reasons). I just got back from an Obama-party, making GOTV calls. I was so psyched to get Montana to call, since I have a dear friend there with whom I’ve argued politics for almost 30 years. I come by my nerdliness honestly, having learned how to read NMR spectra at Caltech in 1980. Next election cycle, we can be even more geek-chic by doing Bayesian analysis of electoral projections. OK, everybody, now get out there and vote! -robin

  • Sam Wang

    Robin – I learned to read NMR spectra at Caltech as well, in 1982. Chem 3a, I believe.

    In regard to Bayesian analysis, the Meta-Analysis is only one step away. It uses posterior probabilities. It has been suggested to me that movement in other states and national polls makes some interpretations more likely than others. Using these other polls as priors would give a slightly different result for calculating state win probabilities – a Bayesian analysis.


  • Ming

    Sam, Thanks for the great website! 2 minor suggestions:
    1) Add major events option to the Meta-Margin graph like you do for the Median EV estimator?
    2) Include Sept 15 major event of Lehman failure/500 pt fall in Dow?

    Looking forward to the Bayesian analysis : )

  • robin

    OMG! (Apologies to non-techers.) We just missed one another at tech! Probably just as well, since I disqualified myself from ever holding public office on at least several independent axes while at tech! ;^)

    -robin (RC), Page House Veep, ASCIT Sec, BOC rep at large, started dating my (now) spouse playing opposite in the Caltech musical (Fiddler).

    I need to think some more about this Bayesian thing (I have been using MrBayes -should be RevBayes- for phylogenetic analysis). But first we need to finish off this election! Go nerds!

  • Mathphysto

    Not sure anyone will bother checking this now but…

    Has anyone hear ever heard of trying to analyze election data using stochastic methods similar to ARMA/ARCH models in quant finance? Seems to me like the poll volatility would have an autoregressive heteroskedastistic nature, and other features also seem to suggest a good degree of model portability from finance to elections.

    However, is the polling data even useable for such analyses, or is it too sparse and noisy? Any quants on this board who could weigh in?

Leave a Reply to Mathphysto (Cancel)