Princeton Election Consortium

A first draft of electoral history. Since 2004

The collapsing distribution

October 10th, 2008, 12:22pm by Sam Wang


I love the daily commentary by Andrew Tanenbaum at Electoral-vote.com. His analysis hits a sweet spot of providing one layer of judicious data analysis and a daily dose of spot-on commentary that usually hits the key point. His site also contains hidden riches of local race coverage. Today he covers the financial crisis, the threat to capitalism, and the lameness of traditional media coverage of the EV horserace. If you only have time to read one website for election data and analysis, make it his. (Then come back here.)

Normally I don’t like to write about specific combinations of states, which runs counter to the rationale for the Meta-Analysis. As you may have guessed, I think too much attention to specific states and demographics creates the risk of missing the bigger points. I have formed the firm conviction that it’s time to stop fussing over the Presidential race and focus on downticket races.

But things are getting so simple that state-by-state analysis of the national race is easy.

For about a week the EV estimator has basically been immobile. The fundamental reason? As I wrote before, Obama is near his ceiling of support. With few exceptions, McCain’s remaining states are too red to be captured. Conditions could change, but it would take a really big push to get Obama over 380 EV.

Look at the probability distribution, which is more spiky every day. This reflects the fact that based on current state polls (medians and original data), only four states have win probabilities between 5% and 95%: West Virginia (Obama win probability 12%), North Carolina (65%), Missouri (69%), and Ohio (93%). Assuming OH goes to Obama and WV goes to McCain, there are four major permutations formed by who wins NC (15 EV) and OH (20 EV). These possibilities contain a whopping 74% of the total distribution (raw values available here):

Total EV Probability Scenario
338 EV 8% McCain wins both
349 EV 18% Obama MO, McCain NC
353 EV 15% McCain MO, Obama NC
364 EV 33% Obama wins both

If there is a change, likely next stops are down to 349 EV or up to 364 EV. 364 EV is the mode, the first combination that comes up if you start clicking the pop-up map of current polls. It is:

In an August interview, Obama campaign manager David Plouffe listed his 18 target states. Plouffe might be regretting leaving WV out in favor of, say, Georgia (15 EV). But then again, regret may not be high on his current list of emotions.

A second way to look at the situation is the Meta-Margin, the amount of swing needed to reach a perfect toss-up. In the last 5 days the Meta-Margin has risen from 4.74% to 6.18% with no corresponding change in the EV estimator. When things were on a knife edge, such a shift would have yielded about 20 EV. Not now.

To summarize, electorally speaking the Presidential race is getting dull. The remaining question is whether McCain will (or can) stop driving his supporters to a dangerous mouth-foaming frenzy.

Tags: 2008 Election

11 Comments so far ↓

  • MarkfromCt

    Thanks for your analysis…As a stats junkie, it is nice to know that after all the mouth-foaming crowds the MSM keep showing, the election is really dull and the only question mark is the effect on down races…again tilting against Republicans.

  • gprimos1

    Shouldn’t that second line be

    Obama OH, McCain NC ?

    338 EV 8% McCain wins both
    349 EV 18% Obama MO, McCain NC
    353 EV 15% Obama NC, McCain OH
    364 EV 33% Obama wins both

  • James

    It looks like there are some places where you’ve typed OH but meant MO.

  • Nicholas J. Alcock

    Dear Sam,
    If I understand you correctly, the “spikiness” is caused by small shifts in polls only affecting these four states and whose small number of permutations of outcomes accentuate the EV shifts? May, I suggest that this volatility is influenced by the four states including one with a large number of EV’s i.e. Ohio and one with a small number of EV’s i.e. West Virginia. Am I right that if all four states had small Ev’s e.g. 3-5 the “spikiness” would be less pronounced? If all four states had identical EV’s wouldn’t this also reduce the “spikiness” ?Am I also correct because small shifts in state polls are common i.e. smaller samples, more less competent pollsters and that polling in some states is sparse e.g. West Virginia this makes the resultant “small number of permutations of unequal EV power ” effect more likely to occur?

  • Sam Wang

    Greg and James, thanks for the correction. From a meta-analytic standpoint, these states kind of look alike.

    Nicholas, your posting is basically wrong on all points. The spikiness arises from the fact that two states are extremely close. 2^2=4 spikes. If there were three close states we would expect 2^3=8 spikes.

    If the states in question had small numbers of EV the peaks would be closer but of the same size. If they had identical numbers of EV then there would be three spikes, not four.

    Small-state polls sample similar numbers of voters as large-state polls. Competence does not scale down strongly. For example, Ohio and Minnesota are largish states, yet the Columbus Post-Dispatch and the Minneapolis Star-Tribune still insist upon inflicting bad data upon the world.

  • Michael S

    Sam, given the polls it appears you might well be settling into a distribution that contains one spike, with no states being close, s=2^0.

  • Ginny Mayer

    What do we currently know about the effectiveness of the strategy of seeking the “dangerous mouth-foaming frenzy” vote? It would be nice to think that moderates and/or independents would find this strategy repugnant and would rush to vote for Obama. But today’s Gallup poll shows a 1-point drop for Obama from yesterday (narrowing the lead to 10 points after two days of 11 points) so there’s no guarantee that the lynch-mob strategy is doomed. Is there any polling data on the effectiveness of this strategy?

  • Sam Wang

    Ginny, you are cherry-picking a one-point fluctuation in one poll. The big risk here is that of a crazy act of violence.

  • Vicki vance

    Sam,

    I hope that you are right that the election is only boring at this point and that Obama will win! Still I don’t sleep well and cannot concentrate really well either while I worry that McCain/Palin will somehow steal it. What about all the stuff about voters being purged from the roles in several swing states? Is it possible that it could make enough difference?

    Love this site!

    Vicki

  • Sam Wang

    Vicki, the answer to your question is no. There is nothing on the horizon that is likely to change the dynamic, including voter purges. Also, by worrying about the Presidential race, you are using up energy that could be used in other races to be of help to your party. If you want to make a genuine difference you will pay attention there.

  • KenW.

    Sam:

    I really like your meta-analysis.. Wish it had been available when I was teaching research methods.

    One interesting point that shows up on Nate Silver’s SuperTracker chart that I think is less obvious on yours is the narrowing of variation over time.

    It’s almost a text book example of increasing precision in the measures.

    Thanks again for your efforts in this project.

    Regards,
    Ken

Leave a Comment