Princeton Election Consortium

A first draft of electoral history. Since 2004

Campaign innumeracy

October 15th, 2008, 12:40pm by Sam Wang


The Median EV Estimator is boring these days because so few states are in play. The mode, which is at Obama 364 EV McCain 174 EV, contains an amazing 37% of the distribution. Near Obama’s red ceiling, the race is stable: the mode is the same even if 2% is added to Obama-McCain margins across the board. I realize this is less exciting than the fluctuations at other sites. That’s the power of meta-analysis.

In numerical news: John Allen Paulos, Temple University mathematician and author of Innumeracy, has an editorial in the Philadelphia Inquirer on the ways that campaigns play fast and loose with numbers. “Think, baby, think!”

Update: when this post went up the median was Obama 354, McCain 184. Now it matches the mode.

Tags: 2008 Election

21 Comments so far ↓

  • ndam

    Sam,
    As long as that line stays up high I am ok with boring. Thanks
    -nd

  • Paul

    I was playing the other day with the raw results you publish, comparing the meta-margin to median EV.

    They track pretty closely, but there are several time where the median EV hits a plateau while the MM keeps moving. It seems like sometime when there is a particularly big spike in the distribution, the median EV gets “stuck” on it. In essence the race is still fluctuating, but the median EV is moving within that big spike.

    I’m concern that this might give a false sense of stability. Not that the daily jitters are meaningful — they seem to have a lot more to do with when polls happen to arrive than with actual small changes in the race — but I’m not sure the periods of stability are meaningful either.

    So I guess I’m making my same old pitch for graphing the meta-margin over time as a potentially better indictor of the ebb and flow of the race, if not as an indicator of the outcome.

  • Snowball2

    Boring is good!

    Leave the fluctuations for the sites that live by the narrative fallacy.

    I would love to hear everyone’s (especially Prof. Wang’s) thoughts on the length of presidential campaigning in the US.

    My feeling is that our knowledge of the candidates and the issues does NOT increase with time. Moreover, it seems that external factors affect the results much more than whatever the campaigns are doing (unless, of course, the campaigns do something extreme and unexpected- such as the surprise Palin pick vs the uneventful Biden pick). For example, the state of the economy -and its prominence in the news- seems to correlate well with Obama’s favorable situation right now (I am also cautiously assuming causality here).

    The bottom line is that -in my opinion- the campaigns are extremely and wastefully long.

  • Frank

    What’s the ceiling on the meta-margin?

  • William

    I’ve been thinking that a graph of Meta-Margin, which arguably matters more than median EV, will definitely be more interesting than median EV.

  • Joe

    I’d also argue for a graph of the meta-margin. At this point, the election will only become close based on national-level changes (debates, foreign policy events, etc.). In such a case two possibilities exist:
    1. The electoral college goes with the popular vote winner. If we consider Florida a true statistical tie (i.e., too close to be meaningfully distinguished within any reasonable margin of measurement (of votes) error), then this rule has never been broken.
    2. Even if the electoral college shakes out differently than the popular vote, any changes that cost Obama 5-8% are surely going to shake up the electoral college relative rankings at least as much.

    Conclusion: The current Electoral results aren’t likely to be as useful to those of us tracking what’s going on, as watching to see what’s happening in terms of the meta-margin as it fluctuates (and possibly takes on real movement in one direction or another).

  • Sam Wang

    Great minds think alike. I’ve been playing with meta-margin data, but won’t be posting for some time. We’re fixing some glitches in the archive. As a reward for comment-thread nerds, here’s a sneak preview:

    The graph is less noisy than I was expecting. This date range shows similarity to the EV estimator: the Palin bounce and crash and a steady rise since then. It looks like things have stabilized for now.

    In regard to the duration of campaigns, if we did away with October, wouldn’t everything move up, and then we’d have a boring September? On the other hand, if Labor Day is a hard limit then that wouldn’t be the case. Also, there would be more time for a smooth transition.

  • Sam Wang

    By the way, is everyone aware that the meta-margins are archived here? Also see the geek archive.

  • Michael S

    The meta-margins should be smooth otherwise it would indicate that the input data is noisy, ie inconsistent and unreliable. What we do know is that the input data is consistent.

    If you look carefully, the rate of change is almost certainly self-similar.

  • Sam Wang

    Michael S, I think that’s correct. I think the argument was that if the EV estimator is saturated the meta-margin might still show some dynamic range and therefore be of interest.

  • Michael S

    You’re no longer looking at signal patterns. That appears set. You are looking at power. That’s all that is left. The question is: Can the Obama campaign crank it up to 11?

  • Sam Wang

    A winning Presidential campaign might run up the score to (1) claim a mandate, (2) demoralize the opposition, and (3) bring in downticket races to ensure achieving policy goals.

    (1) and (2) are especially helpful considering the challenges of the coming few years. (3) is hard in the Senate because Obama/Biden trail the Democratic candidate in close races in Georgia, Mississippi, and Kentucky. I’ll guess that the closing moves will involve winning marginal House races in blue states.

  • IYH

    Geez. 364? I know we’re approaching Obama’s upper bound, but, at this rate, you might need to rescale the y-axis on the median EV estimator plot.

  • Frank

    You just surpassed 538’s EV estimate, maybe for the first time!

    If your EV distribution on election eve were similar to the one at noon today, what would your prediction be? The lower median with low prob, or the higher mode with much higher prob?

  • Sam Wang

    Frank, it depends on what the question is. If the EV total, then I’ll go with the median. If it’s an exacta bet, then the mode. There’s a good chance that they will be identical. A major question is whether to assume any bias factor other than zero.

    I am currently developing an interest in Senate and House predictions. As of today it’s 3:1 in favor of a 58-42 split in the Senate, with most of the remaining probability in 57-43. For now, a 60-40 split is about 90000:1 against. I have not yet tackled the large task of sifting through House race data.

  • William

    What happens if you add 2% either way?

  • Paul

    Today’s 10-poing MEV jump is unlikely to represent a sudden shift in the race; it’s a good demonstration of the hazard of tracking EV over time. Almost a sort of oddly-shaped quantization noise….

    MEV is probably most useful as a predictive metric; it is certainly the right thing to look at the day before the election. However, as Sam wisely pointed out at the start, this is a descriptive analysis, not a predictive one. This discussion convinces me that MM is the right metric for tracking.

  • Frank

    Why does your distribution of Senate Democrats differ so markedly from 538’s (e.g. 25% v 9% at 58 seats)?

    In predicting the Senate, one question is whether the coattail effect is fully captured by the Senate polls or whether the Senate poll result should be adjusted in the direction of the Presidential poll result of that state.

  • Sam Wang

    As you might expect, I’m just offering a snapshot of current polls. Silver is making a projection to Election Day. It’s probably the same reason that he shows some Presidential outcomes in which Obama gets 420 EV: there’s a chance of movement up.

    There is also this “coattail” effect you claim, but Democrats’ median expectation is a gain of 7 Senate seats based on polls alone. I haven’t looked into whether there is good evidence for a Presidential campaign drawing a result away from empirical polls.

  • James

    Today may have marked a slightly unheralded milestone.

    I think this is the first time your projected electoral vote (364 vs. 174) has given Obama a better than 2:1 ratio (359 vs. 179). I’ve seen 350 used as the “landslide” threshold, but a 2:1 ratio would seem to be a cleaner definition for a landslide…

  • Frank

    Your snapshot “Democrats’ median expectation is a gain of 7 Senate seats based on polls alone” matches 538’s median prediction, despite big differences in the two distributions, if I’m following this.

    I guess there could also be a split-the-ticket effect, though McCain (perhaps generously) hasn’t made that case. In any event, this is easy enough to model.

Leave a Comment