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

Testing the Presidential predictor

August 27th, 2012, 11:49pm by Sam Wang

(Update: If you’re coming here from the Huffington Post piece on the Colorado model, you may want to read my response to it.)
The predicted path of a storm can change. Hurricane Isaac was thought to be headed for Tampa (and the GOP convention) until a few days ago.
Hurricane Isaac path 27-Aug-2012 Now it’s headed for New Orleans.

Unchanged is the poll-based prediction for the 2012 Presidential race:
2012 prediction 27-Aug-2012

Here are the hard numbers:
President Obama re-elect probability: 88%.
Electoral prediction: 283-353 EV (1 sigma, the red zone), 250-360 EV (2 sigma, the yellow zone).
Vote share prediction: 51.6 +/- 1.1% (1 sigma).

These are long-range predictions, as explained here. They will change between now and the end of September, as more data become available.

Some of you wanted to know what this approach, which is based on polls only, would have said in 2008. Well…it would have¬†looked like this:

The low point at 40 days occurred shortly after the Palin VP nomination.

For 2004, my calculation is approximate since I don’t have access to the Meta-margins at present. It would have been roughly this:

As you can see, the 2004 race was a cliffhanger.

Last week I pointed out that today’s polling snapshot is only predictive on time scales of less than 40 days. It was only in late September and October 2004 that Kerry’s poor prospects became clear, as evidenced by the daily snapshot:

Median EV estimator from 2004 race

So conditions can certainly change.

However, this year’s race seems unusually stable so far. Since June 1 the Meta-margin has been Obama +3.18 +/- 0.99 % (mean +/- SD). That SD is less than half that of 2004 or 2008. So the campaigns are fighting it out over 2% of the voters (or equivalent variations in turnout).

It is for this reason that I find this prediction from political scientists at the University of Colorado to be unbelievable. For Romney to get 320 EV, as they predict, would require a swing of about 7% from current conditions. I am pretty sure that is not going to happen. If Romney wins, it will be by a whisker.

For your purposes, reader, the basic problem with nearly all political science-based models is that they can only reveal the general tone at the start of an election season. Once the campaign starts, polls are a direct measurement, and therefore a far better guide. You don’t need a weatherman to know which way the wind blows.

Tags: 2012 Election · President

15 Comments so far ↓

  • pechmerle

    Sam, I have confidence in your math. But what do you say to the U. of Colorado guys’ claim that their model has been correct in every presidential election since 1980?

    • Sam Wang

      I already responded above. Here’s another take.

      Here’s a question. Why give them any credence? Models of a future election season do not matter when you have actual measurements of opinion. Late August of an election season is not the time to haul out an econometric “model” that was fit to past events. Also..if their model were any good it would say that the 2000 election was basically a toss-up. In the Electoral College, that’s what it was. If they claim to have predicted that…it’s predicting too much.

      Generally one should think of a political science “model” as a research tool for discovery. In this respect they are more like weather researchers than weather forecasters.

      I’ll post on this later.

  • Matt McIrvin

    The Colorado model is a post-diction. It’s not hard to come up with a model that fits a lot of past presidential elections. It’s much harder to come up with one that fits before the elections actually happen.

    Lichtman’s “Keys to the White House” is another model that is claimed to have fit some large number of past elections. He predicted Gore would win in 2000 and then started emphasizing after the fact that he was predicting popular-vote winners (I wonder what he’d have said if the PV/EV discrepancy had swung the other way, as some predicted it would). But Lichtman’s already called 2012 for Obama.

  • The Political Omnivore

    What is the difference between the red and yellow bar areas in the “strike zone”?

    • Sam Wang

      Political Omnivore: They are one-sigma and two-sigma predictions. Approximately 68% and 91% confidence bands. Details are here.

      Matt: I agree with every point you have made.

  • Matt McIrvin

    A similar old game that is the identification of “bellwethers”, states that always vote with the winner. Missouri has been a bellwether for every presidential election since 1904, except for 1956 …and 2008. It’s not clear how much information this gives you. Similarly, political commentators used to play up the significance of the South Carolina Republican primary in picking the nominee, but it failed spectacularly in 2012.

  • Michael K

    Dr. Wang, have you looked at older elections (like 1988, 1992, and 2000)?

    In 1988, Dukakis reportedly led by double digits in March and May, by 6 points before the conventions, and by double digits at the end of July (after the conventions), before ultimately losing badly. Was there unusual volatility in the polls that year to produce a much larger sigma that might have foreshadowed the possibility of such a dramatic reversal?

    In 1992 and 2000, the popular vote also went to the candidate who (according to Gallup) trailed by 5+ points prior to the conventions.

    • Sam Wang

      Michael K: Yes, I have already analyzed that. Read my past posts on this topic. 1988 and 2000 were not re-election years. In re-election years, opinion moves much less.

      Also, you will recall that 2000 was quite close. Considering that the margin that year was smaller than the SD even for re-election campaigns, my calculation would not be very predictive. Nor would other “models.”

      As for 1992, your argument is false and/or misleading. After the third-party candidate, Ross Perot, dropped out, Clinton led G.H.W. Bush all the way.

  • Pat

    You are showing the win probability history for 2004 and 2008.
    Could you also show the same probability evolution for 2012 so far? If you can track this important parameter, it would be useful to see the evolution, like the EV or meta-margin evolution.

  • Matt McIrvin

    The line would be pretty flat, so far.

  • Brad Davis


    I always enjoy reading your election poll coverage, particularly with respect to the U of Colorado predictions. Philosophy (or political science modeling) is only useful in the absence of data.

  • Matt McIrvin

    I suppose that on the basis of past data, I can think of one possible objection to poll-based methods, namely that (like just about everything else) they underestimated the size of the Republican wave in 2010.

    While it’s about Nate Silver rather than you, I’ve seen people arguing that Romney is going to win by adding Silver’s 2010 error to current polling.

    Any further thoughts on where the discrepancy came from? Obviously midterm elections are different in a number of ways.

    • Sam Wang

      In 2008, polls alone did extremely well in predicting House outcomes.

      In 2010, his gap was between an actual gain of 65 seats vs. a predicted +55 seats, which is statistically notable and would give Republicans a 1-2% boost in margin, not quite large enough to give them what they want in 2012. I did about as well using polls alone. District-level polls did seem to be a culprit. There was a hobbyist who did better than either of us, using partisan voting index and campaign funding to fill gaps.

      However, to me this idea smacks of grasping at straws. 2010 was an off-year election. Voter intensity was high. Also, errors like that have not happened for the Presidential race. Or Senate races, for that matter. What you are citing sounds like the kind of desperate rationalization that occurred among Democrats during the 2004 Kerry-Bush race. See my original site for some prime examples of that. Not going down that road again.

Leave a Comment