Princeton Election Consortium

A first draft of electoral history. Since 2004

Meta-Margins for control: House D+1.0% Senate R+4.2% Find key elections near you!

Current single-poll median: Obama 280, McCain 258

August 20th, 2008, 9:10am by Sam Wang


A reader asked about a statement by David Gergen on CNN that based on [an average of ] the most recent polls, McCain would win the Electoral College. Using only the most recent available poll in each state (as opposed to the last three polls, our usual top-line estimate), assigning every state to its more likely winner gives total EV counts of Obama 264, McCain 261, tie 13. This is the mode of the distribution, and is the closest result that I can get to what Gergen said. However, even that’s not quite right…

The meta-analysis, which is probabilistic and therefore takes into account the uncertainty of polls, gives a current median outcome of Obama 280 EV (95% CI 251 to 320 EV), McCain 258 EV, Meta-Margin 0.94%. So it appears that the more likely winner today, by a hair, is unchanged. However, a change could still happen.

[Thanks to reader Jeff for the correction above. By the way, you may mock pundits for merely averaging polls state by state, but in 2004 they weren’t even doing that. Small steps…]

Tags: Uncategorized

5 Comments so far ↓

  • Jeff MacSwan

    Gergen’s actual remark was that an “average of state polls” for the first time show that McCain would win if the election were held today (or yesterday, I guess, when he said it). Sam has commented that the average of polls is less accurate than the meta-analysis. Perhaps an opportunity to make the point clearer?

  • J.W. Hamner

    I love that you report confidence intervals. Seriously. I hardly see it in people who purport to do science, so it’s quite refreshing to see it in someone examining politics.

  • Frank

    Since I appear to have started this, I should clarify.

    Gergen’s Aug 19 remark was this:
    “For the first time, if you look at an average of state by state of the polls, for the first time, today, if you look at the Electoral College, if the election were held today, again to an average of state polls, for the first time, John McCain would win.”
    http://transcripts.cnn.com/TRANSCRIPTS/0808/19/acd.01.html

    On the previous day he wrote this:
    “A web site that averages all significant polls, RealClearPolitics.com, has previously projected that just looking at polls, Obama was ahead in states with over 300 electoral votes; now he is down to 275 — a tiny cushion since 270 is the magical number for winning.”
    http://ac360.blogs.cnn.com/2008/08/18/is-the-tide-turning/#more-5761

    I believe he was referring in both cases to this, which today has McCain up by 10 electoral-college votes:
    http://www.realclearpolitics.com/epolls/maps/obama_vs_mccain/?map=10

  • MichaelL

    Can you somehow measure the volatility in poll shifts over time and factor that into your prediction? So your moving confidence interval would reflect the decreasing volatility as we approach election day. The “if the election were held today” criteria doesn’t really tell us much when we are weeks away from polling day.

  • Sam Wang

    MichaelL, as I have written previously, one could, but it adds considerable uncertainty and is in my view not useful. If you like that kind of thing you can try fivethirtyeight.com and stochasticdemocracy.blogspot.com.