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

New rules

September 30th, 2008, 9:26pm by Sam Wang

As we head into the last five weeks of the campaign, polls are coming more frequently. In response, we are making two changes to the Meta-Analysis: (a) the averaging rule and (b) the update schedule.

In 2004, the median Electoral Vote estimator got noticeably spiky in October. The following graph uses the last-3-polls rule alone:

Median EV estimator from 2004 race

in large part because individual polls passed through the average quickly.

At the very end of the campaign we made a last-minute switch to a last-three-polls or last-7-days (counting back from the date of the most recent poll), whichever gives more data for a state. This performed extremely well on Election Eve 2004, nailing the result. We’ll start using the new rule tomorrow morning. The rest of the procedure – from calculating the median and SEM onward – will stay the same.

To reflect the posting schedule of our data partner, the EV estimator will now be updated four times a day at 8:00am, noon, 5:00pm, and 8:00pm. We will stop doing the midnight update since Eric Dienstfrey and Mark Blumenthal haven’t been putting up data at night. (Good for them – they need their rest.)

Tags: 2008 Election · Meta-analysis · Site News

9 Comments so far ↓

  • David

    Sam, I don’t understand. You say the new 7-day rule worked very well in 2004 towards the end, but the 2004 EV graph looks just as spiky even as we approach the last couple weeks of the election. Why does the 2004 EV graph look even more spiky at the end?

    Also, did you know sometimes users are unable to add comments? The link to post a comment clicks to the post but without textboxes…

    With that said I’d like to make some predictions:

    1) The VP debate will catch Democrats off guard just like Palin’s speech at the RNC convention. I’ve seen youtubes of her previous debates. She can be good when she needs to be. Plus her expectations are so low she could say anything and it would be passing.

    2) Reverend Wright has not yet been unleashed in the general election. But he will, what with McCain tanking in the polls. The Republicans…are waiting…

    3) I bet you the last two weeks of this election and maybe even afterwards will be an all out WAR. The Republicans – especially the administration – know that they could potentially be up for criminal charges when they’re not in power anymore. They will do anything…and I mean ANYTHING…to win this thing.

  • Michael S

    You should have either an RSS-updating post or an email blast that reflects the new numbers with each update.

  • Richard

    Gott and Colley used a one month window for averaging polls in 2004 and predicted every state correctly except HA, which had a paucity of polls in the last month. The length of time used for averaging polls seems quite arbitrary. What window ends up working well no doubt depends on how volatile a race is, but it seems volatility can’t be known a priori. Does the averaging rule amount to anything more than a wild guess?

  • Joe

    I’ve only been coming to this site for a few days and have limited experience with statistics. But am I reading the electoral vote histogram correctly, that today’s polls show there is no possible outcome in which McCain would win? Has that happened before?

  • Sam Wang

    David, the post is clarified now.

    Richard, anybody can calculate averages, which I advocate (of course) but is not novel. Gott and Colley are among the many people who noticed this. Also, in close races (such as Wisconsin in 2004) an element of chance is unavoidable – they got lucky. The Meta-Analysis takes such uncertainty into account to arrive at an accurate overall result.

    In 2004 there seemed to be a measurable event-associated swings in the EV estimator with an average interval of 25-30 days. See this chart. Therefore a suitable interval would be less than that. I chose 7 days, which worked well in 2004. I might tweak it to 10 days, but probably not longer.

    Joe, look at the history plot. Whenever the gray band does not include 269 EV, the win probability is 97.5% or greater. There are possible wins for McCain but they involve a very unexpected combination of polling errors. A more useful quantity is the Popular Meta-Margin (see FAQ for explanation).

  • Sam Wang

    Michael S (and everyone else), we’re not sure how much of a tolerance people will have for tiny updates being beamed at them up to 4 times per day.

    However, we do have a robo-email that is sent every time the top-line estimate changes. If you want to be on that list, contact owsla [at] princeton [dot] e d u.

  • Reid

    I completely agree with David on point 1).

    Palin will win the debate by default. It is set up to play to her strengths (presenting platforms that have been written for her) and avoid her weaknesses (direct confrontation and responses that make her adlib, and demonstrate deeper understanding).

    Furthermore, democrats will dig deeper by claiming she memorized responses lending credence to the sexist-liberal-news-media isn’t giving her a shake argument.

    I expect a bump similar to the one after the RNC speech.

  • Sam Wang

    To reiterate: comments are moderated. Comments should be constructive and have a convincing name and/or email address.

  • Sam Wang

    Reid, it’s hard to imagine a bump for McCain as large as the RNC, which came with a considerable amount of excitement about his VP pick. But at this point the Palin brand is irretrievably damaged. For example, George Will said recently that Palin is “obviously not qualified to be President.”

    I agree that she could avoid major landmines tomorrow night. If she clears this rather low bar for performance, the talk about dropping her from the ticket could diminish.

Leave a Reply to Richard (Cancel)