Subscribe via RSS Feed

Election Prediction

[ 27 ] October 3, 2012 |

This is the time of the season where many if not most of us pore over the various state level election tracking sites available.  Electoral Vote was a big one in 2004, and remains my favorite for daily poll releases; 538 in 2008 and again now under the New York Times in 2012.  One I wasn’t familiar with, inexplicably, until yesterday is Princeton Election Consortium.  I’m assessing the distinctiveness of this one when I have time, which means I haven’t accomplished more than a superficial perusal.  Their current analysis is similar to Silver’s: there appears to be a slight erosion in support for Obama in national polls that is not replicated in state level polls.  The model itself seems far less sophisticated in terms of variables, but this is not necessarily a bad thing in a purely predictive exercise.

It’s worth a look.

h/t Jeffrey Dudas

UPDATE: link fixed.  I hope.

Share with Sociable

Comments (27)

Trackback URL | Comments RSS Feed

  1. Glenn says:

    FYI, for me that Princeton link wouldn’t work until I took the “www” off. Don’t know if that’s just my ISP or what.

  2. Jameson Quinn says:

    Any good House predictions/models out there?

    • Anderson says:

      Here is a good House prediction: we don’t take it back.

      • dave says:

        Wang thinks the House is a coin flip with Dems having a very slight edge. He does note that the lack of polling in House races makes predicting the House more difficult.

        • Ian says:

          Wang is being insufficiently complex in this regard. As Dave Weigel pointed out recently, one has to take redistricting into account. We’re not taking it back until the demographics change again.

          • Craigo says:

            While I respect Weigel as a reporter, I have to ask – where’s the math? historically, what sort of popular vote edge is needed to take the House? How much of an advantage is the redistricting? Is the first number larger or smaller than the second?

            When he does his homework I’ll take him seriously on this issue.

            • Ian says:

              Obviously there is some level of popular vote at which we would take back the House. But it’s not clear to me that it can be effectively modelled by historical comparisons.

              The aggressive redistricting following the 1990 census ushered in an era of Republican dominance, but by the Bush administration demographics had shifted enough that many seats were more or less at parity–thus, a Democratic wave was enough to gain a large majority, and a subsequent Republican wave was enough to get the House back. But now they’ve rejiggered it again, with the intention of securing the gains of 2010.

              If we’re going to compare the post-2010 map with the pre-2010 map, we need to go district by district to find the new partisan lean, and calculate the new overall lead necessary for a Dem victory. But why bother? As far as I can tell, every predictor who is looking at the race district by district thinks it’s hopeless for the Democrats.

              • Njorl says:

                As far as I can tell, every predictor who is looking at the race district by district thinks it’s hopeless for the Democrats.

                The problem with that is their predictions would present a result which is utterly at odds with history. Assuming toss-ups split roughly 50/50, people are predicting a 40 seat Republican advantage while Dems win the cumulative house ballot by about 2 points. That is not going to happen. Something is clearly wrong with the polling.

        • thusbloggedanderson says:

          Wang is drinking happy juice. Time will tell.

          RCP sees 26 tossups. Give ALL of those to Dems. That gets us to 209 vs. 226. I know RCP tilts red, but still, I’m not buying Wang here. Neither is Drum.

  3. J.W. Hamner says:

    The model itself seems far less sophisticated in terms of variables, but this is not necessarily a bad thing in a purely predictive exercise.

    This is precisely backwards. Sam Wang’s model is much simpler and more transparent than Silver’s, but the parsimony of the former, in fact, is what makes it vastly more sophisticated than the latter.

    • mpowell says:

      You cannot just change words to suit your meaning. It could only be more sophisticated in this sense if it relied on an improved fundamental insight that rendered additional details unecessary. But it’s simplicity is hardly proof of this. The question is how well it actually performs and since it is untested on out-of-sample material, we don’t know yet.

      • Craigo says:

        No, he’s correct. Silver’s model makes two mistakes – he overfits (seven econometric variables, seriously?) and overweights (if the econometric variables are affecting public opinion, then it’s already priced in). This is odd, as Silver has previously warned of the danger of overfitting and the general uselessness of econometric models. He is likely including a lot of noise in his dataset, which is why his numbers are relatively volatile compared to models that don’t use econometric data or at all (Sam Wang) or only as a prior (Drew Lizner).

        Wang’s model performed very well in 2004 and 2008, and actually came out ahead of 538 on a few counts four years ago.

        • thusbloggedanderson says:

          overweights (if the econometric variables are affecting public opinion, then it’s already priced in)

          Don’t agree. Economic pain can take a while to sink in and affect people. And IIRC, the closer we get to the election, the less Silver counts his econ variables.

      • J.W. Hamner says:

        All other things being equal, simply including more parameters (even if the inputs are noise) will increase a model’s fit. That’s why when choosing between two distinct models you use a a criteria that penalizes for additional complexity (for the technically minded: AIC). Given that they both did about the same in 2008, but Wang’s model uses only polling data, his model is clearly superior.

        • thusbloggedanderson says:

          Evaluating on the basis of a single election?

          • J.W. Hamner says:

            Does Silver have some secret record of electoral prediction success that I’m not aware of? Neither Wang or Silver are going to be wrong unless the “unskewed polls” guy is right, and there is some massive systematic bias in the polls.

            You can favor whatever model you like, I am just speaking out against the common misconception that the model with the most parameters to tweak and proprietary knobs to twist is the most “sophisticated”… when really the opposite is true.

            • dave brockington says:

              I didn’t intend for my use of “sophisticated” as a word to have normative implications, which several comments seem to assume. It’s a statement of fact, and in order to deflect any judgment assumptions that might be read into it, I added the clause that said something like ‘this is not necessarily a bad thing . . . ‘. FWIW, I have a strong preference for parsimony in modelling, when possible, and I teach the same.

              • dave brockington says:

                To wit: according to Websters, one definition is “highly complicated or developed”. That’s not a judgment, that’s simply empirical fact. If two models, one parsimonious, one “sophisticated” in the sense I used the word, reach the same conclusion, the preference should always be for the former.

  4. [...] At the time I followed Electoral Vote, and Pollster (now part of HuffPo).  The 2008 cycle had me following Nate Silver at 538 (and now at the NYTimes).  TPM’s poll tracker is pretty good, too.  Of course, if reality isn’t your thing, you could also read UnskewedPolls.  Inspired by this post at LGM. [...]

  5. DocAmazing says:

    Guess we’ll find out in a few weeks…

  6. Kori Nevis says:

    Bed Bugs Chicago, Treatment of Bed Bugs Chicago, Removal of Bed Bugs Chicago Don’t stress about those bed bugs anymore! Contact the experts at http://www.bedbugstreatmentchicago.com/

Leave a Reply

You must be logged in to post a comment.