Skip to main content

View Diary: The Meta-Analysis Of State Polls is back! (w/poll) (31 comments)

Comment Preferences

  •  I didn't give odds (0+ / 0-)

    but I did call each race.

    •  Specific calls rely on chance (2+ / 0-)
      Recommended by:
      plf515, SciVo

      I'm impressed. However, that year several races were so close that calling them correctly required a degree of luck.

      This brings us back to the point meta-analysis: Even when it is not possible to predict the outcome of a particular state with confidence, it is still possible to predict the total number of electoral votes with much higher confidence.

      •  That statement (1+ / 0-)
        Recommended by:

        relies on independence of events.

        The interesting question here is "what events have to be independent?"  I think, from what I've read of your method, that the independence is required in "gap between polling average and actual result", but am not sure if that's between pollster (e.g. Rasmussen's error is independent of SUSA's error, using 'error' in its technical sense) or between states (error in NY is independent of error in PA).  Perhaps both.

        And both are possibly not independent.  I've looked at a few pollsters, and all seem to be using racial breakdowns based on 2004 (that is, they think the electorate will be just as 'Black' and 'White' as in 2004).  This seems unlikely --- Obama will increase Black turnout.

        And errors are likely to be correlated regionally, too, I think

Subscribe or Donate to support Daily Kos.

Click here for the mobile view of the site