Skip to main content

View Diary: Beyond the Margin of Error: Presidential Polls Are Almost Always Right, Even When They're Wrong (10 comments)

Comment Preferences

  •  Hmmm (0+ / 0-)

    Something strikes me as odd plotting error against the thing that there is error about. If the data were really random, R^2 would be 1.

    Obama <- rbinom(10, 500, .5)/500
    error <- Obama - .5
    plot(Obama, error)

    in R gives this.

    So, deviations from a perfect line indicate NON-randomness.... Seems odd.

    •  Hmmmmm (1+ / 0-)
      Recommended by:

      Don't know R - It seems you're using a binomial distribution to simulate the error of a bunch of polls of one contest assuming the real %Obama is 50%?  In that case, I see how give a perfect line of error vs % Obama.  

      Wouldn't it make a difference that this is 51 state contests with 51 values of Real Obama %?  Shouldn't we expect in a random universe, that the polling average from Utah might overestimate Obama's performance, while the polling average from Oklahoma might underestimate it, etc?  So that in deep red states, on average, the errors cancel each other out?

Subscribe or Donate to support Daily Kos.

Click here for the mobile view of the site