Skip to main content

View Diary: What's an Obama State? With February predictions. (242 comments)

Comment Preferences

  •  two many variables (3+ / 0-)
    Recommended by:
    kyril, zackamac, EquationDoc

    As a matter of statistical practice, it is a very bad idea to try to predict 26 outcomes with as many as 9 variables.  Of course you can get a great "fit" (explaining 95% of the variance in the data), but what you are really doing is "over-fitting."  With such a very small sample, part of the apparent fit is due to chance, not systematic connections between the variables.

    Still, I appreciate and enjoy the spirit of the analysis; it is better than the "data fits" that folks run "in their heads" and then spout off about -- at least there is some actual statistical method here, albeit pushed farther than I think is plausible.

    •  He could eliminate over-fitting by... (1+ / 0-)
      Recommended by:

      now going back and eliminating the other 17 and only considering the 9. I assumed he'd done that but upon re-reading I don't think he did.

      It would be interesting if the WA, LA and NE data reduce the importance of any of those 9, or if any of the other 17 increase in importance.


      I'm not voting for Barack Obama because he's black. I'm voting for Barack Obama because he's brilliant. --Oprah Winfrey

      by EquationDoc on Sun Feb 10, 2008 at 12:15:09 AM PST

      [ Parent ]

Subscribe or Donate to support Daily Kos.

Click here for the mobile view of the site