Skip to main content

View Diary: Pollster Report Card (42 comments)

Comment Preferences

  •  A+ for effort... less for accuracy (9+ / 0-)

    Error alone doesn't mean anything.  Error compared to other error doesn't mean anything.  Your model, to actually have mathematical accuracy, should include things like standard deviations away from the correct answer and alpha (e.g. a measure for outperforming other pollsters).

    In addition, to be accurate, you'd have to account for the fact that different pollsters might consistently be accurate in some states and consistently suck in others.  For example, the Q poll might be great for Connecticut, but not for Iowa.  And there just isn't a big enough publicly realized sample size to be able to come to these conclusions.

    But your effort in these diaries is always great.  So thanks.

Subscribe or Donate to support Daily Kos.

Click here for the mobile view of the site