Skip to main content

View Diary: Election Predictions Based on Past Polling Errors (12 comments)

Comment Preferences

  •  The problem (1+ / 0-)
    Recommended by:

    with all of this is the lack of data frpm two cycles with two large polling misses 1996 and 2000. In 2000 Nine states were off by 4 or more percent including some as many 7. This means the data set for most analysis is actually not reflective of true volatility.  The problem is most don't have state data before 2002. It took me months to get state level data for 2000 1996 and 92. When you get that data you realize the fundemtal mistakes that their data create

    True volatility is farhigher than the 2002 to 2010  would lead you to believe
    I  poll watching today so I have an excuse for typos thtime

    The bitter truth of deep inequality has been disguised by an era of cheap imported goods and the anyone-can-make-it celebrity myth - Polly Toynbee

    by fladem on Tue Nov 06, 2012 at 09:49:41 AM PST

    •  Well, I'd like to see (0+ / 0-)

      The polling for 2000 would be good to include, simply to increase the amount of data being used.

      I wouldn't want polling for 1996 or 1992 because the three-way races introduce a huge variable into the polling that is not present in most elections.

      For that matter, I have a hunch that 2002 had a lot of one-time only trends due to 9/11. Remember, 9/11 "changed everything" -- at least in the votes in the next election, when Democrats took a hit in a mid-term election, greatly violating the historic record. So I'd venture that 2000 election data is 'more normal' than 2002 data.

Subscribe or Donate to support Daily Kos.

Click here for the mobile view of the site