Comparing the election results to the polling results saved at
http://aics-research.com/election2004july.html
(I just eyeballed this quickly, so I might have a mistake.)

States that voted differently in Nov from how they polled using
traditional polling methods in July:

    AZ IA MO NM NV OH
    and TN was tied at that time

AOL Straw Poll
States that voted differently in Nov from what the AOL straw poll
indicated in July:
     AK AR AZ CO FL IA KS KY MO MT NE NM NV OH TN VA WV WY

    That is, all of the above, plus :
   AK AR CO FL KS KY MT NE VA WV WY



Wirt Atmar wrote:

>Because of the previous discussion on the validity of polling techniques, a
>few days ago I gathered up the predictions of four polls, the first of which is
>text-only and the remaining three graphs, each predicting the outcome of the
>coming election as they see it now at the end of July. The four polls are at:
>
>     http://aics-research.com/election2004july.html
>
>The first poll, as reported by the AP, gives current favor to Bush, based on
>the number of electoral votes that Bush seems to clearly have in his column.
>However, in this predictive poll, neither candidate has enough certain votes to
>pass the 270 mark necessary to win.
>
>This poll could easily be described as the wimpiest of the polls, and I'm not
>at all certain that they have it right.
>
>The second "poll" is not a poll at all. It's merely a historical survey by
>the NY Times on how the various states have voted in past elections. It's
>underlying thesis is that past inertia will dictate a similar voting pattern this
>year as well.
>
>In physical systems where inertia is high, this sort of analysis works
>surprisingly well. Indeed, you can predict tomorrow's weather with a high degree of
>certainty if you simply take the average (and calculate the standard
>deviation) of today's and tomorrow's temperatures for the last 100 years. Tomorrow's
>temperature will be very close to tomorrow's 100-year average. Your estimate can
>be even improved if you take into account today's excursion from today's
>average as a fraction of today's standard deviation and apply that percentage x
>ca. 0.8 for tomorrow (it's presumed that today's excursion will be diminished
>tomorrow as the world returns to more normal conditions).
>
>No weathermen, no synoptic views, and no satellites are necessary for this
>form of prediction. Just local conditions recorded for 100 years. And the
>process works surprisingly well because of the thermal inertia inherent to the
>Earth-Sun system.
>
>However, this form of prediction has no capacity to predict the onset of
>significant future excursions from the norm, such as blizzards, massive warm
>fronts, hurricanes, etc., and that's the reason that it's not used. The NY TImes'
>prediction suffers from the same fault.
>
>The third poll is the one recommended by Ken Hirsch and is a complilation of
>a lot of small, statewide or regional polls conducted by "professionals." This
>poll has Kerry beating Bush by 310 to 217, with Tennessee being too close to
>call, thus the total reported is shy 11 electoral votes.
>
>The fourth poll is a "self-selected" AOL poll. Surprising to some, it's the
>one that I trust the most. At the time of my recording the graph, 130,000 have
>voted. In the states that Kerry leads, he leads decisively. In the majority of
>the states in which Bush leads, he often leads by the slimest of margins,
>although there are exceptions to that statement. In Alabama and Mississippi, Bush
>beats Kerry by 2:1 in the popular vote.
>
>As an aside, you can see the effect of the electoral college in the vote
>totals for each state. In California, with 55 votes, 14,400 people have voted this
>month, with Kerry ahead by a 2:1 margin. In South and North Dakota, each with
>3 votes (the minimum possible, one for each US senator and their one US
>congressman), only ca. 130 AOL members in each state voted, and in the ratio in
>both was ca. 50:49 in favor of Bush. Nonetheless, those 130 people's electoral
>votes weigh much more heavily than the 14,400 Californians' do [130/3 votes =
>43/vote is not equal to 14,400/55 votes = 261/vote]. Indeed, at this ratio, a
>Californian's vote is only worth 1/6th that of a South Dakotan's.
>
>What makes any of these polls accurate (or not) is whether the people
>participating are representative of the people most likely to vote. Although saying
>anything beyond just marking these data at this point in time is fruitless, we
>will nevertheless be able to look back in three months and see which of the
>"polls" was the most accurate.
>
>Wirt Atmar
>
>* To join/leave the list, search archives, change list settings, *
>* etc., please visit http://raven.utc.edu/archives/hp3000-l.html *
>
>

* To join/leave the list, search archives, change list settings, *
* etc., please visit http://raven.utc.edu/archives/hp3000-l.html *