Apparently the ARG Poll which shows Kerry up by 1 point in Florida is also being disputed on Realclear Politics.
There's also this analysis:
The Sept 22 ARG Poll Analyzed
polipundit ^ | Sept. 23, 2004 | DJ Drummond
Yesterday, in the ‘comments’ section of another article, I noted ARG’s nationwide State Polling results (ARG is American Research Group), and wrote that ARG had ‘cheated’. That was a little bit unfair. ARG worked hard to produce a comprehensive poll for the Presidential Election in all 50 states plus D.C., and showed their demographics as well for each and every state. That is certainly worthy of praise, and attention. However, I would still have to say that ARG’s self-review before publishing, appears to be on a par with CBS News. I found what I can only call errors, unless I want to be mean and claim deliberate manipulation. Consequently, I took their numbers and adjusted them to match known and more reasonable demographics, which I will explain further as I go on. Here on Polipundit, I am publishing the overall results. Those interested in the individual states results, may find them over on my personal site, Stolen Thunder.
The first thing which jumped out at me, was the date range for this poll. While some states began polling as early as September 7, others began as late as September 17. Some finished their polling as early as September 9, while others finished as late as September 21. Some took 3 days for their polling, others as much as 6 days, and size was not a factor in the speed; California was the largest state, but only took 3 days, while Oklahoma took 6 days to cover.
There was also a clear fallacy in the gender voting. When I tallied the numbers, women made up 53% of the poll, even though they have never, historically, voted in such majority. A better balance would be 52-48 Women if you want to use the 2000 Demographics, and overall, a 50-50 match makes much better sense.
Race was not mentioned in the poll’s demographics. I expect that with a base response of 30,600 adults, race was discarded as a factor, but this further complicates a valid review of the stated results. Also, there is no statement about whether the respondents are Adults, Registered Voters, or Likely Voters, so from the size of the base and the lack of specific comment, I presume we are discussing Adults only. This is significant, because almost every poll has shown Bush collecting higher support as you move from Adults to Likely Voters.
Then there is the matter of the party alignment. When all 50 states and D.C. are tallied, ARG counted 41.4% Democrats, 35.5% Republicans, and 23.1% Independents, which is an unreasonable makeup, and clearly biased in favor of Democrats.
I took ARG’s numbers for each state and plugged them in. Using their numbers as they reported it, President Bush leads John Kerry by only 0.5%, 46.5% to 46.0%, but Kerry would win the election 271-252 EV (with two states too close to call).
I then went to my own table to Voter Registration (compiled by contacting the Secretary of State for each state keeping such records and noting their alignment as of December 2003, and for states which do not keep registration records by party, noting the party split in Federal Elections from 1994 through 2002), and plugged in those alignments, which balanced things a little better. Using the same party-vote numbers ARG presented, the new tally favors President Bush by 3.5 points, 47.9% to 44.4%, over John Kerry, but Kerry would still win the election, 291 EV to 247.
However, I was not happy with those numbers, not only because they were counting Adults overall instead of Registered or Likely Voters, but because I noticed their strength-of-support numbers seemed very low for what I had seen in internals. I attributed that to the respondent base, and tweaked the numbers to match more conventional strength-of-support numbers, as I have seen since July in every major poll citing that data (Pew, Gallup, IBD, CBS, Fox, Newsweek, Zogby), and with those punched in, President Bush leads John Kerry by 4.7%, 49.1% to 44.4%, and would win the election, 330 EV to 218.
So overall the ARG poll is slanted, but not horribly so. The data is interesting, even where it runs counter to the polls from other groups. It’s also interesting to see that the most significant change comes from noting strength-of-support, which has strongly favored President Bush all year. And finally, for those interested, I have put up the specific numbers for each state over on Stolen Thunder.
http://stolenthunder.blogspot.com/2004/09/september-22-arg-poll-dissected.html