Is Barbaro the guy who ran against Koch for mayor in 1981?

Chris,

Your figures on party ID are annual averages, so they don't show what happened over the course of a year. Party ID can shift substantially over the course of a year, especially an election year.  The NCPP points out that over the course of 1996 the Democrats went from even to 10 points up in party ID.

http://www.ncpp.org/weight_data.htm

National Council on Public Polls'
Polling Review Board

THE GOOD AND BAD OF WEIGHTING THE DATA

"Consider the change in party identification from the Pew Research Center polls throughout 1996. In the beginning of the year the Republican-Democrat split was 30%-30%. On Election Day it was 26%-36%. The number who considered themselves Republicans went down steadily the closer the survey was to the election. Using party identification to weight just the likely voters in a political poll is little better than a guessing game where the pollster is substituting his or her judgement for scientific method."

No pollsters call cell phones because it's illegal.

Still, I very much doubt that the 5% of households with cell phones instead of land lines will make much difference to the accuracy of polls.  

Cell phones are just one of many problems causing polls to get low response rates.  And polls have continued to do quite well in spite of very low response rates.

I just put up a post about this on my blog:
http://ragout.blogspot.com/2004/09/cell-phones-killing-polls.html

Notice that Gallup almost always has the lowest percentage of undecideds/Nader voters (highest percentage of Bush+Kerry).  This holds true until the most recent poll.  Gallup probably pushes the undecideds harder to get them to report which way they're leaning than the other polls.  

I believe that leaners tend to break towards the incumbent in polls, even as they give Bush a low job rating, think the country is going in the wrong direction, and even if they break towards the challenger in actual voting.

There are now seven polls in the last two weeks that  either reported Party ID or had someone back it out.  They show the Republicans with an advantage averaging 4.6 points.  

Newsweek  Rep + 4  9/9-10

Dem Corps Rep + 0  9/6-9

CBS       Rep + 1  9/6-8

ABC       Rep + 6  9/6-8

Gallup    Rep + 10 9/3-5

Time      Rep + 4  8/31-9/2

Newsweek  Rep + 7  8/31-9/1

So there's no doubt in my mind that the Republicans have an advantage in Party ID at the moment.  It's not sampling error and it's not bias due to the particular days the polls were conducted or the incompetance of particular pollsters.  

The implication of the argument that Party ID is very stable is not that this shift should be ignored.  If party ID is usually very stable, Bush  is going to win.

Personally, I think there's a good chance that this shift in party ID is ephemeral, part of the convention bounce.  Is party ID stable or not from month to month?  I've hunted around for info about Party ID before and after conventions, without much success, except that Democracy Corps found little change in Party ID following the Democratic Convention.  The Dems gained something like 4 points in Party ID over the course of 1996 according to Pew:
http://people-press.org/commentary/display.php3?AnalysisID=95

Is there a good study of Party ID from month-to-month that would say whether shifts like this are likely to persist or whether blips are common?  I haven't been able to find one.  I think political scientists are mainly interested in long term trends in Party ID over years and decades, and that the short term dynamics haven't really been studied.

There are now four polls indicating a recent surge in Republican ID (Time, Newsweek, ABC, Gallup) so I think this has really happened.  That surge, combined with some polls that weight by party ID and others that don't, explains a lot about why the polls have been so crazy recently.

See: http://ragout.blogspot.com/2004/09/bush-fires-up-base.html