http://courses.ttu.edu/hdfs3390-reifman/ weighting.htm

Over on my Party ID/Sample Weighting website, I've added data on national polls' partisan composition (to go along with my previous data on Party ID in key states).

My average for today is that self-identified Democrats comprise 38.7% of national samples, whereas Republicans comprise 31.5% (a difference of 7.2%).

This is similar to Rasmussen's latest weighting template (D +6.7%), which is determined from a six-week rolling average of his own polls.

Over at my sample-weighting website, I've created a chart that allows visitors to see "at a glance" some estimates of Party ID composition of various key states.  That way, when a poll result seems odd, you can compare the Party ID composition of the poll (if provided) with some benchmarks.  

http://courses.ttu.edu/hdfs3390-reifman/ weighting.htm

For example, in today's "shock" poll result of Obama leading McCain by 8% in West Virginia, the partisan composition is 55% Democrats, 35% Republican.  As seen on my chart, those Party ID figures are not out of line.

[Cross-posted at Daily Kos]

Now that the general-election campaign is underway, my Party ID/Sample Weighting website is up and running again.  My first entry of the year looks at the newly released Obama-McCain Wisconsin poll.  As we know, many polls over the years appear to have oversampled Republicans (especially in 2004), but this new Wisconsin poll appears to have oversampled Democrats (though no one can know for sure, at least until November).

http://courses.ttu.edu/hdfs3390-reifman/ weighting.htm

There's an excellent diary at Kos, where someone provides maps of what he/she considers some of the most contestable districts:

http://www.dailykos.com/story/2004/11/17/14513/997

I added the following comments...

I wrote this at Kos and thought I'd share it here, too:

http://dailykos.com/story/2004/10/28/132557/30

There is a very simply lesson in all of this. If we get the turnout, we will win--Chris

Steve Soto at The Left Coaster is on the beat and, yes, Gallup oversampled Republicans again (what a shock?).

In fact, based on some D-R-I internals I saw at Daily Kos, Kerry would be winning under a turnout model of the 2000 exit polls:

Bush (.39 X 9 among D's) + (.35 X 93 among R's) + (.26 X 44 among I's) = 3.51 + 32.55 + 11.44 =

47.5

Kerry (.39 X 88 among D's) + (.35 X 5 among R's) + (.26 X 47 among I's) = 34.32 + 1.75 + 12.22 =

48.29

Click here for my sample-weighting website.

Mark Blumenthal over at Mystery Pollster recently wrote up a large piece focusing on the distinction between the Washington Post and ABC polling results, even though the two media outlets work closely together.

In explicating the WP/ABC methods, it became clear that those methods are highly similar to Gallup's.  WP/ABC uses a series of questions similar to Gallup's to determine likely-voter status and, though WP/ABC "partially" weights on party ID, there's still room to have a large R/D discrepancy from previous presidential exit polls.  

Thus, it's no coincidence that the latest WP (Bush up 51-45), ABC (Bush up 51-46), and Gallup (Bush up 52-44) polls have similarly favorable readings for Bush, even though many other polling outfits have the race either tied or with Kerry slightly leading.

http://www.pollingreport2.com

To see Blumenthal's original report and a comment of mine where I address these issues in greater detail (posting as Alan R.), go to:

http://www.mysterypollster.com/main/2004/10/one_less_myster.html

The following article presents a nice overview on pollsters' inability (by law) to survey people over cell phones and the potential implications for survey results.  

http://abcnews.go.com/Politics/wireStory?id=159178