Look at the makeup of Republicans vs. Democrats in Newsweek’s poll from September 11, 2004: NEWSWEEK POLL: Campaign 2004.
391 Republicans (plus or minus 6) 300 Democrats (plus or minus 7) 270 Independents (plus or minus 7)
Compare against the same data from the new poll, which Newsweek is using to claim that Bush’s poll lead has “evaporated:” NEWSWEEK POLL: First Presidential Debate.
345 Republicans (plus or minus 6) 364 Democrats (plus or minus 6) 278 Independents (plus or minus 7)
Did Newsweek choose a lower percentage of Republicans for the first debate to set up Kerry’s “comeback,” or did they stack the deck with more Democrats in the second poll?
(Hat tips to all who emailed about this.)
UPDATE at 10/2/04 6:54:49 pm:
The loons at Daily “Screw Them” Kos are watching this topic: LGFers moan about Newsweek poll.
No.
No pollsters other than Rasmussen "choose" the partisan mix for their calling. They can't. The registration lists have no phone numbers on them.
The most they could do is call, ask, get the response and toss the results that are in excess of their desired mix.
But they don't do this because it would not make sense. Here's why. First, let's discard the conspiracy option. Let's presume they are trying to take a real measurement and not get a desired result.
If they want a real measurement they will sample at random and get the partisan mix that they get. It is not reasonable to conclude that people are changing partisan affiliation every 3-4 weeks (note heavy dem sampling in polls prior to late August, then heavy GOP Aug/Sept, now heavy Dem again). So the question is why do random phone calls yield such wide variances in partisan mix result.
For this Newsweek poll, the answer seems to be that they called only the west coast on Thursday night because the debate ended too late to call the midwest and east. That hugely emphasizes CA, OR and WA, all Dem strongholds.
They also called only Saturday morning because the poll was out mid-day yesterday. Thus, the bulk of calls were Friday night -- and how this skews is unpredictable.
Oddly, Rasmussen looks very good out of all this. He applies a desired partisan mix and does not vary it. His absolute numbers may not be valid because no one knows what mix will vote, but by holding it constant his trend data is the very best we have to go with. It has shown little change all throughout the campaign.