On Thu 24 Sep 2004, Seth Ackerman wrote:
> The obvious solution is to aggregate the party ID responses from all the
> recent polls other than Gallup and CBS and then compare them with
> Gallup/CBS.
There's actually a simpler way: use the CBS numbers. They haven't yet posted the most recent poll, by they have the last one up on the web:
http://www.nytimes.com/packages/html/politics/20040717_poll/20040717_poll_results.pdf
And if you go to page 19, you'll see that they've been doing a tracking poll of party ID every month since 1992. It does vary from month to month, but one thing is very clear: not *once* in the last 12 years has this poll [FN 1] shown more Repugs than Dems. And if you want to go a little bit out on a limb, it looks to me as if, under the variation, there is a remarkably stable distribution it keeps returning to, which is roughly
29% Rep, 35% Dem, 31% Independent and 6% Other
BTW, one thing is worth noting in passing: the question asked here (and I believe this is true of almost all such polls) is not "Are you a registered Repug/Dem/Independent/Other? but "What do you *consider* yourself?"
The CBS question is:
<quote>
Generally speaking, do you usually consider yourself a Republican, a Democrat, an Independent, or what?
<unquote>
which makes sense when you think about it. If we were talking about actual factual registration, you wouldn't get the majority being Independent and Other. Most people who call themselves indies are actually registered with a major party.
It follow from this that the success of registration drive efforts aren't really relevant for explaining variation in this number. You could register people and they could call themselves independent even if they were voting for you.
(One last note: when pollsters talk about "independents" they thus don't mean people who haven't made up their minds. They mostly mean the centrists in both parties -- people registered with one of the major parties who consider voting for the other party a reasonable thing to do.)
> If an aggregate sample of, say, 10,000 respondents gives a party ID
> split in the ballpark of the 2000 VNS numbers while Gallup/CBS gives far
> more Republicans, then you can assume there's something wrong with
> Gallup/CBS.
If there is something seriously non-random and biased in the sampling process, it'll produce the same problem with a bigger sample. Theoretically it's almost entirey the process that gives you the right results, not the size of N (once you've gotten past the minimum).
I feel I should mention that Teixiera was probably being telegraphic when he said the party ID numbers today were probably the same as the 2000 VNS numbers. I'm sure he didn't mean to say what he implicitly seemed to be saying, i.e., that they haven't changed at all in the last 4 years. Because they do vary month to month, and he's very conscious of that fact, because he's written a lot about the controversies that have arisen from it.
So it certainly is possible that there has been a radical shift in what people consider themselves as. It just doesn't fit any known historical pattern in this poll. And it doesn't make sense.
Michael
FN -- The Pew poll did show Repugs as a majority in 2002, in the wake of 9/11 and in the surge of the 2002 elections. And it was news precisely because it hadn't happened before. But it was also an outlier, and Pew too returned to the standard distribution and has remained stable there.