Those Polls

From The Corner on NR:

From today's White House Bulletin... [Rich Lowry]

a point that has been noted in this space before:
* Recent Polls Outside The Historical Norm For Party ID. A spate of recent polls paints a very gloomy electoral outlook for GOP candidates in next month's elections. One reason for that, possibly, is a set of samples in recent polls that do not mirror the historical norm for party ID. A memo circulating among Republicans on the Hill, authored by GOP pollster David Winston, takes a look at the historical spread between Democrats and Republicans in House elections and polling over the last 14 years. According to Winston's analysis, there is a material discrepancy between the party identification listed by people in exit polls (people who actually voted) between 1992 and 2004, and those used over the last few weeks.

In most of the years between 1992 and 2004, Democrats held a slight advantage in party ID. Winston based his data on VNS/Media exit surveys, and concluded in 1992, Democrats held a 3 point advantage; in 1996, they held a 4 point advantage; in 1998, a 1 point advantage; and in 2000, a 3 point advantage. In two election years, 1994 and 2004, the percentages of people identifying themselves as Republicans and Democrats were identical, i.e., no advantage to either party. 2002 was the only year in which Republicans held an advantage over Democrats, with 40% identifying themselves to exit pollsters as Republicans and 38% identifying themselves as Democrats.

In short, between 1992 and 2004, only once did one party enjoy an advantage as large as 4 points over the other in party ID. But in recent polling samples used by eight different polling organizations (USA Today/Gallup, CBS/NYTimes, ABC/Washington Post, CNN/Opinion Research, Newsweek, AP/Ipsos, Pew, and Time), the Democratic advantage in the sample surveyed was never less than 5 points. All these organizations conducted surveys in early October. According to Winston, the Democrats held the following party ID advantages in these early-October surveys:

* USAToday/Gallup: 9 points.
* CBS/NYT: 5 points
* ABC/WP: 8 points
* CNN: did not provide sample party ID details.
* Newsweek: 11 points.
* AP/Ipsos: 8 points.
* Pew: 7 points.
* Time: 8 points.

Party registrations shift over time, and many political operatives believe the country starts to gravitate away from a party that has been in power over an extended period of time. Republicans have controlled the House since 1995. Winston acknowledges that possibility in his memo, writing, "It is certainly not out of the realm of possibility that this year's election could fall outside of historical results, but any survey that does should acknowledge that the data presented are based on a foundation that reflects a structural shift in the way the electorate identifies itself with a party."

Here's more from Big Lizards:

How much to weight for party ID is a weighty question for a very weighty reason: if poll samples consistently come up with significantly more Democrats and Independents than voted in the last comparable election (and consequently fewer Republicans), does that mean that a bunch of registered Republicans now consider themselves more in the Independent or Democratic camps -- hence will vote that way -- or does it mean there is an unidentified but systemic bias in the sample selection that will disappear when voters actually go to the polls?

In other words, should polls be weighted to "correct" the typical "oversampling" in favor of the Left in the pool of Rs, or does that supposed oversampling actually reflect true voter intent -- hence should not be eliminated by weighting?

And there is a related question that even further complicates the situation: assume some number of Republicans are mad at the party, so when asked their party affilliation, they say "Independent" or even "Democrat," and when asked who they will vote for, they say "Casey." What percent of them will, in the end, come back to the fold and vote for Santorum, even if they must hold their noses while doing so? After all, if you believe that a person will "switch" his party affilliation one direction, then he could jolly well switch it back in the voting booth, too.

The reality is that the percent of overpolled Democrats and Independents who are in fact "false-flag" voters -- voters who say they're one party while actually being another -- is neither 0% or 100%; nor will all the false-flaggers actually vote for Democrats: