Here, I brought you Sean Trende's analysis that relative turnout of Republicans, Democrats, and Independents (RD&I) had shifted dramatically in the Virginia and New Jersey governors elections last year, and in the Massachusetts Senatorial election early this year from that in the Presidential election of 2008. You will recall that Trende applied those shifts to explain the troubles the Democrats face in Senate races across the country, and predict, for example, the very popular Joe Manchin's deep trouble in the West Virginia special election to replace Robert "King of Pork" Byrd.
So one problem in understanding the polls this year is to remember that the usual weighting of RD&I segments in them is wrong. But that problem is compounded when the poll includes all adults, or all registered voters. Who will vote? That's the question on which Trende's analysis focuses. A polling organization that reports the opinions of likely voters must decide how to adjust its raw results by excluding those not likely to vote. It may do so before the calls are made, by examining voter registration records to find out who has voted most often, or it can do so after the raw results are in by sifting out those not likely to vote on the basis of answers each respondent gave. Polling organizations call the latter screening, and each uses its own proprietary model to do it. My mental model in comparing registered-voter to likely-voter results is to shift the registered-voter results toward Republicans by 5%.
The Gallup organization's history is to report their Generic Congressional Ballot (GCB) question for registered voters until early October, after which they begin to apply their likely-voter screen and switch to reporting results from likely voters. Here, Frank Newport of Gallup lets a very big cat out of a very deep bag. When Gallup has applied their likely-voter screen internally to recent results, he says, the shift has been greater than 10%! That's consistent with the findings of other polls showing something like a 20% gap in Republican vs. Democrat enthusiasm to vote. It will make a huge difference in their GCB results when they start reporting likely voters next week.
Since Gallup and other registered-voter polls comprise about half of the total sample in the RealClearPolitics GCB average -- now +4% for the Republicans -- this means that average may need to be shifted toward the Republicans by about 5% to get the full likely-voter effect.
Even that assumes that the likely-voter screens of the rest of the polling organizations can be believed. And here is Jay Cost dumping all over CNN/Time for reporting what appears to be a biased sample in the California Senate race. This certainly seems to indicate that a dose of skepticism of polls performed for and released by liberal media outlets is in order from this point forward.
