A Defense and Rejection of Odd Partisan ID in Polling

Nobody knows what turnout is going to be until election day. You can make some pretty good guesses, based on visible enthusiasm for candidates, months out. But you aren't going to nail the actual breakdown in voter ID. Exit polling in 2010, for example, showed a 36R-28I-36D split. Independents broke heavily for the Republicans, and that made the difference in the Red Wave. For months, some firms started shifting over their partisan ID in sampling to the right, and these firms ended up exaggerating Republican strength (prime example : Pat Toomey, who won by 2% but averaged at least double that in polling).

In the hotly contested 2004 Presidential election, voter ID split 37R-26I-37D. Even when Democrats enjoyed their own wave of pain in 2006, voter ID was only 38D-26I-36R, a D+2 edge. Independents, again, made the pain far worse. The 2008 "landslide" for President Obama did see a marked shift in voter ID (one that Gallup and Rasmussen both insist has totally evaporated) from the 2004/2006 cycles to 39D/29I/32R, which marks (so far) the high point for Democratic advantage, D+7. President Obama and the Democrats have taken hits in their image with the voting public since, so that edge is not what it was.

I could understand polling firms showing between R/D+0 and D+4 for a Presidential cycle, as there are more registered Democrats than Republicans and the Republicans make up some of the gap by being more likely to vote. That phenomenon itself is no big secret, we often discuss the difference between registered voter and likely voter polls and results. But polling firms shitting out D+10, D+11, and higher samples have no excuse. It is incredibly laughable that these polls are averaged with those using far more realistic samples (R+0 to D+4), and in so doing prop up the President's numbers.

It is amazing what happens when polling firms switch from registered to likely. Public Policy Polling enjoys a high rate of accuracy for its final polling. When election day comes closer, they shift over from registered to likely voter pools, and, surprise, the Republicans typically (but not always) do better. Just look at their Walker poll already mentioned here. Polling this far out is meaningless not because the electorate can change it's minds, but the polling firms themselves often use voter samples that will, months later, look laughable.

Not accusing anyone here. Just a thought.
Oh, and by the way, Rasmussen is showing the generic congressional ballot preference is 46% for the Republicans, 36% for the Democrats. A bump I can only assume came either entirely from Scott's bald head or a "unity" bump coinciding with the end of the primary.

posted by CAC at 08:26 PM

| Access Comments