I am awaiting the final flurry of House Polls at RCP before I do my final calls today. Not to mention getting some work done (on that note, I will be in meetings all morning tomorrow so I will be posting in the afternoon at the earlies). But let’s set the stage with where we are nationally. As I noted yesterday in the WaPo/ABC News, Pew and Gallup polls on the generic or national mood, the dems have dropped way down from their highs 2-4 weeks ago (the Gallup Poll 2 weeks ago showed a significant drop from the mark 4 weeks ago, indicating Gallup caught wind of this decline earlier than the other two polls). While I think some of this is an actual change in the electorate since all three picked up the movement. It is unclear if it is a change in the access pollster have to conservative voters (fewer hangups) or a change in conservative voter position. In fact, it is probably both but I would wager the cons are finally talking to the pollsters. Either way the pollsters are finally adjusting their turnout models. Hugh Hewitt discusses the five factors that could be shaping these polls, and I agree it is probably a combination of all of these with varying levels of influence. The fact multiple factors are all pushing in one direction means we might finally be seeing the corrective actions in the turnout models that are bring what has been a very unfocused and jumpy image into sharp detail.

If we look at the RCP average of independent, professional polls (which knocks out any considersation of the Time and Newsweek amatuer media polls) we see these last polls giving dems a +4, +6 and +7 lead. The average is 5.6% lead. Which is really nothing historically. We also see in all these polls statements of higher GOP interest. But there is much more to the story, as Ed Morrissey points out. Dems are sliding across the board in the internals

In a month, the Democrats have lost non-minorities altogether. The gap among all whites went from +5 Democrats to +5 GOP, a ten-point swing. White females had supported Democrats by a 15-point margin and a majority (55-40), but now give the GOP a 2-point lead. The Democrats have also lost the middle class, a big problem in this election.

It is these internals that are truly stunning. Did all these people change their minds in a month? Doubtful. It is clear pollsters get a chance to do last minute turn out adjustments this weekend as absentee an early voting results start to be announced. And yes, many people will finally open up and talk to pollsters this weekend, right before the election. The troubling problem for the media and pollster is their consistent bias towards the dems over the years. They still use too much emotion in their turn out models.

I have been looking for the right analogy to convey the inaccuracies of polls. As I tried to show in a post about the FL-13 race (Harris’ old seat) the polls show a small Dem lead while early voting results show a larger Rep lead. But it was a Walter Shapiro handwringing article, and discussions last night with LJStrata, that gave me the perfect example – hurricane predictions.

If you notice any hurricane track prediction it is in the shape of a funnel or cone. As the storm comes closer to land, and variables become less variable, the cone tightens down. These are classic stastical model outputs and the storm’s path, the cone’s center line, is usual pretty accurate. But it can ‘drift’ when an event impacts the path.

Election race results, if done properly, would have this same conic nature. There would be a maximum leftward and rightward boundary with a high probability center path. But the polls always have their center path skewed by unfounded assumptions on turn out. Anyone familiar with statistical models that show a constant bias can adjust the center mark by applying a correction. And that is what I have been doing (and will do soon again) when I make my predictions on the House and Senate. I am removing a constant bias from the model. Now this can be done one of two ways. We can eliminate outliers that push the result one way, or we can include the outliars (pun intended) and simply correct the result. I am still trying to decide which way to go, but one thing is clear: the polls as they sit are meaningless because you need a college degree in math to figure out which poll is biased left or right, or is on the mark.

For example, the Pew poll out yesterday indicated over half their sample have been contacted by GOTV efforts. It also had the smallest Dem margin (4%) and putting the race in a basic tie. The Gallup poll had a signigicantly lower pool of GOTV contacted samples in their poll – and they had the largest Dem lead. I would consider the Pew poll more accurate and give it more weight since the GOTV effort is key in midterm elections. I don’t have time or desire to do this kind of deep analysis, but I think from here on pollsters should be giving us their conic views of where the races are, not a single data point inside the realm of possible results.