A useful read on the recent run of three polls in Oregon on the governor and Senate races in Blue Oregon, speculating on an interesting distinction in the polling for those two offices.
The post notes that polling for the governor's race has been fairly consistent - the three polls (by SurveyUSA, Rasmussen and Davis Hibbitts & Midgehall) all show a close race between the two contenders, Democrat John Kitzhaber and Republican Chris Dudley.
In the Senate race, though, the pollsters diverged more widely. All gave leads to incumbent Senator Ron Wyden over Republican Jim Huffman, but the leads amounted to 10%, 13% and 30% - a big gap.
In the post, Jeff Alworth suggests: "Polls are useful only to the extent that they can predict who will turn out for an election. What I see here is the result of weighting for anti-incumbent votes. . . . But apparently, pollsters are giving a lot of credence to the anti-incumbent theory. You don't see the same effect in the Dudley-Kitzhaber race because neither is an incumbent."
Once, the idea in polling was that you randomly contact people and then report what they say. Over time, in recent years, that has changed: Now results are weighted and adjusted, depending on what factors the pollster thinks happen to be relevant. (Weighting in national polls for party identification has become SOP.)
Hmm. Along with the changes in technology (the old call-em on land lines approach just ain't cutting it the way it used to), you have to wonder just how valid any of these polls are.