Wednesday, October 27, 2010

"pollsters need to address why their results sometimes diverge widely a few weeks or months before the election, only to converge at the end." -Mark Blumenthal

Pollster accuracy is judged almost entirely by their last poll before the election. The idea is they say polls are snapshots not predictions (this is silly we look at them to predict).

LV filters will select different voters depending on the date. Gallup is unusual as they explain their LV filter. One question is something like "do you know the location of your polling place." Clearly not knowing on Oct 30 is an important signal that one is not going to vote. Not knowing in August pretty much just means the voter hasn't voted there before

Another question is something like have you voted there before. Both select against young people. The two put people who have never voted before *or* who have moved since the last election on the edge of exclusion (3 strikes and you're out that is 3 non voter like answers imply that Gallup considers you not likely to vote).

Gallup uses this filter, because just before the election it becomes a good filter. Every 2 years there is a Gallup anomaly where the Gallup LV sample is more Republican than other LV polls (except maybe Rasmussen this year). This is all very predictable and, unlike pollsters, I personally have addressed it repeatedly.

No comments: