Two Takes on the State-National Poll Divergence

National polls show a tie, or perhaps a tiny Romney lead. (Right now, Real Clear Politics average of polls shows a tie, but the larger a poll’s sample size, the better Romney is doing. And fwiw, I find it hard to take the National Journal’s poll with a five-point Obama edge seriously except as a product of a large margin of error, given that it predicts that Democrats will have a greater advantage in party i.d. than they had in the “wave year” of 2008). Extrapolation from state polls shows Obama with a two percent or so lead. Sean Trende, whose work is top notch, examines the discrepancy here, and concludes that there is no way of knowing which set of polls is more accurate. Nate Silver of Blog 538 discusses the discrepancy here, and concludes that we should pay more attention to state polls because they have been more accurate in recent elections.

I’m more persuaded by Trende, for two reasons. First, Trende points out that many 2012 state pollsters, especially in the swing states, have no prior experience in polling those states. The national pollsters, by contrast, are all reputable firms with many years of experience doing national polls.

Second, Silver’s case is based on the fact that “in recent elections — since state polling data became more robust — it’s the state polls that have done a bit better.” But his own data show that national polls were slightly more accurate in 2008, and state polls slight more accurate in 2004. He points out that 1996 was an especially good year for state polls. But does anyone think that the identity of the pollsters, the methodologies they use, and the specific states they have polled (only seventeen in ’96 compared to thirty-nine this year!) have remained largely unchanged in the last sixteen years? Given that, plus the changes in polling techniques necessitated by the rise of cell phones and caller id, 1996 might as well be 1960. None of this means the state polls are wrong. But it does seem to support Trende’s view that there is no way of knowing based on past experience whether the current state or national polls are more accurate.