There’s been a lot of talk recently about party identification, and the impact it has on polling. Presidential polls adjust their results to fit a target figure for what party people say they identify with. Recent shifts in the polls have resulted almost entirely from changes in the target figure, rather than any great changes in the responses of those polled.
For example, a week ago DJ Drummond showed that Obama had risen in the Gallup tracking poll despite losing ground or staying even in every category, due to a shift in Gallup’s weights. (Via Instapundit.) If you were to fix the weights to match historical voting patterns, McCain actually had a six point lead (as of September 18).
What is the appropriate weight for pollsters to use? I certainly don’t know. Clearly, no one thinks that the historical norms are appropriate (assuming Drummond is doing the math properly, which I cannot check), but why not, exactly? This Politico story, saying that the GOP brand is making a comeback, seems relevant. (Via Instapundit.)
In fact, a better question is: why weight by party at all? It makes sense to adjust to fit target figures that we can measure accurately, like race or gender, but why adjust to fit a figure that is pure guesswork like party identification? Certainly, you could determine a party-identification figure by polling, but then there’s no real difference from simply not weighting. I don’t get it.
Internet Scofflaw 