More on Party ID Sampling in Polls

When party ID samples in polls was the au courant topic, I weighed in mocking the conservatives. They were making it sound like pollsters were intentionally stacking party ID numbers to favor Obama. I explained that that was impossible. The only way pollsters can do party ID--especially in a national poll, in a country where people in roughly half the states do not register by party--is to ask people what they consider themselves to be. No one rigs anything.

But that doesn't mean there can't be strange occurrences, and I will confess until someone 'splains me otherwise that today's WaPo poll, Obama +3 among likelys nationally, looks a little odd to me because of the D +9 party ID sample (see the last few questions). I think that's high for right now.

People tell pollsters what party they consider themselves to be based on events of the day. Many weak identifiers switch back and forth. One study of the 2008 electorate found that nearly half the people do so. (Google "Cobb Nie study partisan ID"; I can't link to it, but trust me.) That seems a tad high to me. But whatever. This study also showed that ID tends to solidify after the conventions.

If that's true, and if the Post sample is on the spot, that means we're going to have a very heavily Democratic electorate. Even 2008's electorate was just D +7. But no expert I know of expects that. John Podhoretz thinks the new poll is so rotten the Post shouldn't have published it. 

At the same time Politico has a new poll, also showing Obama ahead but just by one point. Its party ID breakdown is D +4, which seems more realistic.

Advice: Don't get too hopped up about any single poll. Just read Nate Silver. But come election night, go have a look at the exit polls and see what they say about the make-up of the electorate. If it's close to D +7, a lot of experts will be surprised, and Obama will almost certainly win. And if it's only D +2 and he still wins, then I'd chalk it up to the way a certain butterfly flapped his wings in Beijing the day before.