I heard on Fox that the Harris and Pew polls have Kerry and Bush neck and neck, so I take a big gulp and then run across this post from Roger Simon which refers to this Drudge Headilne touting a Gallup poll that shows a Bush "blowout" of 14 points! So, what the heck is going on here? First Bush gets a convention bounce of 11 points, and then that disappears completely without any downward trend, while at the same time there's evidence of a blowout... Very strange.
Well, I have a couple of ideas. They're just theories, or more precisely educated hunches.
One of Roger's commenter, who discusses "weighting," is probably on the right track. Pollsters adjust for a non-random/non-normal sample by constructing what are called "weights." For instance, if you've sampled fewer black people than would be in a random/normal distribution you might count a particular black person's response as a little more than one response. You weight it, when you run your tables.
I worked on a cross-national poll awhile ago with Martin Lipset, and I wouldn't say that the polling firm we worked with (I won't mention the name) really had much of a proprietary weighting scheme. And they're probably more rigorous than Harris, and a bit less rigorous than Pew. They just used census data and "corrected" using a fairly pedestrian heuristic, but it's not a black art. And it's easy to get it wrong, or to inadvertently fail to get the clusters or strata you expect to get (let alone a random sample). Statistically the problem is related to the "ecological fallacy," because you're making assumptions, for instance, about the way black people will skew the results, and you don't really know the opinions of the mystery people you're adding in or subtracting, to "unskew" the sample, because you haven't asked them anything. You just assume those opinions would have been "like" those of the people you have sampled.
But when people aren't behaving the way you expect them to, those assumptions can be wrong. So I think here's what the volatility in the polls means. The assumptions that polling firms, as well as political scientists, sociologists, and all manner of conventional pundits are making about the electorate have been thrown out of whack by... guess who? Just take a guess. A wild one, with sprinkles on top?
There's a deeply intriguing social dynamic going on here that's very "new," and I wish I had the resources to suss it out. Any of you pollsters wanna hire me? I'm available. In fact, I'm fairly cheap for the moment. If I didn't know better though, I'd say that some conventionally Democratic-leaning demographics are "wising up," and fracturing in unfamiliar ways. And these new "crosscuts" in the social fabric are throwing the polling firms into a tizzy because they're running right through some of those "solid voting blocks" that they've assumed are still sitting in the cultural ocean like big icebergs. The bergs are melting. Like I said, it's just an educated hunch.
Update: Vodkapundit argues that the anomalies really have to do with the difference between a "snapshot" and a multiple image "motion picture" (i.e. a cross-section vs. a tracking poll). Al Hunt, on the other hand, thinks it's a matter of registered versus likely voters, and that "screens" used to identify likely voters by some firms tend to over-sample conservatives. Either or both of those conjectures could have merit, but neither really explains the wild fluctuations or the between-sample variance. At least they don't explain it completely. I still tend to think that what this electorate is manifesting is something "on the edge of chaos." There are a large number of voters whose historical and demographic voting preferences are being challenged by a build-up of cognitive dissonance, in the way that a heavy snowfall builds up the potential for an avalanche. And to some extent the snow pack has broken loose.
You could think of it as a series of avalanche "events," where there's a break followed by re-accumulation, followed by another break, etc.. So any snapshot would tend to show an incomplete picture of what's going on (Vodkapundit's thesis) while the imprecision of the "likely voter" filters and/or the demographic weight-adjustments are inaccurately measuring the masses that are moving. But overall I think the entire phenomenon, both the herky-jerky-edgy nature of the electorate and the imprecision about who is likely to vote and how, are some rather new phenomena driven in large part by this very medium: the blogosphere. I could be wrong. It's just a conjecture. (Hat tip: Rusty)
Posted by Demosophist at September 17, 2004 02:33 AM | TrackBack