On Thu, 28 Oct 2004, the Washington Post was quoted saying:
> Pre-election polls in 2000 were the most accurate in nearly three decades.
> Pollsters point to data showing that in 2002, nearly nine out of 10
> candidates who were ahead in surveys conducted immediately before the
> election ended up winning, with the overwhelming majority of these polls
> coming within 3 percentage points of the winner's victory margin.
This shows these guys aren't even trying to account for what's going on. They're in full CYA mode. The last presidential election was notable for its unrandom error: 85% of the polls called it for Bush and Gore won. If they were serious, they'd address that. They never do. Nor that it looks much the same now -- and with a larger turnout (which all predict), and the same methods, there is every reason to expect a larger bias.
> "As far as I can tell, those who don't participate in surveys are busy
> people or people who maybe are unhappy with being deluged by
> telemarketers. But either way, it doesn't seem to be powerfully
> correlated with the kinds of things people measure in polls. Missing
> those people doesn't distort our predictions appreciably."
First off, just in passing: how accurate can a poll be of people who refuse to engage in polls?
I think this article gets close to the real problem and then walks right by it. The cell-phone-only is a small percentage. It's a problem, but it's not yet enormous. What has really gotten large in the last four years is the number of people who don't answer their phone for strangers ever -- for whom it doesn't even cross their mind (except when we're at work). At home, we only accept calls we expect. We do all our business with strangers via email. And the skew of this group is younger, more urban and more educated than the general populace -- and more Democratic.
My working hypothesis is that it isn't a particular killer app that's causing the problem. It's a convergence of a whole host of technologies of call avoidance, and a sufficient period of time of getting used to them, and an almost universal integration of certain social strata into email, that has entirely changed the meaning of the phone. A large swathe of the population has entirely lost the felt obligation to pick up. All the last remaining worries of what you might miss by not doing so have been expunged. And this sizeable group does have a demographic skew. And that's why the all the polls (including the RV ones) are consistently oversampling Republicans. Which is where the bias comes from.
Michael