Reply #16: Not entirely true.

[Teaser]

Any sample is based on a probability distribution. This kind of sample, though, is not a random sample, and, as such, we can't use the central limit theorem to calculate population statistics (like average percent support). However, the biggest problem with this polls is that the distribution it draws from is biased.

What Zogby is looking for, when he does these polls, though is other correlates of that bias. For example, do certain opinions about local government or services or one's shopping habits correlate strongly with an increase or decrease in support for a particular candidate/party?

He's trying to quantify the number of points that gets added to a candidates number based on regressions against these other variables. Then when he finds the adjustment, he can re-weight the data and find the optimal estimate of the "true" random variable (the right percent of people supporting Hilary, for example). Essentially he's trying to find a way around the self selection bias.

Will it work? I'm dubious, but its possible. Indeed, I hope it does work, as it will make opinion research vastly more efficient.

But right now, it's not ready for prime time, and Z should not be releasing this data.