Reply #99: A low-cost DU workaround:

[PaganPreacher]

The "Democratic Underground, 50-State Post-Election Poll"

1. Statistical model:I read posts by amateur statisticians on DU every day. Those folks could develop a statistical model that they agree is accurate and predictive (one that has no assumptions). That model must have one variable, and only one: the number of votes for each candidate.

2. Test bed: Some of our DU-bees could take their local telephone book, call the first number and every tenth number after in the White Pages, and report back the results of two questions:

a. Did you vote in the 2004 Presidential election?
b. (to be asked if the answer to question "a" is "yes"): For whom did you vote?

Each "pollster" provides the raw numbers from his/her canvass, and our in-house savants figure the stats, using the agreed-upon statistical model. That gives us our test batch.

3. Run the poll: The same canvassers go back to their phone books, begin at the second name in the book and call every 10th number. Same two questions. Report back the results, and our homegrown mathematicians apply the numbers to each state's totals, giving us results for all states (not just cherry-picked states, as some here want to do.)

After running the poll, we can see if the results reflect actual poll numbers, DU expectations (the dreaded "f" word), or if the whole idea has an unforseen flaw.

The Pagan Preacher
I don't turn the other cheek.