kvining - 10/6/2004 2:25 PM
The polls are showing partisans selected the winner as would be expected, but everyone is being surprised by the fact that "undecided" groups are scoring Edwards the winner. I didn't expect that, and am left scratching my head as to why, given Cheney's decent performance. Any ideas?
Polls are done by resampling stats. You poll a small number and through various arcane randomization and resampling methods, accumulate results until you reach some stopping criterion.
Because of the small pool of original samples, anything that affects the assumptions when sampling may have a significant effect on the results.
For example, when comparing various polls its a good idea to check whether the sample pool is of anybody who answers the phone, voting-aged people who answer the phone, people who have voted previously in a presidential election, etc.
Also, unplanned effects can be in effect. For example, say a poll taker has a hard time pronouncing Kerry without making a fart-noise. If that person took 1/10th of the samples then there is a great chance that his impediment would bias the results.
One way to avoid systematic errors (like poll-taker error) is to look at polls taken from lots of sources and compare their results. Just looking at one source will give you a greater liklihood of some sort of error. Look at a bunch.
Finally, would you be willing to invest say, your next paycheck that the results of a given poll will correctly predict the winner of the election? If not, then you may wish to greet that poll with an appropriate level of skepticism.
Here's a good source for lots of polls, FWIW.