In fact, when polls report their "margin of error" (e.g., plus or minus 4%), they really take into account only sampling error, because of the four major types of survey errors that's the only one that can be sufficiently quantified. The other biggie sources of errors are coverage error, measurement error, and non-response error. I don't pretend to know much about this stuff, but I know I've fallen into the non-response error category this Presidential election...probably several times: I look at the caller ID before I answer the phone, and if I don't know the caller I'll let voicemail get it 98% of the time (give or take a margin of error of 1.5%
).I know pre-election polls serve valuable purposes, but I wish they weren't as prevalent. Not everyone takes an analytical view of what they see or hear in the media, and I think all the campaign polls--this year more than any other I remember, even more than 2004 (remember when, in 2004, in early October, Newsweek polls had 47% of the vote for Kerry, 45% for Bush, and 2% for Nader?)--have a very real potential to mislead and affect behavior.
Best case, the polls will lead all the newly-registered Democrats to stay home because they believe Obamanos has already won. Worst case, something similar could happen with Republicans and Libertarians.