| BMaker11 said: They only use about 2000 respondents in presidential elections, and those numbers are pretty accurate within a *" "*/-2% margin of error Edit: why won't a plus sign show up? |
Random sampling always has the chance to be inaccurate. Just choosing 2000 people doesn't remotely guarantee that you get an accurate spread. You are just very likely. Laws of probability dictate that eventually you get a bum group. In your example eventually you get 2,000 hardcore partisans one way or the other and get a completely skewed result. The difference being it would be immediately obvious what happened and the results thrown out and try again. If the perceived demand does not match actual sales then the poll is obviously just wrong due to random sampling error.
