Originally posted by: Candy Wright
My first thought...wow, "half of Americans"? Guess I wasn't home when they called.
Then I read the rest---fifty percent of poll respondents. Oh, OK, 1000 registered voters were polled.
Not exactly the same as "half of Americans."
One thing that uneducated people don't understand about polls--and about statistics in general--is that with the proper methodology, a quite small sample in relationship to the whole is mathematically representative of the whole. The degree to which this is true for a given sample size is expressed in "margin of error," which represents the range of results that the exact number actually represents. For instance, in this poll, 50 percent with a 3% margin of error means that the actual finding was 47-53%, with no mathematical distinction between any points within that range.
This is one of the features of statistics (and math in general) that is counterintuitive. It leads to people conducting studies with unnecessarily large (sometimes ludicrously so) sample sizes. There's another common misunderstanding about sampling that leads to the "birthday paradox," where you pick people at random from a group (like an audience) and then ask them what their birthdays are--and it's even money that you find a duplicate after you ask just 23 people. Many would say that you would need to ask 183 people before that happens.
To state it another way, there's a 97% chance that the poll results were indeed representative of the electorate in general, and a 3% chance that they were not. The poll was a binary choice with 1,000 samples, and the chance of error experienced a cumulative reduction as the sample size grew larger. With a sample size of, say, 10,000, the margin of error would still have been close to 1%. Therefore, the best way to validate poll results is to repeat them with different samples but otherwise identical methodology.
It should also be noted that a poll is a snapshot in time and its results cannot be extrapolated to later situations and events.