Originally posted by Airscrew
The trick there is scientifically and without bias. This particular survey just doesnt seem very accurate to me.
[/b]
Why does it not "seem" accurate?
Usually its not the survey data thats suspect anyway, its the interptation of the data and how the data is used that leads me to distrust surveys.
[/b]
The real issue that should worry you about survey data is not so much the interpretation as the question wording and question order. You can draw a perfectly random and representative sample of the population, but if you ask poorly-worded questions or lead people on through question order, you can bias the results. I suppose that's possible in this survey, and I didn't read the text of survey itself (and note that all good survey articles should include a link to the actual survey questions), but asking fact-based questions like naming the rights in the First Amendment or name the Simpsons doesn'tseem prima facie biased to me.
I think that before I made a statement like this "It's disappointing that Americans continue to be ignorant of First Amendment freedoms," I would try to survey a larger sample, 10,000 or more
[/b]
Again, why? You may achieve a 3-4% reduction in your margin of error by doing that, but you do so at a tenfold increase in the administrative costs of the survey. Thus while the survey, if unbiased in every other way, contained 10,000 respondents instead of 1,000, the estimators would in fact be more accurate but not substantially so. We would achieve 100% accuracy with a total population census, but would it be worth the increase in cost from 10,000 respondents? We're talking tens of billions of dollars in administrative costs now.
The fact is that, unlikely as it seems, a survey drawn from 1,000 respondents may be representative of the general population if done properly and scientifically.
-- Todd/Leviathn