Believe me, as the proud possessor of a BS in geography, I know all about the statistical problems of demographic surveying. (Ask a geographer or demographer, or anyone doing work involving government funding and disprivileged groups, about selection bias in US census results sometime. Go on, I dare you.)

The problem is, any survey involving human subjects is going to have self-selection problems, because even if you take a random sample, you can't compel people to answer, and even if you hypothetically you could, can't compel truth. (The only way you could maybe get a truly random survey if you could zap people out of their beds into a truth machine, which is unlikely any time soon, and will also make people very angry at you and your survey and unlikely to listen to your results regardless.)

So even a random sample off archives will exclude people who a) choose not to answer and b) choose not to archive. You can either do the survey anyway and do your best to take the inevitable refusal bias into account (as Kustritz did in her dissertation), or you can work only with publically available data and accept the fact that you're leaving out the people who don't make data available publically (as I did.)

(Why, yes, I did prefer the surveys that were about cataloging types of forb in a square of lawn to the surveys that were about talking to people. So much simpler when the subjects can't refuse to participate!)

ETA: And thank you for the new poll numbers! I did not mean to leave that out, I swear.