It's very clear statistically that with a truly random poll, they can be quite accurate/valid with a quite small sample size.
The large polls, like the online Facebook, Twitter, AFA, Catholic's Online, Mark Levin, etc., etc., show a very different set of numbers.
All of the above "polls" are utterly worthless, because there is not even an attempt at a random sample, and have a self-selecting sample.
You really need to take a class on statistics. I continue to be utterly astounded at the complete lack of understanding of the basic concepts of polling around here. "Online" polls and "straw" polls couldn't possibly be more irrelevant.
Of course, you like the latter set of "polls" because they're much more likely to show what you want.