I understand that one of the reasons why a small sample size can still be useful is to do with random samples: e.g. if all the samples are pulled completely randomly, then correlations even at small numbers can be really significant.
Additionally, if the samples give you additional information that can allow them to be weighted (e.g. geographical location) towards some other greater body of data then a small sample can be usefully extrapolated.
Conversely, that is why twitter and forum polls are generally quite useless at getting towards any real statistical insights that include offline opinion: the way responses to these polls are generated is the opposite of random = they are self selecting, based on who knows the user/follows that thread already etc.
A poll on a TGA in say the destruction sub-forum is probably not necessarily that good at guaging the opinions of even all TGA users, let alone all GW hobbiers, for the reason that a TGA user might have really strong opinions on destruction but still never visit the destruction sub-forum.
(I am not saying that polls like that are not fun or cannot generate insights, just that they're not really going to be good at guaging the entire hobby community)