They do. They don’t just call 3000 people and whoever answers that’s who they go with. They weight the polls in various ways to make the sample representative of America(or a particular state). that involves making some assumptions. On one hand, it makes the polling more scientific, but on the other it can mean the pollsters can get it wrong.
No one’s made money betting against the polls, but that does not mean they are always right. Most of them are working on the assumption that 2008 will see growth in minority turnout comparable to the last few Presidential elections. Problem is, 2008 saw a big jump in minority turnout and a big fall in white turnout, so it’s more likely that the electorate in 2012 will look more like the 2004 electorate than the 2008 electorate. That’s the assumption Gallup is working with, which is why they’ve been showing better results for Romney than even Rasmussen most of the time.
What I want to know is how Nate Silver readjusts his system if a minority of pollsters got it right while the majority got it wrong. Is there even a way to account for that? He works from an assumption that an average of polls will probably get it right, but if the majority of pollsters are doing it wrong, then that doesn’t work.