"It's an interesting balance. The proposals are supposed to be a test of your ability to design a convincing research project - you don't actually have to do what you say, and the NSF doesn't care if you do or not, so it doesn't matter if your idea is wrong (as long as it's not so wrong that the judges notice). On the other hand, it's easier to be convincing if you have some data to support your claims."
What do you mean that it doesn't matter if your idea is wrong? After I submitted my proposal I found some literature showing that I picked an inappropriate substrate to work with...I kind of figured I'd killed my proposal even though the background and idea was good.