Public Polling and the Heisenberg Uncertainty Principle (A nerd two-fer!)

On Twitter, Steve Silberman links to a new report showing that only 52% of Americans know that vaccines don't cause autism. As another Twitterer points out, the veracity of online surveys is always tough to measure, since they tend to be self-selecting. But it raises another question for me, one I haven't seen addressed very much: Does the very nature of polling people about empirical facts make them more likely to think that the topic is more subjective than it really is? That is, is there a Heisenberg effect in polling, whereby one cannot effectively measure public opinion without also shaping it?