Sometimes we go into a situation with such strong prior beliefs about the truth (or the state of the world or whatever) that even very strong evidence that contradicts those priors doesn't persuade us. As a long-time academic and supposed scholar, I'd like to hope that doesn't happen very often with me, but I expect it does. [For more on the use of prior beliefs in statistical analysis, see this and the related links on Bayes.]
But here is a pretty impressive example from Slate.
This morning, out of nowhere, Public Policy Polling revealed to the world that it had polled the Colorado recalls, been on the money, but failed to release it. "In a district that Barack Obama won by almost 20 points," wrote PPP's Tom Jensen, "I figured there was no way that could be right and made a rare decision not to release the poll."
I expect that PPP could not possibly believe their poll results were accurate. So rather than report what they feared was flawed and what might sully the reputation of their firm, they chose instead not to report the results.
Well, at least this way they can say, "Our polling techniques are better than we thought they were."