If you want evidence you can trust, you need well executed scientific experiements, which in this case means well executed usability tests with controls and a decent sample size, and ideally those devices which can monitor where the eye is focused on the page as a user is interacting with it.
Unfortunately we don't usually have that luxury or that data available, and we have to make do with our own experience and the advice of other professionals. But you should really take both of those things (especially our own experience) with a healthy dose of skepticism -- one of the primary goals of the scientific method is to put our own notoriously unreliable but deeply felt instincts in check.
That doesn't mean "real life experience" is worthless, or shouldn't be considered. It just means you should draw a hard line between the truths of life experience and truths established through well conducted experiments.