Good question. To go from a purely conceptual model such as this to one with real empirical substance you would have to be precise about such concepts as confidence and success and how they are quantitatively measured. I think a group of smart psychologists could pull it off, as long as they were open about their definitions and quantitative scales, so people could accept or reject their thesis accordingly.
Good point. There are founders that have valuable inside knowledge of their industry (domain expertise, etc) who can be more reasonably sure of their success in some B2B venture than a YCombinator project that wants to be the next Twitter. Can we say that if both examples are equally confident of their success, that the former is less deluded than the latter?