One last point about ELS and paternalism, on the third objection I identified--that paternalism reduces people's opportunities to learn. And with this one I'll be brief.
Again, there are at least two reasons to encourage learning from mistakes. One is so that people come to know an objectively better solution to a particular question or issue or judgment; a second is that such learning can be a rewarding, valuable experience in itself. I referred to Klick & Mitchell's recent work on paternalism, where they discuss the learning issue; David Shapiro phrased the second theme nicely some time ago, suggesting that "the very ability to choose--which necessarily implies the ability to make poor choices by some objective standard--is critical to the growth of our diverse intellectual, emotional, and volitional capacities."
Of course, for learning to be helpful in either case there must be not only the opportunity, but also the ability and the motivation to learn. Well-known psychological research (e.g., reviewed here), as well as more recent neuropsychological findings, show that people are prone to confirmation biases or "motivated reasoning," where individuals prefer information that will support already held beliefs or attitudes. Personality factors (such as dogmatism), knowledge factors, and social factors (such as our perceived similarity to someone who might make a similar mistake) influence whether we learn from our mistakes--the psychologist to whom Klick & Mitchell refer suggests in his review that because of such factors, "learning is by no means automatic when people are shown the errors of their ways."
And, of course, there must be a second chance. The New Hampshire drivers who suffer fatal accidents because of not wearing a seat belt don't learn to use one the next time. Some scholars have suggested that paternalism may thus be more appropriate in contexts where decisions can have "irreversible consequences."
I said I'd be brief. And in the blog context, my goal is more to raise issues for discussion and empirical research than to refute or defend particular lines of argument. But I would suggest that for the "learning" objection, as with the knowledge-of-preferences and autonomy arguments, empirical research--rather than just ipse dixits--will be helpful in furthering such discussions.