The second piece of evidence is more research-based, but still anecdotal. John Anderson's group at Carnegie-Mellon University has created several intelligent tutoring systems that teach Pascal and Lisp. These have been found to be successful at teaching students the language. However I noted that they never asked the question, "Can the students who learn Pascal or Lisp using the cognitive tutors program in those languages afterward?" I asked Ken Koedinger this question once. His answer was, "No way!" The cognitive tutors lead students to write correct programs. They never *run* the programs. They never *debug* the programs. (Why would you debug something that's already right?) These students have a really good shot at writing a program correctly the first time. Any mistakes at all, and they're hosed. I know of no paper from CMU where they actually tested this claim and published the results. (If you find one, please send me a link!)
Jennifer responds: "I did so have to debug Lisp, with all those damn parenthesis."