Let me ask you a 30,000-foot question. For either one of you, based on the data you're getting, the kinds of inputs you're getting and the kinds of programs you know are out there that you are evaluating with the data, what is working? What kinds of measures are the kinds of measures that work?
