Learning from a discrete choice experiment