I recommend Robert Kass' article "Statistical inference: the big picture" for a related point of view on that issue, which should be obvious but it's easily forgotten in practice.