Why is a carefully designed experiment sometimes a better way to test a hypothesis than only collecting observational data to describe a pattern that occurs in nature?