It's mostly about the difficulty in actually interpreting data. It's very easy to construct an experiment that influences behavior to make a feature seem useful or useless. Intentionally or accidentally.

Science has known this for a long time. Designers seem intent on having to rediscover this on their own.