First, test if observation is valid (Is X > Y?). If valid (and robustly so), then proceed, to determining what might predict it.

Maybe there's something geniunely in the observation. But there are steps we ought to take (and teach) for an observation of the real-world to be considered ...