I'll give a rather silly example: I recently saw a couple of people debating whether betting odds should give more accurate predictions of the outcomes of sporting events than other predictions. One person claimed that people are more risk-averse when their money is on the line and therefore should be expected to consider the probabilities more rigorously, while the other person claimed that it shouldn't matter whether one is putting up money. In other words, they were both making a priori arguments for why betting odds would or would not be most predictive. But in fact it is empirically verifiable that betting odds are more predictive than other models; no amount of a priori reasoning will lead you to the correct answer, only observation will.