In a recent Nature article, Five Ways to Fix Statistics, the overwhelming consensus of the authors were that science needs to incorporate more philosophy.
Though not using those terms specifically, it is interesting to note that, on the heals of decades of philosophy-bashing from the science elite (i.e., scientists who are disproportionally represented in the media), Nature is recognize that this is taking a toll on the scientific data and reasoning.
The thing that most people forget – and they forget this all the time – is that the things that they do regularly must be connected to foundational principles. If they are not, then people will either fail to do them, or start to misapply them, or both. This doesn’t happen immediately, and it requires us to have a longer-term perspective to spot the trends.
In morality, for instance, people argued that because X or Y law was passed, and the society didn’t immediately come crashing down the next year, that there was no harm in the law. This neglects the fact that there is a recursive effect between the law and social moral philosophy. When this link breaks, the breakdown doesn’t happen immediately, but in the next generation. People are taught to do things by rote instead of by principle. When this happens, then either they do it by rote (and therefore unthinkingly apply the logic improperly) or, noticing the lack of principle, drop the practice.
This has evidently happened in statistics. People are unthinkingly applying p-values, without thinking about what p-values are or why they are important. The dominance of p-values in research funding decisions is a failure to understand the philosophy of science, and to view it as a mechanistic process instead of an adventure of curious investigation. p-value hacking keeps the form but removes the substance. This is all evidence of a science that has forgotten philosophy.
On a second but related note, I should point out that I have put a proposal on the table for replacing p-values with a system that is a little less susceptible to hacking, and quite a bit less arbitrary than p-values. It is towards the end of the video below (note – you need to watch the whole video to understand the proposal).