The thing about statisticians is that they want problems to be clean. To the statistician, everything follows from the likelihood equation. The statistician takes the problem at hand, and contorts it until it makes sense to analyze it in terms of the likelihood.
This problem solving strategy makes sense maybe 20% of time in my experience. Real world problems don’t comport with likelihood, but the statistician doesn’t have very many other tools in her tool belt.
Statisticians think less about methods and more about problems. If a method can be applied to solve a problem, that’s great. If not, statisticians need to look beyond conventional practice.
This is one reason we have lost so much ground to machine learning in recent years, but it’s my feeling that machine learners are also starting to fall into the methods > problems trap.