I mean that’s probably true for many applications and why you should use the simplest statistical model possible as not to introduce unnecessary errors or bias (in the classical sense, not the machine learning sense)