Amir says in his headphone review preface for example, that he does not do averaging. I am not sure what is his logic.
I explained at length when I first started to measure headphones. Averaging is a type of low pass filter. It is used to gain insight into data that to humans seems random, or hard to quantify. It is also highly sensitive to extreme values (geometric mean is better in this regard).
In the case of headphone measurements, the graph is not hard to understand at all. I give you two instances in stereo measurements. Your brain can easily eyeball what the average of those two is, and you are welcome to average them if you like.
In my view, it is fool's gold to try to get to high accuracy in headphone measurements. Nothing about is precise. Targets are averaged. Fixtures comply with some average. Position variations, part variations, etc. all work to make actually resolution of the data far lower than 100%.
The measurements give us a guide to follow and confirm. This is what I do with EQ testing and listening tests in tandem. I deviate from measurements as needed to get pleasant sound.
A key goal of the target seems to be lost in all of this: in some ways, it doesn't matter what the target is. We just need one. Not five, but one. If every headphone complied with it, both in production of music and consumption, then we as consumers can EQ to taste and be done. With multiple people chasing some target with different fixtures, we lose this. For this reason, I am disappointed to see a couple of reviewers jumping on 5128 bandwagon. Why on earth would you do this? Is it some kind of race to keep up with head-fi? Why on earth would you adopt a fixture that research shows needs a well researched target to produce correct target? Makes no sense at all.