As you say, it depends on how you count. Do you count total accidents?
Accidents per participant? Accidents per hour? Per mile? Per
passenger mile?


You could do any of these, but you have to do the same thing on the top
and bottom, and with cars as well as planes. Include busses if you
include non-spamcans.


But the more significant factor is, I think: do you
count the accidents that were caused by circumstances that you never
place yourself in?


You discount accidents that don't apply (such as helicopters and jumbo
jets, perhaps). But you don't discount accidents that result from
errors "you'd never make".

Do you count Vmc accidents in twins if you never fly a twin?

Don't count twin accidents at all. Don't divide by the number of twin
hours (miles, whatever) either.


Do you count fuel exhaustion and inadvertent VFR into IMC if
you are absolutely religious about checking your fuel, having plenty of
margin, have an instrument rating, stay current, and always file IFR if
there's a cloud within 500 nm?


Yep. That's a case of "it will never happen to me" wherein it just
might happen to you. That's the definition of "accident".

Do you count stall-spin accidents in Tomahawks if you fly a Cirrus?

Probably. You can stall-spin a cirrus. It obeys the same laws of
aerodynamics.

Do you count icing accidents in the mountains at night if...?

It depends on what lie you want to promulgate. If you want to
figure out the truth of the matter, it is important to ask the right
questions. You need enough data to be meaningful, and you need to pare
it enough to be relevant.

Jose