... I can't see any way of empirically working out if it is less safe having 19 in the back of a 225 than 14 when it is upside down at night in cold NE seas. There aren't enough data sets out there.
These are just a first few thoughts about this.
I think you would start by using the available sets of real data, then add to that with work in the pool, and use all of that to set up the parameters of a computer simulation. Once you have run enough computer simulations and identified some possible good and bad practices, go back to the pool and test these wet. Use the results of the further pool work to refine the computer simulation and then you are probably starting to produce something useful.
The real data will not all be from inversion survivors. The relationship between seat position, injury and outcome in the likes of Cougar91 would be relevant. The decision-making of survivors of ditchings who didn't even get their feet wet would be too.
--------------------
Somebody else said 14. My thoughts are that a reduction of one or two, if sufficiently well thought through, could make a major difference if done along with changes in layout and other factors that prove relevant.
At the moment, I am not sure we even have a grasp of what is relevant.