One collision in 20 years is merely a mathematical probability. It doesn't have to actually happen.
So how is that probability calculated with NO data to go on?
My previous career (30+years) was out on the airfield at large London aerodromes. We had our own separation criteria for aircraft from fixed objects, buildings and so on. It took all of those 30 years to witness sufficient incidents to be able to demonstrate that these distances, laid down many years ago, are adequate or a little conservative. 3 come to mind, an American carrier with a 747 coming close to the South Terminal at Gatwick and again a 757 leaving the paved surface and running along the grass adjacent, again at Gatwick. You could also include the 747 at LHR who forgot to straighten its nosewheel steering and went charging off to the side on takeoff (27R, remember that one, HD?). Thus we have a small amount of data upon which to base decisions about whether or not to change these criteria. Where is your database for airborne collisions between public transport aircraft in the vicinity of an aerodrome? The PT mid-airs I can recall (4) were all in the en-route phase. There's one PT/GA midair that I can recall in the States, but we're not discussing that issue here.
As has already been said in posts above, all you can do in the absence of meaningful data is to take an educated guess and say that millions and millions of incident-free movements have shown the system to work. It may well be that you could now reduce these separation minima as has already been done over the Atlantic etc. We're allowed 500' vertical separation outside controlled airspace on a purely procedural basis without incident (Quadrantal Rule), why not with Radar cover?
TheOddOne