I am currently working on an accident database that does produce stats for this kind of thing and I think that WhatLimits mentioned my biggest problem up to now, differing definitons of what is or is not an accident, which does seem to change nationally from decade to decade.
However with data from about 15 different national authorities it is extremely difficult to assemble all this into a common format, even within one country they record data differently now than they did years ago.
With 12000+ accidents up to now, causal factors are not always obvious to extract from the reports, so what I have decided to do is to work backwards and try to encode the data in a meaningful way, but this is nowhere near complete.
It does show interesting trends in my eyes with regards to time of year and certain types (usually weather related).
Gary