Given that, from year to year, you can get that much of a change in accident rates through mere chance, it's not a very reliable metric. There is no real comparison, except to say that accident rates at the place where the camera was placed dropped by an amount that is actually within the bounds of statistical error. You can have a 100% change in accident rates on a particular stretch of road from year to year without changing anything, so any claim made for a camera would have to be very, very thoroughly researched and evidenced. So far that hasn't been the case, and what evidence has been presented is usually found lacking on closer examination. As in, they claim one figure and then later claim another.
What about all the places where cameras haven't been placed?
What about accidents that happen just out of the range of the camera?
And what about all the accidents that happen within the bounds of the speed limit? Despite the propaganda the problem isn't speed, it's lack of awareness and distraction caused by too much street furniture - including speed cameras. Bad driving, in other words, to which speed merely adds a little more energy. The number of accidents I've seen happen just before passing a camera are too many to list (and, yes, anecdotal - but they were all caused by people who were, up to that point, driving safely though a little over the proscribed speed, suddenly hitting the brakes in order to avoid getting flashed by the camera with the inevitable result of a rear-end collision).
Speed can increase the damage caused by an accident but speed, by itself, doesn't kill. You can be killed by someone driving at 20 if they hit you right.