That
RTTM (click) work has already been done. Table H7 of the Four Year Evaluation Report showed that for a set of ~200 urban cameras with an average KSI fall of 55%, 10% of that fall was the scheme effect, 35% was RTTM, and 10% was long-term trends (non-local safety measures such as car design, post crash response and care, national publicity campaigns, etc).
Early work showed that the RTTM effect is
even higher for non-urban roads, but no further work or analysis was done after the release of that report!
The 10% 'scheme effect' can by any safety measure applied within the defined scope of a camera site and isn't necessarily limited to just the presence of the camera, such as road/junction re-engineering, pedestrian crossing/barrier. This effect I call 'bias on selection' (as opposed to ‘bias
by selection’ [RTTM]). The setup on the M3 J2 southbound is an excellent example of this: lanes re-arranged for much greater safety (each major route gets its own lane), then enforce it and claim how great the camera enforcement is!
This last confounding factor is logically obvious and could well be just as significant as RTTM (thus possibly masking a detrimental local effect from cameras), but no work has been done quantifying it