Researchers at the North Carolina Urban Transit Institute were unsatisfied with the overly simplistic methods used in prior insurance industry funded studies of the effects of red light cameras on accidents. So they conducted a U.S. Dept. of Transportation funded study that looked at a 57-month period and accounted for dozens of variables such as weather and traffic ignored in previous studies. All told, 17,271 observations went into their conclusions.

Their own summary says it best: "The results do not support the view that red light cameras reduce crashes. Instead, we find that RLCs are associated with higher levels of many types and severity categories of crashes."

Complete study in 366k PDF file.

Key Statistic:
[Rear End Accidents] However, in the before/after table (Table 4.2), the raw data show approximately a 10 percent increase at the RLC sites. This difference alone accounts for roughly a 35 percent difference attributable to the RLC placement.

[Total Crashes] The model is estimating that, had an RLC not been placed at a particular intersection, we may have seen a 42% decrease in the accident rate at that intersection (if we
could hold all other factors constant). Similar to what was seen in the raw data in Section 4,
the sites with RLCs are not experiencing the decreasing trend in accidents seen elsewhere.
Additionally, the other characteristics of intersections with RLCs are not explaining the
difference in accident rates.

The results do not support the conventional wisdom expressed in recent literature and popular press that red light cameras reduce accidents.... Our findings are more pessimistic, finding no change in angle accidents and large increases in rear-end crashes and many other types of crashes relative to other intersections. We did find a decrease in accidents involving a vehicle turning left and a vehicle on the same roadway, which may have been included as an angle accident in some other studies. However, given that these left turn accidents occur only one third as often as angle accidents, and the fact that we find no benefit from decreasing severity of accidents suggests that there has been no demonstrable benefit from the RLC program in terms of safety. In many ways, the evidence points toward the installation of RLCs as a detriment to safety.