Article from: www.thenewspaper.com/news/00/53.asp
Nearly every study and crash analysis reviewed had some experimental design or analysis flaw. In many cases the flaw in the analysis was because of the lack of a proper control group, which would allow a valid comparison of the observed changes, increases or decreases, with changes in signalized intersections that did not have cameras. Cameras tend to be installed at problem locations; those with higher than average crash experience. Because of the manner in which crashes occur over time at a given location, these types of locations can experience reductions in subsequent years even without intervention. To account for this "regression-to-mean" phenomenon, control or reference sites are needed. In some cases, a reduction was observed, which when exposed to statistical testing proved to be statistically insignificant. This statistically insignificant finding is often attributed to small sample sizes pertaining to sites and the crash frequency at each site.Source: Impact of Red Light Camera Enforcement on Crash Experience (Transportation Research Board, 4/10/2003)