For some reason I still feel some kind of duty to write corrective letters to papers I read. Probably a waste of time, but here goes, anyway:
As you point out in your article (May 6th), the debate on speed cameras in Britain has been surprisingly controversial, and has seen much abuse of statistics. The claims made by the anti-cameras lobby in particular explore a rich vein of nonsense which even numerically-challenged British tabloids wisely avoid.
But the statement that `deaths and injuries drop on average by 35%' where cameras are installed is also misleading. As you state, cameras are installed in locations where above-average numbers of accidents have occured. All else being equal, a period in which an above-average number of serious accidents occur at a site is very likely to be followed by a period in which fewer accidents occur, whether or not a camera is installed there.
This effect -- known to statistics as `regression to the mean' -- is well-known and widely understood. But sadly it is rarely addressed properly in studies of speed camera effectiveness. Without controlled trials free from such bias, we do not yet know how effective speed cameras really are.
The very controversy of speed cameras is a cost that should be weighed up when making road safety policy. However effective, cameras are unpopular, and they sour the usually good relationship between citizens and the police. Without good data on the effects of cameras, it is not possible soberly to assess the costs and benefits of using them.