Further the data show that the cameras are correlated with a definite decrease in crashes that are directly attributable to red light running, a definite increase in rear-end crashes, a possible decrease in angle crashes, a net decrease in injury crashes attributable to red light running, and an increase in total injury crashes. More time is needed to determine whether the severity of the eliminated red light running crashes was greater than that of the induced rear-end crashes.
I think it'd also be interesting to know the geography of the area leading up to the light. One in my hometown is at the top of a steep hill, so you pretty much have to be accelerating rather hard to even get to the light. There's an area where you can be in when it switches to yellow where you HAVE to go or destroy your brakes, but it puts you pretty solidly in the middle of the intersection.
"We found that the program cannot conclusively demonstrate that it has reduced traffic collisions, thereby increasing public safety."
It goes on to recommend that data collection be improved to answer the question of what kinds of collisions and violations are being observed, in greater detail.
I think this variety of accident happens when the light has been red for quite some time (i.e. more than a few seconds). The people who get tickets (from what I've seen personally, and from talking to police) are people who just missed the yellow light by a few tenths of a second. The latter group are the ones a traffic light cameras are affecting (the ones alert enough to slam on the brakes or speed up). People who run stale red lights (the ones who cause major accidents) are either oblivious to the fact the light is red at all (e.g. drunk drivers), or they don't care (e.g. high speed pursuits).
http://www.iihs.org/research/qanda/rlr.html
In addition to reducing red light violations, cameras have been shown to reduce intersection crashes. In Oxnard, California, significant citywide crash reductions followed the introduction of red light cameras, and injury crashes at intersections with traffic signals were reduced by 29 percent.9 Front-into-side collisions — the crash type most closely associated with red light running — were reduced by 32 percent overall, and front-into-side crashes involving injuries were reduced by 68 percent. An Institute review of international red light camera studies concluded that cameras reduce red light violations by 40-50 percent and reduce injury crashes by 25-30 percent.10
Some studies have reported that while red light cameras reduce front-into-side collisions and overall injury crashes, they can increase rear-end crashes. Because the types of crashes prevented by red light cameras tend to be more severe than rear-end crashes, research has shown there is a positive aggregate benefit.
Exactly what I said, though I got the requisite downvote from the came-from-Reddit "red light cameras are just money grabs!" ignorant crowd. Though I suppose the IIHS -- the foremost authority on traffic safety -- has less sway than various anecdotes.
People accelerating to catch red lights are usually within the last 100 feet before the crosswalk. At 35mph and accelerating (~50 feet per second), assuming the light turned yellow immediately at the 100 foot mark and lasted for 1 second, the driver should be able to reach the intersection within the first second of the red light. The other drivers would take about 0.5 seconds to react and then accelerate from a stop into the intersection (remember that the law says you have to wait for the intersection to clear before proceeding), so they would likely not be in danger. It would be close, but usually these aren't the situations that cause deadly accidents. When the light has been red for a few seconds or more and someone runs it, there is a much greater chance for a deadly accident.