I can be much more verbose now, I have time.
I don't see how it could reduce accidents unless the yellow was too short to allow all vehicle types time to stop.
NHTSA and other associated organizations have done lots of math and studies to determine and verify 'best' designs for nearly every kind of intersection out there, which includes proper yellow light timing.
To put it in context - while I kept it generic I could have said 'increase yellows to NHTSA or similar standards'. If you look at the studies, yellows shorter than NHTSA recommendations tend to have more accidents and red light running. Yellows longer than NHTSA standards don't do much, so it seems they've done their work. Indeed, if all intersections utilize the same standards, you gain the benefit that drivers learn to expect how long the yellow will be at a given intersection - 'Level, 35 mph, the yellow will be 'this' long and I have time to make it/not make it'. No matter where the intersection is.
Thing is though, local governments are discovering that a properly designed intersection that meets NHTSA and similar department rules regarding a well designed intersection won't have enough red-light runners to justify the expense of cameras, and that designing it right(or fixing the flaws) makes it safer than what red light cameras can provide. Of course, this presents a problem - the safest course is to fix the intersection, but that costs money. Cameras at least theoretically 'make' money, so they're preferred in areas concerned more about revenue.
Of course, courts throwing out fines left and right(including forcing the government to pay back all collected fines in select situations) when it's discovered that the reason for excessive red light running is an improperly designed or programmed intersection alters the finances. Much less when it's discovered that somebody shortened the yellow because they weren't making enough, as has occasionally happened. Heads really tend to roll then. The problem is that even if the government and camera company select intersections that happen to have a short yellow, when they're forced to retime the light to standards suddenly revenue drops. Running the cameras are no longer worth it.
If the stats don't pan out, that's interesting.
I suggest reading the sources. It's noted all over that increasing yellow duration at problematic intersections works(at least most of the time). Nearly always there's some problem to be corrected - yellow timing is easy though. Sometimes all you need is a warning light earlier on.
I don't think fatalities is the big issue here. Drunk driving is its own problem that won't be solved with traffic signals. No one wants to deal with collision damage from lower speed accidents.
Bingo. Independent studies have shown that red light cameras don't really reduce serious accidents(T-Boning at speed and such), but can drastically increase the number of rear-end collisions. Not that I want to argue your experience, but there's reasons why I kept it all statistical - accidents will happen no matter what as long as humans are still behind the wheel. All we can do is minimize them.
I think the way we design intersections needs a rethink.
Actually, following NHTSA and similar standards tends to be very effective at reducing accident rates.