Knowledge Resources

A national study of the effectiveness of red-light cameras (RLC) at132 treatment sites across seven jurisdictions found that the RLC sites had a decrease in right-angle crashes but an increase in rear-end crashes. The study estimated the economic effects of RLC-treated intersections by calculating per-crash costs by crash type and severity. Results showed that RLC treatment resulted in a modest aggregate crash cost benefit because the lower costs from the decrease in right-angle crashes was not negated by the higher costs from the increase in rear end crashes.

This study used a methodology that was noteworthy in two respects. First, the study had an unusually robust crash data sample obtained from 132 RLC treatment sites that spanned seven jurisdictions across three states (California, Maryland and North Carolina). Second, the study employed an empirical Bayes before-after research design that corrected for statistical biases frequently found in RLC evaluations. The method compensated for the problem of overstating the safety benefits by not correcting for regression-to-the-mean bias, and the problem of underestimating the safety effect by not considering the “spillover” effect (or benefit) of RLC intersections on nearby non-RLC ones. The research design also adjusts for volume differences between the before-treatment and after-treatment periods. Finally, the study used a comparison group of un-treated intersections. The results of the study revealed that the RLC effect was more modest than previously indicated. Nonetheless, results found a net benefit due to the fact that cost reductions from the decrease in the right-angle crashes were only partially negated by the increased costs from the higher numbers of rear end crashes.

FINDINGS

• A crash reduction benefit of approximately $39,000 per site year when property-only-damages (PDO) crashes are included.
• A crash reduction benefit of approximately $50,000 per site year when excluding PDOs.

The study compared the number of crashes by type (right-angle versus rear end) and by injury at RLC sites during the after-period to estimates of the number of crashes by type and injury in the before-period. In the table below, the results from the 132 RLC treatment sites are grouped by crash type and injury. (Note that "definite injury" crashes are defined as K (killed), A (disabling injury) and B (evident injury) according to the KABCO severity scale used by police officers at the crash scene. "Definite injury" does not include the "possible injury" crashes captured by C (possible injury) or O (no apparent injury.)

at RLC treatment sites	Right-angle crashes		Rear end crashes
	Total crashes	(Definite injury)	Total crashes	(Definite injury)
Count of crashes observed in the after-period	1,163	296	2,896	163
Estimate of percentage change (standard error)	-24.6 (2.9)	-15.7 (5.9)	14.9 (3.0)	24.0 (11.6)
Estimate of the change in crash frequency	-379	-55	375	32

The study estimated the spillover effect of RLC on un-monitored intersections. The data show a modest effect on crash frequency at non-RLC sites, suggesting that RLC have a deterrence effect that generalizes to intersections without RLC.

at spillover intersections	Right-angle crashes	Rear end crashes
Empirical Bayes estimate of crashes expected in the after-period without RLC	3,430	3,802
Count of crashes observed in the after-period	3,140	3,873
Estimate of percentage change (standard error)	-8.5 (2.2)	1.8 (2.3)