U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations
|This report is an archived publication and may contain dated technical, contact, and link information|
Publication Number: FHWA-HRT-05-048
Date: April 2005
Safety Evaluation of Red-Light Cameras
PDF Version (621 KB)
PDF files can be viewed with the Acrobat® Reader®
Red-light running at signalized intersections is a significant problem in the United States; it results in more than 95,000 crashes and approximately 1,000 deaths per year. Red-light-camera systems aimed at reducing this problem have become a popular tool in local jurisdictions. Their use has not been without controversy, primarily related to the use of private firms to implement the program, and questions concerning changes in signal timing during program implementation. Part of the controversy has stemmed from the lack of sound research concerning the effects of RLCs on intersection crashes. Many studies of RLC effectiveness were conducted in jurisdictions outside the United States, and most of the U.S. and non-U.S. studies have experienced methodological problems, as was documented by the critical review of literature conducted in this effort. This current study was an attempt to overcome these methodological issues and to examine the crash-related effects in multiple U.S. jurisdictions to see if consistent results were found.
At the beginning of the study, the FHWA oversight panel defined a series of first-priority questions to be addressed. Following is a list of questions, followed by the study findings for each.
The results of empirical Bayes crash-frequency analyses at the treated intersections indicate that RLCs have effects similar in direction but somewhat smaller in magnitude than those indicated in past studies. Right-angle crashes (the surrogate for "red-light-running" crashes) decrease significantly and rear end crashes increase. Table 17 shows the combined results from all seven jurisdictions, indicating a 24.6 percent reduction in total right-angle crashes and a 15.7 percent reduction in right-angle (definite) injury crashes. Total rear end crashes increased by 14.9 percent, and rear end (definite) injury crashes increased by 24 percent. While the results varied some across the seven jurisdictions, the direction and degree were remarkably consistent, particularly given the differences in crash-reporting practices between jurisdictions.
The results were not as clear for effects at other signalized intersections in the same jurisdiction, known as the "spillover" effects. Here, a modest decrease in right-angle crashes was seen (table 15), but, because the results did not show the expected companion increase in rear end crashes, there is some question of whether the observed difference is to the result of "spillover," the difficulty in defining the before-and after-periods for these untreated signalized intersections (because the treatments in all jurisdictions stretched across multiple years), or to other changes at these untreated locations during the study period. It appears that a well-designed prospective study will be needed to more confidently establish any spillover effect from RLCs.
Finally, because the decrease in right-angle crashes was coupled with an increase in rear end crashes, because there can be more rear end crashes at intersections, and because the severities of the two crash types differ, it was important to combine both frequency- and severity-related effects into one analysis to determine the overall effect of RLCs. This was estimated as the difference between the economic costs of crashes expected and observed in the after-period at the treated intersections, with the former cost based on the empirical Bayes expected crash frequency. Updated estimates of comprehensive and human capital costs per crash were developed for 22 crash types in this project, and those defined for right-angle, rear end, and other signalized intersection crashes were used in this analysis. The combined results from the seven jurisdictions indicated a positive aggregate economic benefit of approximately $39,000 per site per year when property-damage-only (PDO) crashes are included and $50,000 per site per year when PDO crashes are excluded (table 21). These results indicate that the increase in rear end crash costs (due to the increase in frequency, with a lower severity) do not negate the savings in right-angle crash costs.
The analysis efforts here were focused on identifying signal-related factors that could increase or decrease RLC effects. Because such factors can affect both angle and rear end crashes, the outcome was based on the aggregate economic effects per year for a given treatment site. Univariate analyses identified factors associated with higher economic benefit, and regression models were used to verify the direction of univariate effects. (It was not possible to use the models to define either relative size or statistical significance of the individual effects, because this would require a much larger database and the ability to more precisely link a given signal attribute with a given location. In the existing database, as would be the case for almost all signalized intersections, signal timing changes both across years and even within a given day.) In general, it appears that the aggregate economic benefit increases with total entering AADT, an increasing ratio of right-angle crashes to rear end crashes, an increasing proportion of total traffic being on the major road, shorter cycle lengths, and shorter intergreen periods, and is greater for locations with one or more protected left-turn phases as opposed to intersections without such protection. Other factors such as traffic signal actuation, signal coordination, presence of turn restrictions, major road speed limit, and number of approach legs were also investigated; for these, the inability to detect a clear-cut effect may have been caused by the small samples for one level of the factor.
It is again noted that determining the relative importance of each of these factors was not possible, partly because of the noted modeling issues, but also because many of the variables are highly correlated with each other. It is further noted that these findings, even though not as detailed as might be desired, do provide guidance for implementers who want to maximize the potential benefit of RLC programs through good site choice. Given a set of potential RLC locations, an implementer should give the highest priority for RLC implementation to the sites with most or all of the positive binary factors present (e.g., left-turn protection) and with the highest levels of the favorable continuous variables (e.g. higher ratios of right-angle crashes to rear end crashes). Based on the combined univariate analyses and modeling, as well a logical consideration of the result of the crash effects analysis that rear end crashes increase and right-angle ones decrease following RLC implementation, it would appear that the most important determinant of site choice would be a high ratio of right-angle crashes to rear end crashes.
Using the same aggregate-economic-effect analysis methodology described in the preceding bullet, it appears that warning signs located at both the treated intersection and at the city limits were associated with a larger benefit than warning signs at intersections only.
It was not possible to differentiate RLC effects from the effects of other changes at the time of treatment simply because there were virtually no sites where other changes were made at the same time according to information provided to the project team.
Again, using the same methodology, high publicity level was found to be associated with a greater benefit than medium publicity level.
In summary, the multijurisdiction database developed and the crash-based and economic analyses used made it possible to answer most of the questions posed by FHWA. This economic analysis represents the first attempt in the known literature to combine the positive effects of right-angle crash reductions with the negative effects of rear end crash increases, and to identify factors that might further enhance the effects of RLC systems. Larger crash sample sizes would have added even more information. The following primary conclusions are based on these current analyses:
The authors close with two additional findings from the study. First, safety studies related to local road systems, as opposed to State highway systems, will be more expensive and difficult to conduct, and this should be considered in the research planning process. While most State DOTs have developed computerized databases containing crash, roadway inventory, and traffic data that can be used in research (including FHWA's multi-State Highway Safety Information System), this study has further documented that such files (particularly historical data other than crash data) are not usually available in local jurisdictions. If available, the data are often not computerized, they are in multiple formats, and they are maintained by different offices; thus, jurisdiction choice is a critical part of such research planning, with the availability, types, and quality of data being key issues.
Second, research involving signalization parameters such as yellow interval and cycle length is difficult. The difficulties surface not only because historic signal-change data may be difficult to acquire, but the research may be made more difficult by adaptive signal systems where the parameters change within a day. This is particularly true if traffic volumes by intersection approach cannot be linked to the specific sets of signal parameters; AADT is not usually sufficient. Research on these topics is important, but careful planning will be required to ensure its success.