Effects of automated speed enforcement in Montgomery County, Maryland, on vehicle speeds, public opinion, and crashes

ABSTRACT Objectives: In May 2007, Montgomery County, Maryland, implemented an automated speed enforcement program, with cameras allowed on residential streets with speed limits of 35 mph or lower and in school zones. In 2009, the state speed camera law increased the enforcement threshold from 11 to 12 mph over the speed limit and restricted school zone enforcement hours. In 2012, the county began using a corridor approach, in which cameras were periodically moved along the length of a roadway segment. The long-term effects of the speed camera program on travel speeds, public attitudes, and crashes were evaluated. Methods: Changes in travel speeds at camera sites from 6 months before the program began to 7½ years after were compared with changes in speeds at control sites in the nearby Virginia counties of Fairfax and Arlington. A telephone survey of Montgomery County drivers was conducted in Fall 2014 to examine attitudes and experiences related to automated speed enforcement. Using data on crashes during 2004–2013, logistic regression models examined the program's effects on the likelihood that a crash involved an incapacitating or fatal injury on camera-eligible roads and on potential spillover roads in Montgomery County, using crashes in Fairfax County on similar roads as controls. Results: About 7½ years after the program began, speed cameras were associated with a 10% reduction in mean speeds and a 62% reduction in the likelihood that a vehicle was traveling more than 10 mph above the speed limit at camera sites. When interviewed in Fall 2014, 95% of drivers were aware of the camera program, 62% favored it, and most had received a camera ticket or knew someone else who had. The overall effect of the camera program in its modified form, including both the law change and the corridor approach, was a 39% reduction in the likelihood that a crash resulted in an incapacitating or fatal injury. Speed cameras alone were associated with a 19% reduction in the likelihood that a crash resulted in an incapacitating or fatal injury, the law change was associated with a nonsignificant 8% increase, and the corridor approach provided an additional 30% reduction over and above the cameras. Conclusions: This study adds to the evidence that speed cameras can reduce speeding, which can lead to reductions in speeding-related crashes and crashes involving serious injuries or fatalities.


Introduction
Speeding persists as a major factor in motor vehicle crashes. In the United States in recent years, speeding-defined as driving too fast for conditions, exceeding posted speed limits, or racing-has consistently been involved in about one third of crash deaths (Insurance Institute for Highway Safety 2016). About half of speeding-related fatalities in 2014 occurred on roads with speed limits lower than 55 mph, and one-quarter occurred on streets with speed limits of 35 mph or less.
Publicized traditional police enforcement can reduce vehicle travel speeds and crashes (Stuster 1995), although effects can be localized and temporary unless increased enforcement is sustained (Barnes 1984;Hauer et al. 1982). However, many enforcement agencies do not have sufficient resources to mount and sustain publicized speed enforcement programs. Traditional speed enforcement also can be difficult, if not hazardous, at some locations and during periods of heavy traffic. About 1 in 10 U.S. drivers reported being stopped for speeding during the past year, CONTACT Wen Hu whu@iihs.org Insurance Institute for Highway Safety,  North Glebe Road, Suite , Arlington, VA . Associate Editor Clay Gabler oversaw the review of this article.
Supplemental data for this article can be accessed on the publisher's website.
even though 70% were identified as habitual or sometime speeders (Schroeder et al. 2013). Widely used around the world to supplement traditional police enforcement, speed cameras monitor traffic speeds and photograph vehicles traveling at or above a speed threshold, usually set well above the speed limit. A systematic review of controlled studies of speed camera effectiveness, mostly conducted in Europe or Australia, reported 14-65% reductions in the percentage of vehicles traveling above the speed limits or at or above designated speed thresholds (Wilson et al. 2010). Speed camera enforcement was associated with reductions in crashes involving fatalities and serious injuries of 11-44% in the vicinity of camera sites and 17-58% over wider areas.
There is less use of speed cameras in the United States and limited research on their effectiveness. The odds of drivers exceeding speed limits by more than 10 mph declined substantially after speed cameras were introduced on residential streets in Montgomery County, Maryland (Retting, Farmer, and McCartt 2008); on a major highway in Scottsdale, Arizona (Retting, Kyrychenko, and McCartt 2008); and on city streets in the District of Columbia (Retting and Farmer 2003). In Scottsdale and Montgomery County, speeds were reduced by smaller amounts at locations not targeted by cameras, suggesting broader spillover effects.
U.S. research on the effects of speed camera enforcement on crashes is scarce. Shin et al. (2009) found substantial reductions in injury crashes, property damage-only crashes, and total crashes during nonpeak periods on the Scottsdale highway where cameras were placed. However, Retting, Kyrychenko, and McCartt (2008) found that the reductions in speeds attributable to the camera program spilled over to some of the control sites in the study of crashes by Shin et al. (2009), so the crash effects estimated by the study were probably underestimated.
The current study extends the earlier evaluation of the Montgomery County speed camera program (Retting, Farmer, and McCartt, 2008). In May 2007, the county implemented an automated speed enforcement programs. Following a 1-month warning period, citations began to be issued. Over the years, the scope of the program has expanded considerably. The current study evaluates the longer term effects of the program on travel speeds, drivers' attitudes, and crashes.

Program description
A suburb of the District of Columbia, Montgomery County, Maryland, had a geographic area of 491 square miles, a population of more than 1 million, and an average yearly median household income of $98,704 during 2010-2014 (U.S. Census Bureau 2016).
Maryland's law allowed cameras to be placed only on residential roadways with speed limits of 35 mph or less or in school zones. Cameras photograph the rear license plate of vehicles exceeding the citation threshold. Citations carry fines of $40 for the registered vehicle owner but no license penalty points. Initially citations were issued for vehicles traveling at least 11 mph above the speed limit. To reflect changes in the state statute allowing the speed camera program, effective October 1, 2009, the speed threshold was changed to 12 mph above the speed limit, and school zone camera operations were restricted to 6 a.m.-8 p.m. on weekdays. In May 2012, some cameras began to be used in a roadway corridor approach in which cameras were periodically moved throughout the length of a roadway segment. This approach aimed to encourage compliance with the speed limit for the entire stretch of the roadway rather than at specific locations only.
The speed camera program gradually expanded from 18 mobile cameras when the program began to 56 fixed cameras, 30 portable cameras, and 6 mobile speed camera vans in 2014. In December 2014 there were 73 speed camera corridors and 61 speed camera sites located outside these corridors. Double-blind checks of the speeds with speed radar were conducted quarterly to ensure the speeds recorded by speed cameras were accurate.
Selection of camera sites was based on several factors including crash data, vehicle speed data, and input from citizen advisory boards. The "Safe Speed" publicity campaign focused initially on the dangers of speeding and the role of speed cameras and then informed drivers that cameras were in use. When initiated, the program received considerable news coverage, and the news media also covered a press conference announcing the corridor approach. Throughout the program, signs advising motorists of photo enforcement have been posted on major roadways entering the county and "photo enforced" placards installed below speed limit and school zone signs on roads designated for cameras. "Speed Camera Corridor" signs are located at entrances to enforcement corridors.

Vehicle speed measurements and analysis
Free-flow, off-peak travel speeds in September-October 2006, about 6 months before camera enforcement began, were compared with speeds in November 2014, about 7½ years after camera enforcement began.
One year in advance of the program, county police identified 40 locations as potential camera sites, and 20 were randomly selected as study sites. Nineteen sites were on residential streets with speed limits of 25-35 mph. The other site, located within a school zone, and one residential street site were excluded from analyses because of road construction or redesigns during the study period. As of November 2014, speed camera signs had been posted on the roads of all 18 remaining sites, and cameras had been deployed near the observation site or on the same road for 16 sites. The 18 sites are hereafter referred to as "camera sites. " To examine potential spillover effects, 10 sites were randomly selected from 20 Montgomery County locations that had similar characteristics (e.g., roadway characteristics, traffic volumes, residential land use) as the potential camera sites but were ineligible for cameras because they had a 40 mph speed limit. Analyses excluded one site due to a roadway redesign.
As controls for the camera sites, speeds were measured at sites on residential streets in Arlington County and Fairfax County, Virginia, counties that are adjacent or proximate to Montgomery County but that have not used speed cameras. All 3 counties are located in the Washington, D.C., metro area and are similar in terms of demographic characteristics and traffic conditions. For example, the average yearly median household income during 2010-2014 was $98,704 in Montgomery County, $105,120 in Arlington County, and $112,102 in Fairfax County (U.S. Census Bureau 2016). They also all are heavily traveled communities. Ten control sites were randomly selected from 20 locations on residential streets in Fairfax and Arlington counties with speed limits of 25-35 mph and roadway characteristics similar to the Montgomery County camera-eligible streets. One control site was excluded due to the addition of speed bumps during the study period.
At all study sites, speeds of passenger vehicles were recorded electronically using speed cameras deployed covertly by photo enforcement vendors not affiliated with the Montgomery County program. At each location, speeds were measured during each study period during 10 a.m.-4 p.m. on a weekday. Despite consistent observation periods at each site in the before and after periods, changes in traffic volume at some sites led to large differences in before and after sample sizes. To ensure consistent representation of each site in the study periods, overall statistics for each site group in the after period were computed as a weighted mean of the statistics for each site, with weights equal to the proportion of vehicles observed at each site during the before period.
Linear regression models estimated changes in mean speeds associated with the speed camera program at camera sites, using the natural logarithm of speeds as the dependent variable. Logistic regression models estimated the effect of the program on the odds of vehicles exceeding speed limits by more than 10 mph. In all models, the independent variables included hourly vehicle counts during observation periods, individual site indicators, a study period indicator (2014 vs. 2006), and an indicator for camera vs. control sites during the after period. Because odds ratios (ORs) derived from logistic regression models are not good approximations for relative risk ratios (RRs) when the incidence of an outcome is not rare in the study population (i.e., greater than 10%), as is true for speeding, odds ratios were transformed into relative risks as RR where P 0 represents the proportion of vehicles exceeding speed limits by more than 10 mph in the before period for the control group (Zhang and Yu 1998). A P value less than .05 was considered statistically significant.
The effects of the speed camera program on speeds at potential spillover sites in Montgomery County were not modeled, due to the lack of appropriate control sites in Virginia.

Telephone survey
To assess awareness and attitudes with regards to the speed camera program, a telephone survey of Montgomery County drivers was conducted in November 2014. The numbers called were selected with random-digit dialing; 31% of these numbers were cellphone numbers. The cooperation rate, defined as the percentage of completed surveys out of the numbers reached, was 36%. Of 2,470 households reached, 36% declined participation, 25% did not qualify, 3% began but did not complete the interview, and 36% completed the interview. Interviews were completed by 900 licensed drivers ages 18 and older. Responses were weighted to reflect the age (18-34, 35-64, and 65+) and gender distribution of the population ages 18 and older of the county in 2014. Significant differences in responses were evaluated using chi-square (χ 2 ) tests of homogeneity (P < .05).

Analyses of police-reported crashes
Police-reported crashes during January 2004-December 2013 in Montgomery County and the control community of Fairfax County were examined, using electronic crash files provided by the Maryland State Police and the Virginia Department of Motor Vehicles. In both Virginia and Maryland, crashes involving injuries or exceeding a specified amount of property damage are reported. Red light cameras were operated at 13 intersections in Fairfax County during the earliest part of the study period (January 2004-June 2005; crashes at these intersections during the entire study period were excluded. Montgomery County used red light cameras throughout the entire study period.
Crashes on all camera-eligible roads in Montgomery County-that is, residential roads with 25-35 mph speed limits whether or not cameras were deployed on them-were compared with crashes on residential roads with 25-35 mph speed limits in Fairfax County. To explore any potential spillover effects, crashes on roads with 40 mph speed limits in Montgomery County, excluding crashes on roads with speed cameras in school zones, were compared with crashes on roads with 40 mph speed limits in Fairfax County.
The before study period was January 2004-April 2007. The after study period, when speed cameras were used, was June 2007-December 2013. Crashes during May 2007 were excluded because only warnings from speed cameras were issued during this month. October 2009-December 2013 represented the after period following the speed camera law change, and June 2012-December 2013 represented the period when the corridor approach was used.
Analyses focused on the proportion of crashes that were coded by police as involving incapacitating or fatal injury. Traffic volume data for roads with different speed limits were not available to use as a measure of exposure, and serious crashes are sensitive to changes in travel speeds. Higher speeds increase the likelihood that a collision will result in serious injuries (Elvik 2005) so that it is reasonable to expect that lower speeds would be associated with a lower proportion of crashes that involve an incapacitating injury or fatality. Crashes involving serious injuries or fatalities have been used in prior studies to evaluate the safety benefits of speed cameras (Wilson et al. 2010).
Logistic regression analyses evaluated the effects of the speed cameras, 2009 law change, and corridor approach on the likelihood that a crash involved an incapacitating injury or fatality on camera-eligible roads and on potential spillover roads. Odds ratios were transformed into estimates of relative risk. The dependent variable was a binary crash indicator (crash involving incapacitating or fatal injury or not). Independent variables were the number of years since 2004; indicators for quarter of year, time of day (9 p.m.-6 a.m. vs. daytime), study group (Montgomery vs. Fairfax County), road surface condition (wet/snowy/icy vs. dry), road alignment (curved vs. straight), and pedestrian involvement (yes vs. no); and study period indicators (entire after period vs. before period, law change period vs. before period, corridor approach period vs. before period). Speed limit indicators (30 vs. 25 mph, 35 vs. 25 mph) also were included in the models of crashes on camera-eligible roads.
The models included interaction variables for study group and study period indicators as measures of the effects of speed cameras, additional effects of the law change over and above camera effects, and additional effects of the corridor approach over and above the camera and law change effects. From estimated parameters for these interactions, changes in the likelihood that a crash involved an incapacitating/fatal injury beyond what would have been expected absent the speed cameras, law change, or corridor approach were calculated (Zhang and Yu 1998). For example, if the parameter for the interaction between study group and the entire camera period vs. before period was −0.2302 in the model of crashes on camera-eligible roads, the odds ratio was calculated as 0.79[exp(−0.2302)]. With 6.8% of crashes involving incapacitating/fatal injuries at control sites, there was an estimated 19.4% reduction in the likelihood that a crash involved an incapacitating/fatal injury compared with the expected likelihood without speed cameras (RR = (0.79/[(1 − 0.068) + (0.068 × 0.79)]). If the parameter for the interaction between study group and the law change vs. before period was 0.0828, and the parameter for the interaction term between study group and the corridor approach vs. before period was −0.3762, the odds ratio was 0.59[exp(−0.2302 + 0.0827 − 0.3762)]. With 6.8% of crashes involving incapacitating/fatal injuries at control sites during the before period, there was an estimated 39.1% reduction in the likelihood that a crash involved an incapacitating/fatal injury compared with the expected likelihood without any of the treatments (RR = 0.59/[(1 − 0.068) + (0.068 × 0.59)]). A P value less than .05 was considered statistically significant.
The Appendix (see online supplement) summarizes logistic regression analyses evaluating the effects of the speed cameras, law change, and corridor approach on the likelihood that a crash was speeding related.

Vehicle speeds
The mean speed and proportion of vehicles exceeding speed limits by more than 10 mph declined for all 3 study site groups from 6 months before to 7½ years after the speed camera program was implemented (Table 1). Percentage declines in the mean speed and the proportion of speeding vehicles at the camera sites were much larger than declines at potential spillover sites (roads with 40 mph speed limits) in the county or at the Virginia control sites.
According to the linear regression model, the mean speed at camera sites declined by 10.2% relative to the Virginia control sites; this reduction in speeds attributable to speed camera enforcement was significant (Table A-1, online supplement). Based on the logistic regression model, the likelihood that a vehicle exceeded the speed limit by more than 10 mph at camera sites decreased by 62% relative to the control sites, a significant reduction (95% confidence interval of relative risk = 0.35, 0.41; Table A-2, online supplement).

Telephone survey
When interviewed in November 2014, about 7½ years after the program began, 56% of Montgomery County drivers agreed that speeding was a problem on residential streets in the county. Almost all drivers (95%) knew that speed cameras were in use on residential streets; 62% supported their use and 38% opposed it. Support for school zone cameras was significantly higher than support for cameras on residential streets (86% vs. 62%), χ 2 = 130.5, P < .0001.
Among drivers who were aware of the camera program, 76% said that camera enforcement had caused them to reduce their speeds on residential streets and in school zones; 59% had received at least one speed camera citation; and 75% knew someone else who received a citation. Fourteen percent were aware that in 2009 the speeding threshold for camera citations was raised from 11 to 12 mph, and 76% said the number of speed cameras had increased over the past several years.

Likelihood that a crash involved an incapacitating or fatal injury
In both Montgomery County and Fairfax County, yearly counts of police-reported crashes on roadways with 25-35 mph speed limits and on roadways with 40 mph limits declined during the study period. The mean yearly crash counts on 25-35 mph limit roadways and on 40 mph limit roadways were 6,005 and 2,299, respectively, in Montgomery County and 7,929 and 1,096, respectively, in Fairfax County.
There also were downward trends in the yearly proportion of crashes that involved an incapacitating/fatal injury on cameraeligible roads and potential spillover roads in Montgomery County and on the associated control roads in Fairfax County (Figure 1). Logistic regression models estimated the effects of speed cameras, the additional effects of the 2009 law change and  the corridor approach, and the effects of other predictors on the likelihood that a crash involved an incapacitating/fatal injury. Separate models were developed for crashes on camera-eligible roads and crashes on potential spillover roads. Table A-3 (see online supplement) provides descriptive statistics of the variables included in the models. Based on these models (summarized in Tables A-4 and A-5, online supplement), the estimated effects of the speed cameras, law change, and corridor approach are summarized in Table 2.
For crashes on camera-eligible roads, based on the interaction between study group and the entire after period vs. before period, the estimated likelihood that a crash involved an incapacitating/fatal injury was 19.4% lower than would have been expected without cameras, a significant difference. The estimated additional effect of the law change over and above the effects of cameras was based on the interaction between study group and the law change vs. before period. The likelihood that a crash involved an incapacitating/fatal injury was 8.0% higher than would have been expected without the law change, a nonsignificant difference. Based on the interaction between study group and the corridor approach period vs. before period, the estimated additional effect of the corridor approach over and above the effects of the speed cameras and law change was a significant 29.9% reduction in the likelihood that a crash involved an incapacitating/fatal injury. The estimated combined effect of the cameras, law change, and corridor approach was based on the 3 interaction terms. The likelihood that a crash involved an incapacitating/fatal injury was an estimated 39.1% lower than would have been expected without any of the treatments, a significant difference.
For crashes that occurred on potential spillover roads, the likelihood that a crash involved an incapacitating/fatal injury was an estimated 26.6% lower than would have been expected without any of the treatments, a significant reduction.
As described in the Appendix, the likelihood that a crash was speeding related was 8.4% and 22% lower than would have been expected without any of the treatments on camera-eligible roads and on potential spillover roads, respectively. Both differences were significant.

Discussion
The current evaluation of the Montgomery County speed camera program found long-lasting significant reductions in speeds on roads with camera warning signs relative to speed changes at the Virginia control sites; cameras were deployed at or near 16 of the 18 sites.
The surveys of county residents provide some insights into the program's effect on drivers' behaviors. The proportion of drivers who viewed speeding as a problem on residential streets was much lower in 2014 (56%) than in surveys conducted 6 months before (71%) and 6 months after (74%) the program began (Retting, Farmer, and McCartt 2008). This is consistent with an increase in the proportion of drivers who said they had reduced their travel speeds due to the cameras (76% in 2014 vs. 59% in the 6 months after survey). Three-quarters of drivers in 2014 said they knew someone who received a speed camera citation, and 59% had received a citation themselves. These responses appear reasonable because more than 500,000 speed camera citations were issued in fiscal year 2015 alone, compared with the county's population of over 1 million. Because receiving a citation is likely to discourage speeding, at least for a while, all of these findings are consistent with the reductions in observed travel speeds associated with the cameras.
Automated enforcement is controversial in many communities. Montgomery County sought to minimize controversy by educating the public about the speeding problem and the safety benefits of speed cameras, and an initial publicity campaign and road signs sought to inform drivers about the camera enforcement, including the enforcement along the special corridors. Although a majority of drivers favored automated speed enforcement on residential streets when interviewed in Fall 2014, a sizeable minority, 38%, opposed it. Support was stronger, at 86%, for cameras in school zones.
The current research is the first known U.S. study of the association between speed camera enforcement on residential streets and crashes. Camera enforcement was associated with a 19% reduction in the likelihood that a crash involved an incapacitating/fatal injury on camera-eligible roads. This is consistent with findings from the extensive international research that automated speed enforcement reduces serious crashes (Pilkington and Kinra 2005;Wilson et al. 2010). The 2009 law change was not significantly associated with a reduction in the severity of crashes. This is not surprising because most respondents in the 2014 survey were unaware of the law change, and the 1 mph increase in the speeding threshold was likely too small to change travel speeds substantially. However, the corridor approach was associated with a 30% reduction in the likelihood that a crash involved an incapacitating/fatal injury, over and above the reduction associated with speed camera enforcement. It should be noted that the period (June 2012-December 2013) for examining effects of the corridor approach was relatively brief. Overall, the estimated combined effect of the speed cameras, law change, and corridor approach was a 39% reduction in the likelihood that a crash involved an incapacitating/fatal injury on camera-eligible roads.
In the analysis of crashes on potential spillover roads in Montgomery County, the estimated combined effect of the speed cameras, law change, and corridor approach was a significant 27% reduction in the likelihood that a crash involved an incapacitating/fatal injury. The apparent spillover effect is consistent with the findings of the international evaluations of speed camera programs (Wilson et al. 2010).
Increasing the perceived risk of detection is an important objective of speed enforcement strategies (Ostvik and Elvik 1991). A countywide effect of camera enforcement on crashes on camera-eligible roads and potential spillover effects on other roads would be expected because of the relatively large scale of the program as well as the high level of awareness of the enforcement, as documented in the 2014 survey of drivers. The effects of the Montgomery County program might be even stronger if periodic publicity campaigns about the program were mounted.
Several limitations of the study are worth noting. Potential spillover effects on vehicle speeds on roads with 40 mph speed limits could not be rigorously examined due to a lack of baseline speed data on a sample of control roads with 40 mph speed limits in Virginia. It was impossible to examine changes in crashes close to the speed camera locations relative to other locations on camera-eligible roads because of the lack of information on the location of crashes relative to camera locations. However, a review of the international literature (Wilson et al. 2010) concluded that the effects of speed cameras on vehicle speeds and crashes extend to wider areas beyond camera sites. The criteria for reporting crashes to police did not change during the study period in Montgomery County. However, in Virginia, the property damage threshold for reporting crashes increased from $1,000 to $1,500 in 2009. If this reduced the number of crashes that were reported, the effects on the likelihood that a crash involved a fatal/incapacitating injury may have been overestimated. However, any effect of the change in reporting crashes would likely have been small because the current threshold still reflects minor vehicle damage. Even though the counties of Fairfax and Montgomery are similar in economic, demographic, and traffic characteristics and the logistic regression models controlled for county differences, it still is possible there are unknown county differences that were not fully accounted for. Bias due to regression to the mean is another possible limitation of the crash analysis, but the likelihood of this was minimized in the study design. Random variation was reduced by focusing on crashes that occurred on all roads with speed limits of 25-35 mph and 40 mph over a number of years, when compared with, for example, crashes at sites with speed cameras over a short time period.
In summary, although automated traffic enforcement is not a panacea for reducing traffic violations, the current study indicates that speed cameras can result in long-term substantial reductions in speeding and crashes involving serious injuries or fatalities. This evidence should be considered by communities considering ways to keep their roads safer.