4/28/2016 EVALUATING THE SAFETY EFFECTS OF INTERSECTION SAFETY DEVICES AND MOBILE PHOTO ENFORCEMENT AT THE CITY OF EDMONTON Karim El Basyouny PhD., Laura Contini M.Sc. & Ran Li, M.Sc. City of Edmonton Office of Traffic Safety Introduction 2 1
Introduction Fixed Enforcement at Intersections ISD Cameras Estimate the collision reduction effect at both an Intersection level and approach specific level Investigate I t the spillover effects to unenforced approaches Identify intersection characteristics that impact ISD success Mobile Photo Enforcement on Arterials Estimate the collision reduction effect on enforced arterial segments Compare different enforcement strategies (continuousvs. discontinuous) Investigate the spillover effects on unenforced adjacent approaches 3 Intersection Safety Devices Laura Contini im.sc. City of Edmonton Office of Traffic Safety 4 2
ISD Evaluation Contents Introduction Background Methodology Results Conclusion What are Intersection Safety Devices? Collisions at Intersections Summary of Previous Works Before and After Evaluation Data Description Intersection Level Results Approach hlevel lresults Summary of Findings Further Benefits of ISD 5 ISD Research Breakdown Safety evaluation of Intersection Safety Devices (ISD) Intersection Level Analysis Approach Level Analysis Spillover Effects Site Selection Criteria 6 3
Intersection Safety Cameras Introduction Red light Cameras (RLC) Targets drivers who enter the intersection after the red light Intersection Safety Devices (ISD) Targets drivers who are speeding through the intersection and drivers who enter the intersection after the red light 7 Background Problem of Red light Running and Speed RLR is a common cause of intersection collisions RLR collisions tend to result in right angle collisions RLR occurs in 30 35% of signal phase cycles 1,2 Drivers who are speeding are more likely to run a red light 1. Porter, B.E., and K.J. England, Predicting red-light running behavior: a traffic safety study in three urban settings. Journal of Safety Research 3, 2000, pp.1 8. 2. Retting, R.A., and A.F. Williams. Characteristics of red light violators: results of a field investigation. Journal of Safety Research 27, no. 1, 1996, pp. 9-15. 8 4
Background Problem of Red light running and Speed Speeding increases the risk of collision as well as the severity of collisions At higher speeds there is less time to observe, react and maneuver to avoid collisions Due to increased kinetic energy, collisions severity increases at higher h speeds In Edmonton 57% of all collisions and 72% of injury collisions occur at intersections 9 RLC and ISD Previous Research Background Several studies have been conducted regarding RLC safety effectiveness, but few have been focused on ISD RLC and ISD evaluations tend to focus on changes in overall collisions, severe, angle collisions and rear end collisions Most studies focus on intersection level changes, not approach level Current research generally shows a reduction in angle collisions (6 46%), but an increase in rear end collisions (9 44%) 10 5
Data Description Methodology Traffic Volumes Traffic volumes were required for each year Missing data was extrapolated based on population changes Geometric Data Collected using google maps and city CAD database Information included lane width and configuration, intersection size, etc. Collision Data Collision history was collected for each site Information included collision cause, severity and travel direction Collision data was aggregated into 5 collision categories 11 Collision Categories Methodology Collision Seve erity Total Collisions Severe Collisions Property Damage Only (PDO) Collisions Sum of PDO collisions and Severe collisions All fatal and injury collisions Collisions which do not result in an injury or fatality Collision Type Rear end Collisions Angle Collisions Collisions in which a leading vehicle is stuck by the following vehicle Collisions involving at least two vehicles traveling in perpendicular directions 12 6
Methodology Safety Performance Functions Methodology Safety Performance Functions (SPF) are equations used to predict the number of collisions which will occur Variables include: Traffic volumes Number of lanes Average lane width Lane separation Other roadway characteristics The SPF is created using data from a group of non-treated sites 13 Reference intersections were used to create SPF Methodology These sites should be similar to treated t sites A buffer was used to avoid affects from ISD spilling-over to reference sites 14 7
Before and After Safety Evaluation Background Empirical Bayes (EB) Before and After Number of observed collisions compared to the number of expected collisions in the after period Accounts for sources bias 15 Background Safety Evaluation Methods Sources of Bias Regression to the Mean the (RTM) o The statistical phenomenon collision numbers fluctuate o Treatment sites are generally chosen due to high collision frequencies Overestimating! 16 8
Background Safety Evaluation Methods Sources of Bias Changing gcollision Trends Changes in collision frequency due to general trends across all locations Example population change or level of enforcement, etc. Confounding Factors Temporary events or conditions that would impact collisions Examples include weather patterns, new collision countermeasures, etc. 17 Results & Discussion Intersection Level Results Approach Level Results Spillover Effects Site Selection Criteria 18 9
Results & Discussion Intersection Level Analysis Total Collisions Severe Collisions PDO collisions Rear End Collisions Angle Collisions % Collision Reduction 25.47* 3.99 6.35* 10.74* 33.44* * Significant at 95% level Reductions were observed in all the collision categories Large reductions in Total and Angle collisions Unlike most other ISD or RLC studies, significant reductions were found in rear-end collisions 19 Results & Discussion Approach Level Analysis ISD cameras are not located on all intersection approaches In this analysis only approaches with cameras are included Collision Reductions Total Collisions * Significant at 95% level Severe Collisions PDO collisions Rear End Collisions Angle Collisions 11.55* 3.06 11.74* 13.63* 43.06* 20 10
Results & Discussion Approach Level Analysis Spillover Effects Spillover effects are changes at untreated locations cause by the treatment If collisions have been reduced at non ISD approaches it may indicate spillover effects Some previous studies of RLC and ISD cameras suggests that there are not strong spillover effects 21 Results & Discussion Approach Level Analysis Spillover Effects ISD Approaches Non ISD Approaches (No Signs) Non ISD Approaches (Signs) Total Collisions 11.55* 4.30 430 4.18 418 Severe Collisions 3.06 33.01 2.10 PDO collisions 11.74* 1.17 8.11 Rear End Collisions 13.63* 0.29 3.64 Negative sign indicates increase in collisions Angle Collisions 43.06* 79.84* 35.55* * Significant at 95% level o o The reductions are much greater at the ISD equipped locations suggesting that spillover to other approaches may not occur at the approach level There were significant reductions in angle collisions at all approaches, suggesting that spillover effects may occur for angle collisions 22 11
Results & Discussion Approach Level Analysis Site Selection Criteria Current ISD sites are chosen based on a history of high collision frequency high collision frequency Collision How reduction can we is prioritize not uniform future across ISD sites? sites, some experience greater reductions Site selection criteria is evaluated in two ways in order to identify factors which influence ISD success. 23 Results & Discussion Site Selection Criteria Collision Frequency Collision Rate Traffic Volume Threshold Total Severe Angle Collisions Collisions Collisions Low 10.17 33.10 35.35* Medium 16.80* 9.00 49.90* High 26.39* 29.06* 47.90* Low 6.33 37.91 40.10* Medium 18.84* 5.99 54.05* High 22.71* 19.44 29.54 Low 24.74* 507 5.07 41.83* Medium 2.10 31.03 54.68* High 10.02 13.53 37.64* ocollision reductions were greatest at sites with high collision frequency (> 15 yearly collisions) ono clear trend between traffic volume and collision reduction was observed osignificant reduction in severe collisions i at high h collisions i locations * Significant at 95% level 24 12
Results & Discussion Site Selection Intersection Characteristics o Models were created to relate intersection characteristics to collision reductions o Independent variables related to approach characteristics - number of lanes, speed limit, right turn separation. Total Severe PDO Rear End Angle Collisions Collisions reductions collisions were Collisions Collisions Intercept 1.98 greater 2.725 at approaches 1.384 with more through lanes 2.854 1.258 Number of Lanes 0.142 0.135 0.190 0.704 Speed Limit 0.041 0.045 0.031 0.058 Right Turn Separation 1.29 Collision reductions were greater at approaches with lower speed limits 25 Results & Discussion Site Selection Intersection Characteristics Results suggest reductions were greatest at sites with: Speed limits 60 km/hr Separated right turn lane >5 lanes Speed Limit Right Turn Lane Separation Number of Lanes Threshold 50 km/hr 60 km/hr 70 km/hr No Yes 2 4 5 7 Group Size 7 25 8 8 32 19 21 Total 2.90 20.50* 12.79 14.37 11.22* 10.95 11.95 * Angle 27.77 46.14* 46.21* 0.36 49.87 * 28.00 53.89 * * Significant at 95% level 26 13
170 St & 87 Ave Southbound Approach Results & Discussion Average 15 collisions per year 6 lanes Separated right turn lane 60 km/hr speed limit 65% reduction in total collisions 27 Further enforcement opportunity right turn on red Further ISD Benefits Diagnosis of problem behavior It was discovered that vehicles in the 3 rd lane had a higher proportion of red-light running so additional signal heads were added Collision investigations 28 14
Automated Mobile Speed Enforcement on Urban Arterial Roads Ran Li M.Sc. City of Edmonton Office of Traffic Safety 29 Previous Works Study Carnis & Blais (2013) Enforcement Program French Speed Camera Program Major Findings 21% reduction in fatality rate Diminishing effects on injury collisions Newstead (2001) Queensland Random Road Watch 31% reduction in fatal collisions Benefit-cost ratio reached 55:1 Goldenbeld & Mobile Enforcement 21% reduction in injury collisions Schagen (2005) in a Dutch Province 21% reduction in serious traffic casualties Chen et al. (2000) Photo Radar Program in British Columbia 25% reduction in daytime speed-related collisions and 17% in daytime fatalities 30 15
Previous Works Results Summary Consistent effectiveness of automated mobile enforcement The results revealed 17% to 31% collision reductions More reductions in fatal and injury collisions 31 Gaps in the Literature More focused on the system wide collisions rather than segment based collisions Deficienciesin in methodology (RTM, traffic volume, trend, and other confounding factors) Little discussion on enforcement strategy and spillover effect on the opposite approach 32 16
DATA DESCRIPTION 33 Data Description Study Period: January 2005 December 2012 Road Segment (93 enforced 266 reference): Deployment Hour Traffic Counts (AADT) Segment Length Road Geometry Data Unsignalized Intersection Density Median Midblock Collision Data Severe Collision Property Damage Only Collision (PDO) Total Collision Speed-related PDO Collision Speed-related Collision 34 17
Data Description GIS Data Collection 35 Deployment Hour Data Description ID 2005 2006 2007 2008 2009 2010 2011 2012 1 35.81667 55.31667 17.1 2 2 109.4667 5.4 3 2.766667 16.2 1.783333 4 16.78333 1.316667 1 5.083333 5 7.3 27.28333 6 14.08333 28.9 1.5 7 14.36667 7.15 8 23.68333 6.733333 9 Before Period After Period 24.05 10 81.35 99.05 41.91667 11 94.15 93.53333 48.93333 38.63333 12 11.85 8.3 13 13.48333 8.633333 14 47.06667 147.4667 160.3333 54.93333 15 52.88333 16.76667 18.41667 43.18333 16 126.4667 86.71667 167.0833 218 17 108.6 103.8 117.9 145.2667 18 13.25 39.93333 11.55 83.18333 70.05 16.06667 36 18
METHODOLOGY 37 Safety Performance Function (SPF) Generalized linear model (negative binomial) Model form: 1 2 V L exp( Mdi 0 3UNSD 4Median ) V AADT, L Length, UNSD unsignalized intersection density, Median presence of median dummy Selection of reference group: Arterial segments with similar AADT and collision No enforcement and not adjacent to enforced segments Calibration: SAS GENMOD procedure GOF: scaled deviance and Pearson 38 19
Yearly Calibration Factor (YCF) Annual fluctuation confounding factors Weather conditions, roadway improvement, general trend in safety, etc. Assumption the impacts of the confounding factors on collision variation are similar for both the reference segments and the enforced segments. Calculation (by year by collision type) YCF All sites All sites Observed Predicted Collisions Collisions Adjusted predicted number of collisions: YCF 39 Empirical Bayes Method (EB) Step 1: expected number of collisions for before period E B w (1 w) Obs B B Step 2: expected number of collisions for after period A EA EB B Step 3: overall odds ratio of collision reduction N sites ( All sites A E A All All sites 1 Var E A) ( E All sites A ) 2 40 20
Methods Comparison Before-After Methods Naïve Comparison Group Empirical Bayes Regression-to-the-Mean Effect General Trend of road safety External Factors Change in Traffic Volume 41 RESULTS 42 21
SPF Estimation Results All the SPFs fit the data well Most parameters are statistically significant Severe PDO Total Speed-related Speed-related Collision Collision Collision PDO Collision Collision Intercept -10.25* -5.87* -6.00* -6.48* -6.73* AADT 1.05* 0.74* 0.78* 0.75* 0.81* Length 044* 0.44 036* 0.36 038* 0.38 040* 0.40 041* 0.41 UNSD 0.06* 0.07* 0.07* 0.07* 0.06* Median -0.28* -0.32* -0.31* -0.15-0.18** Dispersion Parameter 0.38* 0.34* 0.34* 0.34* 0.34* * Significant at 0.01 level ** Significant at 0.05 level 43 Overall Evaluation Reduction Percentage by Collision Severity/Type 25% 20% 20.1% 17.9% 18.5% Collision Reduction (%) 15% 10% 14.3% 14.5% 5% 0% Severe Collision PDO Collision Total Collision Speed-Related PDO Collision Speed-Related Collision 44 22
Overall Evaluation Reduction Percentage by Collision Severity/Type 25% 20% 20.1% 17.9% 18.5% Collision Reduction (%) 15% 10% 14.3% 14.5% 5% 0% Severe Collision PDO Collision Total Collision Speed-Related PDO Collision Speed-Related Collision 45 Overall Evaluation Site Selection Criteria Collision frequency Collision rate AADT Deployment Hours Total hours Yearly average hours Example: Total Collisions Example: Total Collisions Collision Frequency < 1.5 [1.5, 3.5) 3.5 Total Hours < 15 [15, 70) 70 2.9% 7.1% 19.4%* 13% 8.5% 18.9%* * Significant at 0.05 level * Significant at 0.05 level 46 23
Enforcement Strategy Continuous vs. Discontinuous Strategies Collision Reduction (%) ID 2005 2006 2007 2008 2009 2010 2011 2012 Reduction Percentage by Collision Severity/Type 1 35.81667 55.31667 17.1 2 2 109.4667 5.4 35% 3 32.1% 2.766667 16.2 1.783333 4 28.7% 16.78333 1.316667 1 5.083333 30% 5 27.7% 27.3% 7.3 27.28333 26.7% 6 25% 14.08333 28.9 1.5 7 14.36667 7.15 20% 8 23.68333 6.733333 17.9% 9 15.0% Continuous 24.05 15% 10 81.35 13.4% 99.05 Discontinuous 41.91667 11 8.5% 8.6% 94.15 93.53333 48.93333 38.63333 10% 12 11.85 8.3 13 13.48333 8.633333 5% 14 47.06667 147.4667 160.3333 54.93333 0% 15 52.88333 16.76667 18.41667 43.18333 16 Severe Collision PDO Collision Total Collision Speed-Related 126.4667 86.71667 Speed-Related 167.0833 218 17 PDO Collision 108.6 Collision 103.8 117.9 145.2667 18 13.25 39.93333 11.55 83.18333 70.05 16.06667 47 Enforcement Strategy Continuous vs. Discontinuous Strategies Reduction Percentage by Collision Severity/Type 35% 32.1% 30% 28.7% 27.7% 27.3% 26.7% Collision Reduction (%) 25% 20% 15% 10% 17.9% 8.5% 8.6% 15.0% 13.4% Continuous Discontinuous 5% 0% Severe Collision PDO Collision Total Collision Speed-Related PDO Collision Speed-Related Collision 48 24
Spillover Effects Unenforced Adjacent Approach (39 pairs) Enforced Segment Unenforced Adjacent Approach 49 Spillover Effects Enforced vs. Unenforced Segments Reduction Percentage by Collision Severity/Type 30% 25% 26.1% Collision Reduction (%) C 20% 15% 10% 14.6% 14.1% 15.4% 9.2% 14.0% 11.1% Enforced Unenforced 5% 2.5% 1.8% 4.5% 0% Severe Collision PDO Collision Total Collision Speed-Related PDO Collision Speed-Related Collision 50 25
Spillover Effects Enforced Approach Reduction Percentage by Collision Severity/Type 30% 25% 26.1% Collision Reduction (%) C 20% 15% 10% 14.6% 14.1% 15.4% 9.2% 14.0% 11.1% Enforced Unenforced 5% 2.5% 1.8% 4.5% 0% Severe Collision PDO Collision Total Collision Speed-Related PDO Collision Speed-Related Collision 51 Spillover Effects Unenforced Adjacent Approach Reduction Percentage by Collision Severity/Type 30% 25% 26.1% Collision Reduction (%) C 20% 15% 10% 14.6% 14.1% 15.4% 9.2% 14.0% 11.1% Enforced Unenforced 5% 2.5% 1.8% 4.5% 0% Severe Collision PDO Collision Total Collision Speed-Related PDO Collision Speed-Related Collision 52 26
CONCLUSIONS 53 Conclusions ISD Summary Significant Collision Reductions in the following categories Intersection level 24% Total collisions 33% Angle collisions 11% Rear end collisions Approach level 12% Total collisions 43% Angle collisions 14% Rear end end collisions 29% Severe collisions locations with high collision frequency (>15) 54 27
Conclusions ISD Summary Site Selection Collision frequency was found to be successful as site selection criteria Traffic volume was not a recommended site selection criteria Intersection characteristics related to greater reductions at ISD approachesincluded: numberof lanes, speedlimit and right turn lanes 55 Conclusions 20% reduction in severe collisions and 15% reduction in total collisions Continuous strategy is able to achieve better results than discontinuous strategy 15% reduction in PDO collisions on adjacent unenforced segments 56 28
Conclusions Speed effects of enforcement Safety effects on other types of road Enforcement effectiveness models Drivers attitude towards enforcement 57 Thank You Ran Li, M.Sc. City of Edmonton Office of Traffic Safety Ran.Li@edmonton.ca Laura Contini, M.Sc. City of Edmonton Office of Traffic Safety Laura.Contini@edmonton.ca 29