Incorporating Safety into Transportation Planning for Small and Medium-Sized Communities Teng Wang 10/23/ Program of Study Committee: Dr. Nadia Gkritza Dr. Reg Souleyrette Dr. Alicia Carriquiry
10/23/ Thesis Outline
` 10/23/ Introduction Review of MPOs State-of-the Practice PLANSAFEusRAP–style Risk Mapping Conclusion, Recommendation & Limitation Data Collection & Descriptive Analysis Empirical Bayes
10/23/ Introduction
10/23/ Motivation & Background Summary Safety facts: o over 40,000 fatalities/yr in the US ( ). o about 1,000 crashes per year in Ames ( ). Historically, safety was not explicitly considered in the transportation planning process SAFETEA-LU emphasizes safety performance measures & transparency. Desire to move from “reactive” to “proactive” safety management project-level to systems-level Lack of resources, staff and experience is a challenge to make and monitor progress towards transportation safety goals, need tools new tools and methods are available … do they “fit” the small urban planning model?
10/23/ Research Objective and Tasks Literature Review Data Collection and Descriptive Data Analysis Calibrate safety systems-level predictive PLANSAFE models Develop Empirical Bayes Models for identification of high crash locations usRAP application Conclusions, Recommendations and Limitations
10/23/ Review of MPOs State-of-the Practice
10/23/ Ames Area MPO Population: 50,731 (Census, 2000) 56,510 (Iowa Data Center, 2008) Area: 21.6 square miles (Census, 2000) MPO: 36 square miles, designated in year 2003 Iowa State University : 28,682 students (as of Fall 2010)
10/23/ Ames Area MPO AAMPO Final 2035 Long Range Transportation Plan Chapter 2.2 Goals and Objectives Incorporate strategies to promote safety and security across the entire network. Chapter 10 Safety and Security descriptive crash data analysis such as the crash counts by severities GIS crash density map safety candidate locations roundabouts and access management
10/23/ Summary MPOs' TIP and LRTP CriteriaAAMPOJCCOGCAMPOWVTCLCVMPOBENDMPO Mention Safety PlanningXXXX X Tool or Methodology of Safety Planning Safety Performance Listed in Goals/ObjectivesXXXXXX Consider all Modes of TransportationXXXX X Candidate Sites to be improvedXXXXXX GIS-based Crash MapXX X
10/23/ Data Collection & Descriptive Analysis
10/23/ Crash Data Geocoded crash points: severity, users, and collision type.
10/23/ Fatal and Injury Crashes, 2002 to 2008
14 Crash Statistics, City of Ames, YearTotal CrashesFatalitiesMajor Injuries Minor/Possible Injuries Total Year Total Ames Iowa Percentage % Source: Iowa DOT statewide geocoded crash database
10/23/ Road Network Data Geocoded road network data: functional class, AADT and segment length.
10/23/ Risk Mapping Data Summary (Ames Metropolitan Area ) : Note: Only non-zero AADT road segments are used in the usRAP style risk mapping analysis
10/23/ Socio-Demographics Population, housing, children, working adults, population density, etc.
10/23/ PLANSAFE
10/23/ PLANSAFE: For forecasting and estimating safety due to changes in socio- demographics, traffic demand, road network and planning-level countermeasures. NCHRP 08-44: Incorporating Safety into Long-Range Transportation Planning NCHRP 08-44(02): Transportation Safety Planning: Forecasting the Safety Impacts of Socio-Demographic Changes and Safety Investments
10/23/ Developing PlanSafe SPFs: 1. perform geospatial analysis to assign crashes (points) to the road network (lines) and then assign road network to TAZs (polygons) 2. aggregate crash and road network data to the TAZ-level 3. aggregate census data from block or block group level to the TAZ-level 4. build log linear regression crash frequency models based on the data collected Note: 80 TAZs for the Ames MPO
10/23/ Total Crashes (KABCO) Fatal and Incapacitating Injury Crashes (KA) Fatal and Injury Crashes (KAB) Pedestrian Crashes Bicycle Crashes Property Damage Only Crashes (O) PLANSAFE crash frequency models
10/23/ Due to the small sample size, only two crash frequency models could be calibrated: Total Crashes Minor Injury Crashes
10/23/ Total Crash Frequency Model EQ1: Total crash frequency (per TAZ) = exp( (POP_PAC) (PNF_0214) (POP16_64) (HH_INC)) – 1 As PNF_0214 increases, total crash frequency increases As HH_INC increases, total crash frequency decreases
10/23/ Total Crash Frequency Model
10/23/ Minor Injury Crash Frequency Model EQ 2: Minor injury crash frequency = exp( (TOT_MILE) (HU) (PNF_0214) (POP16_64) (INT) – (HH_INC) – (POPTOT) (ACRE)) – 1 As TOT_MILE increases, minor injury crash frequency decreases. As HU increases, minor injury crash frequency increases. As HH_INC increases, minor injury crash frequency decreases.
10/23/ Minor Injury Crash Frequency Model
10/23/ PLANSAFE Software Analysis User Interface and Analysis Steps: 1. Select Analysis Area and Units 2. Prepare Current Baseline Data 3. Select Target Area 4. Prepare Future Baseline Data 5. Predict Baseline Safety 6. Evaluate Safety Projects
10/23/ PLANSAFE Software Analysis Safety project evaluation results report
10/23/ PLANSAFE Key Findings models are system-level Requires road network, crash and socio-demographic data Software is user friendly
10/23/ Empirical Bayes
10/23/ Why Empirical Bayes (EB)? For MPOs not currently quantifying safety For MPOs using traditional frequency, rate candidate list o to correct for the “regression-to-mean” bias if only “high crash” locations are being considered o to increase the precision of estimation of future crash frequency in the absense of change (for improved B/A analysis) Required: 1. for a road segment or intersection in question: historical crash count/frequency. 2. Data from similar sites to develop or calibrate an appropriate SPF. 3.Regression parameters (in order to weight between observation and mode predictions)
10/23/ EB Estimate of the Expected Crashes for an entity = Weight * Crashes expected on similar entities + (1 – Weight) * Count of crashes on this entity, where 0 ≤ Weight ≤ 1 Where W = weight applied to model estimate μ = mean number of crashes/year from model φ = overdispersion parameter Y = the number of years during which the crash count was taken
10/23/ SPF: general expression for the Negative Binomial regression model where EXP (εi) is a Gamma distribution with mean 1 and variance α. The Negative Binomial regression model has an additional overdispersion parameter Phi (φ) Model Specification Whereμ = number of crashes/year from model L = Length of the road segment in mile e = base of natural logarithms AADT = Annual Average Daily Traffic of the road segment α = Intercept β = parameter for AADT Note: for intersections, L is taken as 1 and AADT = DEV
10/23/ Average crashes for each type of road to build SPFs No. of Ave. Crashes2LArterial2LCollect2LLCOAL4LD4LUFreewayRAMPTotal SPF SPF SPF SPF SPF SPF SPF # of obs Total Length % of Length
10/23/ Summary statistics Model(SPFs) Crash Mean(Variance) Crash Max./Min. AADT Mean(Std Dev.) AADT Max./Min. # of Obs. 2LArterial(SPF02-08)1.9(4.9)12/07200(3100)15,100/ LCollect(SPF08)1.9(10.7)22/03200(2300)8700/5066 2LLCOAL(SPF07- 08)0.30(1.1)12/0680(1030)15,600/6790 4LD(SPF05-08)4.5(57.4)39/010,000(4500)22,700/ LU(SPF07-08)6.0(90.5)47/010,000(4800)24,200/110055
S 10/23/ Negative binomial estimated equations by road type ROADTYPESPFs 2LArterial(SPF02-08)Crashes/year = LENGTH*e *AADT LCollec(SPF08)crashes/year = LENGTH*e *AADT LLOCAL(SPF07-08)crashes/year = LENGTH*e *AADT LD(SPF05-08)crashes/year = LENGTH*e *AADT LU(SPF07-08)crashes/year = LENGTH*e *AADT 1.01
10/23/ Overdispersion parameter Phi (φ) for all each SPFs calibrated on different years of crashes Phi (φ)2LArterial2LCollect2LLCOAL4LD4LUFreewayRAMP SPF N/A SPF N/A SPF N/A SPF N/A SPF N/A SPF N/A SPF N/A # of obs Total Length % of Length These Phi values in bold are the largest Phi values among SPFs in each type of road (Note: for 2 lane arterial (2LArterial), the largest Phi value is from the SPF08, but the variables in the SPF08 model are not significant, so I used the second largest Phi value from SPF02-08 instead).
10/23/ EB Analysis Results EB estimates are calculated by using the largest “Phi value” SPFs
10/23/ EB Analysis Results EB estimates are calculated by using the most comprehensive crash data from to build SPFs
10/23/ EB Analysis Key Findings the EB method is better than the average crash method for predicting crashes, as indicated by the smaller RMSEs. generally, RMSEs become smaller when more years of crash data are used However, this trend only holds for up to 5 years. Crash data over 5 years do not accurately represent the current safety situation for the site EB more accurate than the average crash method. Crash history over 4 years, EB is not preferred.
10/23/ usRAP-style Risk Mapping
10/23/ usRAP Risk Mapping and RPS protocols were investigated for applicability Risk Mapping included four maps: crash density crash rate crash rate ratio potential crash savings usRAP has never been used on urban local roads
10/23/ usRAP-style crash density risk map - crashes per mile - Useful for road authorities to identify high crash segments High: top 5% crash density of all segments by total mileage, medium-high: 5%-15%, medium: 15%-35%, low-medium: 35%-60% Low: 60%-100%.
10/23/ usRAP-style risk crash rate map -Crashes per 100M vehicle mile traveled - Useful to motorists who want to reduce risk
10/23/ usRAP-style risk crash rate ratio map -Rate compared to the “average” road in that class - useful to identify poorly performing roads
10/23/ usRAP-style risk Potential Crash Savings map -the number of total crashes saved per mile in seven years if crash rate were reduced to the average crash rate for similar roads - most useful map
10/23/ usRAP-style Risk Mapping Key Findings Clear and easy to understand can be used to identify road segments that may have the highest potential for improvement (engineering or enforcement)
10/23/ Conclusions & Limitations Conclusions: All three tools may be readily applied Methods provide quantitative, proactive tools PLANSAFE and usRAP can be used at the system-level EB is quantitative – may require more data for reliable models usRAP-style risk mapping provides visual explanation of safety issues Limitations: All three tools require large amounts of detailed data (for Ames, risk maps can be produced with available data – EB and PlanSafe models may require more comparable data to improve models)
10/23/ Recommendations: follow up work with usRAP RPS test of policy sensitivity of PlanSafe presentation of results to MPO staff, decision makers and public for feedback
10/23/ Questions?
10/23/ Other MPOs MPO Database from FHWA: area less than 1,000 sq miles and population up to 140,000 similar area, population with Ames Five MPOs: Johnson County COG (JCCOG) Major City: Iowa City, IA. Area: 89 Sq. Miles Populations: Corvallis Area MPO (CAMPO) Major City: Corvallis, OR. Area: 38 Sq. Miles Population: Wenatchee Valley Transportation Council (WVTC) Major City: Wenatchee, WA. Area: 41 Sq. Miles Population: Lewis-Clark Valley MPO (LCVMPO) Major City: Asotin, WA, ID. Area: 43 Sq. Miles Population: Bend MPO Major City: Bend, OR. Area: 46 Sq. Miles Population: