1 Topic 2 – Network Screening CEE 763 CEE 763 Fall 2011 2 OBJECTIVES Identify locations for further study which have both A high risk of crash losses An economically justifiable opportunity for reducing the risk Identify countermeasure options and priorities which maximize the economic benefits It is as much about exclusion of sites from consideration as it is about inclusion CEE 763 Fall 2011 2 3 NETWORK SCREENING Key tool in a highway safety improvement program Definition– A process which aims to identify locations within the road system where correctable crashes are found in order to develop appropriate and cost-effective treatments to reduce the frequency or severity of crashes CEE 763 Fall 2011 3 4 EFFECTIVENESS It is important to identify sites with the most “promise” for improvement as engineering studies are expensive. Agencies have limited budgets, and if a site with potential is not identified, an opportunity to substantially improve safety is missed. CEE 763 Fall 2011 4 5 SOME TYPICAL NAMES High crash location High accident potential Black spot High risk location Top 5% Crash concentration CEE 763 Fall 2011 5 6 Terms: Site and Facility Site – a basic safety study location, e.g., a segment (homogeneous), an intersection, and a freeway ramp Facility – a contiguous set of sites Freeway (segments, ramps) Urban and suburban arterials (segments, intersections): divided, undivided, signalized, TWSC etc. Rural highway (segments): two-lane, multi-lane HSM only covers predictive methods for certain facility types CEE 763 Fall 2011 7 NETWORK SCREENING PROCESS Establish focus Sites with potential to reduce crash frequency Specific crash types or severity Identify sites and reference population Type of site: segments, intersections, ramps Sites of similar characteristics Select performance measures Select screening method Frequency, rate, severity, etc. Ranking, sliding window, peak searching etc. Screen and evaluate results CEE 763 Fall 2011 8 ESTABLISH FOCUS CEE 763 Fall 2011 8 9 PERFORMANCE MEASURES Crash frequency* Crash rate* Quality control* Excess predicted crash frequency using method of moments Critical rate Crash severity* Equivalent property damage only (EPDO) crash frequency Relative severity index Level of service of safety Excess predicted average crash frequency using SPFs* Probability of specific crash types exceeding threshold proportion Excess proportion of specific crash types Expected crash frequency with EB adjustment* Excess expected crash frequency with EB adjustment CEE 763 Fall 2011 9 10 CRASH FREQUENCY Method Benefits Rank locations with highest count of crashes for investigation Simple Focuses on areas with most crashes limitations Does not account for exposure Favors high-volume, urban locations Engineering fix may not be present CEE 763 Fall 2011 10 11 CRASH RATE Method Benefits Rank locations by rate of crashes Accounts for exposure Relatively simply Efforts focused on potential problem not just high volume locations Limitations Favors low volume, low collision sites Cannot compare cross different volumes CEE 763 Fall 2011 11 12 INTERSECTION RATES Crashes per million entering vehicles (MEV) N Ri MEV TEV n 365 MEV 1,000,000 Ri = intersection crash rate N = number of crashes in the study period n = number of years in the study period TEV = the sum of volumes entering from all approaches, in Average Daily Traffic CEE 763 Fall 2011 12 13 EXAMPLE Observed 46 crashes in two years. The ADT for the minor approach was 3000 and the major approach was 6000. Note - volumes includes both directions. What is the crash rate? CEE 763 Fall 2011 13 14 SEGMENT RATES Crashes per million vehicle miles of travel (MVMT) N Rs MVMT V L n 365 MVMT 1,000,000 Example Observed 40 crashes on a 17.5 mile segment in one year. The ADT was 5,000. CEE 763 Fall 2011 14 15 CRASH AND VOLUME CEE 763 Fall 2011 15 16 FREQUENCY-RATE CRITERIA Method Rank by combination of frequency and rate based methods Various ways to combine rankings for composite rankings Benefits Simple Address drawbacks of both the frequency and rate methods Drawbacks Final ranking dependent of combination CEE 763 Fall 2011 16 17 EXAMPLE Five intersections have the following crash frequency and crash rate. Crash Data Intersections 1 2 3 4 5 Frequency 7 12 4 14 10 Rate 0.5 1.5 2.1 1.0 1.8 If a critical frequency is set at 10, and a critical crash rate is set at 1.5, which intersection(s) should be ranked as high crash locations? CEE 763 Fall 2011 17 18 QUALITY CONTROL Rate or Frequency Method Rank location if the crash rate or frequency at a site is statistically significantly higher than a predetermined rate or frequency for locations of similar characteristics Benefits Based on Poisson distribution Seems to identify locations with possible treatments Drawbacks More data is required Categorization is key CEE 763 Fall 2011 18 19 QUALITY CONTROL Method 1) Select average rate or frequency for similar facility 2) Calculate the critical rate or frequency 3) Compare actual rate or frequency 4) Flag or rank if exceeds Ra 1 RC Ra P M 2M RC = critical rate or critical frequency Ra = the average rate or frequency for similar facility P = probability constant based on desired level of significance (1.645 for 95%) M = millions of VMT or entering vehicles CEE 763 Fall 2011 19 20 EXAMPLE There were 40 observed crashes on a 17.5 mile segment in one year. The ADT was 5,000. Given the average rate for similar segments is 1.02 MVMT, does the subject segment exceed the critical rate at 95% confidence? CEE 763 Fall 2011 20 21 SEVERITY Method Rank locations by weighting the severity of crashes Benefits Adds severity to the frequency method Usually relates to benefit/cost selection Drawbacks Dependent on weighting, may concentrate on fatal collisions Weights are essentially arbitrary since it assigned from global crash costs CEE 763 Fall 2011 21 22 EQUIVALENT PROPERTY DAMAGE ONLY (EPDO) CRASH FREQUENCY EPDO f i Ni i EPDO = Equivalent property damage only crashes fi = weight for crash type I Ni = number of crashes of type i Severity Cost Weight Fatal (K) $4,008,900 542 Injury (A,B,C) $82,600 11 PDO (O) $7,400 1 CEE 763 Fall 2011 22 23 EXAMPLE A location has experienced 2 fatal, 12 injury A, 30 injury B, 40 injury C, and 140 PDO crashes in 5 years. What is the EPDO crashes? • Fatal = $3,400,000 • A = $260,000 • B = $56,000 • C = $27,000 • PDO = $4,000 CEE 763 Fall 2011 23 24 RELATIVE SEVERITY INDEX (RSI) n RSIi RSIi RSI j 1 j Ni = relative severity index cost for intersection i RSIj = relative severity index cost for crash type j Crash Type Number of Cost per Crash Crashes Rear End 19 $13,200 Sideswipe 7 $34,000 Angle 5 $61,100 Fixed Object 3 $94,700 CEE 763 Fall 2011 24 25 RSI EXAMPLE An intersection has the following crashes. Determine the RSI for this intersection Crash Type Number of Crashes Cost per Crash Rear End 19 $13,200 Sideswipe 7 $34,000 Angle 5 $61,100 Fixed Object 3 $94,700 CEE 763 Fall 2011 25 26 SAFETY INDICES Method Benefits Rank locations by creating an index which includes a number of factors such as rates, frequencies, severities, and possibly site data. A weighted average or scores are then combined to calculate a composite index. The “Relative Severity Index” discussed earlier is one of these types. Simple and attempts to combine criteria Drawbacks Rank is sensitive to weights of scores which are usually assigned “arbitrarily” CEE 763 Fall 2011 26 27 *ODOT SAFETY PRIORITY INDEX SYSTEM (SPIS) Composite score assigned for frequency, severity, and rate 3 years data, 0.10 mile sections Maximum index is 100 • 25 points max for frequency • 25 points max rate • 50 points max severity Total score = Sum of Indicator values (IV) of Frequency, Rate, and Severity CEE 763 Fall 2011 27 28 *SAFETY PRIORITY INDEX SYSTEM IVFreq LOGTotalCrashes 1 25 min 25, LOG150 1 Note: Max SPIS score is 100 TotalCrash es 1 , 000 , 000 LOG 1 3 yr 365days ADT 25 IVRate min 25, LOG7 1 100FATAL INJ A 10INJ B INJC PDO 50 IVSeverity min 50, 300 CEE 763 Fall 2011 28 29 EXAMPLE 0 Fatal, 1 A, 0 B, 3 C, 4 PDO. ADT 14,200. IVFreq LOGTotalCrashes 1 25 min 25, LOG150 1 TotalCrash es 1 , 000 , 000 LOG 1 3 yr 365days ADT 25 IVRate min 25, LOG7 1 100FATAL INJ A 10INJ B INJC PDO 50 IVSeverity min 50, 300 CEE 763 Fall 2011 29 30 EXAMPLE 0 Fatal, 1 A, 0 B, 3 C, 4 PDO. ADT 14,200. LOG8 1 25 10.95 IVFreq LOG150 1 81,000,000 1 LOG 3 yr 365days14,200 25 4.99 IVRate LOG7 1 1000 1 103 4 IVSeverity 50 22.33 300 Answer: SPIS Score = 38.27 CEE 763 Fall 2011 30 31 POTENTIAL ACCIDENT REDUCTION Method Benefits Rank or flag locations where the difference between observed and expected crash experience will maximize benefits if their crash history can be reduced to the expected value. Most uses frequency rather than rates Can account for “regression to the mean” Drawbacks Data hungry, expected values must be predicted CEE 763 Fall 2011 31 32 EXCESS PREDICTED CRASH FREQUENCY USING METHOD OF MOMENTS Calculate average crash frequency per reference population Calculate crash frequency variance n VAR i 1 obs ,i N pa ) 2 n 1 Calculate adjusted observed crash frequency per site N adj N obs (N N pa VAR ( N pa N obs ) N adj adjusted N pa populationaverage N obs observed Calculate potential for improvement (PI) per site PIi Nadj N pa Rank site according to PI (highest to lowest) CEE 763 Fall 2011 32 33 EXAMPLE An unsignalized intersection has observed 11 crashes in a year. Suppose among all the unsignalized intersections, the average crashes per year is 8, and the standard deviation of crash for all the intersections is 3. Calculate the PI for this intersection. CEE 763 Fall 2011 33 34 EXCESS PREDICTED CRASH FREQUENCY USING SAFETY PERFORMANCE FUNCTIONS Calculate expected crash frequency using SPF Calculate excess predicted average crash frequency Nexcess,i Nobs ,i Nexpected Rank site according to the excess frequency CEE 763 Fall 2011 34 35 EXAMPLE An unsignalized intersection has observed 11 crashes in a year. According to the SPF developed for all the unsignalized intersections, the predicted crash frequency per year is 8. What is the excess predicted crash frequency? CEE 763 Fall 2011 35 36 EMPIRICAL BAYES METHODS E{ k / K } E( k ) ( 1 )K 1 VAR{ k } 1 Y E{ k } 1 1 YE(k ) / Crash Frequence K - Observed # of crashes E{k/K} is best estimate for the expected # of crashes SPF E(k) -Modeled # of crashes E(k) is the predicted value at similar sites, in crash/year Y is the analysis period in number of years Volume φ is over-dispersion factor CEE 763 Fall 2011 36 37 SAMPLE DATA CEE 763 Fall 2011 38 SAMPLE DATA CEE 763 Fall 2011 39 CRASH FREQUENCY WITH EB ADJUSTMENT Step 1 – Calculate the predicted average crash frequency using an SPF Step 2 – Calculate annual correction factor Cn N predicted,n N predicted,1 Year Predicted Average Correction factor 1 2 3 2.5 2.5 2.7 1.0 1.0 ? CEE 763 Fall 2011 40 CRASH FREQUENCY WITH EB ADJUSTMENT Step 3 – Calculate EB weighting factor, Note: rely on dispersion factor or variance. 1 1 YE( k ) / 1 N 1 1 / N predicted ,n n 1 Year Predicted Average 1 2 3 2.5 2.5 2.7 Dispersion factor 1 / 0.49 1 N 1 1 / N predicted ,n ? n 1 CEE 763 Fall 2011 41 CRASH FREQUENCY WITH EB ADJUSTMENT Step 4 – Calculate first year EB adjusted average crash frequency. N N expected,1 N predicted,1 ( 1 ) N n 1 observed ,n N C n 1 n Year Predicted Average Observed Crashes 1 2 3 2.5 2.5 2.7 11 9 14 Nexpected,1 ? CEE 763 Fall 2011 42 CRASH FREQUENCY WITH EB ADJUSTMENT Step 5 – Calculate final year EB adjusted average crash frequency. Nexpected,n Nexpected,1 * Cn Step 6 – Calculate the variance (optional) Cn Var( n year ) N expected,n * ( 1 ) Cn th Step 7 - Rank sites based on the EB adjusted expected average crash frequency for the final year. CEE 763 Fall 2011 43 OTHER CRITERIA Level of service safety (LOSS) Konokov et al. (Colorado DOT) Method of moments PIARC manual Proportions testing Exceeding a particular crash type Rank locations bases on the current annual cost of crashes based on average cost of crash by accident type CEE 763 Fall 2011 43 44 WHICH CRITERIA TO USE? Little consensus on methods The key issue is how the criteria adopted direct the analyst to consider sites which contributes to the overall road safety goal, namely the maximization of benefits of road safety treatments CEE 763 Fall 2011 44 45 METHOD USAGE All of the methods are in use either alone or in combination In US states • Crash frequency by 15% • Crash rate or RQC by 15% of agencies • Crash severities by 50% of agencies • Indices by 18% • Other by 16% CEE 763 Fall 2011 45 46 MORE PRECISE DEFINATION OF SITE Three alternatives (Hauer et al., TRR 1784 – Screening the road network for sites with promise) Based on “Section” Based on a uniform length of a roadway, e.g., 0.1 mi Based on a minimum segment that identifies the highest accident frequency while satisfying the statistical limits (i.e., CV). CEE 763 Fall 2011 47 SEARCHING ALGORITHMS Expected Segment average Segment average does not correspond to the highest Expected Segment average CEE 763 Segments of different length with the highest crash Fall 2011 48 SLIDING WINDOW 0.3-mile window with 0.1 increment Roadway Segment 0.0 mi 0.1 mi 0.2 mi 0.3 mi 0.4 mi 0.5 mi 0.6 mi Win # 1 Win # 2 Win # 3 Win # 4 *The window that has the highest risk is used to rank the segment. CEE 763 Fall 2011 49 EXAMPLE A roadway network has ten segments composed of three types of facilities. Using the sliding window method and the crash rate to rank Segments 1 and 2. CEE 763 Fall 2011 50 More Data Segment 1 starts at mile post 1.2 and ends at 2.0. Segment 2 starts at mile post 2.0 and ends at 2.4. Segment 1 1.2 1.3 1.4 CEE 763 1.5 Segment 2 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3 Fall 2011 2.4 51 SLIDING WINDOW 0.3-mi window with 0.1-mi increment Site No. 1 Second Sliding Window W = 0.3 mi MP 1.0 0.1 mi MP 2.6 Sliding window is moved incrementally by 0.1 mi along the roadway segment. 0.2 mi 0.3 mi 0.4 mi 0.5 mi First Sliding Window W = 0.3 mi CEE 763 Fall 2011 52 SLIDING WINDOW 0.3-mi window with 0.1-mi increment Site No. 11 MP 21.4 Site No. 12 MP 22.2 MP 23.0 Sliding Window W = 0.3 mi CEE 763 Fall 2011 53 SLIDING WINDOW 0.3-mi window with 0.1-mi increment Site No. 23 MP 35.4 Site No. 24 MP 36.2 0.1 mi MP 36.37 Site No. 25 MP 36.7 0.17 mi 0.03 mi Sliding Window Concepts: Bridging Three Contiguous Roadway Segments CEE 763 Fall 2011 54 SLIDING WINDOW 0.3-mi window with 0.1-mi increment A Site No. 31 Site No. 32 MP 53.5 MP 54.3 0.2 mi B Site No. 32 MP 54.3 0.1 mi C Site No. 31 MP 53.5 MP 54.48 0.1 mi Site No. 31 MP 53.5 Site No. 33 Site No. 33 MP 54.48 0.18 mi Site No. 32 MP 54.3 Site No. 33 MP 54.48 0.3 mi Sliding Window Concepts: Window Positions at the End of Contiguous Roadway Segments When Window is Moved Incrementally by 0.1 Miles CEE 763 Fall 2011 55 SLIDING WINDOW 0.3-mi window with 0.1-mi increment Site No. 22 MP 35.5 Site No. 24 Site No. 23 MP 35.6 MP 36.2 Window No. 1 Site No. 25 MP 36.7 Window No. 9 Window No. 2 Window No. 10 Window No. 3 Window No. 11 Window No. 4 Window No. 12 Window No. 5 Window No. 13 Window No. 6 Window No. 7 Window No. 8 Sliding Window Concepts: Example of Position and Location of Sliding Windows and Subsegments CEE 763 Fall 2011 56 SLIDING WINDOW 0.3-mi window with 0.1-mi increment Site No. 22 MP 35.5 Site No. 24 Site No. 23 MP 35.6 Site No. 25 MP 36.2 Sum( X Y ( EPDO ) ) 36.74 Window No. 1 Window No. 2 Window No. 12 Sum( X Y ( EPDO ) ) 34.51 Window No. 6 Sum( X Y ( EPDO ) ) 33.11 Window No. 11 Sum( X Y ( EPDO ) ) 45.96 Window No. 5 Sum( X Y ( EPDO ) ) 39.28 Window No. 10 Sum( X Y ( EPDO ) ) 50.43 Window No. 4 Sum( X Y ( EPDO ) ) 44.85 Window No. 9 Sum( X Y ( EPDO ) ) 37.50 Window No. 3 Sum( X Y ( EPDO ) ) 24.11 Window No. 13 Sum( X Y ( EPDO ) ) 34.51 Sum( X Y ( EPDO ) ) 39.28 Window No. 7 Note: Sum(XY(EPDO)) expressed as acc/mi MP 36.7 Sum( X Y ( EPDO ) ) 36.25 Window No. 8 Sum( X Y ( EPDO ) ) 46.85 limiting value: 40 acc/mi/yr Sliding Window Concepts: Ranking Example CEE 763 Fall 2011 57 EXAMPLE A segment with 2 lanes, rural ADT= 6000 Limiting frequency: 10 SPF: Intercept:-3.63 ADT coefficient: 0.53 Over dispersion Parameter: 0.5 CEE 763 Fall 2011 58 EXAMPLE 0.043 0.147 0.231 0.231 0.240 0.251 0.251 0.287 0.287 0.287 0.310 0.311 0.325 0.329 0.433 0.434 0.440 0.440 0.440 0.441 0.452 0.454 0.483 0.493 CEE 763 Accident locations (mile) 0.533 0.598 0.636 0.636 0.658 0.743 0.806 Site A: 0-0.4 mile 0.806 0.808 0.822 0.823 0.848 0.862 Site C: 0.9-1 mile Non contiguous 0.862 Site B: 0.4-0.9 mile Contiguous 0.901 0.948 0.983 Fall 2011 59 PEAK SEARCHING 0.1-mile window Roadway Segment 0.0 mi 0.1 mi 0.2 mi 0.3 mi 0.4 mi 0.5 mi Win # 2 Win # 3 Note: Window length = 0.1 mi 0.07 mi 0.03 mi Win # 1 0.6 mi 0.67 mi Win # 4 Win # 5 Win # 6 Win # 7 CEE 763 Fall 2011 60 PEAK SEARCHING 0.2-mile window Roadway Segment 0.0 mi 0.1 mi 0.2 mi 0.3 mi 0.4 mi 0.5 mi 0.6 mi 0.67 mi 0.07 mi 0.03 mi Win # 1 Win # 2 Win # 3 Note: Window length = 0.2 mi Win # 4 Win # 5 Win # 6 CEE 763 Fall 2011 61 PEAK SEARCHING 0.4-mile window Roadway Segment 0.0 mi 0.1 mi 0.2 mi 0.3 mi 0.4 mi 0.5 mi 0.6 mi 0.67 mi Win # 1 Win # 2 Win # 3 Win # 4 Note: Window length = 0.4 mi CEE 763 Fall 2011 62 EXAMPLE A roadway segment is 0.47 miles long. Using a window length of 0.1 miles, the following crash data were obtained for each sub-segment. Calculate the CV for each sub-segment, and determine whether the search should continue with longer window sizes (assume the limiting CV is 0.25). CEE 763 Fall 2011 63 EXAMPLE-continued Sub-segment Position Excess Expected Crash Frequency B1 0.00-0.20 6.50 B2 0.10-0.30 4.45 B3 0.20-0.40 3.80 B4 0.27-0.47 7.15 CEE 763 C.V. Fall 2011