Rational road safety management Rational road safety management Practice and TheoryPractice and Theory

Bhagwant PersaudBhagwant PersaudRyerson UniversityRyerson UniversityToronto, CanadaToronto, Canada

OverviewOverviewRoad infrastructure safety Road infrastructure safety management processmanagement process

Evolution of road Evolution of road infrastructure safety infrastructure safety managementmanagement

Current management Current management initiativesinitiatives

Highway Safety ManualHighway Safety ManualAASHTO strategic highway safety AASHTO strategic highway safety planplan

Current research directionsCurrent research directionsTheoretical challengesTheoretical challengesApplications Applications

Future directionsFuture directions

THE ROAD INFRASTRUCTURE SAFETY THE ROAD INFRASTRUCTURE SAFETY MANAGEMENT PROCESSMANAGEMENT PROCESS

NETWORK EVALUATIONNETWORK EVALUATION

SAFETY CONSCIOUS PROCEDURESSAFETY CONSCIOUS PROCEDURES

IMPROVEMENTS TO STANDARDS, POLICIES IMPROVEMENTS TO STANDARDS, POLICIES AND PROCEDURESAND PROCEDURES

NETWORK EVALUATION NETWORK EVALUATION (Of Roads in Service)(Of Roads in Service)

NETWORK SCREENINGNETWORK SCREENINGidentification of sites (road segments, intersections, etc.) witidentification of sites (road segments, intersections, etc.) with promise of h promise of safety improvementsafety improvement

DIAGNOSISDIAGNOSISwhat is wrong?what is wrong?

•• consider deficiencies consider deficiencies –– human, engineering factorshuman, engineering factors

DEVELOPMENT OF REMEDIESDEVELOPMENT OF REMEDIESselection of countermeasures for specific locationsselection of countermeasures for specific locationsprioritization of projectsprioritization of projects

EVALUATION OF REMEDIESEVALUATION OF REMEDIESWhat works, what does notWhat works, what does notFor what works, For what works,

•• When/where does it work? When/where does it work? •• How to make it work?How to make it work?•• What works best?What works best?

Evolution of road safety managementEvolution of road safety management

PrePre--19501950’’s:s: emphasis on providing mobilityemphasis on providing mobility> 1950:> 1950: increasing importance of congestion increasing importance of congestion management (1management (1stst edition of the Highway edition of the Highway Capacity Manual)Capacity Manual)19601960’’s:s: Increasing interest of administrators: Increasing interest of administrators: Start of revolutionary practical resolutionsStart of revolutionary practical resolutions

GuideGuide--rail, barriers and crashrail, barriers and crash--cushionscushions•• Reduced severity, but increased hitsReduced severity, but increased hits

BreakawayBreakaway--poles poles –– eliminate severe accidents eliminate severe accidents •• but also reduce nonbut also reduce non--severe ones!severe ones!

Evolution of road safety management, ctd.Evolution of road safety management, ctd.

Early 1980Early 1980’’s:s: Start of the knowledge revolutionStart of the knowledge revolutionRecognition of problems with existing knowledge on Recognition of problems with existing knowledge on safety implications of roadway improvementssafety implications of roadway improvements

•• RegressionRegression--toto--mean in before after studies mean in before after studies –– safety benefits safety benefits overestimatedoverestimated

•• Knowledge from crossKnowledge from cross--section studies section studies ---- safety benefits tend to safety benefits tend to be underestimatedbe underestimated

•• Better methods for beforeBetter methods for before--after and crossafter and cross--section studiessection studies•• 5 papers on safety analysis submitted to TRB in 1985.5 papers on safety analysis submitted to TRB in 1985.

MidMid--19801980’’s:s: From congestion management to From congestion management to safety managementsafety management

Managing congestion to improve safetyManaging congestion to improve safety

Is it better to be Is it better to be ………………

DEAD?DEAD?OR ALIVE AND OR ALIVE AND STUCK IN TRAFFIC??STUCK IN TRAFFIC??

Evolution Evolution ……. ctd.. ctd.

MidMid--19801980’’s to mids to mid--19901990’’s:s: Dramatic increase in road safety Dramatic increase in road safety research, researchers and practitionersresearch, researchers and practitioners

Mostly ad hoc researchMostly ad hoc researchRoad Safety Auditor as a professionRoad Safety Auditor as a professionSafety considerations in design, construction and maintenanceSafety considerations in design, construction and maintenance

1997:1997: Ezra HauerEzra Hauer’’s landmark book on observational befores landmark book on observational before--after studiesafter studies

Late 1990Late 1990’’s:s: Formal recognition by administrators of need Formal recognition by administrators of need for science based road safety management;for science based road safety management;

Increase in research directed to facilitate the management proceIncrease in research directed to facilitate the management processss•• Identify locations for detailed safety investigationIdentify locations for detailed safety investigation•• Detailed safety investigations of sites and development of remedDetailed safety investigations of sites and development of remediesies•• Prioritization and implementation of remediesPrioritization and implementation of remedies•• Evaluation of remediesEvaluation of remedies

Evolution of road safety management, ctdEvolution of road safety management, ctd..TURN OF THE CENTURYTURN OF THE CENTURY:: Formalizing safety Formalizing safety considerations in the design, construction and considerations in the design, construction and maintenance processesmaintenance processes

Explicit consideration of safety in design guides and manualsExplicit consideration of safety in design guides and manualsInteractive Highway Safety Design ModelInteractive Highway Safety Design ModelSafety auditsSafety audits

20002000’’s: s: AASHTO Strategic Highway Safety PlanAASHTO Strategic Highway Safety PlanHighway Safety ManualHighway Safety ManualPIARC Road Safety Manual PIARC Road Safety Manual

2008:2008: 100+ papers on highway safety analysis submitted 100+ papers on highway safety analysis submitted to Transportation Research Board Annual Meetingto Transportation Research Board Annual Meeting

IMPLEMENTATION OF RATIONAL IMPLEMENTATION OF RATIONAL ROAD SAFETY MANAGEMENTROAD SAFETY MANAGEMENTINITIATIVESINITIATIVES

Highway Safety ManualHighway Safety ManualAASHTO Strategic Highway Safety PlanAASHTO Strategic Highway Safety Plan

THEORETICAL CHALLENGESTHEORETICAL CHALLENGESProblems with existing analysis methods, knowledgeProblems with existing analysis methods, knowledge

RECENT RESEARCHRECENT RESEARCHEvaluation of safety measuresEvaluation of safety measures

FUTURE DIRECTIONSFUTURE DIRECTIONS

INITIATIVESINITIATIVES

THE HIGHWAY THE HIGHWAY SAFETY MANUAL SAFETY MANUAL

–– a new era in highway safety a new era in highway safety analysisanalysis

HSM PurposeHSM Purpose

To provide the best factual information and To provide the best factual information and tools in a useful form to facilitate roadway tools in a useful form to facilitate roadway planning, design, operations, and planning, design, operations, and maintenance decisions based on explicit maintenance decisions based on explicit consideration of their safety consideration of their safety consequences.consequences.

The The ‘‘visionvision’’ of the HSM of the HSM ---- a a document akin to the HCMdocument akin to the HCM

Definitive; represents Definitive; represents quantitative quantitative ‘‘statestate--ofof--thethe--artart’’ informationinformation

Widely accepted in Widely accepted in professional practice professional practice

ScienceScience--based; based; updated regularly to updated regularly to reflect new researchreflect new research

First edition outlineFirst edition outlinePART A: PREFACE TO THE HSMPART A: PREFACE TO THE HSM

PART B: ROADWAY SAFETY MANAGEMENT PROCESSPART B: ROADWAY SAFETY MANAGEMENT PROCESSNetwork ScreeningNetwork ScreeningDiagnosisDiagnosisSelect CountermeasuresSelect CountermeasuresEconomic AppraisalEconomic AppraisalPrioritize ProjectsPrioritize ProjectsSafety Effectiveness EvaluationSafety Effectiveness Evaluation

PART C: PREDICTIVE METHODSPART C: PREDICTIVE METHODSTwoTwo--lane rural roadslane rural roadsMultilane rural roadsMultilane rural roadsUrban and suburban arterialsUrban and suburban arterials

PART D: ACCIDENT MODIFICATION FACTORSPART D: ACCIDENT MODIFICATION FACTORS

EXAMPLE EXAMPLE –– Part C MethodologyPart C MethodologyFor 4For 4--legged signalized intersections on legged signalized intersections on twotwo--lane roadslane roads

Base conditions:Base conditions:NO LEFT turn lanes on major (or minor)NO LEFT turn lanes on major (or minor)NO RIGHT turn lanes on major (or minor)NO RIGHT turn lanes on major (or minor)

AMF for 1 left turn lane on major = 0.67AMF for 1 left turn lane on major = 0.67AMF for 1 right turn lane on major = 0.975AMF for 1 right turn lane on major = 0.975AMF for 2 right turn lanes on major = 0.95AMF for 2 right turn lanes on major = 0.95

)0.20lnADT0.60lnADT5.73exp(N 21bi ++−=

INITIATIVES: INITIATIVES: AASHTO STRATEGIC HIGHWAY SAFETY PLANAASHTO STRATEGIC HIGHWAY SAFETY PLANA comprehensive plan to substantially reduce vehicleA comprehensive plan to substantially reduce vehicle--related fatalities and injuries on U.S. highwaysrelated fatalities and injuries on U.S. highways

Achieve less than 0.6 fatalities per 100 million vehicleAchieve less than 0.6 fatalities per 100 million vehicle--kmkm

Produce guidebooks for emphasis areasProduce guidebooks for emphasis areasIdentify proven, tried and experimental countermeasures Identify proven, tried and experimental countermeasures

Not just a plan!Not just a plan!Stimulate/support research to fill in gaps in countermeasure Stimulate/support research to fill in gaps in countermeasure knowledgeknowledgeMarket research to ensure implementation of resultsMarket research to ensure implementation of results

EmphasisEmphasis areasareasDriversDrivers

YoungYoungUnlicensed/Suspended/RevokedUnlicensed/Suspended/RevokedOlderOlderAggressiveAggressiveImpairedImpairedDistracted/FatiguedDistracted/FatiguedSeat Belt UseSeat Belt UseSpeedSpeed

Special UsersSpecial UsersPedestriansPedestriansBicyclistsBicyclists

VehiclesVehiclesMotorcyclesMotorcyclesHeavy TrucksHeavy Trucks

HighwaysHighwaysTreesTreesRun Off the RoadRun Off the RoadHorizontal CurvesHorizontal CurvesUtility PolesUtility PolesUnsignalized IntersectionsUnsignalized IntersectionsHeadHead--On CollisionsOn CollisionsWork ZonesWork Zones

Emergency Medical Emergency Medical ServicesServices

Rural EMSRural EMS

ManagementManagementDataDataIntegrated Safety Integrated Safety Management ProcessManagement Process

Guide for addressing runGuide for addressing run--offoff--road road collisions collisions –– Strategies to keep vehicles Strategies to keep vehicles

from encroaching on the roadsidefrom encroaching on the roadsideshoulder rumble stripsshoulder rumble stripsedgeline edgeline ““profile marking,profile marking,”” edgeline rumble strips edgeline rumble strips midlane rumble stripsmidlane rumble stripsenhanced shoulder or inenhanced shoulder or in--lane delineation and marking lane delineation and marking for sharp curvesfor sharp curvesimproved highway geometry for horizontal curvesimproved highway geometry for horizontal curvesenhanced pavement markingsenhanced pavement markingsskidskid--resistant pavement surfacesresistant pavement surfacesshoulder treatmentsshoulder treatmentsshoulder dropshoulder drop--offs offs wider and/or paved shoulderswider and/or paved shoulders

Theoretical challenges Theoretical challenges

Rational management requires use of accident Rational management requires use of accident datadata

Accident counts can be randomly high or low, Accident counts can be randomly high or low, therefore:therefore:

•• Sites can be wrongly identified as safe or unsafe in network Sites can be wrongly identified as safe or unsafe in network screeningscreening

•• Sites with high counts will experience a reduction in Sites with high counts will experience a reduction in accidents in future if nothing is done (regressionaccidents in future if nothing is done (regression--toto--the the mean).mean).

This reduction can be wrongly attributed to an implemented This reduction can be wrongly attributed to an implemented remedyremedy

Example of random fluctuation over timeExample of random fluctuation over time

Long term average = 4.23 accidents per year

0123456789

10

1 6 11 16 21

Year number

Acci

dent

cou

nt

Real evidence of regression to the meanReal evidence of regression to the mean742 California rural stop controlled untreated intersections742 California rural stop controlled untreated intersections

Intersections with given number of

accidents in 1994-96

Accidents/ intersection in

1994-96

Accidents/ intersection in 1997-99

Percent Change

110 0 0.96 Large

increase 136 1 1.65 65.4% 101 2 2.64 32.2% 101 3 2.97 -1.3%

37 6 4.89 -18.5%

16 10 8.55 -14.4% 10 11 8.79 -20.0%

REGRESSION TO THE MEAN IN INSTALLING REGRESSION TO THE MEAN IN INSTALLING GATES AT RAIL CROSSINGS WITH FLASHERSGATES AT RAIL CROSSINGS WITH FLASHERS

Accidents beforeAccidents before = 286= 286Accidents afterAccidents after = 114= 114

Apparent savings Apparent savings = 172 = 172 (60% reduction)(60% reduction)

The Reality:The Reality:Accidents expectedAccidents expected = 208= 208

Actual savingsActual savings(208 (208 –– 114)114) = 94 = 94 (45% reduction)(45% reduction)

Implication of temporal fluctuation Implication of temporal fluctuation ---- regressionregression--toto--thethe--meanmean

A high accident count will on average be followed A high accident count will on average be followed by a reduced count even if site is unchangedby a reduced count even if site is unchanged

Therefore, if a site is selected for treatment Therefore, if a site is selected for treatment because of a high count it will experience an because of a high count it will experience an apparent improvement in safety even if the apparent improvement in safety even if the treatment is uselesstreatment is useless

Therefore, this regressionTherefore, this regression--toto--thethe--mean effect must mean effect must be accounted for in:be accounted for in:

Selecting sites for treatmentSelecting sites for treatmentEvaluating the safety effect of treatmentEvaluating the safety effect of treatment

Accounting for regression to the meanAccounting for regression to the mean

0

1

2

3

4

5

6

7

1993 1995 1997 1999 2001 2003 2005 2007Year

Cras

hes

Long-term average (m)

3-year average ‘before’ (Xa)

True safety effect

Observed safety effect

3-year average ‘after’

Basics of rational safety analysisBasics of rational safety analysisEmpirical Bayes MethodEmpirical Bayes Method

Use safety Use safety performance functions performance functions to estimate expected to estimate expected (long term) accident (long term) accident frequency, frequency, mm..

m m is a weighted is a weighted average of the safety average of the safety performance function performance function estimate (SPF) and the estimate (SPF) and the observed accident observed accident frequency (frequency (XX) of a site) of a site

* SPC

rash

/mile

-yea

r

AADT

* X* m SPF

(Observed frequency)

(Expected frequency)

Recent EB safety evaluations with Recent EB safety evaluations with latest methods (for AASHTO SHSP) latest methods (for AASHTO SHSP)

Flashing beaconsFlashing beaconsRaised pavement markersRaised pavement markersCentre line rumble stripsCentre line rumble stripsImprove skid resistanceImprove skid resistanceOffset left turn lanesOffset left turn lanesTwo way left turn lanesTwo way left turn lanesRoad DietsRoad Diets

Flashing Beacons Flashing Beacons Accident reductions at 106 sitesAccident reductions at 106 sites

AngleAngle Injury & Injury & FatalFatal

13.3% 13.3% 10.2% 10.2%

SAFETY EVALUATION OF CENTRE SAFETY EVALUATION OF CENTRE LINE RUMBLE STRIPSLINE RUMBLE STRIPS340 km at 98 U.S. sites340 km at 98 U.S. sites

% reduction% reduction

All Impact All Impact typestypes

1414--15%15%

Opposite Opposite direction and direction and sideswipesideswipe

2121--25%25%

SAFETY EVALUATION OF SAFETY EVALUATION OF Pavement markersPavement markers

Previous comparison group studies had found an increase in crashPrevious comparison group studies had found an increase in crasheses

% Reduction in crashes (EB Study)% Reduction in crashes (EB Study)NightNight WetWet

New JerseyNew JerseyNonNon--SelectiveSelective174 miles174 miles

0.90.9 3.23.2

New YorkNew YorkSelectiveSelective82 miles82 miles

12.712.7 20.220.2

Improve skid resistance at targeted Improve skid resistance at targeted locations (New York) locations (New York) (High skid (High skid

numbers and wet weather accident numbers and wet weather accident frequency)frequency)

% reduction in crashes% reduction in crashes

LocationLocationTypeType All CrashesAll Crashes WetWet--roadroad RearRear--end end

WetWet--roadroad

SegmentsSegments 23%23% 56%56% 43%43%

Intersection Intersection approachesapproaches 20%20% 57%57% 68%68%

Offset left turn lanes (Wisconsin)Offset left turn lanes (Wisconsin)

Total Injury Left-Turn

Rear-End

EB estimate of crashes expected in the after period without strategy 233.77 95.88 94.85 72.76

Count of crashes observed in the after period 155 62.0 59 50

Estimate of percent reduction (standard error)

33.8

(6.0)

35.6

(9.0)

38.0

(8.9)

31.7

(10.9)

Percent reduction in crashes Percent reduction in crashes ((significant at the 5% levelsignificant at the 5% level) )

RearRear--endend InjuryInjury TotalTotal

38.738.7 26.126.1 20.320.3

Addition of TWLTL to 2-Lane Road

Rural and urban locations in NC, IL, CA and AK

Where do we go from here?Where do we go from here?…… Future directions ... Future directions ... From EB to FBFrom EB to FB

From EB to FBFrom EB to FB……..Using Markov Chain Monte Carlo simulation techniquesUsing Markov Chain Monte Carlo simulation techniques

(WINBUGS software)(WINBUGS software)

Model development and posterior distributions of estimates Model development and posterior distributions of estimates are integrated .. are integrated .. λλ estimated in single step estimated in single step

Smaller sample sizes possibleSmaller sample sizes possibleFlexibility to choose crash distribution (EB assumes Flexibility to choose crash distribution (EB assumes

negative binomial)negative binomial)Provides the posterior distributions of outcomes (Provides the posterior distributions of outcomes (λλ))

……usually not normal as for EBusually not normal as for EBCan easily consider spatial correlation between sites in the Can easily consider spatial correlation between sites in the

model formulation model formulation Can easily accommodate very complex model forms (EB Can easily accommodate very complex model forms (EB

uses generalized linear models typically)uses generalized linear models typically)

Application to conversion from 4Application to conversion from 4--lane crosslane cross--section to 3 lanes plus twosection to 3 lanes plus two--way left turn laneway left turn lane

Data for TWLT lane evaluationData for TWLT lane evaluation

16 treatment sites16 treatment sites

16 yoked comparison sites (used for FB)16 yoked comparison sites (used for FB)

256 similar but untreated 256 similar but untreated ““referencereference”” sitessitesUsed to develop safety performance function required Used to develop safety performance function required for EBfor EBUsed for FB to see if results differ from using 16 yoked Used for FB to see if results differ from using 16 yoked sites sites

RESULTS RESULTS ---- ROAD DIET EVALUATIONROAD DIET EVALUATION% reduction in crashes (and standard error)% reduction in crashes (and standard error)

Number of reference sites EB FB 15 (yoked) Not done 53% (1%)296 47% (2%) 49% (1%)

RESULTS RESULTS Evaluation of 28 rural traffic signal installationsEvaluation of 28 rural traffic signal installations

% reduction in crashes (and standard error)% reduction in crashes (and standard error)

Crash Type FB Approach EB Approach Total 22% (6%) 16% (6%)

Rear-end -27% (23%) -26% (27%) Right-angle 79% (3%) 72% (5%)

Left-turn 52% (6%) 49% (7%)

CONCLUSIONS CONCLUSIONS –– EB/FB EB/FB COMPARISONCOMPARISON

FB is quite complicated and only worth the FB is quite complicated and only worth the effort if: effort if:

Reference sample is too small to develop Reference sample is too small to develop reliable safety performance functionsreliable safety performance functionsReference data are spatially correlatedReference data are spatially correlatedInformative priors availableInformative priors availableGood Ph.D. students are availableGood Ph.D. students are available

My vision for the future of road safety My vision for the future of road safety managementmanagement

Road safety as public health issueRoad safety as public health issueResearch funding/interest from Health InstitutesResearch funding/interest from Health InstitutesPublications in public health journalsPublications in public health journals

Interdisciplinary/multidisciplinary approachesInterdisciplinary/multidisciplinary approachesAccommodation in all road engineering disciplinesAccommodation in all road engineering disciplinesCanadian multidisciplinary examplesCanadian multidisciplinary examples

•• Network Centre of excellence: AUTO21Network Centre of excellence: AUTO21•• Fatigue related accidents .. Canadian Institutes for Health ReseFatigue related accidents .. Canadian Institutes for Health Researcharch

International cooperationInternational cooperationResearch transferability Research transferability –– Is it possible?Is it possible?

•• E.g., transferability of accident prediction modelsE.g., transferability of accident prediction models

High level research using microscopic and related dataHigh level research using microscopic and related dataSHRPIISHRPIISurrogate measuresSurrogate measures