A Planning-Level Model for Assessing Pedestrian Safety

MISSING IMAGE

Material Information

Title:
A Planning-Level Model for Assessing Pedestrian Safety
Physical Description:
1 online resource (59 p.)
Language:
english
Creator:
Jermprapai, Khajonsak
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Master's ( M.E.)
Degree Grantor:
University of Florida
Degree Disciplines:
Civil Engineering, Civil and Coastal Engineering
Committee Chair:
SRINIVASAN,SIVARAMAKRISHNAN
Committee Co-Chair:
ELEFTERIADOU,AGELIKI
Committee Members:
BEJLERI,ILIR

Subjects

Subjects / Keywords:
planning -- safety -- traffic
Civil and Coastal Engineering -- Dissertations, Academic -- UF
Genre:
Civil Engineering thesis, M.E.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract:
Crash-prediction models are useful tools to identify locations that have higher risk of crashes and to prioritize projects.The focus of this study was on developing macroscopic or planning-level models for pedestrian safety. Crash data from multiple years (2005 -2009) and land use data from the entire state of Florida are used in developing models for pedestrian crashes. Four models were developed to determine the crash frequency for each census block group. The estimated models capture the effects of several socioeconomic, transportation, land use, and contextual variables. The models were used to determine the expected number of crashes for all the census block groups in the state. This predictive assessment exercise serves to highlight the value of planning models. Specifically, if safety assessments are made purely based on crash history, all the locations with zero observed crashes will be deemed equally “safe”. However, the predictive model highlights that there is a significant variability in crash risk across these locations because of differences in land use and socioeconomic patterns. Thus the planning models developed in this study can be powerful tools in statewide safety funds allocation and prioritization of safety projects.
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility:
by Khajonsak Jermprapai.
Thesis:
Thesis (M.E.)--University of Florida, 2013.
Local:
Adviser: SRINIVASAN,SIVARAMAKRISHNAN.
Local:
Co-adviser: ELEFTERIADOU,AGELIKI.

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2013
System ID:
UFE0046430:00001


This item is only available as the following downloads:


Full Text

PAGE 1

APLANNING-LEVELMODELFORASSESSINGPEDESTRIANSAFETYByKHAJONSAKJERMPRAPAIATHESISPRESENTEDTOTHEGRADUATESCHOOLOFTHEUNIVERSITYOFFLORIDAINPARTIALFULFILLMENTOFTHEREQUIREMENTSFORTHEDEGREEOFMASTEROFENGINEERINGUNIVERSITYOFFLORIDA2013

PAGE 2

c2013KhajonsakJermprapai 2

PAGE 3

ToMomandAuntieNee 3

PAGE 4

ACKNOWLEDGMENTS IwouldhavetothankyouDr.SivaSrinivasanwhoactasmyadvisorfromtherstdayincampus.Heisalwaysprovidingmetheknowledgeandthesuggestionformebothinlectureclassandthecompletionofthisthesis.IwouldalsolikethankmycommitteeDr.LilyElefteriadouandDr.BejleriIlirwhoserveasmycommittee.Iattendtheclassofbothpeopleandfoundithelpmealotinthecontextofhowtoconductagoodresearchandthebettervisionoftrafcandsafetywork. 4

PAGE 5

TABLEOFCONTENTS page ACKNOWLEDGMENTS .................................. 4 LISTOFTABLES ...................................... 7 LISTOFFIGURES ..................................... 8 ABSTRACT ......................................... 9 CHAPTER 1INTRODUCTION ................................... 10 1.1Motivation .................................... 10 1.2OutlineofThesis ................................ 11 2LITERATUREREVIEW ............................... 12 2.1Thecomparisonofstatisticalmodeling .................... 12 2.1.1ModelMethodology ........................... 12 2.1.2GeographicUnitSelection ....................... 13 2.2EmpiricalFindings ............................... 14 2.2.1ImpactsofSocioeconomiccharacteristic ............... 15 2.2.2ImpactsofTransportationCharacteristicandRoadInventory ... 16 2.2.3ImpactsofBuiltenvironmentandLanduse ............. 17 2.3Summary .................................... 18 3DATA ......................................... 23 3.1CrashData ................................... 23 3.2ExplanatoryVariables ............................. 26 3.2.1Socioeconomiccharacteristic ..................... 26 3.2.2TransportationCharacteristic ..................... 27 3.2.3LanduseandBuilt-environment .................... 28 3.2.4Locationcontext ............................ 29 4EMPIRICALRESULT ................................ 45 4.1Impactofexplanationvariable ......................... 45 4.1.1Impactofsocioeconomiccharacteristic ............... 45 4.1.2ImpactoftransportationCharacteristic ................ 46 4.1.3ImpactoflanduseandBuilt-environment .............. 47 4.1.4impactoflocationContext ....................... 48 4.2ModelApplication ............................... 49 4.3Conclusion ................................... 49 5

PAGE 6

5SUMMARYANDCONCLUSION .......................... 53 5.1Summary .................................... 53 5.2Conclusion ................................... 54 5.3Futurework ................................... 54 REFERENCES ....................................... 56 BIOGRAPHICALSKETCH ................................ 59 6

PAGE 7

LISTOFTABLES Table page 2-1Overviewofmajormodels .............................. 19 2-2Impactofsocioeconomiccharacteristicvariablesinpedestriancrashpredictionmodel ......................................... 20 2-3ImpactofTrafcCharacteristicandRoadInventoryvariablesinpedestriancrashpredictionmodel ................................ 21 2-4ImpactsofTrafcCharacteristicandRoadInventory ............... 22 3-1DescriptivestatisticofcensusblockpedestriancrashinFloridafrom2005-2009 43 3-2Descriptivestatisticofvariables ........................... 44 4-1Empiricalresult .................................... 51 4-2Goodnessoftofthemodel ............................ 52 4-3Descriptivestatisticofpredictionforcensusblockgroupwithzerocrashreport 52 7

PAGE 8

LISTOFFIGURES Figure page 3-1Percentageofcrasheventsbyseverity ....................... 31 3-2Locationdistributionofpedestriancrash ...................... 31 3-3Timedistributionofpedestriancrash ........................ 32 3-4Dayofweekcrashesdistribution .......................... 32 3-5TimeDistributionofweekdaypedestrianCrash .................. 33 3-6TimeDistributionofWeekendPedestrianCrash .................. 33 3-7Distributionofcrashesbylightcondition ...................... 34 3-8Distributionoffatalcrashesbylightcondition ................... 34 3-9Pedestrian-Vehicleinteractionofcontributiontocrash .............. 35 3-10Usageofdrug/alcoholonpedestrianinvolvedcrash ............... 35 3-11Usageofdrug/Alcoholonpedestriancrashfatality ................ 36 3-12Theinteractionofalcoholusagebetweendriverandpedestrianontheeventofalcoholinvolvedpedestriancrash ........................ 36 3-13Mapofaggregatecrashincensusblockgroup .................. 37 3-14Mapofaggregateseverecrashincensusblockgroup .............. 38 3-15Mapofaggregatefatalcrashincensusblockgroup ............... 39 3-16Mapofaggregatenighttimecrashincensusblockgroup ............ 40 3-17Frequencydistributionofcrashes ......................... 41 3-18Frequencydistributionofseverecrashes ..................... 41 3-19Frequencydistributionoffatalcrashes ....................... 42 3-20Frequencydistributionofnighttimecrashes ................... 42 8

PAGE 9

AbstractofThesisPresentedtotheGraduateSchooloftheUniversityofFloridainPartialFulllmentoftheRequirementsfortheDegreeofMasterofEngineeringAPLANNING-LEVELMODELFORASSESSINGPEDESTRIANSAFETYByKhajonsakJermprapaiDecember2013Chair:SivamarakrishnanSrinivasanMajor:CivilEngineeringCrash-predictionmodelsareusefultoolstoidentifylocationsthathavehigherriskofcrashesandtoprioritizeprojects.Thefocusofthisstudywasondevelopingmacroscopicorplanning-levelmodelsforpedestriansafety.Crashdatafrommultipleyears(2005-2009)andlandusedatafromtheentirestateofFloridaareusedindevelopingmodelsforpedestriancrashes.Fourmodelsweredevelopedtodeterminethecrashfrequencyforeachcensusblockgroup.Theestimatedmodelscapturetheeffectsofseveralsocioeconomic,transportation,landuse,andcontextualvariables.Themodelswereusedtodeterminetheexpectednumberofcrashesforallthecensusblockgroupsinthestate.Thispredictiveassessmentexerciseservestohighlightthevalueofplanningmodels.Specically,ifsafetyassessmentsaremadepurelybasedoncrashhistory,allthelocationswithzeroobservedcrasheswillbedeemedequallysafe.However,thepredictivemodelhighlightsthatthereisasignicantvariabilityincrashriskacrosstheselocationsbecauseofdifferencesinlanduseandsocioeconomicpatterns.Thustheplanningmodelsdevelopedinthisstudycanbepowerfultoolsinstatewidesafetyfundsallocationandprioritizationofsafetyprojects. 9

PAGE 10

CHAPTER1INTRODUCTION 1.1MotivationPedestriancrashesareaproblemallovertheworld.Morethanhalfoftotalfatalitiesfromworldwideroadaccidents(65%)happenforpedestrians.IntheUS,thepedestriancrashesaccountforabout13%oftotalroad-accidentrelatedfatalities.Thereareapproximately6500pedestriancrashesperyearwith487fatalitiesintheStateofFlorida.Whilethefrequencyofpedestriancrashisonly2%ofthetotalcrashinFlorida,thepedestriancrashesfatalityareaccountedfor19%oftotalfatality(NHSTA,2012)[ 2 ].Crash-predictionmodelsareusefultoolstoidentifylocationsthathavehigherriskofcrashes.Ingeneral,thesemodelscanbeclassiedasmicroscopic(orprojectlevel)andmacroscopic(orplanninglevel).Microscopicmodelsassessthesafetyofanindividualroadsegmentoranintersection.AperfectexampleforthistypeofmodelisthepredictivemethodprescribedbytheHighwaySafetyManual(PartC).Thesemodelstakeintoconsiderationseveraldetailedcharacteristicsoftheroadwaygeometryandtrafcoperations.Assuchtheycanbeusedtoassessthesafetybenetsofanyroadwayprojectbydeterminingtheexpectednumberofcrashesbefore-andafter-theplannedroadwaytreatment.Themacroscopicmodels,ontheotherhand,lackdetailedsiteconditions.Rather,theaggregatenumberofcrashesonthegeographicunit(suchascensustract,censusblockgrouportrafcanalysiszone)areanalyzed.Thesemodelstakeintoconsiderationaggregatesocioeconomicandlanduseconditionsoftheregionasinputs.Suchmodelscanbeusedforplanningpurposestoallocatefundsacrossthedifferentregionsofastate.Thefocusofthisstudyisondevelopingmacroscopicmodelsforpedestriansafety.CrashdataandlandusedatafromtheentirestateofFloridaareusedindevelopingmodelsforpedestriancrashes.Suchmodelscanhelpidentifylocationsofhighestrisk 10

PAGE 11

andtherebyinformdecisionstoallocatefundsforimplementingcountermeasurestoimprovesafety. 1.2OutlineofThesisChapter2willbetheliteraturereview.Overviewandthecurrentstate-of-artofthestudywillbediscussedandidentiedofthelimitation.Chapter3willdiscussaboutthesourceofthedataandthedataprocessingwhichisthepreparationfortheempiricalanalysisthatwillbediscussedinChapter4.Itwillbeaboutthediscussionabouttheeffectofvariablebyvariablefromtheempiricalmodel.Itwillalsoincludethediscussionofhowmultiplelevelvariableinclusionaffecttheoverallmodelandthecombineeffectofvariable.Finally,theChapter5willbetheconclusionandlistofworkthatwouldconductfurtherinthestudy. 11

PAGE 12

CHAPTER2LITERATUREREVIEWThischapterpresentsthecurrentstateofartinpedestriancrashpredictionmodeling.Section2.1presentsadiscussionofthestatisticalmodelsincludingthemodelstructureandgeographicalunit.Thediscussionsaboutempiricalndingsfromthesemodelsarepresentedinsection2.2.Finally,thechapterwillbeconcludewithanoverallsummaryandbyidentifyingthecontributionofproposedwork. 2.1Thecomparisonofstatisticalmodeling 2.1.1ModelMethodologyThenaturesofaggregatecrashdatacausesomeproblemsontheselectionofregressionmodel.Normally,themostbasicmodelforthecountdataisPoissonregressionmodel.CottrillandThakuriah(2010)[ 7 ]usedPoissonRegressionintheearlyphaseofthestudy.However,asmostoftheaggregategeographicunitcrashesdatawillbezero.Theproblemofover-dispersionoccurred.OneofthebasicassumptionsforPoissonregressionisthatthedispersionparameterisequalto1.Duetothisreason,mostofthepreviousmodelsareshiftedtowardthenegativebinomialregression.Ukkusurietal.(2011)[ 18 ],CottrillandThakuriah(2010)[ 7 ],Charkravarthyetal.(2010)[ 5 ],greenetal.(2011)[ 8 ]andPulugurthaetal.(2013)[ 16 ]areallusenegativebinomialregressionintheirmodel.HSMandTorbicetal.(2010)[ 17 ]modelwhichareconsideredtobemicroscopiclevelmodelarealsousenegativebinomialregression.TheotherpotentialmodeltouseisZero-InatedModelduetothenatureofexceedingzero-valueofaggregatecrashdata.Eventhoughthereiscurrentlynoapplicationofthistypeofmodelonpedestriancrash,someofapplicationsongeneraltrafccrashareshowninthestudyofHuangandChin(2010)[ 10 ]andAguero(2013)[ 3 ].ThemeaningofZero-inatedmodelontrafccrash;however,hasbeencriticizebysomeofthepaststudyLordetal.(2007)[ 13 ].Duetotheseparationbetweencountandzerostate,Lordetal.(2007) 12

PAGE 13

Thereisalsotheapplicationofspatialregressiontechniqueontheplanninglevelmodel.GeologicalWeightRegressionisthetechniquethatconductstheregressionanalysisforeachgeographicunitseparately.GWR(localregression)isdifferentfromtraditionalglobalregressionmodelinthesensethatvariablescanhavedifferenteffectforeachgeographicunit.GWRwillconsidertheeffectofvariableforonlywithinthebandwidthradiusofgeographicunitwhileglobalregressionwillconsiderallofthedataset.Pastresearch(Lietal.(2013)[ 12 ],Hadayeghietal.(2013)[ 9 ],Pirdavanietal.(2013)[ 15 ],Zhengetal.(2013)[ 19 ])suggestthatGWRhastheadvantageintermofaccuracy.However,GWRmodelisstrictlytransferable.Inthatcase,globalregressionmodelisstillthemethodologyofchoice. 2.1.2GeographicUnitSelectionAsthemacroscopicmodelispredictiontheaggregatenumberofcrashingeographicalunit,theselectionofgeographicalunitbecomeimportant.Fromthepaststudy,thereare2groupofgeographicunituseinthistypeofmodel,censusgroupandtrafczonegroup.Censusblockisthesmallestunitincensusgeographicunit.Censusblockareterrainwhichdividebyvisiblefeaturesuchasroad,waterwayandrail.Duetothetoosmallsizeoftheblockandthelimitedavailabilityofsocioeconomicdata,itisrarelyuseasthegeographicunitintheanalysis.Thebiggercensusunitiscensusblockgroupwhichisthecombinationofseveralblockstocontain600-3000people.CensusblockgroupisthesmallestcensusunitthatUScensusbureaupublishedthecensusdata.TheusagesofcensusblockgroupintrafcsafetystudyareMarshallandGarrick(2012)[ 14 ],Melikeretal.(2004)andAbdel-Atyetal.(2013)[ 1 ].Thelargestgeographicunitthatfrequentlyuseintrafcsafetystudyiscensustract.Censustractaregenerallycontain12008000peoplethoughtherearecensustractthatcontainalotlesspeopleinruraloruninhabitedarea.TheadvantageofcensustractisthatittendstobepermanentasthedelineationaredenebyCensusbureauopposingtocensusblockgroupwhichdefybylocalgovernment.Censustractarealsohascensusdatapublish 13

PAGE 14

bycensusbureausoitisagoodcandidateformacroscopicanalysis.Thesizeofcensustract;however,sometimesaretoobigtoperformgeographicanalysisespeciallytheareawherehighlypopulatedcensustractanduninhabitedcensustractlocatedtogether.TrafcAnalysisZone(TAZ)isthenalgeographicunitthatcurrentlyused.TAZhasbeendelineatingbylocalDOT.Itwasintendtobeuseintransportationplanning.ThesizeofTAZiscomparabletocensusblockgroup.ThestudyofAbdel-Atyetal.(2013)[ 1 ]isdirectlyrelatedtotheissueofgeographicselection.Abdel-Atyetal.concludethatthemacroscopicmodelforcensusblockgroupandTAZarecomparable.TheTAZbasedmodelhastheadvantageontheavailabilityoftrafcrelatedfactor.Censusblockfamilybasedmodel,ontheotherhand,hasbetteraccesstomoresocioeconomicfactorofcommuter.Sotheselectionofgeographicunitisuptotheanalystonwhattobefocus.Forthemicroscopicmodel,thegeographicalunitofchoicecanberoadsectionandseveraltypeofintersection.Asthisstudyisfocusingonmacroscopicmodelsothedetailofgeographicunitselectionformicroscopicmodelwillnotbegiven. 2.2EmpiricalFindingsEventhoughthepastresearchconductwithdifferentcombinationofvariable,wecangroupindependentvariableinto4groups,commuterandsocioeconomiccharacteristic,builtenvironment/landuseandtrafccharacteristicandlocationcontext.Inthissectionwewilldiscussedaboutthededicatedpedestriancrashmodelonly(Ukkusurietal.(2011)[ 18 ],CottrillandThakuriah(2010)[ 7 ],Charkravarthyetal.(2010)[ 5 ],greenetal.(2011)[ 8 ],Pulugurthaetal.(2013)[ 16 ]andAbdel-Atyetal.(2013)[ 1 ]aseventhoughthemethodologyforthemodeliscomparable,thefactorthataffectgeneralcrashandpedestriancrashisdifferent. 2.2.1ImpactsofSocioeconomiccharacteristicThisgroupofvariabledemonstratesthesocialandeconomicconditionofthegeographicunit.Ithasthepossibilitytoanswerthequestionabouthumanfactorinthecauseofaccident.Thetotalpopulationistherstvariableinthisgroup.Allmodelhas 14

PAGE 15

includethiskindofvariable.Abdel-Atyetal.(2013)[ 1 ]andUkkusurietal.(2011)[ 18 ]usethetotalnumberofpopulation.Chakravarthyetal.(2010)[ 5 ]andCottrillandThakuriah(2010)[ 7 ];however,usethedensityofpopulationintheirstudy.Amongthegroupsthatusetotalpopulation,Abdel-Atyetal.(2013)[ 1 ]hasinconsistentresultthantheothertwo.Ukkusurietal.(2011)[ 18 ]havetheresultthattotalmoremeanmorecrash.AsthestudyofAbdel-Atyetal.(2013)[ 1 ]isnotonlydividingtheeffectoftotalpopulationto2agegroupsbutthegeographicunitalsodividedinto3cases(censustract,censusblockgroupandTAZ).TheresultforcensustractmodelisthetotalnumberoflowageresultinthereductionofnumberofcrashwhilecensusblockgroupandTAZmodeldonothavesignicanteffectinallage.OnespeculationisthatAbdel-Atyalsoincludesthedensityofchildren(underK-12)anddensityofhouseholdinthemodelsotheabnormalresultcancausebytheinternalcorrelationofthisvariable.FortheChakravarthyetal.(2010)[ 5 ]modelandCottrillandThakuriah(2010),thedensityofpopulationcausestheincreasingofcrashfrequency.Fromtheperspectiveofmacroscopicanalysis,theusageofdensityinsteadoftotalnumberofpopulationismakingmoresenseinthecaseofvariationinunitsizeandpopulation.Theageofpeopleinthegeographicunitistheothervariablethathasbeenincludeinall4models.Despitethedifferentformofagevariableinall4models,theeffectonthenumberofcrashesisthesame.Thelowerageofpopulationinthegeographicunitresultinhighernumberofcrash.Thisimpliesthatforthepedestriancrash,thelowagepeoplearemoresusceptible.TheminoritypeopleareothergroupthatlikelytohavemoresusceptibilitytopedestriancrashasshowninbothUkkusurietal.(2011)[ 18 ]andAbdel-Atyet(2013)Alstudies.CottrillandThakuriah(2010)[ 7 ]modelisthemodelthatalsofocusesonthisgroupofminority.Withthisnding,wecanconcludethattheminoritygroupingeographicunitshouldbegivenmoreconcern.ThemedianhouseholdincomeistheanothervariablethatareincludedinCottrillandThakuriah(2010)[ 7 ]andChakravarthyetal.(2010)[ 5 ].Theirresultsarethesameasthehighermedianincome 15

PAGE 16

meanlowercrashes.Thepossibleexplanationisthatthepeoplewithlowincomemaywalkmuchmorebecauseoftheunaffordabilityofautomobile;hence,thehigherriskofpedestriancrash.TheeffectofeducationonthenumberofcrashisalsointerestingasshowninUkkusurietal.(2011)[ 18 ]andChakravarthyetal.(2010)[ 5 ]study.Theunitwithhigherlevelofeducationhaslessnumberofcrashesthantheunitwithlowereducationlevel.TheabilitytospeakEnglish(CottrillandThakuriah(2010)[ 7 ]andChakravarthyetal.(2010)[ 5 ])canbethemissinglinkbetweentheeffectofeducationandminorityasitshownsimilareffect.ItisthecommonknowledgethatminoritypeopleareusuallylacktheeducationandabilitytospeakEnglishuently.Sothisiscanbetherealcauseofsusceptibilityforminoritypeople.Thenalsubgroupofvariableinthisgroupisthebehaviorofcommuterwhichistheusingoftransitandthewalkingtowork.Thisgroupofvariablehasdirecteffectasthemorewalkingthemoreexposetotheriskofpedestriancrash.greenetal.studywhichbasedonthestatisticdatafromEnglishhasincludethevariableforpersonwhoworkathome.Thisvariablegivesthenegativeeffectwhichseemstobereasonable.Thepointthatwecaninferfromthisvariableisthepossibilityeffectofthedurationfromhometoworkplacewhichshouldalsohavetheeffectonthenumberofcrash.ThesummaryofsocioeconomicvariableeffectareshowninTable 2-2 2.2.2ImpactsofTransportationCharacteristicandRoadInventoryThetotallengthofroadandtrafcvolumeinthegeographicunitareincludedinallmodelexceptChakravarthyetal.(2010)[ 5 ]whichdoesntincludedanytrafccharacteristicintheanalysis.Themoretotalroadlengthandtrafcvolumeyieldmorepedestriancrashasthesecharacteristicsincreasetheexposureofcrashtothepedestrian.ThetotalnumberofintersectionwhichincludedinAbdel-Atyetal.(2013)[ 1 ]modelprovidethesimilarresultasitisthesamekindofexposureincrease.Thenalgroupofvariableisthetransitavailability.Ukkusurietal.(2011)[ 18 ]usethetotal 16

PAGE 17

numberofbusstopandsubwaystationtorepresentthischaracteristicwhileCottrillandTakuriah(2010)usethevariablecalledTrafcAvailabilityIndex(TAI)whichcalculatedfromthetimetoaccess,frequencyandhourofserviceofthetransitsystemincensustract.Despitethedifferenceinmethodology,theresultsareinthesamedirection.Thegeographicunitwithmoretransitavailabilityhasmorepedestriancrash.Thepossibleexplanationistheclusterofpedestrianthatwillbehappenonthetransitstationarea.Thiscanbeinferringthatthesafetyfortransitareaisinneedofimprovement.ThesummaryoftransportationcharacteristicvariableeffectareshowninTable 2-3 2.2.3ImpactsofBuiltenvironmentandLanduseThebuiltenvironmentandlandusenotonlyreecttheenvironmentalfactoronthecauseofaccidentbuttheeffectintripgenerationandattractionisperhapsthehigherreasonforthehighernumberofcrashes.Oneofthedifferencesbetweenautomobilecrashpredictionmodelandpedestriancrashpredictionmodelistheavailabilityoftrafcvolume.Atthecurrentstate,thevehiclevolumeisusuallyavailablefortheuse.Thepedestrianvolume;however,arelessavailable.Sothebuiltenvironmentandlandusecanbeservesasanindirectmethodtocountertheneedofpedestrianvolume.Amongthe4previousmodel,Ukkusurietal.(2011)[ 18 ]andCottrillandThakuriah(2010)arethestudythathasfocusonlanduseandbuiltenvironmentthanothers.Bothhasincludedtheeffectofschoolzoneandopenareainthemodelandhavethesimilarresult.Fromtheirstudy,wecanconcludethatschoolzoneandopenareaaretheareasthathavetheriskofpedestriancrashoccurrenceasthemoreschoolandopenareainthegeographicunitthemorenumberofcrashesoccur.Ukkusurietal.(2011)[ 18 ]modelalsoincludedtheeffectofcommercialzoneandindustrialzone.This2typeofzoneareknowntobethetripattractionwhichcanyieldmorenumberofpedestrian;hence,morenumberofcrashes.Theresultofthemodelsupportthisspeculationasthegeographicunitwithhigherproportionofcommercialzoneandindustrialzonehasmorenumberofcrashesthantheunitwithlowerproportioncommercialandindustrialzone.Therelation 17

PAGE 18

ofnumberofcrimeandnumberofcrashesareinthesamedirectionasshownbystudyofCottillandTakhuriah(2010)[ 7 ].Forthisgroupofvariable,Pulugurthaetal.(2013)[ 16 ]hasincludemanyspeciclandusevariablesuchasthePUDstatus,Researchdistrictandmixuseddevelopmentwhicharegivenmoreconsiderationbyurbanplannerthantrafcengineer.ThesummaryoflandusevariableeffectareshowninTable 2-4 2.3SummaryThischapterservesastheoverviewofthecurrentstate-of-artinpedestriancrashpredictionmodel.Atthecurrentstate-of-art,themacroscopicplanninglevelmodelismostlyconductoncensuslevelortrafcanalysiszonelevelwitheithernegativebinomialorGWRmodel.Alimitationthatworthwhileformorestudyisthecurrentmodelsaremostlyconsideredonlytheinternalfactorwithingeographicunit.Theyarelackingthevariablethatrepresentstheinuencefromneighborgeographicunit.Thereispossibilitythatexternaleffectfromneighborhoodandlargerscalecharacteristicwillinuencetheexposureandriskofpedestriancrashincensusblockgroup.Thegoalofthestudyisaimtoanswerthisquestion. 18

PAGE 19

Table2-1. Overviewofmajormodels MicroorMacroRegressionTypeRequirePedestrianCountY/NSpatialAnalysisDataSourceYearofDataRandomErrorTermDedicatedmodelforpedestrianFocusAgeFocusZonenote Ukkusurietal.(2011Macro(censustract)NegativeBinomialngeomappingNY2002-2006yy--CottrillandThakuriah(2010)[ 7 ]Macro(censustract)NegativeBinomialnclusteranalysischicago2005ny-EJAreaEnvironmentalJusticeAreaismainlyaboutmi-norityandpovertytopicCharkravarthyetal.(2010)[ 5 ]Macro(censustract)NegativeBinomialnoverlayCADOTOrangeCounty2000-2004ny---Abdel-Atyetal.(2013)[ 1 ]Macro-CT,BGandTAZPoissonLog-normal(bayesian)n-FDOTPinellas&Hillsbor-ough2005-2006ny--Thispa-permainlydiscussedtheeffectofscaleGreenetal.(2011)[ 8 ]Macro(LSOA)NegativeBinomialnnEngland2000-2005nychild--Pulugurthaetal.(2013)[ 16 ]Macro(TAZ)NegativeBinomialnnCharlotte,NC2005nn---ChiouandFu(2013)[ 6 ]Micro(seg-ment)Generalizedpoissonn-Japan2005yn---AgueroVal-Verde(2013)Micro(seg-ment)Fullbaye,zero-inatedandlog-normaln-PennRMS2003-2006yn---torbicetal.(2010)Micro(inter-section)NegativeBinomialY-torontoandcharlotte1999-2005ny---AziziandChiekolemi(2013)[ 4 ]Micro(U-Turn)EmpiricalBayen-Tehran(Iran)2001-2009nn---Lietal.(2013)[ 12 ]Macroscopic(CountyLevel)PoissonnGWRCADOT2007-2010nn---Hadayeghietal.(2013)[ 9 ]MacroPoissonnGWRToronto2001nn---Pirdavanietal.(2013)[ 15 ]MacroPoissonnGWRFlanders,Belgium2004-2007nn---Zhengetal.(2013)[ 19 ]MacroOLSnGWRHamptonRoadRegion,Virginia2006nn--19

PAGE 20

Table2-2. Impactofsocioeconomiccharacteristicvariablesinpedestriancrashpredictionmodel StudySocioeconomicTotalPopula-tionAverageAgeMinorityEmploymentIncomeEducation.EnglishSpeakerPernocarpertransitperwalkWorkatHomeSingleParentCommuteByCab Ukkusurietal.(2011)[ 18 ]+-+N/AN/A-N/AN/AN/AN/AN/AN/AN/ACottrillandThakuriah(2010)[ 7 ]heterogeneityMarginalEffect+-N/AN/A-N/A++++N/AN/AN/ACottrillandThakuriah(2010)[ 7 ]underreport-ingMarginalEffect+-N/AN/A-N/A-N/AN/AN/AN/AN/AN/ACharkravarthyetal.(2010)[ 5 ]+-N/AN/A---N/AN/AN/AN/AN/AN/AAbdel-Atyetal.(2013)[ 1 ]--+-N/AN/AN/AN/AN/AN/AN/AN/AN/APulugurthaetal.(2013)[ 16 ]N/AN/AN/AN/AN/AN/AN/AN/AN/AN/AN/AN/AN/AGreenetal.(2011)[ 8 ]N/AN/AN/AN/AN/AN/AN/AN/AN/A+-+20

PAGE 21

Table2-3. ImpactofTrafcCharacteristicandRoadInventoryvariablesinpedestriancrashpredictionmodel StudyTrafcandTransitCharacteristicTrafcVolume.ROADLengthIntersectionSubwayBusStopParkArea Ukkusurietal.(2011)[ 18 ]N/AVaried(mostly+)varied(mostly+)++-CottrillandThakuriah(2010)[ 7 ]heterogeneityMarginalEffect++N/A+N/ACottrillandThakuriah(2010)[ 7 ]underreport-ingMarginalEffect++N/A+N/ACharkravarthyetal.(2010)[ 5 ]N/AN/AN/AN/AN/AN/AAbdel-Atyetal.(2013)[ 1 ]+++N/AN/AN/APulugurthaetal.(2013)[ 16 ]N/A+N/AN/AN/AN/AGreenetal.(2011)[ 8 ]0N/AN/AN/AN/AN/A 21

PAGE 22

Table2-4. ImpactsofTrafcCharacteristicandRoadInventory StudyBuiltEnvironmentandLanduseHousingDensitySchoolOpenLandIndustrialCommercialOtherMiscLanduseCrime Ukkusurietal.(2011)[ 18 ]N/A++++N/AN/ACottrillandThakuriah(2010)[ 7 ]heterogeneityMarginalEffectN/A++N/AN/AN/A+CottrillandThakuriah(2010)[ 7 ]underreport-ingMarginalEffectN/A++N/AN/AN/A+Charkravarthyetal.(2010)[ 5 ]N/AN/AN/AN/AN/AN/AN/AAbdel-Atyetal.(2013)[ 1 ]+N/AN/AN/AN/AN/AN/APulugurthaetal.(2013)[ 16 ]++N/AN/A+AvailableN/AGreenetal.(2011)[ 8 ]N/AN/AN/AN/AN/AN/A+ 22

PAGE 23

CHAPTER3DATAThischapterpresentsanoverviewofthedatausedinthisstudy.Section3.1focusesoncrashdataandsection3.2focusesonalltheexplanatoryvariablesusedinmodeling. 3.1CrashDataInthisstudyweuse20052009crashdataprovidedbyFDOTintheformofGISles.Aftertheprocessofcleaningupthereare33,132pedestriancrashrecordsavailablefromthatperiod.Among33,132pedestriancrashes,mostareendwithonlyminorinjury(20,180crashes).Thereare2,564crashesyieldthehighestseverityleveloffatality(Atleast1personinvolvedinthecrashdiedwithin30dayaftercrashevent).7053crashesendwithincapacitatinginjury.Theother3,335caseareproperty-damageonly(PDO).Inthecaseoffatalitycrashes,97%offatalitiesarepedestrian.ThesepercentagesareshowninFigure 3-1 .ThelocationdistributionofcrashesisshowninFigure 3-2 .Thecrashesthathappenonintersectionareaccountedfor41%oftotalpedestriancrashes.Generalroadsectionthatisnotbridge,ramp,tollbooth,etc.areaccountedfor48%.Thisresultindicatesthatintersectionisexposurepointtotheriskofpedestriancrash.Soweexpectedthattheareawherethereishighnumberofintersectionwillhaveoverallhighernumberofpedestriancrashes.Thetimedistributionofcrasheseventaredifferentamongthetypeofcrashseverity.Figure 3-3 .showsthetimedistributionofpedestriancrashevent.WecouldseethatthecrasheventhappenmostlyatthetimeofAMpeakandPMpeakwiththehighestfrequencyhappenatthePMpeak.Onespeculationisthepeakoftotalcrashhappensearlierthanthepeakofsevereandfatalitycrash.Thisinfersthatthevisibilityhaseffectontheseverityofcrashes. 23

PAGE 24

Figure 3-4 .showsthedistributionofpedestriancrasheventamongtheweek.Thereisvisiblydifferencepatternbetweensevere/fatalcrashesandoverallnumberofpedestriancrashes.Thedaythathashighestfrequencyoffatal/severepedestriancrashesisSaturdayopposedtooverallcrashesthathappenmostlyonFriday.ThetimedistributionofcrashesisalsodifferentbetweenweekendandweekdayasshowninFigure 3-5 .andFigure 3-6 .Thereisavisiblydifferencebetweenweekendandweekdaypedestriancrashfrequency.ThecrashfrequencypatterninweekdayisbasedonAMpeakandPMpeakwithmorecrashhappenonPMpeak.Ontheweekend,however,thereisnovisibleAMpeakinthecrashprole.Thisisperhapsreectiveofthedifferenceintheoveralltravel-demandprolebetweenweekdaysandweekends(i.e.,weekdaysexhibitastrongerpeakingoftraveldemandsthanweekends)MostofthecrashesarehappenduringdaylighttimeasshowninFigure 3-7 .However,ifweconsideredonlythefatalcrashes(Figure 3-8 .),thepercentageofpedestriancrasheswhenthenaturallightconditionisdarkishigher(irrespectiveofwhetherstreetlightingispresentornot).Thisresultsuggeststhatthelightconditionhaseffectontheseverityofthepedestriancrash.Figure 3-9 .showstheinteractionofcontributionbetweendriverandpedestrian.Mostofthecrasheseventsarecontributebypedestrianaloneat35%.Thepercentageofeventthatcontributebydriverorbothdriverandpedestrianarealmostequalat19%.Anotherfactorthatisaninterestingissueintermofriskfactoristheusageofalcoholanddrug.Incaseofoverallpedestriancrashes,only10%ofpedestrianinvolvedwithcrashareunderinuencedofalcoholordrug(Figure 3-10 ).However,incaseoffatality,thepedestrianwhoisunderinuencedareaccountfor59%ofthefatality(Figure 3-11 ).Thisresultconrmstheotherinterestingfactaboutalcoholinvolvedpedestriancrashisthepercentageofinteractionbetweendrunkdriveranddrunk 24

PAGE 25

pedestrian.FromFigure 3-12 .,thepercentageofpedestriancrashthathappenwhenonlypedestrianisdrunkarehigherthanwhendriverdrunk.Duetotheneedtofocusingonthesocioeconomicaspect,wechoosecensusblockgroupasanalysisgeographicunitratherthantrafcanalysiszone.Withtheassumptionthatbuiltenvironmentandsocioeconomicdoesnotchangemuchintheperiodofanalysis,weuse2010censusdatainthisstudy.Thenthecrashdataaremappingonthecensusblockgroupandcalculatedasaggregatecrashdata.Duetotheerrorofgeocodingofsomecrashesevent,thetotalnumbersofcrashesarefurtherreducingto32,917.ThemappingdataisshownonFigure 3-13 .,Figure 3-14 .,Figure 3-15 .andFigure 3-16 .Theshadingofthemapsrepresentsthenumberofaggregatecrashincensusblockgroup.Thecensusblockgroupwithwhiteshadingindicatethatitshasnorecordedcrashevent.Thedarkertheshadingindicatesthehighernumberofpedestriancrashcount.Mapfortotalcrash,severecrashandnighttimecrashmodelhavesimilarscaleofdisplay(lightgrey1-5crashes,darkgrey6-15crashes,blackmorethan15crasheswithintheperiodof5years)whilefatalcrashmapusedifferentscaleasithasgenerallylessnumber.Among11442censusblock-groupinFlorida,thereare8,233blockgroupsthathaveatleastonerecordofpedestriancrashinthatperiod.ThehighestrecordfortotalnumberofpedestriancrashincensusblockgroupisfromablockgroupindowntownJacksonville.ThedescriptivestatisticsoftotalpedestriancrashincensusblockgroupareshowninTable??.DescriptivestatisticofcensusblockpedestriancrashinFloridafrom2005-2009.Thefrequencyofcrashwillbeusedasthedependentvariableoftheregressionmodel.Thereare4typeofcrashthatweconsiderinthisstudy.First,thetotalnumberofcrashisthetotalnumberofcrashinthecensusblockgroupregardlessoftheseverityofthecrash.Severecrashisthecrashthatresultinginatleast1severeinjuryorfatality.Fatalcrashisthetotalnumberofcrashthathasatleast1fatalityintheevent.Finally, 25

PAGE 26

nighttimecrashisthecrashthathappensfrom6pmto6amofeachday.Eachofthesevariableswillhavetheirownseparatemodel.ThefrequencydistributionisshownonFigure 3-17 .,Figure 3-18 .,Figure 3-19 .andFigure 3-20 3.2ExplanatoryVariablesTheexplanatoryvariablesusedintheanalysisarebroadlyclassiedintofourcategories:socioeconomicvariables,trafcandtransit,landuseandbuiltenvironment,andlocationcontext.Thissectiondedicatedtothedescriptionoftheconstructionofthesevariables.ThedescriptivestatisticsofthedataisshowninTable 3-2 3.2.1SocioeconomiccharacteristicFirstcategoryofexplanationvariableisSocioeconomiccharacteristic.Allofthevariableinthisgrouparederivedfrom2013census.TotalPopulationisdirectlyderivedfrom2010censusdata.Itisthetotalnumberofpopulationincensusblockgroup.Among11442censusblockgroupinFlorida,thereare86non-inhabitedcensusblockgroup.Sowecouldsaythatmostofthecensusblocksgrouparewellinhabited.However,thisvariableisnotconsideredthesizeofcensusblocksowecouldnottelltheactuallivingdensity.LandAreaisthetotallandarea(excludewater)ofthecensusblockgroup.Thisactasthecontrolvariableonthetermofsizedifferentofcensusblockgroup.Thereare46censusblockgroupswithnolandareainFloridawhichwillnotincludeintheanalysis.Themeanvalueof12138392.03squaremetersandthemaximumvalueof2,320,655,475showthevariability.Thisvariableshouldalwaysconsideralongwithtotalpopulationforthebettersenseofactualinhabitedcondition.Thepopulationdensityiscalculatedfromthetotalpopulationdividebytotallandareaofcensusblockgroup(0forthecensusblockgroupsthathavenolandarea).Thisvariabledisplaystheactualinhabitedconditionofcensusblockgroup.Thecensusblockgroupsthatlocatedinmoreurbanizedareashouldhavehighervalueofpopulationdensity.NextispercentageofpeoplewhocouldnotspeakEnglishuently.Themedianvalueis1.23whilethemaximumvalueis50.Thisindicatesthevariabilityofthedata.This 26

PAGE 27

variableisincludedbecausethepastresearchindicatethatpopulationwhocouldnotspeakEnglishverywellaretendtohavehigherriskofpedestriancrash.Themedianhouseholdincomeandpercentageofpopulationthatliveunderpovertylinerepresenttheeconomiccharacteristicofcensusblockgroup.Themedianhouseholdincomecouldtellthelargedetailabouttheeconomicincensusbloggroup.However,asthereareotherfactorsthatcouldaffectthepovertycondition,thepercentageofpovertyisaddedtothemodel.ThepovertythresholdisdefybyCensusBureauandconsideredbythetotalhouseholdincomeandnumberoffamilymember.Themeanmediandividedbytotalpopulation.Themeanandmedianofthevalueareveryclosenumberat60%.householdincomeforcensusblockgroupinFloridais51,839withthemedianof46,136.Onaverage,14.39%ofpopulationforeachcensusblockgroupwillliveunderthepovertylinewiththemedianvalueof11.29%.FinalvariableinthisgroupisPercentageofHighschoolGraduate.Thisvariableisdenedasthepercentageofpopulationwhoareatleasthighschoolgraduate 3.2.2TransportationCharacteristicThesecondgroupofexplanationvariableisTransportationCharacteristic.Firstvariablefromthisgroupisthetotallinearlengthofthenon-accessedcontrolroaddividebythetotalareaofcensusblockgroup(Metre/Sq.Metre).Thisvariablehasbeenconstructedbyaggregationofthetotallengthofnon-accessedcontrolroadinthecensusblockgroup.Thisvariablerepresentthedensityoftheroadcompareincensusblockgroupwhichconcerningthescalingofcensusblockgrouparea.Themeanis10.9x10-4m/sq.mwhilemedianis6.5x10-4m/sq.m.Nextisthetotalnumberofintersectioninthecensusblockgroup.ThesourceofthisvariableisthedatafromFDOTinformofGISle.Themeanandmedianare17.9and13.Themaximumvalueis291.Intersectionisanexposurepointforpedestriancrashespeciallyatthecrossingpath.Totalworktripsperweekinthecensusblockgrouprepresentbothvehicletripandpedestriantripinthecensusblockgroup.Thisvariableisincludedin2010census. 27

PAGE 28

Thisvariablealsoactsastheproxyfortrafcvolumeoftheroadasinsomesectionoftheroadthetrafcvolumemaynotberecord.NextisNumberoftransitstation.Thisvariableistheaggregationoftotalnumberofxed-waytransitstationinthecensusblockgroup.Theserviceofxedway(lightrailandsubway)inFloridaisstilllimittoMiami-Dadecounty,BrowardCounty,PalmBeachcountyandJacksonville.Thisisoppositetotheserviceoftransitbuswhichhasmuchmorecoverageinotherpartofthestate.Transitstationisatypeofareathathashighexposureofcrashrisk.Duetothelackoftruebusstopdatafromsomecounty,thetotallengthofbusrouteacrosscensusblockgrouphasbeenusedinstead.Thisvariableistheaggregatetotallineardistanceofbusrouteinthecensusblockgroup.Despitenotbeingthedirectreplacement,thisistheonlyavailabledataoftransitsystemthatavailablestatewide.Theassumptionistheareathathashighdensityoftransitsystemwillhavemoreexposureforthepedestriantotheriskofthecrash.ThelasttransportationcharacteristicvariableisMedianAge.Thisvariablerepresentthemedianageofpopulationincensusblockgroup.Thisvariableisdirectlyderivedfromthe2010Census.Themeanvalueis41yearsold.Pastresearchsuggestthatvictimofpedestriancrasharefrequentlyatyoungerage.Theaveragevalueis42.9yearswiththemedianof41.2years. 3.2.3LanduseandBuilt-environmentFirstvariablefromthisgroupiscountofeducationalfacility.Thisvariableistheaggregatenumberofeducationfacilityincensusblockgroup.ThepositionofalleducationfacilityinFloridahasbeenmappingandaggregatingforthetotalnumberneglectingthedifferenceoftype.Thisvariableisincludedbecauseofthehighpedestrianclusternatureofeducationfacility.SecondisPercentageofresidentialarea.Thisvariablerepresentsthepercentageofnon-waterlandthatisfunctioningasresidentialarea.Thisvariableisderivedfromgeneralizelanduseparcelmap.Theaveragepercentageofresidentialuseis35%.Therearealsotheareawhereallterrainhasbeenuseasresidentialareaandlandwithoutpermanentinhabitant. 28

PAGE 29

Similartopercentageofresidentialarea,Percentageofcommercialarearepresentsthecommercialarea.Itsalsoderivedfromthegeneralizedlanduseparcelmap.Therearecensusblockgroupthatuseallterrainforcommercialarea.theaveragevalueforcensusblockgroupinFloridais6.67%withthemeanof2.87%.Thelastvariablethatderivedfromgeneralizedlanduseparcelmapispercentageofindustrialarea.Itisthepercentageofindustrialzoneovertotalnon-waterterrain.Thedifferenceisthattheratioofusageasindustrialzoneissmallerthanresidentialandcommercialwiththemeanandmedianofonly0.9%and0.4%.Themaximumpercentageis63%. 3.2.4LocationcontextFirstvariablefromthisgroupisDistanceofbigcity.ThisvariableistheEuclidiandistancefromthecenterofcensusblockgrouptocenterpointofbigcity.Theoretically,thisvariablewillreecttheurbanizationofthecensusblockgroup.Thehighervalueofthisvariableindicatesthattheareasarelocatedinthemoreremotearea.Inthisstudyweselectthecutpointforbigcityasthecitywiththetotalpopulationmorethan249,999.Urbanandurbanclusteristhedummyvariablethatindicatedthatthecensusblockgroupislocatedinurbanandurban-clusterareaornot(1forurbanandurban-cluster,0forrural).Outof11,442censusblockgroupofFlorida,9446censusblockgrouparelocatedinurbanizedorurban-clusterarea.Despitethehighernumberofcensusblockgroup,urbanareaareaccountedforonly11.7%(4.95from42.08millionacres).Thisisbecausethesizesofcensusblockgroupinruralareaaretendingtomuchlarger.Nextiscountylevelpopulationdensityisthedensitywhichrepresenttheoverallpictureoftheinhabitantconditionincountylevel.Countylevelmedianhouseholdincomeisdirectlyderivedfromcensus2010.Itisrepresentthesocioeconomicconditioninthebiggerscalethantheinternaleffectwithincensusblockgroup.Countylevelpercentageofresidential,commercialandindustrialzoneisthepercentageoflandareathathasbeenuseasresident,industrialandcommercialareainthecounty.Typically 29

PAGE 30

only1or2percentofthecountyareawillbeusedasindustrialorcommercialarea.Theresidentialareawillaccountedaround14%ofthetotalcountyarea. 30

PAGE 31

Figure3-1. Percentageofcrasheventsbyseverity Figure3-2. Locationdistributionofpedestriancrash 31

PAGE 32

Figure3-3. Timedistributionofpedestriancrash Figure3-4. Dayofweekcrashesdistribution 32

PAGE 33

Figure3-5. TimeDistributionofweekdaypedestrianCrash Figure3-6. TimeDistributionofWeekendPedestrianCrash 33

PAGE 34

Figure3-7. Distributionofcrashesbylightcondition Figure3-8. Distributionoffatalcrashesbylightcondition 34

PAGE 35

Figure3-9. Pedestrian-Vehicleinteractionofcontributiontocrash Figure3-10. Usageofdrug/alcoholonpedestrianinvolvedcrash 35

PAGE 36

Figure3-11. Usageofdrug/Alcoholonpedestriancrashfatality Figure3-12. Theinteractionofalcoholusagebetweendriverandpedestrianontheeventofalcoholinvolvedpedestriancrash 36

PAGE 37

Figure3-13. Mapofaggregatecrashincensusblockgroup 37

PAGE 38

Figure3-14. Mapofaggregateseverecrashincensusblockgroup 38

PAGE 39

Figure3-15. Mapofaggregatefatalcrashincensusblockgroup 39

PAGE 40

Figure3-16. Mapofaggregatenighttimecrashincensusblockgroup 40

PAGE 41

Figure3-17. Frequencydistributionofcrashes Figure3-18. Frequencydistributionofseverecrashes 41

PAGE 42

Figure3-19. Frequencydistributionoffatalcrashes Figure3-20. Frequencydistributionofnighttimecrashes 42

PAGE 43

Table3-1. DescriptivestatisticofcensusblockpedestriancrashinFloridafrom2005-2009 TotalCrashesSevereCrashesFatalCrashesNighttimecrashes Total(5years)329179551255311992Numberofblockgroupswith0crashes3164654994496047Maximumnumberofcrashesinablockgroup8521825Meancrashesperblockgroup2.890.840.221.05Varianceincrashesperblockgroup17.882.070.323.3 43

PAGE 44

Table3-2. Descriptivestatisticofvariables MedianMeanSD LandArea(Sq.M)1.20E+061.22E+0768198007PopulationDensity(Person/Sq.M)0.0011290.0016670.0023MedianHouseholdIncome(Dollars)461365183927376.68PercentageofPoverty11.2914.3912.6155PercentageofPopulationwhodonotspeakEnglishwell.1.243.9446.2212Percentageofpopulationwhograduatehighschool.61.0360.915.8988MedianAge(Years)41.142.9411.3271 TotalWeeklyWorkTrip591713.2577.466LinearDensityofnon-accessedcontrolroad(M/Sq.M)0.0006560.0011040.0019NumberofIntersection1317.9818.7381LengthofBusRoute(M)97.3536.11079.266NumberofTransitStation(GuidedRailonly)00.007020.1425 TotalNumberofEducationalFacility00.65141.0726Percentageofresidentialareacensusblockgroup35.9735.721.0107Percentageofcommercialareacensusblockgroup01.7019.9869Percentageofindustrialareaincensusblockgroup2.90886.69315.0188 DistancefromBigCity(M)119802171626502.66CensusBlockGrouplocatedinUrbanorUrbanClusterArea?(1foryes)10.82870.3768Populationdensityofcounty(Person/Sq.M)3.03E-043.65E-040.0003Medianhouseholdincomeofcounty(Dollars)45624448744929.621Percentageofresidentialareaincounty11.175414.18488.5823Percentageofcommercialareaincounty1.652772.239231.7037PercentageofIndustrialAreaincounty1.043951.305511.0654 44

PAGE 45

CHAPTER4EMPIRICALRESULTThischapterwilldiscusstheempiricalresultofthestudy.Inthisstudy,4planninglevelpedestriancrashesmodelhasbeenintroducedwhicharetotalcrashmodel,severecrashmodel,fatalcrashmodelandnighttimecrashmodel.Thetotalnumbersofcrashesinacensusblockgrouparerelatedtotheexplanatoryfactorsviathenegativebinomialmodel.Therearefourmodelsinthisstudy,totalcrashmodel,severecrashmodel,fatalcrashmodelandnighttimecrashmodel(Table 4-1 ).Thevariablearedividinginto4groups,socioeconomiccharacteristic,transportationcharacteristic,landuseandbuiltenvironmentandlocationcontext.VarianceInationFactor(Vif)hasbeenincludedintheTable 4-1 .toobservetheproblemofmulticolinearity.Pastresearch(Kock,N.,Lynn,G.S.(2012)[ 11 ])suggestvifvalueof5forthevariabletohaveproblemofmulticolinearity.FromtheresultTable 4-1 .,novariablehavemorevifvaluethan5sotheproblemofmulticolinearityisnotconcerned.Table 4-2 .showsthegoodnessoftforthemodel.Thesegoodnessoftvaluesarelog-likelihood,Pearsonpseudor-squareandMcfaddenpseudor-square. 4.1ImpactofexplanationvariableThissectionwilltalkabouttheempiricalresultofeachexplanationvariable.Theexplanationvariablearedivideintofourgroups.Theimplicationofimpactforsomevariableshouldnotbeconsideredaloneasthebetterpictureoftheireffectmighthappenwhentheinteractionwithothervariablesareconsidered. 4.1.1ImpactofsocioeconomiccharacteristicCensusblockgrouplevelpopulationdensityhavenegativecoefcientonall4modelsissomewhatcounterintuitive.However,thereisthepossibilitythatmostofthecrasharenothappenonresidentialarea.Theeffectsofothercoefcientconrmthisspeculation.Thecensusblockgrouplevelmediumhouseholdincomeisincludedinthemodelasanothersocioeconomiccharacteristic.Asthepastresearchsuggestthat 45

PAGE 46

themediumhouseholdincomehascorrelationwiththenumberofpedestriancrashinthegeographicunit.Thenegativecoefcientofthemediumhouseholdincomeareinthesamedirectionwithpastmodel(CottrilandThakuriah(2010)[ 7 ]andChakravarthyetal.(2010)[ 5 ]inthesensethatcensusblockgroupwithhighermedianhouseholdincometendtohavelesspedestriancrash.Asthemediumhouseholdincomemaynotreectthetruenumberofhouseholdwholivebelowthepovertyline,thepercentageofpeoplewholivebelowpovertylineisincludedisrequiredforthecoverage.Thisvariablehassignicancepositivecoefcientontotalcrashmodelandseverecrashmodel(Thisvariabledidnothavesignicanceeffectonfatalcrashmodel).Thepositivecoefcientinfersthatthepeoplewholivebelowthepovertylinehavemoresusceptibilitytocrash.Thiscouldbetheresultofmorewalkingtripthanpeoplewhoareliveoverpovertyline.PercentageofpeoplewhocouldnotspeakEnglishuentlyshowpositivecoefcientsonall4models.ItmeanthatthecensusblockgroupswithhigherpercentageofpeoplewhocantspeakEnglishuencytendtohavemorecrashthanthecensusblockgroupwithlowerpercentage.ThisresultissimilartopastmodelsuchasCottrilandThakuriah(2010)[ 7 ]andChakravarthyetal.(2010)[ 5 ].Pastresearch(Ukkusurietal.(2011)[ 18 ]andChakravarthyetal.(2010)[ 5 ])suggestthatthelevelofeducationhaseffectonthereductionofnumberofcrashes.Thepossibleexplanationisthepeoplewithlesseducationlevelaretendtohavemorewalkingtripthanpeoplewithhigheducationespeciallyworktrip.Asitgivepositivecoefcientinallmodelsexceptnighttimecrash.Finally,themedianagehasnegativecoefcientintotalcrash,severecrashmodelandnighttimecrashmodel(Itdoesnthavesignicanceeffectonfatalcrashmodel).TheseresultsaresimilartothestudyofUkkusurietal.(2011)[ 18 ]andAbdel-Atyetal.(2013)[ 1 ]. 4.1.2ImpactoftransportationCharacteristicThetotaltripsperweekinthecensusblockgrouprepresentthebothvehicletripandpedestriantripinthecensusblockgroup.Asthepedestriancrashrequireboththe 46

PAGE 47

automobileandpedestriantohappensotheusedoftotalnumberoftripisrelevant.Thepositivecoefcientconrmsthatthecensusblockgroupswithhighertotalnumberoftriparelikelytohavemorepedestriancrashes.LinearRoadDensityandIntersectioncountarethevariablesthatrepresenttheexposuretotransportationinthecensusblockgroup.Intersectioncounthavesignicancepositivecoefcientinallthreemodelswhilethelinearroaddensityonlyhavesignicancepositivecoefcientontotalcrashmodelandnighttimecrashmodel.Theresultsimplyaboutthedifferenteffectofseverity.Despitedensityofroadhastheeffectonthenumberofcrash,thecrashesarelikelytohavelowerseveritythatyieldonlypropertydamageandlightinjury.Thisimplicationisreasonableasthepedestriancrashthathappenonthesideoftheroadaretendtolessseverethantheonethathappenattheintersectionduetobothcrashdirectionandthelowerspeed.LengthofBusrouteandcountofxed-railtransitstationarethevariablesthatshowtheavailabilityofpublictransitinthecensusblockgroup.Fromthepastresearch,wecouldseethatthepublictransitstationisavulnerablepointforthepedestriancrash.Thesevariableshavesignicancepositivecoefcientsonlyinthetotalcrashesmodelsimilartothelinearroaddensity.Thisisalsoimplyingofthedifferenceseverityofthecrashes.Itissomewhatreasonableasdespitetheexposureofcrash,thecrashesthathappenduetothetransitsystemarelikelytohavelowerseveritysuchaspropertydamageandminorinjury. 4.1.3ImpactoflanduseandBuilt-environmentFrompreviousliterature,theschoolzoneisavulnerablezoneforthepedestriancrash.Thisissimilartothecurrentmodelasthepositivecoefcientincountofeducationalfacilityinall4modelsimpliedthatcensusblockgroupwithhighernumberofeducationfacilityistendtohavehighernumberofpedestriancrash.Percentageofcommercialandindustrialareainthecensusblockgrouphaveonlysignicancepositivecoefcientinallmodelscontrarytothepercentageofresidentialareaincensusblockgroupwhichhasnosignicanceeffectexceptthenighttimecrashmodel.Thisisthe 47

PAGE 48

implicationthatmostofthepedestriancrashesarenothappeninresidentialareawhichexplainsthereasonwhypopulationdensityhasnegativecoefcient.Thisisalsoshowthedifferenteffectsofsimilarvariablefromthedifferentscalewhichwillbediscussagainlater. 4.1.4impactoflocationContextThenegativecoefcientofDistancefrombigcityindicatethatremoteruralaretendtohavelessactivityandreceivedlessattractionfromthebiggercitywhichconrmbythenegativecoefcientofthevariable.Theeffectofurbanandurbanclusterdummyvariablehavethemostpolaroppositeeffectamong4models.Ithaspositivecoefcientontotalcrashesmodel,non-signicancecoefcientonsevereandnighttimecrashesmodelandnegativecoefcientonfatalcrashesmodel.Thepossibleexplanationisthelevelofseverityofcrashes.Fromthiscontext,themoreseverecrashesespeciallythefatalitiesaremorelikelytohappenonruralarea.Onefactorthatcouldaffectthisisthedrivingspeedwhichcouldbehigherintheruralareathantheurbanarea.Theotherfactorthatcouldntbeoverlookedistheavailabilityoftheemergencymedicalservicewhichcouldreducethecausality.Contrarytothecensusblockgroupcounterpart,countylevelpopulationdensityhaspositivecoefcient.Thisshowsthedifferentofsimilareffectfromthedifferentscale.Thecountylevelcounterpartshowsthepictureofoverallactivitythatcouldgeneratemorepedestriantriphencetheexposuretothecrashrisk.Fromthecombinationofeffectwecouldconcludethattheriskiestcensusblockgroupwillbethecensusblockgroupwithlowpopulationdensitywithinthehighpopulationdensitycounty.SimilartoCountylevelpopulationdensity,thecoefcientofcountylevelmedianhouseholdincomeisontheoppositedirectionofthecensusblockgroupcounterpart.Thecombinationofeffectindicatethatthecrasharelikelytohappenincensusblockgroupwithlowmedianincomeinthecountywithhighmedianincome.Thisisimplyingtheeffectofeconomicgap.Finallythenegativecoefcientofcountylevelpercentageofresidentialareasuggeststhatthecrashesarelikelytohappeningin 48

PAGE 49

thecountywithhavelowerpercentageofresidentialarea.Thewithcombinationeffectwithcensusblockgrouplevellandusevariable,itsuggestthatthetypeoflandusethathavemorecrashriskarecommercialandindustrialarea,especiallytheonethatlocateinhighdensitywithlowavailabilityofresidentialarea.Nighttimecrashmodelisanexception,percentageofcommercialareaispositiveinthiscasewhichindicatethedifferenceinthepatternofcommutationbetweennightandday. 4.2ModelApplicationThissectionwillbeabouttheusageofproposedmodelinthecurrentstateofsafetyplanning.Duetotherandomnatureofcrashevent,thecensusblockgroupthathaszerocrashesreportismaybefarfromperfectlysafe.Asthezero-reportwithintheperiodofstudyisabundant(accountaround40%ofallcensusblockgroup)sowewillfocusonthatcensusblockgroupinthisstudy.ThedescriptivestatisticofpredictionforcensusblockgroupwithreportzerocrashisshowninTable 4-3 .FromTable 4-3 .,eventhoughthestatisticalrecordforthesecensusblockgroupreportzerocrash.Theprojectedcrashesfromthemodelarereportacertainvalue.Thesevalueindicatethatthesafetyconditionofeachcensusblockgrouparenotthesamedespiteithassimilarzerocrashreport.Intheallpedestriancrashmodel,fromthemedianpointonward,thiscensusblockareevenhaveprojectedvalueofcrashesmorethan1.Therearecensusblockgroupthathassamesafetyconditionasthecensusblockgroupwithcrashreport.Astheeventofpedestriancrashwhiledidnotfullyrandom,itisstillhavemanyuncontrollablefactor. 4.3ConclusionThischapterisabouttheempiricalresultandtheirimplication.Inthisstudy,4modelshavebeenproposed.Firstisthebasicplanninglevelpedestriancrashmodel.Thenthereare2modelthathasbeensegmentationbythedifferenceseverityoftheoutcomewhichareseverecrashmodelandfatalcrashmodel.Finallythelastmodelisthemodelthatdedicatedtothecrashthathappensonthenighttime.Theeffectof 49

PAGE 50

variablesisdifferentamongall4models.Chapter5isthesummaryandconclusionofthestudy. 50

PAGE 51

Table4-1. Empiricalresult 2Variables TotalCrashModel SevereCrashModel FatalCrashModel NighttimecrashModel Estimate tvalue vif Estimate tvalue vif Estimate tvalue vif Estimate tvalue vif Constant 5.26E-02 0.39 -9.51E-01 -5.43 -1.77E+00 -7.33 -8.58E-01 -4.94 Socioeconomic PopulationDensity -1.32E+01 -2.18 1.69 -5.50E+01 -5.7 1.82 -7.75E+01 -4.75 1.65 -2.04E+01 -2.47 1.59 MedianhouseholdsIncome -7.06E-06 -12.09 1.72 -7.69E-06 -9.38 1.76 -1.23E-05 -9.49 1.45 -8.31E-06 -10.15 1.71 PercentageofPoverty 1.24E-02 11.51 1.82 1.01E-02 7.3 1.85 1.30E-02 9.67 1.73 PercentageofPopula-tionwhodonotspeakEnglishwell. 1.04E-02 5.15 1.61 1.63E-02 6.28 1.63 2.86E-02 7.66 1.47 1.87E-02 7.76 1.39 Percentageofpopula-tionwhograduatehighschool. -2.78E-03 -2.55 2.58 -3.29E-03 -2.31 2.59 -4.74E-03 -2.88 1.39 MedianAge -1.03E-02 -7.08 2.1 -5.68E-03 -3 2.08 -1.50E-02 -9.96 1.26 TransportationChar-acteristic TotalWeeklyWorkTrip 2.08E-04 10.63 1.33 2.32E-04 9.78 1.34 2.33E-04 7.12 1.28 1.99E-04 8.17 1.34 LinearDensityofnon-accessedcontrolroad 3.87E+01 7.1 1.14 3.39E+01 5.07 1.13 3.63E+01 5.37 1.12 NumberofIntersection 1.68E-02 26.48 1.47 1.64E-02 22.2 1.39 1.53E-02 14.67 1.43 1.82E-02 23.73 1.44 LengthofBusRoute 3.86E-05 4.04 1.05 NumberofTransitSta-tion(GuidedRailonly) 1.26E-01 2.09 1.02 BuiltenvironmentandLanduse TotalNumberofEduca-tionalFacility 8.47E-02 8.4 1.22 5.23E-02 4.16 1.22 4.03E-02 2.12 1.37 5.98E-02 4.7 1.21 Percentageofresiden-tialareacensusblockgroup 1.74E-03 2.13 1.39 Percentageofcommer-cialareacensusblockgroup 5.89E-03 2.91 1.15 2.67E-02 21.58 1.13 2.45E-02 13.22 1.18 2.83E-02 21.82 1.18 Percentageofindustrialareaincensusblockgroup 2.73E-02 26.59 1.11 8.50E-03 3.43 1.12 1.49E-02 4.22 1.13 7.80E-03 3.01 1.16 DistancefromBigCity -3.79E-06 -7.23 1.51 -3.23E-06 -4.84 1.46 -4.27E-06 -4.17 1.27 -2.57E-06 -3.95 1.43 LocationContext CensusBlockGrouplocatedinUrbanorUrbanClusterArea?(1foryes) 2.14E-01 5.89 1.42 -2.44E-01 -3.49 1.51 Populationdensityofcounty 7.86E+02 11.74 3.45 4.80E+02 5.44 3.56 Medianhouseholdincomeofcounty 1.54E-05 6.32 1.18 1.28E-05 4.06 1.16 1.38E-05 2.86 1.17 1.52E-05 4.88 1.15 Percentageofresiden-tialareaincounty -2.28E-02 -10.24 3.19 -9.26E-03 -3.15 3.28 Percentageofcommer-cialareaincounty 2.43E-02 2.59 1.29 PercentageofIndustrialAreaincounty 51

PAGE 52

Table4-2. Goodnessoftofthemodel ModelLog-likelihoodPearsonR2McFaddenR2convergenceNull TotalCrash-22798-25166.780.350.09FatalCrash-13243-14363.630.090.07SevereCrash-6135.8-6607.590.180.08NighCrash-14685-16071.930.180.09 Table4-3. Descriptivestatisticofpredictionforcensusblockgroupwithzerocrashreport AllCrashSevereCrashFatalCrashNighttimecrash P50.60.060.230.24P953.970.481.51.81Mean1.780.20.670.78Median1.370.160.520.6Variance7.290.040.350.59Count3164654994496048 52

PAGE 53

CHAPTER5SUMMARYANDCONCLUSIONThischapterisabouttheoverviewofthestudy.Theresultandapplicationtothecurrentstateofsafetyworkwillalsobediscussed.Finally,thechapterwillbeclosedwiththesuggestedlistoffuturework. 5.1SummaryThefocusofthisstudywasondevelopingmacroscopicorplanning-levelmodelsforpedestriansafety.Crashdatafrommultipleyears(2005-2009)andlandusedatafromtheentirestateofFloridaareusedindevelopingmodelsforpedestriancrashes.Atthisstateofwork,threemodelsweredevelopedtodeterminethecrashfrequencyforeachcensusblockgroup.Thesearemodelsfortotalcrashes,severecrashes,andfatalcrashes.Ofthe33,132crashesthatinvolvepedestriansintheanalysissample.9551crashesinvolveatleastoneinjury(i.e.,severecrash)and2553crashesthatinvolveatleastonefatality.Theestimatedmodelscapturetheeffectsofseveralsocioeconomic,transportation,landuse,andcontextualvariables.Inparticular,themodelsincludetheeffectofcertainvariablesatmultiplespatialscalesyieldinginterestingresults.Forexample,theeffectofpopulationdensityattheblockgroupisopposite(negative)tothatoftheeffectofcountyleveldensity(positive).Thisindicatesthatalow-densityblockgroupinhighdensitycountyistheriskiestcombinationintermsofpedestriancrashes.Similarly,theeffectofincomeisalsodifferentwhenexaminedatthetwoscales(negativeatblock-groupandpositiveatcounty)suggestingthatalow-incomelocationwithinahigherincomecountyisriskiest.Finally,byexaminingtheeffectsoflandusesatthetwospatialscales,itisalsoevidentthatcensusblockgroupsthataremorecommercial/industrialbutlocatedwithincountiesthathavemoreresidentsareriskierintermsofpedestriancrashes.Allthesecongurations(basedonpopulationdensity,income,andlanduses)clearlycapturethelocationswhicharelikelytohavealarger 53

PAGE 54

volumeofconictingvehicularandpedestrianmovementsandtherebymakingthemriskier. 5.2ConclusionPedestriansafetycontinuestobeoneofthecriticaltransportationissuesfacingthesociety.Crash-predictionmodelsareusefultoolstoidentifylocationsthathavehigherriskofcrashesandtoprioritizeprojects.Despitethereleaseofhighwaysafetymanual,thereisstillhavingtheneedtoestablishthemodelforpredictingpedestriancrashinplannerlevel.Themodelsproposedinthispapercouldbeusedtodeterminetheexpectednumberofcrashesforallthecensusblockgroupsinthestate.Thispredictiveassessmentexerciseservestohighlightthevalueofplanningmodels.Specically,ifsafetyassessmentsaremadepurelybasedoncrashhistory,allthelocationswithzeroobservedcrasheswillbedeemedequallysafe.However,thepredictivemodelhighlightsthatthereisasignicantvariabilityincrashriskacrosstheselocationsbecauseofdifferencesinlanduseandsocioeconomicpatterns.Thustheplanningmodelsdevelopedinthisstudycanbepowerfultoolsinstatewidesafetyfundsallocationandprioritizingofprojects. 5.3FutureworkAtthecurrentstateofthemodel,themodelcouldpredictthenumberofpedestriancrashthathappeninthecensusblockgrouplevel.Therearesomelimitationsthataretheresultofgeographiccharacteristicofthemodelsuchastheboundarycondition.SotheconsiderationofusingspatialanalysismethodsuchasGeographicallyWeightedRegression(GWR)shouldbeconsideredasthefuturework.Also,forthepracticaluseofpolicymakerandplanner,themodelneedstobeadjustedoftheparameterandvariableaccordingtotheirdemand.Lastbutnotleast,atthetimeofthestudy,therearetheproblemoftimelaggingbetweencrashdataandsocioeconomicandlandusedata.Asthesedataarechangefromtimetotimesotheimprovementofdatacollectionneedto 54

PAGE 55

beimproved.Alsothecrashpredictionmodelshouldbeupdateperiodicallytoimprovetheaccuracyandprecision. 55

PAGE 56

REFERENCES [1] Abdel-Aty,M,Chundi,SS,andLee,C.Geo-spatialandlog-linearanalysisofpedestrianandbicyclistcrashesinvolvingschool-agedchildren.JournalofSafetyResearch38(2007).5:571.CitedReferencesCount:12FHPERGAMON-ELSEVIERSCIENCELTDTHEBOULEVARD,LANGFORDLANE,KIDLINGTON,OXFORDOX51GB,ENGLANDISIDocumentDeliveryNo.:245FH. [2] Administration,NationalHighwayTrafcSafety.TrafcSafetyFacts-Pedestrians[August2012].2012. [3] Aguero-Valverde,Jonathan.FullBayesPoissongamma,Poissonlognormal,andzeroinatedrandomeffectsmodels:Comparingtheprecisionofcrashfrequencyestimates.AccidentAnalysisPrevention50(2013).0:289297.URL http://www.sciencedirect.com/science/article/pii/S0001457512001522 [4] Azizi,LandSheikholeslami,A.SafetyEffectofU-TurnConversionsinTehran:EmpiricalBayesObservationalBefore-and-AfterStudyandCrashPredictionModels.JournalofTransportationEngineering-Asce139(2013).1:101.CitedReferencesCount:13IZASCE-AMERSOCCIVILENGINEERSALEXANDERBELLDR,RESTON,VA20191-4400USAISIDocumentDeliveryNo.:069IZ. [5] Chakravarthy,B,Anderson,CL,Ludlow,J,Lotpour,S,andVaca,FE.TheRelationshipofPedestrianInjuriestoSocioeconomicCharacteristicsinaLargeSouthernCaliforniaCounty.TrafcInjuryPrevention11(2010).5:508.CitedReferencesCount:38NLTAYLORFRANCISINCCHESTNUTST,SUITE800,PHILADELPHIA,PA19106USAISIDocumentDeliveryNo.:654NL. [6] Chiou,YCandFu,C.Modelingcrashfrequencyandseverityusingmultinomial-generalizedPoissonmodelwitherrorcomponents.Ac-cidentAnalysisandPrevention50(2013):73.CitedReferencesCount:58SDPERGAMON-ELSEVIERSCIENCELTDTHEBOULEVARD,LANGFORDLANE,KIDLINGTON,OXFORDOX51GB,ENGLANDISIDocumentDeliveryNo.:079SD. [7] Cottrill,CDandThakuriah,P.Evaluatingpedestriancrashesinareaswithhighlow-incomeorminoritypopulations.AccidentAnalysisandPrevention42(2010).6:1718.CitedReferencesCount:44HSPERGAMON-ELSEVIERSCIENCELTDTHEBOULEVARD,LANGFORDLANE,KIDLINGTON,OXFORDOX51GB,ENGLANDISIDocumentDeliveryNo.:655HS. [8] Green,J,Muir,H,andMaher,M.Childpedestriancasualtiesanddeprivation.AccidentAnalysisandPrevention43(2011).3:714.CitedReferencesCount:43URPERGAMON-ELSEVIERSCIENCELTDTHEBOULEVARD, 56

PAGE 57

LANGFORDLANE,KIDLINGTON,OXFORDOX51GB,ENGLANDISIDocumentDeliveryNo.:742UR. [9] Hadayeghi,A,Shalaby,AS,andPersaud,BN.DevelopmentofplanningleveltransportationsafetytoolsusingGeographicallyWeightedPoissonRegression.AccidentAnalysisandPrevention42(2010).2:676.CitedReferencesCount:30GOPERGAMON-ELSEVIERSCIENCELTDTHEBOULEVARD,LANGFORDLANE,KIDLINGTON,OXFORDOX51GB,ENGLANDISIDocumentDeliveryNo.:568GO. [10] Huang,HLandChin,HC.Modelingroadtrafccrasheswithzero-inationandsite-specicrandomeffects.StatisticalMethodsandApplications19(2010).3:445.CitedReferencesCount:40SVSPRINGERHEIDELBERGTIERGARTENSTRASSE17,D-69121HEIDELBERG,GERMANYISIDocumentDeliveryNo.:644SVFunding:ThisstudywassupportedbyNationalUniversityofSingapore.TheauthersalsowanttothankDr.MohamedA.QuddusatLoughboroughUniversityforcollectingthedatausedinthecasestudy. [11] Kock,N.andLynn,G.Lateralcollinearityandmisleadingresultsinvariance-basedSEM:Anillustrationandrecommendations.JournaloftheAssociationforInforma-tionSystems13(7)(2012):546. [12] Li,ZB,Wang,W,Liu,P,Bigham,JM,andRagland,DR.UsingGeographicallyWeightedPoissonRegressionforcounty-levelcrashmodelinginCalifornia.SafetyScience58(2013):89.CitedReferencesCount:43WZELSEVIERSCIENCEBVPOBOX211,1000AEAMSTERDAM,NETHERLANDSISIDocumentDeliveryNo.:162WZFunding:ThisresearchissupportedbytheNationalKeyBasicResearchProgram(NKBRP)ofChina(No.2012CB725400),theNationalHigh-techRDProgramofChina(863Program)(No.2012AA112304),aswellastheScienticResearchFoundationofGraduateSchoolofSoutheastUniversity(No.YBPY1211). [13] Lord,DandMannering,F.Thestatisticalanalysisofcrash-frequencydata:Areviewandassessmentofmethodologicalalternatives.Trans-portationResearchParta-PolicyandPractice44(2010).5:291.CitedReferencesCount:166GZPERGAMON-ELSEVIERSCIENCELTDTHEBOULEVARD,LANGFORDLANE,KIDLINGTON,OXFORDOX51GB,ENGLANDISIDocumentDeliveryNo.:599GZFunding:IntheUS,theNationalStrategicHighwayResearchPrograminitiatedaseriesofstudiesinrecentyearswiththeobjectiveofaddressingmanyofthefundamentalquestionsrelatingtocrashcausationandinvolvement(see,forexample,Dingusetal.,2006).Thesestudiesarebasedonnaturalisticdrivinginformation,whereinaselectedpoolofdriversisobservedoverprolongedperiodsintermsoftheircrash,near-crashandincidentinvolvements.Shankaretal.(2008)haveattemptedtoconstructoneplausiblestatisticalapproachtoextractinsightsfromnaturalisticdrivingdata. 57

PAGE 58

However,forthemostpart,naturalisticdrivingdatahavenotyetprovidedsignicantnewinsightswithbroadapplicability.Theuseandstatisticalanalysisofthesedataarealsohamperedbyprivacyissuesrelatingtodriver-identifyingvariablesandotherpotentiallitigationissues. [14] Marshall,WEandGarrick,NW.Doesstreetnetworkdesignaffecttrafcsafety?AccidentAnalysisandPrevention43(2011).3:769.CitedReferencesCount:48URPERGAMON-ELSEVIERSCIENCELTDTHEBOULEVARD,LANGFORDLANE,KIDLINGTON,OXFORDOX51GB,ENGLANDISIDocumentDeliveryNo.:742UR. [15] Pirdavani,Ali,Brijs,Tom,Bellemans,Tom,andWets,Geert.SpatialanalysisoffatalandinjurycrashesinFlanders,Belgium:applicationofgeographicallyweightedregressiontechnique.92ndTRBAnnualMeeting.ed.TransportationResearchBoard.TransportationResearchBoard,????,1. [16] Pulugurtha,SS,Krishnakumar,VK,andNambisan,SS.Newmethodstoidentifyandrankhighpedestriancrashzones:Anillustration.Acci-dentAnalysisandPrevention39(2007).4:800.CitedReferencesCount:22JUPERGAMON-ELSEVIERSCIENCELTDTHEBOULEVARD,LANGFORDLANE,KIDLINGTON,OXFORDOX51GB,ENGLANDISIDocumentDeliveryNo.:186JU. [17] Torbic,DJ,Harwood,DW,Bokenkroger,CD,Srinivasan,R,Carter,D,Zegeer,CV,andLyon,C.PedestrianSafetyPredictionMethodologyforUrbanSignalizedIntersections.TransportationResearchRecord(2010).2198:65.CitedReferencesCount:9TJNATLACADSCIENCESCONSTITUTIONAVENW,WASHINGTON,DC20418USAISIDocumentDeliveryNo.:730TJFunding:TheresearchdescribedherewasfundedbyNCHRP.TheauthorsthankStevenKodamaoftheCityofTorontoandCharlieJonesoftheCharlotteDOTfortheirsupport. [18] Ukkusuri,S,Hasan,S,andAziz,HMA.RandomParameterModelUsedtoExplainEffectsofBuilt-EnvironmentCharacteristicsonPedestrianCrashFrequency.TransportationResearchRecord(2011).2237:98.CitedReferencesCount:39ECNATLACADSCIENCESCONSTITUTIONAVENW,WASHINGTON,DC20418USAISIDocumentDeliveryNo.:892ECFunding:ThisworkwasfundedbytheNewYorkCityDepartmentofTransportation. [19] Zheng,Libing,Robinson,R.Michael,Khattak,Asad,andWang,Xin.AllAccidentsareNotEqual:UsingGeographicallyWeightedRegressionsModelstoAssessandForecastAccidentImpacts.InternationalConferenceonRoadSafetyandSimulation.2013. 58

PAGE 59

BIOGRAPHICALSKETCH KhajonsakJermprapaiwasborninSamutPrakarn,ThailandinMarch,1984.Hereceivedbachelor'sdegreeincivilengineeringfromChulalongkornUniversity,Thailandin2009.HebecameacertiedcivilengineerandbeganhisjobatBureauoftrafcsafety,DepartmentofRuralRoadofThailand(DRR)in2010.KhajonsakisalsoamemberofIRFFellowshipclassof2013.WithinthetwoyearsperiodofworkinginDRR,Khajonsakreceivedhands-onexperienceintheeldoftrafcsafetywork.HehasbeenworkingonseveralscasesofroadsafetyissueasthesituationofroadsafetyprobleminThailandisveryseverewith19.90fatalitiesper100,000populationsperyear.HereceivedascholarshipfromtheDepartmentofRuralRoadandstartedhismaster'sresearchatUniversityofFloridaunderthesupervisionofAssistantProfessorSivamahakrishnanSivain2012.Hisresearchhasbeenfocusingonthefactorrelatedtopedestriancrash. 59