valuation of r1-6 gateway treatment lternatives …a series of studies evaluated the gateway...

MichiganDepartmentofTransportation

EVALUATIONOFR1-6GATEWAYTREATMENTALTERNATIVESFORPEDESTRIANCROSSINGSFinalReport

Preparedby:WesternMichiganUniversityT.Y.LinInternationalFebruary2016

EvaluationofR1-6GatewayTreatmentAlternativesforPedestrianCrossings


1. Report No. RC-1638

2. Government Accession No. N/A

3. MDOT Project Manager Carissa McQuiston 5. Report Date

4. Title and Subtitle Comparison of Alternative Pedestrian Crossing Treatments 6. Performing Organization

Code N/A

7. Author(s) Ron Van Houten, Jonathan Hochmuth

8. Performing Org. Report No. N/A 10. Work Unit No. (TRAIS) N/A 11. Contract No. 2013-0069

9. Performing Organization Name and Address Western Michigan University 1903 West Michigan Avenue Kalamazoo, Michigan 49008

11(a). Authorization No. Z2 13. Type of Report & Period Covered Final Report 10/01/13 to 2/28/2016

12. Sponsoring Agency Name and Address Michigan Department of Transportation Research Administration 8885 Ricks Rd. P.O. Box 30049 Lansing MI 48909

14. Sponsoring Agency Code N/A

15. Supplementary Notes 16. Abstract. A series of studies evaluated the Gateway configuration of R1-6 in-street signs. A Gateway configuration consists of an R1-6 sign on the centerline, an R1-6 signs on both edges of the roadway, and R1-6 signs on the lane lines if it is a multilane road. The Gateway treatment was markedly more effective than the use of signs only on the centerline or on lane lines. The results showed: 1. That yielding was related to the narrowness of the gaps between signs; that a Gateway configuration consisting of R1-6 signs was markedly more effective than a Gateway made up of blank signs of the same size and background color. Research also demonstrated that mounting edge signs on the curb face was nearly as effective of placing signs in the gutter pan, and that the use of a robust delineator in place of the R1-6 signs on the lane lines was almost as effective as the use of the R1-6 signs on lane lines. This study also examined the efficacy of the Gateway on different types of crosswalk applications, the long-term persistence of effects and sign survival. 17. Key Words In-street sign, R1-6 Sign, Gateway Treatment of R1-6 Sign, Gateway

18. Distribution Statement No restrictions. This document is available to the public through the Michigan Department of Transportation.

19. Security Classification - report Unclassified

20. Security Classification - page Unclassified

21. No. of Pages 98

22. Price N/A


TABLEOFCONTENTSExecutiveSummary 1Introduction 4Chapter1–ConfigurationsInfluencingtheEfficacyoftheGatewayTreatment 11

Chapter2-EvaluationofVariationsthatImprovetheSurvivaloftheGatewayTreatment 26Chapter3-EvaluationoftheGatewayTreatmentinVariousCrosswalkApplications 35

Chapter4-PotentialRoadwayCharacteristicsInfluencingtheefficacyoftheGatewayTreatment 48Chapter5–EvaluationoftheLongTermEfficacyandSurvivaloftheGatewayTreatment 51

Chapter6-CostsforInstallationandMaintenanceoftheGatewayTreatment 53Chapter7Conclusions 54

Bibliography 57Appendix–StatisticalDataAnalysis 59

LISTOFTABLESTableA-1Numberofin-streetsignsitesreportedby11poolfundstates 9

TableA-2Numberofin-streetsignsitesreportedin24cities 10Table1-1Percentageofdriversyieldingtopedestrians 18

Table1-2Meanpercentageofdriversyieldingtopedestriansduringbaseline,theblanksigngatewayandtheR1-6signgatewayconfigurations 22

Table1-3Meanpercentageofdriversyieldingtopedestriansduringbaseline,thefullgateway,R1-6signsonlyattheedgeoftheroadway,andR1-6signsonlyonthelanelines 23

Table2-1Percentofdriversyieldingtopedestriansduringthebaseline,gutterpanplacementandcurbtopplacementconditions 29

Table2-2Percentofdriversyieldingtopedestriansateachcrosswalkduringthebaseline,thegatewaywithallR1-6signs,andthegatewaywithdelineatorcondition 33

Table3-1Percentofdriversyieldingtopedestriansduringeachconditionatthetwotrailcrossings 44

Table3-2PercentofdriversyieldingatthemidblockcrosswalkinAlleganwhentheRRFBwasactivatedandnotactivated,duringbaselineandafterthegatewaywasinstalled 46

Table4-1SelectroadwaycharacteristicsforMDOTproject120239(ResearchonComparisonofAlternativePedestrianCrossingTreatments)andMDOTproject114527(EvaluatingPedestrian

SafetyImprovement)andyieldingresults 49Table5-1Percentofdriversyieldingtopedestriansduringbaseline,initialtemporaryinstallation,

andafterpermanentinstallationateachofthetreatmentsites 51Table6-1Costsforinstallationandmaintenanceofthegatewaytreatment 53

Table6-2Removalandreinstallationtimesforgatewaytreatment 53

LISTOFFIGURESFigure1-1Diagramfortypicalsignplacementforgatewaytreatment 12

Figure1-2R1-6Signmountedonacurbtypebaseaffixedtotheroadway 16Figure1-3AblanksignthesamesizeastheR1-6signmountedtoaremovablebase 16

Figure1-4Arobustdelineatorattachedtoabasecementedintothepavement 17


Figure1-5DriveryieldingattheMarshallTrafficCircleNWcrosswalkleg 19Figure1-6In-streetsignshowingagatewaywithallR1-6signs 20

Figure1-7In-streetsignshowingagatewaywithallblanksigns 20Figure1-8Percentageofdriversyieldingtopedestriansduringeachconditionatthetwocrosswalks

onRoseStreetatAcademy 21Figure1-9Percentageofdriversyieldingtopedestriansduringeachconditionatthetwocrosswalks

onRoseStreetatKVCC 22Figure1-10Driverspeedapproachingthecrosswalkatthedilemmazone(blueline)andatthe

crosswalk(redline)duringbaselineandgatewayconditiononRoseStreet 24Figure2-1GatewayconfigurationwiththeedgesignonthecurbatacrosswalkonHuronStreet 27

Figure2-2Percentofdriversyieldingduringeachconditionoftheexperiment 29Figure2-3Percentofdriversyieldingduringeachconditionoftheexperiment 30

Figure2-4Percentofdriversyieldingduringeachconditionoftheexperiment 30Figure2-5Percentofdriversyieldingduringeachconditionoftheexperiment 31

Figure2-6Gatewaytreatmentwithdelineatorsplacedonwhitelanelines 32Figure2-7PercentageofdriversyieldingtopedestriansatRoseandAcademyduringbaseline,the

R1-6gatewayconditions,andthegatewaywithdelineatorcondition 33Figure2-8PercentageofdriversyieldingtopedestriansatRoseandKVCClocationduringbaseline,

R1-6gateway,andgatewaywithdelineatorconditions 33Figure2-9PercentageofdriversyieldingtopedestriansatWestnedgeandRaneyduringbaseline,

theR1-6gatewayconditions,andthegatewaywithdelineatorcondition 34Figure3-1Left:widegateway.Right:narrowgateway 35

Figure3-2DriveryieldingattwocrosswalksatatrafficcircleinMarshall,Michigan 36Figure3-3Gatewayconfigurationatoneoftheroundaboutlocations 37

Figure3-4Percentageofdriversyieldingduringthebaselineandgatewaytreatmentsatthetworoundaboutlocations 37

Figure3-5Ratioofdriversyieldingtopedestriansuponenteringandexitingtheroundaboutduringthebaselinecondition 38

Figure3-6PhotographofthesidewalkapproachingthenorthentrancetoInterstateI-94atWestnedgeAvenue 39

Figure3-7Left:wideconfigurationofthegateway.Right:narrowconfigurationwithdelineator 39Figure3-8PercentageofdriversyieldingtopedestriansatnorthandsouthentrancerampstoI-94

duringeachconditionoftheexperiment 41Figure3-9WidegatewayconfigurationonOaklandStreet 42

Figure3-10WidegatewayconfigurationonatthetrailcrossingonGardenLane 43Figure3-11Percentofdriversyieldingtopedestriansduringeachconditionatthetwotrail

crossingsduringeachtreatmentcondition 44Figure3-12RRFBlocationinAllegan,Michigan 45

Figure3-13DiagramofintersectionofMainStreetandBennettStreet,ThreeRivers,Michigan 46Figure3-14Percentofdriversyieldingatthelegofanintersectionwithgatewaytreatmentand

otherlegofintersectionthatwasnottreated 47Figure4-1StudyAreaMap 50


LISTOFACRONYMSAverageDailyTraffic ADT

Baseline BLFederalHighwayAdministration FHWA

InstituteofTransportationEngineers ITEManualonUniformTrafficControlDevices MUTCD

MichiganDepartmentofTransportation MDOTMilesperHour mph

MinnesotaDepartmentofTransportation MNDOTNationalCooperativeHighwayResearchProgram NCHRP

NationalHighwayTrafficSafetyAdministration NHTSAPedestrianHybridBeacon PHB

RectangularRapidFlashingBeacon RRFBWesternMichiganUniversity WMU

EvaluationofR1-6GatewayTreatmentAlternativesforPedestrianCrossings 1

EXECUTIVESUMMARYTheGatewayconfigurationofR1-6,instreetsigns,hasbeendocumentedtoproduceamarkedincreaseindriveryieldingtopedestriansovertheuseofasinglesign.FigureashowsaGatewayconfigurationwith a signon the centerline, a signs at the edge of the road, and signson the

whitelanelinesthatdivideslanes.In2013,theMichiganDepartmentofTransportation(MDOT)initiated a multi-year study with Western Michigan University (WMU) in order to evaluatefactorsrelatedtotheefficacyofagatewaytreatmentusingR1-6signs(gatewaytreatment);the

longtermeffectsofpermanentinstallations;configurationsthatcontributetotheeffectivenessofthetreatment;andthelong-termsurvivalofthetreatment.

Figurea.ApictureofaGatewayconfigurationofR1-6signs.

Aseriesofstudieswerecompletedinordertodeterminetheeffectivenessofthein-streetsigngatewaytreatment,determinewheretheyshouldbeused,anddeterminethecostbenefitsofusingthein-streetsigngatewaytreatmentincludingoperation,maintenanceandreplacement

costsandcomparativeanalysiswiththeRRFBandHybridBeacon.Inordertoaddresstheseobjectives,theWMUteamconductedanumberofactivities;eachiscapturedasanindividualchapterofthisreport.Althougheachoftheseinterventionswereonlycomparedatarelatively

smallnumberofsites,eachconditionwasintroducedmultipletimesateachsiteproducingmultiplereplicationsofeachcondition,theobviouslargechangesindriverbehavior,andtheconsistencyofresultsatthesesitessuggestthatthesefindingarerobust.Thisconclusionis

supportedbythestatisticalanalysisreportedintheappendixtothisreport.Alloftheconfigurationstestedincreaseddriveryieldingbehavior.

One hypothesis examined was that the narrowness of the gap between the signs would beinversely related to treatment effect size. This hypothesiswas confirmedwith narrower gaps


leading to larger increases in driver yielding right-of-way to pedestrians. Another hypothesis

tested was that the sign message itself had little influence on driver yielding behavior. Thishypothesiswasfoundtobeincorrect;agatewayconfigurationconsistingofallblanksignswassignificantly lesseffectivethantheGatewayconfigurationwiththemessagepresent(standard

R1-6signs).Aconfigurationanalysisalsoshowedthatthepositionofthesignisacriticalfactorinfluencing

driveryieldingbehavior.Notallpositionsusedinisolationresultedinthesamedegreeofdriveryielding right-of-way to pedestrians. Signs placed on the white lane line alone exertedmorecontrol over driver yielding behavior than signs placed at the edge of the roadway positions

alone at all sites and both types of placements produced less yielding than the full Gatewaytreatment. However, the partial effects produced by all of the partial gateway configurationsshowsthatifasignishitduringaseasonandnotreplaceduntilthefollowingyearwhenallthe

signsarereinstalledafterwinter,theremainingsignswouldstillbeofbenefittopedestriansfortheremainderoftheseason.

SpeeddatacollectedatonesiteshowedtheGatewaytreatmentisassociatedwithlargespeedreductions evenwhen pedestrianswere not present at the crosswalk and that drivers beganslowingatthedilemmazone.Thesedatasuggeststhatthespeedreductionislikelygradual.This

is an important finding because reduced speed gives drivermore time to respond to avoid acrash.Thisfindingrequiresfurtherreplication.

Although these studies and previous conducted research (2) demonstrate that the gatewaytreatmentproducedchanges indriveryieldingbehaviorat crosswalksonmultilane roads thatrivalstreatmentsthatareoneandtwoordersofmagnitudemoreexpensivetoinstall,itisalso

importanttoshowthatthistreatmentwillnotrequireexcessivemaintenanceefforts. Inareasrequiringsnowremoval,thesignswouldneedtoberemovedduringthewintermonths.Mostofthesignstestedcanberemovedandreinstalledquicklyaftertheinitialinstallation.Although

in-street signs are designed to rebound after being struck, sign survival withmultiple strikeswouldbeexpectedtobeanissue.

Two configurations were tested that should greatly increase sign survival only producedmoderatedecrementsintheefficacyoftheGatewaytreatment.Thefirstconfigurationtested

involved installingsignsontopof thecurb facerather than in thegutterpan (requiredFHWApermissiontoexperiment).Placingsignsontopofthecurbwasonlyassociatedwithonlyasmallreductionintheeffectivenessofthetreatment.Signsontopofthecurbfacearelesslikelytobe

struck thansignsplaced in thegutterpananddonotpresentaproblemfor sweepersnordotheyintroducepotentialdrainageissues.

The most vulnerable element of the sign configuration is the signs placed on white linesseparating lanes carrying traffic in the same direction. The substitution of a robust flexibleyellowgreendelineatordevicefortheR1-6signwouldbeexpectedto increasethesurvivalof


signs placed on thewhite lane lines. A delineatorwas selected thatwas tested by the Texas

TransportationInstituteandfoundtosurvive100strikesat60mph.Thistypeofadeviceshouldhaveanevenlongerlifespanonurbanroadswithspeedlimitsof35mphorless.DatareportedinChapter2showsthattheuseofarobustdelineatoronlyproducedamodestreductioninthe

effectivenessoftheGatewaytreatment.Ifthereisamedianorrefugeislandtheuseofcurbtopplacement and use of a delineator at the lane line covers all installations with devices thatshouldprovideatroublefreeinstallationformanyyears.

MostoftheapplicationstudiedinanearlierMDOTstudyexaminedtheefficacyoftheGatewaytreatmentatcrosswalksonarterialorcollectorroadsandatuncontrolledcrosswalkslocatedat

the intersectionwith a stop controlledminor road. Amap of locations used for this study isprovidedinChapter4.ThisresearchalsoevaluatedtheGatewaytreatmentinavarietyofnewcrosswalk applications, including traffic circles, roundabouts; interstate entrance ramps, and

trails crossings. The results of these studies showed: theGateway treatmentwasmoderatelyeffectiveatthetrafficcircleandtworoundaboutlocations,particularlyatcrosswalksexitingtheroundabout.Theapplicationat interstateentrancerampsonlyproducedmarginal increases in

yieldingbehavior.TheGatewayisthereforenotrecommendedforthistypeofapplication.TheGatewaywashighlyeffectiveatoneofthetwotrailcrossingsbutonlymoderatelyeffectiveatthe second crossing that had a higher operating speed. Another study replicated an earlier

findingthattheGatewaytreatmentcanincreaseyieldingatanRRFBsitetoveryhighlevels.Thefinal study in this series found that treating only one crosswalk leg at an intersectionwith aminorroadcouldimproveyieldingattheuntreatedcrosswalkleg.

Permanent installation of these sites provided preliminary evidence of the long-termpersistenceofincreaseddriveryieldingright-of-waytopedestriansatGatewaylocationsandthe

long-term survival of the Gateway treatment. Results only demonstrated that the signsmaintained their effectiveness in Ann Arbor over two months and at the sites in thesouthwesternsideofthestateforthree-monthsbecausethesesignsneededtoberemovedfor

winter. Because the signs were only installed for two months in Ann Arbor and only threemonths at the sites in southwest Michigan, supplemental data will be required in order toprovide clear evidence that the effects persist over time. Preliminary data on sign survival

indicate that Gateway signsmounted in the roadway on installable curb basesmay bemorerobust than signs installed flush with the roadway and that the robust flexible delineator

installedona lane linecansustainmanyhitsbutmaynotsurviveforanentireseasonatsiteswithhigherspeedsandhighADT.

ThefinalchapterprovidedinformationonmaterialsandapproximatesinstallationcostsforeachoftheelementsoftheGatewaytreatment. Thesedatashowthat it isarelatively inexpensivetreatment. Most items can be removed easily in winter and reinstalled easily in the spring.

Estimated removal and reinstallation costswere also provided. The final chapter provides anoverviewoftheconclusionsreachedbytheresearchteamaspartofeachtaskofthemulti-yearstudyandcostdataforvariousGatewaypedestriancrossingconfigurations.


INTRODUCTIONIn2013, theMichiganDepartmentofTransportation (MDOT) initiatedamulti-year studywithWesternMichiganUniversity(WMU)inorderto:1.Evaluatefactorsrelatedtotheefficacyofa

gatewaytreatmentusingR1-6signs(gatewaytreatment),2.Determinethelong-termeffectsofpermanent installations,and3.Examineconfigurations that contribute to theeffectivenessofthetreatment.

As MDOT would like to increase its focus on reducing the number of pedestrian crashes inMichigan as part of the Toward Zero Deaths statewide safety campaign, the WMU/T.Y. Lin

International team (hereafter referred to as the “WMU team”) proposed the followingobjectives:

1. Determinetheeffectivenessofdrivercompliancewithgatewaytreatmentincomparisontotherectangularrapidflashbeacon(RRFB)andthepedestrianhybridbeacon(PHB).

2. Determinewhereandwhenthegatewaytreatmentshouldbeusedandthemosteffective,configurationsoftheR1-6signs.

3. Determinethecostbenefitsofusingthegatewaytreatmentincludingoperation,maintenanceandreplacementcostswiththeRRFBandPHB.

Inordertoaddresstheseobjectives,theWMUteamconductedanumberofactivities;eachiscaptured as an individual chapter of this report. The following provides an outline of the

individualchaptersandtherebytheactionstakenaspartofthismulti-yearstudy:Chapter1–ConfigurationsInfluencingtheEfficacyoftheGatewayTreatment.

Chapter2–EvaluationofConfigurationsthatImprovetheGatewayTreatmentSurvival.Chapter3–EvaluationoftheGatewayTreatmentConfigurationinVariousApplications.Chapter4–PotentialRoadwayFactorsInfluencingtheEfficacyoftheGatewayTreatment.

Chapter5–LongTermEfficacyandSurvivaloftheGatewayTreatment.Chapter6–CostsAssociatedwithInstallationandMaintenanceoftheGatewayTreatment.Chapter7–Conclusions.


LITERATUREREVIEWTraditionalApplicationsoftheIn-StreetSignAlthoughpedestriancrashesaccountforonly1percentofreportedmotorvehiclecrashesinthe

UnitedStates,theyaccountfor14percentoffatalcrashes(1).Zegeeretal.(2)comparedcrashesat1,000markedand1,000matchedunmarkedcrosswalksin30U.S.cities.Thestudyfoundnosignificantdifferenceincrashesbetweenmarkedandunmarkedcrosswalksatsites

withonelaneineachdirection,buthighercrashratesatmarkedcrosswalksonmultilaneroadswithanuncontrolledapproachwhentheroadhadaverageannualdailytraffic(AADT)above12,000withoutaraisedmedian,andabove15,000witharaisedmedian.Theyalsoobserveda

higherincidenceofmultiplethreatcrashesatthesesites.Thesedatashowtheneedforlowcostcountermeasurestoincreaseyieldingtopedestriansatcrosswalksonmultilaneroadswithmoderatetohighlevelsofaveragedailytraffic(ADT).Currenttreatmentsincludethe

rectangularrapidflashingbeacon(RRFB),whichcostsaround$20,000perinstallationandthepedestrianhybridbeacon(PHB)thatcosts$100,000perinstallation.Boththesetreatmentsshouldbeusedwithadvanceyieldorstopmarkingstoencouragedriverstoyieldfurtherfrom

thecrosswalk.Whenthesetreatmentsareusedtogether,theRRFBorPHBincreasesyieldingwhiletheadvancestopandyieldmarkingsprimaryinfluencesafetybyreducingtheprobabilityofamultiplethreatcrash.Itshouldbenotedthatthecostsassociatedwiththeinstallationof

theRRFBandPHBtreatmentslimittheirdeployment.Onewaytoimprovethesafetyatpedestriancrossingsistheuseofthein-streetsign.Thissignis

installedintheroadwayandremindsdriversthatitisthelawtoyieldtopedestrianswithincrosswalks.Oneadvantageofthisdeviceisitrequiresnoactionfromthepedestriantoactivatethedeviceanditisthereforeactiveforeverycrossing.Manystudieshavedocumentedthat

placing“YieldtoPedestrianSigns”intheroadwaycanincreasethepercentageofmotoristsyieldingtopedestrians(3,4)Huang,Zegeer,andNassi(3)evaluatedtheeffectsofin-street“STATELAW:YIELDTOPEDESTRIANSINYOURHALFOFROAD1”signsplacedinthemiddleofthe

crosswalkondriveryieldingbehaviorat7locations.Yieldingincreasedfrom70%beforethesignswereinstalledto81%afterthesignswereinstalled.Thein-streetsignproducedlargereffectsthananoverheadcrosswalksignorapedestrianregulatorysigninthisstudy.The

authorsalsonotedthattherewerereportsthatsomemotoristsranoverthein-streetsignsintentionally.Althoughin-streetsignsaredesignedtorecoverwhenstruckbyvehicles,repeatedstrikes,orhigh-speedstrikescanpermanentlydamagethesesigns.

In-streetsignswerealsoevaluated(4)inastudyjointlyfundedbytheTransitCooperativeResearchProgramandtheNationalCooperativeHighwayResearchProgramthatcompared

severaltreatmentstoimprovemotoristyieldingtopedestriansatunsignalizedintersections.Theresearchteamcollecteddataonmotoristyieldingbehaviorat42crosswalksindifferent

1 It should be noted that Michigan does not have a state law on yielding to pedestrians incrosswalks, instead local ordinances or local adoption of the Uniform Traffic Code need toaddressROWinMichigan(otherthanatsignalizedintersections).


regionsoftheUnitedStates.Theresultsindicatedthatin-streetsignswereassociatedwith

yieldingratesof87%ontwolaneroads.In-streetsignsweresuperiortoyellowoverheadflashingbeacons,pedestriancrossingflags,andin-roadwaywarninglights.Onlyaredsignalorredbeacondevicesproducedhigheryieldingbehaviorthanin-streetsigns.Theresultsofthe

studyshowedin-streetsignstobehighlycosteffectiveinincreasingyieldingbehavioratcrosswalksthattraverseroadswithonelaneineachdirection.Dataalsoshowedthatin-streetpedestriansignsperformedequallywellonroadswith25and30mphspeedlimits.Theywere

notevaluatedonroadswithhigherspeedlimits.Theauthorsalsoconcludedthatthesesignswerenoteffectiveonmulti-laneroadsandonlyrecommendedtreatmentswitharedindicationatthesesites.

Onestudyexaminedtheeffectofplacingthesesignsonthecenterlineatthecrosswalkline,20feetinadvanceofthecrosswalkline,and40feetinadvanceofthecrosswalklineatthree

crosswalksondriveryieldingbehaviorontwo-laneroads(5).Atoneofthethreelocationsplacingthesignatthecrosswalklineorinstallingthreesignswassignificantlymoreeffectivethaninstallingthesignat20or40ftinadvanceofthecrosswalk.Atanothersiteinstallingthe

signatthecrosswalklineorinstallingallthreesignswassignificantlymoreeffectivethaninstallingthesign40ftinadvanceofthecrosswalk.Overallitappearedthatinstallingthesignsatthecrosswalklinewasaseffectiveormoreeffectivethaninstallingthesigninadvanceofthe

crosswalkorinstallingthesignsatallthreedistancesfromthecrosswalk.Atoneofthethreesites,pedestrianswereoftentrappedinthemiddleoftheroadduringthebaselinecondition;thein-streetsignwashighlyeffectiveinreducingthenumberoftrappedpedestriansregardless

ofwhereitwasplaced.Itshouldbeexpectedthatlargeincreasesinyieldingshouldbeassociatedwithreductionsinthepercentageofpedestrianstrappedinthemiddleoftheroadwaywhereverthisproblemexits.Thesedatasuggestthatthein-streetsignsaremostlikely

effectivebecausetheplacementinthestreetisparticularlysalienttodrivers.Anotherstudy(6)replicatedtheVanHouten,et.al.(5)studybycomparingplacingsignsatthe

crosswalkline,withplacingthem30ft,60ft,90ftand120ftinadvanceofthecrosswalklineon8two-laneroads.Theyfoundthatthein-streetsignplacedatthecrosswalkwasmoreeffectivethansignsplacedinadvanceofthecrosswalk.Theyalsoexaminedwhethertheinstallationsof

anin-streetsigninfluencedvehiclespeedandfoundthein-streetsignssignificantlylowervehiclespeedsatthecrosswalkinbothdirectionsat7ofthe8sites,andinonedirectionatthe

remainingsite.Theseresultsarepromising,howeverallsiteshadverylowpostedspeedlimits(25mphat5sitesand20mphatthe3remainingsites).Kannel,Souleyrette,&Tenges(7)alsoreportedasmallreductioninspeedatasingleconventionalin-streetsignlocationpostedat30

mph.EvaluationofaGatewayInstallationonMultilaneRoads

Onelimitationofthein-streetsignisitsfailuretoproducehighlevelsofyieldingonmultilaneroads.Shurbutt,VanHouten,Turner,andHuitema(8)documentedhowtheefficacyofanotherpedestriancrosswalkcountermeasure,theRRFB,couldbeincreasedbyintroducingathird


deviceonthemedianislandproducingatypeof“gateway”visualeffectforapproachingdrivers.

BennettandVanHouten(9)usedagatewayconfigurationofthein-streetsign(theuseofasignonthelanelineandtworoadwayedgesigns-threesignsforeachtwolaneapproachtothecrosswalk)producedamarkedimprovementinyieldingatmultilaneuncontrolledcrosswalks

thatwascomparabletothoseproducedbyanRRFBorPHB(10,11).Onereasonthegatewayin-streetsignconfigurationwassoeffectivemayhavebeentheperceivednarrowingoftheroadwayproducedbyaddingsignsonbothsidesoftheroadoutsidethelaneseventhoughthe

widthofthetravelwayitselfwasnotactuallynarrowed.Itisalsolikelythatthreesignsweremorevisiblethanonesign,particularlyifvehiclesaheadofamotoristapproachingthecrossingblockedthemotorist’sviewofthelocationofthesinglesign.

Itisalsopossiblethatthegatewayeffectacquiredcapturesdriverattentionindependentofaperceivednarrowingeffect.Thisrationaleissupportedbyresearchshowingthegateway

configurationoftheRRFBalsoproducedamarkedeffecteventhoughbeaconswerenotintheplacedintheroadway.In-streetsignsmayalsoprovidebetterdelineationoftheedgeoftheroadwaywhentheyarepresent.Specifically,theboundariesoftheroadareextendedvertically

viathesigns.Adrivermayignoretheboundariesoftheroadwhiledriving.However,whenthevisualboundariesaremademoresalient,itmaycausedriverstoslowdown.Thiseffectmayoccurbecauseitreducesthenarrowingofperspectivethatoccurswhensomeoneisdriving

makingitmorelikelythedriverwillattendtoapedestrianstartingtocrossattheedgeoftheroad.

BecausethedatareportedbyBennettandVanHouten(9)werecollectedovermonths,theyalsocapturedtheeffectivenessoftheinterventionovertime.However,itisnotknownwhethertheseresultswillbesustainedoverlongerperiodsoftime.Researchisneededtoaddressthis

questionaswellasdataonhowthein-streetsignperformsatnight.InstallationoftheIn-StreetSignatRRFBandPHBSites

DatacollectedbyDr.VanHoutenforMDOT(12)alsorevealedthattheRRFBandPHBproducedpooreryieldingresultsinMichiganthandatacollectedinalarge-scaleFHWAevaluation.DatacollectedinareviewconductedbyDr.VanHoutenalsoshowedthatthisproblemwasnot

uniquetoMichigansiteswithsimilarresultsreportedinKansasandPortland,Oregon(13,14).Itislikelythepoorerresultsareafunctionofthesmallernumberofunitsinstalledinsometest

communitiesandtherelativelylowerlevelofoutreacheffortsatsitesthatwerenotpartoftheFHWAresearchproject.DatafromtheMichigansitesseemtoconfirmthisfinding.DatacollectedbyintheMDOTstudyalsosuggestthattheuseofthein-streetsignatnewRRFBand

PHBsitesmaybearelativelylowcostalternativetolarge-scaleoutreachefforts.BennettandVanHouten(9)foundthatasinglein-streetsignatRRFBandPHBsitesinMichiganproducedhigherlevelsofyieldingthanRRFBorPHBtreatmentsusedalone.

Itisimportantthatadditionalstudiesexaminetheinteractionofthein-streetsignwiththeRRFBandPHBinordertodeterminewhyitiseffect.Onequestionthatneedstobeaddressedis


whetherdriverslearntorespondtobettertotheRRFBandPHBinstallationswithoutthein-

streetsignafterithasbeenpairedforaperiodofmonthswiththein-streetsign.

TheRoleofAdvanceStoporYieldSignandMarkingsWheneverdriversyieldrightofwaytopedestriansonmultilaneroadwaysandstopsclosetothecrosswalkline,theycanscreentheviewofpedestriansfromvehiclesapproachinginadjacent

lanes,increasingtheprobabilityofmultiplethreatcrash.Aseriesofexperiments(15,16,17,18,19)hasdemonstratedthatthecombinationofadvanceyield/stopmarkingsalongwith“Yield/StopHereforPedestrian”signsinadvanceofthecrosswalkreduceddriver/pedestrian

evasiveconflictsby67%to87%andproducedalargeincreaseinthedistancemotoristsyieldedinadvanceofthecrosswalk.Thesedatastronglysupportalwaysusingadvancemarkingsatuncontrolledcrosswalksitesassociatedwithincreasedriskofmultiplethreatcrashes.These

sitesincludethosewithADTabove12,000withoutaraisedmedianand15,000witharaisedmedian.

QuestionstobeAddressedbyFurtherResearchTheresultsofthisresearchwillassistMDOTindetermininghowtomaximizepedestriansafetybenefitswithlimitedfinancialresources.Evaluatingtheimpactofthein-streetsigngateway

treatmentinavarietyofdifferentcrossingapplicationscouldhelpidentifypotentialapplicationsofthistreatmentoption.Itisalsoimportanttodeterminehowmuchthevisualnarrowingeffectofthegatewaytreatmentcontributestotheeffectandwhethersimilareffectscouldbe

obtainedbyusingdelineatorswithoutthein-streetsignmessage.Anotherissuethatrequiresstudyishowtoextendtheusefullifeofin-streetsigns.Onewaytoreducedamagetothesesignsistoplacethematroadwaylocationswheretheyarelesslikelytobehit.Therearea

numberofplacementstrategiesthatcouldextendthelifeofthesesigns,whichshouldbeevaluated.Additionalresearchshoulddeterminetheconditionswherethein-streetsigngatewaytreatmentcanbesubstitutedformoreexpensiveRRFBandPHBtreatments.

Newtrendsintrafficengineeringresearchhavefocusedondeterminingtheconditionsunderwhichatreatmentiseffective,andhowtooptimizethetreatment.Forexampleintworecently

fundedFederalHighwayAdministration(FHWA)studies,EllisandVanHouten(20)documentedhowmatchingcountermeasurestocrashtypesinMiamicouldreducepedestriancrashesinhigh

crashzonesby50%,andShurbutt,VanHouten,Turner,andHuitema(8)documentedhowtheefficacyoftheRRFBcouldbeincreasedbyfinetuninghowitisinstalledandhowitisoperated.ArecentlypublishedNationalHighwaySafetyAdministration(NHTSA)fundedstudy(21)has

alsodemonstratedhowthein-streetsigncouldbeusedaspartofanpedestrianright-of-wayenforcementprogramtoproduceacultureshiftinyieldingright-of-waytopedestriansonacitywidebasis.Ithasbeenlongknowthatresultsmayvaryfollowingtheimplementationofsignage

andtrafficcontroldevices.Inordertomaximizeresultspractitionersneedtobetterunderstandwhere,andhowtoimplementthesedevices.Itisalsoimportanttodeterminewhatvariablesinfluencingthelifespanorsurvivalofthesesignssincethisimpactsreplacementand


maintenancecosts.Researchonthein-streetsignshouldaddresstheseissuesaswellas

examinethelifespanofthesesignsusingdifferentstrategiestodevelopguidelinestobestinsureinstallationsarepracticalaswellascosteffective


SURVEYONUSEOFR1-6SIGNA survey was sent out on our behalf from FHWA to states participating in the pooled fundresearchprogram.Thesurveyaskedrespondentstoidentifythestate,anestimateofhowmany

in-street pedestrian crossing sign (R1-6, or R1-6a) have been installed in their state,whethertheywereaware ifanyonehadperformedacrashanalysisof theeffectivenessof thesesigns,andifacrashanalysishadbeenperformed,contactinformationtolearnmoreabouttheresults

of the study. We received feedback from the following 11 states: Florida, Iowa, Kansas,Minnesota, Missouri, Nebraska, New Hampshire, North Carolina, Oregon, Pennsylvania, andSouthCarolina.Noneofthestatepoolfundcoordinatorswereawareofanyresearchonthese

signs. With the exception of Pennsylvania, Minnesota, New Hampshire, and North Carolina,stateofficialsstatedtheyonlyreportedthenumberinstalledonstateroadsanddidnotattempttoestimatethenumberofsignsinstalledbycitiesandmunicipalities.Therefore,thislistshould

be considered to be a conservative estimate because inmost cases they only included thoseinstalledonstateroads. Pennsylvaniamentionedthattheypaidforall in-streetsignsinstalledby municipalities in the state since 2001, and therefore, the number they provided can be

viewed as accurate. New Hampshire mentioned that their number included in-street signspurchasedformunicipalitiesbutthattheyhadnoknowledgeofadditionalsignsmunicipalitiesmay have directly purchased and installed. Table A-1 shows the results of the pooled fund

survey.ThestateofPennsylvaniahadthelargestnumberinstreetsigns(7500).Itisnotknownwhetheranyofthesesignsareinstalledonmultilaneroads,andthereisnoindicationwhetheranyofthesesignsareinstalledinagatewayconfiguration.

TableA-1.Thenumberofin-streetsignsitesreportedby11poolfundstates.

State NumberofInstallations

Florida 24

Iowa 15

Kansas Afew

Minnesota 500

Missouri 0

Nebraska 5

NewHampshire 36

NorthCarolina 400

Oregon 0

Pennsylvania 7,500

SouthCarolina 0

Additionaldataonin-streetsigninstallationswerecollectedfromavarietyofcitiesaspartofaNCHRP study that is examining crash modification factors for various pedestrian safetycountermeasures. In-streetsignswerenotselectedasacountermeasurebecause thesesigns

were rarely reported be installed on roads with higher crash rates, such as crosswalks atmultilanelocations.Thenumberofin-streetsignsitesforeachcityisshowninTableA-2.Thesedatashowthattheuseofthistreatmentvariesconsiderablefromcitytocity.Atlantareported


having the most crosswalk sites with a total of 300, withWashington, D.C. reporting having

several 100 sites. The cityofArlington said theywill only replace in-street signs2or3 times.Atlantasaidtheyusedthemonalltypesofroads.ThecityofPittsburgsaidtheytaketheresignsdownatnightandthecityofCambridgesaidtheyremovethesignsduringthewintermonths.It

is clear thateach jurisdictionhasestablished theirownprocedures for theuseof thesesigns.Onlyonecity,Gainesville,FL,reportedusingagatewaytreatmentandtheyweresatisfiedwiththe performance of the gateway in-street sign installations on multilane roads. Gainesville

installed gateway in-street signs after learning of the data reported by MDOT. Subsequentinformation showed use of the gateway treatment in San Antonio, Texas and Palm Springs,California.

TableA-2.Thenumberofin-streetsignsitesreportedin24cities.

City State NumberofSites Notes

Phoenix AZ None

Scottsdale AZ 2 “YieldtoPedestrians”inDowntownArea

Tucson AZ 0

LaMesa CA 6

LosAngeles CA Many

SanFrancisco CA Afew Notsurehowmanyremain

SantaMonica CA Afew

Boulder CO 6-12 Text on in-road sign is “State Law - Yield toPedestrians”

Washington DC 100’s

Gainesville FL 6 Workwell,twogatewayinstallations

St.Petersburg FL None Nonein-road;mountedbehindcurbatstopbar

Atlanta GA 300 Alltypesofroads

Chicago IL 39 Installedin2011-2012,“StopforPedestrians”

Cambridge MA 12 Removedduringsnowmonths

Columbia MO ~25

Springfield MO 15-20

Charlotte NC 20-25 Mostwithrefugeisland

NewYork NY 0 They’redoingsignreduction;veryfewofthese

Portland OR Afew

Pittsburgh PA 20 Takethemdownatnight

Alexandria VA ~50

Arlington VA Some Don’treplaceafter2-3replacements

Kirkland WA 0

Milwaukee WI 40-50


CHAPTER 1 – CONFIGURATIONS INFLUENCING THE EFFICACY OF THEGATEWAYTREATMENT

INTRODUCTION

Pedestrian fatalities steadily decreased nationally from 5,801 pedestrian fatalities in 1991 to

their record low of 4,108 in 2009. Pedestrian fatalities began to rise in 2010 to 4,302 andcontinued to rise until 2012 when there were 4,818 pedestrian fatalities. 2013 saw a slightdecreaseof2%inpedestrianfatalitieswithatotalof4,735(22).

AttemptstoimprovepedestriansafetydatebacktoancientRome.The2,000-year-oldruinsofPompeiicontainedraisedstonesforpedestrianstousetocrossroads.Theseraisedcrosswalks

had gaps for the wheels of the cart to pass through. This design served as a form of trafficcalmingbecause thecartdriverneeded to slow inorder toalign thecartwheelswithgaps inraised crossways. Such a configuration couldbe thoughtof as a typeof gateway through the

crosswalk.Bennett,ManalandVanHouten(9)placed in-streetsignsoneachsideofamulti-laneroadat

uncontrolledcrosswalks,on the lane linesandon thecenter line.Thisnovel interventionwasreferred to as the gateway treatment. A diagram showing the placement of signs for thegateway treatment is shown in Figure 1-1. Additional configurations for varying roadway and

crosswalkcharacteristicsareprovidedintheUserGuide.

The 2009 version of the MUTCD (23) states that In-Street Pedestrian Crossing sign shall beplacedintheroadwayatthecrosswalklocationonthecenterline,onalaneline,oronamedianisland.Itdoesnotpermitplacementonasignpostatthesideoftheroadorontopofthecurb.

ThereforeFHWApermissiontoexperimentisrequiredtoplacethemontopofthecurbattheoutsideedgesoftheroadbutnotonthecurbofarefugeormedianisland.


Figure1-1:DiagramofTypicalSignPlacementforGatewayTreatment

Thistreatmentusessixsignsforafour-laneroaddividedbyamedianorrefugeisland,andfivesigns for a crosswalk without a median or refuge island. The gateway treatment produced

markedincreasesinthepercentageofdriversyieldingatavarietyofcrosswalklocationsacrossMichigan.Yieldingratesincreasedfromabaselineaverageoflessthan25%to79%atonesiteand from23%to82%atanother site.Prior researchhaddemonstrated thata single in-street

sign installed on the centerline was effective on two lane roads with a travel lane in eachdirection but were relatively ineffective on streets with two or more travel lanes in eachdirection(4).

Thegatewaytreatmentalsoproducedhighyieldingresultswhenusedinconjunctionwithtwoprovenalternativeinterventions,therectangularrapidflashbeacon(RRFB),andthepedestrian

hybrid beacon (PHB)(9). The price of the gateway in-street sign treatment is relatively low at(approximately$200to$300persignandbase)andthusmaybemorecosteffectivethantheRRFB (estimated cost of $20,000) or the PHB (estimated cost of $100,000). However, little is

knownabouthowoftenthesignsmayneedtobereplaced.TheBennett,ManalandVanHoutenexperimentdidnottesttheeffectsofnarrowingtheroadusingthesigns,theeffectofthesignmessage,ortheeffectsofthesignsatdifferenttypesofpedestriancrosswalks.

The gateway treatment is both a traffic control device aswell as a geometric designelementbecause it involves the perceived narrowing of the roadway at the crosswalk. The gateway

treatmentcanbeviewedasa trafficcalmingdevicebyvisuallynarrowing the travelpathofadriver, and thereby inducing the driver to slow down when approaching the gap. Researchassistants all noticed an increase in driving scanning for pedestrians when the gateway was

IN-STREET PEDESTRIAN CROSSING SIGN EDGE ALIGNED WITH PARKING LANE

FLEXIBLE DELINEATOR

IN-STREET PEDESTRIAN CROSSING SIGN PLACED IN GUTTER PAN


present.Futureresearchshoulddirectlymeasuredriver-scanningbehavior.Thesignsmayalso

bemore visible to drivers because of their position on the sides and center of the roadway.Pedestriansareoftenpositionednearthe locationof thesidesigns inthegatewaytreatment.Theextentthatthesignfunctionsasatrafficcalmingdevicewascomparedwiththeeffectsof

thesignsasapromptfordriveryieldingbehaviorbymanipulatingavarietyofsignfeaturessuchasthepresenceorabsenceofthesignmessageandthenarrownessofthegap.

This chapter examines the effects of the gatewaywithdifferent configurations tounderstandhow they impact driver yielding to pedestrians in crosswalks. The configurations includevariationsingatewaywidthandsignmessage,andpartialgatewayconfigurations.Thegateway

effectonreducingvehiclespeedisdiscussed,aswell.

METHODOLOGY

DependentVariablesThenumberofmotoristswhodidanddidnotyieldtopedestriansincrosswalkswasmeasured.Driver yielding was measured in reference to an objective dilemma zone (a location beyond

which a driver can easily yield if a pedestrian enters the crosswalk). A formula used todeterminewhetheradrivercouldhavesafelystoppedatatrafficsignal,wasusedtodeterminewhetheradrivercouldhavestoppedforapedestrianstandingwithone foot in thecrosswalk

(24). This formula takes into account driver reaction time, safe deceleration rate, the postedspeed, and the grade of the road to calculate this interval for the yellow traffic light. Thisformulawasusedtodeterminethedistancetothedilemmazoneboundarybymultiplyingthe

time(y)bythepostedspeedlimitinfeetpersecond:

v y = t +

2a + 2Gg

wheret=theperceptionandreactiontimeinseconds(S);v=thespeedofapproachingvehiclesinfeetpersecond(thepostedspeedwasusedforapproachspeed);a=thedecelerationrate,recommendedat10 feet/S2;G=accelerationduetogravity (32 feet/S2);andg=thegradeof

theapproach.Toaidobserversinidentifyingthedilemmazone,thezonewasmarkedbyeitherasprinklerflag locatedadjacenttothecurborwithbrighttapethatextendedfromtheraisedconcreteofthecurbfaceintothegutterpan.

Motorists who had not passed the outer boundary of the dilemma zone when a pedestrianenteredthecrosswalkwerescoredasyieldingornotyieldingbecausetheyhadsufficienttime

andspacetostopsafelyforthepedestrian.Motoristswhoenteredthedilemmazonebeforethepedestrianplacedafootinthecrosswalkcouldbescoredasyielding,butcouldnotbescoredasfailing to yield because the motorist did not have adequate distance to yield based on the

calculated distance. However, the signal timing formula is relatively lenient; hence, many


vehiclesthatpassedthedilemmazonecouldyieldsafely,particularlythosetravelingbelowthe

speedlimit.ResearchDesign

A replication logic reversal designwasused in theseexperiments. In replication logic reversaldesign,atreatmentisintroduced,removed,andreintroducedtoisolatetheeffectivenessofthetreatment on driver behavior independent of other environmental factors. In all preliminary

studiesinChapters1through4temporarysignswereinstalledthatcouldbeeasilymovedandremoved by the research team. Evaluations in Chapter 5 evaluated more permanentinstallationsthatwouldonlyberemovedforwinter.

A trial, or staged crossing, beganwhen a researcher demonstrated an intention to cross thestreetbyplacingonefootwithinthecrosswalkwithhisorherheadturned inthedirectionof

theapproachingvehicle.Aresearchassistantrecordedtheresultsofthetrialonaclipboard.Eachsessionconsistedof20trials(pedestriancrossings).Thepercentageofdriverswhoyielded

the right-of-way to pedestrians was calculated for each session by dividing the number ofdriversthatyieldedtheright-of-waybythenumberofyieldingdriversandnon-yieldingdrivers.Datawerecollectedduringdaylighthoursbetween10:00a.m.and8:00p.m.Mondaythrough

SaturdayinMaythroughNovember.Datawerenotcollectedwhenitwasraining.Scoring

Drivers in the first two travel lanes nearest the pedestrianwere scored for yielding after thepedestrian had entered the crosswalk. This procedure was used because it conforms to theobligations of motorists specified in the Universal Vehicle Code and local ordinances in

Kalamazoo, and Ann Arbor regarding who has the right-of-way at what time. Drivers in thesecondhalfoftheroadwaywerescoredasaseparatetrial iftherewasapedestrianrefugeormedianislandseparatingthetravelway.Iftherewasnoisland,driversinthesecondhalfofthe

roadwerescoredwhenthepedestrianapproachedthecenterofthelasttravellaneadjacenttotheyellowcenterlineseparatingopposinglanesoftraffic.Motoristswerethenscoredusingthesametrialmethodasthecrossingforthefirsthalfoftheroadway.

DataCollectorTrainingProcedure

Researchersweretrainedtousetheoperationaldefinitionofyieldingbehavior.Theypracticedrecordingtogetheruntiltheyobtainedinter-observeragreement(seebelowforadescriptionofinter-observer agreement) of 90% or better for two consecutive sessions (a total of 40

observations). Researcherswere also trained on how to use awalkingwheel tomeasure thedistancetothedilemmazone,andhowtoinstalltheflagsorlaythetape.

DataCollectionSetupTheresearcherssetupthedilemmazonebeforebeginningtrials.Awalkingwheelwasusedtomeasurethedistancefromthenearestcrosswalklinetothedilemmazone.Duringthemarking


process,oneoftheresearchersservedasaspottertoensurethatthepersonusingthewalking

wheelwasclearoftraffic.Bothpersonsworeorangevestsduringthemarkingprocesstomakethemmorevisibletodrivers.Theresearchersthenmarkedthelocationwiththenecessaryflags,tape,orboth.

Inter-observerAgreementInter-observeragreementwascalculatedforatleast34%ofallobservationsinallexperiments,

and data were collected during each condition of each experiment in order to validate theobservationaldata.Eacheventthatwasscoredthesamebybothobserverswascountedasanagreement, and each event that was scored differently by each observer was scored as a

disagreement.Inter-observeragreementwascalculatedbydividingthenumberofagreementsduringeachsessionbythesumofagreementsplusdisagreementsforthatsession.Theresultofthiscalculationwasthenconvertedtoapercentage,asexpressedbytheformulabelow:

Agreements per session Inter-observer agreement (%) = (Agreements per session+ Disagreements per session)

During sessions in which agreement data were collected, the two observers stood several

metersapart at a locationwithanunobstructedviewof the crosswalk.Whenmore thanonepedestriancrossedataparticularcrosswalk,theprimaryobserveridentifiedthepedestrianforwhomyieldingbehaviorwastobescoredbydescribingadistinctivefeaturesuchaswhetherthe

person was a male or female or the color of his or her clothing. They then independentlyrecordedmotoristyieldingbehavioranddidnotdiscusswitheachotherhowtheyscoredanyofthetrials.Thisprocedurecontrolledforpotentialobserverbias.

Inter-observer agreement on the percentage of drivers yielding to pedestrians averaged 96%overallofthestudiescompletedinthisresearchwitharangeof88%to100%.

ApparatusThreetypesoftrafficcontroldevices(TCD)swereusedinthisstudy.ThefirstTCDdevicewasthe

R1-6in-streetsign.ThesecondTCDwasablanksignoftheexactsamesizeastheR1-6signandusing a similar diamond grade fluorescent yellow green background color. This was done todetermine whether the language and symbols on the sign influenced the yielding rate of

motorists. The third TCD was a robust flexible delineator that is designed to withstand 100impactsat60mph.ThisdevicewasthesamecolorasthebackgroundoftheR1-6sign. Itwasexamined in more vulnerable gateway treatment locations to address concerns about

survivabilityoftheintervention.

Figure1-2showsaR1-6signmountedonacurbdeviceaffixedtotheroadway.Figure1-3showsablank signmountedona removablebase,andFigure1-4 showsa flexibledelineator that isscrewedintoabasecementedintotheroadway.


Figure1-2:R1-6Signmountedonacurbaffixedtotheroadway

Figure1-3:AblanksignthesamesizeastheR1-6Signmountedtoaremovablebase


Figure1-4:Aflexibledelineatorattachedtoabasecementedintothepavement

EVALUATIONOFTHEIMPORTANCEOFGATEWAYWIDTH

The first study in this seriesexamined theeffectsof gatewaywidthonyieldingbehavior. Thehypothesis was that the narrowing of the gateway was a factor associated with yielding

behavior.Thisstudyteststhishypothesis.Locations

ThecomparisonofgatewaywidthsmeasuredthelanewidthfromtheinsideedgeofoneR1-6signtotheinsideedgeofthecorrespondingR1-6signontheothersideofthetravellane.TheeffectofgatewaywidthwasmeasuredattwositesattheMarshalltrafficcircle,onecrosswalk

was on the NE side of the traffic circle and the other crosswalk was on the SW side of thecrosswalk. Itwasalsomeasuredattwotrailcrossings inPortageMi.OnetrailcrossingwasonOaklandStreetsouthofMilhamRoad,andsecondthetrailcrossingwasonGardenLanewestof

Kingston Dr. The freeway crosswalk siteswere at the two uncontrolled I-94 entrance rampsfromSouthWestnedgeAvenue.

ResearchDesignAreplicationlogicreversaldesignwasemployedinthisstudy.Followingmultipledailybaselinemeasures (sessions duringwhich no in-street sign treatmentwas present) the treatmentwas


introducedat onewidth formultipledays.Next, thewidthof the gatewaywas changed, and

observation data was collected again for multiple days. These changes were replicated byrepeatingmeasuresforeachofthegatewaywidthsseveraltimes.

Table1-1showsyieldingbehaviorforvariousgatewaywidths.Inspectionofthesedatarevealsthat in each case, narrower configurations lead to a higher driver yielding rates than widerconfigurations. The gap size (the difference between the wide and narrow width) varied

between2.4 feet to 9.4 feet across sites. Although thesedata show that gatewaywidth is afactor influencing yielding, it is clear that other variables also have an influence on driveryielding. This becomes apparent when one ranks sites by gateway treatment width, which

reveals that some narrow locations provide better yielding rates than other sites with evennarrowerconfigurations.Forexample,yieldingishigherattheMarshalltrafficcirclesitethanatthe Oakland trail crossing and I-94 entrance ramp even though the latter two sites are

considerablynarrowerthantheMarshalltrafficcircle.TheconsistencyofthesedatacanbeseenfrombyviewingdatacollectedovertimeattheMarshallNWleginFigure1-5.

Table1-1.Thepercentageofdriversyieldingtopedestrianswithawidevs.anarrowgateway.Thelastcolumnshowsthechangeinwidthbetweenthewideandnarrowgateway

WideConfiguration NarrowConfiguration Location Width Yielding Width Yielding GapWidth

Reduction Baseline Gateway Baseline Gateway MarshallTrafficCircleCrossingSEleg

44.8ft. 11% 15% 36.8ft. 11% 29% 8ft.

MarshallTrafficCircleCrossingNWleg

30.3ft. 13% 19% 22.3ft. 13% 48% 8ft.

OaklandSt.TrailCrossingSofMilham

17.5ft. 2.7% 10% 11.7ft. 2.7% 39% 5.8ft.

GardenLn.TrailCrossingE.ofS.Westnedge

19.5ft. 21% 67% 17.1ft. 21% 75% 2.4ft.

I-94RampE.BoundRampEntrance

20.0ft. 2% 17% 10.6ft. 2% 31% 9.4ft.

.


Figure1-5.DriveryieldingattheMarshallTrafficCircleNWcrosswalkleg

EVALUATIONOFTHEIMPORTANCEOFTHESIGNMESSAGEIn order to determine whether the message on the signs in the roadway controlled driverbehavior, the gateway configuration with sign blanks with the reflective background but no

messagewerecomparedtothestandardR1-6signattwodifferentcrosswalklocations.ThefirstlocationwasatacrosswalkataT-intersectionthattraversedamulti-laneroadwithtwotravellanesineachdirectionandon-streetparking.Thesecondlocationwasatamidblockcrosswalk

withtwotravellanesineachdirection,apedestrianrefugeislandandnoparking.GatewayConfigurations

Two types of gateway configurations were used in this study. The first configuration was agatewaywithallR1-6 in-streetsigns.Thesecondtypeofconfigurationwasagatewaywithallblanksignsoftheexactsamesize,signsheetingbackgroundcolor(diamondgradefluorescent

yellow green), and shape as the R1-6 in-street sign to determine if themessage on the signinfluencedtheyieldingrateofmotorists.Figure1-6showsaphotogatewayconfigurationofthein-street sign with all R1-6 signs placed in the roadway. Figure 1-7 shows a gateway

configurationwithallblanksigns.Signplacementwasidenticalforthetwoconfigurations.

LocationsTwodifferentsitesinthecityofKalamazoo,Michiganwereusedinthisstudy.Thepostedspeedlimitatbothsiteswas35mph.ThefirstcrosswalkwasonRoseStreetattheT-intersectionwith

AcademyStreet.RoseStreethasfourlaneswithtwotravellanesineachdirectionandon-streetparking, the second sitewas amidblock crosswalk on Rose Street near the KalamazooValleyCommunity College Campus (KVCC) with two travel lanes in each direction and a pedestrian


refuge islandseparatingnorthandsouthbound traffic.TheAnnualDailyTraffic (ADT)onRose

Streetwas6,820.

Figure1-6:In-streetsignshowingagatewayatRoseSt.atAcademySt.withallR1-6Signs

Figure1-7:In-streetsignshowingagatewayatRoseSt.andAcademySt.withallblanksigns

ResearchDesignBecauseoftherobustchangesproducedbythistreatmentinthepreviousstudies,areplication

logic reversal design was employed in this study. Followingmultiple daily baselinemeasures(sessions during which no treatments were present) the treatments were evaluated in a


counterbalancedorder formultipledays.Datawere returned tobaseline andback to various

treatmentconditionsmultiple times toconfirmthe robustchanges indriverbehavior throughrepeateddirectreplications.

ResultsTheaveragepercentageofdriversyieldingright-of-waytopedestriansduringeachconditionatboth locations ispresented inTable1-2.AtRoseStreet. atAcademyStreet, yieldingbehavior

duringbaselineconditionaveraged6%.Duringthegatewaywithblanksconfiguration,yieldingaveraged32%.DuringthegatewaywithR1-6signsconfiguration,yieldingaveraged80%.AttheRose Street at KVCCmidblock location yielding during baseline also averaged 6%. During the

gatewaywithblanksconfiguration,yieldingaveraged36%.DuringthegatewaywithR1-6signsconfigurationyieldingaveraged78%.Figures1-8and1-9showstheaveragepercentofdriversyieldingduringeachsessionateachsite.Thesedatashowthatyieldingbehaviorwasrelatively

consistent at each site and that results did not varywhen the first treatmentwas either thegatewaywithblanksconfigurationorthegatewaywithR1-6signsconfiguration.Thestabilityofthe effect over multiple replications provides clear evidence that the differences in driver

yieldingbehaviorwerearesultofthetwodifferenttreatmentconditions.

RoseStreetatAcademyStreet RoseStreetMidblockatKVCC

Configuration MeanPercentYielding Configuration MeanPercentYielding

Baseline 6% Baseline 6%

BlankSigns 32% BlankSigns 36%

R1-6Signs 80% R1-6Signs 78%

Table1-2.Themeanpercentageofdriversyieldingtopedestriansduringbaseline,theblank

signgatewayandtheR1-6signgatewayconfigurations

Figure1-8.PercentofdriversyieldingtopedestriansatRoseSt.atAcademySt.


Figure1-9.PercentofdriversyieldingtopedestriansduringeachconditionatRoseStreetatKVCC.

EVALUATIONOFPOSITIONOFSIGNS

This study compared the full Gateway configuration with R1-6 signs only on the white lines

separatingtravellanesinthesamedirectionandR1-6signsonlyattheedgeoftheroadwayattwosites.

LocationsTwodifferentsitesinthecityofKalamazoo,MIwereusedinthisstudy.Thepostedspeedlimitat both siteswas 35mph. The first crosswalkwas on Rose Street at the T intersectionwith

Academy Street. Rose Street has four lanes with two lanes in each direction and on-streetparking,thesecondsitewasatacrosswalkontheintersectionofSouthWestnedge,aone-wayroadwith two southbound lanes and on-street parking on both sides of the street, at the T-

intersectionwithRanneyStreet.TheADTonRoseStreetwas6,820andonSouthWestnedgeitwas14,709.

ResearchDesignBecauseoftherobustchangesproducedbythistreatment inthepreviousstudy,areplicationlogic reversal design was again employed in this study. Following multiple daily baseline

measures (sessionsduringwhichno treatmentswerepresent) the treatmentswereevaluatedformultiple days. Datawere returned to baseline and back to various treatment conditionsmultipletimestoconfirmtherobustchangesindriverbehaviorthroughdirectreplication.The

treatmentsevaluatedwere1.ThefullGatewaywithR1-6signsattheedgesoftheroadandonalllanelines;2.TheplacementofR1-6signsonlyontheedgesoftheroadway;3.TheplacementoftheR1-6signsonlyonthewhitelanelinesseparatingtravellanescarryingtrafficinthesame

direction.


Results

The results of this experiment are presented in Table 1-3. At the Rose Street at Academylocationthebaselineyieldingwas6%,thefullGatewayproducedyieldingof80%,theedgesignsaloneconditionwasassociatedwith36%yieldingandthesignsonlyonthecenterlineproduced

52%yielding.At theSouthWestnedge siteyieldingwas0%duringbaseline,89%with the fullGateway,10%withtheedgesignsalone,and18%withsignsonlyonthewhitelanelines.

Table 1-3. Mean percentage of drivers yielding to pedestrians during baseline, the fullgateway,R1-6signsonlyattheedgesoftheroadway,andR1-6signsonlyonthewhitelanelines.

RoseSt.@AcademySt. Condition PercentYielding

Baseline 6%

GatewaywithallR1-6Signs 80%

CenterlineandEdgesignsonly 36%

R1-6SignsonlyonLaneLines 52%

S.WestnedgeAve.atRanneySt. Condition PercentYielding

Baseline 0%

GatewaywithallR1-6Signs 89%

R1-6SignsonEdgealone 10%

R1-6SignonLaneLinealone 18%

GATEWAYEFFECTONVEHICLESPEEDSpeed data were collected at one crosswalk in the presence and absence of the gatewaytreatmenttodeterminehowmuchdriversslowedwhentraversingthecrosswalk.

SettingThe participants were 2,000 motorists using two southbound traffic lanes approaching thecrosswalkonRoseStreetattheintersectionwithAcademyStreet.Driverswereexcludedifthey

changed lanes after their speed was read at the dilemma zone and if they parked at thesouthbound meters on Rose Street. It is important to note that drivers were excluded ifpedestrianswere attempting to enter orwithin the crosswalk and ifmotorists in northbound

traffic laneswere turningorattempting to turnwhileadriverwas in thedilemmazone.Thusthesedatashowtheeffectofthegatewayintheabsenceofpedestrians.

MethodThe dependent variablewas the vehicle’s speed at the dilemma zone,which begins 183 feetsouthofthecrosswalkandendsatthethresholdofthecrosswalk.Allspeedswerereadusing

laserradarlocatedinavehicleparkedsouthofthecrosswalk.


Results

TheresultsarepresentedinFigure1-10.Thesedatashowdriverswereacceleratingduringthebaseline speedmeasures when the gateway was absent.When the gateway was introducedaveragespeeddecreasedfrom26.8mphto23.1mphatthedilemmazoneandfrom28.3mph

to18.1mphatthecrosswalk,a10mphdrop.AsingleR1-6signlocatedontheyellowlineatthecenteroftheroadwasassociatedwithaverysmallreductioninspeedsimilartothatreportedinresearchthatonlyusedonesign(7).

Figure1-10.Driver speedapproaching the crosswalkat thedilemmazone (blue line)andatthecrosswalk(redline)duringbaselineandgatewayconditiononRoseStreetatAcademy.

Discussion Thepurposeofthisseriesofstudieswastoevaluatevariablescontributingtotheimprovementproducedbythegatewaytreatmentondriveryieldingright-of-waytopedestriansatcrosswalks.

Althoughtheconfigurationswereonlycomparedatalimitednumberofsites,theconsistencyoftheresultsproducedforeachconfigurationintroducedmultipletimesateachsite,theobviouslargechangesindriverbehavior,andtheconsistencyofresultsbetweendifferentsitessuggest

thatthesefindingarerobust.Alloftheconfigurationstestedincreaseddriveryieldingbehavior.However, no configurationmatched the effectiveness of the complete gateway configurationwithallR1-6signs.

One hypothesis testedwaswhether the distance between the signs had an influence on theefficacyof thegatewayR1-6treatment.Theanalysisofgapwidthbetweensignsshowedthat

narrower configurations weremore effective at each site where this variable was examined.


However,dataalsoshowedthatsignplacementhadaneffectonsignefficacy.Placingsignson

thewhite lane linepositionsaloneproducedbetterdriveryieldingbehavior than signsplacedonlyontheroadwayedgeatallsitesandthefullgatewayproducedthehighestlevelofdriveryieldingbehavior.

Another hypothesis tested was whether the content of the signs influenced driver yieldingbehavior. The blanks configurationwas significantly less effective than the configurationwith

thesignmessagepresent.Speed data suggest that the gateway treatment has a marked effect on vehicle speeds

approachingthecrosswalk,andthatdriversbeganslowingatthedilemmazonewhentheysawthe gateway ahead at the crosswalk. These data were collected because research assistantsfrequently observed drivers slowing as they approached the crosswalk. Not only was the

magnitudeof the reduction clearly visible, but it also reduced vehicle speedbelow the speedassociated with fatal crashes (25). Speed reductions can reduce both the probability of apedestriancrashbygivingdriversmoretimetoreact,andreducingthetunnelvisionassociated

withhighervehicle speeds.Reducedspeedcanalsodecrease the severityof injuries shouldacrashoccur.

InordertodeterminewhethertheGatewayincreasedthepercentageofhardbrakingdatawererecordedonhardbreakingattwositesinAnnArbor,MI:NixonRd.atBluettRd.,andS.DivisionSt. at E. Jefferson, St. At the Nixon Rd. crosswalks 758 vehicles that slowed at the crosswalk

wereobservedandnoinstancesofhardbrakingwereobserved.AttheS.Divisioncrosswalk912vehiclesthatslowedwereobservedandonlyoneinstanceofhardbrakingwasobserved.Thesedatawerecollectedseveralweeksafterthepermanentinstallations.


CHAPTER2–EVALUATIONOFCONFIGURATIONSTHATIMPROVESURVIVALOFTHEGATEWAYTREATMENT

INTRODUCTION

Although the gateway R1-6 treatment in some applications has proven to be as effective atincreasingyieldingright-of-waytopedestriansasmoreexpensivetreatmentsliketheRRFBand

PHB, it ismoreeasilydamagedthantreatmentsmountedonamastarmoronthesideoftheroad.ThePHBhasbeenshowntoincreasethepercentageofdriversyieldingtopedestriansinMichigantobetweenfrom61%and95%andtheRRFBhasbeenshowntoincreaseyieldingin

Michigantobetween55%and89%(12).Thereareseveralwaystoreducethevulnerabilityofthegatewaytreatment.Onewaytoimprovesignsurvival istoinstallthesignsattheroadwayedgepositionofthegatewayoutofthetravelwayinthegutterpan.Gutterpaninstallationis

lesslikelytobehitbuthasseveralpotentialdrawbacks.First,itcouldpresentdrainageissuesatsomelocationswhendebrisaccumulatesaroundthesign.Second,gutterpanplacementcouldbeaproblemforstreetsweepers.Third,theremaynotbeagutterpanandtheentirelanemay

be needed for larger vehicles. Fourth, narrow bike lanesmay preclude gutter pan placementbecausethesignwouldbetooclosetoriders.

Onealternativetogutterpanplacementisplacementontopofthecurb.Thisplacementwouldbelessvulnerabletocollisionanddoesnothaveanyofthepotentialdrawbacksofgutterpansignplacementandcouldbeusedatedgesatthesideoftheroadandattheedgesofrefuge

islandsoramedianisland.Signsplacedontopofthecurbaspartofagatewaytreatmentlookvery similar to signsplaced in thegutterpan toapproachingdriversandmayproduce similareffects.However,curbtopplacementisonlypermittedonmedianandrefugeislandsbutnotat

theedgeoftheroad.Placementattherightedgeoftheroadcurrentlyrequirespermissiontoexperiment from the Federal Highway Administration (FHWA). Therefore, permission toexperimentwasobtainedfromFHWAforthisstudypriortoplacingR1-6signsontopofthecurb

attherightedgesoftheroad.

Anotherway to increase thesurvivalof in-street signsplaced inmorevulnerable locationsonwhile lines dividing lanes carrying traffic in the same direction is the use of a flexible robustdelineator at these gateway positions rather than a R1-6 sign. These delineators have been

documentedtotakelargenumberofhitsatspeedsgreatlyinexcessofthoseexpectedatmostcrosswalklocationsandhenceshouldbeexpectedtohavearelativelylonglife.Atsiteswithtwolanesineachdirectionandapedestrianrefugeislandseparatingopposingtraffic,edgesignson

bothsidesof thegatewaycouldbe installedoutof thetravelwayontopof thecurbonbothsidesoftheroad,ontopoftheedgeoftherefugeisland,andaflexibledelineatorpostcanbeinstalledonthewhitelinesdividingtravellanes.Onroadswithoutapedestrianrefugeisland,a

R1-6 signwould be required on the yellow centerline. Signs placed on a curb type basemayshowthebestsurvivalinthiscondition.


The purpose of this series of studies is to compare the effects of gutter pan versus curb top

placement and R1-6 sign versus robust delineator placement on white lane lines on driveryieldingright-of-waytopedestrians.InExperiment1,gutterpanversuscurbtopplacementwascompared at four sites. In Experiment 2, the use of the R1-6 sign on thewhite lane linewas

comparedwiththeuseoftherobustdelineatoronthelaneline.

EVALUATIONOFGUTTERPANVERSUSCURBTOPPLACEMENT

InordertodeterminethedifferenceinyieldingbehaviorbetweenplacementofR1-6signsinthegutterpanandon topof curb, thegatewayconfigurationedgesignsplaced in thegutterpanweredirectlycomparedwiththeedgesignsplacedontopofthecurbatfourlocations.

GatewayConfigurationsTwo types of gateway configurations were used in this study. The first configuration was a

gateway with all R1-6 edge signs in the gutter pan. The second type of configuration was agatewaywithallR1-6signsontopofthecurbonthesideoftheroadorifpresentthesideofthemedianisland.Figure2-1showsagatewayconfigurationwiththeedgesignsonthecurb.

Figure2-1.GatewayconfigurationwiththeedgesignonthecurbatacrosswalkonEastHuron

Street.Locations

Fourdifferentcrosswalks,threeinthecityofAnnArbor,MIandoneinthecityofKalamazoo,Michiganwere studied in this experiment. The first crosswalkwasonEastHuronbetweenN.Thayer St. andN. Ingalls St. EastHuronat this locationhas two travel lanes in eachdirection

separatedbyapedestrianrefugeislandandaspeedlimitof30mph.ThesecondcrosswalkwasamidblocklocationonSouth7thStreetnorthofPioneerSchoolDriveandhadapostedspeedof35mph.Thisstreethastwolanesandabikelaneinonedirectionandonelaneandabikelane

intheseconddirectionwithamedian islandseparatingthetwodirectionsof travel.Thethird


crosswalkwasonNixonRd.atBluettRd.NixonRd.hastwolanesinonedirectionandonelane

intheotherdirectionandapostedspeedlimitof30mph.AtalloftheAnnArborcrosswalks,theR1-6asignwasusedwithastopsignsymbolinplaceoftheyieldsignsymbolbecauseAnnArborhad a stop rather than a yield ordinance. The fourth crosswalkwas on Rose Street at the T-

intersectionwith Academy Street in Kalamazoo. Rose Street has four laneswith two lanes ineachdirection,on-streetparking,aspeedlimitof35mphandanADTof6,820.

ExperimentalDesignAreplicationlogicreversaldesignwasemployedinthisstudy.Followingmultipledailybaselinemeasures(sessionsduringwhichnotreatmentswerepresent)theedgesignswerefirstinstalled

inthegutterpanandevaluated,thenplacedontopofthecurbandevaluated.Afterareturntothe baseline (no gateway) condition the signs were again introduced and evaluated in acounterbalancedorder formultipledays.Datawere returned tobaseline andback to various

treatmentconditionsmultipletimestoconfirmthechangesindriverbehaviorthroughrepeateddirectreplications.

ResultsTheaveragepercentageofdriversyieldingright-of-waytopedestriansduringeachconditionatboth locations ispresented inTable2-1.At themidblock locationonE.HuronStreet,yielding

behavior averaged 62% during the baseline condition, 97% during the gutter pan placementcondition, and 92%during the curb top placement condition. At themidblock location on 7thStreet,driveryieldingaveraged15%during thebaselinecondition,70%during thegutterpan

placementcondition,and54%during thecurb topplacementcondition.At theNixonRd.andBluett Rd. Site yielding averaged 40% during baseline, 93% during the gutter pan placementcondition,and86%duringthecurbtopcondition.AttheRoseSt.andAcademySt.site,yielding

averaged6%duringbaseline,82%duringthegutterpanplacementcondition,and72%duringthe curb top placement condition. Average performance across all four siteswas 31% duringbaseline, 86% during the gutter pan placement condition, and 76% during the curb top

placementcondition.Althoughyieldingwas10%lowerduringthecurbtopplacementcondition,it was still markedly higher than baseline. Figures 2-2, 2-3, 2-4, and 2-5 show the averagepercentofdriversyieldingduringeachsessionateachcrosswalk.Thesedatashowthatyielding

behaviorwas relatively consistent at each site and that results did not varywhen gutter panplacementwas the first treatment or curb top placement conditionwas introduced first. The

stabilityof theeffectovermultiplereplicationsprovidesclearevidencethat thedifferences indriveryieldingbehaviorwerearesultofthetwodifferenttreatmentconditions.


Table 2-1. The percent of drivers yielding to pedestrians during the baseline, gutter panplacementandcurbtoplacementconditionsusingR1-6signsonatemporarybasis.

Location Baseline GutterPanPlacement CurbTopPlacement

E.HuronMidblock 62% 97% 92%

Midblock7thStreet 15% 70% 54%

NixonRd.atBluettRd 40% 93% 86%

RoseSt.atAcademySt. 6% 82% 72%

Mean 31% 86% 76%

Figure2-2.Percentofdriversyieldingduringeachconditionoftheexperiment


Figure2-5.Thepercentofdriversyieldingduringconditionoftheexperiment.

EVALUATIONOFROBUSTDELINEATORONTHEWHITELANELINE

Oneofthemostvulnerablepositionsinagatewaytreatmentisthewhitelanelineonmultilane

roads.Theuseofarobustdelineatoroffersseveraladvantages.First,itisnarrowerthanaR1-6sign.Second,itcansurvivesignificantlymorehitsathigherspeeds.Thedelineatorselectedfortestinginthisstudyhadbeentestedandfoundtowithstand100hitsatspeedsof60mph.The

purposeofthisstudyistodeterminehowmuchtheeffectivenessofthegatewaydeclineswhenayellowgreendelineatorissubstitutedfortheR1-6signsinthewhitelanelineposition.

GatewayConfigurationsTwo types of gateway configurations were used in this study. The first configuration was a

gatewaywith threeR1-6 signs ineachdirection (twoedge signsandoneon thecenterwhiteline).ThesecondtypeofconfigurationwasagatewaywithtwoedgeR1-6signsandadelineatoron thewhite lane line ineachdirection.Figure2-6 showsaphotoofagatewayconfiguration

withadelineatoronthewhitelanelines.


Figure2-6.Gatewaytreatmentwithdelineatorsplacedonwhitelanelines

LocationsThreecrosswalk locationsinthecityofKalamazoo,Michiganwereusedinthisstudy.Allthree

crosswalkshadmultiple travel lanes in the samedirectionof travel. The first locationwasonRoseStreetattheT-intersectionwithAcademyStreet.RoseStreethasfourlaneswithtwolanesineachdirectionandon-streetparking,thesecondsitewasamidblockcrosswalkonRoseStreet

nearKVCCwithtwolanesineachdirectionandapedestrianrefugeislandseparatingnorthandsouthboundtraffic,andthethirdcrosswalkwasonaone-wayroadwithtwosouthboundlanesand on-street parking on both sides of the street on south Westnedge Avenue at the T-

intersectionwithRanneyStreet.TheADTonRoseStreetwas6,820andonSouthWestnedgeitwas14,709.

ExperimentalDesignAreplicationlogicreversaldesignwasemployedinthisstudy.Followingmultipledailybaselinemeasures, the treatmentswere evaluated in a counterbalancedorder formultiple days.Data

werereturnedtobaselineandbacktovarioustreatmentconditionsmultipletimestoconfirmthechangesindriverbehaviorthroughrepeateddirectreplications.

ResultsTheaveragepercentageofdriversyieldingright-of-waytopedestriansduringeachconditionis

showinTable2-2.Driveryieldingright-of-waytopedestriansaveraged5%acrossthethreesitesduringbaseline.ThecompletegatewaywithallR1-6signs increasedyieldingto82%whilethegatewaywiththedelineatorwasassociatedwith60%increaseinyielding.Figures2-7,2-8and


2-9 show that the individual level of yieldingwas very similar for each condition during each

replication.Table 2-2. The percentage of drivers yielding to pedestrians at each crosswalk during the

baseline,thegatewaywithallR1-6signs,andthegatewaywithdelineatorcondition.

Location BaselineGateway

withallR1-6SignsGateway

withDelineator

RoseStreetatAcademyStreet 7% 79% 60%

RoseStreetatKVCC 7% 77% 60%

SouthWestnedgeAvenueatRanneyStreet

0% 89% 59%

Mean 5% 82% 60%

Figure 2-7. The percentage of drivers yielding to pedestrians at Rose and Academy duringbaseline,theR1-6gatewayconditions,andthegatewaywithdelineatorcondition.

Figure2-8.ThepercentageofdriversyieldingtopedestriansatRoseatKVCClocationduringbaseline,theR1-6gatewayconditions,andthegatewaywithdelineatorcondition.


Figure2-9.ThepercentageofdriversyieldingtopedestriansatWestnedgeandRaneyduringbaselinetheR1-6gatewayconditionsandthegatewaywithdelineatorcondition.Discussion The purpose of this series of studieswas to evaluatewhether gateway configurationswith apotentiallyhigherlikelihoodofsurvivalcouldyieldsimilarresultsasthoseproducedbythefull

R1-6gatewayconfiguration.Theresultsofthefirststudyshowedthecurbtopplacementoftheedgesignscould increasedriver’syieldingright-of-waytopedestriansto levelsalmostasgoodasthoseproducedbygutterpanplacement.Althoughsignsplacedinthegutterpanareashould

not be struck very often, they could present problems for drainage, street sweepers, and atlocations with narrow bicycle lanes. Placement on top of the curb would not share theselimitations. It is also the case that signs placed on top of the curb should also receive less

intentionalstrikesthensignsplacedinthegutterpan.Anotheradvantageofcurbtopplacementis that the sign bases could remain in place during the winter plowing season in northernclimates.However,itisnotclearwhetherthesignsthemselvescouldstanduptothesnowload

fromplowing.ThesecondstudyexaminedwhetherreplacingtheR1-6signsinstalledonthewhitelanelineon

multilaneroadswithaflexibledelineatorcouldproduceasimilarpercentageofdriversyieldingright-of-way topedestrians. The results of this study showed that a gatewaywithdelineatorsproduced much better yielding than baseline but 22% fewer drivers yielded right-of-way to

pedestriansduring thisconditionthanduring theR1-6signonwhite lane linescondition.Oneadvantage offered by the delineator is that they can withstand a large number of strikes.Anotheradvantageisthattheycanbeeasilyscrewedoutbeforethestartoftheplowingseason

innorthernclimates,andcanbereplacedinamatterofminutes.


CHAPTER3–EVALUATIONOFTHEGATEWAYTREATMENTINVARIOUSCROSSWALKAPPLICATIONSThischapterevaluatestheuseofthegatewaytreatmentinavarietyofcrosswalkapplications,includingtrafficcircles,roundabouts,freewayentranceramps,andtrailcrossings.Theseapplicationsrevealedstrengthsandweaknessesofthegatewaytreatmentlocationandhelpeddeterminewheretheymaybemosteffectivelydeployed.Also,thechapterevaluatestheuseofagatewayinconjunctionwithaRRFB.Thefinalstudyinthischapterexamineswhetherplacingthegatewayatonecrosswalkatanintersectionwouldhaveaneffectofyieldingattheuntreatedleg.

TRAFFICCIRCLEINSTALLATIONSLocationsAgatewaywasinstalledattwocrosswalksatatrafficcircleinMarshall,Michigan.Bothgatewayswereinstalledbetweenthepaintedareaadjacenttothecenterislandofthetrafficcircleandthegutterpanontheoutsideofthecircle.Onecrosswalklocationhadawideconfigurationwithasinglelaneroadwaywidthof36.8feetbetweenthepaintedareaandtheoutercurb.Thesecondcrosswalklocationhadanarrowconfigurationwithasinglelaneroadwidthof22.3feetbetweenthepaintedareaandtheoutercurb.TheleftframeofFigure3-1showsthewideconfigurationcrosswalkandtherightframeshowsthenarrowconfigurationcrosswalk.ResultsThedata,asshowninFigure3-2,illustratesthatyieldingincreasedmoreatthesitewiththenarrowconfiguration.

Figure3-1.Left:widegateway.Right:narrowgateway.


Figure3-2.DriveryieldingatthetwocrosswalksatatrafficcircleinMarshall,Michigan.DiscussionGatewaytreatmentsattrafficcirclesimproveddriveryielding.However,narrowingthegatewaygapresultedinahigheryieldingpercentage.DatareportedinChapter1alsocomparedwideandnarrowgatewayconfigurationsatbothofthesesitesandfoundbetteryieldingwiththenarrowerconfiguration.

ROUNDABOUTINSTALLATIONSLocationsThegatewayconfigurationwasexaminedattwosingle-laneroundaboutlocationsinBentonHarbor,Michigan.Eachroundabouthassplitterislandsateachentryandexitpointthatservedaspedestrianrefugeislands.Thereisalsoanislandinthemiddleofeachroundaboutwithlowgrowingfoliage.YieldingwasonlymeasuredatthefourcrosswalksonEastMainStreet.Thepostedspeedlimitwas35mphatbothroundabouts.Therewaslittlepedestriantrafficateachsite,butalargeamountofvehicletraffic.Figure3-3showsthegatewayconfigurationatonecrosswalklegoftheroundaboutattheEastMainStreet.Notethatallsignswereplacedinoradjacenttothegutterpan.Itshouldbenotedthattherewasalotoftrucktrafficatthissite.

ResearchDesignAmultiplebaselinedesignwasemployedinthisexperiment.Afterdatawerecollectedduringthebaselineconditionatbothsites,onesitereceivedthegatewayin-streetsigntreatmentwhilethesecondlocationremainedinthebaseline(untreated)condition.Afteraconvincingincreasewasnotedatthefirstsite,thegatewaytreatmentwasaddedatthesecondlocation.Someoftheyieldingdatawasseparatelycodedfordriverenteringandexitingthecrosswalk.

0

10

20

30

40

50

60

70

80

90

100

WideCrossing NarrowCrossing

Percen

tDriversYielding

Baseline

Treatment


Figure3-3.Gatewayconfigurationatoneoftheroundaboutlocations.ResultsFigure3-4showsthepercentageofdriversyieldingright-of-waytopedestriansateachlocation.BaselinedriveryieldingbehaviorattheEastMainSt.atRiverviewDr.andEastMainSt.at5thSt.roundaboutsaveraged9%and19%respectively.TheintroductionofthegatewayattheEastMainStreetandRiverviewDr.siteincreasedyieldingto43%.TheintroductionofthegatewayattheEastSt.and5thsiteincreasedyieldingto45%.

Figure3-4.Thepercentageofdriversyieldingright-of-wayduringthebaselineandgatewaytreatmentsatthetworoundaboutlocations.


DataforDriversEnteringandExitingRoundaboutsBecausethefrequencyofdriveryieldingright-of-waytopedestriansseemedhigherfordriversenteringtheroundaboutthanexitingtheroundabout,datacollectorswereinstructedtoseparatelymeasuretheyieldingbehaviorofdriversenteringandexitingtheroundabout.Figure3-5showstheratioofdriversyieldingright-of-waytopedestrianswhenenteringtheroundaboutcomparedtoyieldingwhenexitingtheroundabout.Driverswerealmosttwiceaslikelytoyieldright-of-waytoapedestrianwhenenteringascomparedtoexitingtheroundaboutatthesetwosites.

Figure3-5.Ratioofdriversyieldingtopedestriansuponenteringandexitingtheroundaboutduringthebaselinecondition.DiscussionDriverswerelesslikelytoyieldright-of-waytopedestriansincrosswalkswhenexitingtheroundaboutthanwhenenteringit.Althoughthetreatmentincreasedyieldingwhendriverswereenteringorexiting,thisrelationshipcontinuedtoholdwithdriversmuchlesslikelytoyieldwhenexitingthanentering.Althoughtheeffectsatroundaboutsweremodest,theywouldmakeiteasiertocross.BecauseotherdatashowthatthecombinationofthegatewaywithaRRFBleadstoveryhighlevelsofyielding,thistreatmentshouldbeconsideredatroundaboutswithanRRFBlocationwithrelativelypoordriveryieldingbehavior.

FREEWAYENTRANCERAMPINSTALLATIONSAnotherpossibleapplicationforthegatewaytreatmentisatfreewayentranceramplocations.Pedestrianscrossingentrancerampseitherapproachthecrosswalkfacingapproachingtrafficorwithapproachingtrafficattheirback.Crashriskshouldbehigherforpedestriansapproachingthecrosswalkwithtrafficapproachingfrombehind.

0

0.5

1

1.5

2

2.5

Mainat5th MainatRiverview

Rad

oExidngto

Entering


LocationsThegatewayconfigurationwasexaminedatthesouthandnorthentrancerampstoI-94onSouthWestnedgeAvenue.Bothrampshavealongdedicatedapproachlanetotherampentrance.Figure3-6showstheapproachtothenorthramp.TheleftframeofFigure3-7showsaphotographofthewidegatewayconfigurationatthenorthentrancerampandtherightframeshowsaphotographofthenarrowgatewayconfiguration.ThewideconfigurationconsistedoftwoR1-6signsinstalledinthegutterpan,whilethenarrowconfigurationaddedtwoflexibledelineatorpostsonthelanelines.

Figure3-6.PhotographofthesidewalkapproachingthenorthentrancetoInterstateI-94atWestnedgeAvenue.


Figure3-7.Thephotoontheleftshowsthewideconfigurationofthegatewayandthephotoontherightshowsthenarrowconfigurationwithaddedflexibledelineatorpostsonthelanelines.ResearchDesignAmultiplebaselinedesignwasemployedinthisexperiment.Afterdatawerecollectedduringthebaselineconditionatbothsites,thenorthentrancerampreceivedthewidegatewayin-streetsigntreatmentwhilethesouthentranceramplocationremainedinthebaseline(untreated)condition.Afteraconvincingincreasewasnotedatthefirstsitethewidegatewaytreatmentwasaddedatthesecondlocation.Nextthenarrowgatewayconfigurationwasintroducedatthenorthramp.ResultsTheresultsofthisexperimentareshowninFigure3-8.Driveryieldingduringthebaselineconditionwasnegligibleatbothentranceramps(1%atthesouthrampand2%atthenorthramp).Duringthewidegatewayconditionyieldingincreasedto17%atthesouthrampandto13%atthenorthramp.Whenthenarrowgatewayconditionwasintroducedatthenorthramp,yieldingincreasedto31%,untilatruckstruckoneofthetemporarydelineatorsdestroyingit.Giventhelimitednumberoftemporarydelineatorsavailable,theWMUteamdecidedtoenddatacollectionattherampexitsites.


Figure3-8.PercentageofdriversyieldingtopedestriansatthesouthandnorthentranceramptoI-94duringeachconditionoftheexperimentDiscussionAlthoughthewidegatewayplacedinthegutterpanatanentrancerampproducedsomeincreaseindriveryieldingtopedestriansinthecrosswalks,itwasnotsufficienttorecommendthegatewayforthisapplication.Althoughtherewasanincreaseinyieldingproducedbythenarrowconfigurationwithaddeddelineators,thelevelsobtainedwerestillfarlowerthantheeffectsobtainedatmidblocklocationsandatuncontrolledcrosswalksatintersectionswithastop-controlledminorcrossstreet.Heavytrucktrafficatfreewayentrancerampsalsomakeitunlikelythatanarrowconfigurationwouldsurviveatthistypeoflocation.Theonlyinstallationthatwouldlikelysurviveforlong-termusewouldbecurbtopplacementanditisunlikelythattheywouldproducemeaningfulincreasesinyieldingatthistypeofsite.


Onereasonwhythegatewaywassoineffectiveatthistypeofsiteisthetendencyfordriverstowanttoincreasespeedtomatchtrafficonthefreewayatthetopoftheramp.Onceadriveristheprocessofaccelerationtheybecomelesslikelytomaketheefforttoswitchtheirfootfromtheacceleratorpedaltothebrake.Thisphenomenonisalsolikelyatworkatexitstoroundaboutswheredriversareintheprocessofaccelerating.

TRAILCROSSINGINSTALLATIONSLocationsThegatewaytreatmentwasexaminedattwotrailcrossinglocationsinPortage,Michigan.ThefirstsitewasamidblocktrailcrossingonOaklandStreetjustsouthofMilhamRoad.Thistrailcrossinghasarefugeisland.Thereisalsoaconstantflashingyellowbeaconinadvanceofthissiteandapushbuttonactivatedflashingyellowbeaconatthecrosswalklocation.ThesecondsitewasatrailcrossingonGardenLaneontheCeleryCreektrail.ThewidegatewayconfigurationatOaklandStreetisshowninFigure3-9.ThewidegatewayconfigurationatGardenLaneisshowninthelowerframeofFigure3-10.

Figure3-9.WidegatewayconfigurationonOaklandStreet


Figure3-10.WidegatewayconfigurationatthetrailcrossingonGardenLane.ResearchDesignAreversaldesignwasemployedatbothsites.Inareversaldesignatreatmentisintroduced,removed,andreintroducedinordertoreplicatetheeffectofthetreatmentondriver’sbehavior.ResultsTable3-1showstheaveragepercentageofdriversyieldingtopedestriansduringeachconditionatbothsites.Thepercentageofdriversyieldingright-of-waytopedestriansduringbaselinewas3%attheOaklandStreettrailcrossingand21%attheGardenLanetrailcrossing.AtbothcrossingsthenarrowR1-6gatewayproducedthehighestlevelofyieldingbehaviorwith39%yieldingattheOaklandStreetlocationand75%yieldingattheGardenLanelocation.ThewidetreatmentconditionwithdelineatorsplacedintothepositionusedfortheR1-6signsforthenarrowconditionwasassociatedwith18%yieldingattheOaklandStreetlocationand72%yieldingattheGardenlanelocation.BecausetherewasapedestrianactivatedflashingyellowbeaconattheOaklandsiteitwastestedaloneandwiththewideandnarrowR1-6gatewayconditions.Thepedestrianbeaconalonewasassociatedwith2%yielding,thewidegatewayconfigurationyieldingincreasedto23%andthebeaconwiththenarrowconfigurationwasassociatedwith57%yielding.ThelowerframeofFigure3-11showsyieldingattheOaklandStreetlocationandtheupperframeofFigure3-11showsyieldingattheGardenLanelocation.


Table3-1.Thepercentageofdriversyieldingright-of-waytopedestriansduringeachconditionatthetwotrailcrossings.OaklandStreetTrailCrossing GardenLaneTrailCrossingCondition Percent

YieldingCondition Percent

YieldingBaseline 3% Baseline 21%R1-6gatewayWide 10% R1-6gatewayWide 67%R1-6gatewayNarrow 39% R1-6gatewayNarrow 75%RobustDelineatorAlone 5% RobustDelineatorAlone 61%R1-6gatewayWide+RDNarrow 18% R1-6gatewayWide+RDNarrow 72%Baseline+Beacon 2% Baseline+Beacon NAR1-6gatewayNarrow+Beacon 57% R1-6gatewayNarrow+Beacon NAR1-6gatewayWide+Beacon 23% R1-6gatewayWide+Beacon NA

Figure3-11.Percentageofdriversyieldingtopedestriansduringeachconditionatthetwotrailcrossingsduringeachtreatmentcondition.DiscussionThegatewaywashighlyeffectiveatoneofthetrailcrossingsbutonlymoderatelyeffectiveatthesecondcrossing,whichhadahigherADT(nearly17,000)andspeed(35mph).Thecrosswalk


onOaklandStreethadatrafficsignalnorthofthemidblocktrailcrossing.Itwasnoticedthatvehiclesacceleratingfromafreshgreensignalseemedtobelesslikelytoyieldasweredriversapproachingagreensignalfromtheoppositedirection.Observersalsothoughtmanyvehiclesweretravelingoverthespeedlimitatthissite.Thissitealsohadalargenumberofplatoonedvehiclesduetothecloseproximityofthetrafficsignalswhichreleasedthevehicleswhenthelightturnedgreen.Thismaybeafactorworthyoffurtherstudy.

INSTALLATIONATANRRFBLOCATIONAnRRFBisanotherwaytoincreasemotoristyieldingright-of-waytopedestriansatcrosswalks.AnotherpossibleapplicationforthegatewaytreatmentisatRRFBsites(8,11,26).PreviousresearchhasdocumentedtheefficacyoftheR1-6signusedaloneontheyellowlineorinagatewayconfigurationtoincreaseyieldingatcrosswalkscontrolledbyanRRFBorPHBatlocationswheretheRRFBorPHBproducedlessyieldingthanreportedinnationalstudies(9).LocationsThegatewayconfigurationwasexaminedwithaRRFBatamidblocklocationatasmallparkonMonroeStreetEastofChestnutStreetinAllegan,Michigan.Thecrossinghasapedestrianrefugeisland.Theedgesignsadjacenttotherefugeislandwereplacedinthegutterpanandthesignsadjacenttothecurbwereplacedonthewhitelanelines,asshowninFigure3-12.

Figure3-12.RRFBlocationinAllegan,Michigan.ResearchDesignAmultipleelementdesignwasemployedinthisexperiment.Duringthebaselinephasetwodatasetswerecollectedeachday,onewithoutandonewiththeRRFBactivated.Subsequently,thegatewaydeviceswereinstalledanddatacollectedwiththeRRFBactivatedandnotactivated.


ResultsTheresultsofthisexperimentareshowninTable3-2.Driveryieldingduringthebaselineaveraged18%whenthedevicewasnotactivatedand65%whentheRRFBwasactivated.Thisresultissomewhatlowerthanthatfoundinnationalstudies(8,11,25).Afterthegatewaywasinstalled,yieldingwhentheRRFBwasnotactivatedincreasedto82%andaveraged96.5%whentheRRFBwasactivated.ItisinterestingtonotethatthecompletegatewayaloneinSeptemberwasmoreeffectivethantheRRFBbeforethegatewaywasintroduced.AsinpreviousstudiesthegatewayproducedamarkedincreaseintheefficacyoftheRRFB.OneoftheedgesignsonthelanelinewasdamagedbeforeOctober2015andNovember2015datawerecollected.Thisproducedadropinyieldingtothepartialgateway.Thissignwouldlikelyhavesurvivedifitwereplacedinthegutterpanorontopofthecurb.Table3-2.PercentofdriversyieldingatthemidblockcrosswalkinAlleganwhentheRRFBwasactivatedandnotactivated,duringbaselineandafterthegatewaywasinstalled.BeforegatewayInstalled AftergatewayInstalled

Sept Oct NovBaseline 18% gatewayalone 82% 50%* 57%*RRFB 65% gateway+RRFB 95% 98% 96%*TheRRFBonthenorthedgeofthecrossingwaslostduetoanimpact.

INSTALLINGAGATEWAYATONLYONELEGOFANINTERSECTIONSometimespedestrianexposureismuchhigheratthecrosswalkononesideofanintersectionthantheother.Thisstudyexaminedifagatewaytreatmentinstalledatthebusierlegproducedanincreaseinyieldingonthecrosswalkattheuntreatedlegoftheintersection.Thepurposeofthisstudywastodetermineifagatewayinstalledononelegofanintersectionhadaneffectonyieldingattheuntreatedlegoftheintersection.ParticipantsandSetting ParticipantsweremotoristsattheintersectionofMainStreetandBennettStreetnearthedowntownareaofThreeRivers,Michigan.Theintersectionconnectstwochurches,apark,andanelementaryschool.Thenorth-southtraffic(MainStreet)travelsintwolanes,withamiddleturnlane,withapostedspeedof3Omph.Theeast-westtraffic(BennettStreet)hasa25mphspeedlimitandiscontrolledwithstopsigns.Figure3-13showsadiagramoftheintersection.

Figure3-13.DiagramoftheintersectionofMainStreetandBennettStreetinThreeRivers,Michigan.


ResearchDesignAreversaldesignwasusedtodeterminetheeffectivenessofagatewaytreatmenttoadjoiningcrosswalks.Followingabaselineconditiontoestablishthepercentageofdriversyieldingright-of-waytopedestrians,thegatewaytreatmentwasintroducedatonelegofthecrosswalkanddriveryieldingwasassessedatbothlegsofthecrosswalk.ResultsTheresultsarepresentedinFigure3-14showthatthegatewaytreatmentproducedanincreaseinyieldingatbothcrosswalksalthoughtheeffectonthecrosswalklegwiththegatewayissomewhatlargerthantheeffectoncrosswalklegwithoutthegateway.Bothincreasesinyieldingwerelarge.Onereasonwhythegatewayinfluencedyieldingatbothlegswasmayhavebeenbecausethegatewayreduceddriverspeedattheintersection.Thisreductioninspeedbenefitspedestrianusingboththegatewaytreatedlegandtheuntreatedleg.

Figure3-14.Percentofdriversyieldingatthelegofanintersectionwiththegatewaytreatmentandotherlegofintersectionthatwasnottreated.

0

10

20

30

40

50

60

70

80

90

100

Baseline GatewayLeg OtherLeg

Percen

tDriversYielding


CHAPTER 4 - POTENTIAL ROADWAY CHARACTERISTICS INFLUENCING THEEFFICACYOFTHEGATEWAYTREATMENT

Table4-1listscharacteristicsforeachexperimentationsiteswheregatewayinstallationswereinstalled.Mostofthesesiteswereinstalledduringthepresentstudy,buttheTrowbridgeRoadsite,theLivernoissiteandthetwoFarmingtonsiteswereinstalledduringanearlierMDOTcontract.Thesehavebeenincludedtoincreasethediversityofsites.AmapoflocationsisprovidedinFigure4-1.Thedatadoshowsometrends.Thegatewayseemstoworkbestatmidblockcrosswalksanduncontrolledcrosswalksatanintersectionwithaminorstreetwithstopsigncontrol.Itdoesnotseemtomatterwhetherthesitehasarefugeislandormedian,butitisknownthatthepresenceofthesefeaturesisassociatedwithacrashmodificationfactor.Itisalsothecasethatamorerobustgatewaycanbeinstalledatamidblocksitewitharefugeislandbyinstallingthesignsontherefugeislandontopofthecurbandthesignsontherightedgeoftheroadinthegutterpanorontopofthecurbunderpermissiontoexperiment.Roadwaycharacteristicsmayinfluencetheprobabilitythatdriverswillyieldright-of-wayatacrosswalklocation.Onecharacteristicisvehiclespeed,whichmayormaynotrelatetothepostedspeed.Theprobabilityofadriveryieldingright-of-waytoapedestrianisknowntobeafunctionofvehiclespeed.Onereasonwhydriversarelesslikelytoyieldright-of-waywhentheyaretravellingfastisthatittakesmoreeffortandtimetobrake.Anotherreasonisthatdriversshowincreasedtunnelvisionwithincreasingvehiclespeed.Undertheseconditionstheymaynotdetectpedestriansenteringthecrosswalkfromthesideoftheroadway.Unfortunately,wedonothavedataonoperatingspeedsonmostoftheroadsstudiedinthisseriesofexperiments.Speedseemstobethemajorfactorwith61%ofdriversonaverageyieldingtopedestrianson35mphroad,78%ofdriversyieldingonaveragetopedestrian’s30mphroads,and83%ofdriversyieldingon25mphroads.Anothercharacteristicthatmightinfluencewhetheradriveryieldsright-of-waytoapedestrianisADT.However,TheGatewayworkedverywellatseveralsiteswithahighADTandaspeedlimitof30mph(E.HuronSt.,TrowbridgeRd.,andLivernoisAve.)butitdidnotworkwellattwositeswithahighADTandaspeedlimitof35mph(OaklandRd.andWestnedgeAve.).Theaveragepercentageofdriversyieldingtopedestriansaveraged79%forwhenAADTwasbelow10,000,69%whenAADTwasbetween10,000and20,000,and71%whenAADTwasover20,000.OverallthiseffectsizeforADTwassmallerthanthatforspeed.Yieldingaveraged69%withonevehicletravellaneineachdirectionand76%withtwomotorvehicletravellanesineachdirection.Cautionshouldbeusedwheninterpretingtheseresultsbecausetheyarenotbasedonmanysites.However,itdoesappearthatspeedmaybealargerfactorthanAADTandthattheGatewaytreatmentmayworkbestonroadswitha25or30mphspeedlimit.Anothercharacteristicmaybethenumberofengineeringdevicespresentthatinfluencedriveryieldingright-of-waybehavior.Forexample,advanceyieldorstoplines,aRRFB,oraPHBusedinconjunctionwithasingleR1-6signoracompletegatewaytreatmentmayproducehigherdriveryieldingbehaviorthanthepresenceofonlyonesafetyfeature.Clearandconsistentevidenceexistsshowingacumulativeeffectofmultiplesafetyfeatures.


Table4-1SelectroadwaycharacteristicsforMDOTproject120239(ResearchonComparisonofAlternativePedestrianCrossingTreatments)andMDOTproject114527(EvaluatingPedestrianSafetyImprovements)andyieldingresults


Figure4-1.StudyAreaMap


CHAPTER5–EVALUATIONOFTHELONGTERMEFFICACYANDSURVIVALOFTHEGATEWAYTREATMENTThischapterprovidesanexaminationofthelong-termpersistenceofdriveryieldingright-of-waytopedestriansandthelong-termsurvivalofthegatewaytreatment.Dataforlong-termefficacyofthegatewaytreatmentispresentedinTable5-1.Table5-1.Percentofdriversyieldingtopedestriansduringbaseline,initialtemporaryinstallation,andafterpermanentinstallationateachofthetreatmentsites.

PercentofDriversYieldingCrosswalkLocation

Baseline TemporaryInstallation

PermanentInstallation

FollowupPeriod X X 1Month 2Month 3MonthNixonRd.atBluettRd. 40 86 93 89 XS7thSt.midblockNorthofPioneerRd. 15 54 64 70 XS.DivisionSt.atE.JeffersonSt. 3 94 94 93 XE.HuronSt.westofN.IngallsSt. 40 86 X X XN.MainSt.NorthofE.MichiganAve. 6 X 64 53 50RoundaboutE.MainSt.at5thSt. 19 45 61 60 *33RoundaboutEMainatRiverviewDr. 9 43 44 44 44MarshallTrafficCircleNECrosswalk 13 54 71 71 50MarshallTrafficCircleSWCrosswalk 11 29 26 38 34S.WestnedgeatRanneySt. 0 59 33 29 *NAMonroeSt.eastofChestnutRRFBOff 6 X 82 *50 *56MonroeSt.eastofChestnutRRFBOn 65 X 95 98 96Mean 19 61 66 63 *Gatewayelementwasidentifiedasdamagedordestroyed

LONGTERMEFFICACYTheinstallationofthetemporarygatewayledtoanincreaseinthepercentageofdriversyieldingtopedestriansateachsite.Duringbaselineconditions,yieldingaveraged19%witharangeof0to40%.Atemporaryinstallationwasevaluatedafterthebaselineconditionateachsite.Whenthetemporarygatewaytreatmentwasevaluated,yieldingincreasedto61%witharangeof29%to94%.Oncethepermanentgatewaytreatmentswereinstalledforone-monthyieldingincreasedto66%witharangeof26%to95%.Twomonthsafterthepermanentinstallationyieldingwas63%witharangeof29%to98%.Datacollectedafterthefirsttwomonthsafterpermanentsignswereinstalledwasverysimilartodatacollectedwiththetemporarysigns.ThesignsinAnnArborwereinstalledonemonthlaterthanthesignsinsouthwestMichigansothereisnodataforthethirdmonthatthesesites.ThedatafromthesitesinsouthwestMichiganweresimilarforthethirdmonthwithexceptionoftwositesthathadelementsofthegatewaydamaged.ThesiteattheRoundaboutat5thinBentonHarborlostoneofthetwosignsplacedontheedgelanelineattheentrancetotheroundabout.Thissignwassheeredfromitsbase.AtthesiteatWestnedgeatRanneyinKalamazoo,thedelineatorwasseriouslydamagedreducingthegatewayeffect.Thissignshowedevidenceofmanymorestrikesthanothersigns.Manyofthesignsshowednoevidenceofbeingstruck.IfoneassumesthedatawillholdupforthethirdmonthinAnnArbor,themeanwouldbe65%,whichisthesameasthelevelproducedbythetemporarygatewayinstallations.Itshouldbenotedthatthesignsare


muchmoreeffectiveatsomesites.However,inallcasespedestriansdidnotneedtowaitlongtocrossafterthegatewaywasintroduced.Forexample,with50%yieldingeveryseconddriveryields;evenwith30%yieldingnearlyeverythirddriveryields.However,withlowbaselinelevelsof0%,3%,6%or9%,lessthanonedriverintenyields.

SURVIVALSignswerecheckedfordamageonaregularbasis.Manysignsshowednoevidenceofbeingstruck.Generally,signsmountedonacurbshowednoevidenceofbeingstruckandnoneweredamaged.Twoofthesignsinstalledintheroadwaywithflushmountedbaseswereseriouslydamaged.Onlyoneoftheflexibledelineatorpostswasdestroyedandthiswasatthesitewiththehighestnumberofstrikes.Preliminarydatasuggestthattheuseofacurbtypebasesmountedinthegutterpanorsignsmountedontopofthecurbmayshowgoodlong-termsurvival.Theuseoftheflexibledelineatorpostlookslikeitcansurvive,howeverthedataatWestnedgeandRanneyshowthissigncanbedestroyedifitisstruckonaregularbasis.Itshouldbenotedthatonesignthatwasdestroyedandtheotherthatwasseverelydamagedweresignswithflushmountedbaseslocatedontheedgelaneline.Thesedatasuggestthatsignsmountedflushtothestreetarelessrobustthanthesignsmountedoncurbtypebases.Itisunlikelythesesignswouldhavebeendamagediftheywereplacedontopofthecurb.Inclimatesthatrequireplowingsnowinwinter,itisnecessarytoinstallthesignsafterthesnowseasonendsandtoremovethesignsbeforethestartofthesnowseason.Becausetheinitialcostofinstallationisgreaterthanthecostofremovalandreinstallation,removalforwinteroperationsisnotlikelyamajorburdenfortheuseofthesesigns.Typicallypedestrianactivityisgreaterduringthespring,summerandfallthanduringwintermonths,particularlyinareaswithhightourismexposure.Thisiskeybecausetheonlysignsthatwouldlikelysurviveinwinterarethosemountedonthetopofthecurbattherightsideoftheroadandthosemountedontopofthecurbonarefugeormedianisland.However,itisnotcertainwhetherthesnowloadingfromtheplowcoulddamagethesignsifitpushesthemtotheside.


CHAPTER6–TYPICALLYCOSTSFORINSTALLATIONANDMAINTENANCEOF

THEGATEWAYTREATMENT(NOTEINSTALLATIONCOSTMAYVARY)

Table6-1.CostsforinstallationandmaintenanceofthegatewaytreatmentCOSTOFVARIOUSGATEWAYELEMENTS

R1-6signmountedonacurbtypebase $300

R1-6signmountedabasecementedontopofthecurbbesidetheroad $200

Costofaflexibledelineatorpost,base,capandepoxy $130

INSTALLATIONTIME(MOBILIZATIONANDTRAVELNOTINCLUDED)R1-6Signonacurbbasemountedintheroadway 5min.

R1-6signonabasecementedtothetopofthecurb 5min.

FlexibleDelineatormountedintheroadway 10min.

INSTALLATIONMATERIALCOSTANDTIME,PERCONFIGURATION

FOUR-LANECROSSINGWITHREFUGEISLAND COST TIMEGateway,curbmountedR1-6signswithcurbbases $1,820 30min.

Gateway,curbmountedR1-6signs,robustdelineatoronwhitelanelines $1,480 40min.

THREE-LANECROSSINGWITHNOREFUGEISLAND

GatewaywithcurbmountedR1-6signswithcurbbases $1,210 20min.

TWO-LANECROSSINGWITHNOREFUGEISLAND

GatewaywithcurbmountedR1-6signswithcurbbases $910 15min.Note:Costforallconfigurationswouldbelowerwithedgesignsplacedontopofcurb.ThisconfigurationrequiresFHWArequestforexperimentation

INDIVIDUALINSTALLATIONCOSTITEMSR1-6Signmountedonacurbbase $190

R1-6signmountedontopofcurbbesideroadway $190

Flexibledelineatorpostinstalledonwhitelaneline $110

All signs can be removed and taken out for winter operation very quickly. The flexibledelineators come with a recessed cap that protects the base from water damage. The curb

bases are removable and the holes need to be filled with silicone caulking to protect theroadwayfromwaterfreeze/thawdamageduringwinter.

Table6-2.RemovalandreinstallationtimesforgatewaytreatmentRemovalofcurbtypebaseandsign 2min.

Reinstallationofcurbtypebaseandsign 2min.

Removalofsignmountedontopofcurb 1min.

Reinstallationofsignmountedontopofcurb 1min.

Removalofflexibledelineator,installcapforwinter 40sec.

Reinstallationofflexibledelineator,removecapinspring 90sec.


CHAPTER7-CONCLUSIONSTheresultsofthefirstseriesofstudiesexploredseveralgatewayconfigurationsinfluencingtheefficacyofthegatewaytreatment.Onehypothesisthatwasconfirmedwasthatthenarrowness

ofthegapbetweenthesignswasinverselyrelatedtothetreatmenteffect.Thishypothesiswasconfirmed with narrow gaps leading to larger increases in driver yielding right-of-way topedestriansatseveraldifferenttypesofcrosswalkapplications.

Anotherhypothesistestedwaswhetherthesignmessageitselfhadaneffectondriveryieldingbehavior. It was found that the blanks configuration was significantly less effective than the

gatewayconfigurationwiththesignmessagepresent.A configurationanalysisalso showed that thepositionof the sign isa critical factor fordriver

yieldingbehavior.

• Gatewaytreatmentpositionsusedinisolationresultedinlowerdriveryieldingright-of-

waytopedestriansthanthefullgateway.• Signs placed on the white lane line alone exerted more control over driver yielding

behaviorthantheedgepositionsaloneatallsites,butlessyieldingthanthefullgateway

treatment.Itappearsthattheeffectisgreatestwhenmotoristsneedtodriveinalanebetweentwosigns.

However,theeffectsproducedbythepartialgatewaytreatmentshowthatifasignishitduringa season and not replaced until the signs are reinstalled after winter, the partial gateway

treatmentwouldstillbeofsomebenefittopedestrians.Another finding was the reduction in speed associated with the presence of the gateway

treatment.Driversbeganslowingatthedilemmazoneandweretravelling10mphslowerwhentheyentered thecrosswalkat theonly sitewhere speedwasmeasured.This speed reductionshould decrease the probability of hitting a pedestrian, as well as the seriousness of a crash

shouldoneoccur.Althoughthisstudyandpreviousstudies(9)demonstratethatthegatewaytreatmentproduced

changes indriveryieldingbehaviorat crosswalksonmultilane roads that can rival treatmentsoneandtwoordersofmagnitudemoreexpensivetoinstall,itisalsoimportanttoshowthatthistreatmentwillnotrequireexcessivemaintenanceefforts. In locationswhereplowing isnotan

issue thesignscanremainupyear round.However, inareaswheresnowremoval isan issue,the signswouldneed tobe removed inwinter.Mostof the signs testedcanbe removedandreinstalledquicklyaftertheinitialinstallation.Althoughin-streetsignsaredesignedtorebound

afterbeingstruck,signsurvivalwithmultiplestrikescouldbeanissue.


Chapter 2 examined two configurations that could increase the survival of the gateway

treatment.Thetwoconfigurationstested,whichshowedincreasedsignsurvival,onlyproducedmoderatereductions intheefficacyofthegatewaytreatment.First, installationofthesignontopofthecurbfacewasassociatedwithonlyasmallreductioninthetreatmenteffectiveness.

Signsontopofthecurbfacearelesslikelytobestruckthansignsplacedinthegutterpananddonotpresentaproblemforsweepersordrainageissues.However,thistypeofconfigurationwouldbe lesseffectiveat locationswithon-streetparkingwhereparkedvehiclescouldscreen

theviewofasignaswellaspedestrians.Onesolutiontothisproblemwouldbetoinstallacurbextension. The curb extension would prevent parked cars from screening the view of thepedestrian,aswellasallowingtheinstallationofaR1-6edgesignsontopofthecurb.

Themostvulnerableelementofthegatewayconfigurationarethesignsplacedonwhite linesseparatinglanescarryingtrafficinthesamedirection.Theuseofarobustyellowgreenflexible

delineatordevicecanaddtothesurvivalofthegatewaytreatment.ThedelineatorselectedwastestedbytheTexasTransportationInstituteandfoundtotake100strikesat60mph.Thistypeofadeviceshouldhavea long lifespanonurbanroadswithspeed limitsof35mphor less. If

thereisamedianorrefugeisland,theuseofcurbtopplacementoftheR1-6edgesignsandadelineatoratthelanelineshouldprovideatroublefreeinstallationformanyyears.

Most of the applications studied in an earlierMDOT study examined the gateway treatmentefficacy at midblock crosswalks and crosswalks at uncontrolled locations on an arterial orcollector road at the intersection of a stop-controlled road. Chapter 3 examined the gateway

treatment in a variety of new crosswalk applications, including traffic circles, roundabouts,freewayentranceramps,andtrailcrossings.Theresultsofthisstudyshowedthatthegatewaytreatmentwaslesseffectiveattrafficcirclecrosswalks,particularlyatthewiderconfigurations.

The gateway treatment was also less effective at roundabout locations, particularly atcrosswalks exiting the roundabout. The treatment at freeway entrance ramps only producedmarginalincreasesinyieldingbehavior.Thegatewaytreatmentisthereforenotrecommended

forthistypeofapplication.Onereasonwhythegatewaytreatmentwasineffectiveatthistypeofapplication is thatdriversareacceleratingas theyapproachanentrance rampandare lesslikelytoyieldright-of-waytopedestrians.Thegatewaytreatmentwashighlyeffectiveatoneof

the two trail crossingsbutonlymoderatelyeffectiveat the secondcrossing thathadahigheroperatingspeed.

Preliminarydatasuggest:

• That the gateway treatment is most effective at midblock crosswalk locations,

uncontrolledcrosswalklocationsatanintersectionwithastop-controlledroad,andtrailcrossings.

• Data also indicated that a gateway installed at an RRFB location could bring yielding

behaviortoalevelthatcomesclosetothatproducedatatrafficsignal.• Dataalsosuggestthatthegatewaymaynotbeaseffectiveonroadswithaspeedlimit

above30mphiftheADTisover15,000.


Chapter5providedpreliminaryevidenceof the long-termpersistenceofgateway increases indriversyieldingright-of-waytopedestriansovertimeandthelong-termsurvivalofthegatewaytreatment.Because the signswereonly installed for twoor threemonths, supplementaldata

will be required in order to provide clear results. However, the signs did maintain theireffectivenessinAnnArboroveratwo-monthperiodandatthesitesinthesouthwesternsideofthestateforathree-monthperiod.

Preliminary data on sign survival indicate that gateway signs mounted in the roadway oninstallablecurbbasesmaybemorerobustthanthoseinstalledflushwiththeroadway,andthat

therobustdelineator installedona lane linecansustainmanyhitsbutmaynotsurviveforanentireseasonatsiteswithhigherspeedsandADT.Atlocationswithon-streetparking,theedgesignsneed tobeplacedon theparking lane line.However suchplacementwouldmake them

very vulnerable to turning vehicle movements. Truck traffic hitting the signs likely woulddamagethesigns.Onealternativementionedabovewouldbetheuseofacurbextensionatthecrosswalk.Thiswouldallowforeithergutterpanplacement,curbtopplacementorplacement

intheroadwayattheleadingedgeofthecurbextension.Chapter6providedinformationonmaterialsandinstallationcostsforeachoftheelementsof

thegatewaytreatment.Thesedatashowthatthegatewaytreatmentisarelativelyinexpensivetreatment. Most items can be removed easily in winter and reinstalled easily in the spring.Estimatedremovalandreinstallationcostswerealsoprovided.

Although this treatment increases the percentage of drivers yielding to pedestrians atcrosswalkswithtwoormoretravel lanes ineachdirection, it isnotclearwhether itwouldbe

associatedwith a reduction in crashes at these sites.Oneway to increase the likelihood thatgatewaytreatmentreducescrasheswouldbetocombineitwithadvancestoporyieldmarkings(15,16,17,18,19).Itisalsoknownthatin-streetsignsareassociatedwithareductionofvehicle

speedatcrosswalks(7).Futureresearchshoulddeterminewhetherthefullgatewaytreatmentconsistentlyleadstoalargerspeedreductionthanasinglein-streetsign.Suchafindingwouldbeimportantbecausetherelationshipbetweenspeedreductionandreductionincrashseverity

hasbeenclearlyestablished(24).


BIBLIOGRAPHY1 NHTSA. (2011). Traffic Safety Facts 2011.National Highway Traffic Safety Administration,

Washington,D.C.

2 Zegeer, C. V., J.R. Stewart, H. Huang, and P. Lagerwey. (2001). Safety Effects ofMarkedversus unmarked crosswalks at uncontrolled locations. Transportation Research Record,1773,56–68.

3Huang,H.,&Zegeer,C.(2000)EffectsofInnovativePedestrianSignsatUnsignalizedLocations.TransportationResearchRecord,1705.43-52.

4 Turner, S., Fitzpatrick, K., Brewer, M., & Eun Sug Park, S. (2006). Motorist yielding to

pedestrians at unsignalized intersections: Findings form a national study on improvingpedestriansafety.TransportationResearchRecord,1982,1-12.

5 Ellis,R.,R.VanHouten,&J.L.Kim.(2007).“In-Roadway“YieldtoPedestrians

Signs”:PlacementDistanceandMotoristYielding.”TransportationResearchRecord,2002,84-89.

6 Gedafa, D., Kaemingk, B., Mager, B., Pape, J., Tupa, M. & Bohan, T. (2014). Impacts of

alternativeyieldsignplacementonpedestriansafety.TransportationResearchSafety,2464,DOI:http://dx.doi.org/10.3141/2464-02

7 Kannel, E. J., Souleyrette, R. R., & Tenges, R. (2003). In-Street yield to pedestrian sign

application in Cedar Rapids, Iowa. Ames, IA: Center for Transportation Research andEducation,IowaStateUniversity,http://www.ctre.iastate.edu/reports/pedyield.pdf.

8. Shurbutt, J.,VanHouten,R.,Turner,S.&Huitema,B. (2009).AnAnalysisof theEffectsof

Stutter Flash LEDBeacons to Increase Yielding to PedestriansUsingMultilaneCrosswalks.TransportationResearchRecord.No.2073,69-78.

9. Bennett,M.&VanHouten,R.(2014).AComparisonofGatewayIn-StreetSignTreatmentto

otherDriver Prompts to Increase Yielding to Pedestrians at Crosswalks Journal of AppliedBehaviorAnalysis.47,1-13.

10. Fitzpatrick,K. andPark, E.S. (2010). SafetyEffectivenessof theHAWKPedestrianCrossing

Treatment.FHWA-HRT-10-042,FederalHighwayAdministration,Washington,DC.Availableat: http://www.fhwa.dot.gov/publications/research/safety/10042/10042.pdf. Site lastaccessedJanuary5,2011.

11. VanHouten,R.,Ellis,R.&Marmolejo,E.(2008).TheUseofStutterFlashLEDBeacons toIncreaseYieldingtoPedestriansatCrosswalks.TransportationResearchRecord,2073,69-

78.

12. VanHouten,R.,LaPlante,J.&Gustafson,T.(2012).EvaluatingPedestrianSafetyImprovements.MichiganDepartmentofTransportation.ReportRC-1585.

13. City of Portland Bureau of Transportation (2010). FHWA Experimentation #4-298(E)

ModifiedHAWKSignalandBikeSignal.DraftReport


14. Godavarthy,R.P.&Russell,E.R.(2010).EffectivenessofaHAWKBeaconSignalatMid-block

Pedestrian Crossings in Decreasing Unnecessary Delay to the Drivers. TRB 89th AnnualMeetingCompendiumofPapersDVD

15. Huybers, S.,Van Houten, R. &Malenfant, J.E.L. (2004). Reducing conflicts betweenmotor

vehiclesandpedestrians:Theseparateandcombinedeffectsofpavementmarkingsandasignprompt.JournalofAppliedBehaviorAnalysis,37,445-456.

16. VanHouten,R. (1988). Theeffectsofadvancestop linesandsignpromptsonpedestrian

safety incrosswalkonamultilanehighway. JournalofAppliedBehaviorAnalysis.21,245-251.

17.VanHouten,R.&Malenfant,L.(1992).TheInfluenceofsignspromptingmotoriststoyield

50 feet (15.5 m) before marked crosswalks on motor vehicle-pedestrian conflicts atcrosswalkswithpedestrianactivatedflashing lights. AccidentAnalysisandPrevention,24,217-225.

18. Van Houten, R., McCusker, D., and Malenfant, J.E.L. (2001). Reducing motor vehicle-pedestrian conflicts at multilane crosswalks with uncontrolled approach. TransportationResearchRecord1773.p69-74.

19. Van Houten, R.McCusker, D. Huybers, S., Malenfant, J.E.L., & Rice-Smith, D. (2003). AnexaminationoftheuseofadvanceyieldmarkingsandfluorescentyellowgreenRA4signsatcrosswalksatwithuncontrolledapproaches.TransportationResearchRecord1818,p.119-

124.20. Ellis,R.&VanHouten,R. (2009).ReductionofPedestrianFatalities, Injuries,Conflicts,and

Other Surrogate Measures in Miami-Dade, Florida: Results of Large-Scale FHWA Project.

TransportationResearchRecord.No.2073,55-62.21. Van Houten, R., Malenfant, L., Huitema, B. & Blomberg, R. (2013) The effects of High

Visibility Enforcement on Driver Compliance to Pedestrian Right-of-Way Laws.

TransportationResearchRecord,2393,41-49.22. National Highway Traffic Safety Administration. (2014). Traffic safety facts 2013 data:

Pedestrians(DOTHS812139).Washington,DC:U.S.DepartmentofTransportation.

23. FederalHighwayAdministration(FHWA).(2012).ManualonUniformTrafficControlDevicesforStreetsandHighways(2009Ed.,IncludingRevision1and2).Washington,DC:Author.

24. Institute of Transportation Engineers (1989). Determining Vehicle Change Intervals:

Proposedrecommendedpractice.ITEJournal,32,126-129.25. Rosen, Erik& Sander, (2009)U. Pedestrian fatality risk as a functionof car impact speed.

AccidentAnalysisandPrevention.41,536-542.26.FHWA.(2010).EffectsofYellowRectangularRapid-FlashingBeaconsonYieldingatMultilane

uncontrolledCrosswalks.PublicationNo.FHWA-HRT.10-043,Washington,D.C.


APPENDIXASTATISTICALDATAANALYSIS

Thepurpose of this section is twofold. First, details regarding the statisticalmethods used to

analyze the outcomes of the studies are explained. Second, the outcomes themselves arepresented. Because the research was carried out at different Kalamazoo locations under avariety of conditions the research design, methods of analysis, and results are presented

separately for each of the following sites: (1) Rose Street at Kalamazoo Valley CommunityCollege(KVCC),(2)WestnedgeAvenueatRanneyStreet,and(3)RoseStreetatAcademy.

RoseStreet-KVCCSite

ResearchDesign

The data collected at the Rose Street-KVCC site were obtained under a 14-phase time-seriesdesignthatincluded68observationdays.ThesedataarepresentedinFigure1.Theconditionlabels for each phase in the 14-phase sequence are as follows: (1) Baseline, (2) Blanks, (3)

Gateway, (4)CityPost, (5)Gateway, (6)CityPost, (7)Gateway, (8)CityPost, (9)Baseline, (10)Blanks, (11) City Post, (12) Gateway, (13) Baseline, and (14) Blanks. This complex designprovideselaboratereplicationcomponentsthatallowrepeateddemonstrationsofintervention

effects.VisualAnalysis

Phases1,9,and13wereexposedtotheBaselinecondition.ItcanbeseeninFigure1thattheyieldingpercentagesareverylowbutsimilarforthesethreephases.TheBlanksconditionwas

appliedduringphases2,10,and14;eachofthesephaseshasayieldingpercentagethatcanbeseentobeapproximately30pointshigherthanduringadjacentbaselinephases.TheGatewayconditionwasappliedduringphases3,5,7,and12.Itcanbeseenthattheyieldingpercentages

duringthesephasesareover70pointshigherthanduringnearbybaselinephases.TheCityPostconditionwasappliedduringphases4,6,8,and11.Theyieldingpercentagesforthesephasesareover50pointshigherthanforbaselinephasesthatareneartheminthetimesequence.

Hence,avisualanalysisoftheresultsindicatesthatyieldingpercentagesareverylowduringallbaseline phases and that yielding is much higher for each non-baseline condition. Formal

statistical analyses that confirm these visual impressions and provide detailed quantitativeevaluationsaredescribednext.


Figure1.YieldingPercentageDuring14Phases(RoseStreet-KVCCSite–Daytime).

StatisticalMethodsModelingApproach

Atime-seriesregressioninterventionanalysisofthetypedescribedinHuitema(2011;Chapters18-21) was carried out on the 14-phase design implemented at the Rose-KVCC site. This

modeling approach begins by fitting a complex model that accommodates several types ofintervention effects and errors; subsequent analyses are then carried out to evaluate thenecessityofretainingthevariousparametersintheinitialmodelandtoevaluatetheadequacy

ofpotentiallysimplermodels.


The initial analysis involved fitting a complex (28-parameter) model; additional analyses

summarized in Table 1 were then carried out to determine if simpler models adequatelydescribethedata.

Table 1. Analyses Used in Model Identification and Evaluation (Rose Street-KVCC Site –

Daytime)

1.Model-comparisontest.ThecomparisonoftheinitialFullModel(includinglevelchangeandtrendchangeparameters)versustheRestrictedModel(includinglevelchangeparametersonly)yieldsF = 0.38 (df = 14, 40) andp = 0.97. It is concluded thatwithin-phase trend and trend

change parameters are not necessary and that the simpler (level change only) model issatisfactory.

2.Test forhomogeneouswithin-conditionphase levels. AnoverallF-test forequalityofphaselevelswithinallconditionsisrejected(p=0.04).Separatefollow-upF-tests(computedforeach

condition)revealthatthreeoffourconditionshavehomogeneouswithinconditionphaselevels.Onecondition(CityPost)hasstatisticallysignificant(p<0.04)heterogeneityofwithin-condition

phase levels,but thedegreeof theheterogeneity is insufficient to invalidatepooling. Hence,levelsamongphaseswithin the four treatmentconditionsare treatedashomogeneous in thefinalinterventionanalysis.

3. Testoftheassumptionof independenterrors. Autocorrelationamongresidualsofthefinal

analysis(a4-condition levelchangemodel) isnotpresent(r1=-0.09,Huitema-McKeantestp-value=0.98;Ljung-Boxtestonthecompleteautocorrelationfunctionthrough lag-17yieldsp-value = 0.85). Autoregressive parameters are not required in themodel because the errors

appeartobeindependent.

4. Testoftheassumptionofanormalerrordistribution. TheAnderson-Darlingtestappliedtothe residuals of the 4-condition level change model does not reject (p-value > 0.05) thenormalityassumption.Approximatenormalityisimplied.

5.Testoftheassumptionofhomogeneousofwithin-conditionvariances. Levene’stestrejects


(p<0.05)thehypothesisofvariancehomogeneity.Subsequenttestscomparingconditionlevels

are modified (using Welch-F and Games-Howell-q approaches) to accommodate thisheterogeneity.

ModelSimplification

One model simplification that is sometimes possible involves the removal of parametersdescribing trend within phases or conditions. The first test listed in Table 1 is a “modelcomparison”testthatdetermineswhethermeasuresoftrendwithinphasesarerequiredinthe

model.Becausethistestisclearlynotstatisticallysignificant(p=0.97)itisconcludedthatthedataarewellmodeledwithoutparametersdescribingtrendwithinphases.Theonlyparametersneededtodescribetheeffectsoftheinterventionsarethosethatdescribethechangeinlevel

fromonephasetothenext.A second concern in evaluating the adequacy of the initialmodel is the variation among the

phase levels within conditions. Therewere three Baseline phases, three Blanks phases, fourGatewayphases,andfourCityPostphases.Ifthelevelsofthethreeorfourphaseswithineachof these conditions are essentially homogeneous it is sensible to pool data. That is, on a

condition-by-condition basis, consolidate the data from the various phases within eachcondition.

Alternatively,ifthephaselevelswithinconditionsarenothomogeneousitmaybemisleadingtopoolthedata;todosointhiscaseistoimplythatasingleconditionisinvolvedwhenactuallyitmaybeimportanttoconsiderdifferentvariantsofacondition.Thedecisionmadewithrespect

topoolingleadstoamajordifferenceinthecomplexityofthefinalanalysisinthisexperiment.Without pooling there are 14 phases and 182 potential comparisons to be analyzed; withpooling there are just four conditions and six pairwise comparisons. This simplicity occurs

because a single parameter adequately describes a condition level regardless ofwhich phasewithintheconditionisinvolved.Forthesereasonsformalstatisticaltestsofthehomogeneityofthephaselevelswithinconditionswerecarriedout;theyaresummarizedasthesecondentryin

Table1.Theoveralltestrejectsthehypothesisthatwithinconditionphaselevelsarehomogeneous(p=

0.01). This implies that the phase levels are heterogeneouswithin at least one condition. Aseparate test of homogeneity of phase levels was then carried out for each of the fourconditions.ThesefourtestsrevealedthatonlytheCityPostconditionhasheterogeneousphase


levels. Homogeneity of the phase levels is present within the Baseline, Blank, and Gateway

conditions.Althoughstrictheterogeneityofphase levelswithintheCityPostconditionwas identified,the

more important issue isdecidingwhether theheterogeneity is largeenoughtobeofpracticalimportanceindescribingtheoveralleffectsoftheinterventions.Itturnsoutthatitisnot.ThiswasconcludedbecausethecomparisonofeachCityPostphaselevelwiththeoveralllevelsfor

the other three conditions yields the same pattern regardless of the phase chosen. That is,regardless of which City Post phase is chosen for comparison with the three remainingconditions (viz.,Baseline,Blanks,andGateway) theorderofeffectiveness isalways thesame:

Gateway>CityPost>Blanks>Baseline.AlthoughthispatternisnoteasilyobservedinFigure1,it is fairy obvious in Figure 2. Notice that the Baseline condition has the lowest yieldingpercentageregardlessofwhichofthethreeBaselinephasesisconsidered,theBlankscondition

is always second lowest, theCityPost yieldingpercentage is secondhighest regardlessof thephase,andtheGatewayconditionalwaysproducesthehighestyieldingpercentage.Thereisnoinstance in which this pattern does not hold. Because this ordering is constant for the four

phaseswithintheCityPostcondition,thesephaseswerepooled(justastheywerefortheotherconditions)toprovideasinglelevelestimateforthiscondition.

Figure2.YieldingPercentageDuringFourConditions(Rose-KVCCSite–Daytime).


FinalSimplifiedModelThemodelcomparisontestandthehomogeneitytestsdescribedabovewereappliedinorderto

justify simplifying the initial complex model. It was concluded from these tests that trendparameters were not necessary and that phases within conditions could be pooled. Thesedecisionsledtoafarsimplermodel,whichiswrittenasfollows:

where

Ytistheyieldingpercentagemeasuredondayt, β0 istheregressionintercept, β1 isthelevelchangebetweentheBaselineandBlanksconditions, D1t isthedummyvariableindicatingtheBlanksconditionondayt, β2 isthelevelchangebetweentheBlanksandGatewayconditions, D2t isthedummyvariableindicatingtheGatewayconditionondayt, β3 isthelevelchangebetweentheGatewayandCityPostconditions, D3t isthedummyvariableindicatingtheCityPostconditionondayt,and εtistheerrorofthemodel. Whereas the original 14-phase model has 28 parameters, notice that the simplified four-

conditionmodel has only four parameters. This simplification results in a more transparentinterpretationofparameters,increasedpoweroftheinferentialtests,andnarrowerconfidenceintervals.

After the simplified model was initially identified it was necessary to evaluate its adequacy.Entries 3, 4, and 5 in Table 1 focus on three issues relevant to evaluating thismodel. These

issues are all related to the assumptions regarding theproperties of theerrors of themodel.Theseassumptionsarerelevantbecausetheinferentialtestsrestonthem.

The first assumption is that the errors are independent. This assumption is violated if theresiduals are autocorrelated. It can be seen in Table 1 that the residuals of the fitted four-condition level-change model are not autocorrelated. Both the H-M and L-B tests for

autocorrelation yield very high p-values (viz., 0.98 and 0.85, respectively). This implies thatautoregressive parameters are not needed and that independence of errors of the adoptedmodelcanbeassumed.

Yt = β0 + β1D1t + β2D2t + β3D3t + ε t


The secondassumption is that thedistributionof themodel errors is normal. Table1 shows

that the Anderson-Darling test of normality does not reject this assumption; approximatenormality isaccepted. This implies thatneither transformationsof theoutcomemeasurenoralternativeestimationmethodsarerequiredforadequatetestsofinterventioneffects.

Thelastassumptionisthatofhomogeneouswithinconditionvariances.AscanbeseeninTable1 (last entry) this assumption is violated. It can be seen in Table 2 that the variance for the

Baselinecondition isconsiderablysmaller thanthevarianceassociatedwitheachof theotherthreeconditions.Forthisreasontheconventionaltestsappliedtoevaluateinterventioneffectsunder this model were modified (using Welch-F and Games-Howell-q methods) to

accommodatepotentialproblemsassociatedwithheterogeneousvariances.Insummary,allassumptionsexcepthomogeneityofvariancearemetfortheproposedoutcome

analysis. Plotsoftheresidualsofthemodel (notshownhere)confirmtheresultsof thetestssummarized above; it is concluded that the identified four-parameter intervention model isappropriateforthefinalanalysisifmodifiedteststhataccommodateheterogeneityofvariance

areused.Theresultsofapplyingthisanalysisarepresentedbelow.

ResultsMeasuresoftheeffectsofintroducingconditionchanges(Blanks,Gateway,andCityPost)atthe

RoseStreet–KVCCsiteareprovidedinthissection.Thesemeasuresarebasedontheidentifiedfour-conditiontime-series regression interventionmodel thatwas identifiedabove. Estimatesofthelevelandvarianceparametersassociatedwiththefourconditions(i.e.,Baseline,Blanks,

CityPost,andGateway)arelistedinTable2.Thelevelparametersareessentiallytheaverageyielding percentages for the four conditions. Notice that very large effects seem obvious bysimplyinspectingtheyieldingpercentagesshowninTable2forthefourconditions.

Inferentialtestsondifferencesamongtheconditionlevelsconfirmimpressionsbasedonsimpleinspection. The overall test for differences among the condition levels yields an F-ratio of

378.23andap-valuethatis<0.001.Hence,thereisnodoubtthatatleastoneconditionleveldifferssignificantlyfromtheotherthree.Pairwisecomparisonsamongthefourconditionswere

carriedouttoidentifythespecificconditionlevelsthatdifferedsignificantlyfromeachother.Table 3 lists the pairwise differences among the estimated levels, the results of inferential

multiple-comparison testson thesedifferences, and the standardizedeffect sizes.Notice thateach contrast of two conditions is associatedwith ap-value that is less than .005; hence, allpairwisecomparisonsarestatisticallysignificant.Moreimportant,thestandardizedeffectsizes

indicatethateachtwo-conditioncomparisoneasilyexceedsthedefinitionofalargeeffectsizeusingCohen’scriterion(viz.,≥.80).Similarly,verylargeeffectsoftheinterventionsarerevealed


bythe statistic,whichis0.906;thisisinterpretedastheproportionofthetotalvariationin

dailyyieldingbehaviorthatisexplainedbytheinterventionconditions.

AgraphicrepresentationoftheleveldifferencescanbeseeninFigure3asthedotsinthecenteroftheillustratedlines. Numbers1,2,3,and4inthisfigurecorrespondtothefourconditionspreviouslydescribedasBaseline,Blanks,Gateway,andCityPost, respectively. Hence,“2–1”

shown in the upper left of the figure identifies the row with a line that contains a dotcorrespondingtothedifferencebetweenthelevelofthefirstcondition(Baseline)andthelevelofthesecondcondition(Blanks).Noticethatthisdotfallsaboveapointontheabscissathatis

abouthalfwaybetween20and40;actually it isexactly30.34.(This isalsothevalueshowninTable3asthedifferencebetweenthelevelsforthetheseconditions.)Theleftandrightendsofthe lines in the figure indicate the lower and upper limits on the simultaneous 95 percent

confidenceintervals.Wecanbeatleast95percentconfidentthatthewholesetofintervalsiscorrect; that is,weareat least95percent confident that thewhole collectionof six intervalsindeed trap the true process effect (i.e., level differences) within the limits illustrated here.

Noticethatnoneoftheintervals isnearthezeroline;thismeansthatzeroisfarfrombeingacrediblevalueforthetruedifferencebetweenanypairoftrueconditionmeans.

Although theconventional confidencecoefficientof95percenthasbeenused inconstructingthe lines in this figure, themuchmore stringent confidence coefficient of 99.5 could also beused.Ifthisweredoneitwouldstillbefoundthatnoneoftheintervalswouldcontainzeroand

thesetof intervalswouldbeconsistentwith thesetof results shown incolumn3ofTable3.Notice that each p-value in this table is less than .005. This is simply an alternative way ofindicatingthateachofthesixpairwisedifferencesamongconditionlevelsfarexceedssampling

errorand,moreimportantly,representsalargeandpersuasiveeffectoftheinterventions.

Table2.YieldingLevelandVarianceforFourConditions(Rose-KVCCSite–Daytime)

Condition YieldingPercentageLevel Variance

1.Baseline 5.83 19.65

2.Blanks 36.17 96.92

3.Gateway 78.52 82.15

4.CityPost 59.82 112.96

η̂2


Table 3. Differences Among Yielding Levels, Familywise p-values, and Standardized Effect

Sizes(Rose-KVCCSite–Daytime)

Conditions

Compared

LevelDifference Familywise p-

value

Standardized

EffectSize

Blanks-Baseline 30.34 <0.005 3.38

Gateway –Baseline

72.69 <0.005 8.10

City Post -

Baseline

53.99 <0.005 6.02

Gateway-Blanks 42.35 <0.005 4.72

CityPost-Blanks 23.65 <0.005 2.64

City Post -Gateway

-18.70 <0.005 -2.08


Figure 3. Simultaneous 95PercentConfidence Intervals on LevelDifferences for EachPair ofConditions(Rose-KVCCSite–Daytime).

RoseStreetatKVCCSite(Night)Theanalysispresentedabove isbasedon thedaytimedatacollectedat theRoseStreet-KVCC

site; a second experiment was carried out at the this site at night. The nighttime data areanalyzedhere.

ResearchDesignThedesign used at night ismuch simpler than the daytime version because only four phases

(ratherthan14)areinvolvedandthenumberofdatacollectiondaysis18(ratherthan68).Datawerecollectedunderthreeconditions:Baseline(phasesoneandfour),Gateway(phase2),andCityPost(phase3).ThesedatacanbeseeninFigure4.

VisualAnalysis

ItisobviousinFigure4thattheyieldingpercentageisverylowduringtheBaselinephases(oneandfour)andthatbothexperimentalconditionsareassociatedwithmuchhigheryielding.Therankorderof theeffectivenessof thedifferentconditionsappears tobequiteconsistentwith

thefindingsreportedearlierforthemain(daytime)studyatthissite.


Figure4. YieldingPercentageUnderBaseline,Gateway,andCityPostConditions (Rose-KVCCSite–Night).

StatisticalMethodsThestatisticalanalysisofthesedataissimilartotheapproachdescribedearlierindetailforthe

daytimedata. Thenatureofthenighttimedesign,however, leadstoseveralsimplifications inthedataanalysis.

ModelIdentificationAsummaryofthetestsusedtoidentifythemostadequatemodelareshownbelowinTable4.

Tests shown in the first two entries support the use of a three-condition level-changemodelwithpooledbaselinephases. Hence,thefinalmodelhasthreeparameters:onedescribesthelevel of the Baseline condition, one describes the level change from Baseline to Gateway

conditions,andonedescribes level change fromGateway toCityPostconditions. Thismodelalso provides information needed to evaluate the difference between Baseline and City Postconditions.


Table4.SummaryofTestsUsedinModelIdentificationfortheRose-KVCCNighttimeData

1.Model-comparisontest.ThecomparisonoftheFullModel(includinglevelchangeandtrendchangeparameters)versustheRestrictedModel(includinglevelchangeparametersonly)yieldsF = 0.79 (df = 4, 10) andp = 0.56. It is concluded thatwithin-phase trend and trend change

parametersarenotnecessaryandthatthesimpler(levelchangeonly)modelissatisfactory.

2.Test for homogeneouswithin-condition Baseline phase levels. AnF-test for equality of thetwoBaselinephaselevelsisretained(p=0.37);thedifferencebetweenthetwoiseasilywithinthe rangeexpected fromsamplingerroralone.Hence, thedata fromthe twoBaselinephases

aretreatedashomogeneousinthefinalinterventionanalysis.

3. Testoftheassumptionof independenterrors. Autocorrelationamongresidualsofthefinalanalysis(a3-conditionlevelchangemodel)isnotpresent(Durbin-Watsonstatistic=1.74;H-M

statistic=1.31,p= .19;Ljung-Boxteston lag-1autocorrelationyieldsQ= .30,p-value=0.58).Autoregressive parameters are not required in the model because the errors appear to beindependent.

4. Testoftheassumptionofanormalerrordistribution. TheAnderson-Darlingtestappliedto

the residuals of the 3-condition level change model does not reject (p-value = 0.19) thenormalityassumption.Approximatenormalityisimplied.

5.Testoftheassumptionofhomogeneousofwithin-conditionvariances. Levene’stestrejects(p=0.05)thehypothesisofvariancehomogeneity.Subsequenttestscomparingconditionlevels

are modified (using Welch-F and Games-Howell-q approaches) to accommodate thisheterogeneity.

FinalSimplifiedModel

Testresultsshownasentries3,4,and5inTable4leadtotheconclusionthattheassumptionsof normality and independence of the errors are approximately met; the assumption of

homogeneous variances is not met. Consequently, the final tests for the identified three-parametermodelaremodifiedusingWelch-FandGames-Howell-qmethodsforaccommodatingvarianceheterogeneity.Theoutcomeofapplyingthesemethodsispresentednext.


ResultsAninspectionoftheconditionlevels(Table5)andthedifferencesamongthem(Table6)leaves

nodoubt that largeeffectswereobtained. The standardizedeffect sizes listed inTable6areconsistentwiththisconclusion.Inaddition,mostofthevariationintheexperimentisexplained

bytheinterventions( statistic=0.96).

Inferential tests and confidence intervals essentially eliminate the argument that either theoverallresultsoranyindividualdifferencebetweenconditionlevelsmayhaveoccurredonlyasaresultofchance. TheWelch-Fomnibustestfordifferencesamongallcondition levelsyieldsa

valueof155.23 (p< .001). Hence, theobserveddifferencesamongthethreecondition levelsaremuchtoolargetobereasonablyattributedtosamplingerror. TheGames-Howellmultiplecomparisontestappliedtoeachindividualpairwisedifferencebetweenconditionlevelsreveals

astatisticallysignificant(p<.02)effectforeachcomparison.Correspondingly, the simultaneous 98 percent confidence intervals can be seen in Figure 5.

Notethatnoneoftheintervalstrapszero;wecanbe98percentconfidentthatthewholesetofintervalstrapthetruedifferencebetweenconditionlevels.Becausenoneoftheintervalstrapszeroitcanbeconcludedthatzeroisnotacrediblevalueforthetrueeffectofanycontrast.

Acomparisonoftheresultsofthedaytimeandnighttimeresultsrevealsamarkedconsistencyinboththeoverallpatternsandthespecificoutcomeeffects.InbothcasestheBaselineyielding

level is lessthansixpercent,theGatewaylevel isthehighest(roughly60-80percent),andtheCityPostlevelisintermediateatroughly35-60percent.

Table5.YieldingLevelandVarianceforThreeConditions(Rose-KVCCSite–Nighttime)


1.Baseline 1.49 6.22

2.Gateway 59.77 111.58

3.CityPost 38.92 16.35

Table 6. Intervention Effect Estimates, Familywise p-values, and Standardized Effect Sizes

(Rose-KVCCSite–Nighttime)

Conditions LevelDifference Familywise p- Standardized

η̂2


Compared value EffectSize

Gateway -Baseline

58.28 <0.02 9.76

City Post -

Baseline

37.43 <0.02 6.27

City Post -

Gateway

-20.85 <0.02 -3.49

Figure5.Simultaneous98PercentConfidenceIntervalsonDifferencesAmongConditionLevels.Condition:1=Baseline,2=Gateway,3=CityPost.(RoseStreet-KVCCSite–Nighttime).

WestnedgeAvenueatRanneyT-IntersectionSite

ResearchDesign


The data collected from theWestnedge Avenue at Ranney T-intersection site were obtainedunderaneight-phasetime-seriesdesignthatincludes37datacollectiondays.Theeightphasesarelabeledasfollows:(1)Baseline,(2)Edge,(3)Center,(4)FullGateway,(5)CityPost,(6)Edge,

(7)Baseline,and(8)CityPost.VisualAnalysis

Aconventionaltime-seriesplotoftheyieldingdatacollectedduringeachoftheeightphasesispresentedinFigure6.Itisobviousthattherearemajorshiftsfromphasetophase.Becausethe

designhaseightphasesbutonlyfiveconditions,someconditionshavemorethanonephase.Itmay be easier to visualize the overall differences in yielding between conditions in Figure 7,becausedifferentphasesbelongingtoagivenconditionareconnectedwithlines.Forexample,

nearthebottomofFigure7thefirstandseventhphasescontainthebaselinedata;thesephasesareconnectedwithastraightline.

It is obvious that there is essentially no yielding during the baseline condition (regardless ofphase)andthatallotherconditionshavehigheryielding.Morespecifically,itcanbeseenthatthehighestyieldingisassociatedwiththeFullGatewayconditionandthatthedataforallother

conditions fall between these two extremes. It appears that the Edge and Center conditionsresulted inminor tomoderate improvements,and theCityPost conditionhadamajoreffect,although the latter effect appears to be below that of the Full Gateway condition. Formal

statisticalevaluationsoftheeffectsarepresentednext.


Figure6.YieldingPercentagebyPhase(WestnedgeatRanneySite).


Figure7.YieldingPercentagebyCondition(WestnedgeatRanneySite).

StatisticalMethods

ModelingApproachThe general modeling approach applied to theWestnedge-Ranney data is similar to the one

used for the Rose Street-KVCC site analyses. That is, a complex time-series regressionintervention model was initially estimated; tests were then used to determine whethersimplificationswerejustified,andthefinalmodelwasevaluated.Table7presentsasummaryof

thesetestsandtheassociatedresults.


Table7. SummaryofTestsUsedinModel IdentificationandEvaluation(Westnedge-RanneySite).

1.Model-comparisontest.ThecomparisonoftheFullModel(includinglevelchangeandtrend

changeparameters)versustheRestrictedModel(includinglevelchangeparametersonly)yieldsF=0.754 (df=8,21)andp=0.65. It is concludedthatwithin-phase trendand trendchangeparametersarenotnecessaryandthatthesimpler(levelchangeonly)modelissatisfactory.

2.Testforhomogeneouswithin-conditionphase levels. Thetestfortheequalityofthephase

levelswithinconditions(F=.09,df=3,29,p=.98)isretained;poolingallthedatawithineachconditionisjustified.

3. Testoftheassumptionof independenterrors. Autocorrelationamongresidualsofthefinal

analysis (a5-condition level changemodel) isnotpresent (Durbin-Watsonstatistic=2.54,p>.05;H-Mstatistic=-.41,p=.68;Ljung-Boxtestforlag11throughlag-9autocorrelationyieldsQ=6.41, p-value = 0.70). Conclusion: Autoregressive parameters are not required in themodel

becausetheerrorsappeartobeindependent.

4. Testoftheassumptionofanormalerrordistribution. TheAnderson-Darlingtestappliedtotheresidualsofthe5-conditionlevel-changemodelrejectsthenormalityassumption(p-value<0.01). Approximate normality is not present. Amodified test insensitive to non-normality is

indicated.

5.Testoftheassumptionofhomogeneousofwithin-conditionvariances. Levene’stestrejects(p<0.05)thehypothesisofvariancehomogeneity.Subsequenttestscomparingconditionlevels

aremodified(usingrankbasedWelch-FandGames-Howell-qapproaches)toaccommodatethisheterogeneity.

ModelSimplificationThe initial eight-phase analysis contains 16parameters that includemeasures of level change

andtrendchangefromphasetophase. Testswerethencarriedouttodetermineifasimpler


modelisadequateforthesedata.First,amodelcomparisontest(reportedasthefirstentryin

Table7)supportsamodelthatdoesnotincludemeasuresoftrend,becausesystematictrendswerenotpresentwithinphases.Onlylevel-changeparametersarerequiredinthemodel.

Second, a testwas carried out to determinewhether it is appropriate to pool the data fromdifferentphaseswithinconditions.TheresultsofthistestarepresentedasthesecondentryinTable7. Thistest justifiespoolingthedata.Becausetheyieldingpercentagesbetweenphases

within conditions are essentially the same, these phases were combined. That is, the twobaseline phases were combined to form one condition. Similarly, the two Edge phases werecombinedtoprovidetheoverallEdgecondition,andthetwoCityPostphaseswerecombinedto

provide the overall City Post condition. Hence, even though there are eight phases in thedesign,themodelselectedforthefinalanalysisinvolvesthecomparisonofonlyfiveconditions(i.e.,Baseline(combined),Edge(combined),Center,FullGateway,andCityPost(combined).

FinalModel

Themodelidentifiedasmostappropriateforthesedataincludesfiveparameters.Itissimilartothe four-parameter model identified for the Rose-KVCC (daytime) site (described in detailearlier), but the present model includes one additional level-change parameter. The five

parameters measure the initial level during baseline as well as change from condition tocondition.Thelastthreeentries inTable7summarizeteststhatevaluatetheadequacyofthisfive-parameterlevel-changemodel.

All three testsof independence retain thehypothesisof independenterrors;hence, themostimportantassumptionappearstobemet.Thetestsfornormalityandhomogeneityofvariance

conclude that these assumptions are not met. These problems occur because the Baseline-condition data are essentially constant. An alternative analysis (a rank-based approachincorporating theWelch-F andGames-Howell-qmethods) that is little affectedbyeithernon-

normalityorheterogeneityofvarianceswasappliedtothedata;ityieldsthesameconclusionsregarding the effects of the interventions as does the originally selected level-changemodel.Forthisreason,onlytheresultsofthelatterarereportedhere.

ResultsAn inspectionof the condition levels (Table8)and thedifferencesamong them(Table9)leaves no doubt that large effects were obtained. The standardized effect sizes listed inTable9areconsistentwiththisconclusion.Allstandardizedeffectsizes,excepttheoneforthe Edge-Center comparison, exceed the criterion for a large effect. In addition,approximately90percentofthevariationintheexperiment(asmeasuredbythe statistic)is

explainedbytheinterventions.

η̂2


Inferential statistical tests and confidence intervals essentially eliminate the argument that

either the overall results or any individual difference between condition levels may haveoccurred only as a result of chance. The Welch-F omnibus test for differences among allconditionlevelsyieldsavalueof85.88(p<.001). Hence,theobserveddifferencesamongthe

three condition levels aremuch too large tobe reasonablyattributed to samplingerror. TheGames-Howellmultiplecomparisontestappliedtoeachindividualpairwisedifferencebetweencondition levels reveals a statistically significant (p < .05) effect for each of the 10 pairwise

differences,exceptforthe7.64-pointdifferencebetweentheEdgeandCenterlevels.Correspondingly, the simultaneous 95 percent confidence intervals can be seen in Figure 8.

Notethatnoneoftheintervalstrapszeroexceptthe“3–2”contrast,whichistheCenter–Edgecomparison. We can be 95 percent confident that the whole set of intervals trap the truedifferencebetweenconditionlevels.BecauseonlytheCenter-Edgeintervaltrapszeroitcanbe

concludedthatzeroisnotacrediblevalueforthetrueeffectinthecaseofanyoftheotherninecomparisons.

Table8.YieldingLevelandVarianceforFiveConditions(Westnedge-RanneySite)



2.Edge 9.86 10.81

3.Center 17.50 67.50

4.FullGateway 89.14 250.81

5.CityPost 58.67 345.25


Table9.TenInterventionEffectEstimates,Familywisep-values,andStandardizedEffectSizes

(Westnedge-RanneySite).

ConditionsCompared Level

Difference

Familywise p-

value

Standardized

EffectSize

Edge-Baseline

9.81 <0.05 0.81

Center-Baseline

17.45 <0.05 1.44

FullGateway-Baseline 89.09 <0.05 7.37

CityPost–Baseline

58.62 <0.05 4.85

Center–Edge

7.64 >0.05 0.63

FullGateway–Edge 79.28 <0.05 6.56

CityPost–Edge

48.81 <0.05 4.04

FullGateway-Center 71.64 <0.05 5.93

CityPost–Center

41.17 <0.05 3.41

City Post – FullGateway

-30.47 <0.05 -2.52


Figure8.Simultaneous95PercentConfidenceIntervalsonthe10PairwiseDifferencesBetweentheFiveConditionLevels(Westnedge-RanneySite).1=Baseline,2=Edge,3=Center,4=Full

Gateway,and5=CityPost.

RoseStreetatAcademySite

ResearchDesign

A 19-phase time-series research design was used at the Rose Street & Academy site. Theoutcome data series consists of 87 observations (daily measures of yielding). One of the


followingsevenconditionswas ineffecteachday: (1)Baseline, (2)Narrow, (3)Blanks, (4)City

Post,(5)Wide,(6)Edge,or(7)Center.

VisualAnalysisThe yielding percentage data for the Rose Street-Academy site are illustrated in Figure 9 for

eachofthe19phasesinthisdesign.Itcanbeseenthatdatawereobtainedon87daysandtheconditionassociatedwitheachphaseisindicatedinthelegend.Itissomewhatcumbersometograsptheoveralloutcomefrominspectingthisfigurebecausetherearesomanyphases.

The outcome may be somewhat easier to comprehend in Figure 10, which identifies allobservations associated with a single condition using a line to connect the within-condition

phases. Although the observations associated with the Baseline, Blanks, and City Postconditionsarefairlyeasytotrack,thedataforsomeoftheremainingconditionsappeartoberathercommingled.

Figure11presentsboxplotsthatgreatlyclarifytherelativestandingoftheoutcomeunderthesevenconditions,althoughtheseplotseliminateinformationregardingthetimestructureofthe

data.Eachboxplothasalinerunningthroughtheboxthatindicatestheconditionmedian.Thebottomandtopportionsofeachboxcanbe interpreted,approximately,as the firstandthirdquartilesoftheindicatedconditiondistribution.

Inspectionof theseboxplotsmakes itclearthattheBaselineconditionhasthe lowestyieldingrate by far. The Narrow and Wide conditions appear to have the highest overall yielding

percentages, but the Narrow condition appears to have a somewhatmore consistent effect.Notice that the Narrow condition has less variation away from the center of the distributionthandoestheWidecondition(asindicatedbythedistancebetweenthetopandbottomofthe

box). Theother conditionsappear tohaveeffects thataremoremoderate than thoseof theNarrow and Wide conditions, but all of them are much better than the Baseline condition.These brief descriptive results are supplemented with a more thorough statistical analysis

below.


Figure9.YieldingPercentagefor19PhasesIdentifiedbyCondition(Rose-AcademySite).


Figure10.YieldingPercentageforSevenConditions(Rose-AcademySite).


Figure11.YieldingPercentageBoxplotsforSevenConditions(Rose-AcademySite).

StatisticalMethods

ModelIdentification

Asummaryof the testsused to identifyandevaluate thebestmodel for theoutcomedata ispresented in Table 10. The first and second entries in this table indicate that (1) there is noneed for trend parameters in the model (i.e., level parameters are sufficient) and (2) the

variationbetweenphasemeanswithineachconditionissmallenoughtojustifypoolingalldata(from different phases) within conditions. The implications of the pooling are that the finalmodelevaluateschangesamong7conditionsratherthanamong19phases.Thisreductionhas

majordescriptiveandinferentialadvantages;amongthemissimplicity inthe interpretationofresults.Thereare21pairwisecomparisonsamongconditions(amanageablenumber)whereasthereare171pairwisecomparisonsamongphases.Table11liststhespecificphasesincludedin

eachconditionaswellastheconditionmeans.


Thefinalmodel issimilartothemodelusedfortheanalysisoftheRose-KVCCdata,but ithassevenparameters insteadof four. The firstparametermeasures the initialBaseline levelandthe remaining six parameters measure change from one condition to the next and provide

informationneededtocompareallconditionswitheachother.

Table10.SummaryofTestsUsedinModelIdentification(Rose-AcademySite)

1.Model-comparisontest.ThecomparisonoftheFullModel(includinglevelchangeandtrendchangeparameters)versustheRestrictedModel(includinglevelchangeparametersonly)yieldsp > 0.05. It is concluded that within-phase trend and trend change parameters are not

necessaryandthatthesimpler(levelchangeonly)modelissatisfactory.

2. Test for homogeneous within-condition phase levels. Tests for the equality of the phaselevelswithinconditionshaveBonferroni-correctedp-values≥.14;thehomogeneityhypothesisisretained.Poolingdatafrommultiplephaseswithinconditionsappearsjustified.

3. Testoftheassumptionof independenterrors. Autocorrelationamongresidualsofthefinalanalysis (a 7-condition level changemodel) is not present (Ljung-Box test for autocorrelationbased on lags-1 – 22 of the autocorrelation function yields Q = 27.38, p-value = 0.20).

Conclusion:Autoregressiveparametersarenotrequiredinthemodelbecausetheerrorsappeartobeapproximatelyindependent.

4. Testoftheassumptionofanormalerrordistribution. TheAnderson-Darlingtestappliedtotheresidualsofthe7-conditionlevelchangemodeldoesnotrejectthenormalityassumption(p-

value=0.24).Approximatenormalityisassumed.

5.Testoftheassumptionofhomogeneouswithin-conditionvariances.Levene’stestrejects(p=0.03) the overall hypothesis of variance homogeneity. Consequently subsequent tests

comparing the seven condition levels are modified (using Welch-F and Games-Howell-qapproaches)toaccommodatethisheterogeneity.


Table 11. Phases Pooled Within Each of Seven Conditions and Level Estimate for EachCondition(Rose-AcademySite)

Condition Phases Pooled Within EachCondition

PooledLevelEstimate

(1)Baseline 1,4,18 6.10

(2)Narrow 2,6,8,11,15 79.71

(3)Blanks 3,5,19 30.22

(4)CityPost 7,9 58.27

(5)Wide 10,12 78.81

(6)Edge 13,17 36.24

(7)Center 14,16 52.26

FinalModelEvaluation

The finalmodelwas estimated and the residuals from it were used to evaluatewhether theunderlyingmodelassumptionsaremet.Entries3,4,and5inTable10summarizetheresultsoftestsregardingtheassumptions. Entry3describesthetestoftheassumptionof independent

errors; the results of the Ljung-Box test imply that no additional parameters are required tomeasure dependency across time. The model errors appear to be correctly specified asapproximatelyindependent.Entry4describestheresultsofthetestforapproximatenormality

of the error distribution; this assumption appears to be met. The test described in Entry 5indicatesthattheassumptionofhomogeneousconditionvariancesmustberejected.Thistestresult does not invalidate the basic model or the estimation of effect estimates, but it does

change the methods required to provide valid inferential results. That is, the conventionalmethods for estimating error variances, hypothesis tests, and confidence intervals must bemodified to appropriately accommodate heterogeneous variances. Twomethods were used

here; they are known as the Welch-F omnibus test and the Games-Howell-q multiplecomparisontest.

Figure12 illustratesseveraldifferentwaysofplotting theresidualsof themodel. Theseplotsprovide visual confirmation of the test results described above. The plot in the lower rightquadrantofthefigureillustratesthebehavioroftheresidualsacrosstime.Itcanbeseenthat,in

general,thevalueofaresidualobservedononespecificdayisnotpredictiveofthevalueoftheresidualonsubsequentdays.ThisiswhytheLjung-Boxtestforautocorrelatedresidualsisnot


statisticallysignificant.Theplotsintheupperleftandlowerleftquadrantsprovidedescriptive

information regarding the normality assumption. The upper left “normal probability” plot isused to identifydepartures fromnormality;when thedotsdepart little fromthestraight line,approximatenormalityispresent. Becausemostdotsshowninthisplotarefairlyclosetothe

straight line it appears that approximate normality is present. The plot in the lower leftquadrant, which is simply the distribution of the observed residuals, can be seen to beapproximatelynormal.Theplotintheupperrightquadrantindicatestheresidualsforeachof

thesevenconditions. Thedistance from left toright represents theyieldingpercentage. ThedistributionofdotsthatappearontheleftoftheplotaretheBaselineresiduals.Noticethatthisdistribution has less variation than do the six other distributions (i.e., the non-Baseline

distributions). These discrepancies represent the variance heterogeneity that is identified byLevene’stest.

Figure 12. Residual Plots Indicating ConformityWith or Departure From Assumptions of the

FinalModel(Rose-AcademySite).

Theconclusionsofthemodelevaluationstagearethat(1)thedataconformtotwoofthethree

assumptionsofthefinalmodeland(2)thethirdassumption(homogeneityoferrorvariances)is


violated,butalternativestatisticalmethodsthataccommodatethisproblemareappropriatefor

theinferentialaspectsoftheanalysis. Theoutcomeofthestudy,basedonthefinalmodel, ispresentednext.

Results

TheaverageyieldingpercentageassociatedwitheachconditionisshowninthesecondcolumnofTable12. TheWelch-Fomnibus test fordifferencesamong these sevenmeansproducesa

valueof233.15(p< .001). Itcanbeconcludedthatdifferencesamongthesemeanestimatesare far too large to be explained as sampling fluctuation. Most of the variation in the

experiment isexplainedbybetweenconditiondifferences( =0.89). Becausethisevidence

supports the argument that overall intervention effects exist, more specific tests (and/orconfidenceintervals)evaluatingdifferencesbetweenpairsofconditionsareofinterest.

There are 21 pairwise comparisons in this seven-condition experiment. The two conditionsassociatedwitheachcomparisonarelistedininthefirstcolumnofTable13.Forexample,thefirstentryinthefirstcolumnis“Narrow–Baseline”andthevalueinthenextcolumnis73.61;

thelatteriscomputedastheNarrowmean(79.71)minustheBaselinemean(6.10). Thethirdcolumnliststhep-valueassociatedwiththetestonthedifferencebetweenthetwoconditionmeans, and the last column lists the standardized effect size. The standardized effect size is

simplytheeffectestimate(i.e.,thedifferencebetweenconditionmeans)dividedbythepooledwithin-conditionstandarddeviation(9.84).(BaselinedatawerenotincludedinthecomputationofthispooledvaluebecausetheBaselinestandarddeviationunderestimatestheerrorvariation

presentinallotherconditions.)

AnexaminationofthefirstsixrowsofTable13revealsthateachexperimentalconditionhasanaverageyieldingpercentage that ishigher than theBaselinepercentagebyat least24points.The range for the size of the effects is 24.12 points for the least effect condition (Blanks) to

73.61forthemosteffective(Narrow).Theorderoftheobservedeffectsofthesixactiveinterventionconditionsisasfollows:Narrow>

Wide > City Post > Center > Edge > Blanks. It should be pointed out, however, that thedifferencebetween theNarrowandWideeffectsarebothdescriptively trivialandstatisticallyunimportant. Similarly, both the (Blanks– Edge) and (Center - CityPost) differences arewell

within the rangeofvariation thatcanbeattributed tosamplingerror. Thiscanbeconcludedfromthenonsignificant(ns)p-valuesshownincolumnthree.

Analternative(andperhapsmoretransparent)analysisofthesamedatacanbeseenasthesetof confidence intervals presented in Figure 13. There are 21 intervals in this figure. Theseintervalsarepresentedinexactlythesameorder(fromtoptobottom)asthetestresultsshown

η̂2


Table13. Noticethatthreeconfidenceintervalstrapzero. Thesearetheconfidenceintervals

associated with the (Narrow – Wide), (Blanks – Edge), and (Center – City Post) differences.Whenan intervalcontainszerothismeansthatzero isacrediblevalueforthetruedifferencebetweenconditionmeans.

Just as 18 of the 21 familywise p-values in Table 13 are less than 0.05 (and therefore aredeclared statistically significant), 18 of the confidence intervals in Figure 13donot trap zero.

Bothapproachesleadtotheconclusionthatthese18differencesaretoolargetobeexplainedbysamplingerror.Theadvantagesoftheconfidenceintervalapproacharethat(1)thedegreetowhichanintervaldeviatesfromzerocanbeobserveddirectlyand(2)thewidthoftheinterval

provides a clear indicationof the amountof uncertainty associatedwith themeandifferenceestimate.

The intervals provided in Figure 13 are called “simultaneous 95 percent confidence intervals”becausetheprobability is0.95that thewholesetof21 intervalscontainthetruedifferences.Becausethisprobabilityvaluereferstothewholesetofintervals,thedegreeofconfidencethat

anindividualintervaltrapsthetruevalueisactuallymuchhigherthan.95.

Table12.YieldingLevelandVarianceforSevenConditions(Rose-AcademySite)

Condition MeanYieldingPercentage Variance


2.Narrow 79.71 111.05

3.Blanks 30.22 59.00

4.CityPost 58.27 56.70

5.Wide 78.80 180.59

6.Edge 36.24 61.65

7.Center 52.26 110.50


Table13.InterventionEffectEstimates,Familywisep-values,andStandardizedEffectSizesfor21PairwiseConditionComparisons(Rose-AcademySite)

ConditionsCompared Difference BetweenMeans

Familywisep-value Standardized EffectSize

Narrow-Baseline 73.61 <0.05 7.48

Blanks–Baseline 24.12 <0.05 2.45

CityPost-Baseline 52.17 <0.05 5.30

Wide–Baseline 72.70 <0.05 7.39

Edge-Baseline 30.14 <0.05 3.06

Center–Baseline 46.16 <0.05 4.69

Blanks–Narrow -49.49 <0.05 -5.03

CityPost–Narrow -21.44 <0.05 -2.18

Wide–Narrow -0.91 >0.05(ns) -0.09

Edge–Narrow -43.47 <0.05 -4.42

Center–Narrow -27.45 <0.05 -2.78

CityPost-Blanks 28.05 <0.05 2.85

Wide–Blanks 48.58 <0.05 4.94

Edge–Blanks 6.02 >0.05(ns) 0.62

Center–Blanks 22.04 <0.05 2.24

Wide–CityPost 20.53 <0.05 2.09

Edge–CityPost -22.03 <0.05 -2.24

Center–CityPost -6.01 >0.05(ns) -0.61

Edge–Wide -42.56 <0.05 -4.32

Center–Wide -26.54 <0.05 -2.70

Center-Edge 16.02 <0.05 1.63


Figure13. Simultaneous95PercentConfidence Intervals for21PairwiseDifferencesBetween

SevenConditionMeans(Rose-AcademySite).

Thestandardizedeffect sizes listed in the final columnofTable13 for the first six rowsarguethattheeffectsoftheinterventionconditionsareofpracticalimportance.Eachofthesixvalues

easilyexceedstheconventionalcriterion(viz.,0.80)fordefininga“large”treatmenteffect.Moreimportantly,eventheleasteffectiveintervention(Blanks)hasayieldingpercentagethatis

fivetimesthesizeoftheBaselinepercentage,andthetwomosteffectiveinterventions(NarrowandWide)haveyieldingpercentagesthatareapproximately13timestheBaselinepercentage.Regardlessofthestatisticalmeasureusedtoevaluatethesizeorimportanceoftheoutcomethe

conclusionisthesame:largeeffectsaredemonstratedinthisexperiment.


Also, the last15 rows in the last columnofTable13 indicate thatall comparisonsamong theactive (i.e., non-Baseline) intervention conditions except the previously mentioned (Wide-Narrow), (Edge-Blanks), and (Center-City Post) comparisons are statistically significant and

greatlyexceedthecriterionforalargeeffect.

valuation of r1-6 gateway treatment lternatives …a series of studies evaluated the gateway...

Documents