Modelling Departure Time Choice in the Context of

1ModellingDepartureTimeChoiceintheContextofActivitySchedulingBehaviorDickEttemaandHarryTimmermans1UrbanPlanningGroupEindhovenUniversityofTechnologyP.O.Box5135600MBEindhovenTheNetherlands[July2002,RevisedNovember2002]Abstract.Thispaperreportsthedevelopmentofamodelofdeparturechoicebehaviorinthecontextofactivity-travelschedulingbehavior.Basedonabriefcharacterizationoftheliterature,somekeypropertiesofthedesiredmodelareidentified.Then,thetheoreticalframeworkisoutlinedandanoperationalmodelisderived.Next,themodelisestimatedusingactivity-traveldiarydata,collectedinthecontextoftheAlbatrossmodelsystem.Theresultsarepromising.Avenuesoffutureresearcharediscussed.1.IntroductionNetworkperformanceincongestedareasdependsnotonlyontheamountoftrafficonthenetwork,butalsoonthedistributionoftripsovertime.Therefore,modelingtripdeparturetimeisanimportanttopictounderstandandpredicthowcongestionarisesfromindividualtraveldecisions.Inparticular,thequestionishowindividualsadjusttheirdeparturetimeinresponsetocongestionoccurringonthenetworkandhowdeparturetimeisaffectedbysuchpoliciesasimprovedaccessibility,pricing,flexibleworkhoursandimprovedtrafficinformation.Theresearchthathasbeencarriedoutinthisareatodatehasfocusedonvariousaspectsofdeparturetimechoicedecisions,suchasschedulingconsiderationsregardingthemorningcommute,dailyvariabilityindeparturetimechoice,theeffectoftraveltimeuncertainty,activitydurationsandimplicationsfortheactivitypattern.Severalmodelingapproacheshavebeenusedtopredictdeparturetimechoice.Forexample[1],[2]and[3]usedadiscretechoiceframework,basedonmicro-economictheory.Morerecently,discretechoicemodelsofdeparturetimechoicehavebeenextendedtoaccountfortraveltimeuncertainty([4],[5],[6]).Otherscholarshavefocusedonduration.Forexample,Ettemaetal[7]usedahazardmodeltomodeltheprobabilityofswitchingbetweenactivitiesandofcontinuinganactivityovertime.Similarly,BhatandSteed[8]usedanon-parametricbaselinehazardtomodelthedeparturetimeofshoppingtrips,accountingforunobservedheterogeneity.Inadditiontosuchsingle-facetmodels,thereisalsoaliteratureonthecombinedchoiceofdeparturetimechoiceandactivityduration(e.g.,[9],[10]).Itreflectstheviewthatdeparturetimedecisionscannotbefullyunderstoodwithoutmodelingtheirimpacton1Papersubmittedforpresentationatthe82ndAnnualConferenceoftheTransportationResearchBoard,Washington,D.C.DickEttemaandHarryTimmermansarerespectivelyAssistantProfessorandProfessorattheUrbanPlanningGroup,EindhovenUniversityofTechnology,TheNetherlands.Phone:[phone];fax:[phone];e-mail:[e-mail],[e-mail],Joh,ArentzeandTimmermans[11]proposedamodelofactivityreschedulingbehavior,whichpotentiallycouldalsobeusedasamodelofschedulingbehavior,includingdeparturetime.Theirmodeldoeshowevernotincorporateuncertaintyandisdifficulttoestimate.Themodelpresentedinthispaperextendstheexistingtheorybydevelopingamoreintegratedframework,incorporatingtiminganddurationofactivitiesincludedinthefullactivitypattern.Thedeparturetimeofahome-basedtripandthereturntriparemodeledinmutualcoherence.Asimplifiedversionofthemodelistestedempirically.Duetotheabsenceofsuitabledataontraveltimevariabilitybytimeofday,themodelwasestimatedundertheassumptionofconstanttraveltimes.Thepaperisorganizedasfollows.First,section2presentstheframeworkthathasbeendevelopedtodescribetriptimingdecisionsinthecontextofactivityschedulingandparticipation.Then,section3describesthecalibrationofthemodelusingactivitydiarydata.Section4presentstheresultsofaseriesofsimulationsthatwereconductedtotestthepredictivecapabilityofthemodel.Finally,section5drawsconclusionsregardingtherelevanceofthetheoryandaddressessomeavenuesoffutureresearch.2.AmodelofdeparturetimechoiceTheliteratureondeparturetimechoicesuggeststhatseveralfactorsinfluencethisdecision.First,departuretimehasanimpactonthepossibilityofallocatingtimetoactivitiessuchastoachievemaximumutility.Incaseofcongestion,whichvariesbytime-of-day,travellerswilltradeofftheallocationofavailabletimetotravelortoanactivity.Forinstance,ifonedepartsatatimewithhighercongestionlevels,travelitselfwilltakelonger,andlesstimeremainsforparticipationinactivitiesresultinginashorteractivityduration.Ifmarginalutilityisnotdependentontimeofday,theutilityderivedfromactivityparticipationwilldecrease(e.g.[10]).Secondly,departuretimewillaffecttheopportunitytoparticipateinactivitiesthatarerestrictedbyclock-time.Forexample,ifonedepartstoworkveryearly,itisnolongerpossibletohavebreakfastwithone’sfamilymembersiftheydonotadjusttheirschedules.Also,whenarrivingtoolateatwork,itmaynotbepossibletoparticipateincertainworkactivities,whicharescheduledearly.Following[1],wetermtheseeffectsschedulingcosts:costsofdeviatingfromone’spreferredschedule.However,schedulingcostsusuallyrefertotheactivity,conductedatthedestination.Ourexampleillustratesthatschedulingcostscanalsooccurattheorigin.Infact,schedulingcostswilldependonthemarginalutilit