An Evaluation of Pittsburgh's Dynamically‐Priced Curb Parking Pilot

Abstract

This paper analyzes a small, dynamically‐priced curb parking pilot that took place in Pittsburgh, PA from 2013 to 2015. Dynamic pricing of curb parking is a recent innovation – one which is designed to manage parking congestion through price manipulation in order to optimize occupancy and reduce traffic congestion. I find that prices declined during the pilot, revenues rose and occupancies moved towards target ranges set by program administrators. In the scant few studies of such pricing schemes, disagreement has arisen amongst scholars as to whether elasticities are appropriate to measure the effect of price changes on driver behavior. This paper demonstrates the use of the regression discontinuity statistical model in estimating the effects of price change effects on driver behavior. Regression discontinuity models suggest that prices had the intended effect on driver behavior during the Pittsburgh pilot, but such effects took a couple weeks to develop.

Full text






AnEvaluationofPittsburgh’sDynamically‐Priced
CurbParkingPilot


PresentedtotheFacultyoftheSchoolofDesign
ofthe
UniversityofPennsylvania

Inpartialfulfillmentoftherequirementforthe
degreeofMasterofCityandRegionalPlanning

By

MichaelA.Fichman

Philadelphia,Pennsylvania
May6th,2016

Fichman,2

Abstract
Thispaperanalyzesasmall,dynamically‐pricedcurbparkingpilot
thattookplaceinPittsburgh,PAfrom2013to2015.Dynamic
pricingofcurbparkingisarecentinnovation–onewhichis
designedtomanageparkingcongestionthroughprice
manipulationinordertooptimizeoccupancyandreducetraffic
congestion.Ifindthatpricesdeclinedduringthepilot,revenues
roseandoccupanciesmovedtowardstargetrangessetbyprogram
administrators.Inthescantfewstudiesofsuchpricingschemes,
disagreementhasarisenamongstscholarsastowhether
elasticitiesareappropriatetomeasuretheeffectofpricechanges
ondriverbehavior.Thispaperdemonstratestheuseofthe
regressiondiscontinuitystatisticalmodelinestimatingtheeffects
ofpricechangeeffectsondriverbehavior.Regression
discontinuitymodelssuggestthatpriceshadtheintendedeffecton
driverbehaviorduringthePittsburghpilot,butsucheffectstooka
coupleweekstodevelop.

Introduction
InOctoberof2014,Pittsburgh’sCityCouncilauthorizedthePittsburghParkingAuthority
tousea“dynamicpricingmodel”tosetparkingpricescity‐wide.Theperceivedsuccessofa
smallpilotundertakenstartinginJanuary,2013wasthebasisforPittsburgh’scity‐wide
legislation(Zullo2014).Intheory,thedynamicpricingmodelworksasfollows:parkingdemand
andsupplywillequilibratesothattherearealwaysatleastafewspacesavailableoneach
block,minimizingthetimeittakestofindparkingandminimizingcongestionandpollution
(Shoup2004).PittsburghjoinsSanFrancisco,LosAngelesandafewothersinusingthisnew
wayofpricingstreetparking.Withfewpilotscompleted,andfewerstudiedintensely,more
informationisneededtoallowimplementationofdynamic,demand‐basedparkingwitha
thoroughunderstandingofprobableeffects.Dotheprogramsworkasintended?Dotheyvary

Fichman,3

fromplace‐to‐place?Howisdemandaffectedbypricechanges?Howfrequentlyshouldprices
change?Byhowmuch?
Thisstudyuncoversmoreevidenceoftheeffectsofmarket‐basedcurbparkingschemes
byexaminingPittsburgh’spilotprogram.Iconcludethatthepilotworkedasintended‐overall
revenuesrose,occupancymovedtowardstargetrangesonmoststreetsandpricechangeshad
theintendedimpactondriverbehavior.Drivers’generalpricesensitivitywassimilartothat
measuredinotherstudies.Furthermore,thisanalysisdemonstratestheutilityofthe
regressiondiscontinuity(RD)statisticalmodelinanalyzingthecausaleffectsofpricechangesin
theabsenceofrobustcontrols.Furthermore,IusetheRDmodeltoshowthattheeffectof
pricechangesondriverbehaviorwerelargerseveralweeksafterthechangestookplace–
suggestingalaginconsumerresponsetopricechanges.However,therobustnessofthese
findingsisdiminishedbythefactthatapplicationofpricechangesbyadministratorswas
haphazardandresultedindecreasedsamplessizesinRDmodels.
Bycomparingandcontrastingthedynamicsofthispilotprogramtothoseofother
pilots,specificallySanFrancisco’sSFPark,amorenuancedpictureofdynamic,demand‐based
parkingpricingbeginstoemerge.Thisinformationisthenusedtoformthebasisforsome
observationsaboutgeneralizabilityandprogramdesignandadministration.Iconcludeby
offeringrecommendationstothoseplannersinterestedincreatingandoperatingdemand‐
based‐pricedparkingprograms.
Cruising‐theimpetusfordemand‐basedpricingofon‐streetparking
Fordecades,planningscholarshavearguedthatcurbparkingtendstobeunderpriced
andthereforeovercrowded.DonaldShoupsummarizedthehistoryofliteratureonthetopicin

Fichman,4

histomeTheHighCostofFreeParking(2004):Whencurbsideparkingspacesareovercrowded,
andvacantspacesareunavailable,drivers“cruise”insearchofparking.Cruisingcreates
negativeexternalitiesincludingcongestion,airpollutionandunnecessaryfuelconsumption.
Furthermore,congestionpromptedthecreationofminimumparkingrequirementsformost
typesofzoningdistricts.Shouparguespersuasively(inHighCostandelsewhere)thatthese
minimumsareinefficient(Shoup1999)(PierceandShoup2013).
Invariousstudiesundertakenduringthepast90years,cruisinghasbeenobservedto
comprisealargeproportionofoverallcitytraffic.Theaveragepercentageofcruisingasa
proportionofoveralltrafficobservedin10studiesfrom1927‐2011was34%(Shoup2004).As
longagoas1954,Nobel‐winnerWilliamVickreysuggestedthatcurbpricescouldbe
manipulatedinordertosuppressorstimulatedemandwiththegoalofensuringenough
parkingavailabilitytoeliminateordecreasecruising(Vickrey1954).Trafficengineershave
determinedthatapproximately80‐85%averageoccupancyrepresentsanoptimalpointat
whichonespaceperblockshouldbeavailable(Shoup2004)(Millard‐Ball,Weinbergerand
Hampshire2014).ArnottandIncicreatedseveraltheoreticalmodelsofparkingandcongestion
(2006).Theirmost“robust”modelsuggestedthat“itisefficienttoraisetheon‐streetparking
feetothepointwherecruisingforparkingiseliminated.”Thealternativewouldbetoincrease
theamountofparkingtoachievethesameeffect–aninefficientsolutionforreasonsexplained
byShoup(2004).Parkingminimumscreateseriouslandmarketdistortionsandpromotesprawl
andautomobile‐centricdevelopment.Brooke,IsonandQuddus(2014)surveyedstreet
parking‐relatedliterature.Theysuggestthatthereisagreementthatsettingadjacenton‐and
off‐streetparkingtodemand‐basedpricesinconcertcouldeliminatecruising.

Fichman,5

Surveyofexistingpilotprogramsandrelatedresearch
Theemergenceofdigitalparkingmetersnowallowsfordemand‐basedprice
manipulationtoachievedesiredoccupancyorvacancylevels.Furthermore,sincemodern
kiosksacceptcreditcards,pricescanberaisedtolevelswhichwouldhavepreviouslyrequired
driverstocarryonerousamountsofcoinage.Recently,therehavebeenseveralfederally
fundedpilotprogramsdemonstratingthedynamicpricingconcept(USDepartmentof
Transportation2015).Duetothenoveltyofsuchprograms,researchregardingtheeffectsof
demand‐basedpriceregimesisscant.
Atpresent,academicliteratureappearstobelimitedtothecasesofSanFrancisco,
SeattleandLosAngeles.Pittsburgh’spilotprogramwasconductedbytwoCarnegieMellon
professorsasademonstration,andasofthiswriting,hasnotbeenanalyzedinapeerreviewed
study.
Thefourpilotsdifferedinsmallways.Theyeachuseddifferentheuristicoccupancy
targets,pricechangeintervalsandpriceincrements.Somecitiesusedtechtoolsto
communicateratestodrivers.Somevariedpricebytimeofday.Somecitieshadsensorsto
measureoccupancy,someusedestimatesbasedontransactiondata.Theparametersand
characteristicsofthesefourpilotprogramscanbecomparedinFigure1.


Fichman,6


CityTarget
Occupancy
Price
Change
Time
Interval
Price
Increment
($/hr)
Time‐o
f
‐
Day
Pricing?
Consumer
Price
Information
App?
Measured
or
Estimated
Occupancy
SanFrancisco*60‐80%10
changes
(2011‐13)
$0.25 Yes YesMeasured
LosAngeles**70‐90%Nomore
thanonce
permonth
$1(or
150%
baserate)
Yes YesMeasured
Seattle+71‐86%Nottruly
“dynamic”
‐onlyone
change
Variable
bydistrict
No No Estimated
Pittsburgh++60‐80%Nomore
thanonce
permonth
$0.25 No No Estimated
Figure1.Characteristicsofselecteddemand‐basedpricingpilotprogramsintheUS.Alloftheseprogramsexcept
Pittsburgh’shavebeenanalyzedinpeer‐reviewedliterature.Characteristicsofparticularprogramsreflectonlythe
periodsunderstudy.
Informationisdrawnfromthefollowingsources:
*(Millard‐Ball,WeinbergerandHampshire2014),(PierceandShoup2013);**(Ghent2015);
+(Ottosson,etal.2013);++(Fichman2015)

Thesepilotshavebeenstudiedwithvaryingdegreesofrigor.Seattle’spilothasbeen
econometricallyanalyzedinthepeerreviewedliterature(Ottosson,etal.2013),butwasnot
truly“dynamic”asithadonlyonepricechange.TheOttosson,etal,studyofSeattle’s2011
demand‐basedpricecorrectionswasdoneinordertoopportunisticallyexploitanatural
experimentbroughtaboutbyapricechangewhichappliedtowholeneighborhoods.Muchof
theheterogeneityamongstthesediverseneighborhoodsdoesnotappeartohavebeen
persuasivelycontrolled.LosAngeles’LAExpressParkpilothasnotyetbeensubjecttoan
econometricanalysis.SanFrancisco’sSFParkprogramisthemostthoroughlystudiedandmost
programmaticallysophisticatedoftheseprogram.ThePittsburghpilotisinmanyways
influencedby,andsimilartotheSFParkdesign.ThisanalysiswillthereforediscussonlySFPark
andthePittsburghpilotindetail.

Fichman,7

SFPark–design,studyandcriticism
In2011,theSanFranciscoMunicipalTransportationAuthority(SFMTA)adopteda
pricingprogramknownasSFPark.Thisprogramincorporatedthedynamicpricingstrategies
advocatedbyShoupandVickrey.SFParkwasfundedinpartbytheUSDepartmentof
Transportation(USDepartmentofTransportation2015).Thegeneralsetupoftheprogramwas
asfollows:
‐256on‐streetpilotblockswereestablishedincommercialcorridors,some
blocksweresetupascontrols

‐Onpilotblocks,pricesweresetforthreedifferent“timebands”andpriced
differentlyforweekendsandweekdays

‐Whenaverageoccupancyratesonagivenblockfellbelow60%,priceswere
lowered$0.25,whenratesroseabove80%,rateswereraised$0.25.Rateswere
changed10timesduringa2011‐2013study(Millard‐Ball,Weinbergerand
Hampshire2014).

ThereisdisagreementamongstscholarsabouthowSFPark’spricesaffecteddriver
behaviorandwhethertheprogrambroughtaboutoptimaloutcomes.A2013paperbyPierce
andShoupreportedthatparkingdemandinSFPark’sfirstyeartendedtoberelativelyinelastic
butvariedbylocation,time,day,initialpriceandsizeofpricechange.Occupancyonover
2/3rdsofblocksoverorunder60‐80%occupancymovedintothetargetrangeinYear1(Pierce
andShoup2013).SuchananalysisshowsSFPark’spricingregimequicklybringingparking
occupancyintosociallyoptimalequilibrium.Subsequently,severalpapersemergedthat
paintedalessrosypicture.
ChatmanandManvilleconductedastudywhichloggedrepeatedobservationsofSFPark
blocksandfoundthatalthoughoccupancyfellaspricesrose,overalllevelsofavailability(the
timeduringwhichaspaceisavailableonagivenblock),didnotincrease(Chatmanand

Fichman,8

Manville2014).Thisperhapspointstoaweaknessintheoccupancy‐drivenheuristicbasisof
demand‐basedpricingmodels.
Millard‐Ball,WeinbergerandHampshirepublishedtwopapersaboutSFPark–bothof
whichwerecriticalofPierceandShoup’sresearch.Thefirst,publishedin2014in
TransportationResearchPartA,concludedthatratechangesachievedtheSFMTA’soccupancy
goaloverthecourseoftwoyears.However,theyalsoassertedthatPierceandShoup’s(2013)
one‐yearelasticitieswereoverstatedbecauseofafailuretoaddressendogeneityrelatedto
pricechanges.Essentially,theysuggestthatthecircularrelationshipbetweenprice
manipulationanddemandbiasesanymodelusedtoderiveelasticities.Subsequently,Millard‐
Ball,WeinbergerandHampshire(2014)submittedapublishedcommentinresponsetothe
PierceandShoup,publishedintheJournaloftheAmericanPlanningAssociation(JAPA).They
expandedontheirpreviousassertionthatendogeneityispresentdynamic‐pricingelasticity
estimations,addinganassertionthatthechangingconditionswhichdrivetraffictooneplaceor
anotherintroducemorebiasintothesemodels.Subsequently,theycomparedtheone‐year
elasticitiescalculatedbyPierceandShoup(2013)usingtheSFParkdatatoarandomsimulation
wherepricewasdesignedtohavenoimpactondemandandobservedthatPierceandShoup’s
findingswere“largelyspurious,causedbythestatisticalphenomenonofreversiontothe
mean.”(Millard‐Ball,WeinbergerandHampshire2014)
OneoftheprimaryweaknessesintheSFParkdesignwasthelocationofthecontrol
blocks.Mostofthe“control”areasareincompletelydifferentpartsofthecityfromthepilot
areas,leadingtothepossibilityofspatialvariationsthatcannotbeeffectivelycontrolled.For

Fichman,9

example,thedowntowncoreandtouristyFisherman’sWharfarehardtocomparetocontrol
areasinlow‐riseresidentialInnerRichmond.
Millard‐Ball,WeinbergerandHampshire(2014)useda“fuzzy”regressiondiscontinuity
modeltodisaggregatedatabygeographicareaandbyindividualrateadjustment.Theirresults
confirmedtheirskepticismthatindividualpricechangesinSFParkwereinfluencingdemandin
theshortterm.TheRDdesignhaspracticalapplicationforanalysisofratechanges,particularly
inPittsburgh’spilotarea,wheretheprimarypurposewastodemonstrateatooltopolicy
makers,notcreateaniron‐cladexperimentaldesignandapplicationoftreatmenteffectswas
uneven.
AsaresultofMillard‐Ball,WeinbergerandHampshire’sdemonstration,Idecidedtouse
the“sharp”RDmodeltoevaluatetheeffectsoftreatmentinPittsburgh.ThedetailsoftheRD
designarediscussedtowardstheendofthefollowingsection.
PittsburghData,EmpiricalSettingandMethods
DescriptionofPilotProgram
In2012,theCityofPittsburghraisedparkingratesinareasadjacenttoCarnegieMellon
University(CMU)to$2/hour,doublingthepreviouspriceanddrivingparkingoccupancytolow
levelsadjacenttotheuniversity.Thispricechangeprompteddistortedparkingbehavior,as
commutersbeganparkinginfar‐flungresidentialareasandbeganleavingCMU’scurbsides
under‐occupied(Fichman2015).InSeptemberof2012,theCityinstalledelectronickiosks,
meaningthatpricescouldnowbemoreeasilymanipulatedanddriverscouldpayforparking
usingcreditcards,obviatingtheneedtocarrylargequantitiesofcoinage.IntheFallof2012,
StevenSpearandMarkFichman,professorsattheTepperSchoolatCMUpersuadedtheCityof

Fichman,10

Pittsburghtoletthemadministeradynamic,demand‐basedpricingpilotonfivestreetsnear
theUniversity.Unlikethelarge‐scaleSFParkandLAExpressParkpilots,thiswasnotaFederally
fundedproject.ThepilotbeganinJanuary,2013andranuntilDecember,2015.Thepurposeof
thepilotwastodemonstratetotheCitythatoccupancy‐drivenpricingwouldproducebetter
outcomesforbothconsumersandgovernmentthantheCity’ssingle‐price,time‐limitedregime
(Fichman2015).Inlate2014,PittsburghCityCouncilenableddynamicpricingcity‐wide
startinginFebruary,2015(Zullo2014)–spellingthebeginningoftheendforthe
demonstrationpilot.
Figure2.StudyArea

Fichman,11

Fivestreetswereincludedintheoriginalpilot‐TechStreet,MargaretMorrisonStreet,
SchenleyDriveandtwosectionsofFrewStreet(oneknowninthisstudyas“Frew5001”forits
solemeterID).Becausethepilotwasdesignedfordemonstrationandnotspecificallyfor
empiricalstudy,therewereno“control”metersorstreetsinthepilotarea.However,there
weresomefactorsthatmadetheareaeasytostudy.Mostpilotmeterswerenotadjacentto
anyotheron‐streetmeters.Thelandusesurroundingthemeterswasalmostexclusively
institutional,providingareasonableamountofspatialhomogeneitytothestudyarea(Figure
2).Theparkingandtrafficintheareaarealmostentirelyrelatedtotheuniversity–nearby
residentialareashaveamplestreetparking.SchenleyDrivepresentstheexceptiontothis
generalhomogeneity.SchenleyDrivebordersa3‐hourtime‐limitedparkingareaaswellas
somefreeparkingareaswhichareusedbytouristvisitorstoPhippsConservatory.Itcanbe
Figure3.Monthlyratesforeachstreetinthepilot.Thedashedlinerepresentsthe$2pricewhichthe
PPAestablishedinJanuary,2012.

Fichman,12

arguedthatSchenleydoesnotbelonginthestudy,butitmostcertainlyrepresentsapartofthe
CMUparking“market”anditsidiosyncrasiescanbecontrolledusingfixedeffects.
ThePittsburghpilotwasinspiredbytheparametersofSFPark,butoperatedundera
morefluidsetofconditions.Thecityinitiallylimitedpricechangesto$0.25,withnomorethan
onechangepermonth.Eachstreethadonlyoneprice,allday,eachdayforamonth.Parking
wasmeteredforthehours8AM‐6PM.Astimewenton,theadministratorsmadelargerprice
changesattheendoftheschoolyear.After$0.25reductionsseemedtooslowtocorrespond
withdrasticallylowersummer‐timedemand,theadministratorsbeganusinglargerdecreases
attheendoftheschoolyearand$0.50increasesattheendofthesummer.Atimelineofprice
changesforallfivestudystreetscanbefoundinFigure3.Onenotableidiosyncrasyinthe
pricingregimeisthatthe5001sectionofFrewStstartedoffasahigher‐priced“Premium”
parkingarea,butunderthatpricingregimeitsawoccupancyinthe10‐20%range.Thatblockis
extremelysteepandperhapsseenasanundesirableparkinglocation.Administratorsswitched
itovertoan“Economy”pricingregimebylate2013–droppingitspricefrom$2to$1.
Theadministratorsperiodicallyestimatedgrossandmeandailyoccupancy,using
samplesofmid‐weekdata,andchangedpricesbasedinpartonastatedgoaltokeepmid‐
week,school‐yearoccupanciesbetween60‐80%.Theyalsochangedpricesinreactionto
repeated“maxingout”atcertainlocations(Fichman2015).

Fichman,13

Signsindicating“Economy”and“Premium”parkingzoneswerepostedtodifferentiate
higherandlowerrateareasandbygraphicdisplaysattheelectronickiosks,howeverdriversdid
notlearntheactualpriceuntilvisitingthecurb‐sidekiosk(Figure4).Thisisanotabledifference
withtheconsumerexperienceofSFPark,wheredriverswereabletoexaminepricesusingan
internetapplication.AdriverinPittsburghwhodidn’tlikethepricepostedatakioskhadto
decidewhethertopayorreturntothecartofindotherparking.Giventhesecircumstances,it
ispossiblethatsomedriversmadeaquickdecisiontopayanundesirablepriceandusetheir
newfoundinformationaboutpricesontheirnextvisit.Itisalsopossiblethatsomedriversre‐
enteredtheirvehiclesandcontinuedtocruiseforparking.
Figure4.Signageandconsumer‐facinginformationatthestudysite

Fichman,14

Data
TheCityofPittsburghandtheCMUadministratorscollectedadatasetthatconsistedof
point‐levelkiosktransactionsfortheentirecityandconveyedthedatasettomeuponrequest.
Thecity‐widedatasetwhichwasusedforthisstudyconsistedof13,348,153pointlevel
observations.Eachdatapointcontainedthefollowinginformation:purchasetimeanddate,
minutespurchased(units),amountpaid,kioskID(indicatingStreetLocationanduniqueID),
paidintervalstartandend,maskedpayerIDandsomeextraneousinformation(Figure5).The
Cityalsoprovidedthelatitudeandlongitudeofallkiosks.
ItemNameExample
PurchaseDateLocal11/21/20144:51
Terminal‐TerminalID410153‐MARMOR5103
PayUnit
–
NameCard
TicketNumber12241
Amount8
Units514
MaskedPAN443057*8412
PayIntervalStartLocal11/21/20148:00
PayIntervalEndLocal11/21/201414:24
TariffPackage
–
NamePgm42
ArticleNameArticle1
NodeOakland
TariffPackage
–
ID42
OccupancyEstimation
Iestimatedoccupancybykioskinfifteenminuteintervalsthroughouteachdayofthe
studyperiod.Ithencoalescedthesekioskestimationstothestreet‐level.It’snecessarytodeal
withoccupancyatthestreetorblocklevelbecauseA)thatisthelevelatwhichpricesare
determinedandB)becauseadrivercanpayatanykioskonagivenstreet‐eachindividual
kioskmaylogaparkingobservationfromanywhereonthestreet.Mymethodsweredesigned
Figure5.Sampleofrawdata

Fichman,15

toapproximatetheapproachoftheprogramadministrators,soastounderstandprogram
dynamicsfromasimilarstandpoint.
Therewerenumerousconsiderationsandassumptionsnecessarytoundertakethese
estimations.UnlikeSFParkandLAExpressPark,Pittsburghdoesnotusesensorstodetermine
actuallengthofstay.Iassumedthatonaverage,acaroccupiesaspotforaslongastimehas
beenpurchased.Perhapsthisisunreasonable,butforlackofanyotherinformationabout
behaviorrelatedtodepartures,thisisanecessaryassumption.
Itwaschallengingtoestimatemaximainordertoestimatepercentageoccupancies.
WiththeexceptionofstretchesofFrewStreet,eachstreetinthestudyareahasparallel
parking.Parallelparkingareascanaccommodateavaryingnumberofvehiclesdependingon
vehiclesizeandspacingonagivenday.Isurveyedthestudysiteandobservedparallelparking
behavior.Iconsideredthisinformationalongsidestreetlengthmeasurementsdoneusing
GoogleEarthandestimatesofaverageparallelparkingspacelengthdescribedinother
Pennsylvaniadesigncodes(PhiladelphiaMayor'sOfficeofTransportationandUtilities2009).
Pittsburgh’szoningandsubdivisionregulationsdonotspecifythelengthofanon‐streetspace
(CityofPittsburgh2015).Iestimatedaparallelparkingspacetobeapproximately18feetin
lengthanddeterminedparkingmaximaalongeachstreetincludedinthestudyarea(Figure6).
Street(PseudonyminDataSet)MaximumOccupancy PriceRegime
FrewSt.(FREWST)170 Premium
FrewStWest(FREW5001) 14 Economy
TechStreet(TECHST)34 Premium
SchenleyDrive(SCHEDRCMU)107 Economy
MargaretMorrisonSt(MARMOR)54 Economy
Myalgorithmicestimationofoccupancy,writtenusingthestatisticalsoftwareR,
functionedroughlyasfollows:
Figure6.Characteristicsofstreetsinstudyarea

Fichman,16

1.Foreachmeter,foreachday,a96‐itemvectorrepresentsthe96fifteen‐minuteperiodsof
theday.Considereachiteminthevectorasbeinga“bin.”
2.Eachtimeadriverlogsatransaction,dividethenumberofminutespaidbyfifteen.Consider
eachincrementoffifteenminutestobea“token.”Let’ssayadriverpurchasesonehourof
parking‐thatwouldbefourtokens.
3.Giventhestarttimeofthetransaction,roundedtothenearestfifteenminutes–dropthe
firsttokeninthebinthatcorrespondstothattimeandoneineachsuccessivebinuntilthere
arenomoretokens.Soifadriverbuysonehourofparkingat8AM,heisawardedfourtokens‐
hedropsatokenintothe8:00‐8:14bin,oneinthe8:15‐8:29andsoon,untilhe’soutoftokens.
4.Repeatthisprocessforeachtransaction.
5.Sumthelikebinsforallmetersonastreetfortheday.
6.Calculatethepercentageoccupancyineachstreetbinrelativetothemaximumoccupancy,
takethemeanpercentageoccupancyforeachstreet‐dayforthehours8AM‐6PM.Ifthisfigure
isabove100%(asitcanbegiventhefactthatmultipledriverscanpayforaspaceshouldone
leavebeforehistimeisup),assigna100%occupiedor“maxedout”valuetothatday.These
estimationsareeasilyinterpretableviaahistogramwhichshowstheoccupancyofparkedcars
onagivendayshowninFigure7.

Fichman,17

AdditionalDataCleaning
Thepilotprogramwasdesignedtousepricingtomanipulateoccupancyduringmid‐
weekperiodsduringtheschoolyear.Figure8showstheaveragedailyoccupanciesforFrewSt
throughoutthestudyperiod.Itisnoticeablethatduringthesummerandwinterbreakperiods,
occupancyislower–withmanydayslogginglittletonooccupancy.Similarly,thereweredays
duringtheschoolyearwithalmostnooccupancy.Allstreetssharedtheseseasonal
characteristics.Recallthattheconditionsofthepilotonlyallowedfor$0.25pricechanges,one
pricechangepermonth.Despitethefactthatthepilotimplemented$0.50pricechangesat
thebeginningandendofsummerin2014,Ireasonthatthesemonthsofextremelylow
Figure7.EstimatedoccupancyforFrewStreet,September10th,2014

Fichman,18

demandweresubjecttoconditionsinappropriateforsuchaconstrainedpricingregime.Given
theseconditions,IdecidedtoexcludealldatapointsfromthemonthsJune,JulyandAugust
andallSaturdayandSundayobservationsfromregressionmodelsratherthanjustcontrolling
forthem.Theremainingdatasetcontains3840street‐dayobservations.711ofthesefall
withineither30daysbeforeor90daysafteraschool‐yearpricechange.
VariableNameDescriptionExample
streetStreetNameFixedEffect FREW5001
weekdayDayofWeekFixedEffect Monday
dateDateusingPOSITXtimeconvention 2014‐10‐20
off_sessionNoClassesFixedEffect(1for“noclass/exams”) 0
monthlyPricePriceChargedinUSD 1.00
monthDumbMonthFixedEffect OC
T

yearDumbYearFixedEffect YEAR13
avgOccupancyMeanDailyGrossOccupancy 7.6
pctTrueOccupancyMeanPercentageOccupancy 54%
prevOccMeanMid‐WeekOccupancyinPreviousMonth 86%
trueMaximumMaximumBlockOccupancy 14
tMonthPriorToPriceChangeFixedEffect 0
t_plus_1Month1AfterPriceChangeFixedEffect 0
t_plus_2Month2AfterPriceChangeFixedEffect 0
t_plus_3Month3AfterPriceChangeFixedEffect 0
Figure8.AveragedailyoccupanciesonFrewSt.Dashedlinesrepresent60and80%estimatedmeandaily
occupancylevels.RedobservationsrepresentdayscodedashavingnoclassorexaminationsatCMU.
Figure9.Sampleofrawandtransformeddatausedtoestimateregressionmodels

Fichman,19

IalsoexaminedCarnegieMellon’spubliclyavailableacademiccalendarsfortheyears
includedinthepilotstudyandcodedallstreet‐dayswithoutclassorexaminationswitha
dummyvariabletocontrolfor“offsession”universitybusinessinregressionmodels.Figure9
showsasampleofcleanedandtransformeddatausedintheregressionmodelsinthisstudy.
EstimationofPriceElasticitiesofDemand
InordertodeterminetheeffectofpricechangesIundertooktwosetsofregression
models–first,anOrdinaryLeastSquares(OLS)regressiontoestimatetherelationshipbetween
priceandoccupancywhilecontrollingforseveraltimeandlocationalvariables.Equation1
describestheOLSmodel.



⋯

Equation1.OLSmultipleregression

ThenullhypothesistestedbyanOLSmodelisthatthecoefficientindicatingthe
relationshipofpricewithdemand(occupancy)isstatisticallyindistinctfromzero.The
alternativehypothesiswouldbethat“allelseequal,”thereisastatisticallysignificantchance
(p<0.05)thatthemagnitudeofthevalueassociatedwiththeindependentvariable“price”is
notzero.ThecoefficientsusedtoestimateregressionsareshowninFigure9.
SeveralassumptionsmustbemetinorderforOLStocorrectlyfitamodeland
determinethesignificanceofthevalueassociatedwithprice.First,theindependent
variablesshouldhavenostrongmulti‐collinearity.Multi‐collinearitycanbeassessedby
measuringvarianceinflationfactors(VIF)usingthe“car”packageforthelanguageR.Second,
therelationshipbetweenthedependentandindependentvariablesmustbelinear.Third,the
errortermsinthemodel(thedifferencebetweentheobservedpointsandthemeanfunction)
mustbenormallydistributed.Iassumedmodelerrortermstobeapproximatelynormally

Fichman,20

distributed.Lastly,thevarianceortheseerrortermsshouldbeuncorrelatedwiththeother
variables(homoscedastic).
IestimatedthepriceelasticitiesofdemandDforpricePandquantityQatpointEusing
Equation2,whichincorporatesthecoefficientassociatedwithpricefromEquation1.
ED  ΔQ/ΔP ∗ P/Q   ∗ P/Q
RegressionDiscontinuityModels
Millard‐Ball,WeinbergerandHampshire(2014)arguethatthetwo‐wayrelationship
betweenprice‐changesandoccupancy(e.g.highoccupancylevelsinducingpricechangesand
pricechangesinfluencingoccupancy)representsanendogenousforcewithinanOLSmodel
usedtoestimateelasticities.TheRDmodelisa“quasi‐experimental”designthatoffersawork‐
aroundtothisproblem.ARDdesignassignsa“treatment”effectinanon‐experimentalsetting
bydisaggregatingdatapointsintothoseaboveandbelowacertainthresholdfortreatment
(ThistlewaiteandCampbell1960),(LeeandLemieux2010).Thisdesignhasrecentlyincreased
inpopularityandhasbecome“ahighlycredibleandtransparentwayofestimatingprogram
effects,”accordingtoLeeandLemieux(2010).
Inthecaseoftheparkingpilot,ideally,whenmeanmonthlyoccupanciesdroppedbelow
60%onagivenblock,theyweretobesubjecttoapricedecreaseinthesubsequentmonth.
Similarly,whenoccupanciesexceeded80%,theyweretobesubjecttoapriceincreaseinthe
subsequentmonth.Hypothetically,thiswouldallowfortheapplicationofRDmodelsto
estimatetheeffectofpricechangesoneithersideofthe60%and80%boundariesduringthe
monthpriortoapricechange.BecausethePittsburghpilothadno“control”blocks,theRD
Equation2.PriceElasticityofDemandforPointE

Fichman,21

designisappealing.Allelseequal,onewouldassumeobservationswith59%and61%
occupanciesinagivenmonthtobeverysimilar,exceptforthefactthatthe59%blockwillhave
itspricemanipulatedthefollowingmonth.Thatallowsonetodetectpossibleeffects.
TheRDdesignwhichestimateseffectsoneithersideofastrictcutoffvalueisknownas
a“sharp”RDmodel.SuchamodelisidealizedinFigure10.Inthisfigure,considerthe
relationshipbetweenmonthtandmontht+1.Ifoccupancyonagivenblockinmonthtis
below60%,bydesign,inmontht+1thepriceonthatblockshouldbelower,presumingastrict
applicationofthepricingregimen.Givenanalternativehypothesisthatthedecreaseinpricein
montht+1wouldbeassociatedwithariseinoccupancyrelativetomontht,onewouldexpect
alinefittedtothediscontinuousfunctioninFigure10,wherethetreatment(priceincreaseor
decrease)isrepresentedbyadummyvariableDwhichaffectstheconditionalexpectation
Figure10.GraphicalrepresentationofhypothesizedRDdesignatthe60%occupancythreshold.Redarrow
representsthehypotheticaloccupancyeffectofa$0.25pricechange.E[Y(0)|X]istheconditionalexpectationofY
givenXintheabsenceofthepricechangetreatmenteffect.E[Y(1)|X]istheconditionalexpectationofYgivenXin
thepresenceof(orinteractedwith)thetreatmenteffect(1).

Fichman,22

E(Y|X)ofoutcomesoneithersideofthecutoffvaluec=60%forauser‐specifiedbandwidthb
(Equation3).
D 0  
D 1  
BecausetheassignmentvariableXisindependentofpotentialoutcomesY(0)andY(1),the
causaleffectofthetreatmentisrepresentedbythecoefficientρ–representingdiscontinuity
betweenestimatedfunctionsoneithersideofthecutoffvalue(Equations4‐6).
EY0|X α 
Y1Y
0ρ

α ρD η
Equations3‐6areadaptedfromImbensandLemieux(2008).Thespecificationofbandwidthis
importanttopreventnon‐parametricfunctions(whichfunctiononlargerscalesinthedata)
fromdeceptivelyapproximatingjumpsinvalueatthecutoffpoint(AngristandPischke2015).
Inpractice,theRDmodelcanbeusedasatooltoestimateOLSregressionsoneithersideofthe
discontinuityandcomparetheirbehavioratthecutoff.ThisisthemethodIutilize.However,it
isalsocommonpracticetoestimateasinglemodelandassesstheeffectasthecoefficientρ
associatedwiththeinteractionbetweenthetreatmentfixedeffectandeachofthe
independentvariablesinordertoreducethemagnitudeofstandarderrors(Shaman2016).
TheRDdesigncanbeillustrativeinimplementinga“data‐mining”approach,and
comparingtheeffectoftreatmentonone‐week,two‐week,one‐monthandtwo‐month
Equation3.AssignmentofdummytreatmentvariableD
Equation6.Combinedregressionterms
Equations4&5.Conditionalexpectationgivenlackoftreatmentvariable.

Fichman,23

timescales.Thisapproachallowsonetoperceivethechangeorlagintheeffectofthe
treatmentoneithersideofthethresholdovertime,shouldsuchaneffectexist.RDcanbe
implementedusingthebaseOLSregressionmodelsinRandgraphicallyrepresentedusingthe
“lattice”package.
SomepracticalproblemswiththedatalimitthepowerofRDmodelsinthisinstance.I
examinedtheoccupancydataandpricechangeregimeanddeterminedthatinpractice,the
administratorsdidnotadheretosuchstrictrulesregardingthe60%and80%thresholds–
insteadworkingtominimizespikesinoccupancyandgenerallymaintainoccupancieswithinthe
60‐80%range.Thismeansthattherewereseveralcircumstancesinwhichstreetsthathadmid‐
weekmeanoccupanciesoflessthan80%hadtheirpricesraised,or,inonecircumstance,for
reasonsIcouldnotdetermine,FrewStreet’s“5001”blockunderwentapricehikedespitemean
Figure11.Observationsofdailymeanoccupancyratesformidweek,schoolyearkiosksasafunctionofprevious
occupancyinmontht.Thisplotshowsonlyobservationswhichhadexperiencedpricechangesthemonthprior.

Fichman,24

occupanciesnear20%‐perhapsaresponsetoseveralmaxedoutperiodsintandemwith
coterminousareasonTechStreetandFrew’smaindrag.Administratorscommunicatedtome
thattheseunevenapplicationsoftreatmentwereinpartafunctionofuncertaintyregarding
theoccupancyestimations.Theywerealsoaproductofadministrativeapplicationof
specializedlocalknowledge–forexample,ifidiosyncraticcampuseventslikegraduationledto
inflatedmeanoccupancies,theseobservationsweredisregarded.Also,theadministrators
weresamplingdayseachmonth,ratherthanlookingatthefulldataset–whichmighthaveled
themtomakedecisionswhilemissinginformationthatisincludedinthisstudy(Fichman2015).
Furthermore,becausethedatahadtobetrimmeddowntoomitsummer‐timeobservations,
thismeansthataftercleaning,littledataremainednearthe80%threshold(Figure11)–
certainlynotenoughtocreateanyreliableestimatesforeffectsnear80%.Mostoftheblocks
whichhadschool‐yearrateincreaseshadexperiencedpreviousaverageoccupanciesof100%.
Figure12.Observationsofdailymeanoccupancyratesformidweek,schoolyearkiosksasafunctionofprevious
occupancyinmontht.Observationswhichfollowedapricechangeareshowninred.

Fichman,25

The60%thresholdhasareasonableamountofdataoneitherside,butforsomemonthswith
lessthan60%occupancythetreatmentwasnotapplied(Figure12).Thismaybeaccountedfor
byusingonlytreatedblocksbelowc=60%totheexclusionofun‐treatedblocksbelow60%.
Inordertodetermineifremovaloftheuntreatedobservationsbelow60%wouldunduly
biasthesample,IperformedaWelch,IndependentTwo‐Samplet‐testtotestthenull
hypothesisthatthemeanmid‐week,school‐yearoccupancyinmonthsoflessthan60%
occupancypriortopricedecreasewasstatisticallynodifferentthanmeanoccupancyinmonths
oflessthan60%occupancywhichdidnothavetheirpricesmanipulated.Shouldthesamples
provestronglydistinct,onemightarguethatremovinguntreatedobservationsinordertorun
anRDmodelwouldresultinseriousbiasunlessuntreatedandtreatedobservationsare
essentiallyindistinctfromoneanother.TheWelchtestismoreappropriatethantheStudent’s
t‐testinthiscircumstancebecausethesampleshaveunequalsizesandvariance,buttheWelch
test(sometimesknownasthe“UnequalVariance”t‐test)performswellundersuchan
assumption(Ruxton2006).Ifailedtorejectthenullhypothesisthatthemeanswerestatistically
indistinct,suggestingnospecificbiasinapplyingthetreatment(Figure13).Ibelievethatthe
non‐applicationoftreatmentwasnotdonewithanyspecificgoalofoccupancymanipulation,
Figure13.F‐TestandWelch’sT‐Testperformedondailymeanweekday,school‐yearoccupancyobservations
below60%occupancyforwhichtherewasandwasnotapricechangeinthesubsequentmonth
F‐Statistic p‐value Lower Upper Denominator Numerator RatioofVariances
0.777 0.128 0.548 1.075 96 236 0.777
t‐statistic df p(twotailed) Lower Upper NoPriceChange PriceChange
0.238 160.16 0.813 ‐0.037 0.047 0.389 0.383
F‐TesttoCompareVariances
95%ConfidenceInterval df
Welch'sT‐TestAssumingUnequalVarianceandSampleSize
95%ConfidenceInterval SampleMeans

Fichman,26

butwasratheraproductofhaphazardinterventionbroughtaboutforreasonsspecifiedearlier.
Theprobabilityofasub60%observationreceivingthetreatmentwas0.27.
Results
DescriptiveAnalyses
Duringthecourseofthepilot,overallrevenuesincreased(Figure15).Totalrevenuesfor
thefivestreetsanalyzedinthispaperwere$2,139,328duringthestudyperiod.Recallthatfor
moststreetsinthepilotprogram,ratesdecreasedduringthestudyperiod(Figure3)andall
streetsendeduphavinglowerpricesthanthe$2City‐imposedpricewhichwastheimpetusfor
thepilot.
Bytheendofthepilot,occupanciesformoststreetsmovedtowardsthetarget60‐80%
rangespecifiedbytheadministrators(Figure16).Thepilotseemedespeciallysuccessfulin
elevatingtheoccupancyofblockswhichhadpreviousoccupanciesbelow60%andseemedto
havelittleeffectonblockswhichhadveryhighoccupancies(Figure17).Theplotbelowshows
Figure15.Grossmonthlyrevenuesandlineartrend.

Fichman,27

asimilargraphictoonepublishedbyMillard‐Ball,WeinbergerandHampshire(2014).Thisplot
isnotablysimilarinthatafittedlinedoesnotshowadistinctdiscontinuitynear60%or80%.In
thecaseofSFPark,thislackofdiscontinuitywastakenbytheauthorstomeanthatprice
changeswerenotworking.However,inthecaseofthePittsburghdata,thehaphazard
applicationofpricechangetreatmentsmeansthatoneshouldexpectnosucheffectatthose
thresholds.
Figure16.Street‐by‐streetdaily,on‐session,midweekmeanoccupancieswithlineartrend

Fichman,28


PriceElasticityofDemand

Figure18showsthePriceElasticitiesofDemandforallstreetsbasedonweekday
observationsduringtheentirestudyperiod.TheseelasticitieswerecalculatedbasedonOLS
regressioncoefficientsassociatedwithon‐streetparkingrateestimatedusingthemodelshown
inAppendixI.Thecoefficientsinthesemodelsrepresenttheestimatedeffectofpriceonmean
dailyoccupancywhencontrollingforstreet(inthecaseofthestudy‐wideregression),dayof
Street(PseudonyminDataSet) WeekdayPriceElasticityofDemand
FrewSt.(FREWST)‐0.728
FrewStWest(FREW5001)‐0.299
TechStreet(TECHST)‐0.730
SchenleyDrive(SCHEDRCMU) ‐1.211
MargaretMorrisonSt(MARMOR) ‐0.867
Figure18.Priceelasticitiesofdemandforstreetsinvolvedinthepilot
Figure17.Observationsofdailymeanoccupancyratesformidweek,schoolyearstreet‐daysinmontht+1asa
percentageofpreviousoccupancyinmontht.Loessestimatedfit.Dottedlinerepresents100%.Oneoutlierof
1000%changewasremoved

Fichman,29

week,year,andwhetherCMUclasswasinsession.Thesefactorsexplainedalargeproportion
oftheoverallvarianceinoccupancy.R‐squaredvaluesrangedfrom0.435inthecaseof
SchenleyDriveto.699forthe5001blockofFrewStreet.Itshouldbenotedthatwhetheror
notschoolwasinsessionhadaverylargeeffectonoccupancy.Fortheentiredataset,the“Off
Session”fixedeffectwasassociatedwitha29.7%reductioninmeandailyoccupancy.Ididnot
controlfor“month”oftheyearfixedeffectsbecauseofstrongmulti‐collinearitywiththe“Off
Session”effect.SchenleyDrivewasthemostelastic,whichseemsintuitivegivenitsrelative
distancetomaincampusattractionsanditsrelativeproximitytosomefreeon‐streetparking
nearPhippsConservatory.Thetwopremiumareas,FrewSt.andTechSt.hadverysimilar
elasticities,whiletheEconomy‐pricedMargaretMorrisonStreetwasslightlymoreelastic.Frew
St.’s5001blockwastheleastelasticbyfar.Icomparetheseelasticitiestothosemeasuredfor
SanFranciscoandSeattleinthediscussionsectionofthisstudy.
RegressionDiscontinuity
TheRDmodelssuggestedthatpricechangeswereindeedhavinganeffect(Figure19).
However,someproblemswithsamplesizelimittheconclusivenessoftheseanalysesandthe
resultsshouldbetreatedwithsomereserve.Atthe60%threshold,therewasadistinct
discontinuityforobservationsinmontht+1,estimatedasa12.17%changeinoccupancy
associatedwitha$0.25changeinprice(Figure20).Splittingthediscontinuityintoweeks1‐2
andweeks3‐4ofmontht+1providessomeinterestinginsightintothebehaviorofdrivers.In
weeks1‐2,theestimatedeffectofa$0.25changeinpricewas6.19%(Figure21),whereasthe
estimatedeffectinthemonth’slasttwoweekswas18.14%(Figure22),implyingthattherewas
alagindriverresponsivenesstopricechanges.

Fichman,30

TheregressionlinesrepresentthefittedvaluesoftheOLSmodelsshowninAppendixII.

Relativelysmallsamplesizescausedsomeproblemsandshouldgivesomereasonto
taketheSRDresultswithcaution.First,therewastoolittledataonthetreatmentsideofthe
discontinuitytouseabandwidthsmallerthanthatoftheentiredata‐setwithoutseeingan
incrediblylargediscontinuity–over50%.Thereislittletosuggestthatthislineardiscontinuity
isactuallytracinganon‐linearfunction,butthefunctionmaybeover‐fitted.Second,the
estimationsofdiscontinuitiesforweeks1‐2andweeks3‐4ofmontht+1havetoofew
observationstoincludeasmanycontrolsaswereincludedwiththeestimationofthefullmonth
t+1(AppendixII).Eachoftheseestimationswascarriedoutusingareducednumberof
controlstoeliminatethestrongmulti‐collinearitybetweenstreetfixedeffectsandpricein
TimePeriodChangeinOccupancyat60%Threshold
Montht+112.17%
Montht+1,Days1‐146.19%
Montht+1,Days14‐End18.14%
Figure20.SRDplotshowingdiscontinuityinoccupancyinmontht+1.Blockswhichreceivedpricemanipulation
treatmentbelowthe60%discontinuitythresholdareshowninred.
Figure19.Estimateddiscontinuityat60%thresholdforthreeSRDmodels

Fichman,31

montht.Inthissense,thesebi‐weeklyregressionsarenotcompletelycomparabletothatof
thewholemontht+1.

Discussion
Thedynamicpricingregimeusedduringthepilotpushedoccupancytowardstarget
rates,increasedrevenuesandloweredprices.Furthermore,RDmodelssuggestthattheprice
changeshadtheintendedeffectondriverbehavior.ThestaticratestheCityhadbeenusing
priortotheimplementationofthepilotdidnotaccuratelypriceparkingaccordingtodemand.
Consideringhowrelativelyelasticdriverresponsewasduringthestudyperiod,over‐charging
forparkingduring2012mostlikelycreatedincentivefordriverstomodifytheirparking
behaviorinwaysthathadpotentiallynegativeeffects.Thoughitisdifficulttodiscernwhere
driversparkedoutsidethestudyareawhenpricesweretoohigh,it’sclearthatbehaviorwas
Figures21and22.RDplotsshowingdiscontinuityinoccupancyinmontht+1,weeks1‐2(left)and3‐4(right).
Blockswhichreceivedpricemanipulationtreatmentbelowthe60%discontinuitythresholdareshowninred.

Fichman,32

affected.Forexample,the5001blockofFrewStreethadextremelylowoccupancywhen
pricesweresetat“Premium”levelsearlyinthestudyperiod(atasimilarratetothepre‐study
rateof$2/hr),butsawincreasedoccupancywhenpriceswerereduced.
CityEstimatedElasticity
SanFrancisco(PierceandShoup,2013) ‐1.04to‐0.03
Seattle(Ottossonetal.,2013) ‐0.4 (mean)
Pittsburgh‐1.21to‐0.29
Therangeofelasticitiesestimatedforstreetsinthepilotprogramroughlycorresponds
toestimatesinbothSanFranciscoandSeattle,(Figure23).UnlikeSeattleandSanFrancisco,
thispilotdidnotestimateforblockscity‐wide,andtheabsenceofobservationsincentral
businessareasprobablycreatedapictureofamoreelasticparkingmarket.Forreasons
discussedearlier,simplepriceelasticitiesofdemandareprobablynotextremelyusefulin
determiningthespecificeffectofpricemanipulationbecauseofissueswithendogeneity.
Furthermore,positiveelasticitiesmeasuredbyChatmanandManville(2014)associatedwith
ratehikessuggestthatissueswithlatentdemandmayfurtherobscuretheuseofsuchametric
inmeasuringtheeffectivenessofapricingregime.However,elasticitiesareusefulforpurposes
ofcomparisonconsideringhowscantthepublishedliteratureisatthisjuncture.
Withcaveatsaboutsamplesizetakenunderconsideration,RDmodelssuggestthat
pricechangesdoindeedhaveaneffectondriverbehaviorwhichcorrespondstoanegative
elasticity.However,thesemodelsalsosuggestthatdriversdidnotrespondinstantaneouslyto
pricechangesduringthepilot.Theexistenceoftheselagsimpliesthatmoreprecise
communicationofpricemayreducelag‐timeindriverresponsivenessbycommunicating
informationmorequickly.Suchcommunicationcouldcomeintheformofaconsumer‐facing
Figure23.EstimatedelasticitiesforPittsburghincomparisontootherpublishedstudies.

Fichman,33

internetandphoneapplicationshowingpricinginformationandperhapsoccupancy
estimations.Recallthatduringthepilotprogram,specificpriceswerenotcommunicated
directlytodriversexceptatthekiosk.Althoughtherewerestreetsignsindicating“Economy”
and“Premium”parkingzones,adrivercouldnotlearnthespecificpriceuntilheparkedthecar
andapproachedthekiosk.Bythattime,thedrivermaybemuchlesslikelytogetbackinthe
carandchooseanotherlocationandhemayjustpayforthetransactionandapplynewpricing
knowledgeonthenextvisit.Forone‐timevisitors,ratesmayhavelittleeffect.
Dynamically‐pricedparkingprogramsshouldbedesignedtoallowforrobust
measurementofeffects.Thisanalysisdemonstratesthepotentialutilityoftheregression
discontinuitydesigninmeasuringtheeffectofpricechanges.Theregressiondiscontinuity
modelshouldallowplannerstomonitortheeffectofdynamicpricingwithoutworryingabout
thebiascausedbymyriadspatialandtemporalfactors,nottomentiontheendogeneity
inherentinoccupancy‐drivenpricemanipulations.Inordertoeffectivelyusethistool,
programsshouldadheretooneormoreregimesofrule‐baseddecision‐makingthatcanbe
carefullycomparedandmeasured.Furthermore,randomizedcontrolareasshouldbe
implementedinanypilotprogramtoprovidemorebasisforrobustestimationoftreatment
effects.
Inorderforademand‐based‐pricingprogramtofunctioneffectively,itisimportantto
accuratelymeasureoccupancy.Duringthepilot,fundamentaluncertaintyabouttheaccuracy
ofoccupancymeasuresseemstohavemadeadministrativedecisionspronetoerror.
Knowledgeofthisuncertaintymayhavepromptedadministratorstoimplementratechanges
conservatively.Itisperhapsforthisreasonthatpositiveratechangescamemostlyasaresult

Fichman,34

ofmaximumoccupancyevents,itwasalsoforthisreasonthatpricereductiontreatmentwas
appliedhaphazardly.
Estimatingoccupancyisadifficultthingtodowell–ifsensorscannotbeusedto
monitortrueoccupancy,itwouldbenecessarytomodelarrivalanddepartureratesinrelation
tooccupancytoestimatewhenpeoplehavelefttheirspaces.Spatialelementsofparkingmight
alsobedifficulttomodel–ablockcansupporttencompactcarsorperhapssixpickuptrucks,
withsomemarginforerrorbasedonhowclosedriversdecidetoparktooneanother.Thereis
awaytoreducesomeuncertaintycreatedbythisspatialproblem.WhenadriverinPittsburgh
paysforparkingtheyinputtheirlicenseplatenumber,sothatparkingauthoritiescan
appropriatelyassignticketstoscofflawdrivers.Thispersonalidentifyinginformationismasked
inthedataforprivacypurposes.However,itseemsconceivablethatlicenseplateinformation
couldbejoinedwithinformationaboutthemakeandmodeloftheidentifiedvehicle,which
couldbeassociatedwithadatabaseofknowndimensionsofthesevehicletypes.Thiswould
allowonetoestimatethetotallengthofallthevehiclesparkingatanygiventime.This
naturallyintroducessomeprivacyconcerns,sosubsequenttothisjoinofinformation,perhaps
theID,makeandmodelofthecarcouldbeobscuredbeforeanyadministratorscouldviewit.
Thedifficultypilotadministratorshadininfluencingbehaviorduringsummermonths
pointstotheimportanceofusinglaxconstraintsonpricechanges.Theuseofmultiple“time‐
bands”wouldallowformoresensitiveandflexibleimplementationofpricesinorderto
accommodatefluctuationsinparkingpatternsduringtheday,weekoryear.Oneofthe
reasonsrobuststatisticalanalysiswasdifficultinthisstudywasthatsummerobservationswere
nexttouselessbecausepricescouldnotbeadjustedquicklyenoughtodealwithchangesin

Fichman,35

demand.Althoughadministratorsbeganusinglargerpricechangesduring2014,itwasstilltoo
experimentalandtooslowtoprovideusefuldatawhereinsupplyanddemandweresubjectto
theequilibratingforcesofpricemanipulation.
Ultimately,demand‐basedpricingisdesignedtoreducetrafficcongestionbyeliminating
cruising.Occupancyandvacanciesareonlyproxiesoftherealdependentvariableofinterest.
Totrulyassessanydemand‐basedparkingprogram,trafficcountsshouldbetakenatstrategic
locationsthatallowforcontrolledobservationofeffects.InthecaseofthePittsburghpilot,the
PennsylvaniaDepartmentofTransportation,SouthwesternPennsylvaniaCommissionand
Pittsburgh’scity’splanningdepartmentrespondedtomyinquiriesabouttrafficcount
inventories.Asagroup,theyhadtakenfewerthanahalf‐dozentrafficcountsintheareanear
theuniversity,noneofthemontherelevantstreets.Purposebuiltcountersorvideomonitors
arecheapandeasytooperateandshouldbeintegratedwithanypricemanipulationschemeso
astomeasurethemostimportantofalleffects–roaddecongestion.
Ifdemand‐basedprogramsaremonitoredsuchthatpricescanbemanipulated
effectively,andifrulesareflexibleenoughtoallowforadministratorstoeffectivelyequilibrate
supplyanddemand,thiswillbringparkingalittleclosertobeingagoodwhichisnon‐
excludable.Unlikeatrue“publicgood,”itwillstillberivalinconsumptionwithprivateparking
garages.However,privategarageownersalreadyintuitivelyusethedemand‐basedpricing
strategy–iflotsareempty,theyarechargingtoomuch,andviceversa.Ifpubliclyowned
parkingcanbemoreefficientlypricedsuchthatitcompeteseffectivelywithprivateparking,
privateparkingownersmaycryfoul,buttheirlandwillsoonbepurposedtohigherandbetter
uses–anetsocialbenefit.Aholistictransportationstrategy,completewithdynamicallypriced

Fichman,36

parkingandavailable,reliablepublictransportation,couldcreateanenvironmentinwhichto
makeapoliticallyfeasiblepushtoreduceoreliminateminimumparkingrequirementsina
zoningcodeandmaketheactualpriceofdrivingandparkingtransparenttoconsumers.
I’dliketoacknowledgetheadviceandsupportofFrancescaAmmon,ErickGuerra,Kenneth
Steif,MarkFichman,PaulShaman,IvanTereshenko,CarnegieMellonUniversityandtheCityof
Pittsburghinhelpingmecompletethisproject.
WorksCited
Angrist,JoshuaD,andJorn‐SteffenPischke.2015.Mastering'Metrics.Princton,NJ:PrincetonUniversity
Press.
Arnott,Richard,andErenInci.2006."Anintegratedmodelofdowntownparkingandtrafficcongestion."
JournalofUrbanEconomics60:418‐442.
Brooke,Sarah,StephenIson,andMohammedQuddus.2014."On‐StreetParkingSearch,Reviewand
FutureResearchDirection."TransportationResearchRecord:JournaloftheTransportation
ResearchBoard2469:66‐75.
Chatman,DanielG,andMichaelManville.2014."Theoryversusimplementationincongestion‐priced
parking:AnevaluationofSFPark2011‐2012."ResearchinTransportationEconomics52‐60.
CityofPittsburgh.2015.CodeofOrdinances,CityofPittsburgh,Pennsylvania.Pittsburgh,PA,August15.
AccessedMarch28,2016.
https://www.municode.com/library/pa/pittsburgh/codes/code_of_ordinances.
Colonico,Sebastian,MathiasD.Cattaneo,andRocioTitiunik.2015."rdrobust:AnRPackageforRobust
NonparametricInferenceinRegression‐DiscontinuityDesigns."TheRJournal7(1).
Fichman,Mark,interviewbyMichaelFichman.2015.Personalemailcommunication(October12).
Gelman,Andrew,andJenniferHill.2009.DataAnalysisUsingRegressionandMultilevel/Heirarchical
Models.NewYork:CambridgeUniversityPress.
Ghent,Peer.2015."OptimizingPerformanceObjectivesforProjectsofCongestionPricingforParking."
TransportationResearchRecord:JournaloftheTransportationResearchBoard2530:101‐105.
Imbens,Guido,andThomasLemieux.2008."Regressiondiscontinuitydesigns:Aguidetopractice."
JournalofEconometrics142:615‐635.
Lee,DavidS,andThomasLemieux.2010."RegressionDiscontinuityDesignsInEconomics."Journalof
EconomicLiterature48:281‐355.
Millard‐Ball,Adam,RachelRWeinberger,andRobertCHampshire.2014."EvaluatingtheImpactsof
Performance‐BasedParking,CommentonPierceandShoup."JournaloftheAmericanPlanning
Association79(4):330‐336.

Fichman,37

Millard‐Ball,Adam,RachelRWeinberger,andRobertCHampshire.2014."Isthecurb80%fullor20%
empty?AssessingtheimpactsofSanFrancisco'sparkingpricingexperiment."Transportation
ResearchPart!76‐92.
Ottosson,DadiBaldur,CynthiaChen,TingtingWang,andLinHaiyun.2013."Thesensitivityofon‐street
parkingdemandinresponsetopricechanges:AcasestudyinSeattle,WA."Transportation
Policy25:222‐232.
PhiladelphiaMayor'sOfficeofTransportationandUtilities.2009.PhiladelphiaCompleteStreetsDesign
Handbook.Philadelphia,June4.AccessedMarch28,2016.
http://www.philadelphiastreets.com/images/uploads/resource_library/cs‐handbook.pdf.
Pierce,Gregory,andDonaldShoup.2013."GettingthePricesRight."JournaloftheAmericanPlanning
Association79(1):67‐81.
Ruxton,Graeme.2006."Theunequalvariancet‐testisanunderusedalternativeetoStudent’st‐testand
theMann–WhitneyUtest."BehavioralEcology17(4):688‐690.
Shaman,Paul,interviewbyMichaelFichman.2016.PersonalInterview(April12).
Shoup,Donald.2004.TheHighCostofFreeParking.Chicago,IL:PlannersPress,AmericanPlanning
Association.
Shoup,Donald.1999."Thetroublewithminimumparkingrequirements."TransportationResearchPart
A33:549‐574.
Thistlewaite,Donald,andDonaldTCampbell.1960."Regression‐discontinuityanalysis:Analternativeto
theexpostfactoexperiment."JournalofEducationalPsychology51(6):309‐317.
USDepartmentofTransportation.2015.ValuePricedPilotProgramFunding.August27.Accessed
February3,2016.
http://ops.fhwa.dot.gov/congestionpricing/value_pricing/projects/funding.htm.
Vickrey,William.1954."Theeconomizingofcurbparkingspace."TrafficEngineering62‐67.
Zullo,Robert.2014."Pittsburghtoraiseparkingmeterrates,chargeatnight."PittsburghPost‐Gazette,
December11.AccessedMarch23,2016.http://www.post‐
gazette.com/local/city/2014/12/10/Pittsburgh‐to‐raise‐parking‐meter‐rates‐charge‐at‐
night/stories/201412100208.


Fichman,38

Appendix–Regressiontables
AllStreets FREWST FREWST
5001
SCHENL EY
DRIVE TECHST
MARGARET
MORRISON 
ST
MondayFixedEffect ‐5.551
***
‐3.193
**
‐6.763
***
‐1.248 ‐7.856
***
‐7.836
***
Standa rdError 1.049 1.543 2.494 2.392 1.706 2.551
ThursdayFi xedEffect 0.094 1.581 1.217 0.675 ‐2.15 ‐0.608
Standa rdError 1.027 1.473 2.451 2.348 1.678 2.525
Tues da yFi xedEffect ‐1.29 0.261 ‐2.619 0.522 ‐5.172
***
0.41
Standa rdError 1.019 1.453 2.473 2.323 1.66 2.501
WednesdayFixedEffect ‐1.338 ‐0.535 ‐2.07 0.867 ‐3.869
**
‐1.756
Standa rdError 1.024 1.461 2.475 2.374 1.671 2.474
FrewStFixedEffec t 19.537
***
Standa rdError 1.111
Mar garetMorrisonFix edEffect 10.487
***
Standa rdError 1.117
SchenleyDriveFi xedEffect ‐16.646
***
Standa rdError 1.118
TechStFixedEff ect 26.968
***
Standa rdError 1.232
Price(Doll ars ) ‐30.070
***
‐21.251
***
‐9.870
***
‐49.593
***
‐18.291
***
‐57.791
***
Standa rdError 1.014 2.722 3.527 3.031 2.217 4.088
Year 2014FixedEffect 8.724
***
2.258 38.447
***
18.735
***
‐2.475
*
28.012
***
Standa rdError 0.763 1.675 3.879 1.788 1.262 2.191
Year 2015FixedEffect 12.332
***
4.067
**
41.385
***
22.715
***
0.08 33.491
***
Standa rdError 0.882 1.764 3.616 2.518 1.499 2.544
OffSessionFixedEffec t ‐29.767
***
‐24.823
***
‐35.170
***
‐30.071
***
‐33.048
***
‐34.777
***
Standa rdError 0.688 1.01 1.641 1.692 1.152 1.83
Constant 95.616
***
100.242
***
49.806
***
90.769
***
108.331
***
124.302
***
Standa rdError 1.962 6.013 7.479 3.491 5.103 4.298
Observations 2,942 596 562 590 610 584
R
2
0.548 0.578 0.699 0.435 0.584 0.45
Adj u s tedR
2
0.546 0.572 0.694 0.428 0.578 0.443
Resi dua lStd.Error 17.592(df
=2929)
11.377(df
=587)
18.491(df
=553)
17.939(df
=581)
13.074(df
=601)
19.171(df=
575)
FStatis tic
295.783
***
(df=12;
2929)
100.496
***
(df=8;
587)
160.333
***
(df=8;
553)
55.982
***
(df=8;
581)
105.325
***
(df=8;
601)
58.872
***
(df=8;575)
Note:
*
p
**
p
***
p<0.01
OLSModelstoEstimateElasticities
Dependentvariable:
PercentageMeanDa il yOccupa ncy(%)
AppendixI‐OLSregressionsusedtoestimatepriceelasticityofdemand.Basecasesforfixedeffectsare:Friday
(Weekday),2013(Year)andFrewSt5001(Street)andOn‐Session(Off‐Session)

Fichman,39


AppendixII‐OLSregressionsusedtoestimateregressiondiscontinuities.Basecasesforfixedeffectsare:Friday
(Weekday),FrewSt5001(Street)andOn‐Session(Off‐Session)
WholeMonth
t+1,0‐60%
Previous
Occupancy
WholeMonth
t+1,60‐80%
Previous
Occupancy
Week1‐2,
Montht+1,0‐
60%Previous
Occupancy
Week1‐2,
Montht+1,60‐
80%Previous
Occupancy
Week3‐4,
Montht+1,0‐
60%Previous
Occupancy
Week3‐4,
Montht+1,60‐
80%Previous
Occupancy
PreviousMonthMean DailyOccupancy(%) ‐0.463 0.207 ‐2.049
***
0.075 0.542 0.349
*
Standa rdError 0.38 0.141 0.385 0.198 0.437 0.199
MondayFixedEffect ‐0.469 ‐2.189 ‐2.492 ‐1.703 1.914 ‐2.649
Standa rdError 4.512 1.869 3.858 2.535 5.912 2.729
ThursdayFi xedEffect 0.906 1.81 0.427 ‐0.521 ‐0.125 4.177
Standa rdError 4.253 1.814 3.767 2.508 5.206 2.595
Tuesday FixedEffect 2.227 ‐0.951 5.207 ‐1.934 ‐2.432 0.06
Standa rdError 4.493 1.801 4.229 2.486 5.361 2.576
WednesdayFixedEffect ‐0.923 ‐0.899 0.847 ‐1.47 9.047 ‐0.607
Standa rdError 4.526 1.748 3.775 2.483 6.444 2.439
FrewStFixedEffect 12.742
***
11.292
*
12.667
**
Standa rdError 4.319 6.283 6.016
Mar garetMorrisonFixedEffect 7.349
***
2.437 11.146
***
Standa rdError 1.925 2.773 2.677
SchenleyDriveFixedEffect 22.469 ‐20.329
***
67.647
***
‐23.260
***
‐42.416
**
‐17.990
***
Standa rdError 17.555 2.14 15.03 3.153 17.711 3.05
TechStFixedEffect 70.800
***
25.022
***
131.890
***
26.261
***
‐15.295 22.005
***
Standa rdError 14.614 6.007 14.419 8.628 20.181 8.473
Price(Dollars) ‐22.098 ‐39.465
***
‐17.138 ‐45.367
***
‐32.226
***
Standa rdError 15.163 6.518 10.402 9.397 09.197
OffSessionFixedEffect ‐16.890
***
‐36.341
***
‐8.275
**
‐38.069
***
‐67.738
***
‐36.524
***
Standa rdError 4.022 1.356 3.841 2.129 12.546 1.879
Constant 72.046
**
110.992
***
80.616
***
130.596
***
62.946
***
90.220
***
33.093 15.246 22.348 21.833 15.275 21.407
Observations 79 538 39 259 40 279
R
2
0.798 0.685 0.93 0.664 0.869 0.718
Adj us ted R
2
0.772 0.678 0.909 0.649 0.835 0.706
Residual Std.Err or 0.123(df=
69)
0.132(df=
526)
0.075(df=
29)
0.128(df=
247)
0.109(df=
31)
0.135(df=
267)
FStatis tic 30.373
***
(df
=9;69)
103.832
***
(df
=11;526)
43.006
***
(df
=9;29)
44.461
***
(df=
11;247)
25.659
***
(df
=8;31)
61.754
***
(df=
11;267)
Note:
*
p
**
p
***
p<0.01
OLSModelstoEstimateRegre ssion Discontinuity
Dependentvariable:
MeanPercentageOccupanc yForDays inTimePeriodt+i