PublicDisclosureAuthorizedPublicDisclosureAuthorized

PolicyResearchWorkingPaper11181

MeasuringJobAccessibilityDifferentMethodsandNewData

AtsushiIimi

WORLDBANKGROUP

TransportGlobalDepartmentAugust2025

Averifiedreproducibilitypackageforthispaperisavailableat

,click

here

fordirectaccess.

ProducedbytheResearchSupportTeam

PolicyResearchWorkingPaper11181

Abstract

ThepaperreexamineshowtomeasurejobaccessibilityinenvironmentswithlimiteddataavailabilityandappliesdifferentmethodstoAntananarivo,thecapitalofMada-gascar.Jobcreationandaccessibilityareattractingrenewedinterestindevelopingcountries,whereunemploymentratesremainpersistentlyhigh.Thepaperfindstwotypesofjobaccessibilitymeasuresthatparticularlyimpactemploy-ment:proximitytopublictransportandaveragetraveltime

weightedbyavailablejobopportunities.Forthelatter,thepaperalsofindsthatnewopen-sourcedata,suchastheOpenBuildingsdataset,areeffectiveinidentifyingexistingjobopportunities.Usingthemeasuredresults,themarginalimpactofjobaccessibilityonemploymentisestimatedatabout?0.05to?0.06afterthepotentialendogeneityofaccessibilitymeasuresiscontrolled.

ThispaperisaproductoftheTransportGlobalDepartment.ItispartofalargereffortbytheWorldBanktoprovideopenaccesstoitsresearchandmakeacontributiontodevelopmentpolicydiscussionsaroundtheworld.PolicyResearchWorkingPapersarealsopostedontheWebat

/prwp.Theauthormaybecontactedat

aiimi@.Averifiedreproducibilitypackageforthispaperisavailableat

http://reproducibility.worldbank.

org

,click

here

fordirectaccess.

Y

C

I

A

E

RES

L

O

P

H

C

R

S

TRANSPARENT

P

E

R

W

O

R

K

I

ANALYSIS

A

NGP

ThePolicyResearchWorkingPaperSeriesdisseminatesthefindingsofworkinprogresstoencouragetheexchangeofideasaboutdevelopmentissues.Anobjectiveoftheseriesistogetthefindingsoutquickly,evenifthepresentationsarelessthanfullypolished.Thepaperscarrythenamesoftheauthorsandshouldbecitedaccordingly.Thefindings,interpretations,andconclusionsexpressedinthispaperareentirelythoseoftheauthors.TheydonotnecessarilyrepresenttheviewsoftheInternationalBankforReconstructionandDevelopment/WorldBankanditsaffiliatedorganizations,orthoseoftheExecutiveDirectorsoftheWorldBankorthegovernmentstheyrepresent.

MeasuringJobAccessibility:DifferentMethodsandNewData

AtsushiIimi

?

EasternandSouthernAfrica

TransportGlobalPractice

WorldBank

Keywords:Accesstojobs;Jobcreation;Probitmodel.

JELclassification:C25,J08;O14;O18

?Correspondingauthor.

-2-

I.INTRODUCTION

Thispaperaimstoreconsiderjobaccessibilitybycomparingdifferentmeasuresusedintheliteratureandexamininghowtheybehaveinconnectionwithlabormarketoutcomes,especiallywithinthecontextofdevelopingcountries.Inrecentyears,employmenthasbeenincreasingly

discussedinthedevelopingworld(e.g.,ILO,2022a,2022b,2024;BanerjeeandSequeira,2023;CarranzaandMcKenzie,2024).However,howtomeasurejobaccessibilityremainsrelatively

understudied.Variousmethodsandconceptsarediscussedintheliterature.Thepaperreviews

theprosandconsofexistingmethodsandproposesanewapproachusinganovelglobaldataset,OpenBuildings2.5DTemporalDataset(Sirkoetal.,2021),whichisapplicableevenindata-

scarceenvironments.Thepapertheninvestigateswhichaccessibilitymeasurementisthemosteffectivetoevaluateitspotentialimpactonlabormarketoutcomes,suchaslaborforce

participation.

Whiletheunderlyingbasicconceptislargelysimilar,jobaccessibilityisdefineddifferentlyacrossexistingstudies.Forinstance,ShahandSturzenegger(2022)estimatethatSouthAfricansspendabout60percentoftheirincomeontransportcosts.Inthiscontext,jobaccessibilityis

simplymeasuredbycommutingtimeandcosts.BanerjeeandSequeira(2023)alsodiscussthe

effectivenessofpublictransportsubsidiestosupportjobsearches.Proximitytopublictransportisanothermeasureoftenusedtoexamineitsimpactonemployment(e.g.,Holzeretal.,2003;

MayerandTrevien,2017;Tyndall,2017).Incar-dependentcountries,suchastheUnitedStates,congestionanddrivingtimeonhighwaysmaybemorerelevant(e.g.,GoodwinandNoland,

2003;BurrowsandBurd,2024).

BaradaranandRamjerdi(2001)discusstheperformanceofdifferentjobaccessibility

measures,classifyingthemintofivegroups:travelcost,gravity,timeconstraint-based,utility-based,andcompositeapproaches.Whilethetravelcostapproachismoststraightforward,the

gravityapproachexaminespotentialopportunities.Thetimeconstraint-basedapproachcanalsoaccountforpeople’stimelimitationsforotherdailyactivitiesthatrequiretransportation.The

utility-basedapproachismoredataintensiveasittakesintoaccountpeople’spreferencesoverdifferenttransportoptions.Thecompositeapproachisevenmorecomplex,combiningthe

-3-

constraint-andutility-basedmethods.MerlinandHu(2017)emphasizetheeffectofcompetitionforjobs,classifyingmethodsslightlydifferentlyintofourgroups:cumulativeopportunity,

gravity,Shen’smodel,andinversebalancingfactorapproaches.Mostrecently,MazzullaandPirrone(2024)provideacomprehensivereviewofaccessibilitymeasures,developinga

systematicclassificationframework.

Despitedifferentclassifications,allmeasuresfocusontheavailabilityofopportunitiesandtransportationbarriersacrossdifferentlocations,differingmainlyinhowtheseimpedimentsandopportunitiesarequantifiedandaggregatedintoonemeasurementundertheoreticalassumptionsanddataconstraints.Thereareseveralmethodsthatarerelativelysimplemethodsandapplicableeventodevelopingcountrieswheredataavailabilityisoftenlimited.Insuchcountries,detailedorigin-destination(OD)dataarenotavailable,firmregistrydataareincomplete,anddataon

existingtransitsystemsareoftenlargelymissing.

Thefirstmethodinvolvestraditionaltravelcostortime-basedmeasures.Thistypeofdataisconceptuallyunderstandableandcanberelativelyeasilycollected.Soetal.(2001)examinetherelationshipbetweenwagesandcommutetimeobservedinthecensusdatafortheUnited

States.Ruppertetal.(2009)andLeBarbanchonetal.(2021)analyzeasimilarissueinFrance.

Commutetimeandcostsmaybeavailableinstandardhouseholdsurveysevenindeveloping

countries.AccordingtothenationaltravelsurveydatainSouthAfrica,averagecommutingtimeisabout50minutesoneway(ShahandSturzenegger,2022).Usinghouseholdsurveydata,

Lozano-GraciaandYoung,2014)showthatmedianhouseholdexpenditureontransportationnormallyrangesfrom2to8percentofincome.

Apotentialempiricalchallengeinusingthesemeasuresisthepartialobservabilityofdata.Inparticularindevelopingcountries,informalityisveryhigh(Dodmanetal.,2017;ILO2017;

Lalletal.,2017;OECD,2022).Over90percentofthetotalworkforceisengagedininformal

jobsorself-employment(ILO,2023).Inaddition,manyAfricanswalktowork.InAntananarivo,Madagascar,halfofthecityresidentssimplywalktotheirworkplaces(Iimi,2023).Thus,the

traditionalassumptionontime-minimizingtravelbehavior,giventheavailabilityof

transportationoptions,maynothold.Observedcommutetimesandcostsareinherentlytruncated

-4-

atzero.Thiscensoredregressorproblem,thoughoftenignored,maycausesignificantbiasifnongenuinezeroobservationsaredroppedorassumedtobezero(RigobonandStoker,2007).

Second,theempiricalliteraturealsousestheproximitytotransportinfrastructureand

servicestomeasurejobaccessibility.Unliketheabovetraveltimeorcostapproach,thismeasureisobservableforallindividualswithknownresidences.Ithasbecomemorepracticalwiththe

increasingavailabilityofspatialdataandtoolsevenindevelopingcountries.Thechoiceofthisapproachdependsontheassumptionthatpeopleusethesetransportpublicservicestoaccess

jobs.Thismayormaynotbethecaseindevelopingcountriesbecausepeoplemaynotuseanytransportmeans.However,thismethodcanbeusefulifitismeasuredwithinformalpublic

transportincluded,whichisadominanttransportmeansinmanydevelopingcities(e.g.,ShimazakiandRahman,1996;CDIA,2011;Iimi,2023).

Theliteraturecommonlyusesa500to1,000-meterdistancefrompublictransportto

identifypotentialbeneficiaries(e.g.,Wasfietal.,2013;LachapelleandBoisjoly,2023).Tennoyetal.(2022)examinewalkingdistancestopublictransportinNorway,supportingaconventionalthresholdof500metersora4to6-minutewalktopromotepublictransportusage.Scholletal.

(2018)usealongerdistanceof1.5kmtodefinepotentialBRTcommutersinLima.

Theavailableevidencegenerallysupportsapositiveemploymentimpactofproximityto

publictransport(e.g.,seeBastiaanssenetal.(2020)forameta-analysis).Holzeretal.(2003)

findapositiveemploymenteffectofreversecommutetosuburbanareascausedbyanextendedmetrolineinSanFrancisco.Similarly,MayerandTrevien(2017)showthattheprogressive

developmentofRegionalExpressRailaroundParisledtoincreasedemployment.ProximitytoanewBRTinLimaalsohasapositiveimpactonemployment(Scholletal.,2018).Usingquasi-experimentalnaturalevents,Tyndall(2017)showsthatmetroservicedisruptionscausedby

HurricaneSandyincreasedunemploymentintheaffectedareasofNewYork.Anegative

correlationbetweenemploymentandhouseholddistancefromthenearestrailstationisalsofoundinMumbai,India(Alametal.,2021).

-5-

Third,thedistanceortraveltimetoworkplaces,namely,thecentralbusinessdistrict

(CBD),isanotherstraightforwardmeasureofjobaccessibility.Theaboveapproachmeasuringaccesstotransitmerelyrepresentspartofoverallaccessibility.Bycontrast,thismeasureis

potentiallycompleteandalignswellwithurbaneconomics.TraditionalurbanstudiescenteronaccessibilitytotheCBD,highlightingatradeoffbetweencommutingcostsandhousingcosts.

Forinstance,Levinson(1998)showsthatjob(andhousing)accessibilitydecreaseswithdistancefromtheCBDinWashington,DC.Similarly,Weber(2003)findsthatinPortland,Oregon,thedistancefromtheCBDisimportanttodeterminethenumberofopportunitiesandpeople’s

mobility.ThisinverserelationshipbetweenjobaccessibilityanddistancefromtheCBDisconfirmedinmostmajorU.S.cities(Ermagun,2021).

TheCBD-basedapproachassumesmobilitypatternswithintheframeworkofthe

monocentricurbanmodel,whichmayremainapplicabletomanyAfricancities.Primacyratesofthelargestcityindevelopingcountriesaregenerallyhigh.Unlikeotherregions,the

transformationfrommonocentrictopolycentricurbanstructuresisstilllimitedinAfricathoughsomesecondarycitiesareemerging(e.g.,Agyemangetal.,2019;OECD2022).InmanyAfricancities,theCBDcanbedefinedrelativelystraightforwardly.Thisisastrongadvantagefromtheempiricalpointofview.

Finally,thegravityapproachisperhapsthemostcommonmethodtomeasurejob

accessibilityintheliterature,withvariousextensionsdeveloped(e.g.,BaradaranandRamjerdi,2001;MerlinandHu,2017;JinandPaulsen,2018;Huang,2020;MazzullaandPirrone,2024).Itcantakeintoaccountpotentialopportunitiesacrossmultiplelocationsandweighsthem

negativelybytransporttime,costsorotherimpediments.Datarequirementsincrease

considerablywithgranularityinanalysis.Indevelopingcountries,itmaynotbeeasytoobtainhighlydisaggregatedspatialdataonhouseholds,employment,orfirms.

Thispaperaimsatcontributingtoidentifyingareliableandpracticaljobaccessibility

measureinthecontextofdevelopingcountries,especiallyinrelationshipwithemployment

outcomes.Usingvarioussourcesofdata,thepapercomparesdifferentmeasuresofjob

accessibility,usinguniquedatafromAntananarivo,thecapitalofMadagascar.Itnotonlyusesa

-6-

detailedlandusemaptoidentifyjoblocationsbutalsotestsanovelglobaldatasettoproxyjobopportunities.TheevidenceindicatesthatthejobaccessibilitymeasurederivedfromtheOpenBuildings2.5DTemporalDataset(Sirkoetal.,2021)ispromising.Thisisthefirstcontributionofthepaper.

Inaddition,thepaperalsocontributestodemonstratingapracticalwayofcontrollingforendogeneityofjobaccessibility.Aspredictedbytheory(e.g.,Rupertetal.,2009),commutingvariablesorjobaccessibilityarelikelyendogenousinthecontextoflaborforceparticipation.

Thispaperemploystheinstrumentalvariable(IV)techniqueusingsimpletopographicandmeteorologicaldata,withtheirempiricalvalidityreaffirmedbyconventionalHansen’sJ

statisticsandGuevara’sRefutabilityteststatistics(Guevara,2018).

Therestofthepaperisorganizedasfollows:SectionIIbrieflyexplainsthecountrycontextandmeasuresjobaccessibilitybydifferentmethods.SectionIIIdevelopsourempiricalstrategy.SectionIVdescribesthedata.SectionVpresentsthemainresults.SectionVIconcludes.

II.MEASURINGJOBACCESSIBILITYINANTANANARIVO

MadagascarisoneoftheleastdevelopedcountriesinAfrica,withaGDPpercapitaof

approximatelyUS$520.About21millionMalagasypeople,orroughly80percentofthe

population,livebelowthepovertyline.LikemanyotherAfricannations,Madagascarhasbeenexperiencingrapidurbanizationinrecentyears.Thetotalpopulationstandsataround26

million,

1

with3.3millionresidinginGreaterAntananarivo,thecapital.

2

Antananarivoisthe

fourthlargestcityintheEasternandSouthernAfricanregion,followingDaresSalaam,Nairobi,andAddisAbaba(Figure1).Othersecondarycities,suchasToamasinaandMahajanga,arealsoexperiencinggrowth.However,GreaterAntananarivoremainstheprimaryeconomichub.

Antananarivoexperiencessignificantcongestion,duetolimitedlandallocatedfortransportinfrastructure,historicbuildingsdatingbacktotheearly1600s,andsurroundinghillymountains.

1Accordingtothelatestnationalcensusdata(INSTAT2018).

2AccordingtotheUNHabitatestimate(UNESA2018).

-7-

Thecity’smainpublictransportservices,knownastaxi-be,areinefficientandunreliable.Over6,000minibusesoperateonabout130routes,manyofwhichareduplicatedandfragmented.

Minibusservicesareconcentratedalongmajornationalcorridorswithhighdemand,all

convergingtowardsthecitycenteraroundLakeAnosy.ThisareahistoricallyconstitutestheCBD,hostingmanyjobs(Figure2).

Inthispaper,fourmetricsareconsideredtomeasurejobaccessibility.First,dataon

reportedcommutingtimeorcostcanbeused.Arecentcity-levelhouseholdsurvey,MeasuringUrbanLivingStandards(MULS)inAntananarivocarriedoutin2016,includedabout9,300

individualsfromabout2,300households.Accordingtothisdata,peopleinthecityspendan

averageofabout15minutescommutingtowork(Figure3).

3

Onaverage,menspend21minutesforcommuting,whilewomenspend11minutes.Notably,however,asignificantnumberof

peopleeitherdonotworkorworkathome.Whenexcludingzerocommutetimes,theaveragecommutingdurationisabout29minutes(32minutesformenand25minutesforwomen).

ComparedtootherAfricancities,thisstilllooksrelativelymodest:50minutesonewayinSouthAfrica(ShahandSturzenegger,2022),74minutesinDaresSalaam(WorldBank,2016),and

100minutesdailyinAccra(Carmichaeletal.,2024).

Ingeneral,theriskofoverestimatingjobaccessibilityisevidentwhenusingthereport-

basedcommutetimemeasure,especiallyiftherearemanypeoplewhodonotworkorworkathomeorneartheirresidences,therefore,reportingminimalorzerocommutetime.Among

OECDcountries,wherepublictransportationispresumablymoreefficientandavailablethanrapidlygrowingdevelopingcitiessuchasAntananarivo,theaveragecommutetimeisabout28minutes(33minutesformenand22minutesforwomen)(LeBarbanchonetal.,2021).IntheUnitedStates,forinstance,theaverageone-waycommutetimeisabout26minutes(BurrowsandBurd,2024).Theaveragetransitcommutetimeis49.3minutes(Barkleyetal.,2018).Itisessentialtocarefullyinterpretshortcommutetimesindevelopingcities.

3Theseaveragesarecalculatedwithallavailabledata,someofwhicharenotusedinthefollowingregressionanalyses.

-8-

Similarly,reportedcommutingcostsinAntananarivoarealsosignificantlyzero-inflated.

Thesimpleaverageone-waycommutecostisabout117Malagasyariary(Figure4).Whilemen

spendonaverage162ariary,womenspend83ariary.Buthalfofthosesurveyedreportedzero

commutingcosts.Morewomenwalktoworkthanmen(Figure5).Whenthosewhowalkordo

notspendmoneyoncommutingareexcluded,theaveragecommutecostisabout500ariaryforbothmenandwomen.Theseinflatedzerovaluesincommutetimeorcostsmaycausebiaswhenexaminingtheeffectonemployment.Zerotimeorcostcouldindicateverylimitedaccessibilityratherthangoodaccess.Thefollowingsectionswillcastlightonthisproblembyexamininghowthesemeasuresbehaveinrelationwithlabormarketoutcomes.

Figure1.Urbanpopulationinmajorcities

Figure2.AccessibilitytoCBDandbusoperations

Source:UN-Habitat(2016).

-9-

Figure3.Commutetimesreported

Figure4.Commutingcostsreported

0.01.02.03.04.05

05.0e-04.001.0015.002

Density

Density

050100150200

010002000300040005000

Commutetimeoneway(minutes)

Commutingcostsoneway(Ariary)

Figure5.Modalshares

020406080100

Modalsharetogotoaworkplace

(%,multipleanswers)

36.6

...

43.4

2.67.22.9

57.7

43.9

Female

Car

BicycleMinibus

MaleWalk

Microbus&Taxi

Second,theproximitytopublictransportationismeasuredusingspatialsoftware.Unlikedevelopedcountrieswithhighcardependency,privatecarownershipisstilllimitedin

Madagascar.Themostimportantmeansoftransportareminibusesandcitybuses(seeFigure2above).AmonghouseholdsincludedintheMULSsurvey,theaveragegeographicdistanceto

nearestbusrouteisabout225metersora3-minutewalk,assumingawalkingspeedof4km/h(Figure6).Notethatthisdistanceismeasuredfromthebusroutesratherthanbusstops,astherearenostrictlydesignatedbusstopsintheMalagasyminibussystem.Practically,passengerscanembarkanddisembarkatanypointalongtheroutes.

Themeasureddistancevariessignificantly,rangingfromnearlyzerotoover2km,becausetransitaccessibilitybecomesextremelypoorasonemovesoutfromthecitycenter.Thefindingisconsistentwiththeliterature(e.g.,Ermagun,2021),highlightingthatthosewholiveoutside

-10-

thereachofcurrentminibusoperationsarelikelytofacesubstantiallimitationsinjob

accessibilityinAntananarivo.Recallthataglobalnormforproximitytopublictransportistypically500to1,000meters(e.g.,Wasfietal.,2013;Tennoyetal.,2022;LachapelleandBoisjoly,2023).

Figure6.Distancetothenearestbusroute

0.001.002.003.004

Density

0100020003000

Distancetothenearestbusroute(meter)

Third,theCBD-basedtraveltimeiscomputedbasedontheassumptionoftime-minimizingcommutingbehavior.LakeAnosyisusedasthecentralpointoftheCBDforthiscalculation.

Theshortesttriptimeisestimatedwithroadclassandtrafficspeeddatatakenintoaccount.Foreachroadsection,averagedrivingspeedisassumedtobe20kmperhourforNationalRoadsinthecityofAntananarivo,40kmperhourforNationalRoadsinsuburbanareas,and20kmperhourforotherroads.Inaddition,trafficspeedisassumedtobereducedbyhalfifthereare500minibusesoperatingonagivenroute.Thisisbecauseminibusescausetrafficcongestionin

Antananarivo.

ThemeasuredaccessibilitytotheCBDlooksmoreevenlydistributed,capturingpotential

jobopportunitiesbetter(Figure7).Byconstruction,thereisnozero-valueforthisvariablecreatedregardlessofwhethertheyactuallycommuteornot.Theaveragetraveltimeisestimatedatabout28minutesamonghouseholdscoveredbytheMULSsurvey.Importantly,thismeasureof

accessibilitycanpresentdetailedspatialgranularityofaccessibility(Figure8).ItisclearthattheaccessibilitytotheCBDisparticularlyhighwithina5-kmradiusoftheCBD.Accessibility

dropssharplybeyonda10-kmradius.Ingeneral,traveltime-basedaccessibilityisnot

necessarilyassociatedwithdistance-basedmeasurements.InAntananarivo,however,thetwo

-11-

metricsseemhighlycorrelatedbecauseofitsclassic,monocentriccitystructure.AllmajorroadsconvergearoundLakeAnosy,wheremanyminibusoperationsareconcentrated,attractingmorepeopleandfirms.

Figure7.EstimatedtraveltimetotheCBD

0.01.02.03

Density

020406080

TraveltimetoCBD(minute)

Figure8.SpatialdistributionoftraveltimetotheCBD

Finally,despitethehighconcentrationoffirmsandjobsaroundtheCBD,severalnew

businesscentersarealsoemerginginAntananarivo.Toaddressthispolycentric,thegravity

approachispredominantlyusedintheliterature(e.g.,DeMontisetal.,2011;JinandPaulsen,2018;Dixonetal.,2019;Huang,2020).Thispaperconsidersthesimplestversionofgravity-basedaccessibilitymeasure:

Thisrepresentstheaveragetriptimefromindividualitojoblocationm,weightedbythesizeofpotentialjobopportunitiesatm.Whiledimistransportdeterrencebetweeniandm,Ymrepresentsthesizeofjobopportunitiesexistingatlocationm.Thus,ym/Σmymcanbeinterpretedasthe

relativeimportanceofjobsizeatlocationmoraweight.Unlikeexistingstudieswhichoften

-12-

calculatetheaveragenumberofopportunitiesinverselyweightedbytraveldeterrence,

4

weusethisweightedaveragecommutetimetoreachpotentialjobopportunities.Althoughthetwo

formulasareessentiallyequivalentbutshowtheoppositedirections,thetime-basedmatricismoreusefulforcomparisonwithothermeasuresdiscussedinthispaper.

Anempiricalchallengeariseswhenmeasuringthesizeofjobopportunities,Y,inadata-

scarceenvironment.Twodatasourcesareusedhere.First,acity’slandusemapisusedto

quantifythesizeoflandareasdedicatedtoadministrative,commercialandindustrialzones,

culturalfacilities,hotelbuildings,marketareas,andofficebuildings.Thistypeofdata,though

notalwaysavailableindevelopingcountries,isconsideredasidealforidentifyinglocationsofpotentialjobopportunities,especiallyintheformalsector.Atotalof1,094landparcelswere

identified,rangingfrom0.007hato37ha(Figure9).Usingtheavailableadministrativefirmdataatacommunelevel,itwasconfirmedthatthesemeasuredlandareasarehighlycorrelatedwiththenumberofregisteredbusinessesinthesamearea.Theconnectivitydeterrencedimisagain

measuredbytheshortesttriptimebetweenlocationianddestinationm.

Apotentialdrawbackisthatinformaljobsareprobablyignored,whicharetypically

widespreadallovertheplace.Toaddressthisissue,wealsousetherecentlydevelopedglobal

dataset,OpenBuildings2.5DTemporalDataset(Sirkoetal.,2021),whichisderivedfromthe

Sentinel-2satelliteimageryandcontainsspatialdataofbuildingpresenceandheightsatabout4-meterresolution(Figure10).Unfortunately,theclassificationofbuildingsbytypehasnotyet

beenidentified.Nonetheless,inmanyAfricancountries,buildingstypicallyhavelowerheights,andinformalemploymentiswidespreadeverywhere.Thus,allbuilt-upareasareassumedas

potentialjobplaces.Weonlyusedatawithaconfidencelevelofbuildingpresenceexceeding75percent.Forcomputationalsimplicity,thebuildingdataisspatiallyaggregatedtoa200-meter

resolution(Figure11).Eachcellisconsideredasajobplace,andthebuildingareawithineachcellisusedasaproxyofYm.AsthetruejoblocationsmightbesomewherebetweenthetwomapsshowninFigure9andFigure11,wewillcomparethetworesults.

4Thatis,Σmymf(dim)wheref()isadecayfunctionoftraveltimeorcosts.

-13-

Figure9.Commercialareasandnumberoffirms

Figure10.OpenBuildingsDataforAntananarivo

Source:OpenBuildings2.5DTemporalDataset

Figure11.DistributionofbuildingsinAntananarivo

Source:BasedonOpenBuildings2.5DTemporalDataset

-14-

Amongabout2,300householdssurveyedbytheMULS,sixjobaccessibilitymeasurementsarecomparedinTable1.

5

Asdiscussed,thereportedcommutetime,denotedbyTIMEObserved,

maybeunderestimated.Theproximitytothenearestminibusroute(TIMEBus)isonly3minutes.Thestraight-linedistancetotheCBD(KMCBD)isonaverage4.3km.Thetraveltime(TIMECBD)is28minutes.However,noteveryonecommutestotheCBD.Withothercommercialareastakenintoaccount,theweightedaveragecommutetime,denotedbyTIMEComBuild,isincreasedto41

minutes.Furthermore,includingallbuildinglocations,thecommutetime,TIMEAllBuild,isabout44minutes.

Notsurprisingly,allmeasurementsarepositivelycorrelatedwithoneanotherexceptfor

TIMEObserved(Table2).Thisisbecauseofthezero-inflatedcommutetimereportedinthe

householdsurvey.Ontheotherhand,jobaccessibilitymeasuredbasedonthelandusemap,

TIMEComBuild,showshighcorrelationwiththeonederivedfromtheOpenBuildingsdataset,

TIMEAllBuild.Thecorrelationcoefficientis0.98,suggestingthatthelatteropen-sourcedata

approachisreliableforevaluatingjobaccessibilityinAntananarivoandpotentiallyother

developingcitieswithsimilarcharacteristics.Thisisanimportantfindingofthispaper.Thereisalsoacorrelationbetweenproximitytominibusroutesandotheraccessibilitymetrics.Buttransitaccessibi