Anselin.spatial-econometrics

SPATIALECONOMETRICSLucAnselinBrutonCenterSchoolofSocialSciencesUniversityofTexasatDallasRichardson,[e-mail]26,1999ABSTRACT:Spatialeconometricmethodsdealwiththeincorporationofspatialinteractionandspatialstructureintoregressionanalysis.Thefieldhasseenarecentandrapidgrowthspurredbothbytheoreticalconcernsaswellasbytheneedtobeabletoapplyeconometricmodelstoemerginglargegeocodeddatabases.Thereviewpresentedinthischapteroutlinesthebasicterminologyanddiscussesinsomedetailthespecificationofspatialeffects,estimationofspatialregressionmodels,andspecificationtestsforspatialeffects.SPATIALECONOMETRICSLucAnselin1.IntroductionSpatialeconometricsisasubfieldofeconometricsthatdealswiththetreatmentofspatialinterac-tion(spatialautocorrelation)andspatialstructure(spatialheterogeneity)inregressionmodelsforcross-sectionalandpaneldata[PaelinckandKlaassen(1979),Anselin(1988a)].Assuch,thefieldissimilartogeostatisticsandspatialstatistics,whichformanimportantpartofthemethodologi-caltoolboxofthephysicalscientist[e.g.,Ripley(1981),Cressie(1993)].However,spatialecono-metricsisdistinctfromspatialstatisticsinthesamesenseaseconometricsisdistinctfromstatisticsingeneral.Thisalbeitsubtledistinctionboilsdowntothecentralroleattributedtothetheoreticalmodelratherthanthedatawhenitcomestodictatingthetypesofspecificationsthatareofinterestinspatialeconometrics.Afocusonlocationandspatialinteractionhasrecentlygainedamorecentralplacenotonlyinappliedbutalsointheoreticaleconometrics.1Inthepast,modelsthatexplicitlyincorpo-rated“space”(orgeography)andthereforeapplicationsofspatialeconometricswereprimarilyfoundinspecializedfieldssuchasregionalscience,urbanandrealestateeconomicsandeco-nomicgeography[e.g.,recentreviewsinAnselin(1992a),AnselinandFlorax(1995a),AnselinandRey(1997),Paceetal.(1998)].However,morerecently,spatialeconometricmethodshaveincreasinglybeenappliedinawiderangeofempiricalinvestigationsinmoretraditionalfieldsofeconomicsaswell,including,amongothers,studiesindemandanalysis[Case(1991)],interna-tionaleconomics[Aten(1996)],laboreconomics[Topa(1996)],publiceconomicsandlocalpublicfinance[Case,RosenandHines(1993),Murdoch,RahmatianandThayer(1993),Holtz-Eakin(1994),Brueckner(1998)],andagriculturalandenvironmentaleconomics[BenirschkaandBin-kley(1994),Murdoch,SandlerandSargent(1997),NelsonandHellerstein(1997),BellandBocks-tael(1999)].Inaddition,thereisagrowingmethodologicalliteratureineconometricsthatdealswithalternativemodelspecifications,estimatorsandteststatisticsgearedtomodelsthatincor-poratethe“geography”ofthedata[recentexamplesinclude,amongothers,Conley(1996),DriscollandKraay(1998),PinkseandSlade(1998),KelejianandPrucha(1999a)].1Theoriginsofthisreactiontotheso-calledAnglo-SaxonfocusineconomicsonthetimedomaincanbetracedtoIsard(1956).—2—Thisnewattentiontospecifying,estimatingandtestingforthepresenceofspatialinter-actioninthemainstreamofappliedandtheoreticaleconometricscanbeattributedtotwomajorfactors.Oneisagrowinginterestwithintheoreticaleconomicsinmodelsthatmoveawayfromtheatomisticagentasadecisionmakeractinginisolationtoanexplicitaccountingfortheinter-actionofthatagent(itspreferences,utility,etc.)withotherheterogeneousagentsinthesystem.Thesenewtheoreticalframeworksspecifyandstudy“direct”interactionbetweenagents,intheformofsocialnorms,neighborhoodeffects,copy-cattingandotherpeergroupeffects,andraiseinterestingquestionsabouthowtheindividualinteractionscanleadtoemergentcollectivebehaviorandaggregatepatterns.Examplesofsuchmodelsarefoundinthenewmacroeconom-icsofAoki(1994,1996),intheoreticalmodelsofsocialinteraction[BrockandDurlauf(1995),Akerlof(1997)],interdependentpreferences[AlessieandKapteyn(1991)],modelsofevolvingtradingstructures[Ioannides(1990,1997)],neighborhoodspillovereffects[Durlauf(1994),Borjas(1995),Glaeser,SacerdoteandScheinkman(1996)],andyardstickcompetition[BesleyandCase(1995),BivandandSzymanski(1997)].Theseframeworksalsoformsomeoftheunderpinningsforempiricalmodelsthatencompassstrategicinteractionbetweenagents[e.g.,Case,RosenandHines(1993),Murdoch,SandlerandSargent(1997),Brueckner(1998)].Muchofthisliteratureisinspiredbyprinciplesdevelopedinstatisticalmechanics,suchasthestudyofinteractingparticlesystemsandrandomfieldmodels[forareview,seeDurlauf(1997)],which,interestingly,alsoformsabasisforsomeofthespatialMarkovfieldmodelsdevelopedinspatialstatistics[e.g.,Cressie(1993,Ch.6)].Relatedtothisstrandintheliteratureistherevivedemphasisonthespa-tialaspectsofMarshallianexternalities,agglomerationeconomiesandotherspilloversthatarecentraltotheneweconomicgeographyreflectedintheworkofArthur(1989),Krugman(1991a,1991b,1998),Glaeseretal.(1992)andothers.Parallelingandcomplementingthistheoreticalmotivation,i.e.,theneedforamethodol-ogytohandlespatialmodels,therehasalsobeenaseconddriverbehindtheincreaseindemandfortechniquestodealwithspatialdatafromapractical,appliedperspective.Irrespectiveofthenatureofthemodelofinterest(whetherspatialornon-spatial),theexplosivediffusionofgeo-graphicinformationsystems(GIS)technologyandtheassociatedavail