Concentration, agglomeration and the size of plant

Concentration,agglomerationandthesizeofplantsMirenLafourcadea,GiordanoMionb,⁎aUniversitéd'Évry(IUT-GLTDepartment)andParis-JourdanSciencesEconomiques(UMR8545CNRS-EHESS-ENPC-ENS),PSE,48boulevardJourdan,75014Paris,FrancebCORE-UCLandFNRS,CORE,VoieduRomanPays34,B-1348LLN,BelgiumAccepted20April2006Availableonline11September2006AbstractThispaperinvestigateswhetherthegeographicdistributionofmanufacturingactivitiesdependsonthesizeofplants.UsingItaliandata,wefind,asinKim[Kim,S.,1995.Expansionofmarketsandthegeographicconcentrationofeconomicactivities:thetrendsinU.S.regionalmanufacturingstructure,1860–1987,QuarterlyJournalofEconomics110(4),881–908.],HolmesandStevens[Holmes,T.J.,andStevens,J.J.,2002.Geographicconcentrationandestablishmentscale,ReviewofEconomicsandStatistics84,682–690.],andHolmesandStevens[Holmes,T.J.andStevens,J.J.,2004.SpatialdistributionofeconomicactivitiesinNorthAmerica,in:J.V.HendersonandJ.F.Thisse,eds.,HandbookofRegionalandUrbanEconomics,Vol.4,(Elsevier-NorthHolland,Amsterdam).],thatlargeplantsaremoreconcentratedthansmallplants.However,consideringdistance-basedpatternsviaspatialauto-correlation,wefindthatsmallestablishmentsactuallyexhibitagreatertendencytobelocatedinadjacentareas.Theseapparentlycontradictoryfindingsraiseameasurementissueregardingco-locationexternalitiesandsuggestthatlargeplantsaremorelikelytoclusterwithinnarrowgeographicalunits(concentration),whilesmallestablishmentswouldratherco-locatewithinwiderdistance-basedclusters(agglomeration).Thispictureisconsistentwithdifferentsizeplantsengagingindifferenttransport-intensiveactivities.©2006ElsevierB.V.Allrightsreserved.JELclassification:C21;L11;R12;R30;R34Keywords:Concentration;Spatialauto-correlation;Heterogeneousfirms1.IntroductionEconomists,geographersandhistoriansshareaconsiderableinterestinanalyzingthecausesofregionalspecialization.Amongthemyriadofdeterminantswhichhavebeenexplored,particularRegionalScienceandUrbanEconomics37(2007)46–68⁎Correspondingauthor.Tel.:[phone];fax:[phone].E-mailaddresses:[e-mail](M.Lafourcade),[e-mail](G.Mion).URL's:(M.Lafourcade),~mion(G.Mion).0166-0462/$-seefrontmatter©2006ElsevierB.V.Allrightsreserved.doi:10.1016/j.regsciurbeco.2006.04.004attentionhasbeenpaidtoregionalendowmentsorraw-materialintensity,comparativeadvantage,localizationexternalities,or,morerecently,transportcostsandmarketpotential.Inthispaper,wefocusononeparticularaspectofthiscomplexsetofmechanismswhichhasremainedrelativelyunexplored:thesizeofplants.Ourmaincontributionistoinvestigatewhetherthegeographicdistributionofmanufacturingactivitiesisrelatedtoplantsize,andinparticulartolookforaseparateroleforphysicaldistanceinexplaininglocation.WeaddressthisquestionusinganumberofyearsofItaliancensusdataonmanufacturingindustries,atvaryinggeographicandindustrialscales.ByextendingtheempiricalfocustoEurope,wecomplementtheseminalempiricalstudiesofKim(1995),HolmesandStevens(2002),andHolmesandStevens(2004),whichfocusedonNorthAmerica.Kim(1995)reportsapositivecorrelationbetweenconcentrationandbothaverageplantsizeandtheintensityofrawmaterials,acrossU.S.manufacturingindustries.HolmesandStevens(2002)findstrongevidenceofthesamephenomenonwithinindustries:plantslocatedinareaswithhigherindustrycon-centrationarelarger,onaverage,thanthoselocatedoutsidesuchareas,andthisholdsespeciallyforthemanufacturingsector.HolmesandStevens(2002)extendKim'sfindingsbyemphasizingthatthispositiverelationshipisrobusttocontrollingfortheestablishment'sownsizeeffectonconcentration.Weextendtheanalysisoftherelationbetweenplantsizeandspatiallocationpatternstoincludespatialdependenceamonggeographicunits.Anumberofrecentpapers(forexample,Arbia,2001a;DurantonandOverman,2005;MarconandPuech,2003)haveunderlinedsignificantdifferencesintheconcentrationpatternsobtainedfromcontinuousdistance-basedmeasurescomparedtomorestandardindicatorsliketheindexproposedbyEllisonandGlaeser(1997).Sincelaborproductivityispositivelyrelatedtoemploymentdensity(CicconeandHall,1996;Ciccone,2002),aclearerpictureofhowestablishmentdensityvarieswithinindustriesandofhowthisdependsonthenumberofemployeesremainshighontheregionalpolicyagenda.Althoughourapproachbuildsonthisrecentrevivalofdistance,wecontinuetoworkwithadiscretevisionofspaceasafinitenumberofspatialunits(whichcanbeaggregatedtovariousdegrees).WecanthereforecomplementtheEllisonandGlaeserconcentrationindex,whichmeasuresthestrengthofco-locationspilloverswithineachgeographicunit,withanindicatorofspatialauto-correlation(theMoranindex)thataccountsforpossibledistance-basedco-locationpatternsacrossgeographicalunits.Spatialauto-correlationexistsif,foraparticularindustry,dataonthelocationofplantsinregioniis‘linearly’informativeaboutthelocationofotherplantswithinthesameindustryinregions‘close’toi.1Spatialauto-correlationasafeatureoflocationdecisionshasreceivedrelativelylittleattentioncomparedtoconcentrationandcanberelatedtotheminimizationoftransportcosts,asintheNewE