The$Pennsylvania$State$University$ The$Graduate$School$ College$of$the$Liberal$Arts$ $ $ $

Transcription

1 ThePennsylvaniaStateUniversity TheGraduateSchool CollegeoftheLiberalArts RACE%AND%PLACE:%INVESTIGATING%RACIAL%DIFFERENCES%IN%THE%SPATIAL%% % PATTERNING%OF%MORTALITY%THROUGH%SOCIAL%CAPITAL%THEORY% % % % % AThesisin SociologyandDemography by ChristopherPrather 2013ChristopherPrather SubmittedinPartialFulfillment oftherequirements forthedegreeof MasterofArts December2013

2 ii ThethesisofChristopherPratherwasreviewedandapproved*bythefollowing: GlennFirebaugh RoyC.BuckProfessorofAmericanInstitutions ProfessorofSociologyandDemography ThesisAdviser BarrettLee ProfessorofSociologyandDemography StephenMatthews AssociateProfessorofSociology,AnthropologyandDemography CourtesyappointmentinGeography Director,GraduatePrograminDemography JohnIceland ProfessorofSociologyandDemography DepartmentHead,DepartmentofSociologyandCriminology *SignaturesareonfileintheGraduateSchool.

3 iii ABSTRACT Thispapershowsthathealthoutcomes,measuredbymortality,havedistinct spatialpatterning.focusingonthetenleadingcausesofmortalityintheunited States,Iutilizethelensofsocialcapitaltheorytolinkenvironmenttohealth, differentiatingbetweencontextualandcompositionaleffectsandtheirplaceinthe roleofsocialcapital.ialsoanalyzeracialdifferencesinmortality,andconnectsocial capitaltheorywithmortalitytoexaminewhetherthespatialpatterningofmortality isduetospatialprocessesthatoccurinthecauseofdeath,orspatialpatterningof theunderlyingprocessesthatresultinmortality.thisinsightisusedtohelpshed lightonwhytheseracialdifferencesmightexist. % % % % % % % % % % % % % % % % % % % % % % %

4 TABLE%OF%CONTENTS% % ListofTables.. v ListofFigures..vi Chapter1.RACE,PLACE,ANDSOCIALCAPITAL TheLinkBetweenEnvironmentandHealth TheRoleofSocialCapital Compositionvs.Context RacialDifferencesinMortality...8 SpatialPatterningofMortality...9 Chapter2.DATAANDANALYSIS...14 Data Analysis..18 All]CauseMortality.25 HeartDisease..27 Cancer.30 ChronicLowerRespiratoryDisease.32 CerebrovascularDisease.34 Accidents...36 Alzheimer sanddiabetes 39 Nephritis/Nephrosis..41 Influenza/Pneumonia...43 IntentionalSelf]Harm 45 Chapter3.CONCLUSIONANDDISCUSSION.46 Conclusion 46 AppendixA:AdditionalTables 50 AppendixB:Figures1]14 54 References 68 iv

5 v LISTOFTABLES % Table1.GlobalMoran sivaluesforindependentvariablesintheanalysis.. 20 Table2.ResultsofOLSandSpatialRegressionModelsforMortalitydue toallcausesinu.s.counties,2005] Table3.ResultsofOLSandSpatialRegressionModelsforMortalitydue toheartdiseaseinu.s.counties,2005] Table4.ResultsofOLSandSpatialRegressionModelsforMortalitydue tocancerinu.s.counties,2005] Table5.ResultsofOLSandSpatialRegressionModelsforMortalitydue tochroniclowerrespiratorydiseaseinu.s.counties,2005] Table6.ResultsofOLSandSpatialRegressionModelsforMortalitydue tocerebrovasculardiseaseinu.s.counties,2005] Table7.ResultsofOLSandSpatialRegressionModelsforMortalitydue toaccidentsinu.s.counties,2005] Table8.ResultsofOLSandSpatialRegressionModelsforMortalitydue toalzheimer sinu.s.counties,2005] Table9.ResultsofOLSandSpatialRegressionModelsforMortalitydue todiabetesinu.s.counties,2005] Table10.ResultsofOLSandSpatialRegressionModelsforMortalitydue tonephritis/nephrosisinu.s.counties,2005] Table11.ResultsofOLSandSpatialRegressionModelsforMortalitydue toinfluenza/pneumoniainu.s.counties,2005] Table12.ResultsofOLSandSpatialRegressionModelsforMortalitydue tointentionalself]harminu.s.counties,2005] Table13.ICD]10CodesforEachCauseofDeath. 51 Table14.LagrangeMultiplierDiagnosticsforSpatialAutocorrelation 52 Table15.DescriptiveStatisticsforMortality 53

6 vi LISTOFFIGURES Figure1.IndependentVariablesDistribution 55 Figure2.IndependentVariablesDistribution2 56 Figure3.FacilityCrowdingDistribution...57 Figure4.NeighborsHistogram...57 Figure5:CancerDistributionandResiduals..58 Figure6.ChronicLowerRespiratoryDiseaseDistributionandResiduals..59 Figure7.HeartDiseaseDistributionandResiduals...60 Figure8.CerebrovascularDiseaseDistributionandResiduals..61 Figure9.AccidentDistributionandResiduals..62 Figure10.Alzheimer sdistributionandresiduals. 63 Figure11.DiabetesDistributionandResiduals 64 Figure12.NephritisDistributionandResiduals..65 Figure13.InfluenzaDistributionandResiduals..66 Figure14.SuicideDistributionandResiduals...67

7 1 CHAPTER1.RACE,PLACE,ANDSOCIALCAPITAL ThestudyofhealthinequalityintheU.S.atanationalscaleoftenassumes thatthevariablesunderexamination,whethertheyaretheexplanatoryoroutcome variables,donotvarybylocation.or,iflocationisconsidered,itisincludedasa controlforcensusregion,suchassouthorwest.thisstudyaimstoshowthat healthoutcomes,measuredbymortality,havedistinctspatialpatterning,and examineswhetherthisisduetospatialprocessesthatoccurinthecauseofdeath,or spatialpatterningoftheunderlyingprocessesthatresultinmortality.focusingon thetoptencausesofmortalityintheunitedstates,ialsoinvestigatehowthese spatialpatternsresultindifferentmortalityratesforblackandwhitemen,usingthe lensofsocialcapitaltheorytodifferentiatebetweenthecontextualand compositionaleffectsofplace. Whilemanystudieshaveexaminedthelinkbetweensocialandphysical environmentandhealth,fewhaveexaminedtheeffectsofspecificfeaturesandor havehypothesizedspecificwaysthatthesefeaturesmightworktoaffecthealth outcomes(macintyre,ellawayandcummins2002).toaddressthislackofresearch, Iexplorehowthespatialpatterningofmortalityvariesbyraceforblackandwhite menthroughthelensofsocialcapitaltheory,touchingonfivekeypoints:(1)the linkbetweenenvironmentandhealth,(2)theroleofsocialcapital(3)thedifference betweencompositionalandcontextualeffects,(4)racialdifferencesinmortality, and(5)thespatialpatterningofmortality.

8 2 The%Link%Between%Environment%and%Health% Whetheritbethroughthecontextoftheplaceorthecompositionofthe place,itisevidentthatplaceisimportant.untiltheearly1990 s,veryfewstudies attemptedtoconnectfeaturesofthesocialorphysicalenvironmenttohealth outcomesinsociology,geography,orevenepidemiology,andthisdearthof literaturereachesbacktothe1940s(macintyre,ellaway,andcummins2002). Historically,placeeffectshavebeenaddressedaswhatisleftoverinsomeresidual category,comprisingtheeffectsthatcouldnototherwisebeexplainedorcontrolled forintheanalysis.itwasnotuntilrecentlythatthespecificimpactofplaceon healthbegantobestudied(macintyreandellaway2000).beginninginthelate 1980s,butappearingfullforceinthe1990s,researchersbegantoexaminethe connectionsbetweenplaceandhealth,linkingmortalityandmorbidityto neighborhoodandcommunitycharacteristics.forexample,haanetal(1987)find thatthesocio]physicalenvironmentconnectslowsocioeconomicstatustoexcess mortalityevenaftercontrollingforindividualbehaviors,whilewaitzmanandsmith (1998)discoverthatpovertyarearesidenceislinkedwithexcessmortality.Diez] Rouxetal.(1997)additionallyfindthatneighborhoodcharacteristicsareassociated withtheprevalenceofcoronaryheartdisease,whilekennedy(1998)findsa connectionbetweenincomeinequalityandself]reportedhealth,afteraccountingfor personalincome. Thesestudiesindicateanimportantlinkbetweenboththesocialand physicalenvironmentthatanindividualoccupiesandtheirhealthoutcomes.some

9 3 researchhasbeenconductedinanattempttodiscoverwhatspecificallyitiswithin theseexternalsocialcontextsthatresultsindifferencesinmortality.cohen,farley, andmason(2003)findthatcollectiveefficacyaswellasproximitytophysical disorderbothhaveaneffectonprematuremortalityaswellashomicideand cardiovasculardisease.othersocialfactorsthathavebeenlinkedtomortalityare socialsupport(kawachiandkennedy1997),socialcapital(sampson,raudenbush andearls1997;votrubaandkling2005)andpsychosocialattributessuchas anxietyanddepressionaswellasexposuretostress(rossandwu1995;latzetal 2005;SchnittkerandMcLeod2005). % The%Role%of%Social%Capital% Thisstudyoperationalizesthesocialandphysicalenvironmentthroughthe lensofsocialcapital,usingadefinitionofsocialcapitalsimilartoputnam(1993) andportes(1998),stressingmembershipinorganizationsandnetworksandthe benefitsthataccruetoindividualsduetothesememberships,specificallylookingat healthoutcomes.socialcapitalhasincreasinglyemergedasanareaofstudyin healthliterature,withthenumberofarticlespublishedrisingfrom20before1981 toover1,000publishedbetween1996and1999.organizationssuchastheworld Bankhavealsoshownincreasinginterestinpoliciesrelatingsocialcapitaltohealth outcomes(baum2000).onereasonforthisheightenedinterestistheideathat socialcapitaluniquelycapturessocialprocessesthatallowindividualsaccessto resourcesthatcanresultinbetterhealthoutcomes.justlikemanyotherresources,

10 4 socialcapitalisunevenlydistributedinsociety,accordingtofactorssuchasage, race/ethnicity,socialclass,andgender(ferlander2007). Socialcapitalisoneofthemorecommonconceptualizationsofthesocial environmentinhealthresearch,andcanbedefinedinanumberofways.according tobourdieu(1986),socialcapitalis theaggregateoftheactualorpotential resourceswhicharelinkedtopossessionofadurablenetworkof institutionalized relationships, whilecoleman(1990)seessocialcapitalasaproductiverelationship betweenentitiesthat makepossibletheachievementofcertainendsthatwould notbeattainableinitsabsence. Finally,Putnam(1993)haspopularizedsocial capitaltocapture featuresofsocialorganizationssuchastrust,normsand networks thatareableto improvetheefficiencyofsocietybyfacilitating coordinatedactions. Morerecently,Portes(1998)hasattemptedtocombinethese threedefinitionsofsocialcapital,synthesizingittomean theabilitytosecure benefitsthroughmembershipinnetworksandothersocialstructures. Thoughcommonlyoperationalizedasapositiveresource,socialcapitaldoes notalwaysoperateinabeneficialway.insomeinstances,itcanhavenegative consequences.forindividualswithinthecommunitythatsocialcapitaloperates,it canconstrainopportunitiesiftheydonothavetherightconnections,itmightplace excessivedemandsontheirtimeorresources,oritcouldleadtodelinquent behaviorwherecapitalisgainedthroughmembershipislessthanreputablegroups (HaweandShiell2000).Kawachi(1999)exploresthisevenfurther,notingthat negativehealthbehaviorscanbecomereinforcedinclose]knitcommunitieswith highsocialcapital,suchassmokingoralcoholuse.therefore,iconsiderthatsocial

11 5 capitalmightnotresultinonlypositivehealthoutcomes,butmightworktoproduce negativehealthoutcomesaswell,dependingonhowsocialcapitalisoperatingfor eachspecificcauseofdeath. Priorstudiesofsocialcapitalandhealthhaveconnectedsocialcapitaltoself] ratedhealth(kawachietal1999),riskbehaviors(lindstrometal2001),violent crime(wilkinsonetal1998),children swelfare(jackandjordan1999)and developmentaloutcomesforpre]schoolchildren(runyanetal1998).analyzingthe well]knownalamedacountystudy(bellocetal.1972),yenandkaplan(1999)find thatnetofrace/ethnicity,perceivedhealth,smokingbehaviors,bodymassindex, andalcoholintake,lowsocialenvironment,orsocialcapital,isassociatedwith significantlyhighermortalityrisk. Inastudyof6,000malesinFinland,Kaplanetal(1988)comparedtwo groups:thosewithhighsocialconnectednessandthosewithlowsocial connectedness,andfoundthathigherlevelsofsocialsupportreducedmortalitydue tocardiovasculardiseaseby5.0to12.1deathsper1,000,dependingonthe endpointtime.asimilarstudyintheu.s.comparedtheneighboringcommunitiesof Roseto,PAandBangor,PAover50yearsandfoundthatsocialcohesionwas associatedwithareductionof2.9deathsper1,000foroverallmortality(bruhnetal 1966;Egolfetal1992).Lomas(1998)arguesthatthesestudiesprovideevidence thatinterventionstoincreasesocialcapital,socialsupport,andsocialcohesion shouldbeconsideredjustasrelevanttopolicydecisionsasaccesstomedicalcare. %

12 6 Composition%vs.%Context% Withinthesocialcapitalliterature,acommonthemeistherelationship betweencompositionandcontext(bernardetal.2007),whichisusedtodividethe conceptualizationofspaceintoadualityofcompositionaleffectsandcontextual effects.thosewhosupportthecompositionalexplanationfocusonhowhealth outcomesaregeographicallyclusteredaccordingtothecommoncharacteristics sharedbyresidentsofthesamegeographicalarea.someofthemorecommon characteristicsascribedtothecompositionalexplanationareeducation,income,and employment.individualsmay,forexample,choosetoliveinthesameplacebecause theysharesimilarlevelsofpersonalresources,acommonculture,orsimilar languageskills.itisimportanttonotethatthesechoicesarenotalwaysentirely voluntary.whileanaffluentindividualmightchooseaspecificneighborhoodtolive nearpeoplewithsimilarsocioeconomicstatus,alessadvantagedindividualmight berelegatedtoanundesirableneighborhoodduetolackofresources.regardlessof individualagencyinchoosingtheresidentialcontext,thecompositionalviewholds thatindividualswhosharesimilarcharacteristicshaveapropensitytoliveinclose proximitytooneanotherandthistendencyexplainspartoftheassociationbetween placeandhealth(harvey1973).forexample,waitzmanandsmith(1998)findthat excessmortalityisassociatedwithpovertylevelsintheareaofresidenceevenafter controllingforindividualbehaviorssuchassmoking,drinking,exercise,andother characteristics.tocapturecompositioninthisstudy,iincludemeasuresofpercent

13 7 black,percentwithoutinsurance,percentinpoverty,unemploymentrate,and percentofthepopulationovertheageof65. Thecontextualexplanation,ontheotherhand,holdsthatplaceshavecertain ecologicalattributesthataffectthehealthoutcomesofindividualswithinthesame geographicboundaries(macintyre,ellaway,andcummins2002).neighborhood attributessuchaslanduse,cleanliness,andpresenceofparksorpublic transportationexistindependentlyfromthecharacteristicsoftheindividualswho resideintheneighborhood.theseecologicalattributesworktoexplainthe associationbetweenplaceandhealthevenafterindividualcharacteristicsare accountedfor.haanetal.(1987),forexample,findthattherearespecificphysical propertiesofthesocio]physicalenvironmentthatexplaintheassociationbetween lowsocioeconomicstatusandhighmortalitynetofindividualcharacteristicsand behaviors.thisshowsastrongeffectofplaceoverandabovetheeffectsofcommon individualcharacteristics.tocapturecontextinthisstudy,iincludemeasuresof accesstodoctors,accesstomedicalfacilities,facilityexpenditures,andfacilityand doctorcrowding. Inotherwords,thecompositionalviewarguesthatplaceandhealthare connectedindirectlythroughtheattributesofthepeoplecontainedintheplace, whilethecontextualviewarguesthatplaceandhealthareconnecteddirectly throughtheecologicalattributesoftheplace. % %

14 8 Racial%Differences%in%Mortality% Whethertheyarephysicalresourcesthatareapartofthecontextual attributes,suchasaccesstohealthfacilities,orsocialresourcesthatareabyproduct ofthecompositionalattributesofanarea,differentialaccesstoresourcesis frequentlypatternedbyrace/ethnicity.becauseofthis,blackscommonlyfind themselvesdisadvantagedrelativetowhites.inhealthliterature,thisresultsin worsehealthoutcomesforblacks.asof2002,lifeexpectancyforblacksinthe UnitedStateswasfiveandahalfyearslessthanthelifeexpectancyforwhites (Cutler,Deaton,Lleras]Muney2006).Thisgapinlifeexpectancyisfrequently investigatedbydifferencesinindividualcharacteristics,especiallyindividual socioeconomicstatus,andthefactthatmortalityvariesbymeasuressuchas education,income,andoccupationisalreadywellestablished(elo2009).whatis lesscleararetheprocessesthatlinkrace,ethnicity,andsocioeconomicstatusto mortality.assampsonandmorenoff(2005)noteintheiranalysisofracialand ethnicdisparitiesinviolence,belongingtoacertainraceorethnicityisnotthecause ofviolenceormortality.followingthis,receivingaparticularlevelofincomeor educationdoesnotdirectlycausemortalityeither.rather,race,ethnicity,income, education,andotherindividualfactorsactasmarkersforparticularexternalsocial contextsthatbecomeallocateddifferentiallyaccordingtothesefactors. Forthisreason,Icomparemortalityratesofnon]Hispanicblackmalesto mortalityratesofnon]hispanicwhitemales,aimingtodiscoverwhateffectspatial processesaswellascontextualandcompositionaleffectshaveonmortality

15 9 outcomesforbothgroups.whileoveralllifeexpectancyhasrisenbetween1900and 2006from47yearsto78years,thegainsforwhiteshaveoutpacedthegainsfor blacks(cutler,deaton,lleras]muney2006).measuredanotherway,thereareupto 100,000excessdeathsforblacksannuallyrelativetowhites(HummerandChinn 2011).Areaswithhighblackconcentrationsfacesimilaroutlooks;evenafter accountingforeconomicandindividualcharacteristics,areaswithhighblack concentrationarefoundtohavehigherratesofmortalityrelativetoneighborhoods withlowblackconcentration(cubbinetal;deatonandlubotsky2003;geronimus etal2001).thesizeofthegapbetweenthetwoisquitestartling,aswell.while whiteslivingindisadvantagedareasexperiencehighermortalityratesthanthose wholiveinmoreaffluentareas,whitesindisadvantagedareasstillexperience mortalityratesbelowtheaveragemortalityexperiencedbyblacks(geronimusetal 1996).Thisextremedichotomyexemplifiestheeffectoftheunequaldistributionof resources,resultinginthespatialpatterningofhealthinequalities(bernardetal. 2007). Spatial%Patterning%of%Mortality% Asaresultofthelackofstudiesexplicitlyexaminingtheeffectsofsocialand physicalenvironmentalfeaturesonhealthoutcomes,thereislittletheoretical groundworkestablishingwhichvariablesareimportanttoincludeinananalysis. Therefore,researcherscommonlychoosevariablesthatarereadilyavailablerather thantheoreticallyapplicabletocharacterizeanarea(mitchelletal.2000).

16 10 Additionally,thespatialscaleisoftendeterminedinthesameway,resultingin contrastingresultsforareaeffects.thispapercarefullyconsiderstheincluded variablesaswellasthegeographicscaleofanalysisinordertoadvanceamore theoreticallybasedratherthanexploratoryanalysis. Forsocialcapital,thereismuchdebateastotheoperativelevel,aswellasits natureandmeasurement.evenstill,significantquestionsremainabouthowsocial capitalrelatestohealthoutcomes(macinkoandstarfield2001).onecommon critiqueofpriorresearchpointsoutthatarticlesoftenfailtodiscusswhyacertain conceptualizationofsocialcapitalisutilizedinananalysis,orwhysocialcapital shouldbetheorizedtooperateatthechosengeographicscale.manystudiesof socialcapitalalsoignorespatialautocorrelation,assumingthatconditionsinone locationhavenoeffectonconditionsthataffectmortalityinneighboringareas (Cumminsetal2007). Forthisstudy,thegeographicunitofanalysisistheU.S.county.Asnotedby Bernardetal.(2007),Giddens structurationtheorypositsthatneighborhood environments involvetheavailabilityof,andaccessto,health]relevantresourcesin ageographicallydefinedarea. Here,thereisareciprocalrelationshipbetweenthe socialstructureandindividualagency,orcompositionandcontext,withsocial structuresconstrainingthebehaviorsofindividualsandindividualsinturn reproducingoralteringthesesocialstructures(seegiddens StructurationTheory, 1984).TheU.S.countyfitsthisdefinitionofaneighborhoodenvironmentinthat individualsareexposedtosocialandphysicalhealth]relevantresources.although someresearchersarguethatsocialcapitalshouldbeanalyzedatasmaller

17 11 geographicscale,portes(1998)arguesthatintrinsically,thereisnothingwrong withsocialcapitalanalyzedatalargeaggregatedlevelratherthanattheindividual level,itjustrequiresgreatercareinanalysis.infact,inastudyofneighborhoods, Stephens(2008)findsthatindividual ssocialconnectionsarenotprimarilylocated inneighborhoods,andconcludesthatsocialcapitalisbestunderstoodinalarger contextthatincludescitizensacrossmanyneighborhoods.neighborhoodsarenot alwaysthesiteofanindividualssenseofcommunity(szreter2000),whichcanbe, andoftenis,basedinalargergeographicarea,encompassingplaceswherethey work,attendreligiousservices,andparticipateincommunityactivitiessuchas sportsgroups,allpartsofbroadersocialconnections(szreterandwoolcock2004). Additionally,usingsmallgeographicalareastoanalyzesocialcapitalhasresultedin widelyvaryingresultsduetolargevariationsinthemeasures,meaningthat interpretingresultsatsmalllevelsmightbemisleading(kawachietal1999). Toaddresstheissueofspatialautocorrelation,Iexplorethespatial patterningofeachindependentvariable,aswellasthespatialpatterningoftheten leadingcausesofmortality.therearetwospatialprocessesbywhichacounty mightbeaffectedbysurroundingcounties:diffusionandspatialexternalities.baller etal.(2001)andmorenoff(2003)outlineanimportantdistinctionbetweenthese twoprocesses.diffusionmodels,modeledbyspatiallag,wouldmodelmortality outcomessuchasinfectiousdisease,wherethespatialprocessisinherentinthe causeofdeath,suchasthediseasespreadingthroughmechanismslikesocial networksthatcancutacrosscountyboundaries(cohenandtita1999).thespatial lagmodelisrepresentedbytheequation:

18 12 = "# + " + (1) whererepresentsavectorthatcontainstheobservationsofthedependent variable,andareparameters,"isthespatiallylaggeddependentvariablefor theweightsmatrix,xisamatrixcontainingallobservationsontheexplanatory variables,andisavectorofindependentandidenticallydistributederrorterms (Anselin2005).Inpractice,thespatiallagmodelisestimatedusingmaximum likelihoodofaspatialregressionmodelspatiallylaggeddependentvariable. Spatialexternalities,modeledbyspatialerror,areexemplifiedbycausesof mortalitysuchasheartdisease.whileheartdiseasecannotspreadinthesameway asinfectiousdiseases,itcanstillbespatiallyconditioned.thiswouldoccurifthere arespatialprocessesthatproducethemortalityoutcome,suchasstressfromhigh mortalityinneighboringcounties.inthisinstance,thespatialprocessisnot inherentintheoutcomeitself,butratherinthefactorsthatcanleadtotheoutcome. Thespatialerrormodelisrepresentedbytheequation: = " + (2) with = "# + (3) whererepresentsavectorthatcontainstheobservationsofthedependent variable,andareparameters,isamatrixthatcontainsobservationsonthe explanatoryvariables,isthespatialweightsmatrix,isavectorthatcontains spatiallyautocorrelatederrorterms,andisavectorofindependentand identicallydistributederrors(anselin2005).themethodologicalaspectsofboth

19 13 modelsareoutlinedinanselin(1988)aswellinanselinandbera(1998),andthe sameestimationalgorithmisusedinbothmodels,showninsmirnovandanselin (2001).

20 14 CHAPTER2.DATAANDANALYSIS Data% SimilartoSparksandSparks(2010),Iexaminecounty]levelmortalityrates usingspatialanalyticmethods,stressingtheimportanceofmodelchoicebasedon theoryastohowtheunderlyingprocessesoperateontheoutcome.sparksand Sparks(2010)ultimatelyconcludethataspatialerrormodelbestfitsthedata, notingthatthealternativespatiallagmodelwouldsuggestadiffusiveprocessof mortality,whilethespatialerrormodelsuggestautocorrelationamongomitted variablesinthemodel.thisseemstobereasonableforall]causemortality,butall] causemortalityiscomprisedofmanydifferentcauses,someofwhichmightbe bettermodeledbyadiffusiveprocess.inordertotestthis,aswellastheassociation betweencontext,composition,andsocialcapitalwithcause]specificmortality,irun aseriesofordinaryleastsquaresandspatialmodelsforthetoptencausesof mortalityintheunitedstates.ialsotestwhethertheseprocessesoperatethesame byrace,includingseparatemodelsfornon]hispanicblackandwhitemales. Formyanalysis,Iconsideradefinitionofsocialcapitalmoreinlinewith Putnam(1993)andPortes(1998),wherefeaturesofsocialorganizationsfacilitate coordinationandresultinaresourcefromwhichindividualsofacountycandraw. Inthecaseofhealthoutcomes,thismightbeassimpleasemotionalormonetary supportorascomplexasreductionofbehaviorsthatresultinnegativehealth outcomesandencouragementofbehaviorsthatresultinpositivehealthoutcomes.

21 15 ThemeasureofsocialcapitalIuseisanindexofsocialcapitaldevelopedby Rupasingha,GoetzandFreshwater(2006),andproducedviaprincipalcomponent analysisfromalargevarietyofvariables.thefirstcomponentaggregates participationinreligiousorganizations,professionalorganizations,labor organizations,bowlingcenters,physicalfitnessfacilities,publicgolfcourses,and sportsclubs,managersandpromoters.thesearedividedbythenumberappearing per10,000populationandagaindividedby10toproducethefirstfactorofthe index.thesecondcomponentisameasureofvoterturnout,thethirdisthecensus responserateforthecounty,andthefourthcomponentisthenumberofnon]profit organizationswithadomesticapproachper10,000people.thesefourfactorsare combinedwithafifthfactor,acountofthepopulationofthecounty,toproducethe finalsocialcapitalindex. Withregardstocontextandcomposition,thecompositionalvariablesthatI includearethepercentageofthepopulationthatisblack,percentagewithouthealth insurance,percentageinpoverty,countyunemploymentrate,percentageofthe populationthatisover65yearsold,andacontrolforpopulationdensity.the percentageofthepopulationthatisblackisincludedasanindicatorofblack/white diversity,drawingonraciallyconcentrateddisadvantageliterature(masseyand Denton1987;MasseyandDenton1993).DeatonandLubotsky(NBERWorking Paper)notethatpercentageblackandhighermortalityarecorrelatednot necessarilyduetotheconnectionbetweenlowerblackincomesandhigher mortality,butratherduetothefactthatwhitemortalityratesarealsohigherin areaswithahighpercentageblack.percentageofthepopulationinpovertyand

22 16 countyunemploymentrateareincludedasmeasuresoftheoveralldisadvantage. Independentofindividualcharacteristics,low]incomeandhigh]deprivationareas arefoundtoexperienceworsehealthstatus(howardetal2000).additionally, individualsarelesstrustingwhereincomedifferencesaregreat,resultinginless cohesivesocialrelations(wilkinson1999).finally,thepercentageofthepopulation thatisover65isimportantforcausesofdeathsuchasalzheimer s,whichare primarilyaconditionofoldage. Contextualvariablesincludedinthemodelareprimarilyindicatorsofhealth careandhealthcareaccess,andincludeaccesstodoctors,facilityexpenses,hospital access,facilitycrowdinganddoctorcrowding.accesstodoctorsandhospitalsare measuresofhoweasyitisforanindividualtogettoamedicalfacilityorgetan appointmenttoseeadoctor.themeasureofaccesstohospitalsaccountsforthe numberofhospitalstherearepersquareperson]mileandisintendedtoreflectthe averageindividual saccesstoahospitalwithintheircounty.accesstodoctorsisa measureofhowmanylicensedandactivemdstherearepersquareperson]mile andreflectsaccesstoknowledgeablestaff.whilesomeareasmighthavemedical facilities,theymaynotnecessarilybestaffedbylicensedmds,relyingratheron nursepractitioners.facilitycrowdingisameasureofthenumberofannual inpatientdaysperbedinafacility.forthisvariable,avalueof365wouldindicate thateverybedwasfilledeverysingledayoftheyear.doctorcrowdingismeasured bythenumberofoutpatientvisitspermdineachfacilityandindicateshowmuchof adoctor stimeisspentonoutpatientprocedures.thesetwovariablescapturenot justthephysicalproximitytohospitalsandmedicalprofessionals,buttheactual

23 17 availabilityofspaceandtimeformedicalcare.facilityexpensesaretheactual expendituresofthemedicalinstitutionsonmedicalcare,includingequipmentand payroll.thisisdividedbythenumberofpeopleservedinthecountytocapturethe annualcostofprovisionsofmedicalcare. Allofthecontextualandcompositionalvariablesarefromthe2005]2009 AreaResourceFilesprovidedbytheU.S.DepartmentofHealthandHumanServices, andallvariablesareaveragedovertheentireperiodtoprovideasinglestable measure.currentlythereare3,144countiesintheu.s.,anddataareincludedfor 3,143ofthesecounties.Theexcludedcounty,Broomfield,Colorado,wasestablished in2001asanincorporatedcity]countyandwasnotincludedinthesocialcapital measure,soisnotconsideredinthisanalysis.theregressionanalysesconsiderall countiesthatcontainalargeenoughnumberofblackandwhitemalemortalitysoas toprovidereliableratesasdeterminedbytheu.s.censusbureau,whichisover three]quartersofthecountiesintheu.s.becausethesocialcapitalmeasureincludes afactorforthepopulationcountofeachcounty,idonotweightthecountiesby populationsize.oneconcernintheanalysisistemporalvariationinexposure,orat whattimeperiodsexposuretodifferentfactorsmightmatter.whilethisisanissue, SampsonandMorenoff(2006)notethatinequalityisquitedurable,findingastrong correlation(r=.87)betweenneighborhoodpovertyin1970and1990anda moderatetostrongcorrelation(r=.73)betweencollectiveefficacywhenmeasured attheneighborhoodlevel.theyalsonotethatevenwhenareasdochange,the relativerankordertendstoremainthesame.thismeansthatwhileexposureover timemightbeaproblem,itislikelyaminoroneasindividualswithinthesame

24 18 areaswouldnotlikelyexperiencewidelyvaryingexposuresregardlessofthetime dimension. Thedependentvariable,causeofdeath,comesfromtheCDCCompressed MortalityFileandisbrokendownbyrace,gender,andHispanicorigin,andage] standardizedto2000censuswhitemalefiguresinordertocontrolfordifferences intheagestructure.forthisanalysis,ifocusonthecomparisonbetweennon] Hispanicwhiteandnon]Hispanicblackmalesinordertosimplifythenumberof comparisonsnecessary.futureresearchshouldconsidergenderdifferencesaswell. DuetothevaryingsizeofU.S.counties,Iaveragemortalityratesoverthe2005] 2009periodtoproducestableestimatesandcapturemortalitythatmightnothave occurredineveryyear.forexample,acountymighthaveoneblacknon]hispanic maledeathduetodiabetesoneyear,butthreethenext,whichwouldappeartobea triplingofthedeathratewhenanincreaseoftwodeathsformoneyeartothenext inamorepopulouscountywouldhardlyregisterasanincreaseatall.additionally,i restricttheanalysistotheleadingtencausesofdeathintheu.s.toensurethatthere willbeanadequatenumberofdeathssoastonotbiastheresults.thecausesof deatharedeterminedusingicd]10classifications,andthespecificicd]10codesare includedinappendix(tobeadded). Analysis% Thefirststepinmyanalysisistomakesurethatmymodelisproperly specifiedandthatmyvariablesdonotviolateanyimportantassumptionsofthe

25 19 regressionmodelsiwillbeusing.ifoundthatunemploymentrateswereslightly right]skewed,butnotenoughtobeaconcern.theindexforsocialcapitalwasfound tohaveoneoutlier,withedgefield,southcarolinahavingavalueforthesocial capitalindex68%higherthanthenexthighestcounty,butwithnoreasonto suspectthatitisanerrorinthedataileaveitintheanalysis.accesstomdsand hospitalaccessaremoderatelycorrelated,butiarguethattheycaptureslightly differentprocesses,andleavebothintheanalysis.manyofthestaffmembersina hospitalarelicensedmds,buttherearealsomdsatsmallerfacilitieslikefamily medicaloffices,whichmightbemoreprominentinruralcountiesandshouldnotbe excludedduetothemoderatecorrelation.ialsodetectminorheteroskedasticity, anduserobuststandarderrorstocontrolforthis. Toinformmyuseofspatialautoregressivemodels,Itestforthepresenceof autocorrelationinmyvariables.ifcountieswithhighmortalityratesbycauseare surroundedbyothercountieswithhighmortalityrates,forexample,orhavelow ratessurroundedbylowrates,thiswillsuggesthighautocorrelationinthevariable. Todeterminewhetherthisexists,IcalculateMoran siforbothmydependentand independentvariables.

26 20 Table&1.&Global&Moran's&I&Values&for&Independent&Variables&in&the&Analysis& ObservedI E[I] Z[I] Social&Capital& Composition& PercentBlack PercentWithoutInsurance PercentinPoverty UnemploymentRate PercentofPopulationover PopulationDensity Context& AccesstoDoctors FacilityExpenses HospitalAccess FacilityCrowding DoctorCrowding Z[I]=theobservedIvalue'sstandarddeviateundertheH0ofnoassociation;E[I]= theexpectedvalueofmoran'si. ThecalculationofMoran si%isbasedon999randompermutationsof thedistributionofeachvariablewhichassumethatthereisnospatial autocorrelationpresent.theobserveddistribution(observedi)inthencompared totheresultachievedfromtherandompermutationproceduree[i],andsignificant deviationsintheobserveddistributionsuggestspatialautocorrelation.thevalues formoran sirangefrom]1to1,withnegativevaluesindicatingnegativespatial autocorrelationandpositivevaluesindicatingpositivespatialautocorrelation (Anselin,1995).TheglobalMoran sivaluessuggestthattheremightbespatial autocorrelationpresentinmypredictorvariables,withtheexceptionofhospital access.ialsomapthelocalindicatorofspatialautocorrelation(lisa)statisticina clustermaptofurthercheckforautocorrelationandclustering(seeappendix).this statisticisaglobalmeasureofautocorrelation,andmeasuresthesignificanceof

27 21 spatialclustering(anselin,1995).countiesthatarefoundtobesignificantwitha highvaluethataresurroundedbyotherhighvalues(high]highclustering)arered, whilecountiesthathavealowvalueandaresurroundedbyothercountieswithlow values(low]lowclustering)arecoloredblue.forthemaps,iuseaspatialweights matrixbasedonthe10nearestneighboringcountiestogetaviewofthelarger spatialstructureacrosstheu.s.,whilefortheanalysisiuseaqueen]basedfirst ordercontiguityspatialweightsmatrix.usingthequeencriterionmeansthat countiesareconsideredneighborsiftheyshareanypointincommon,whetheritis anedgeoracorner,asopposedtoarookcriterionthatconsiderscounties neighborsonlyiftheyshareacommonedgeboundary.inmapping,idiscoverthat therearefourcountiesthathavenoneighborsinmyweightingscheme,astheyare islands.ithereforedeletethecountiesofsanjuan,wa,dukes,ma,nantucket,ma andrichmond,ny. Ifindthatthesocialcapitalindexisclusteredwithhigh]highvaluesinthe Midwestandlow]lowvaluesinthesouth/southeast.Percentblackisclusteredhigh] highinthesouthandlow]lowinthewestandmidwest,andpopulationdensityis similarlyclusteredbutwiththehigh]highvaluesinthenortheast.percentuninsured isclusteredhigh]highinthenorthwest/midwestregionwithspotsoflow]low clusteringthroughouttheotherregions.facilityexpensesandcrowdingare clusteredsimilarlytooneanother,withhighlow]lowclusteringinthesouthwest andhigh]highinthenortheast.formortality,deathsduetocancerareclustered low]lowinthemidwestandwest,andhigh]highinthesouthandthemoresouthern regionsofthenortheast.

28 22 Deathsduetoinfluenza/pneumoniaandchroniclowerrespiratorydiseases areclusteredlow]lowinthemidwest/southwestandhigh]hightowardstheeast. Heartdiseasehasasimilarlow]lowclustering,buthigh]highclustering concentratedmoreinthesouth.cerebrovasculardisease,morecommonlyreferred toasastroke,isclusteredlow]lowinthemidwesttrailingtowardsthesouth,with high]highpocketsonthewestcoastandscatteredintheeasternhalfofthecountry. Accidentsseemtofollowaclusteringpatternverysimilartothatofcerebrovascular disease.alzheimer sisclusteredhigh]highonthewestcoastaswell,withrandom clustersofhigh]highmortalitythroughoutthecountry.similarly,diabetesis clusteredhigh]highonthewestcoast,butisalsoclusteredhigh]highinnearlyallof thenortheast,extendingwesttowardthegreatlakes.nephritisisprimarily clusteredlow]low,occurringinthemidwest,andfinally,suicideisclusteredhigh] highonthewestcostandlow]lowinthemidwest.theseclustermapscanallbe foundintheappendix. Havingmappedmyvariables,IproceedtotheOLSandspatially autoregressivemodels.foreachsetofmodels,iusegeoda s(anselin,syabri,and Kho2006)diagnosticsforspatialdependenceinordertodeterminewhethera spatialmodelismoreaccurateforthedata,andifso,whetheraspatiallagorspatial errorbetterreflectstheunderlyingprocesses.itheorizethatcancermightbe modeledbyspatiallagduetothestrongassociationbetweenenvironmental exposuresandincidenceofcancer.similarly,chroniclowerrespiratorydiseaseand influenza/pneumoniamightbebestmodeledbyspatiallag.ontheotherhand,i hypothesizethatheartdisease,cerebrovasculardisease,accidents,alzheimer s,

29 23 diabetesandsuicidewillbebettermodeledbyspatialerror,asthespatial patterningislikelytheresultofun]capturedspatialautocorrelationinthevariables orintheunderlyingprocesses.

30 24 OrdinaryLeastSquares SpatialLag SpatialError WhiteMale BlackMale WhiteMale & BlackMale White&Male& Black&Male& Social&Capital& = * = *** = *** Composition& PercentBlack =0.181 =0.662 =0.503 * =0.980 =0.395 =0.775 PercentWithoutInsurance =5.120 *** =5.816 =4.158 *** = =3.828 PercentinPoverty *** *** *** *** *** *** UnemploymentRate *** = ** = *** =2.114 PercentofPopulationover * * * ** PopulationDensity =0.004 ** =0.003 * =0.005 ** Context& AccesstoDoctors FacilityExpenses =0.221 *** =0.249 * =0.163 *** =0.189 =0.090 ** =0.202 HospitalAccess =0.263 = = =5.788 FacilityCrowding ** * DoctorCrowding * * AkaikeInfoCriterion & 24026& *p<.05**p<.01***p<.001

31 25 All#Cause)Mortality) ) Beforebreakingmortalitydownbycause,itisusefultofirstconsideroverall mortalityintheunitedstates.forbothblackandwhitemales,thespatialerrormodel emergedasthebestfit.thisconclusionfitstheoreticallyaswell,asallacausemortality likelyisspatiallypatternedduetounmeasuredvariablesinthemodelratherthansome sortofdiffusiveprocess.lookingatsocialcapital,itissignificantwithalargecoefficientfor whitemalesregardlessofmodelchoice.thismeansthatregardlessofthespatialprocesses underlyingallacausemortality,itisclearthatsocialcapitalisassociatedwithlower mortalityratesatasignificantlevel.interpretingthespatialerrormodel,percentin povertyandpercentofthepopulationover65aresignificantforbothwhiteandblack males,andareassociatedwithhighermortalityrates,whiletheunemploymentrateis significantonlyforwhitemales.ofthecontextualvariables,facilityexpensesanddoctor crowdingarebothsignificantforwhitemalesonly,withtheformerassociatedwithlower mortalityratesandthelatterassociatedwithhighermortalityrates. ) ) ) ) ) )

32 26 Table&3.&Results&of&OLS&and&Spatial&Regression&Models&for&Mortality&due&to&Heart&Disease&in&U.S.&Counties,&2005B2009& OrdinaryLeastSquares SpatialLag SpatialError Modified SpatialError WhiteMale BlackMale WhiteMale BlackMale White&Male& BlackMale Black&Male& Social&Capital& @ Composition& PercentBlack * *** *** *** PercentWithoutInsurance *** ** *** * *** * PercentinPoverty *** ** *** UnemploymentRate *** *** *** * PercentofPopulationover *** *** *** *** *** *** *** PopulationDensity * * Context& @ @ @0.092 DoctorCrowding *** *** & & AkaikeInfoCriterion & && & *p<.05**p<.01***p<.001

33 27 Heart&Disease& Forheartdisease,theleadingcauseofdeathintheUS,theLagrangeMultiplier(LM) issignificantandsimilarlyvaluedforbothlaganderrormodels(seetable16inappendix). EventherobustLMissignificantforboth.Forblackmales,however,thelagspecificationis amuchhighervaluethantheerror,althoughbotharestillsignificant.thissuggeststhat thespatialprocessesmightworkdifferentlyforblackandwhitemales.astheerrormodel makesmoretheoreticsense,iincludeamodifiedspatialerrormodel,whereiinclude spatiallylaggedvariablesforallindependentvariablestoexplicitlymodelthespatial processes.indoingso,thespatialerrormodelforblackmalesbecomesmoresimilarin diagnosticstothespatiallagmodelthatwaspreviouslyindicatedasthebetterchoice, suggestingthattheerrormodelmighttrulybebestforheartdisease.themodelchoiceis boldedintable3. Regardlessofthemodelchoice,itisevidentthatbothcompositionalandcontextual variablesareimportantforheartdiseasemortality.forwhitemales,percentwithout insurance,unemploymentrate,andpercentofthepopulationover65areallassociated withhighermortalityrates.accesstodoctors,facilitycrowding,anddoctorcrowdingare allsignificantlycorrelatedwithheartdiseasemortality,thoughthecoefficientsarequite small.allthreevariablesareassociatedwithreductionsinmortality,indicatingthat greateraccesstocrowdeddoctorsandhighnumbersofoccupiedbedsarecorrelatedwith lowermortalityrates.whiletheformermakessense,itissurprisingthatcountieswith crowdedfacilitiesalsohavelowermortalityrates.onereasonforthismightbethat

34 28 facilitieswhichexperiencecrowdingaremoremotivatedtomovepatientsthroughthe system,reducingtheirexposuretothehospitalenvironmentandlesseningtheriskof secondaryinfections.forblackmales,percentblack,percentwithoutinsurance,percentin poverty,andpercentofthepopulationover65aresignificantoutofthecompositional variables,whilefacilityexpensesanddoctorcrowdingaresignificantforthecontextual variables. Lookingatthespatialerrormodelforwhitemales,andthemodifiedspatialerror forblackmales,socialcapitalisnonqsignificantforboth,indicatingthatthelevelsofsocial capitalinacountydonotreduceorincreasemortalityratesduetoheartdisease.

35 29 Table&4.&Results&of&OLS&and&Spatial&Regression&Models&for&Mortality&due&to&Cancer&in&U.S.&Counties,&2005A2009& OrdinaryLeastSquares SpatialLag SpatialError WhiteMale BlackMale White&Male& & Black&Male& WhiteMale BlackMale Social&Capital& =5.601 ** =9.773 * =3.907 =7.731 =4.550 =9.050 * Composition& PercentBlack = *** = *** = *** PercentWithoutInsurance *** *** *** PercentinPoverty =0.053 =4.515 *** =0.084 =3.352 *** =0.003 =3.802 *** UnemploymentRate = = =1.143 PercentofPopulationover *** *** *** *** *** *** PopulationDensity Context& AccesstoDoctors = = = FacilityExpenses = *** = *** = *** HospitalAccess =4.012 *** =1.926 =3.990 *** =1.753 =3.896 *** =1.819 FacilityCrowding = = =0.034 DoctorCrowding =0.007 *** =0.006 *** =0.006 *** & & & & & AkaikeInfoCriterion & && && 36437& && *p<.05**p<.01***p<.001

36 30 Cancer' Forcancer,thediagnosticsclearlypointtowardsthespatiallagmodelbeing thebestspecificationforbothwhitemalesandblackmales.asthisfitsboth quantitativelyandtheoretically,iinterprettheresultsofthespatiallagmodels (boldedintable4). Again,bothcompositionalandcontextualvariablesaresignificant,following thesamepatternasforheartdiseasemortality.forwhitemales,percentwithout insuranceandpercentofthepopulationover65aresignificantofthecompositional variables,andbothareassociatedwithincreasedmortalityrates.hospitalaccessis theonlysignificantoneofthecontextualvariables,andisassociatedwithdecreased mortality.likewhitemales,percentofthepopulationover65issignificantand positiveforblackmales.unlikewhitemales,percentblackandpercentinpoverty aresignificant,andbothareassociatedwithincreasedmortality.ofthecontextual variables,facilityexpensesareassociatedwithhighermortalityandaresignificant, whiledoctorcrowdingisassociatedwithlowermortalityforblackmales. SocialcapitalforwhitemalesisnonLsignificantagainforbothblackand whitemales,meaningthatcancer,likeheartdisease,isnotacauseofmortalitythat isaffectedbycountylevelsofsocialcapitalforeitherblackorwhitemen.

37 31 Table&5.&Results&of&OLS&and&Spatial&Regression&Models&for&Mortality&due&to&Chronic&Lower&Respiratory&Disease&in&U.S.&Counties,&2005C2009& OrdinaryLeastSquares SpatialLag SpatialError WhiteMale Black&Male& White&Male& & BlackMale WhiteMale BlackMale Social&Capital& Composition& PercentBlack *** *** *** *** *** *** PercentWithoutInsurance PercentinPoverty * ** * * * * UnemploymentRate *** ** *** PercentofPopulationover *** *** *** PopulationDensity *** *** *** Context& AccesstoDoctors * * * FacilityExpenses *** *** *** HospitalAccess * FacilityCrowding *** *** *** DoctorCrowding *** *** *** AkaikeInfoCriterion & & &.& &.& 20040& & &.& *p<.05**p<.01***p<.001

38 32 Chronic(Lower(Respiratory(Disease( Thediagnosticsforspatialdependencearequiteclearwhenanalyzingthe regressionforwhitemalechroniclowerrespiratorydiseasemortality.thestandardlmis significantforboththelaganderrorspecifications,butwhentherobustlmisused,the errorspecificationisnolongersignificant.thisindicatesthat,asihypothesize,chronic lowerrespiratorydiseaseismodeledmoreaccuratelybyaspatiallagmodel.whenthe diagnosticsforblackmalesareanalyzed,though,thereverseistrue.neitherthelagor errorspecificationsaresignificant,suggestingthatstandardolsisabetterfit.itherefore interpretthespatialerrormodelforwhitemalesandtheolsmodelforblackmales (boldedintable5). Forwhitemales,allofthecompositionalvariableswiththeexceptionofpopulation densityandpercentwithoutinsurancearesignificantforwhitemales,whilepercentblack, percentinpovertyandpopulationdensityaresignificantforblackmales.ofthecontextual variables,accesstodoctors,facilityexpensesandfacilitycrowdingaresignificantforblack males,whileonlydoctorcrowdingissignificantforwhitemales. SocialcapitalisfoundtobenonKsignificantforwhiteandblackmales,onceagain suggestingthatsocialcapitalisnotassociatedwithmortalityduetochroniclower respiratorydisease.

39 33 Table&6.&Results&of&OLS&and&Spatial&Regression&Models&for&Mortality&due&to&Cerebrovascular&Disease&in&U.S.&Counties,&2005C2009& OrdinaryLeastSquares SpatialLag SpatialError ModifiedSpatial WhiteMale BlackMale WhiteMale BlackMale WhiteMale Black&Male& White&Male& Social&Capital& =2.473 * =1.270 =2.316 * =1.180 =2.545 * =1.212 =1.174 Composition& PercentBlack =0.419 *** *** =0.375 *** *** =0.401 *** *** =0.568 *** PercentWithoutInsurance =0.965 ** =0.921 ** =1.064 ** =0.428 PercentinPoverty =0.496 ** =0.445 ** =0.486 ** =0.274 UnemploymentRate *** ** *** *** PercentofPopulationover *** ** *** ** *** ** PopulationDensity Context& AccesstoDoctors =0.225 ** =0.224 ** =0.224 ** FacilityExpenses *** *** *** *** *** *** HospitalAccess =0.752 * *** = *** = ** =1.075 FacilityCrowding =0.014 =0.032 *** =0.012 =0.032 *** =0.013 =0.032 *** =0.012 DoctorCrowding =0.002 *** =0.001 * =0.002 *** =0.001 * =0.002 *** =0.001 * =0.002 * & & AkaikeInfoCriterion & && 15818& *p<.05**p<.01***p<.001 Error

40 34 Cerebrovascular,Disease, Thediagnosticsforspatialdependencewithcerebrovasculardiseaseasthe outcomeclearlyindicatedthespatialerrormodelasthebestchoiceforblackmales, butwaslessclearforwhitemales,withnodiscernabledifferencebetweenthe spatialerrorandthespatiallagmodelwhencomparingthelm.therefore,similarto blackmalecancermortality,iranamodifiedspatialerrormodelforwhitemales, withspatialprocessesexplicitlymodeledbylaggedindependentvariables,resulting inaclearchoiceofthespatialerrormodelwhencomparingthelmandakaike InformationCriterion,ameasureofmodelfit.IanalyzethemodelsboldedinTable 6. Here,allofthecontextualvariablesforblackmalesaresignificantwhileonly doctorcrowdingissignificantforwhitemales.thissuggeststhatblackmalesface additionalbarrierswhenfacedwithcerebrovasculardisease,suchashospital access,whichhasalargecoefficientandisassociatedwithhighermortalityforblack males.forthecompositionalvariables,percentblack,percentinpoverty,and percentofthepopulationover65areallsignificant,thoughonlypercentblackand unemploymentratearesignificantforwhitemales,withunemploymentratehaving ahighpositivecoefficientofover3.similartothepreviouscausesofdeath,social capitalisnonisignificantforeitherblackorwhitemales.

41 35 Table&7.&Results&of&OLS&and&Spatial&Regression&Models&for&Mortality&due&to&Accidents&in&U.S.&Counties,&2005B2009& OrdinaryLeastSquares SpatialLag SpatialError ModifiedSpatial WhiteMale BlackMale WhiteMale BlackMale WhiteMale Black&Male& White&Male& Social&Capital& >7.124 *** >1.915 * >5.464 *** >1.600 >5.398 *** >1.607 >5.664 * Composition& PercentBlack >0.880 * *** >0.721 *** *** >0.876 *** *** >0.974 *** PercentWithoutInsurance > > > >0.101 PercentinPoverty * >0.738 *** * > >0.496 * UnemploymentRate > * PercentofPopulationover ** * * PopulationDensity >0.001 Context& AccesstoDoctors >0.205 >0.248 ** >0.203 >0.252 ** >0.210 >0.258 ** FacilityExpenses > *** > *** *** >0.007 HospitalAccess > ** > ** > ** FacilityCrowding >0.056 *** >0.056 *** >0.056 *** DoctorCrowding >0.002 *** >0.001 >0.002 *** >0.002 *** >0.002 * & & AkaikeInfoCriterion & && 16753& *p<.05**p<.01***p<.001 Error

42 36 Accidents) Formortalityduetoaccidents,thediagnosticsforspatialdependence indicatedspatialerrorasthebestmodelforblackmales,butthelmwassignificant forboththespatiallagandspatialerrormodelsforblackmales.ithereforeincluded amodifiedspatialerrormodel,whichexplicitlymodelsthespatialprocessesofthe independentvariables.rebrunningthediagnostics,thismodifiedspatialerrormodel wasfoundtobethebestfitforwhitemales. LookingattheboldedmodelsinTable7,Ifindresultsoppositetothosefor mortalityduetocerebrovasculardisease.foraccidentmortality,whitemalesare notquiteasabletocapitalizeonthecontextualfactorsofthecounty,only benefittingfromthecompositionalfactors,whileblackmalesexperiencedifferences inmortalityduetobothcompositionandcontext.thecontextualfactorsarenotall beneficial,however,withfacilityexpensesandhospitalaccessincreasingmortality forblackmales.inthiscase,itappearsthatthecontextualfactorsworktoincrease blackmalemortalityoverall. Socialcapitalisfoundtobesignificantforwhitemales,reducingmortality ratesby5.6deaths.blackmales,ontheotherhand,receivenobenefitforsocial capital,suggestingthatsomemechanismexiststhatallowswhitemalestobenefit fromthesocialcapitalavailableinthecountywhileblackmencannot.

43 37 Table&8.&Results&of&OLS&and&Spatial&Regression&Models&for&Mortality&due&to&Alzheimer's&in&U.S.&Counties,&2005E2009& OrdinaryLeastSquares SpatialLag SpatialError WhiteMale BlackMale WhiteMale BlackMale White&Male& & Black&Male& Social&Capital& A A A0.036 Composition& PercentBlack A0.109 *** *** A0.089 *** *** A0.095 ** *** PercentWithoutInsurance PercentinPoverty A0.312 ** A0.041 ** A0.246 * A0.036 * A0.292 * A0.037 * UnemploymentRate A PercentofPopulationover *** A *** *** PopulationDensity A0.001 *** A0.001 *** A0.001 ** Context& AccesstoDoctors A0.128 * A0.007 A0.136 ** A0.007 A0.137 ** A0.007 FacilityExpenses *** *** *** *** *** *** HospitalAccess FacilityCrowding A0.035 *** A0.005 *** A0.033 *** A0.005 *** A0.034 *** A0.004 *** DoctorCrowding A0.001 *** A0.001 *** A0.001 *** & & & & & AkaikeInfoCriterion & && && 16394& && *p<.05**p<.01***p<.001

44 38 Table&9.&Results&of&OLS&and&Spatial&Regression&Models&for&Mortality&due&to&Diabetes&in&U.S.&Counties,&2005A2009& OrdinaryLeastSquares SpatialLag SpatialError WhiteMale BlackMale White&Male& & Black&Male& WhiteMale BlackMale Social&Capital& ;0.866 ;0.353 ;0.704 ;0.336 ;0.824 ;0.370 Composition& PercentBlack ;0.263 *** *** ;0.180 *** *** ;0.210 *** *** PercentWithoutInsurance ; * ; ; PercentinPoverty ;0.004 ;0.401 *** ;0.010 ;0.338 ** ;0.131 ;0.389 *** UnemploymentRate ** ** PercentofPopulationover *** *** *** PopulationDensity ** ** ** Context& AccesstoDoctors ;0.146 * ;0.083 ;0.123 ;0.082 ;0.119 ;0.082 FacilityExpenses *** *** *** *** *** *** HospitalAccess ;0.069 ;0.007 ;0.091 ;0.006 ;0.083 ;0.007 FacilityCrowding ;0.043 *** ;0.033 *** ;0.039 *** ;0.032 *** ;0.040 *** ;0.032 *** DoctorCrowding ;0.001 *** ;0.001 *** ;0.001 *** & & & & & AkaikeInfoCriterion & && && 36383& && *p<.05**p<.01***p<.001

45 39 Alzheimer s+and+diabetes+ MortalityduetoAlzheimer sanddiabetessharesthesamesubstantive patterns,asidefromthemodelchoice.forbothwhiteandblackmales,mortality duetoalzheimer sisbestmodeledbyspatialerror,andmortalityduetodiabetesis bestmodeledbyspatiallag.asidefromthisdifference,theyarequitesimilar,as seenintables8and9.forbothcausesofmortality,compositionalandcontextual factorsareassociatedwiththemortalityrate,andsocialcapitalisnotsignificantfor eitherblackmalesorwhitemales.althoughtheoverallpatternsaresimilar,there areimportantdifferenceswithinthecompositionalandcontextualfactors.for whitemales,thecompositionalvariableshavetheeffectofincreasedmortality, whiletheyreducemortalityforblackmaleswhenconsideringmortalitydueto Alzheimer s.thesameistrueformortalityduetodiabetes,wherethecompositional variablesworktoincreasewhitemalemortalityandtodecreaseblackmale mortality.

46 40 Table&10.&Results&of&OLS&and&Spatial&Regression&Models&for&Mortality&due&to&Nephritis/Nephrosis&in&U.S.&Counties,&2005C2009& OrdinaryLeastSquares SpatialLag SpatialError ModifiedSpatial WhiteMale BlackMale WhiteMale BlackMale WhiteMale Black&Male& White&Male& Social&Capital& >1.477 *** >0.299 >1.242 ** >0.280 >1.304 ** >0.286 >0.304 Composition& PercentBlack >0.052 * *** > *** > *** >0.132 *** PercentWithoutInsurance >0.290 ** >0.222 * > >0.191 PercentinPoverty >0.142 >0.111 >0.092 >0.085 >0.188 * >0.100 >0.128 UnemploymentRate > > > PercentofPopulationover *** > *** *** > ** PopulationDensity * * * Context& AccesstoDoctors >0.102 ** >0.043 >0.119 ** >0.043 >0.118 ** > * FacilityExpenses *** *** *** *** *** *** ** HospitalAccess >0.989 FacilityCrowding >0.030 *** >0.021 *** >0.027 *** >0.021 *** >0.028 *** >0.021 *** >0.014 DoctorCrowding >0.001 *** *** *** >0.001 ** & & AkaikeInfoCriterion & && 12747& *p<.05**p<.01***p<.001 Error

47 41 Nephritis/Nephrosis+ ForblackmalemortalityduetoNephritis/Nephrosis,spatialerrorwas indicatedbytheonlysignificantlmstatisticaswellastheory.forwhitemales,the LMforbothspatialerrorandspatiallagweresignificant,soIagainincludea modifiedspatialerrormodel,whichresultsinthebestmodelfit.iinterpretthe boldedmodelsintable10.allofthecompositionvariables,withtheexceptionof populationdensity,aresignificantforblackmales,andareassociatedwith increasedmortality.forwhitemales,onlypercentblackandpercentofpopulation over65aresignificant,andworkinoppositedirections,withtheformerassociated withreductionsinmortalityandthelatterassociatedwithincreasedmortality.for bothblackandwhitemales,facilityexpensesandcrowdingofeitherfacilities(for blackmales)ordoctors(forwhitemales)aresignificant,withfacilityexpenses workingtoincreasemortalityforblackandwhitemales.socialcapitalissignificant forneitherblacknorwhitemales.

48 42 Table&11.&Results&of&OLS&and&Spatial&Regression&Models&for&Mortality&due&to&Influenza/Pneumonia&in&U.S.&Counties,&2005E2009& OrdinaryLeastSquares SpatialLag SpatialError WhiteMale Black&Male& White&Male& BlackMale WhiteMale BlackMale Social&Capital& =1.159 * =0.059 =1.109 * =0.066 =1.183 * =0.060 Composition& PercentBlack =0.183 *** *** =0.158 *** *** =0.172 *** *** PercentWithoutInsurance = = = PercentinPoverty =0.096 =0.038 =0.062 =0.046 =0.111 =0.041 UnemploymentRate * PercentofPopulationover *** *** *** PopulationDensity * *** *** *** Context& AccesstoDoctors =0.143 ** =0.025 =0.149 ** =0.026 =0.149 ** =0.025 FacilityExpenses *** *** *** *** *** *** HospitalAccess = = =0.011 FacilityCrowding =0.028 *** =0.011 *** =0.027 *** =0.012 *** =0.027 *** =0.012 *** DoctorCrowding =0.001 *** =0.001 *** =0.001 *** & & & & AkaikeInfoCriterion & && 25029& && *p<.05**p<.01***p<.001

49 43 Influenza/Pneumonia. ForInfluenza/Pneumonia,thediagnosticspointtoaspatiallagmodel,asthe LMvalueismuchhigherthanthatforthespatialerrorspecification.Thisholdstrue forblackmalesaswell,butsimilartochroniclowerrespiratorydiseasediagnostics, neitherlagnorerroraresignificant,suggestingthatanolsmodelisthebestfit.i interprettheboldedmodelsintable11. Forwhitemales,onlythepercentblackandpercentageofthepopulation over65issignificant,suggestingthatthecompositionisonlyafactorin influenza/pneumoniamortalityduetoracial/ethnicmakeupandageofthe population.thesamecontextualvariablesforblackmalesandwhitemalesare significant,withtheadditionofdoctorcrowdingandaccesstodoctorsforwhite males,andallcoefficientsareinthesamedirection.theonlyexceptionisfor percentblack,whichisassociatedwithanincreaseinmortalityforblackmales,but adecreaseinmortalityratesforwhitemales.socialcapitalissignificantforwhite males,andisassociatedwithreductionsinmortality,whileitisnonjsignificantfor blacks.

50 44 Table&12.&Results&of&OLS&and&Spatial&Regression&Models&for&Mortality&due&to&Intentional&Self<Harm&in&U.S.&Counties,&2005<2009& OrdinaryLeastSquares SpatialLag SpatialError ModifiedSpatial WhiteMale BlackMale WhiteMale BlackMale WhiteMale Black&Male& White&Male& Social&Capital& =1.427 ** =0.011 =1.346 ** =0.011 =1.433 ** =0.011 =0.641 Composition& PercentBlack =0.184 *** *** =0.147 *** *** =0.164 *** *** =0.271 *** PercentWithoutInsurance * *** PercentinPoverty =0.428 *** =0.045 ** =0.319 ** =0.047 ** =0.412 *** =0.044 * =0.794 *** UnemploymentRate *** PercentofPopulationover = = = PopulationDensity ** *** ** *** * *** Context& AccesstoDoctors =0.123 ** =0.012 =0.135 ** =0.012 =0.131 ** = FacilityExpenses *** *** *** *** *** *** HospitalAccess FacilityCrowding =0.029 *** =0.005 *** =0.026 *** =0.005 *** =0.027 *** =0.005 *** =0.023 ** DoctorCrowding =0.001 *** =0.001 *** =0.001 *** =0.001 *** & & AkaikeInfoCriterion & && 13410& *p<.05**p<.01***p<.001 Error

51 45 Intentional)Self,Harm) ThefinalcauseofdeathIanalyzeisintentionalself5harm,commonlyknown assuicide.similartonephritis/nephrosis,accidents,andcerebrovasculardisease, blackmalemortalityisbestmodeledbyspatialerror,whilethediagnosticsdidnot differentiatebetweenspatiallagandspatialerrorforwhitemalemortalityuntila modifiedspatialerrormodelwasspecified.themodelsinterpretedareboldedin Table12.Ofthecontextualvariables,facilityexpensesandcrowdingaresignificant forblackmales,andfacilityanddoctorcrowdingaresignificantforwhitemales,all associatedwithdecreasesinmortalitywiththeexceptionoffacilityexpensesfor blackmen.thisisinterestinginthecaseoffacilityanddoctorcrowding.itcouldbe thatwhitemalesaresomehowbetterabletoaccesscrowdeddoctorsandfacilities thanothergroups,andthusexperiencegreaterreductionsinmortalitywhenfaced withcrowding. Withtheexceptionofpercentpopulationover65andpopulationdensity,all compositionalvariablesaresignificantforwhitemales,andallbutunemployment rate,percentwithoutinsurance,andpopulationover65aresignificantforblack males.socialcapital,similartonearlyallothercauses,isnon5significantforboth whiteandblackmales.

52 46 CHAPTER3.CONCLUSION Conclusion) Examiningthespatialpatterningoftheleadingtencausesofmortality,I focusedonfivekeypoints:(1)thelinkbetweenenvironmentandhealth,(2),the roleofsocialcapital(3)thedifferencebetweencompositionalandcontextual effects,(4)racialdifferencesinmortality,and(5)thespatialpatterningofmortality. Toaddressthefirstandfifthpoints,Imappedtheclusteringofmortality ratesandcomparedtwodifferentspatialmodels.asihypothesized,cancer,chronic lowerrespiratorydisease,andinfluenza/pneumoniawereallmodeledbestbya spatiallagmodel,indicatingadiffusiveprocessinthecausesofmortality. Surprisingly,diabetesalsoemergedasacauseofdeaththatismodeledbestby spatiallagforbothblackandwhitemen.onereasonforthismightbeastrong associationbetweendiabetesandahealthindicatorsuchaspsychosocialstress,one factorcommonlysuggestedtounderliethespatialdistributionofmortality.geeand Payne5Sturges(2004)suggestthatpsychosocialstressisafactorthatcanresult fromthelocalenvironmentandleaddirectlytoillness,suchasdiabetes.stressorsin thelocalenvironment,suchassegregation,canincreasevulnerabilityto psychosocialstress.alsoconformingtomyhypothesis,heartdisease, cerebrovasculardisease,accidents,alzheimer s,nephritis/nephrosisandintentional self5harmarebestmodeledbyspatialerror,indicatingspatialautocorrelation amonguncontrolled5forvariablesratherthanadiffusiveprocessinthemortality

53 47 outcomeitself.asaresultoffindingthataspatialmodelworksbestformanyofthe causesofmortality,coupledwiththelisamapsthatindicatehighandlowspatial clustering,iconcludethatthereisindeedanassociationbetweentheenvironment andhealth,aswellasastrongspatialclusteringofmortality. Theresultsfromtheseanalysessuggesttwoimportantoverallconclusions, whichrelatetopointstwoandthree,whilebothconclusionsrelatetopointfour. Thefirst,relatingtopointtwo,isthatthereissomethinguniqueaboutsocialcapital thatbenefitswhitemalemortality,resultinginreductionsfor3ofthetop10causes ofmortalityaswellasall5causemortality,whileblackmalesdonotseesuch reductions.thissuggeststhatsocialcapitalisanimportantresourcetoconsider whenworkingtoreducedifferencesinblack5whitemalemortality.contrarytothe suggestionofkaplanetal(1988),andrelatingtopoint4,policiestoincreasesocial supportmightincreasethegapbetweenblackandwhitemales.iftheretrulyis somethinguniqueaboutsocialcapitalthatbenefitswhitemalesbutnotblackmales, increasesinsocialcapitalwouldonlybenefitwhitemales.instead,themechanisms thatallowwhitemenandnotblackmentobenefitfromsocialcapitalneedtobe examinedandaddressedtoallowblackmentobetterbenefitfromsocialcapitalasa resource.likeotherresources,socialcapitalisspatiallypatterned,andthisrelation tospaceneedstobetakenintocarefulconsiderationaswell.ingeneral,placeis importantforresearchonhealthbecauseitbothconstitutesandcontainsphysical resourcesandsocialrelations(cumminsetal2007).whileinthisanalysisitwas, socialcapitalisnotalwaysassociatedwithpositivehealthoutcomes,ascarpiano

54 48 (2007)findsthatforoutcomesotherthanthetoptencausesofmortality,social capitalisassociatedwithsomenegativehealthoutcomes. Thesecondconclusion,relatedtopoint3,isthatbothcompositionaland contextualfactorsdomatterforblackandwhitemalemortality,suggestingthatthe debatebetweencompositionvs.contextshouldmovetowardsconsideringthatboth areassociatedwithmortalityoutcomes,andacombinationofthetwomightbea moreaccurateportrayaloftheactualprocess.afewlimitationsofthisfindingare theuseofcountiesandthetimeperiodexamined.overafive5yearperiod,much migrationtakesplace,whichaltersthecompositionofthearea.additionally,as manyofthecausesofdeatharedegenerativeandtakeplaceoveralongperiodof time,exposuresoveralifetimecanaffectmortalityoutcomesandmightnotbe capturedbythelimitedtimeperiodexamined.bothofthesefactorsmustbe consideredwheninterpretingthecompositionvs.contextargumentaswellasthe results. Indeed,oneofthepreviouscritiquesofthecontextvs.compositionargument pointsoutthedifficultyintrulyparsingoutwhatresultsfromcontextandwhat reallyresultsfromcomposition(mcintyreetal2002).inotherwords,itcanbe difficulttodisentanglewhethertheeffectisentirelyduetocompositionorentirely duetocontext.whilethetheoreticaldistinctionbetweencompositionandcontext maybequitedistinct,intherealworldthesamecannotalwaysbesaid. Characteristicsofindividualscanbeshapedbytheecologicalattributesofwhere theylive,andtheecologicalattributesofaplacecanbealteredbytheindividuals wholivethere.take,forexample,thepresenceofabikingtrail.asanecological

55 49 attribute,thebikingtrailmightinfluenceindividualstoexercisemore,butalongthe samelinesaphysicallyactivecommunitymightactivelyseektobuildsuchatrailif itdidnotalreadyexist.inthiscasesphysicalactivitywouldbepositivelyrelatedto health,butitwouldbehardtodiscernwhethertheattributesofthepeopleaffect theplace,andthenhealth(bybuildingabiketrail),orwhetherplacedirectly influenceshealththroughthepresenceofabiketrail.whilethisisanoverly simplifiedexample,itillustratesthatcompositionandcontextcaneachaffectthe otherincomplicatedways.analytically,thismeansthatusingindividualcontrolsin anattempttoparseoutwhateffectisduetocontextdoesnotalwayswork,asthe individualcontrolscouldbeinterveningvariablesbetweenplaceandhealthrather thanconfoundingvariablesthatshouldbecontrolledfor.contextvs.composition mightinfactbea falsedualism, withareciprocalrelationshipoccurringbetween thetwo(cumminsetal.2007). Finally,directlyrelatingtopointfour,Ifindthatthereareimportant variationsinblackandwhitemalemortality.asnotedpreviously,socialcapitalisa resourcethatbenefitswhitemalesbutdoesnothavethesameeffectforblack males.whilepeoplewholiveinthesameareascanbeassumedtobeexposedtothe sametypesofresources,thisdoesnotnecessarilymeanthattheyhaveequalaccess totheresourcestowhichtheyareexposed,exemplifiedbythisfinding.this suggeststhatthepresenceofsocialcapitalisnotnecessarilythemostaccurate representationoftheconcept,andfutureresearchshouldattempttodiscoverwhat mechanismsallowonegrouptobenefitfromsocialcapitalwhileanothergroupdoes not,assumingequalexposuretotheavailabilityofsocialcapital.

56 50 APPENDIX(A( ADDITIONAL(TABLES( ( ( ( Table13.ICD510CodesforEachCauseofDeath Table14.LagrangeMultiplierStatisticsforSpatialAutocorrelation Table15.DescriptiveStatisticsforMortality

57 51 Table&13.&ICD-10&Codes&for&Each&Cause&of&Death& Cause&of&Death& ICD&Code(s)& HeartDisease I05-I09,I10-I15,I20-I25,I30-I51 Cancer C00-C97,D10-D36,D37-D48 ChronicLowerRespiratoryDisease J40-J47 CerebrovascularDisease I60-I69 TrafficAccidents V01-V99 Alzheimer's G30-G31 Diabetes E10-E14 Nephritis/Nephrosis N00-N07,N17-N19 Influenza/Pneumonia J09-J18 IntentionalSelf-Harm X60-X86

58 52

59 53 Table&15.&Descriptive&Statistics&for&Mortality&& MeanMortalityRate AverageAgeatDeath Whites Blacks Whites Blacks AllCauses HeartDisease Cancer ChronicLowerRespiratoryDisease CerebrovascularDisease Accidents Alzheimer's Diabetes Nephritis/Nephrosis Influenza/Pneumonia IntentionalSelf-Harm

60 54 Appendix(B(( ( Figures(1:14( Figure1:IndependentVariablesDistribution Figure2:IndependentVariablesDistribution2 Figure3:FacilityCrowdingDistribution Figure4:NeighborsHistogram Figure5:CancerDistributionandResiduals Figure6:ChronicLowerRespiratoryDiseaseDistributionandResiduals Figure7:HeartDiseaseDistributionandResiduals Figure8:CerebrovascularDiseaseDistributionandResiduals Figure9:AccidentDistributionandResiduals Figure10:Alzheimer sdistributionandresiduals Figure11:DiabetesDistributionandResiduals Figure12:Nephritis/NephrosisDistributionandResiduals Figure13:Influenza/PneumoniaDistributionandResiduals Figure14:SuicideDistributionandResiduals

61 Figure'1.'Independent'Variables'Distribution' SocialCapital PercentBlack PercentWithoutInsurance MedianHouseholdIncome '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''' ' ' 55

62 56 Figure'2.'Independent'Variables'Distribution'2' ' PercentinPoverty UnemploymentRate PopulationDensity FacilityExpenses ' '

63 Figure'3.'Facility'Crowding'Distribution' ' ' ' Figure'4.'Neighbors'Histogram' FacilityCrowding NeighborsHistogram ' 57 ' ' '

64 Figure'5:'Cancer'Distribution'and'Residuals' ' WhiteMaleCancerMortality BlackMaleCancerMortality ' ' 58