ArcGIS10回归分析教程

Analyzing911responsedatausingRegressionThistutorialdemonstrateshowregressionanalysishasbeenimplementedinArcGIS,andexploressomeofthespecialconsiderationsyou’llwanttothinkaboutwheneveryouuseregressionwithspatialdata.Regressionanalysisallowsyoutomodel,examine,andexplorespatialrelationships,tobetterunderstandthefactorsbehindobservedspatialpatterns,andtopredictoutcomesbasedonthatunderstanding.OrdinaryLeastSquaresregression(OLS)isaglobalregressionmethod.GeographicallyWeightedRegression(GWR)isalocal,spatial,regressionmethodthatallowstherelationshipsyouaremodelingtovaryacrossthestudyarea.BothofthesearelocatedintheSpatialStatisticsTools-ModelingSpatialRelationshipstoolset:Beforeexecutingthetoolsandexaminingtheresults,let’sreviewsometerminology:•Dependentvariable(Y):whatyouaretryingtomodelorpredict(residentialburglaryincidents,forexample).•Explanatoryvariables(X):variablesyoubelieveinfluenceorhelpexplainthedependentvariable(like:income,thenumberofvandalismincidents,orhouseholds).•Coefficients(β):values,computedbytheregressiontool,reflectingtherelationshipandstrengthofeachexplanatoryvariabletothedependentvariable.•Residuals(ε):theportionofthedependentvariablethatisn’texplainedbythemodel;themodelunderandoverpredictions.Thesign(+/-)associatedwiththecoefficient(oneforeachexplanatoryvariable)tellsyouwhethertherelationshipispositiveornegative.IfyouweremodelingresidentialburglaryandobtainanegativecoefficientfortheIncomevariable,forexample,itwouldmeanthatasmedianincomesinaneighborhoodgoup,thenumberofresidentialburglariesgoesdown.Outputfromregressionanalysiscanbealittleoverwhelmingatfirst.Itincludesdiagnosticsandmodelperformanceindicators.Allofthesenumbersshouldseemmuchlessdauntingonceyoucompletethetutorialbelow.Importantnotes:1.ThestepsinthistutorialdocumentassumethedataisstoredatC:\SpatialStats.Ifadifferentlocationisused,substituteC:\SpatialStatswiththealternatelocationwhenenteringdataandenvironmentpaths.2.ThistutorialwasdevelopedusingArcGIS10.0.Ifyouareusingadifferentversionofthesoftware,thescreenshotsandhowyouaccessresults,maybeabitdifferent.TutorialEstimatedtime:1.5hoursIntroduction:Inordertodemonstratehowtheregressiontoolswork,youwillbedoingananalysisof911EmergencycalldataforaportionofthePortlandOregonmetropolitanarea.Supposewehaveacommunitythatisspendingalargeportionofitspublicresourcesrespondingto911emergencycalls.Projectionsaretellingthemthattheircommunity’spopulationisgoingtodoubleinsizeoverthenext10years.Iftheycanbetterunderstandsomeofthefactorscontributingtohighcallvolumesnow,perhapstheycanimplementstrategiestohelpreduce911callsinthefuture.Step1GettingStartedOpenC:\SpatialStats\RegressionExercise\RegresssionAnalysis911Calls.mxd(thepathmaybedifferentonyourmachine)InthismapdocumentyouwillnoticeseveralDataframescontaininglayersofdataforthePortlandOregonmetropolitanstudyarea.EnsurethattheHotSpotAnalysisdataframeisactiveInthemap,eachpointrepresentsasinglecallintoa911emergencycallcenter.Thisisrealdatarepresentingover2000calls.Step2ExamineHotspotAnalysisresultsExpandthedataframeandclickthe+signtotherightoftheHotSpotAnalysisgroupedlayerEnsurethattheResponseStationslayerischeckedonResultsfromrunningtheHotspotAnalysistoolshowuswherethecommunityisgettinglotsof911calls.Wecanusetheseresultstoassesswhetherornotthestations(fire/police/emergencymedical)areoptimallylocated.Areaswithhighcallvolumesareshowninred(hotspots);areasgettingveryfewcallsareshowninblue(coldspots).Thegreencrossesaretheexistinglocationsforthepoliceandfireunitstaskedwithrespondingtothese911calls.Noticethatthe2stationstotherightofthemapappeartobelocatedrightover,orverynear,callhotspots.Thestationinthelowerleft,however,isactuallylocatedoveracoldspot;wemaywanttoinvestigatefurtherifthisstationisinthebestplacepossible.Thecommunitycanusehotspotanalysistodecideifaddingnewstationsorrelocatingexistingstationsmightimprove911callresponse.Step3ExploringOLSRegressionThenextquestionourcommunityisprobablyaskingis,“Whyarecallvolumessohighinthosehotspotareas?”and“Whatarethefactorsthatcontributetohighvolumesof911calls?”Tohelpanswerthesequestions,we’llusetheregressiontoolsinArcGIS.ActivatetheRegressionAnalysisdataframebyrightclickingandchoosingActivateExpandtheSpatialStatisticstoolstoolboxRightclickinaopenspaceinArcToolboxandsetyourenvironmentasfollows:Disablebackgroundprocesses(GeoprocessingGeoprocessingOptions).WithArcGIS10,geoprocessingtoolscanruninthebackgroundandallresultsareavailablethroughtheResultswindow.Bydisablingbackgroundprocessing,wewillseetoolresultsinaprogresswindow;thisisoftenbestwhenyouareusingtheRegressiontools:Inthedataframe,checkofftheData911CallslayerInsteadoflookingatindividual911callsaspoints,wehaveaggregatedthecallstocensustractsandnowhaveacountvariable(Calls)representingthenumberofcallsineachtract.RightclicktheObsData911CallslayerandchooseOpenAttributeTableThereasonweareusingcensustractleveldataisbecausethisgivesusaccesstoarichsetofvariablesthatmighthelpexplain911callvolumes.NoticethatthetablehasfieldssuchasEducationalstatus(LowEd),Unemploym