9212011pp2532602011bytheEcologicalSocietyofAmericaQualityofbasicdataan - PDF document

1 Notes ,92(1),2011,pp.253…2602011bytheEco
Notes ,92(1),2011,pp.253…2602011bytheEcologicalSocietyofAmericaQualityofbasicdataandmethodtoidentifyshapeaffectrichness–altituderelationshipsinmeta-analysis tendencyforH-SHtobemostfrequentlyfoundintheNearcticandPalearcticregions,whichcomplicatedinterpretation.Nonetheless,whereasH-SHrepresentedoutofthetotalPalearcticNearctic(takentogether)SRArelationshipsinthewholedataset,itsrepresentationroseto77inthestringentsubset.Also,theproportionalrepresentationofNearticregionsincreasedbyafactorof1.25inthestringentsubset,whereastheproportionalrepresentationofH-SHpatternsincreasedbyafactorof3.Thus,inspiteofthebiogeographicassociation,thesubstantialchangeintheproportionalrepresentationofdifferentshapesshowninthestringentsubsetcanbehardlyconsideredasmerelytheconsequenceofthepresenceofNearcticandPalearcticSRArelationships.OneeffectofrelaxingtheselectioncriteriaofinclusionofSRArelationships,fromstringenttointermediateandlax,wastoincreasetheproportionalrepresentationofmonotonicdecreasingpatterns.Thissuggestedthatreducingthesamplingeffortinindivid-ualstudies,i.e.,eitherbyloweringthenumberofsamplingpoints,orbyreducingtheproportionofthegradientsampled,mayfavortheperceptionofmono-tonicdecreasingpatterns,andthiscouldaltertherelativedistributionofdifferentSRApatterns.Varia-tioninsamplingefforthasbeenpreviouslyfoundtoaffectothermacroecologicalrelationships(e.g.,bodysize…abundancerelationships;Grif“ths1998).Exam-inedwithinthecontextofSRArelationships,thereductionofthenumberofsamplingpoints,ortheproportionofgradientsampled,maybeassociatedwitheffectsofspatialscaleknowntoaffecttheanalysisofspeciesdiversitypatterns(LyonsandWillig1999,2002,RahbekandGraves2000,Whittakeretal.2001),andspeci“cally,ofaltitudinalrichnessgradients(e.g.,Rahbek2005,JankowskiandWeyhenmeyer2006,Nogues-Bravoetal.2008,Sandersetal.2009).Weshowedthehigherfrequencyofoccurrenceofhump-shapedpatternsinthestringentsubsetofdata,whichencompassedstudiesthatinvolvedgreatersamplingeffort,andspannedoverlongeraltitudinalextentsthantheintermediateandlaxsubsets.Ontheotherhand,allhump-shapedpatternsinthestringentsubsetcameupfromstudiesperformedatlandscapescalewithinthePalearcticandNearcticregions.Giventhelownumberofobservationsinourstringentdataset,alltheseeffectsdeservefurtherattentioninfuturestudies.Anotherconsequenceofourdatamanipulationwasincreasinguncertaintyintheidenti“cationofpatternusingtheintermediateandlaxsubsetsofdata,andalso,aftercontrollingfordifferencesinsamplesize,uncer-taintyoverestimationofthemagnitudeofSRArelationshipsincreasedwiththerelaxationofcriteriaofinclusionofdata.Theoccurrenceofnopattern(NP)wasgreaterintheintermediateandlaxsubsetsofdataratherthaninthestringentsubset,especiallyforthestatisticalmethod.Forthelaxandintermediatesubsets,thestatisticalmethodwasmoreconservativeinthedetectionofpattern,andsuggestedahigherproportionofNPratherthanthevisualmethod.ThevisualmethodandtheconsensushelpedinsortingoutNPdatasetsintorecognizableshapes;however,aconsiderableproportionofdatasetssuggestingnopatternorcontradictoryinformationremained.Thestringentsubsethadlessaverageinter-siteresolution,whichimpliesthatagreatereffectofspatialautocor-relationinthisdatasetmightcontributetodecreaseuncertaintyaboutdetectionofshape.Thereareanumberofattributesinthedesignofstudiesthatcomposedtheintermediateandlaxsubsetsthatmaycomplicatethedetectionofp

2 attern,includinglessand/orunstandardized
attern,includinglessand/orunstandardizedsamplingeffort,andincreasedinter-siteresolutionthatpromotesanincreaseinthescatteringofdata.However,aproportion()ofdatasetsinthestringentsubsetalsoshowednopattern.ThissuggeststhatthepresenceofidiosyncraticvariationintheSRArelationshipsmightberathercommoninarthropods.OurstudyshowedthattheuseofdifferentmethodstoidentifySRAshapesindeedinteractswiththequalityofbasicdatatoin”uencetherelativedistributionofpatterns.OnlythestringentsubsetofdatarevealedaconsistentorderintherelativeimportanceofdifferentSRAshapesthatwasrobusttovariationinmethodofanalysis.ToofewSRArelationshipsremainedinthestringentsubsetastoinfergeneralconclusionsontheextremelyhighlydiversearthropods.However,theconsistencyintheoutcomeinthestringentsubsetmakesreasonabletoinfertheexistenceofmultipleSRAforms,withthepredominanceofhump-shapedpatternsovermonotonicdecreasingones,alongwiththepresenceofconsiderableidiosyncraticvariation.Hadwebasedourinterpretationonthewholedatasetthiswouldhaveledtoadifferentconclusionaboutthepredominantpatternsinarthropods.WeconcludethatdecisionsfollowedtogatherthedataandthemethodweusetoinferSRArelationships(visuallyorstatistically)bothmayin”uenceourperceptionoftherelativefrequencyofpredominantshapes,andnotnecessarilythelargestdatasetisthebestformeta-analysis.J.Gurevitch,K,MergensenandJ.M.Moralesgaveususefulstatisticaladvice.A.F.SilberingandR.J.Whittakercontributedusefulbibliographies.D.J.Currieprovidedinsightfulcomments.V.WerenkrautandA.RuggieroaresupportedbyCONICET(PIP5113)andUniversidadNacionaldelComahue.ITERATUREBrown,J.H.,andM.V.Lomolino.1998.Biogeography.Secondedition.Sinauer,Sunderland,Massachusetts,USA.Burnham,K.P.,andD.R.Anderson.2002.Modelselectionandmultimodelinference.Secondedition.Springer,NewYork,NewYork,USA. January2011 R ESULTS Thetotalof109SRArelationshipstakentogether showeddifferencesintheproportionalrepresentationof differentformsbetweenthetwomethodsofanalysis, andconsensus(Fig.1a…c).Thestatisticalmethod suggestedlackofpatternin ; 40 % ofthedatasets (NP;Fig.1a).DECandL-PLhadasimilar( ; 20 % ) proportionalrepresentation,followedbyH-SH( ; 13) (Fig.1a).Thevisualmethodhalf-decreasedthepropor- tionalrepresentationofNPandL-PL,andincreased considerablytheproportionalrepresentationofH-SH andDEC(Fig.1b).TheconsensussuggestedDECwas themostabundantSRArelationship,followedbyH- SH,andL-PLbecamerarer; ; 20 % outofthetotaldata setsremainedasNP,and ; 25 % showednoconsensus (CONT;Fig.1c). F IG .1.RelativefrequencydistributionsofSRA(speciesrichness…altitude)patternsobservedinsubsetsoftheworldwide arthropoddata:(a…c)wholedataset;(d…f)stringentsubset;(g…i)intermediatesubset;and(j…l)laxsubset.Thepatternswere analyzedbystatisticalandvisualmethodsandbyconsensus.Abbreviationsare:H-SH,hump-shaped;DEC,monotonicdecreasing; L-PL,low-plateau;INC,monotonicincreasing;U-SH,U-shaped;NP,nopattern;OTH,otherpatterns;andCONT,contradictory. Numbersabovebarsaresamplesizes.Studieswithalownumberofsamplingpoints( N ¼ 5)werenotanalyzedbythestatistical method. January2011 257 NOTES patternthathasarichnesspeakatintermediatealtitudes with25 % ormorespeciesthanatthebaseortopofthe mountain(i.e.,thesocalledmid-elevationpeakby McCain2009).AL-PLpatternhad . 300mof consecutivelyhighrichnessatthemountainbaseand thereafterdecreasingspeciesrichness(seeMcCain2009 forfurtherdeta

3 ilsandotherpossibleforms).Datasets thats
ilsandotherpossibleforms).Datasets thatsuggestednoclearSRArelationshipwereassigned toNP,andotherdifferentformswereincludedinOTH. Welookedforaconsensusbetweenthevisualand statisticalmethodstoidentifythe“nalshapeofeach dataset;datasetsshowinglackofconsensuswere labeledascontradictory(CONT). Tofurtherevaluateuncertaintyoveridenti“cationof shape,wesquareroot-transformedthecoef“cientsof determination( R 2 )fromtheordinaryleastsquares (OLS)regressionstoobtaincorrelationcoef“cients( r ) (e.g.Hillebrand2004),whichweretransformedtoan effectsize( z r ,Fishers z -transformation;Hedgesand Olkin1985).Weestimatedacommonmeasureofeffect sizeforthestringent,intermediate,andlaxsubsetsof data,takingintoaccountthatthenonsystematic varianceofestimatesofeffectsizewasinversely proportionaltothesamplesizeofthegradientson whichestimateswerebased(HedgesandOlkin1985). Combinationoflinearandquadratictermsinmeta- analysisrequiresthatallthestudy-speci“cregressions havebeen“ttedwiththesamenumberofterms(K. MergensenandJ.Gurevitch, personalcommunication ). Hence,weconductedthisanalysisseparatelyforeach SRApattern(H-SH,DEC,andL-PL).Weperformed alltheanalysesusingRsoftware(RDevelopmentCore Team2009). Closeexaminationofpatternsinourwholedataset, afterconsensus,showedthat,althoughtherelative frequencyofthemostabundantpatterns(H-SH, DEC,L-PL)wasindependentoftaxonomy(permuta- tion-basedFisher-FreemanHaltontestforsmall-sample categoricaldata[FI] ¼ 20.16,df ¼ 18, P ¼ 0.19),there wasanassociationwithclimate(FI ¼ 30.87,df ¼ 24, P ¼ 0.02),andbiogeography(FI ¼ 23.03,df ¼ 12, P ¼ 0.003) (seealsoAppendixB:Fig.B1).Wetestedthese associationsinthethreesubsetsofdatatoelucidate theextenttowhichtaxonomy,climate,andbiogeogra- phymightin”uencechangesintherelativeproportions ofSRApatternsafterourdatamanipulation.These analyseswereconductedusingStatXact-6(2003). P LATE 1.PanoramicviewofatemperatemountainregioninnorthernPatagonia,showinganexampleofthekindof environmentalchangesthatoccurwithaltitude.Photocredit:V.Werenkraut. NOTES 256 Ecology,Vol.92,No.1 toin”uencetheSRApatterns(Rahbek2005).(3)Numberofsamplingpoints.Whittaker(2010)adopteda10-datapointminimumassuitabletodiscriminatebetweenlinearandunimodalforminthespeciesrichness…productivityrelationship.(4)WerecordedthepresenceofanthropogenicdisturbanceasapotentialconfoundingvariableoftheSRArelation-ship.Whittaker(2010)arguedthatthestudydesignshouldnotinvolvepotentialconfoundingvariablesofthetestedecologicalrelationship.Weassignedeachgradienttoadifferentsubsetofdataaccordingtothefollowingcriteria:(1)stringent(ofthegradientsampled,standardizedorequalsamplingeffortacrossdifferentaltitudes,and10samplingpoints);(2)intermediate(standardizedorequalsam-plingeffortalongwithoneoftwootherpossibleconditions[eithertheproportionofgradientsampledandthenumberofsamplingpointswasorthenumberofsamplingpointswas10,buttheproportionofgradientsampledwas]);and(3)lax(studiesthatinvolvedunstandardizedorunequalsamplingeffortacrossdifferentaltitudesand/orshowedevidenceofanthropogenicdisturbanceasapotentialconfoundingvariable;ifstandardizedsamplingeffortwasapplied,thentheyhad10samplingpointsandofgradientsampled).Mostofthestudiesincludedinouranalysiswere“eldstudiesthatsampledarthropodsatverylocalgrainsizesusingdifferentsamplingmethods;therewereonlyafewstudiesthatusedcollectiondatafrommuseumsordist

4 ributionalinformationfrommaps(seeAppendi
ributionalinformationfrommaps(seeAppendixA:TableA1).Afterclassi“cationofstudiesintothethreesubsets,wecon“rmedthatthestringentsubsetofdataencompassedgradientsof760mofmeanaltitudeextent(meanSD)and45mofmeaninter-siteresolution(i.e.,meandistancebetweensam-plingpoints),withagreaterproportionofstudiesatthelandscapescale(i.e.,lineardistancebetweenthetwomostextremepoints30km)ratherthanatlocalscales(i.e.,distancebetweenthetwomostextremepointskm):62vs.38,respectively.Theintermediatesubsetofdataencompassedgradientsof668mofmeanaltitudeextent,134mofmeaninter-siteresolution,andagreaterpercentageofstudiesatlocal)thanatlandscapescale(29).Thelaxsubsetofdataencompassedgradientsof808mofmeanaltitudeextent,137mofmeaninter-siteresolution,andsimilarpercentageofstudiesatland-scapeandlocalscale.Identi“cationofpatternsWeanalyzed109altituderichnessgradientsbytwomethods:(1)astandardprotocolthatallowedthestatisticaldescriptionofpattern(hereafterreferredasstatisticalmethod)and(2)visualexaminationofshape(visualmethod).Ourpurposeherewasprimarydescriptive,andweusedstandardstatisticalorvisualmethodstoaccountfortheshapeofaltitudinalrichnessgradients(e.g.,RoweandLidgard2009forasimilarStatisticalmethod.„Foreachdataset,weregresseddataofrichness()onaltitude()toevaluatethelikelihoodofthedatagivenfourdifferentmodels.Model1:asimplelinearSRArelationship(istheinterceptandistheslope)thatdescribedamonotonicdecreasing(DEC)orincreasingpattern(INC),dependingonthesignoftheslope.Model2:anonlinearSRArelationshipoftheformwhichdescribedalow-plateaupattern(L-PL;ratherconstanthighrichnessatlowaltitudesfollowedbyadecreaseinrichness)when0.Model3:oftheform,withistheintercept,andregressioncoef“cients,which,dependinguponthesigncoef“cient,describedahump-shapedpattern(H-0),orU-shapedpattern(U-SH;0).Model4:theonly-interceptmodel()thatevaluatedthelackofaltitudinalpattern(NP).ForeachdatasetassignedtoH-SH,wefurthercheckedthatthemaximumrichnessfellwithintherangeofaltitudesencompassedbythedata.ThestatisticalmethodallowedthedetectionofotherformsintheSRArelationship(e.g.,J-shapedorL-shapedpatterns)thatwemaintainedinasinglecategory(OTH).To“ndthebestexplanatorymodel,i.e.,forassign-mentofeachdatasettoadifferentpattern,weusedtheAkaikesinformationcriterioncorrectedforsmallsamples(AIC;BurnhamandAnderson2002,Diniz-Filhoetal.2008),whichallowedorderingthefourmodels“ttedtoeachdatasetfrombesttoworst.WeconsideredthemodelhavingtheminimumAICasthebestmodelsupportedbythedata.Weestimatedthesizeoftheincrementsofinformationloss()foreachmodelcomparedtotheestimatedbestmodel();modelshaving2ofthebestmodelwereconsideredtohaveconsiderablelesssupport(BurnhamandAnderson2002,Diniz-Filhoetal.2008).Modelsthathada2ofthebestmodelwereconsideredequallylikelyforaparticulardataset;inthiscase,weassignedsupporttoeachSRApatterninvolved,inequalproportions,dividing1bythetotalnumberofSRApatternssupported.Visualmethod.„Foreachdataset,weelaboratedascatterplotofthevariationofrichnessasafunctionofaltitudeforvisualidenti“cationofshape:monotonicdecreasing(DEC),monotonicincreasing(INC),hump-shaped(H-SH),U-shaped(U-SH),andlow-plateau(L-PL).Tominimizetheinherentsubjectivityofthismethod,wefollowedMcCain(2009)scriteriaforassignmentofSRArelationshipsintodifferentpatterns.Wede“nedDECandINCasthosepatternsinwhichspeciesrichness,respectively,declinedorincreasedmonotonicallywithelev

5 ation.H-SHwasaunimodal January2011 shape
ation.H-SHwasaunimodal January2011 shape(Ribeiroetal.1998).SuchmultipleformsintheSRArelationshipquestiontheuniversalityofthealtitudinaldiversitygradient.Here,wecompiledSRArelationshipsfromdifferentpartsoftheworldtotestthehypothesisthatidenti“cationofSRApatternsdependsuponthequalityoforiginalstudiesincludedinmeta-analysisandthemethodusedtoidentifyshape.Wetestedthepredictionsthat(1)uncertaintyoveridenti“-cationofshapeincreasesasthecriteriausedtoincludeastudyintotheanalysisbecomesmorelax,and(2)studiesthatconformtostringentselectioncriteriashowrobustaltitudinalrichnesspatternswithrespecttovariationinmethodusedtoidentifyshape.Theselectionofpublishedstudiesusedtocreatedifferentsubsetsofdatahasbeenusefultoexaminetheeffectsofscale,sampling,andarealstandardizationonSRApatterns(Rahbek2005).OurstudywillshowthattheuseofdifferentmethodstoidentifySRAshapesinteractswiththequalityofbasicdatatoin”uencetherelativedistributionofpatterns.Acaveatisneededabouttheapproachadoptedinthepresentstudy,whichuseddatafrommountainsindifferentpartsoftheworldtocomparetherelativefrequencyofSRApatterns,butwithoutdisentanglingtheroleofdifferentenvironmentaldriversonshape.Datacollectedalongaltitudinalgradientsre”ectthecombinedeffectofgeneralclimaticandgeophysicalchangeswithaltitudeandregionalphenomena(e.g.,historyandisolationofmountainbiota;seePlate1);hence,ithasbeensuggestedthatthelackofastandardmountaincomplicatestheinterpretationofdiscrepancybetween“ndingsfromdifferentaltitudinalgradientsbydifferentresearchersifonlyaltitudeistakenintoaccountasexplanatoryvariable(Korner2007).None-theless,throughoutthepresentstudyaltitudewasnotconsideredthedrivingfactorforspeciesrichness,butjustthetemplateforourmeta-analysis.Ourpurposeherewastoevaluatetheextenttowhichdecisionstakenbyresearchersatthetimeofdatacompilationandassignmentofgradientstodifferentshapesmaycomplicatedetectionofrobustpatternsinmeta-analysis;con“denceintheidenti“cationofshapeisneededbeforeanyattempttoidentifyunderlyingdriversofSRArelationshipsismade.SelectionofdataWecarriedoutaliteraturesearchthroughZoologicalRecordandScopus(availableonlineWeusedaltitud*orelevation*richnessordiversityaskeywords.Asterisksareusedtosubstituteforanyothercharacterorcharactersinthesearchstring.Forinstance,searchtermssuchasaltitud*wouldreturnanywordthatbeginswithaltitud,suchasaltitudeandaltitudinal.Wesearchedforpaperswithanyofthe“rsttwowords[altitud*orelevation*]anyofthesecondtwowords[richnessordiversity].Additionalpaperswereobtainedbyexaminingthereferencesoforiginalarticles.Weselectedallpapersthatreporteddataontherichnessofarthropodspeciesforatleast“vedifferentelevations.Toreducetheso-called“ledrawerprob-lem(Rosenthal1979,Csadaetal.1996),weselectedstudiesthatwereoriginallydesignedtotesttheSRArelationshipalongwithothersthatwerenotspeci“callydesignedforthispurpose.Weincludedgradientsprovidedtheyreporteddataonrawspeciesrichness(neitherrare“ednortransformed)foreachaltitude.Forafewstudieswheredataonrichnesswerenotavailable,weestimatedlocalrichnessateachaltitudebasedoncon“rmed(i.e.,notinterpolated)presenceofspecies.Toovercometheproblemofpseudo-replication,weselectedpapersfromthesameauthor/sprovidedtheyworkedwithdifferentdatasets,andweanalyzedrichnessdatafromdifferentyearsinthesamelocationonlyiftheywerefromdifferentsamplingpoints.Whenastudycombineddataonlocalric

6 hnessestimationsfromseveralmountainstode
hnessestimationsfromseveralmountainstodescribearegionalaltitude…richnessrelationship,weselecteddataonthelocalaltitudinalgradientsanddiscardedvaluesatregionalscale.Whenastudyreportedrichness…altitudedataofsubordinatetaxa(e.g.,subfamilies)withinahighertaxonomiclevel(e.g.,family),weseparatelyanalyzedthedataoneachsubordinatetaxonanddiscardedthevaluesreportedathighertaxonomiclevel.Ourselectionprocessresultedinaworkinglistof75studieswithaltitude…richnessdataon109altitudinalgradients(seeAppendixA).DataonlypublishedingraphicformweredigitizedusingDataThiefII1.1.0(Tummers2006;availableonlineOurlastsearchwasinDecember2007,andpaperspublishedsincethenwerenotincludedinourstudy.CriteriausedtoclassifygradientsintodifferentsubsetsofdataWeclassi“edeachaltitudinalgradientwithrespecttofourvariablesthatallowedtheirsubsequentinclusionintodifferentsubsetsofdata(seeAppendixA:TableA1).Foreachstudy,werecordedthreefactorsinvolvingsamplingdesign(points1…3)andoneinvolvingtheimpactsofhumanpresence(point4):(1)Proportionofgradientsampled.McCain(2009)proposedthatanalysisofSRArelationshipsshouldbebasedonstudiesthatcoveratleast70ofthetotalmountainrange.(2)Samplingstandardization.Werecordedthepresenceofstandardizedorequalsam-plingeffortacrossdifferentaltitudes,whichareknownhttp://thomsonreuters.com/products_services/science/science_products/a-z/zoological_recordhttp://www.scopus.com/home.urlhttp://datathief.org/ Ecology,Vol.92,No.1 Csada,R.,P.James,andR.Espie.1996.The“ledrawerproblemofnon-signi“cantresults:doesitapplytobiologicalresearch?Oikos76:591…593.Diniz-Filho,J.A.F.,T.F.L.V.B.Rangel,andL.M.Bini.2008.Modelselectionandinformationtheoryingeograph-icalecology.GlobalEcologyandBiogeography17:479…488.Grif“ths,D.1998.Samplingeffort,regressionmethod,andtheshapeandslopeofsize-abundancerelations.JournalofAnimalEcology67:795…804.Hedges,L.V.,andI.Olkin.1985.Statisticalmethodsformeta-analysis.AcademicPress,NewYork,NewYork,USA.Hillebrand,H.2004.Onthegeneralityofthelatitudinaldiversitygradient.TheAmericanNaturalist163:192…211.Jankowski,T.,andG.A.Weyhenmeyer.2006.Theroleofspatialscaleandareaindeterminingrichness-altitudegradientsinSwedishlakephytoplanktoncommunities.Oikos115:433…442.Janzen,D.H.,M.Ataroff,M.Farinas,S.Reyes,N.Rincon,A.Soler,P.Soriano,andM.Vera.1976.ChangesinthearthropodcommunityalonganelevationaltransectintheVenezuelanAndes.Biotropica8:193…203.rner,C.2007.Theuseofaltitudeinecologicalresearch.TrendsinEcologyandEvolution22:569…574.Lawton,J.H.,M.MacGarvin,andP.A.Heads.1987.EffectsofaltitudeontheabundanceandspeciesrichnessofinsectherbivoresonBracken.JournalofAnimalEcology56:147…Lods-Crozet,B.,E.Castella,D.Cambin,C.Ilg,S.Knispel,andH.Mayor-Simeant.2001.MacroinvertebratecommunitystructureinrelationtoenvironmentalvariablesinaSwissglacialstream.FreshwaterBiology46:1641…1661.Lyons,S.K.,andM.R.Willig.1999.Ahemisphericassessmentofscale-dependenceinlatitudinalgradientsofspeciesrichness.Ecology80:2483…2491.Lyons,S.K.,andM.R.Willig.2002.Speciesrichness,latitude,andscale-sensitivity.Ecology83:47…58.MacArthur,R.H.1972.Geographicalecology.HarperandRowe,NewYork,NewYork,USA.McCain,C.M.2007.Couldtemperatureandwateravailabilitydriveelevationalspeciesrichnesspatterns?Aglobalcasestudyforbats.GlobalEcologyandBiogeography16:1…13.McCain,C.M.2009.Globalanalysisofbirdelevationaldiversity.GlobalEcologyandBiogeography18

7 :346…360.s-Bravo,D.,M.B.Araujo,T.S.Romda
:346…360.s-Bravo,D.,M.B.Araujo,T.S.Romdal,andC.Rahbek.2008.Scaleeffectsandhumanimpactontheelevationalspeciesrichnessgradients.Nature453:216…219.RDevelopmentCoreTeam.2009.R:alanguageandenvironmentforstatisticalcomputing.RFoundationforStatisticalComputing,Vienna,Austria.http://www.Rahbek,C.1995.Theelevationalgradientofspeciesrichness:auniformpattern?Ecography18:200…205.Rahbek,C.1997.Therelationshipamongarea,elevation,andregionalspeciesrichnessinneotropicalbirds.TheAmericanNaturalist149:875…902.Rahbek,C.2005.Therolofspatialscaleandtheperceptionoflarge-scalespecies-richnesspatterns.EcologyLetters8:224…Rahbek,C.,andG.R.Graves.2000.Detectionofmacro-ecologicalpatternsinSouthAmericanhummingbirdsisaffectedbyspatialscale.ProceedingsoftheRoyalSocietyofLondonB267:2259…2265.Ribeiro,S.P.,M.A.A.Carneiro,andG.W.Fernandes.1998.Free-feedinginsectherbivoresalongenvironmentalgradientsinSerradoCipo:basisforamanagementplan.JournalofInsectConservation2:107…118.Rosenthal,R.1979.The“ledrawerproblemandtolerancefornullresults.PsychologicalBulletin86:638…641.Rowe,R.J.,andS.Lidgard.2009.Elevationalgradientsandspeciesrichness:domethodschangepatternperception?GlobalEcologyandBiogeography18:163…177.Sanders,N.J.,R.R.Dunn,M.C.Fitzpatrick,C.E.Carlton,M.R.Pogue,C.R.Parker,andT.R.Simons.2009.Adiversityofelevationaldiversitygradients.Pages75…87rnerandE.Spehn,editors.Dataminingforglobaltrendsinmountainbiodiversity.CRCPress,BocaRaton,Florida,Sanders,N.J.,J.Moss,andD.Wagner.2003.Patternsofantspeciesrichnessalongelevationalgradientsinanaridecosystem.GlobalEcologyandBiogeography12:93…102.StatXact-6.2003.Astatisticalpackageforexactnonparametricinference.CytelSoftwareCorporation,Cambridge,Massa-chusetts,USA.Turner,B.D.,andE.Broadhead.1974.ThediversityanddistributionofpsocidpopulationsonMangiferaindicaL.inJamaicaandtheirrelationshiptoaltitudeandmicro-epiphytediversity.JournalofAnimalEcology43:173…190.Whittaker,R.J.2010.Meta-analysisandmega-mistakes:callingtimeonmeta-analysisofthespeciesrichness…productivityrelationship.Ecology91:2522…2533.Whittaker,R.J.,andE.Heegaard.2003.Whatistheobservedrelationshipbetweenspeciesrichnessandproductivity?Ecology84:3384…3390.Whittaker,R.J.,K.J.Willis,andR.Field.2001.Scaleandspeciesrichness:towardsageneral,hierarchicaltheoryofspeciesdiversity.JournalofBiogeography28:453…470.APPENDIXADetailsofthepapersusedasbasicdata(EcologicalArchivesE092-021-A1).APPENDIXBRelativefrequenciesdistributionsofSRArelationshipsobservedindifferenttaxaandindifferentclimaticandbiogeographicregions(EcologicalArchivesE092-021-A2). Ecology,Vol.92,No.1 ThestringentcriteriaofinclusionofSRAdatacame upwithasubsetof21datasetsthatshowedaconsistent orderintherelativeimportanceofthemostabundant SRApatternsthroughoutbothmethodsofanalysisand consensus(HS-H . DEC . L-PL;Fig.1d…f).The percentagerepresentationofH-SHratherincreased,and L-PLdecreasedwiththevisualmethodandconsensus, comparedtothestatisticalmethod(Fig.1d…f).The percentagerepresentationofNP,andmoreoverof CONT,waslowerthaninthewholedataset(compare Fig.1c,f). RelaxingtheselectionofSRAdata,cameupwithtwo datasubsetsof42(intermediatesubset)and46(lax subset)SRArelationships.Therelativefrequency distributionofdifferentSRApatternsinbothsubsets largelyfollowedtheobservedinthewholedatasetbyall methodsofanalysis(compareFig.1a…candg…l).The statisticalmethodshowedthe

8 highestpercentageofNP (Fig.1j,k).Thevisu
highestpercentageofNP (Fig.1j,k).Thevisualmethodsomewhatdecreasedthe percentageofL-PLandNP,whichincreasedthe percentageofH-SHandofDEC(Fig.1h,k).The consensuscon“rmedthepredominanceofDEC,al- thoughtherewasahighpercentageofNPandCONT (Fig.1l).INCandU-SHpatternswererarethroughout (Fig.1j…l). Aftercontrollingfordifferencesinsamplesize,we furthercon“rmedthatrelaxingtheselectionofSRA dataincreaseduncertaintyintheestimationofthe magnitudeofSRArelationships(Fig.2). Therewasnoassociationofshapewithtaxonomyor climateafterourdatamanipulation(Table1,Appendix B:Figs.B2andB3),butasigni“cantbiogeographical relationshipwasevidentinthestringentsubset(Table 1),whichsuggestedthatH-SHpatternswererecorded onlyintheNearcticandPalearticregions,whereasDEC weremorefrequentintheNeotropics(AppendixB:Fig. B4). D ISCUSSION OurmanipulationofSRArelationshipsintodifferent subsetsofdatashowedthatthequalityofbasicdata selectedformeta-analysisiscrucialtoreliableidentify shape(seeWhittaker2010fordiscussion).Wecon“rmed thetwopredictionsproposedattheoutsetofthepresent study.Ingeneral,uncertaintyoveridenti“cationof shapeincreasedasthecriteriaofinclusionofstudiesinto theanalysisbecamemorelax;studiesthatconformedto stringentselectioncriteriashowedrobustnessinthe SRApatternstovariationinmethodusedtoidentify shape.Rahbek(2005)demonstratedthatdecisions concerningtheanalyticaldesignofindividualstudies cancompletelyturnaroundthestatisticaloutcome relatedtotheshapeoftheSRApattern(butseeRowe andLidgard2009).Ourstudyshowedhowtheseeffects couldalsointeractwithtwomethods(statisticaland visual)usedtoidentifyshapetoaffecttheoverall relativefrequencydistributionsofSRApatterns. ThetotalSRArelationshipstakentogethersuggested thepredominanceofmonotonicdecreasingpatterns afterconsensus.However,theapplicationofstringent selectioncriteriacon“rmedthepredominanceofhump- shapedpatternsovermonotonicdecreasingones.In general,therelativefrequencyofpatternsinthewhole datasetparalleledthosefoundintheintermediateand laxsubsets,whichtakentogetherrepresented . 80 % of theSRArelationshipsinourstudy.Changesinthe proportionalrepresentationofshapesafterdatamanip- ulationwerenotassociatedwithclimateandtaxonomy, althoughwefoundanassociationwithbiogeographyin thestringentsubset.Thisassociationsuggesteda F IG .2.Effectofrelaxingtheselectioncriteriaofstudies, fromstringent(triangles),tointermediate(circles),andlax (squares),overuncertainty(shownas95 % con“denceintervals) intheestimationofacommoneffectsize( z r ,Fishers z - transformation)forH-SH(hump-shaped),DEC(monotonic decreasing),andL-PL(low-plateau)patterns. T ABLE 1.Testsofassociationbetweentherelativedistributionoffrequenciesofthethreemostabundantpatterns(hump-shaped, monotonicdecreasing,andlow-plateau)ofarthropoddistributionafterconsensusandtaxonomy,climate,andbiogeography. Testofassociation StringentIntermediateLax FIdf P FIdf P FIdf P Taxa 3 SRApattern13.648100.11711.086120.82614.080140.706 Climaticregion 3 SRApattern15.425120.17915.534140.15714.333160.987 Biogeographicregion 3 SRApattern13.81060.0039.35680.27111.346120.818 Notes: Abbreviationsare:FI,permutation-basedFisher-Freeman-Haltonstatistic;df,degreesoffreedom; P ,probabilitylevel; SRA,speciesrichness…altituderelationship.See Methods fordescriptionsofthestringent,intermediate,andlaxselectioncriteria. NOTES 258 Ecology,Vol.92,N