Selection bias - Wikipedia

Selection bias is the bias introduced by the selection of individuals, groups, or data for analysis in such a way that proper randomization is not achieved, ... Selectionbias FromWikipedia,thefreeencyclopedia Jumptonavigation Jumptosearch Biasinastatisticalanalysisduetonon-randomselection Selectionbiasisthebiasintroducedbytheselectionofindividuals,groups,ordataforanalysisinsuchawaythatproperrandomizationisnotachieved,therebyfailingtoensurethatthesampleobtainedisrepresentativeofthepopulationintendedtobeanalyzed.[1]Itissometimesreferredtoastheselectioneffect.Thephrase"selectionbias"mostoftenreferstothedistortionofastatisticalanalysis,resultingfromthemethodofcollectingsamples.Iftheselectionbiasisnottakenintoaccount,thensomeconclusionsofthestudymaybefalse. Contents 1Types 1.1Samplingbias 1.2Timeinterval 1.3Exposure 1.4Data 1.5Studies 1.6Attrition 1.7Observerselection 1.8Volunteerbias 2Mitigation 3Relatedissues 4Seealso 5References Types[edit] Samplingbias[edit] Samplingbiasissystematicerrorduetoanon-randomsampleofapopulation,[2]causingsomemembersofthepopulationtobelesslikelytobeincludedthanothers,resultinginabiasedsample,definedasastatisticalsampleofapopulation(ornon-humanfactors)inwhichallparticipantsarenotequallybalancedorobjectivelyrepresented.[3]Itismostlyclassifiedasasubtypeofselectionbias,[4]sometimesspecificallytermedsampleselectionbias,[5][6][7]butsomeclassifyitasaseparatetypeofbias.[8] Adistinctionofsamplingbias(albeitnotauniversallyacceptedone)isthatitunderminestheexternalvalidityofatest(theabilityofitsresultstobegeneralizedtotherestofthepopulation),whileselectionbiasmainlyaddressesinternalvalidityfordifferencesorsimilaritiesfoundinthesampleathand.Inthissense,errorsoccurringintheprocessofgatheringthesampleorcohortcausesamplingbias,whileerrorsinanyprocessthereaftercauseselectionbias. Examplesofsamplingbiasincludeself-selection,pre-screeningoftrialparticipants,discountingtrialsubjects/teststhatdidnotruntocompletionandmigrationbiasbyexcludingsubjectswhohaverecentlymovedintooroutofthestudyarea,length-timebias,whereslowlydevelopingdiseasewithbetterprognosisisdetected,andleadtimebias,wherediseaseisdiagnosedearlierparticipantsthanincomparisonpopulations,althoughtheaveragecourseofdiseaseisthesame. Timeinterval[edit] Earlyterminationofatrialatatimewhenitsresultssupportthedesiredconclusion. Atrialmaybeterminatedearlyatanextremevalue(oftenforethicalreasons),buttheextremevalueislikelytobereachedbythevariablewiththelargestvariance,evenifallvariableshaveasimilarmean. Exposure[edit] Susceptibilitybias Clinicalsusceptibilitybias,whenonediseasepredisposesforaseconddisease,andthetreatmentforthefirstdiseaseerroneouslyappearstopredisposetotheseconddisease.Forexample,postmenopausalsyndromegivesahigherlikelihoodofalsodevelopingendometrialcancer,soestrogensgivenforthepostmenopausalsyndromemayreceiveahigherthanactualblameforcausingendometrialcancer.[9] Protopathicbias,whenatreatmentforthefirstsymptomsofadiseaseorotheroutcomeappeartocausetheoutcome.Itisapotentialbiaswhenthereisalagtimefromthefirstsymptomsandstartoftreatmentbeforeactualdiagnosis.[9]Itcanbemitigatedbylagging,thatis,exclusionofexposuresthatoccurredinacertaintimeperiodbeforediagnosis.[10] Indicationbias,apotentialmixupbetweencauseandeffectwhenexposureisdependentonindication,e.g.atreatmentisgiventopeopleinhighriskofacquiringadisease,potentiallycausingapreponderanceoftreatedpeopleamongthoseacquiringthedisease.Thismaycauseanerroneousappearanceofthetreatmentbeingacauseofthedisease.[11] Data[edit] Partitioning(dividing)datawithknowledgeofthecontentsofthepartitions,andthenanalyzingthemwithtestsdesignedforblindlychosenpartitions. Posthocalterationofdatainclusionbasedonarbitraryorsubjectivereasons,including: Cherrypicking,whichactuallyisnotselectionbias,butconfirmationbias,whenspecificsubsetsofdataarechosentosupportaconclusion(e.g.citingexamplesofplanecrashesasevidenceofairlineflightbeingunsafe,whileignoringthefarmorecommonexampleofflightsthatcompletesafely.See:Availabilityheuristic) Rejectionofbaddataon(1)arbitrarygrounds,insteadofaccordingtopreviouslystatedorgenerallyagreedcriteriaor(2)discarding"outliers"onstatisticalgroundsthatfailtotakeintoaccountimportantinformationthatcouldbederivedfrom"wild"observations.[12] Studies[edit] Selectionofwhichstudiestoincludeinameta-analysis(seealsocombinatorialmeta-analysis). Performingrepeatedexperimentsandreportingonlythemostfavorableresults,perhapsrelabellinglabrecordsofotherexperimentsas"calibrationtests","instrumentationerrors"or"preliminarysurveys". Presentingthemostsignificantresultofadatadredgeasifitwereasingleexperiment(whichislogicallythesameasthepreviousitem,butisseenasmuchlessdishonest). Attrition[edit] Attritionbiasisakindofselectionbiascausedbyattrition(lossofparticipants),[13]discountingtrialsubjects/teststhatdidnotruntocompletion.Itiscloselyrelatedtothesurvivorshipbias,whereonlythesubjectsthat"survived"aprocessareincludedintheanalysisorthefailurebias,whereonlythesubjectsthat"failed"aprocessareincluded.Itincludesdropout,nonresponse(lowerresponserate),withdrawalandprotocoldeviators.Itgivesbiasedresultswhereitisunequalinregardtoexposureand/oroutcome.Forexample,inatestofadietingprogram,theresearchermaysimplyrejecteveryonewhodropsoutofthetrial,butmostofthosewhodropoutarethoseforwhomitwasnotworking.Differentlossofsubjectsininterventionandcomparisongroupmaychangethecharacteristicsofthesegroupsandoutcomesirrespectiveofthestudiedintervention.[13] Losttofollow-up,isanotherformofAttritionbias,mainlyoccurringinmedicinalstudiesoveralengthytimeperiod.Non-ResponseorRetentionbiascanbeinfluencedbyanumberofbothtangibleandintangiblefactors,suchas;wealth,education,altruism,initialunderstandingofthestudyandit'srequirements.[14]Researcher'smayalsobeincapableofconductingfollow-upcontactresultingfrominadequateidentifyinginformationandcontactdetailscollectedduringtheinitialrecruitmentandresearchphase.[15] Observerselection[edit] PhilosopherNickBostromhasarguedthatdataarefilterednotonlybystudydesignandmeasurement,butbythenecessarypreconditionthattherehastobesomeonedoingastudy.Insituationswheretheexistenceoftheobserverorthestudyiscorrelatedwiththedata,observationselectioneffectsoccur,andanthropicreasoningisrequired.[16] AnexampleisthepastimpacteventrecordofEarth:iflargeimpactscausemassextinctionsandecologicaldisruptionsprecludingtheevolutionofintelligentobserversforlongperiods,noonewillobserveanyevidenceoflargeimpactsintherecentpast(sincetheywouldhavepreventedintelligentobserversfromevolving).HencethereisapotentialbiasintheimpactrecordofEarth.[17]Astronomicalexistentialrisksmightsimilarlybeunderestimatedduetoselectionbias,andananthropiccorrectionhastobeintroduced.[18] Volunteerbias[edit] Self-selectionbiasoravolunteerbiasinstudiesofferfurtherthreatstothevalidityofastudyastheseparticipantsmayhaveintrinsicallydifferentcharacteristicsfromthetargetpopulationofthestudy.[19]Studieshaveshownthatvolunteerstendtocomefromahighersocialstandingthanfromalowersocio-economicbackground.[20]Furthermore,anotherstudyshowsthatwomenaremoreprobabletovolunteerforstudiesthanmales.Volunteerbiasisevidentthroughoutthestudylife-cycle,fromrecruitmenttofollow-ups.Moregenerallyspeakingvolunteerresponsecanbeputdowntoindividualaltruism,adesireforapproval,personalrelationtothestudytopicandotherreasons.[20][14]Aswithmostinstancesmitigationinthecaseofvolunteerbiasisanincreasedsamplesize.[citationneeded] Mitigation[edit] Inthegeneralcase,selectionbiasescannotbeovercomewithstatisticalanalysisofexistingdataalone,thoughHeckmancorrectionmaybeusedinspecialcases.Anassessmentofthedegreeofselectionbiascanbemadebyexaminingcorrelationsbetweenexogenous(background)variablesandatreatmentindicator.However,inregressionmodels,itiscorrelationbetweenunobserveddeterminantsoftheoutcomeandunobserveddeterminantsofselectionintothesamplewhichbiasestimates,andthiscorrelationbetweenunobservablescannotbedirectlyassessedbytheobserveddeterminantsoftreatment.[21] Whendataareselectedforfittingorforecastpurposes,acoalitionalgamecanbesetupsothatafittingorforecastaccuracyfunctioncanbedefinedonallsubsetsofthedatavariables. Relatedissues[edit] Selectionbiasiscloselyrelatedto: publicationbiasorreportingbias,thedistortionproducedincommunityperceptionormeta-analysesbynotpublishinguninteresting(usuallynegative)results,orresultswhichgoagainsttheexperimenter'sprejudices,asponsor'sinterests,orcommunityexpectations. confirmationbias,thegeneraltendencyofhumanstogivemoreattentiontowhateverconfirmsourpre-existingperspective;orspecificallyinexperimentalscience,thedistortionproducedbyexperimentsthataredesignedtoseekconfirmatoryevidenceinsteadoftryingtodisprovethehypothesis. exclusionbias,resultsfromapplyingdifferentcriteriatocasesandcontrolsinregardstoparticipationeligibilityforastudy/differentvariablesservingasbasisforexclusion. Seealso[edit] Berkson'sparadox –Tendencytomisinterpretstatisticalexperimentsinvolvingconditionalprobabilities Blackswantheory –Theoryofresponsetosurpriseevents Cherrypicking –Fallacyofincompleteevidence Fundingbias –Tendencyofascientiicstudytosupporttheinterestsofitsfunder Listofcognitivebiases –Systematicpatternsofdeviationfromnormorrationalityinjudgment Participationbias –Typeofbias Publicationbias –Higherprobabilityofpublishingresultsshowingasignificantfinding Reportingbias –Biasinthereportingofinformation Samplingbias –Biasinthesamplingofapopulation Samplingprobability Selectiveexposuretheory –Theorywithinthepracticeofpsychology Self-fulfillingprophecy –Predictionthatcausesitselftobecometrue Survivorshipbias –Logicalerror,formofselectionbias Frequencyillusion References[edit] ^DictionaryofCancerTerms→selectionbias.RetrievedonSeptember23,2009. ^MedicalDictionary-'SamplingBias'RetrievedonSeptember23,2009 ^TheFreeDictionary→biasedsample.Retrievedon2009-09-23.Siteinturncites:Mosby'sMedicalDictionary,8thedition. ^DictionaryofCancerTerms→SelectionBias.RetrievedonSeptember23,2009. ^Ards,Sheila;Chung,Chanjin;Myers,SamuelL.(1998)."Theeffectsofsampleselectionbiasonracialdifferencesinchildabusereporting".ChildAbuse&Neglect.22(2):103–115.doi:10.1016/S0145-2134(97)00131-2.PMID 9504213. ^Cortes,Corinna;Mohri,Mehryar;Riley,Michael;Rostamizadeh,Afshin(2008).SampleSelectionBiasCorrectionTheory(PDF).AlgorithmicLearningTheory.LectureNotesinComputerScience.Vol. 5254.pp. 38–53.arXiv:0805.2775.CiteSeerX 10.1.1.144.4478.doi:10.1007/978-3-540-87987-9_8.ISBN 978-3-540-87986-2.S2CID 842488. ^Cortes,Corinna;Mohri,Mehryar(2014)."Domainadaptationandsamplebiascorrectiontheoryandalgorithmforregression"(PDF).TheoreticalComputerScience.519:103–126.CiteSeerX 10.1.1.367.6899.doi:10.1016/j.tcs.2013.09.027. ^Fadem,Barbara(2009).BehavioralScience.LippincottWilliams&Wilkins.p. 262.ISBN 978-0-7817-8257-9. ^abFeinsteinAR;HorwitzRI(November1978)."Acritiqueofthestatisticalevidenceassociatingestrogenswithendometrialcancer".CancerRes.38(11Pt2):4001–5.PMID 698947. ^TamimH;MonfaredAA;LeLorierJ(March2007)."Applicationoflag-timeintoexposuredefinitionstocontrolforprotopathicbias".PharmacoepidemiolDrugSaf.16(3):250–8.doi:10.1002/pds.1360.PMID 17245804.S2CID 25648490. ^MatthewR.Weir(2005).Hypertension(KeyDiseases)(AcpKeyDiseasesSeries).Philadelphia,Pa:AmericanCollegeofPhysicians.p. 159.ISBN 978-1-930513-58-7. ^Kruskal,WilliamH.(1960)."SomeRemarksonWildObservations".Technometrics.2(1):1–3.doi:10.1080/00401706.1960.10489875. ^abJüni,P.;Egger,Matthias(2005)."Empiricalevidenceofattritionbiasinclinicaltrials".InternationalJournalofEpidemiology.34(1):87–88.doi:10.1093/ije/dyh406.PMID 15649954. ^abJordan,Sue;Watkins,Alan;Storey,Mel;Allen,StevenJ.;Brooks,CarolineJ.;Garaiova,Iveta;Heaven,MartinL.;Jones,Ruth;Plummer,SueF.;Russell,IanT.;Thornton,CatherineA.(2013-07-09)."VolunteerBiasinRecruitment,Retention,andBloodSampleDonationinaRandomisedControlledTrialInvolvingMothersandTheirChildrenatSixMonthsandTwoYears:ALongitudinalAnalysis".PLOSONE.8(7):e67912.Bibcode:2013PLoSO...867912J.doi:10.1371/journal.pone.0067912.ISSN 1932-6203.PMC 3706448.PMID 23874465. ^Small,W.P.(1967-05-06)."LosttoFollow-Up".TheLancet.OriginallypublishedasVolume1,Issue7497.289(7497):997–999.doi:10.1016/S0140-6736(67)92377-X.ISSN 0140-6736.PMID 4164620. ^Bostrom,Nick(2002).AnthropicBias:ObservationSelectionEffectsinScienceandPhilosophy.NewYork:Routledge.ISBN 978-0-415-93858-7. ^Ćirković,M.M.;Sandberg,A.;Bostrom,N.(2010)."AnthropicShadow:ObservationSelectionEffectsandHumanExtinctionRisks".RiskAnalysis.30(10):1495–506.doi:10.1111/j.1539-6924.2010.01460.x.PMID 20626690. ^Tegmark,M.;Bostrom,N.(2005)."Astrophysics:Isadoomsdaycatastrophelikely?".Nature.438(7069):754.Bibcode:2005Natur.438..754T.doi:10.1038/438754a.PMID 16341005.S2CID 4390013. ^Tripepi,Giovanni;Jager,KittyJ.;Dekker,FriedoW.;Zoccali,Carmine(2010)."SelectionBiasandInformationBiasinClinicalResearch".NephronClinicalPractice.115(2):c94–c99.doi:10.1159/000312871.ISSN 1660-2110.PMID 20407272. ^ab"Volunteerbias".CatalogofBias.2017-11-17.Retrieved2020-10-29. ^Heckman,J.J.(1979)."SampleSelectionBiasasaSpecificationError".Econometrica.47(1):153–161.doi:10.2307/1912352.JSTOR 1912352. vteBiasesCognitivebiases Actor–observer Acquiescence Ambiguity Anchoring Attentional Attribution Authority Automation Belief Blindspot Choice-supportive Confirmation Compassionfade Congruence Courtesy Cultural Distinction Dunning–Kruger Egocentric Emotional Extrinsicincentives Fadingaffect Framing Frequency Correspondence Haloeffect Hindsight Horneffect Hostileattribution Impact Implicit In-group Mere-exposureeffect Negativity Normalcy Omission Optimism Out-grouphomogeneity Outcome Overtonwindow Precision Present Pro-innovation Response Restraint Self-serving Socialcomparison Statusquo Time-saving Traitascription vonRestorffeffect Zero-risk Inanimals Statisticalbiases Estimator Forecast Healthyuser Information Psychological Leadtime Lengthtime Non-response Observer Omitted-variable Participation Recall Sampling Selection Self-selection Socialdesirability Spectrum Survivorship Systematicerror Systemic Verification Wet Otherbiases Academic Funding FUTON Inductive Infrastructure Inherent Ineducation Media Falsebalance VietnamWar Norway SouthAsia Sweden UnitedStates Arab–Israeliconflict Ukraine Net Politicalbias Publication Reporting Whitehat Biasreduction Cognitivebiasmitigation Debiasing Heuristicsinjudgmentanddecision-making Lists:General ·Memory vteClinicalresearchandexperimentaldesignOverview Clinicaltrial Trialprotocols Adaptiveclinicaltrial Academicclinicaltrials Clinicalstudydesign Controlledstudy(EBMItoII-1) Randomizedcontrolledtrial Scientificexperiment Blindexperiment Open-labeltrial Adaptiveclinicaltrial Platformtrial Observationalstudy(EBMII-2toII-3) Cross-sectionalstudyvs.Longitudinalstudy,Ecologicalstudy Cohortstudy Retrospective Prospective Case–controlstudy(Nestedcase–controlstudy) Caseseries Casestudy Casereport MeasuresOccurrenceIncidence,Cumulativeincidence,Prevalence,Pointprevalence,PeriodprevalenceAssociationRiskdifference,Numberneededtotreat,Numberneededtoharm,Riskratio,Relativeriskreduction,Oddsratio,HazardratioPopulationimpactAttributablefractionamongtheexposed,Attributablefractionforthepopulation,Preventablefractionamongtheunexposed,PreventablefractionforthepopulationOtherClinicalendpoint,Virulence,Infectivity,Mortalityrate,Morbidity,Casefatalityrate,Specificityandsensitivity,Likelihood-ratios,Pre-andpost-testprobabilityTrial/testtypes Invitro Invivo Animaltesting Animaltestingonnon-humanprimates First-in-manstudy Multicentertrial Seedingtrial Vaccinetrial Analysisofclinicaltrials Risk–benefitratio Systematicreview Replication Meta-analysis Intention-to-treatanalysis Interpretationofresults Selectionbias Survivorshipbias Correlationdoesnotimplycausation Nullresult Sexasabiologicalvariable Category Glossary Listoftopics Retrievedfrom"https://en.wikipedia.org/w/index.php?title=Selection_bias&oldid=1069965732" Categories:Sampling(statistics)ExperimentalbiasScientificmethodCausalinferenceHiddencategories:ArticleswithshortdescriptionShortdescriptionisdifferentfromWikidataAllarticleswithunsourcedstatementsArticleswithunsourcedstatementsfromJanuary2021 Navigationmenu Personaltools NotloggedinTalkContributionsCreateaccountLogin Namespaces ArticleTalk English Views ReadEditViewhistory More Search Navigation MainpageContentsCurrenteventsRandomarticleAboutWikipediaContactusDonate Contribute HelpLearntoeditCommunityportalRecentchangesUploadfile Tools WhatlinkshereRelatedchangesUploadfileSpecialpagesPermanentlinkPageinformationCitethispageWikidataitem Print/export DownloadasPDFPrintableversion Languages العربيةČeštinaDeutschEspañolEuskaraفارسیFrançais한국어ItalianoעבריתLietuviųМакедонски日本語РусскийSimpleEnglishSvenskaไทยTürkçeУкраїнська Editlinks