《Selection bias and sensitivity analysis》由会员分享，可在线阅读，更多相关《Selection bias and sensitivity analysis（17页珍藏版）》请在金锄头文库上搜索。
1、1 SelectionBiasandSensitivityAnalysis1 OverviewofselectionbiasRosenbaum ssensitivityanalysisStatalab runningsensitivityanalysiswithrbounds 2 ReadingsforThisSession Guo S Fraser M 2009 PropensityScoreAnalysis ThousandOaks CA SagePublications Inc Sections8 1 8 3 8 6ofChapter8 Rosenbaum P R 2002 Observ
2、ationalstudies 2nded NewYork Springer Chapter4 Rosenbaum P R 2005 Sensitivityanalysisinobservationalstudies InB S EverittandD C Howell Eds EncyclopediaofStatisticsinBehavioralScience pp 1809 1814 NewYork JohnWiley Sons Ltd 3 Overviewofselectionbias 4 ImportancetoStudySelectionBias Althoughthenewmeth
3、odsofpropensityscoreanalysis PSA addresssomeofthecorechallengesfoundincausalinferencesofobservationalstudies theycannotcompetewithrandomizedcontrolledtrial RCT onakeyissue PSAonlycontrolsforselectionbiasgeneratedbymeasuredcovariates whileRCTbalancedataandcontrolsforbiasgeneratedbybothmeasuredandunme
4、asuredcovariates Itisforthisreason researchersusingPSAshouldbealertedbyselectionbiases particularlybythosegeneratedbytheunmeasuredcovariates 5 SourcesofSelectionBias Self selectionResearcher sselectionTheclassicexample thefailuretopredictHarryS Truman s1948presidentialvictoryThemodernexample interne
5、tsurveyNonprobabilitysamplingAdministrator sselectionGeographicselectionorgeographicmismatchMeasurementerror nonrandomerrorRater seffectinlongitudinalstudySelectionduetoattrition 6 OvertBiasversusHiddenBias Rosenbaum 2002 p 71 Anobservationalstudyisbiasedifthetreatedandcontrolgroupsdifferpriortotrea
6、tmentinwaysthatmatterfortheoutcomesunderstudy Anovertbiasisonethatcanbeseeninthedataathand forinstance priortotreatment treatedsubjectsareobservedtohavelowerincomesthancontrols Ahiddenbiasissimilarbutcannotbeseenbecausetherequiredinformationwasnotobservedorrecorded Overtandhiddenbiasesthataffectcaus
7、alinferencearecontrolledinwell implementedrandomizedexperiments Balancingthedataonhiddenbiasisespeciallyaprobleminobservationalstudies 7 Rosenbaum ssensitivityanalysis 8 TheBasicIdea 1 Correctingthehidden selectionproblemessentiallyinvolvestakingstepssuchasrespecifyingananalyticmodelbyusingadditiona
8、lvariables collectingadditionaldata redesigningastudyasarandomizedexperiment ormakingotherendeavors Evidently thesestrategiescanbeexpensiveandtime consuming Itisoftenpreferable then tostartbyconductingsensitivityanalysestoestimatethelevelofbias Thistypeofanalysisseekstoanswerthequestion Howsensitive
9、arefindingstohiddenbias Althoughsensitivityanalysisisexploratory itisanimportantstepinanalysesusingthecorrectivemodels RosenbaumandRubin 1983 andRosenbaum 2002 2005 haverecommendedthatresearchersroutinelyconductsensitivityanalysesinobservationalstudies 9 TheBasicIdea 2 Thecrucialideaofthismethodis O
10、nemanipulatestheestimatedoddsofreceivingaparticulartreatmenttoseehowmuchtheestimatedtreatmenteffectsmayvary Whatonewantstofindisthattheestimatedtreatmenteffectsarerobusttoaplausiblerangeofselectionbiases Berk 2003 p 231 Ifastudyisfreeofhiddenbias thentwounitswithsameobservedcovariatesxshouldhavesame
11、chancesofreceivingtreatment Sothedifferenceinoddsratioofreceivingtreatmentbetweentwounitswiththesamecovariatesxindicatesthelevelofsensitivitytohiddenselectionbias 10 TheBasicIdea 2 Considertwounitsjandk iftheyhavethesameobservedcovariatesx i e x j x k butdifferentchancesofreceivingtreatment i e j k
12、thenit slikelythatthereishiddenbias Asensitivityanalysisasks Howwouldinferencesabouttreatmenteffectsbealteredbyhiddenbiasesofvariousmagnitudes Supposethe sdifferatagivenx Howlargewouldthesedifferenceshavetobetoalterthequalitativeconclusionsofastudy Rosenbaum 2002 p 106 11 TheBasicIdea 3 Theoddsthatu
13、nitsjandkreceivethetreatmentareand respectively TheoddsratioisThesensitivityanalysisgoesfurthertoassumethatthisoddsratioforunitswithsamexwasatmostsomenumberof 1 thatis 12 TheBasicIdea 4 Bytheabovedefinitions if were1 then j k wheneverx j x k sothestudywouldbefreeofhiddenbias If 2 thentwounitsthatapp
14、earsimilar thathavethesamex coulddifferintheiroddsofreceivingthetreatmentbyasmuchasafactorof2 sooneunitmightbetwiceaslikelyastheothertoreceivetreatment Thus isameasureofthedegreeofdeparturefromastudythatisfreeofhiddenbias Asensitivityanalysiswillconsiderseveralpossiblevaluesof andshowhowtheinference
15、smightchange Astudyissensitiveifvaluesof closeto1couldleadtoinferencesthatareverydifferentfromthoseobtainedassumingthestudyisfreeofhiddenbias Astudyisinsensitiveifextremevaluesof arerequiredtoaltertheinference Rosenbaum 2002 p 107 13 TheBasicIdea 5 Theoriginalframeworkofthesensitivityanalysisforhidd
16、enbiascamefromCornfieldetal 1959 However theCornfieldetalstudyignoredsamplingvariability whichishazardousexceptinextremelylargesamples Rosenbaum 2002 attendedtosamplingvariationbydevelopingmethodstocomputetheboundsoninferencequantities suchaspvaluesorconfidenceintervals Thus foreach 1 weobtainanintervalofpvaluesthatreflectuncertaintybecauseofhiddenbias As increases thisintervalbecomeslonger andeventuallyitbecomesuninformative includingbothlargeandsmallpvalues Thepoint atwhichtheintervalbecomesun
