资源描述:
《A sas application to identify and evaluate outliers.pdf》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、NESUG2010ApplicationsDevelopmentASAS®ApplicationtoIdentifyandEvaluateOutliersRichardHendra,MDRC,NewYork,NYPauletteW.Staum,PaulWaldronConsulting,WestNyack,NYABSTRACTThispaperpresentsanapproachtooutlieridentificationandevaluationthatutilizesmultipleSASprocedurespackagedintoaunifiedapplication.T
2、heoutputincludesreportsandplots,withinformationaboutextremevalues,influencestatistics,andtheeffectofoutliersonamodelofrelationshipsamongvariables.Itproducesacompact,readablereportforeachvariableofinterest.Thefocusisonanalyzingoutlierswithinexperimentalresearchdesigns.INTRODUCTIONThispaperdesc
3、ribesanapplicationthatallowsanalyststoefficientlyidentifyandevaluateoutliers.Theapplicationproducesonepageofoutputforeachvariabletobeanalyzed.Eachpagecontainsmultipleperspectivesonanyoutliersforasinglevariable.Theapplicationincludestechniquesfor4tasks:1.Identifyoutliers2.Combinedifferenttypes
4、ofinformationforeachvariable3.Evaluateoutliers4.RepeattheprocessformanyvariablesAnoutlierisavalueoranobservationthatisquitedifferentfrommostoftheothervaluesorobservationsinadataset.Sometimestheexistenceofanoutlierisobviousfromexaminingthevaluesforonevariableonly(i.e.,fromaunivariateperspectiv
5、e).Atothertimes,anoutlierisonlyobviouswhenexaminingthevaluesforacombinationofvariablesinanobservation(i.e.,fromamultivariateperspective).Anoutlierinfluencesboththemeanandthevarianceofavariable‟sdistribution.Forsmallsamples,thiscaneasilydistorttheresultsofananalysis.Forlargesamples,theeffecton
6、themeanissmaller,buttheincreasedvariancecanchangethestatisticalsignificanceofregressionestimates.Agroupofborderlineoutliers(“fringeliers”)canalsohaveasimilareffectonanalysisresults.Also,theeffectofanoutlieronconclusionsforasub-groupcanbelargerthantheeffectforthefullsample.Outliersappearindata
7、fordifferentreasons,including:validnaturalvariationmeasurementerrorsincorrectselectionofasample(i.e.,millionairesinawelfarestudy)dataentryerrors(i.e.,monthlyorannualwagesenteredashourlywages)Whatshouldyoudoaboutoutliers?Dependingonhowthedataw