A unifiedmodelfordetectingefficientandinefficientoutliersindata envelopment analysis

Data envelopmentanalysis(DEA)usesextremeobservationstoidentifysuperiorperformance,makingit vulnerable tooutliers.Thispaperdevelopsaunifiedmodeltoidentifybothefficientandinefficientoutliers in DEA.Findingbothtypesisimportantsincemanypostanalyses,aftermeasuringefficiency,dependon the entiredistributionofefficiencyestimates.Thus,outliersthataredistinguishedbypoorperformance can significantlyaltertheresults.Besidesallowingtheidentificationofoutliers,themethoddescribedis consistent witharelaxedsetofDEAaxioms.Severalexamplesdemonstratetheneedforidentifyingboth efficient andinefficientoutliersandtheeffectivenessoftheproposedmethod.Applicationsofthemodel reveal thatobservationswithlowefficiencyestimatesarenotnecessarilyoutliers.Inaddition,astrategy to acceleratethecomputationisproposedthatcanapplytoinfluentialobservationdetection.