简介:Inthispaper,weconsidertheempiricallikelihood-basedinferencesforvaryingcoefficientmodelsY=X~τα(U)+εwhenXaresubjecttomissingatrandom.Basedontheinverseprobability-weightedidea,aclassofempiricallog-likelihoodratios,aswellastwomaximumempiricallikelihoodestimators,aredevelopedforα(u).Theresultingstatisticsareshowntohavestandardchi-squaredornormaldistributionsasymptotically.Simulationstudiesarealsoconstructedtoillustratethefinitesamplepropertiesoftheproposedstatistics.
简介:ConsidertwolinearmodelsXi=U′iβ+εi,Yj=Vj′γ+ηjwithresponsevariablesmissingatrandom.Inthispaper,weassumethatX,Yaremissingatrandom(MAR)andusetheinverseprobabilityweightedimputationtoproduce‘complete’datasetsforXandY.Basedonthesedatasets,weconstructanempiricallikelihood(EL)statisticforthedifferenceofXandY(denotedas?),andshowthattheELstatistichasthelimitingdistributionofχ21,whichisusedtoconstructaconfidenceintervalfor?.ResultsofasimulationstudyonthefinitesampleperformanceofEL-basedconfidenceintervalson?arereported.