基于二次推斷函數(shù)的縱向數(shù)據(jù)半?yún)?shù)模型的估計.pdf

1、華東師范大學碩士學位論文基于二次推斷函數(shù)的縱向數(shù)據(jù)半?yún)?shù)模型的估計姓名：柏楊申請學位級別：碩士專業(yè)：概率論與數(shù)理統(tǒng)計指導教師：朱仲義20050501壘逸壘型一生型墮型麴塑這虹AbsttactLongitudinaldataisreferredtodatainwhichindividualsaremessuredrepeatedlythroughtime，SOitcombineselementsofcrOSssectionaldataan

2、dtimeseriesdataTheprominentadvantageoflongitudinaldataisthatitcallanalyzeeffectivelythechangeofindividualsovertimeHowever，itiscomplicatedbyrepeatedmessurementwithinindividuaisandvariationamongindividualsLiang＆Zeger(19861

3、introducedtheingeniousideaofusingaworkingcorrelationmatrixwithasmallsetofnuisanceparametersThegeneralisedestimatingequation(GEE)estimatorsoftheregressionparametersareconsistentevenwhenthetruecorrelationmatrixiSnotaneleme

4、ntoftheclassofworkingcorrelationmatrices，andareefficientwhentheworkingcorrelationiscorrectlyspecifiedWhentheworkingcorrelationism_isspecified，however，theGEEnolongerresultsintheoptimalestimationQueta1(2000)introducedadiff

5、erentstrategyforestimatingtheworkingcorrelationSOthattheestimatoralwaysexisted，and，evenifthecorrelationWasmisspecified，theregressionestimatorremainedoptimalwithintheassumedfamilyTheirideaistepresenttheinverseoftheworking

6、correlationmatrixbythelinearcombinationofbasismatricesThenusethequadraticinferencefunctions(QIF)toconstracttheestimatingequationThepurposeofthispaperistoapplythequadraticinferencefunctionstothesemi—parametricmodelforlong

7、itudinaldataandimprovetheestimatoroftheparametricpartinthemodelWesucceedinprovingtheconsistentandasymptoticnormalityoftheQIFestimatorLast，wecomparetheQIFandGEEmethodsbysimulationsandapplytheQIFestimatortotheCD4data，Keywo