




版權(quán)說明:本文檔由用戶提供并上傳,收益歸屬內(nèi)容提供方,若內(nèi)容存在侵權(quán),請進(jìn)行舉報(bào)或認(rèn)領(lǐng)
文檔簡介
Chapter9
Multileveleventhistorymodels
9.1Eventhistorymodels
Thisclassofmodels,alsoknownassurvivaltimemodelsoreventdurationmodels,haveastheresponsevariablethelengthoftimebetween'events'.Sucheventsmaybe,forexample,birthanddeath,orthebeginningandendofaperiodofemploymentwithcorrespondingtimesbeinglengthoflifeordurationofemployment.Thereisaconsiderabletheoreticalandappliedliterature,especiallyinthefieldofbiostatisticsandausefulsummaryisgivenbyClayton(1988).Weconsidertwobasicapproachestothemodellingofdurationdata.Thefirstisbasedupon'proportionalhazard'models.Thesecondisbasedupondirectmodellingofthelogduration,oftenknownas'acceleratedlifemodels'.Inbothcaseswemaywishtoincludeexplanatoryvariables.
Themultilevelstructureofsuchmodelsarisesintwogeneralways.Thefirstiswherewehaverepeateddurationswithinindividuals,analogoustoourrepeatedmeasuresmodelsofchapter5.Thus,individualsmayhaverepeatedspellsofvariouskindsofemploymentofwhichunemploymentisone.Inthiscasewehavea2-levelmodelwithindividualsatlevel2,oftenreferredtoasarenewalprocess.Wecanincludeexplanatorydummyvariablestodistinguishthesedifferentkindsofemploymentorstates.Thesecondkindofmodeliswherewehaveasingledurationforeachindividual,buttheindividualsaregroupedintolevel2units.Inthecaseofemploymentdurationthelevel2unitswouldbefirmsoremployers.Ifwehadrepeatedmeasuresonindividualswithinfirmsthenthiswouldgiverisetoa3-levelstructure.
9.2Censoring
Acharacteristicofdurationdataisthatforsomeobservationswemaynotknowtheexactdurationbutonlythatitoccurredwithinacertaininterval,knownasintervalcensoreddata,waslessthanaknownvalue,leftcensoreddata,orgreaterthanaknownvalue,rightcensoreddata.Forexample,ifweknowatthetimeofastudy,thatsomeoneenteredherpresentemploymentbeforeacertaindatethentheinformationavailableisonlythatthedurationislongerthanaknownvalue.Suchdataareknownasrightcensored.Inanothercasewemayknowthatsomeoneenteredandthenleftemploymentbetweentwomeasurementoccasions,inwhichcaseweknowonlythatthedurationliesinaknowninterval.ThemodelsdescribedinthischapterhaveproceduresfordealingwithcensoringInthecaseoftheparametricmodels,wheretherearerelativelylargeproportionsofcensoreddatatheassumedformofthedistributionofdurationlengthsisimportant,whereasinthepartiallyparametricmodelsthedistributionalformisignored.Itisassumedthatthecensoringmechanismisnoninformative,thatisindependentofthedurationlengths.
Insomecases,wemayhavedatawhicharecensoredbutwherewehavenodurationinformationatall.Forexample,ifwearestudyingthedurationoffirstmarriageandweendthestudywhenindividualsreachtheageof30,allthosemarryingforthefirsttimeafterthisagewillbeexcluded.Toavoidbiaswemustthereforeensurethatageofmarriageisanexplanatoryvariableinthemodelandreportresultsconditionalonageofmarriage.
Thereisavarietyofmodelsfordurationtimes.Inthischapterweshowhowsomeofthemorefrequentlyusedmodelscanbeextendedtohandlemultileveldatastructures.Weconsiderfirsthazardbasedmodels.
/筱
9.3Hazardbasedmodelsincontinuoustime
Theunderlyingnotionsarethoseofsurvivorandhazardfunctions.Considerthe(singlelevel)casewherewehavemeasuresoflengthofemploymentonworkersinafirm.Wedefinetheproportionoftheworkforceemployedforperiodsgreaterthantasthesurvivorfunctionanddenoteitby
whereisthedensityfunctionoflengthofemployment.Thehazardfunctionisdefinedas
andrepresentstheinstantaneousrisk,ineffectthe(conditional)probabilityofsomeonewhoisemployedattimet,endingemploymentinthenext(small)unitintervaloftime.
Thesimplestmodelisonewhichspecifiesanexponentialdistributionforthedurationtime,
whichgives,sothatthehazardrateisconstantand=.In
general,however,thehazardratewillchangeovertimeandanumberofalternativeformshavebeenstudied(seeforexample,CoxandOakes,1984).AcommononeisbasedontheassumptionofaWeibulldistribution,namely
ortheassociatedextremevaluedistributionformedbyreplacingtbyzt=E.Anotherapproachtoincorporatingtime-varyinghazardsistodividethetimescaleintoanumberofdiscreteintervalswithinwhichthehazardrateisassumedconstant,thatisweassumeapiecewiseexponentialdistribution.Thismaybeusefulwherethereare'natural'unitsoftime,forexamplebasedonmenstrualcyclesintheanalysisoffertility,andthiscanbeextendedbyclassifyingunitsbyotherfactorswheretimevariesovercategories.Wediscusssuchdiscretetimemodelsinalatersection
Themostwidelyusedmodels,towhichweshalldevoteourdiscussion,arethoseknownas
proportionalhazardsmodels,andthemostcommondefinitionisThetermn
denotesalinearfunctionofexplanatoryvariableswhichweshallmodelexplicitlyinsection9.5.It
isassumedthatthebaselinehazardfunction,dependsonlyontimeandthatallothervariation
betweenunitsisincorporatedintothelinearpredictorn.Thecomponentsofnmayalsodepend
upontime,andinthemultilevelcasesomeofthecoefficientswillalsoberandomvariables.
9.4Parametricproportionalhazardmodels
Forthecasewherewehaveknowndurationtimesandrightcensoreddata,definethecumulative
baselinehazardfunctiontandavariablevwithmean,takingthevalue
oneforuncensoredandzeroforcensoreddata.Itcanbeshown(McCullaghandNelder,1987)thatthemaximumlikelihoodestimatesrequiredarethoseobtainedfromamaximumlikelihoodanalysisforthismodelwherewistreatedasaPoissonvariable.ThiscomputationaldeviceleadstotheloglinearPoissonmodelforthei-thobservation
(9.1)
wherethetermgistreatedasanoffset,thatis,aknownfunctionofthelinearpredictor.
Thesimplestcaseistheexponentialdistribution,forwhichwehave.Equation(9.1)
thereforehasanoffsetⅡandthetermIQisincorporatedinton.WecanmodeltheresponsePoissoncountusingtheproceduresofchapter6,withcoefficientsinthelinearpredictorchosentoberandomatlevels2orabove.Thisapproachcanbeusedwithotherdistributions.FortheWeibulldistribution,ofwhichtheexponentialisaspecialcase,theproportionalhazardsmodelisequivalent
tothelogdurationmodelwithanextremevaluedistributionandweshalldiscussitsestimationina
latersection.
/筱
9.5ThesemiparametricCoxmodel
Themostcommonlyusedproportionalhazardmodelsareknownassemiparametricproportionalhazardmodelsandwenowlookatthemultilevelversionofthemostcommonoftheseinmore
detail.
Considerthe2-levelproportionalhazardmodelforthejk-thlevellunit
(9.2)
whereistherowvectorofexplanatoryvariablesforthelevel1unitandsomeorallofthe厭
arerandomatlevel2.Weadoptthesubscriptsj,kforlevelsoneandtwoforreasonswhichwillbeapparentbelow.
Wesupposethatthetimesatwhichalevellunitcomestotheendofitsdurationperiodor'fails'areorderedandateachoftheseweconsiderthetotal'riskset'.Atfailuretimet.therisksetconsistsofallthelevellunitswhichhavebeencensoredorforwhichafailurehasnotoccurredimmediatelypreceedingtimeti.Thentheratioofthehazardfortheunitwhichexperiencesafailureandthesumofthehazardsoftheremainingrisksetunitsis
whichissimplytheprobabilitythatthefailedunitistheonedenotedby(Cox,1972).Itis
assumedthat,conditionalonthe,theseprobabilitiesareindependent.
Severalproceduresareavailableforestimatingtheparametersofthismodel(seeforexampleClayton,1991,1992).Forourpurposesitisconvenienttoadoptthefollowing,whichinvolvesfittingaPoissonorequivalentmultinomialmodelofthekinddiscussedinchapter7.
Ateachfailuretimelwedefinearesponsevariateforeachmemberoftheriskset
whereiindexesthemembersoftheriskset,andj,klevel1andlevel2units.Ifwethinkofthebasic2-levelmodelasoneofemployeeswithinfirmsthenwenowhavea3-levelmodelwhereeachlevel2unitisaparticularemployeeandcontainingnalevellunitswherenzisthenumberofrisksetstowhichtheemployeebelongs.Level3isthefirm.Theexplanatoryvariablescanbedefinedatanylevel.Inparticulartheycanvaryacrossfailuretimes,allowingsocalledtime-varyingcovariates.Overallproportionality,conditionalontherandomeffects,canbeobtainedbyorderingthefailuretimesacrossthewholesample.Inthiscasethemarginalrelationshipbetweenthehazardandthecovariatesgenerallyisnotproportional.Alternatively,wecanconsiderthefailuretimesorderedonlywithinfirms,sothatthemodelyieldsproportionalhazardswithinfirms.Inthiscasewecanstructurethedataasconsistingoffirmsatlevel3,failuretimesatlevel2andemployeeswithinrisksetsatlevel1.Inbothcases,becausewemaketheassumptionofindependenceacrossfailuretimeswithinfirms,thePoissonvariationisatlevel1andthereisnovariationatlevel2.Inotherwordswecancollapsethemodeltotwolevels,withinfirmsandbetweenfirms.
AsimplevariancecomponentsmodelfortheexpectedPoissoncountiswrittenas
(9.3)
wherethereisa'blockingfactor'cforeachfailuretime.Infactwedonotneedgenerallytofitallthesenuisanceparameters:insteadwecanobtainefficientestimatesofthemodelparametersbymodellingarasasmoothfunctionofthetimepoints,using,say,aloworderpolynomialorasplinefunction(Efron,1988).
Forthemodelwhichassumesoverallproportionalityanestimatorofthebaselinesurvivingfractionforanindividualinthek-thfirmattimeh,where,is
andtheestimateforanindividualwithspecificcovariatevaluesX;xis
/筱
(9.4)
Forthemodelwhichassumesproportionalitywithinfirmsthesetwoexpressionsbecomerespectively
Wherewefitpolynomialstotheblockingfactors,thea,areestimatedfromthepolynomial
coefficients,andthesurvivingfractioncanbeplottedagainstthetimeassociatedwitheachinterval.
9.6Tiedobservations
Wehaveassumedsofarthateachfailuretimeisassociatedwithasinglefailure.Inpracticemanyfailureswilloftenoccuratthesametime,withintheaccuracyofmeasurement.Sometimes,datamayalsobedeliberatelygroupedintime.InthiscaseallthefailuresattimesIhavearesponseequalto1.ThisprocedureforhandlingtiesisequivalenttothatdescribedbyPeto(1972)(seealsoMcCullaghandNelder,1989).
9.7Repeatedmeasuresproportionalhazardmodels
Asinthecaseofordinaryrepeatedmeasuresmodelsdescribedinchapter6wecanconsiderthecaseofmultipleepisodesordurationswithinindividualswithbetweenandwithinindividualvariationandpossiblyfurtherlevelswhereindividualsmaybenestedwithinfirms,etc.Themodelsofprevioussectionscanbeappliedtosuchdata,buttherearefurtherconsiderationswhicharise.Whereeachindividualhasthesamefixednumbernofepisodes.Wecantreatthese,asinchapter5,asconstitutingnvariatessothatwehaveann-variatemodelwithan(nxn)covariancematrixbetweenindividuals.Thevariatesmaybeeitherreallydistinctmeasurementsorsimplythedifferentepisodesinafixedordering.ThisisthemodelconsideredbyWeietal(1989)whodefineproportionalityaswithinindividuals.Wecanalsomodelamultivariatestructurewhere,withinindividuals,therearerepeatedepisodesforanumberofdifferenttypesofinterval.Foreachtypeofintervalwemayhavecoefficientsrandomattheindividuallevelandthesecoefficientswillgenerallyalsocovaryatthatlevel.
Oftenwithrepeatedmeasuresmodelsthefirstepisodeisdifferentinnaturefromsubsequentones.Anexamplemightbethefirstepisodeofadiseasewhichmaytendtobelongerorshorterthansubsequentepisodes.Ifthefirstepisodeistreatedasifitwereaseparatevariatethenthesubsequentepisodescanberegardedashavingthesamedistribution,asintheprevioussection.
Anotherpossiblecomplicationinrepeatedmeasuresdata,asinchapter5isthatwemaynotbeabletoassumeindependencebetweendurationswithinindividuals.Thiswillthenleadtoserialcorrelationmodelswhichcanbeestimatedusingtheproceduresdiscussedinchapter6fortheparametriclogdurationmodelsdiscussedbelow.
9.8Exampleusingbirthintervaldata
Thedataareaseriesofrepeatedbirthintervalsfor379HutteritewomenlivinginNorthAmerica(LarsenandVaupel,1993;Egger,1992).Theresponseisthelengthoftimeinmonthsfrombirthtoconception,rangingfromlto160,withthefirstbirthintervalignoredandnocensoredinformation.Thisgives2235birthsinall.
Thereisinformationavailableonthemother'sbirthyear,herageinyearsatthestartofthebirthinterval,whetherthepreviouschildwasaliveordead,andthedurationofmarriageatthestartofthebirthinterval.Sincewehavealargenumberofwomeneachwitharelativelysmallnumberofintervalswehaveassumedoverallproportionality,withfailuretimesorderedacrossthewholesample.Table1givestheresultsforavariancecomponentsanalysisandonewhereseveralrandomcoefficientsareestimated.Afourthorderpolynomialwasadequatetosmooththeblockingfactors.
/筱
Table9.1ProportionalhazardsmodelforHutteritebirthintervals.Inthe
randompartsubscript0referstointercept,1topreviousdeath.
Parameter
Fixed
Intercept
Mother'sbirthyear-1900
Mother'sage(year-20)
Previousdeath
Marriageduration(Months)
Estimate(s.e.)
A
-3.65
0.026(0.003)
-0.008(0.014)
0.520(0.118)
-0.003(0.001)
Estimate(s.e.)
B
-3.64
0.026(0.003)
-0.004(0.014)
0.645(0.144)
-0.004(0.001)
Random
c
。
〇aol
0.188(0.028)
0.188(0.028)
0.005(0.088)
0.381(0.236)
Theonlycoefficientestimatedwithanon-zerovarianceatlevel2waswhetherornotthepreviousbirthdied,butalargesamplechisquaredtestforthetworandomparametersforthiscoefficientgivesaP-valueof0.01on2degreesoffreedom.Anincreaseonthelinearscaleisassociatedwithashorterinterval.Thusthebirthintervaldecreasesforthelaterbormothersandalsoifthepreviousbirthisadeath.Theintervalissomewhatshorterthelongerthemarriagedurationwithlittleadditionaleffectofmaternalage.Thisapparentlackofasubstantialageeffectseemstobeaconsequenceofthehighcorrelation(0.93)betweendurationofmarriageandage.Higherordertermsfordurationandagewerefittedbuttheestimatedcoefficientsweresmallandnotsignificantatthe10%level.Thebetween-individualstandarddeviationisabout0.4whichiscomparableinsizetotheeffectofapreviousdeath.Thebetween-individualstandarddeviationforamodelwhichfitsnocovariatesis0.45sothatthecovariatesexplainonlyasmallproportionofthebetween-individualvariation.Figure9.1showstwoaverageestimatedsurvivingfractioncurvesforawomanaged20,bornin1900withmarriageduration12months.Thehigheroneisforthosewheretherewasapreviouslivebirthandthelowerwheretherewasapreviousdeath.
Figure9.1Probabilityofexceedingeachbirthintervallength;livebirthupper,previousdeathlower.
9.9Thediscretetime(piecewise)proportionalhazardsmodel
Wheretimeisgroupedintopreassignedcategorieswewritethesurvivorfunctionattimeintervall,theprobabilitythatfailureoccursafterthisinterval,ass.Thisgives
Thisgives
whichcanbeusedtoestimatethesurvivorfunctionfromasetofestimatedhazards.
Fortheproportionalhazardsmodel(9.2)anda2-levelmodeltheexpectedhazardisgiven(Aitkinet
al,1989)by
where,asbefore,theaareconstantstobeestimated,oneforeachtimeinterval.Thisleadstoamodelwheretheresponseisabinomialvariate,beingthenumberofdeathsdividedbythenumberintherisksetatthestartoftheinterval(seealsoEgger,1992).Anycensoredobservationsinanintervalareexcludedfromtheriskset.Theestimationfollowsthatforthelogitbinomialmodel
/筱
describedinchapter7,exceptthatwenowrequirethefirstandseconddifferentialsoftheloglogfunction,namely
AsintheCoxmodel,wecanfitapolynomialfunctiontothesuccessivetimeintervals,ratherthanthefullsetofblockingfactors.Thedatawillbeorderedwithinlevel2unitssothatarisksetingeneralwillextendoverseveralsuchunits.Ageneralprocedureistospecifytheresponseforeachlevel1unitasbinary,thatiszeroiftheunitsurvivestheintervalandoneifnot,withtheappropriatecinthefixedpart.Thusa2-levelmodelwillbecomespecifiedasa3-levelmodelwiththebinomialvariationatlevel1andtheactuallevellunitsatlevel2.Themodelcanbefurtherextendedtopolytomousoutcomes,or'competingrisks',whereseveraldifferentkindsoffailurecanoccur.Theanalysisfollowsthesamepattern,butwiththeresponsebeingamultinomial
variateandthecorrespondingmodelsofChapter7canbeappliedwithadifferentlinearpredictorforeachoutcomecategory.
9.10Logdurationmodels
Fortheacceleratedlifemodelthedistributionfunctionfordurationiscommonlyassumedtobeof
theform
where天isabaselinefunction(CoxandOakes,1984).Fora2-levelmodelthiscanbewrittenas
(9.5)
whichisinthestandardformfora2-levelmodel.WeshallassumeNormalityfortherandomcoefficientsatlevel2(andhigherlevels)butatlevel1weshallstudyotherdistributionalformsforthee.Thelevel1distributionalformisimportantwheretherearecensoredobservations.WefirstconsiderthecommonchoiceofanextremevaluedistributionforthelogdurationL,conditionalon ,whichaswenotedabove,impliesanequivalencewiththeproportionalhazardsmodel.Omittinglevelsubscriptswewrite
For(9.5)thisgives
(9.6)
(9.7)
Wherethedifferentialisforuseintheestimationofcensoreddataandiswithrespecttoβinthe
expressionbelow.
ThemeanofLisincorporatedintothefixedpredictor.Ifwehavenocensoreddataweestimatetheparametersforthemodelgivenby(9.5)bytreatingitasastandardmultilevelmodel.Wenotethattheestimationisstrictlyquasilikelihoodsinceweareusingonlythemeanandvariancepropertiesofthelevel1distribution.Ifweassumeasimplelevel1variancethenwecaniterativelyestimatecfromtheaboverelationshipandwealsoobtainforthe2-levelmodel(9.5)
Wherethereiscomplexvariationatlevel1thencwillvarywiththelevellunits.Toestimatethesurvivalfunctionforagivenlevel2unitwefirstconditiononthecovariatesandrandomcoefficients,thatisX,B,andthenuse(9.7).
Wecanchooseotherdistributionalformsforthelogdurationdistribution.Theseincludetheloggammadistribution,theNormalandthelogistic.Thus,forexample,fortheNormaldistributionwehave
/筱
whereφarethecumulativeanddensityfunctionsofthestandardNormaldistribution.
Quasilikelihoodestimatescanbeobtainedforanysuitabledistributionwithtwoparameters.Thepossibilityoffittingcomplexvariationatlevel1canbeexpectedtoprovidesufficientflexibilityusingthesedistributionsformostpurposes.ingthesedistributionsformostpurposes.
Table9.2.LogdurationofbirthintervalforHutteritewomen.Subscript1referstobirthyear,2
toageand3topreviousdeath.
Parameter
Fixed
Intercept
Mother's
Mother'sPrevious
Marriage
Random
Level2
birthyear-1900
age-20
death
duration(Months)
Cmo
o
〇m?
Level1
。
3
-2loglikelihood
Estimate(s.e.)Estimate(s.e.)Estimate(s.e.)
ABC
1.971.961.97
-0.021(0.002)-0.005(0.010)-0.435(0.079)
0.003(0.001)
-0.021(0.002)-0.005(0.010)
0.436(0.079)
0.003(0.001)
-0.021(0.002)-0.005(0.010)-0.438(0.089)
0.003(0.001)
0.127(0.017)
0.121(0.054)
-0.001(0.002)0.0001(0.0001)
-0.005(0.003)
0.0001(0.0001)
0.0006(0.0003)
0.114(0.052)
-0.001(0.002)
0.0001(0.0001)
-0.004(0.003)
0.0001(0.0001)
0.0005(0.0003)
0.549(0.018)0.533(0.018)0.522(0.018)
0.200(0.108)
5305.95295.55290.8
9.11Censoreddata
Wheredataarecensoredinlogdurationmodelswerequirethecorrespondingprobabilities.Thus,forrightcensoreddatawewoulduse(9.7)withcorrespondingformulaeforintervalorleftcensoreddata.Foreachcensoredobservationwethereforehaveanassociatedprobability,sayg,withtheresponsevariablevalueofone.
Thisleadstoabivariatemodel,inwhichforeachlevellunittheresponseiseitherthecontinuouslogdurationtimeortakesthevalueoneifcensoredwithcorrespondingexplanatoryvariablesineachcase.Therearebasicallytwoexplanatoryvariablesforthelevel1variation,oneforthecontinuouslogdurationresponseandoneforthebinomialresponse.Intheformercasewecanextendthisforcomplexlevel1variation,asintheexampleanalysisbelow.Forthelatterweusethestandardlogitmodelasdescribedinchapter7,possiblyallowingforextra-binomialvariation.Therandomparametersatlevel1forthetwocomponentsareuncorrelated.Whencarryingoutthe
computations,wemayobtainstartingvaluesfortheparametersusingjusttheuncensored
observations.
/筱
Sincethesamelinearfunctionoftheexplanatoryvariablesentersintoboththelinearandnonlinearpartsofthismodel,werequireonlyasinglesetoffixedpartexplanatoryvariables,althoughthesewillrequiretheappropriatetransformationforthelogitresponseasdescribedinchapter7.Wealsonotethatanykindsofcensoreddatacanbemodelled,solongasthecorrespondingprobabilitiesarecorrectlyspecified.
Figure9.2.Level1residualsbyNormalscoresforAnalysisBinTable9.2
Wecanreadilyextendthismodeltothemultivariatecasewhereseveralkindsofdurationsaremeasured.Thiswillrequireoneextralowestleveltobeinsertedtodescribethemultivariatestructure,withlevel2becomingthebetween-observationlevelandlevel3theoriginallevel2.Forthelogitpartofthemodelwewillallowcorrelationsatlevel2wherethesecanbeinterpretedaspoint-biserialcorrelations.
Forrepeatedmeasuresmodelswheretherearedifferenttypesofdurationwecanchoosetofitamultivariatemodel.Alternatively,asdiscussedinchapter4,wemaybeabletospecifyasimplermodelwherethetypesdifferonlyintermsofafixedpartcontribution,orperhapswheretherearedifferentvariancesforeachtypewithacommoncovariance.Aspointedoutearlier,wemay
sometimeswishtotreatthefirstdurationlengthseparatelyandthisisreadilydonebyspecifyingit
asaseparateresponse.
9.12Infinitedurations
Itissometimesfoundthatforaproportionofindividuals,theirdurationlengthsareextremelylong.Thus,someemployeesremaininthesamejobforlifeandsomepatientsmayacquireadiseaseandretainitfortherestoftheirlives.Instudiesofsocialmobility,someindividualswillremaininaparticularsocialgroupforafinitelengthoftimewhileothersmayneverleaveit:suchmodelsaresometimesreferredtoasmover-stayermodels.Wecantreatsuchdurationsasiftheywereinfinite.Sinceanygivenstudywilllastonlyforafinitetime,itisimpossibletodistinguishinfinitetimesfromthosewhicharerightcensored.Nevertheless,ifwemakesuitabledistributionalassumptionswecanobtainanestimateoftheproportionofinfinitesurvivaltimes.
Foraconstantθ,givenanunobserveddurationtime,theobservationiseitherrightcensoredwithfinitedurationorhasinfinitedurationsothatwereplacetheprobability,by
Ingeneralθwilldependonexplanatoryvariablesandanobviouschoiceforamodelis
(9.8)
Thecoefficientsin(9.8)mayalsovaryacrosslevel2units.
Wheretheobservationisnotcensoredweknowthatithasafinitedurationsothatfortheinfinite
durationparameterswehavearesponsevariabletakingthevaluezerowithpredictorgivenby 三.Thefullmodelcanthereforebespecifiedasabivariatemodelwhereforobserveddurationswehavetworesponses,onefortheuncensoredcomponentl,andtheonefortheparameters9.Forthecensoredobservationsthereisasingleresponsewhichtakesthevalueone
withpredictorfunction
Wecanextendtheproceduresofchapter7tothejointestimationofB9,notingthatforthe
censoredobservationswhenestimatingβ,wehave
andforestimatingwehave
/筱
9.13Exampleswithbirthintervaldataandchildren'splayepisodes
WefirstlookagainattheHutteritebirthintervaldata.Sinceallthedurationsareuncensoredweapplyastandardmodeltothelog(birthinterval)values.ResultsaregiveninTable9.2.
Weseethatwecannowfittheyearofbirthandageasrandomcoefficientsatlevel2.Ajointtestgivesachi-squaredvalueof10.4with5d.f.P=0.065,andtheyareeachseparatelysignificantwithasignificancelevelof6%.Wehavesignificantheterogeneityatlevel1wherethevariancewithinwomenisgreaterwheretherehasbeen
溫馨提示
- 1. 本站所有資源如無特殊說明,都需要本地電腦安裝OFFICE2007和PDF閱讀器。圖紙軟件為CAD,CAXA,PROE,UG,SolidWorks等.壓縮文件請下載最新的WinRAR軟件解壓。
- 2. 本站的文檔不包含任何第三方提供的附件圖紙等,如果需要附件,請聯(lián)系上傳者。文件的所有權(quán)益歸上傳用戶所有。
- 3. 本站RAR壓縮包中若帶圖紙,網(wǎng)頁內(nèi)容里面會(huì)有圖紙預(yù)覽,若沒有圖紙預(yù)覽就沒有圖紙。
- 4. 未經(jīng)權(quán)益所有人同意不得將文件中的內(nèi)容挪作商業(yè)或盈利用途。
- 5. 人人文庫網(wǎng)僅提供信息存儲(chǔ)空間,僅對用戶上傳內(nèi)容的表現(xiàn)方式做保護(hù)處理,對用戶上傳分享的文檔內(nèi)容本身不做任何修改或編輯,并不能對任何下載內(nèi)容負(fù)責(zé)。
- 6. 下載文件中如有侵權(quán)或不適當(dāng)內(nèi)容,請與我們聯(lián)系,我們立即糾正。
- 7. 本站不保證下載資源的準(zhǔn)確性、安全性和完整性, 同時(shí)也不承擔(dān)用戶因使用這些下載資源對自己和他人造成任何形式的傷害或損失。
最新文檔
- 牛津譯林版七年級英語上冊教學(xué)計(jì)劃(含進(jìn)度表)
- 2025年黨章黨史國史國情知識競賽題庫及答案(共220題)
- 新型家庭醫(yī)生簽約服務(wù)對促進(jìn)轄區(qū)孕產(chǎn)婦管理的效果分析
- 《單片機(jī)技術(shù)應(yīng)用》 課件
- 節(jié)能環(huán)保居間服務(wù)合同范例
- 道路交通規(guī)劃方案介紹
- 低空經(jīng)濟(jì)行業(yè)報(bào)告
- 醫(yī)院裝修大包合同參考范本
- 投資可行性分析報(bào)告包括哪些內(nèi)容
- 低空經(jīng)濟(jì)涉及的行業(yè)
- 2025年防范電信網(wǎng)絡(luò)詐騙知識競賽題庫及答案
- 第12課 遼宋夏金元時(shí)期經(jīng)濟(jì)的繁榮【公開課一等獎(jiǎng)創(chuàng)新教學(xué)設(shè)計(jì)】-【教學(xué)評一體化】大單元整體教學(xué)
- 農(nóng)村荒山地轉(zhuǎn)讓合同6篇
- 《無人機(jī)操控基礎(chǔ)》課件
- 塔設(shè)備技術(shù)問答-化工設(shè)備
- 2025年熔化焊接與熱切割試題(附答案)
- 水池防滲漏施工方案
- 第八單元+中華民族的抗日戰(zhàn)爭和人民解放戰(zhàn)爭+作業(yè)設(shè)計(jì)方案 高一統(tǒng)編版2019必修中外歷史綱要上冊
- 2025年湖北省新華書店(集團(tuán))限公司招聘(93人)高頻重點(diǎn)提升(共500題)附帶答案詳解
- 《鐵路技術(shù)管理規(guī)程》(普速鐵路部分)
- 消防煙感報(bào)警設(shè)備 投標(biāo)方案(技術(shù)標(biāo) )
評論
0/150
提交評論