Treść książki
Przejdź do opcji czytnikaPrzejdź do nawigacjiPrzejdź do informacjiPrzejdź do stopki
Preface
MaunoKetoandErkkiPahkinenconcentrateonaverypressingproblem
ofsampleallocationaimedatmaximisingtheefficiencyofindirectestimates
forsmallareaparameters.Theyproposeanoriginalprocedurecalled
‘experimentalallocation’forthecasewherestratifiedsamplingisusedand
stratacoincidewithareaswhiletheempiricalbestlinearunbiasedpredictor
(EBLUP)isused.Theexperimentalallocationisformulatedonthebasisofan
empiricaldistributionofareasamplesizesassessedinasimulationstudy.
Anothersimulationstudyisthencarriedouttoshowsuperiorityofthe
proposed
allocation
scheme
over
other
allocation
methods
including
proportionalandequalallocation.
PaolaM.Chiodini,RitaLima,GiancarloManzi,BiancaM.Martelliand
FlavioVerrecchiastudytheproblemofsampleallocationfromadesign-based
pointofview.TheiranalysisisbasedontheexperiencesoftheItalianRegister
ofEnterprises(ASIA-ISTAT).Theaimofthestudyistofindthemostsuitable
allocationmethodforafinitepopulationofenterpriseswithpre-determined
stratification.AninnovativeallocationprocedurecalledAOSUisproposed,
whichaimsatreducingtheunwelcomeeffectofemptystratificationcells,by
combiningprinciplesofNeymanallocationanduniformallocation.Several
allocationmethodsarethencomparedbysimulation,withrespecttovarious
errormeasures.Thelistofexaminedmethodsincludes:proportional
allocation,uniformallocation,Neyman’soptimalallocation,ISAEallocation
andAOSUallocation.Simulationresultsjustifytheuseoftheproposed
allocationscheme.
JanuszWywiałcomparestheaccuracyofvariousestimationstrategies
usingtheavailableauxiliaryinformation.Twosamplingdesignsdependenton
observedauxiliaryvariablesareconsidered.Oneofthemisbasedonthe
differenceoforderstatistics.AnotheronewasproposedbySinghand
Srivastavaandassociateswitheachsampleaselectionprobabilityproportional
tothevarianceofauxiliaryvariable.Severalestimatorsofapopulationmean
areconsidered,includingtheclassicHorvitz-Thompsonestimatoraswellas
otherestimatorsdependentonorderstatisticsproposedinthepaper.Their
propertiesarecomparedthroughasimulationstudywhichusesthewell-known
datasetdescribingSwedishmunicipalities.Simulationswerealsocarriedout
fordatacontaminatedwithoutlierobservationsandtheirresultsarealso
presented.
TomaszŻądło’sworkisdevotedtosituationswheredesign-consistent
estimates
of
domain
totals
are
desired
due
to
the
risk
of
model
misspecification.Amodel-assistedpseudo-empiricalbestlinearunbiased
predictorforlongitudinaldata,basedontheGeneralLinearMixedModelwith
element-specificanddomain-specificrandomcomponentsisproposed.The
8
