資源描述：
《SAS Markov Chain Monte Carlo (MCMC) Simulation in Practice》由會(huì)員上傳分享，免費(fèi)在線閱讀，更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫(kù)。
1、PaperSP07?SASMarkovChainMonteCarlo(MCMC)SimulationinPracticeScottDPatterson,GlaxoSmithKline,KingofPrussia,PAShi-TaoYeh,GlaxoSmithKline,KingofPrussia,PAABSTRACTMarkovChainMonteCarlo(MCMC)isarandomsamplingmethodwithMonteCarlointegrationusingMarkovchains.MCMChasgainedpopularityinmanyapplicationsduetoth
2、eadvancementofcomputationalalgorithms?andpower.TheSASMIProcedureprovidesMCMCmethodforfillingarbitrarymissingdataandforsimulatingrandomsamplesbasedoncompletedatainformation.ExtensionsofthisprocedurearecurrentlyavailableinexperimentalformtoperformBayesianstatisticalanalysis.Thepurposeofthispaperistous
3、easimulatedhypotheticalclinicaltrialefficacydatasetandChallenger’sO-ringfailuredataasinputinordertoperformtheMCMCmethodformissingdataimputation,modelparametersimulation,andmodeldiagnostics,andtouseSAStoperformaBayesiananalysisofdatacommonlyencounteredinclinicaltrials.????TheSASV9productsusedinthispa
4、perareSASBASE,SAS/STAT,andSAS/GRAPHonaPCWindowsplatform.INTRODUCTIONMonteCarlomethodsaresamplingtechniquesthatdrawpseudo-randomsamplesfromspecifiedprobabilitydistributions.Inotherwords,MonteCarlomethodsarenumericalmethodsthatutilizesequencenumbersofrandomnumberstoperformstatisticalsimulations.AMonte
5、Carloalgorithminvolvesthefollowingcomponents:1)probabilitydistributionfunctions(pdf’s)–thetargetdistributionmustbespecifiedbyasetofpdf’s,2)randomnumbergenerator–asourceofrandomnumbersuniformlydistributedontheunitinterval,3)samplingrule–aprescriptionforsamplingfromthespecifiedpdf’s,4)scoring–theoutco
6、mesmustbesummarizedintooverallscores,5)errorestimation–anestimateofthestatisticalerror(variance)asafunctionofthenumberoftrials,6)variancereductiontechniques–methodsforreducingthevarianceintheestimatedsolutiontoreducethecomputationaltime,7)parallelizationandvectorization–analgorithmtoallowMonteCarlom
7、ethodstobeimplementedefficientlyoncomputercomputation.Forindependentsamples,thesimulationoutcomescanapply‘LawofLargeNumbers’.ButindependentsamplingfromMonteCarlomethodsmaybedifficult.Theissueofindepen
