ABSTRACT
ThischapterfocusesonBayesianinferenceanddesigninbinaryregressionexperiments.Asacasestudy,weconsiderheartdefibrillatorexperiments,inwhichthenumberofobservationsthatcanbetakenislimited anditisimportanttoincorporateallavailablepriorinformation.Inparticular,bymodelingtheindividual-to-individualvariationintheappropriate defibrillationsetting,wecanuseinformationonpastpatientsinformulatingasensiblepriordistributionfordesigningexperimentsforcurrentpatients.Thefirstpartillustratestheuseofhierarchicalmodelstoobtain suchpriordistributions.Thesecondpartofthechapterconsidersdesign strategies.AnimportantadvantageofaBayesiantechniqueisthatitis conceptuallyeasytoadapttoinformationthataccruessequentially.This isparticularlydesirablewhenearlystoppingoftheexperimentationisof interest.Ingeneral,analyticexpressionsforoptimalsequentialsolutions arenotavailable,andacombinationofapproximationtechniquesand numericalcomputationmustbeused.Herewefocusonfindingoptima withinrestrictedsetsofstrategies.Wecompareanadaptivestrategybased onfixedpercentagechangesintheenergylevelsandvariablesamplesize withastrategyinwhichalllevelsarechosenoptimallybutthesample sizeisfixed.
