A Bayesian model for longitudinal circular data

-

Documents
20 pages
Obtenez un accès à la bibliothèque pour le consulter en ligne
En savoir plus

Description


The analysis of short longitudinal series of circular data may be problematic and to some extent has not been completely developed. In this paper we present a Bayesian analysis of a model for such data. The model is based on a radial projection onto the circle of a particular bivariate normal distribution. Inferences about the parameters of the model are based on samples from the corresponding joint posterior density which are obtained using a Metropolis-within-Gibbs scheme after the introduction of suitable latent variables. The procedure is illustrated both using a simulated data set and a realdata set previously analyzed in the literature.

Sujets

Informations

Publié par
Publié le 01 septembre 2011
Nombre de visites sur la page 35
Langue English
Signaler un problème
 Working Paper 11-27 Statistics and Econometrics Series 020 September 2011   Departamento de Estadística  Universidad Carlos III de Madrid Calle Madrid, 126 28903 Getafe (Spain) Fax (34) 91 624-98-49 A BAYESIAN MODEL FOR LONGITUDINAL CIRCULAR DATA  Gabriel Núñez-Antonio* and Eduardo Gutiérrez-Peña**   Abstract  The analysis of short longitudinal series of circular data may be problematic and to some extent has not been completely developed. In this paper we present a Bayesian analysis of a model for such data. The model is based on a radial projection onto the circle of a particular bivariate normal distribution. Inferences about the parameters of the model are based on samples from the corresponding joint posterior density which are obtained using a Metropolis-within-Gibbs scheme after the introduction of suitable latent variables. The procedure is illustrated both using a simulated data set and a real-data set previously analyzed in the literature.   Keywords: Circular data; longitudinal data; Gibbs sampler; latent variables; mixed-effects linear models; projected normal distribution.   * Department of Statistics, Universidad Carlos III de Madrid, Spain. E-mail address: gabriel.nunez@uc3m.es. **Department of Probability and Statistics, IIMAS-UNAM, Mexico. E-mail address: eduardo@sigma.iimas.unam.mx.    
1IntroductionSeveralapproacheshavebeenproposedforanalyzinglongitudinaldata.Forareviewthereaderisreferred,forexample,toDiggleetal.(2002),Fitzmauriceetal.(2004),HedekerandGibbons(2006),andGelmanandHill(2007).Thesebooksalldiscusslongitudinalmodelsfor‘scalar’(i.e.linear)responsesasopposedtocircular.Incontrast,methodologicalproposalstodescriberelation-shipswithinrepeatedmeasurementsofcirculardataareratherlimited.Thismaybeduetothedifficultiesinworkingwithprobabilitydistributionscommonlyassociatedwithdirectionaldataandtotheintrinsicdependencyinherenttolongitudinalstructures.Circulardataareaparticularcaseofdirectionaldata.Specifically,circulardatarepresentdi-rectionsintwodimensions.ForasurveythereaderisreferredtoFisheretal.(1987),Fisher(1993),MardiaandJupp(2000),andJammalamadakaandSenGupta(2001).SeealsoArnoldandSenGupta(2006)foranoverviewoftheapplicationsofcirculardataanalysisinecologicalandenvironmentalsciences.Fromatheoreticalpointofview,therearethreebasicapproachestodirectionalstatistics,whichmaybecalledtheembedding,wrappingandintrinsicapproaches;see,MardiaandJupp(2000).Consequently,thereareseveralwaysofgeneratingprobabilitydistributionsforcirculardata.Onerelativalystraightforwardwayistoradiallyprojectontheunitcircleprobabilitydistributionsoriginallydefinedontheplane.Inthegeneralcase,letYbeaq-dimensionalrandomvectorsuchthatPr(Y=0)=0.ThenU=||Y||1Yisarandompointontheq-dimensionalunitsphere.Itsmeandirectionistheunitvectorη=E(U),whereρ=||E(U)||,0ρ1;hereE()representstheusualexpectationforrandomvectors,and||∙||representstheusualEuclideannorm.Theparameterρiscalledthemeanresultantlengthandrepresentsameasureofconcentrationfordirectionaldistributions(see,forexample,MardiaandJupp,2000,andPresnelletal.,1998).AnimportantinstanceisthatinwhichYhasaq-variateNormaldistribution,Nq(∙|µ,Λ),withmeanvectorµ=E(Y)andprecisionmatrixΛ=Var(Y)1.InthiscaseUissaidtohaveaq-dimensionalprojectednormaldistribution,heredenotedbyPN(∙|µ,Λ).Inthecircularcase,2