PYXIDA Institutional Repository
and Digital Library
Collections :

Title :Bayesian latent variable models for ordinal data
Contributor :Ntzoufras, Ioannis (Επιβλέπων καθηγητής)
Athens University of Economics and Business, Department of Statistics (Degree granting institution)
Type :Text
Extent :160 p.
Language :en
Abstract :Generally, Latent variables models are used in cases that the under studyvariables cannot be measured directly. Such variables are for example, consumersatisfaction of a new product. Here we analyze a market research dataset with discreteordinal manifest variables and continuous latent features. Thus the methodology oflatent trait models has been applied. In order to express our personal beliefs about theunder study problem, the Bayesian approach has been implemented. Through theBayesian paradigm we assume a prior distribution to express our information aboutthe under estimation parameters. The combination of the prior distribution with thelikelihood, results in the posterior distribution. Estimation of the posterior distributioncan be achieved through Markov Chain Monte Carlo (MCMC) algorithms. For ourdataset the Gibbs sampler was implemented via the WinBungs software.Three different link functions were used, the logit, the probit and the c-loglog.Furthermore, one and two factor latent trait models were fitted. The final choice of themodel was achieved through the Deviance Information Criterion (DIC).The aforementioned methodology was applied to detect a possible linkbetween excessive consumption behaviors with schizotypy. The impulsive and thecompulsive buying behaviors are considered excessive by the experts. On the otherhand, schizotypy is related to a specific gene which increases the probability ofschizophrenia when combined with specific environmental conditions. Suchconditions are stress, anguish or even sadness (generally negative feelings).Schizotypy can be detected by its nine traits through the SPQ questionnaire.Obviously, interest lies in their association with serious psychiatric deceases. The datawere collected for the purposes of a student survey (Iliopoulou, 2004) in the School ofManagement Sciences of the University of Aegean and the Technological EducationInstitutes of Crete and Piraeus.Useful and interesting outcomes have been raised as far as the potentialinfluence of schizotypy on impulsive and compulsive buying behaviors. Furthermore,many proposals for deeper exploration have been occurred but unfortunately arebeyond the purposes of this thesis.
Subject :Bayesian model
Latent profile analysis
Deviance Information Criterion(DIC)
Date Created :14-12-2015
Licence :

File: Prifti_2015.pdf

Type: application/pdf