National Technical University of Athens (NTUA)
Επιλογή οικογένειας μετασχηματισμών σε μπεϋζιανά στατιστικά μοντέλα: μεθοδολογία και εφαρμογές
Abstract
dc:descriptionThe primary purpose of this thesis was to produce a Bayesian methodology for contrasting and comparing several transformation families $T$ with a view of achieving normality of a set of observations. Four transformation families have been considered (Box-Cox, Modulus, Yeo & Johnson and Dual) along with the Identical and Logarithmic transformations. The proposed methodology demarcates the optimal choice of family $T$ and value of the transformation parameter λT within the Bayesian model selection framework using appropriate MCMC algorithms. We investigate different approaches to constructing compatible prior distributions forλT over alternative transformation families. In the case of univariate problems with no covariates, the power prior approach has been employed based on a common set of imaginary data simulated from the reference model of the Id transformation. An alternative prior used was a unit-information normal prior for the transformation parameter λT . Regarding problems with covariates, the role of the imaginary data has been granted to the training samples of the intrinsic Bayes factor and to the fractional parameter of the fractional Bayes factor, thus permitting the formulation of objective improper priors for the transformation parameter. Moreover, the research framework has been enhanced to incorporate the Bayesian variable selection simultaneously with the transformation family selection.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2016
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Charitidou, Efstratia
- Χαριτίδου, Ευστρατία
Subjects
dc:subject × 20- Αλγόριθμοι MCMC
- Πρότερη κατανομή δύναμης
- Επιλογή οικογένειας μετασχηματισμών
- Μπεϋζιανή επιλογή μοντέλου
- Ενδογενής παράγοντας Bayes
- Κλασματικός παράγοντας Bayes
- Συμβατότητα πρότερων κατανομών
- Φανταστικά δεδομένα
- MCMC
- Power prior distribution
- Transformation family selection
- Bayesian model selection
- Intrinsic Bayes factor
- Fractional Bayes factor
- Prior compatibility
- Imaginary data
- Φυσικές Επιστήμες
- Μαθηματικά
- Natural Sciences
- Mathematics
Rights
dc:rights- Statement dc:rights
-
- BY_NC
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/40069