University of Ioannina
Τοπικές φ-αποκλίσεις στη στατιστική θεωρία πληροφοριών και εφαρμογές σε ελέγχους στατιστικών υποθέσεων και επιλογής μοντέλου
Abstract
dc:descriptionThe Dissertation consists of five Chapters the summary of the dissertation in English and the Bibliography. In the first introductory chapter, a brief review on the φ-divergence measures is given as well as the concepts that are necessary for the development and study of the following chapters. Chapter 2 focuses on the definition of a broad class of local divergence measures between two probability measures or between the corresponding densities. The local divergences developed are based on the classical Csiszar divergence and provide a measure of the pseudo-distance (or statistical distance or divergence) between two distributions in a specific area of their joint domain. In this way they provide a useful tool for the quantification of the statistical distance between two distributions, locally, in a specific area of their joint domain. In addition, in Chapter 2 the range of values of the local divergences are studied and the analytical expressions of the proposed local divergences are further determined when the distributions are members of the exponential family and the case of the multivariate normal distribution is also considered. Chapter 2 exemplifies the methodology via a simulation study which illustrates the robust behavior of the proposed local divergence measures in identifying the discrepancies between two populations which cannot be captured using classical global measures. Chapter 3 focuses on the construction and study of parametric statistical tests for goodness of fit (one sample problem) and tests of homogeneity (two sample problem), locally, in a subset of the domain of the distributions under consideration. More precisely, test statistics are defined and their asymptotic distributions are obtained under the null hypothesis. In addition, in Chapter 3 the effectiveness of the developed local parametric tests are investigated by means of simulation studies based on their type I error and power. The behavior of the proposed statistical tests has also been investigated via real data. In Chapter 4, a local model selection criterion (LDiv.IC) is developed and studied using the local BHHJ power divergence. We focus on the BHHJ measures of divergence because their functional expression is particularly useful in the construction of the criterion. Moreover in Chapter 4, simulations are presented in order to evaluate the performance of the proposed model selection criterion in a local setting and three applications of the proposed methodology are given by analyzing real datasets. In Chapter 5, we provide suggestions for further research and possible extensions of this work are indicated, along with and some open problems left to consider in this context.
Degree
thesis:*- Grantor dc:publisher
- University of Ioannina
- Year dc:date
- 2017
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Avlogiaris, Georgios
- Αυλογιάρης, Γεώργιος
Subjects
dc:subject × 48- φ-απόκλιση
- Απόκλιση Csiszar
- Απόκλιση Kullback-Leibler
- Απόκλιση Cressie and Read
- Απόκλιση BHHJ
- Τοπική απόκλιση
- Εκθετική οικογένεια κατανομών
- Τοπικός πίνακας πληροφορίας Fisher
- Τοπικό στατιστικό ελέγχου Wald
- Τοπικό στατιστικό ελέγχου Rao
- Τοπικός έλεγχος καλής προσαρμογής
- Τοπικός έλεγχος ομοιογένειας
- Τοπικό στατιστικό ελέγχου φ-απόκλισης
- Επιλογή μοντέλου
- Τοπική επιλογή μοντέλου
- Κριτήριο πληροφορίας τοπικής απόκλισης
- Τοπικό κριτήριο επιλογής μοντέλου
- Τοπική αναμενόμενη συνολική διαφορά
- Τοπική απόκλιση BHHJ
- Μοντέλα μείξης
- Θεωρία point process
- φ-divergence
- Csiszar divergence
- Kullback-Leibler divergence
- Cressie and Read power divergence
- BHHJ power divergence
- Local divergence
- Exponential family
- Local Fisher information matrix
- Local Rao and Wald tests
- Local goodness of fit test
- Local homogeneity test
- Local φ-divergence test
- Model selection
- Local model selection
- AIC
- LDiv.IC
- LAIC
- Local divergence information criterion
- Local model selection criterion
- Local expected overall discrepancy
- Local BHHJ power divergence
- Mixture models
- Point process theory
- Φυσικές Επιστήμες
- Μαθηματικά
- Natural Sciences
- Mathematics
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/44728