University of Ioannina
Θέματα πολυδιάστατης ανάλυσης με ελλειπτικές κατανομές και μονότονα ελλιπή δεδομένα
Abstract
dc:descriptionOur main aim in this dissertation was to extend the available results in the literature in two aspects: on the one hand adopting elliptically contoured distributions and on the other hand considering monotone missing data. Initially, the analytic form of the maximum likelihood estimators is derived for the mean vector and the scale matrix of the elliptic distributions. The likelihood ratio test statistics are obtained for testing several hypotheses about the mean and scale parameters of the elliptic distributions, in the presence of missing data with a monotone pattern (Batsidis and Zografos 2006, J. St. Plan. Inference, 136, 2606-2629). Afterwards, two rules are proposed for the classification of an observation into one of two elliptic populations, which differ only in their location parameters, under the assumption of 2-step monotone missing training samples. The first is derived if, in the usual linear discriminant rule, the unknown parameters are replaced by their estimators. Hence, the analytic expressions for the MLEs of the mean vectors and the common scale matrix of two elliptic populations based on 2-step training samples are obtained. The second rule is constructed, as a linear combination of two different linear discriminant functions, which are obtained by using two different methods of handling missing data. Based on multivariate t data, simulation results are presented in order to compare the two procedures (Batsidis and Zografos, 2006, Metrika, 64, 221-241). Furthermore, we study the distribution of the probabilities of the errors of misclassification of a new observation into one of two multivariate normal populations, which have distinct mean vectors and common covariance matrix, under the additional assumption that we have to estimate the unknown parameters by using 2-step monotone training samples (Batsidis et al., 2006, Com. St. and Data Analysis, 50, 2600-2634). Finally, we consider the problem of multivariate elliptic linear regression model of p response variables on a q explanatory variables, when the response variables have missing values of a monotone pattern. In this frame, the explicit form for the MLE of the parameters is derived. Moreover, we obtain the likelihood ratio test statistic in order to examine the existence of linear constraints on the regression parameters (Batsidis and Zografos, 2007, Com. Stat.Th. Methods, accepted).
Degree
thesis:*- Grantor dc:publisher
- University of Ioannina
- Year dc:date
- 2005
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Batsidis, Apostolos
- Μπατσίδης, Απόστολος
Subjects
dc:subject × 20- Ελλειπτική οικογένεια κατανομών
- Πολυμεταβλητή ανάλυση
- Ελλιπή δεδομένα
- Στατιστική συμπερασματολογία
- Διαχωριστική ανάλυση
- Σφάλματα μη ορθής ταξινόμησης
- Πολυμεταβλητή γραμμική παλινδρόμηση
- Πολυδιάστατη t κατανομή
- Elliptically contoured distributions
- Multivariate analysis
- Missing data
- Statistical inference
- Discriminant analysis
- Errors of misclassifications
- Multivariate linear regression analysis
- Multivariate t distribution
- Φυσικές Επιστήμες
- Μαθηματικά
- Natural Sciences
- Mathematics
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/24647