National Technical University of Athens (NTUA)
Μη παραμετρική προσέγγιση σε πειραματικούς σχεδιασμούς: γενικευμένες μη παραμετρικές υποθέσεις σε ανισόβαρα F στατιστικά
Abstract
dc:descriptionThe rank transform (RT) of a classical test occurs when we replace all the observations in the classical test statistic with their ranks in the set of all observations. The RT statistics were studied only in special cases and on the basis of these partial results the assertion of their general validity was formulated. Subsequent research has shown that this assertion is wrong and the RT procedure is inappropriate for testing many parametric hypotheses. This is due to two reasons. The first reason is that the parametric hypotheses are not invariant under nonlinear transformations of the data, such as the RT. This problem was solved with the introduction of nonparametric hypotheses by Akritas & Arnold (1994) which are defined in terms of distribution functions and are invariant under nonlinear monotone transformations. The second reason is due to the heteroscedasticity that is produced by the nonlinear transformations. This problem is solved with the use of weighted statistics. In this thesis we develop a methodology that is relevant to the RT methodology. We use subspaces to describe the nonparametric null hypotheses, which leads to a natural extension of the models and the class of nonparametric hypotheses considered there. To test these generalized hypotheses we introduce a new family of midrank statistics which are based on weighted F-statistics and are appropriate for unbalanced heteroscedastic designs with independent observations such as factorial and nested designs, and with dependent observations such as repeated measures designs. The procedures given are also applicable to discrete data and, since continuity of the distribution functions is not assumed, they can also be applied to data with ties. We show that the proposed statistics converge in distribution to central chi-squared distributions under their respective null hypotheses. With simulation studies we examine the speed of convergence of the new statistics and compare their achieved error rate and power with that of competing procedures in unsaturated factorial designs, nested designs and unsaturated repeated measures designs.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2009
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Stavropoulos, Athanasios
- Σταυρόπουλος, Αθανάσιος
Subjects
dc:subject × 20- Μη παραμετρικά μοντέλα
- Γενικευμένες μη παραμετρικές υποθέσεις
- Ανισόβαρα F στατιστικά τάξεων
- Ασυμπτωτική θεωρία
- Ετεροσκεδαστικοί σχεδιασμοί επαναλαμβανομένων μετρήσεων
- Εμφωλευμένοι σχεδιασμοί
- Στατιστικά τύπου λόγου πιθανοφάνειας
- Στατιστικά τύπου Wald
- Nonparametric models
- Generalized nonparametric models
- Weight F rank statistics
- Asymptotic theory
- Heteroscedastic repeated measures designs
- Nested designs
- Likelihood ratio type statistics
- Wald type statistics
- Φυσικές Επιστήμες
- Μαθηματικά
- Natural Sciences
- Mathematics
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/17386