M. L. Menendez, Departamento de Matematica Aplicada, E.T.S de Arquitectura, Universidad Politecnica de Madrid, 28040 Madrid, Spain; D. Morales, L. Pardo, Departamento de Estadistica e I.O., Facultad de Matematicas, Universidad Complutense de Madrid, 28040 Madrid, Spain; M. Salicru, Departamento de Estadistica, Avd. Diagonal 645, Universidad de Barcelona, 08028 Barcelona, Spain
Abstract: Burbea and Rao (1982a, 1982b) gave some general methods for constructing quadratic differential metrics on probability spaces. Using these methods, they obtained the Fisher information metric as a particular case. In this paper we apply the method based on entropy measures to obtain a Riemannian metric based on $(h,\Phi)$-entropy measures (Salicru et al., 1993). The geodesic distances based on that information metric have been computed for a number of parametric families of distributions. The use of geodesic distances in testing statistical hypotheses is illustrated by an example within the Pareto family. We obtain the asymptotic distribution of the information matrices associated with the metric when the parameter is replaced by its maximum likelihood estimator. The relation between the information matrices and the Cramer-Rao inequality is also obtained.
Keywords: $(h,\Phi)$-entropy measures, information metric, geodesic distance between probability distributions, maximum likelihood estimators, asymptotic distributions, Cramer-Rao inequality.
Classification (MSC 1991): 62B10, 62H12
Full text available as PDF (smallest), as compressed PostScript (.ps.gz) or as raw PostScript (.ps).
Access to the full text of journal articles on this site is restricted to the subscribers of Myris Trade.
To activate your access, please contact Myris Trade at myris@myris.cz.
Subscribers of Springer (formerly Kluwer) need to access the articles on their site, which is http://www.springeronline.com/10492.
