Measure compact sets of functions and consistency of statistical models

G. Peškir

Teoriâ veroâtnostej i ee primeneniâ, Tome 38 (1993) no. 2, pp. 431-438

Résumé

Sufficient conditions for consistency of statistical models are deduced by using Fréchet—Ŝhmul'yan's necessary and sufficient conditions for conditionally compactness relative to the topology of convergence in measure being imposed on the family of associated densities by a so-called control measure. The method relies upon the facts established in [4] where sufficient conditions for consistency are deduced by using necessary and sufficient conditions for conditional compactness of the family of associated densities relative to the topology of pointwise convergence. When a statistical model under consideration admits a finite control measure, the present conditions for conditional compactness become weaker and easily verified. However the present approach relies upon the fact that any control measure that yields consistency must be nice enough in such a way that the upper oscillation of densities on infinitesimally small balls behaves smoothly relative to the distribution of the random phenomenon under consideration.

Détail

Keywords: asymptotic likelihood theory, statistical model, reference measure, parameter set, analytic metric space, sample space, the likelihood function the log-likelihood function, the unknown (true) distribution, the (upper, empirical) information function, maximum likelihood estimator, pointwise compact metrizable, separable, conditionally compact, Fréchet–Ŝhmul'yan compactness criteria, control measure, upper semicontinuous, the Hewitt–Savage 0-1 law, the projection theorem, the permutation invariant $\sigma$-algebras.
Mots-clés : consistent