On the Rate of Convergence in Normal Approximation and Large Deviation Probabilities for a Class of Statistics

Document Type

Article

Publication Date

7-11-2011

Identifier/URL

40902518 (Pure)

Find this in a Library

Catalog Record

Abstract

A new class of statistics is introduced to include,as special cases, unsigned linear rank statistics, signed linearrank statistics, linear combinations of functions of orderstatistics, linear functions of concomitants of order statisticsand a rank combinatorial statistic. For this class, the rate ofconvergence to normality and Cramér-type large deviationprobabilities are investigated. Under the assumption that the underlyingobservations are only independent, it is shown that this rate isO(N−δ/2logN) if the first derivative of the score-generatingfunction φ satisfies a Lipschitz condition of order δ,0<δ≤1, that it is O(N−1/2) if φ′′satisfies a Lipschitz condition of order δ≥12, andthat Cramér's large deviation theorem holds in the optimal range0

Comments

Publisher Copyright: © VSP 2003.

Catalog Record

Share

COinS