ON EDGEWORTH EXPANSIONS IN THE MIXTURE CASES
ON EDGEWORTH EXPANSIONS IN THE MIXTURE CASES
复制标题
混合情况下的Edgeworth展开
DOI:
10.1214/aos/1176347029
复制
发表时间:
1989
期刊:
影响因子:
--
通讯作者:
Kesar Singh
中科院分区:
文献类型:
--
作者:
G. Babu;Kesar Singh
where ' denotes transpose. In the lattice case, where each univariate population assigns its entire mass to countably many equidistant points, the expansion is different and an explicit form is known only for a restricted class of sets A. Consider an univariate statistic of the form H(Zn), for a smooth function H. Bhattacharya and Ghosh (1978) used the expansion for Zn to establish the validity of the formal Edgeworth expansion for Vn[H(Zn) - H(E(Zn))] in the nonlattice case. In the lattice case, an explicit form of the expansion is typically not available for nonlinear univariate statistics. See Yarnold (1972) for instance, where an explicit expansion for Pn(A) is given, when A is an ellipsoid. In this paper we consider Edgeworth expansions for H(Zn), when X has a lattice distribution and Y has a continuous distribution. Situations like this do arise in statistical experiments. As a simple example, X could be the age of a plant or animal in years and Y could be its weight; H(Zn) = YJXn. In this article we establish that no correction factor is needed for the lattice character of X in the one-term Edgeworth expansion. Beyond one term, the answer is not known to us at present. Our approach for establishing the above mentioned fact is as follows: Let q be a symmetric random variable having finite third moment and whose characteristic function vanishes outside a compact interval. It is first shown that in the