| Abstract [eng] |
Properties of linear combinations of order statistics (L-statistics), where samples are drawn without replacement, are considered in the thesis. The main object of the thesis is an improvement of the normal approximation to distributions of L-statistics by one-term Edgeworth expansions. An accuracy of these approximations is estimated using the Hoeffding decomposition of finite population symmetric statistics. In the first chapter of the thesis, explicit expressions of the first terms and remainder terms of the Hoeffding decomposition of L-statistics are obtained. The main applications of the decomposition are given in the second chapter: the optimal upper bound for variances of the sample minimum and maximum is obtained; sufficient conditions for the asymptotic normality of L-statistics are established; the one-term Edgeworth expansion for L-statistics is constructed and sufficient conditions for the validity of this approximation are obtained. In the third chapter, estimators of the variance and parameters that define the Edgeworth expansion of an L-statistic are constructed. In the fourth chapter, a one-term Edgeworth expansion for a Studentized L-statistic and empirical Edgeworth expansions are constructed and analyzed. |
