
Accession Number : AD0661353
Title : CONTRIBUTIONS TO ROBUST ESTIMATION,
Corporate Author : NEW YORK UNIV N Y COURANT INST OF MATHEMATICAL SCIENCES
Personal Author(s) : Mike,Valerie
Report Date : SEP 1967
Pagination or Media Count : 133
Abstract : For the problem of efficiencyrobust estimation of the location parameter theta of a family of symmetric pdf's f(xtheta(vertical line) lambda), lambda epsilon Lambda = (1,...,m), theta epsilon Theta = (theta such that minus infinity < theta < infinity), the method of 'mixture models' of Birnbaum is applied to determine generalized Pitman estimators, which are shown to be admissible, with squared error loss function, under broad regularity conditions. With increasing sample size, these estimators are proved to be fully efficient (i.e., asymptotically equivalent, for each value of lambda, to the maximum likelihood estimator which would be appropriate if the true value of lambda were known). Computationally tractable analogous estimators based on k sample quantiles are defined in the context of the model representing their asymptotic normal distributions. It is shown that with increasing k these approach equivalence to the fully efficient estimators based on complete samples. Equivalent estimators are given also for the case of unknown scale parameters. Efficiencyrobust linear unbiased estimators based on sample quantiles are derived and the optimal spacing of quantiles is discussed. (Author)
Descriptors : (*STATISTICAL ANALYSIS, OPTIMIZATION), MATHEMATICAL MODELS, RANDOM VARIABLES, THESES, SAMPLING, APPROXIMATION(MATHEMATICS)
Subject Categories : Statistics and Probability
Distribution Statement : APPROVED FOR PUBLIC RELEASE