• Journal of the Korean Statistical Society
• /
• 제19권2호
• /
• pp.176-181
• /
• 1990

Sung [1990] presented an analogue of the classical Cramer-Rao inequality for median-unbiased estimators with continuous multivariate densities depending upon a vector parameter. In the process, diffusivity, a new dispersion measure relevant to median-unbiased estimators, was defined to be a function of median-unbiased estimator's density height. In this paper we shall elaborate these ideas by defining a second kind of diffusivity and discuss the role of model-unbiasedness in median-unbiased estimation in connection with this seconde kind of diffusivity. In addition, median-unbiased estimation will be compared to mean-unbiased estimation.

• Journal of the Korean Statistical Society
• /
• 제23권1호
• /
• pp.187-198
• /
• 1994

We derive a new Cramer-Rao type lower bound for the reciprocal of the density height of the median-unbiased estimators which improves most of the previous lower bounds and is attainable under much weaker conditions. We also identify useful necessary and sufficient condition for the attainability of the lower bound which is considerably weaker than those for the mean-unbiased estimators. It is shown that these lower bounds are attained not only for the family of continuous distributions with monotone likelihood ratio (MLR) property but also for the location and scale families with strong unimodal property.

• Journal of the Korean Statistical Society
• /
• 제28권1호
• /
• pp.35-44
• /
• 1999

We derive a new upper bound of convolution type for the median-unbiased estimators with respect to an arbitrary unimodal utility functions. We also obtain the necessary and sufficient condition for the attainability of the information bound. Applications to general MLR(Monotone Likelihood Ratio) model and censored survival data re discussed as examples.

• Journal of the Korean Statistical Society
• /
• 제26권4호
• /
• pp.445-452
• /
• 1997

A more generalized version of the information inequality based on diffusivity which is a natural measure of dispersion for median-unbiased estimators developed by Sung et al. (1990) is presented. This non-Bayesian L$_{1}$ information inequality is free from regularity conditions and can be regarded as an analogue of the Chapman-Robbins inequality for mean-unbiased estimation. The approach given here, however, deals with a more generalized situation than that of the Chapman-Robbins inequality. We also develop a Bayesian version of the L$_{1}$ information inequality in median-unbiased estimation. This latter inequality is directly comparable to the Bayesian Cramer-Rao bound due to the van Trees inequality.

• Journal of the Korean Statistical Society
• /
• 제22권1호
• /
• pp.1-12
• /
• 1993

An approach to make the information bound sharper in median-unbiased estimation, based on an analogue of the Cramer-Rao inequality developed by Sung et al. (1990), is introduced for continuous densities with a nuisance parameter by considering information quantities contained both in the parametric function of interest and in the nuisance parameter in a linear fashion. This approach is comparable to that of improving the information bound in mean-unbiased estimation for the case of two unknown parameters. Computation of an optimal weight corresponding to the nuisance parameter is also considered.

• Journal of the Korean Statistical Society
• /
• 제24권2호
• /
• pp.361-373
• /
• 1995

Three remarks are offered, pertaining to classes of estimators Pitman-dominating a given estimator. The first remark concerns incorporating general loss in the construction of such classes. The second remark concerns Pitman domination comparisons amongst the members of such classes. The third remark concerns construction of such a class in the location-scale case.