A Conditional Property of Adaptive Estimators.
Abstract
In adaptive estimation, it is often considered that an estimator has made a mistake if the component estimator chosen for use is not the most efficient for the distribution sampled. Theoretical and simulation results point to a fallacy in this line of thought. The Monte Carlo study involves extension of the Princeton Swindle to distributions conditional on a location-and scale free statistic, and to the uniform. The results give a partial explanation for the sometimes surprising robustness of adaptive L-estimators. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 25, 1979
- Accession Number
- ADA072991
Entities
People
- William C. Parr
Organizations
- Southern Methodist University