Assessing the Behavior of Robust Estimates of Location in Small Samples: Introduction to Configural Polysampling.
Abstract
Configural polysampling offers an effective alternative to both usual Monte Carlo techniques and small sample asymptotics in studying and improving robust estimates. Attention here is restricted to location estimators that are location and scale invariant. The methods are applicable to both simple and compound sampling situations. Several possible aspects of such studies include: (a) the determination of the minimum attainable variance in each sampling situation, (b) the determination of the maximum attainable polyefficiency over several sampling situations, (c) the fine-tuning of robust estimates with the intent of increasing their polyefficiency for small n, and (d) the identification of data configurations where one can a priori expect poor performance with certain estimators. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1981
- Accession Number
- ADA105956
Entities
People
- Daryl Pregibon
- John W. Tukey
Organizations
- Princeton University