Unbiased L1 Estimators and Their Covariances,
Abstract
The parameters in a linear regression model can be estimated by minimizing the sum of the absolute residuals (L sub 1 estimation) instead of the more classical approach of minimizing the sum of squared residuals (least squares estimation). In addition to other nice properties L sub 1 estimators are less sensitive to outliers than least squares estimators. This paper describes a linear programming algorithm and computer program for obtaining unbiased L sub 1 estimators and estimates of their covariances. These estimated covariances are the new feature in this work and are an extremely important ingredient in hypothesis tests and confidence interval construction. Technical Report 65 provides an analogous treatment of L sub 1 estimation subject to linear constraints on the parameters. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 1980
- Accession Number
- ADA113531
Entities
People
- D. Book
- Herman Otto Hartley
- J. Booker
- R. L. Sielken Jr.
Organizations
- Texas A&M University