ROBUST REGRESSION BY MODIFIED LEAST-SQUARES.
Abstract
Estimates of regression parameters are usually found by minimizing the sum of squared differences between observed and predicted values of a dependent variable. As is well known, such estimates can be seriously impaired by the presence of outliers. To combat this effect, I consider minimizing an alternative function of differences. This function is the square for arguments less than a certain value (determined from the data itself) and linear for arguments beyond that. An algorithm for computing the estimate is given, large-sample properties are derived, and small-sample properties are studied by means of Monte Carlo exploration of various error distributions. An extended summary of results is given. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1967
- Accession Number
- AD0664508
Entities
People
- D. A. Relles
Organizations
- Yale University