The topic of robust estimation has appeared often in the literature (see Armstrong, Frome, and Sklar (1980), for instance, for a list of references). Least absolute value (LAV) estimation, in particular, has been shown to be a useful alternative to least squares estimation. In this paper a number of extensions to an LAV algorithm of Armstrong and Kung (1982) are developed and tested, and the computational results reported. The topics are organized as follows: locating the three best subsets; determining the most influential of a selected set of observations; altering the percent-of-optimality figure; and, forcing parameters to be in the model.
展开▼