ROBREG <data>, L / Reino Siren
Performs a robust estimation of the parameters of a linear regression
model by the method of Least Median of Squares (Rousseeuw 1984,
Rousseeuw & Leroy 1987). Instead of minimizing the sum of squared
residuals (OLS), the median of squared residuals is minimized, giving
estimates which are highly robust to outliers in the data.
The model is specified by activating variables with Y and X´s in the
<data>.
The constant term is omitted by setting CONSTANT=0.
The seed number used in subsampling may be specified by SEED=<integer>.
It is optional.
The number of trials or samples from the data used in the estimation,
may be specified by TRIALS=<integer>. The default value is 1500.
ROBREG (continued)
Furthermore the following masks for special output variables from the
LMS -fit may be used:
U is the residual variable,
V is the variable for predicted values,
T is the variable for the standardized lms-residuals.
The printed output includes the LMS estimates of the regression
coefficients and the robust error scale estimate.
The first line of results (L) is optional.
Literature:
Rousseeuw, Peter, J. 1984. Least median of squares regression. J. Am.
Stat. Assoc. Vol 79, 871-880.
- " - & Leroy, Annick, M. 1987. Robust Regression and Outlier
Detection. Wiley. New York.
http://win-www.uia.ac.be/u/statis/publicat/progress_abstr.html
R = More information on regression analysis
Help System (web edition)