An iterative algorithm for fitting a linear model in the case where the data may contain outliers that would distort the parameter estimates if other estimation procedures were used. The procedure uses weighted least squares, the influence of an outlier being reduced by giving that observation a small weight. The weights chosen in one iteration are related to the magnitudes of the residuals in the previous iteration—with a large residual earning a small weight. This is one of a number of methods for robust regression, the weights being related to M-estimates.
Subjects: Probability and Statistics.