Chapter 10 Robust Regression 505 computes the STS estimator, where sc is the measure of scale to be used. By default, the percentage bend midvariance is used. 10.13.3 E-Type Skipped Estimators Skipped estimators remove any outliers among the cloud of data (x i1 , . . . , x i p , y i ), i = 1, . . . n, and then fit a regression line to the data that remain. E-type skipped estimators (where E stands for error term) look for outliers among the residuals based on some preliminary fit, remove (or downweight) the corresponding points, and then compute a new fit to the data. Rousseeuw and Leroy (1987) suggested using least median of squares (LMS) to obtain an initial fit, remove any points for which the corresponding standardized residuals are large, and then apply least squares to the data that remain. But He and Portnoy (1992) showed that the asymptotic efficiency is 0. Another E-type skipped estimator is to apply one of the outlier detection methods in Chapter 3 to the residuals. For example, first fit a line to the data using the the STS estimator described in Section 10.13.1. Let r i (i = 1, . . . , n) be the usual residuals. Let M r be the median of the residuals and let MAD r be the median of the values |r 1 - M r |, . . . , |r n - M r |. Then the ith point (x i , y i ) is declared a regression outlier if