Robust Regression with Censored and Truncated Data

Choosing p suitably gives estimators that have desirable robustness proper­ ties. A well-known robust choice of p is Huber’s score function which is continuous, with p(0) = 0 and

(4) where c represents some measure of the dispersion of F. Using equation (4) in equation (2) is tantamount to applying the method of least squares to “metrically Winsorized residuals”, cf. Huber (1981, p. 180).