The choice of ρ (·) is often based on robustness and computational considerations
(see, e.g., Li, 1985).
However, it is clear that different choices of ρ (·) will affect not
only the efficiency of the estimator and the convergence properties of the algorithm
used in the minimization of the objective function (see, e.g., Li, 1985), but, more
importantly, the interpretation of the estimates.
used in the minimization of the objective function (see, e.g., Li, 1985), but, more
importantly, the interpretation of the estimates.
Consequently, in order to be able to
interpret the robust regression results that have appeared in the literature, it is now
important to study in detail the particular algorithm used to compute them.