Rank-Based Inference for Linear Models: Asymmetric Errors

摘要

In this paper robust, rank-based inference procedures are considered for general linear models with (possibly) asymmetric errors. Approximating standard errors of estimates and testing hypotheses about the model parameters require estimating a scaling functional, and an approach is developed which, unlike previous work, does not require symmetry of the underlying error distribution or replicates in the design matrix. Hence, important asymmetric models such as arise in life testing can now be handled. Further, it is shown that the asymptotic properties of the inference procedures hold with simpler conditions on the design matrix than previously required. In addition an estimate of the intercept is developed without requiring the assumption of a symmetric error distribution. (Author)