We propose a stepwise procedure for the detection of multiple outliers in generalized linear models and nonlinear regressions. The algorithm starts with a high breakdown point estimation method to find the potential outliers, then uses an adding-back iterative approach to confirm such outliers with envelopes derived from simulation. The proposed method can overcome the masking and swamping problems commonly occurred in multiple outliers detection. We apply the procedure to two real examples, and satisfactory results are obtained.
|Number of pages||11|
|Journal||Computational Statistics and Data Analysis|
|Publication status||Published - 1997|