A new method called stepwise local influence analysis is proposed to detect influential observations and to identify masking effects in a dataset. Influential observations are detected step-by-step such that any highly influential observations identified in a previous step are removed from the perturbation in the next step. The process iterates until no further influential observations can be found. It is shown that this new method is very effective to identify the influential observations and has the power to uncover the masking effects. Additionally, the issues of constraints on perturbation vectors and bench-mark determination are discussed. Several examples with regression models and linear mixed models are illustrated for the proposed methodology.
