Karin Stål: On the identification of multiple influential observations in linear and nonlinear regression

It is well known that outliers or influential observations can strongly influence, or even distort the results of the data analysis. If the data contains only one influential observation it is relatively simple to identify it. However, in practice the data can contain more than one influential observation, and the identification of these observations is more complicated. This is partly due to masking effect. Masking occur when an observation is not identified as influential until another observation is deleted first. A lot of work has been done on identifying multiple influential observations in linear regression, especially when masking is present. Much less attention has been devoted to the identification of multiple influential observations in nonlinear regression. In this work we extend the influence measures based on empirical influence curve for identifying single influential observations in nonlinear regression models to the case of multiple observations. The procedures for detecting multiple influential observations will be worked out that also can handle masking effect.