ABSTRACT
Robust statistics is an extension of classical statistics that specifically takes into account the fact that models only provide an approximation to the true underlying random mechanism that generates the data. Model assumptions are almost never exactly satisfied in practice. A fraction of the observations can exhibit patterns not shared by the bulk of the data and therefore be outliers. The occurrence of departures from model assumptions by atypical values may have unexpected and deleterious effects on the outcomes of the analysis. These issues are interwined: an inappropriate model can be the reason of several data anomalies, and many outlying observations may suggest that the model is not adequate. A fundamental concept is that outliers are such only with respect to a certain model. Under the model, these observations are very unlikely or even impossible.
