A review of subsampling methods and their robustness

This paper looks for establishing how robust statistics theoretical models and intensive computation methods interact. The functional approach is discussed for establishing the convergence of estimators and predictors to the parameter, expressed as a functional of the unknown distribution function. It allows the establishment of general conditions that sustain the use of Jackknife and Bootstrap for estimating confidence intervals, hypothesis testing etc., using only sampling information. Some Bootstrap derived methods are described.