The "Hook and Loop" Resampling Plane

We propose a new resampling scheme that takes literally the concept of the non-parametric bootstrap in which new samples are generated from the empirical distribution function. The introduced resampling concept is totally heuristic, but already shows promising results when applied to model selection. We show that for a range of linear models, the proposed resampling scheme outperforms the classical model selection techniques as well as its predecessor, the non-parametric bootstrap. It also simplifies the practical problem of choosing residual scaling or the length of the subsample that exists in the traditional bootstrap based model selection approach.