Imprecise predictive selection based on low structure assumptions

New methods for statistical selection are presented, where the inferences have a nonparametric predictive nature. The basic assumption is Hill's $A_{(n)}$. Assuming that values of some random quantities from $k \geq 2$ independent sources are observed, $A_{(n)}$ provides a predictive probabilistic inference for further unknown random quantities from each source. Selection is mainly based on imprecise probabilities for the event that the next observation from the selected source will be greater than the next observation from every non-selected source. Both selection of a single source and selection of $m (m \leq k - 1)$ sources are presented, considering the probability that all selected sources are better than all non-selected sources. A second related approach, using imprecise previsions with an interpretation as bounds for expected values for future observations, is presented briefly.