Selectivity refers to the number of rows an index returns.
An index that returns a small number of rows has high selectivity, a positive attribute.
An index that return a large number of rows has low selectivity, an attribute so negative that the Optimizer often selects a full-table scan over a NUSI with low selectivity, because the full-table scan can be less costly.
All UPIs and USIs are highly selective by definition, like well-chosen NUPIs. High selectivity is favored for its precision and low cost, involving a small number of disk I/Os, which is a performance-enhancing attribute.