On Lattice Event Probabilities for Levin-Robbins-Leu Subset Selection Procedures
- 2016-08-12 (Fri.), 10:30 AM
- Recreation Hall, 2F, Institute of Statistical Science
- Prof. Cheng-Shiun Leu
- Department Biostatistics, Columbia University
Abstract
The Levin-Robbins-Leu (LRL) family of sequential subset selection procedures admits of a simple formula which provides a lower bound for the probability of various types of subset selection events. Here we demonstrate that a corresponding lower bound formula holds for lattice events when using the non-adaptive family member with binary outcomes. Lattice events are more general than the type of events that we previously considered in demonstrating that the non-adaptive LRL procedure selects “acceptable” subsets with an arbitrarily large pre-specified probability irrespective of the true population parameters. Interestingly, the proof of the lower bound formula for lattice events which we present here simplifies the methods previously used and sheds additional light on why the lower bound formula holds. As for acceptable subset selection, we conjecture that the other LRL family members, which allow for adaptive elimination of inferior candidates and/or recruitment of superior candidates, obey the same lower bound formulas for lattice events.