Screening: Optimal Methods and Anomalies

Abstract: The talk will consist of three parts, each based on a different research paper, all concern a decision maker who uses noisy unbiased assessments to screen elements from a general set:

1. The first part, based on "A Bias of Screening" (AER: Insights), shows that stricter screening not only reduces the number of accepted elements, but possibly reduces their average expected value. 
2. The second part shows that one can actually generate 'lucky coins' as additional binary noise can strictly improve a screening process. We also provide a comparison of different noisy signals under threshold (screening) strategies and optimal ones, and depict several partial characterizations of cases in which one noise is preferable over another. Accordingly so, we establish a novel method to compare noise variables using a contraction mapping between percentiles.
3. The third and last part would show that one-stage screening is possibly preferable to multi-stage screening. In addition, we present a method to preform a perfect screening under rather general conditions.