Selection tests are widely used in job applications, college admissions, training programmes, etc. We examine the validity of the widely-held belief that individuals with high selection test scores are high performers. We find that, under normal assumptions, the correlation between the test scores of those selected and their subsequent performance scores is quite low for low-to-moderate selection ratios. If, however, the selection objective is to select at least r of the top n candidates out of a pool of N candidates, we show that the natural selection rule (selecting the top n selection test scores) has certain optimality properties. We investigate the properties of the natural selection rule under normal assumptions theoretically and via simulation studies.
This is joint work with Rahul Mukerjee, Indian Institute of Management, and Abhijit Das Gupta, University of Washington.