Research · Per Ardua

No Concerns: Conjunctive Hiring and the Dilution of Exceptional-Skill Selection

Selectivity is not aim

AP-7 Applied DOI SSRN

Executive Summary

A hiring process is usually judged by how many candidates it rejects. This paper separates that quantity, selectivity, from aim: how much the process raises the share of hires who have the capability the organization wants. The model is deliberately favorable to the process: skills are measured without error and every rule is applied correctly.

Under those conditions the paper shows three things. When a candidate must clear a threshold on each of k independent skills, enrichment for exceptional ability on any one skill equals the inverse of that skill's pass rate, and at a fixed hire rate it falls toward 1 as k grows. Once each skill is recorded as pass or fail, no rule applied to those marks can do better than the all-pass rule, because the record has discarded how far above the threshold the candidate was. Whether this matters depends on the work: in a team-value model, selection on strength beats selection on the absence of weakness by a wide margin when payoffs are heavy-tailed, loses most of that margin when usable contribution is capped at the 95th percentile, and is the wrong rule for weakest-link work. The paper states the conditions under which the argument holds, the evidence against it, and the tests that would refute it.

Key Findings

  • Dilution: Under a k-threshold conjunctive rule, enrichment for exceptional ability on any one skill equals the inverse of that skill's pass rate, and at a fixed hire rate it falls toward 1 as k grows.
  • 1.00% to 1.26%: A ten-box process that rejects 90% of applicants raises the share of top-1% specialists from 1.00% to 1.26%.
  • Pass/fail ceiling: Once evidence is reduced to pass/fail marks, no decision rule over those marks does better than all-pass.
  • The work decides: Selection on strength wins by a wide margin under heavy-tailed payoffs, loses most of that margin when usable contribution is capped at the 95th percentile, and is the wrong rule for weakest-link work.

Code and Data

Simulation code is archived at Zenodo: concept DOI 10.5281/zenodo.23164405 (resolves to latest).

Download Full Paper

Access the complete research paper with detailed methodology, empirical evidence, and formal proofs.

Download PDF