Bayesian Sequential Experimental Design for Planning Series of Police Lineups.

This article develops a formal Bayesian sequential experimental design framework to evaluate and optimize the configuration of series of police lineups involving multiple witnesses and suspects. It introduces methods to model how information gained from one lineup can inform the configuration of subsequent lineups, thereby increasing the overall evidentiary value beyond treating lineups independently. Using both synthetic data based on signal detection theory and semi-synthetic analyses grounded in human-subjects lineup studies, the authors demonstrate that adaptive policies—particularly those that present weaker witnesses first to break symmetry in suspect guilt probabilities—can yield more informative outcomes. The study compares several lineup configuration policies, finding that a "greedy" policy, which optimizes each lineup based on current knowledge without considering future lineups, often performs near optimally, while random or non-updating policies substantially underperform. The work highlights the potential for these analytic tools to improve investigative procedures and suggests avenues for future research, including incorporating cognitive dependencies, confidence measures, and outcome-weighted objectives that reflect legal principles such as Blackstone's Ratio.