How crime data incompleteness shapes ethnic disparities in predictive policing: evidence from Chicago

Jiang, Mengqin K., Liberatore, Federico ORCID: https://orcid.org/0000-0001-9900-5108 and Preece, Alun ORCID: https://orcid.org/0000-0003-0349-9057 2026. How crime data incompleteness shapes ethnic disparities in predictive policing: evidence from Chicago. Crime Science 15 (1) , 40. 10.1186/s40163-026-00306-9

PDF - Published Version Available under License Creative Commons Attribution. Download (2MB)

Abstract

Purpose: Fairness in Predictive Policing (PP) has been explored by many researchers. However, to the best of our knowledge, no work has quantified the impact of event-level data incompleteness in spatio-temporal prediction on each ethnicity. In this paper, we propose a model to quantify the impact of recorded crime data incompleteness on ethnic disparities in policing. We also examine whether these disparities depend on where in the city data are missing. Methods: We introduce a neighbourhood-level method that allocates PP prediction errors (false positives as potential over-policing, false negatives as potential under-policing) to ethnic groups in proportion to their share of the neighbourhood population. Using Chicago crime records (2022–2024), census data, and HAGEN (a state-of-the-art spatial-temporal PP model), we explore data incompleteness by reducing the original crime data at varying ratios in two settings: (i) 25–100% reductions in neighbourhoods with low racial diversity; (ii) 50% or 100% reductions in communities dominated by a single ethnicity (Hispanic, Non-Hispanic Black, Non-Hispanic White, or Non-Hispanic Asian). Enforcement bias outcomes are then compared across ethnicities and scenarios. Results: Removing records from a single-ethnicity-majority community generally reduces predicted over-policing and increases predicted under-policing for that ethnicity across the whole study area, with Asian-majority scenarios a consistent exception. In low-diversity neighbourhoods, greater missingness amplifies prediction bias against Non-Hispanic Black residents specifically: predicted over-policing falls and predicted under-policing rises, with the opposite holding for other groups. The largest overall shifts occur when data are removed from Black-majority communities. Conclusions: Incomplete crime records could propagate into substantial, ethnically unequal disparities in model-predicted policing exposure, whose direction depends on where missingness concentrates. The findings motivate auditing missingness and its demographic distribution as a prerequisite for deploying event-level PP systems.

Item Type:	Article
Date Type:	Publication
Status:	Published
Schools:	Schools > Computational & Mathematical Sciences Schools > Computer Science & Informatics
Publisher:	SpringerOpen
ISSN:	2193-7680
Date of First Compliant Deposit:	19 October 2026
Date of Acceptance:	30 September 2026
Last Modified:	19 Oct 2026 14:00
URI:	https://orca.cardiff.ac.uk/id/eprint/190310

Actions (repository staff only)

Edit Item