October Research Brief
Here are the studies I found since the last brief that I think are worth your attention. This month there's a randomised trial of AI body-worn camera analytics, a look at what Waze police alerts do to speeding, a study of what happened to an untargeted hacker forum after a police takedown, and some thoughts on how the public picks its police chiefs. As ever, if you spotted something I missed or got wrong, let me know and I'll add it to the next edition.
TL;DR?
AI body-worn camera analytics and officer behaviour — a randomised trial of 314 Arizona troopers found no significant change in stops, citations, arrests, use of force or complaints once results were corrected for multiple comparisons.
Waze police alerts and speeding — speeding fell from over half of drivers to around a quarter once a police alert appeared on the app, but the effect looks short-term and local.
Operation Cookie Monster and an untargeted hacker forum — a major police takedown of one illicit market rippled out to a different forum, bringing a temporary dip in activity alongside signs of displacement.
What the public looks for in a police chief — professional standards mattered most to respondents, with commitment to evidence-based policing tied for second.
Missing data and predictive policing — a Chicago stress test shows that gaps in crime records can shift who a forecasting model would over- or under-police, depending on where the gaps fall.
Recognising vulnerability at first contact — a rapid evidence assessment finds the evidence on vulnerability assessment tools is thin and mostly observational, with no randomised trials.
Early attrition among degree-apprenticeship officers — in one English force, around one in five PCDA recruits and almost one in four DHEP recruits resigned within five years.
Does AI generate a civilizing or de-policing effect? Testing the impact of AI-based body worn camera analytics in a large state police agency
Authors: White, Watts & Malm
Study design: Randomised controlled trial
Setting: United States
Summary: 314 Arizona highway patrol troopers were randomly assigned to have access to AI analysis of their body-worn camera footage, or not, for six months. After correcting for multiple comparisons, there were no significant changes in stops, citations, arrests, use of force or complaints. A rise of about one percentage point in highly professional encounters was not statistically significant. The authors found no sign of either a "civilizing" or a "de-policing" effect.
You'll be interested if: you're weighing up AI tools to review body-worn video. This is a rare randomised test of what they do to officer behaviour, and it suggests the payoff may lie in efficiency rather than in changing how officers act.
Waze to beat the system: a real-world evaluation of crowdsourced enforcement alerts on driver speed
Authors: Anderson
Study design: Observational before-and-after study (no comparison group)
Setting: Australia
Summary: Researchers used a laser device to record more than 9,000 vehicle speeds at 33 sites in South East Queensland, before and after a police alert appeared on the Waze app. Speeding fell from over half of drivers to about a quarter once an alert appeared, with the smallest effect in active school zones. The author suggests the benefit is short-term and local, and that alerts may encourage speeding when none is showing.
You'll be interested if: you work in roads policing or road safety. It's a useful look at how drivers react to crowdsourced warnings, though with no comparison group it can't tell us about longer-term effects.
The ripple effect of Operation Cookie Monster: the impact of an illicit market takedown on an untargeted hacker forum
Authors: Stibbe, Ruiter & Moneva
Study design: Natural experiment combined with a field experiment
Setting: International (online)
Summary: This study asks what happened on a hacker forum that wasn't targeted when law enforcement took down the Genesis Market in 2023. Activity fell among experienced non-paying users and for most types of stolen credential, but rose among paying users and for some account types. Administrators also moderated more selectively afterwards. The authors conclude the takedown brought a temporary spread of benefits alongside displacement, and that offenders' adaptation limits long-term impact.
You'll be interested if: you work in cybercrime, serious and organised crime or intelligence. It's a rare piece of evidence on what a takedown does beyond the platform it hits.
What signals police leadership quality to the public? A look at evidence-based policing and professional standards
Authors: Abner & Mee
Study design: Mixed methods: choice experiment (conjoint) with thematic analysis of open-ended responses
Setting: United States
Summary: 303 members of the public chose between hypothetical applicants for a police chief's job whose credentials varied. Emphasis on professional standards was the most influential factor; the size of the population served and a commitment to evidence-based policing tied for second. Respondents preferred applicants with external credentials over those who had helped implement these practices without such recognition. It measures stated preferences about hypothetical candidates, not real appointments.
You'll be interested if: you champion evidence-based policing or are involved in chief officer selection. It suggests external recognition of EBP commitment is something the public notices, at least in the US.
How crime data incompleteness shapes ethnic disparities in predictive policing: evidence from Chicago
Authors: Jiang, Liberatore & Preece
Study design: Data audit using a simulated "stress test" of a predictive policing model
Setting: United States
Summary: Using Chicago crime records from 2022 to 2024 and a crime forecasting model, the authors deleted records from selected neighbourhoods to see how missing data changes who the model would over- or under-police. Removing records from a neighbourhood where one ethnic group is the majority generally lowered predicted over-policing for that group and raised predicted under-policing. In low-diversity areas, more missing data increased bias against Black residents. This is a simulation, not a live deployment.
You'll be interested if: you work with crime analysis, data quality or the oversight of predictive tools. It's a reminder to audit what's missing from your data, and for whom, before relying on a forecast.
Recognizing vulnerability at first contact: a Rapid Evidence Assessment and agenda for police practice
Authors: Hutton
Study design: Rapid evidence assessment (12 studies, 2018 to August 2025)
Setting: International
Summary: This review looks at how police recognise and assess vulnerability at first contact. The evidence on whether assessments work is limited and uneven: most studies are observational, quasi-experiments are rare and there were no randomised trials. Risk labels were applied inconsistently and did not reliably predict later harm. Better recognition was linked to professional judgement and scenario-based, trauma-informed training. The author calls for stronger evaluations of what police do in practice.
You'll be interested if: you lead on vulnerability, safeguarding or frontline training. It's an honest look at how little we know about whether vulnerability tools work, with a clear agenda for testing them.
Early service attrition among Police Constable Degree Apprenticeship (PCDA) and Degree-Holder Entry Programme (DHEP) officers: evidence from an English force, 2020–25
Authors: Lamb & Tonks
Study design: Mixed-methods analysis of force records
Setting: England
Summary: The study examined all 214 voluntary resignations among 1,016 PCDA and DHEP recruits in one English force between January 2020 and August 2025. Five-year attrition was 19.5% for PCDA and 23.7% for DHEP, against 6.0% nationally across all ranks. Operational unsuitability was the most commonly recorded reason (29%); academic failure was rare (3.7%). Resignations peaked in the first two years, especially after independent patrol status. It describes one force and doesn't test what reduces
attrition.
You'll be interested if: you lead on recruitment, retention or student officer training. It points to the move onto independent patrol, rather than the academic side, as a pressure point in the first two years.
And that's October 2026!
Thanks for reading. If you've found a study you think should feature next month, send it our way and we will take a look.
