I help people use data to reduce crime. Associate Professor at @UCLCrimeScience. Former police officer.
Nottingham and London, UK
Joined April 2009
- Tweets21.1K
- Following1.3K
- Followers6K
- Likes32.3K
When a temporary speed trap (🇦🇺) appears on phone navigation apps (Waze, Google Maps, etc.), average vehicle speed on that road drops by about 4km/h.
One for academics still teaching students that offenders don't make rational choices, I guess. 😉
doi.org/10.1007/s11292-026-0…
Matt Ashby retweeted
I’ve heard Journal of Quantitative Criminology will soon require opens data/code…which will make it the only journal in the entire field to do so. In late 2026.
We should be embarrassed! Complete abrogation of basic scientific responsibility for decades. And seemingly no urgency from the editors of Criminology, Justice Quarterly, Criminology & Public Policy, or Journal of Experimental Criminology.
There was an effort to get there from the last two editors of criminology, but the new editors haven’t followed through.
Meanwhile, what year do you think the first preregistered RCT was published in the fields flagship journal Criminology? If you guessed 2026 then you win (proud to have authored that one 😁).
Chatting with a friend about ongoing efforts in Economics to use AI to replicate every econ paper ever published. This is scary for individual researchers, of course (everyone lives in fear of a coding error), but clearly good for science.
I'd love to see this applied to Criminology, but those journals still don't require authors to post their data or code. 🙄
Embedding mental-health clinicians in police control rooms (🇺🇸) was associated with a small drop (about 7%) in police deployments to mental-health calls. Seems like this would have been a good opportunity for a randomised controlled trial, though.
doi.org/10.1080/0735648X.202…
Repeat victimisation happens to police officers, too, especially when it comes to violent crime. A useful reminder in @PolicingInsight of the importance of leveraging concentrations of victimisation risk when trying to prevent officers being assaulted.
policinginsight.com/feature/…
Of 812 articles in the British Journal of Criminology (2014–2024), only 0.2% (two articles in 11 years!) used experimental/quasi-experimental methods.
(Quasi-)experiments aren't good for everything, but is UK crim research really using the right methods?
doi.org/10.1007/s11292-026-0…
ALT Dumbbell chart comparing the percentage of articles using quasi-experimental (QE, red) and Big Five (dark blue) research designs across 15 criminology journals. Big Five percentages are higher in almost every journal. The largest difference is in the Journal of Experimental Criminology (34% Big Five versus 9% QE), followed by the Journal of Quantitative Criminology (10% versus 7%) and Criminology and Public Policy (8% versus 5%). In most other journals, both percentages are below 5%, with differences generally under 3 percentage points. The British Journal of Criminology and Legal and Criminological Psychology have approximately 0% for both categories.
Mark Rowley hits the nail on the head:
“If we're serious about policing London effectively … Why wouldn't we want people who understand and can connect with every part of one of the world's most globally connected cities?”
lnkd.in/p/e9XTGmvU
Replacing sodium street lights with LEDs (🇺🇸) led to substantial reductions in night-time property and violent crime outdoors, but the effect only lasted for six months before crime reverted to previous levels.
doi.org/10.1007/s10940-026-0…
A randomised trial (🇺🇸) of sending police drones to calls at the same time as dispatching officers found the drones decreased both how quickly police arrived at calls and how quickly calls were resolved.
(Not yet peer reviewed)
doi.org/10.21428/cb6ab371.97…
An excellent blog post on some of the challenges of using generative AI tools in policing. As so often, it’s the non-technical issues that are the hardest to work through!
Definitely worth your time:
stephenrussell996702.substac…
Incorporating street networks into geographic profiles of serial homicides produced results that were no more accurate, despite profiles using street networks requiring more work and taking longer.
doi.org/10.21428/cb6ab371.c0…
This paper finds that the chances of a person with a criminal conviction (🇺🇸) being charged with a further crime drops rapidly every year afterward conviction, such that after 10 years they're no more likely to reoffend than the population at large.
doi.org/10.1111/1745-9125.70…
This study finds police officers (🇨🇦) perceive dealing with an armed robbery or violent domestic assault in progress as much more dangerous than driving to an emergency call, even though vehicle crashes are the top killer of cops (in 🇨🇦 and elsewhere).
doi.org/10.1080/10439463.202…
More undocumented immigrants moving into a neighbourhood (🇺🇸) is associated with a small *decrease* in violent crime rates and no change in property crime. More legal immigrants moving to an area isn't associated with changes in violent or property crime.
doi.org/10.1080/07352166.202…
This paper usefully categorises the ever-expanding use of AI in crime based on whether AI is an enabler or the principal executor of the crime, and whether the behaviour is human-directed or autonomous.
doi.org/10.21428/cb6ab371.ec…
A hopefully useful (free) opportunity to hear from the statisticians who create our crime statistics on what they're working on.
eventbrite.co.uk/e/annual-cr…
This huge replication (🇺🇸) of the 'Koper curve' on how long hotspot patrols should last reinforces the original findings: patrols are effective once cops are in a hotspot for 15 mins, after which staying in that area has no extra crime-prevention effect.
doi.org/10.1111/1745-9133.70…
Useful estimates in this paper of how long it takes for crimes of different types of be reported and recorded (🇺🇸), i.e. how long it takes for crime data to be 'stable' enough for research and analysis. For most types, >90% are reported within 14 days.
doi.org/10.1080/07418825.202…
This paper finds the 'iron rule' that most crimes happen in a small number of places applies equally in an African city (Abuja, 🇳🇬) and when measured using data from a door-to-door survey rather than police data (which is limited by under-reporting).
doi.org/10.1057/s41300-026-0…
