Beyond the Training Room: An RCT of Virtual Reality De-escalation Training on Field Behavioral Outcomes and Citizen Complaints

Virtual reality (VR) is increasingly used in police de-escalation training, but evaluations have rarely followed officers out of the training room. We conducted a randomized controlled trial in a municipal public security bureau in central China, assigning 480 officers who handled 110 dispatches at local police stations to VR training or to traditional scenario-based training matched in content, duration, learning objectives, and debriefing. Over the 12 months after training, blind coders rated de-escalation behavior in 5,259 body-worn camera recordings of real encounters, and research staff extracted documented use of force and public complaints. In the modified intention-to-treat sample ( n = 457), with standard errors clustered by training session, VR officers displayed more de-escalation behavior than traditional officers ( d = 0.38, 95% CI [0.18, 0.59]), with differences in verbal, relational, and tactical behavior. Their documented use-of-force rate was 20% lower after adjustment for workload and prior force (IRR = 0.80, 95% CI [0.66, 0.97]), a smaller and less precisely estimated difference than the behavioral effect. Fewer VR officers received a public complaint (16.9% vs. 21.2%), but this difference was not statistically reliable (OR = 0.76, 95% CI [0.47, 1.22]). Both groups used slightly fewer de-escalation behaviors as time since training increased, but the between-group difference was similar across four three-month intervals, and the behavioral result held across all sensitivity analyses. The findings show that a content-matched VR program can produce measurable transfer to real police work over a full year, and that the estimated effect becomes smaller and less certain as outcomes move further from the trained behavior, from coded conduct to documented force to public complaints.