Self-defense: what the data actually says
“Ninety percent of fights end up on the ground.” It's the line that sold jiu-jitsu to the world, it comes from a real study, and it doesn't say what everyone repeats. Here are the numbers, the sources, and the limits of each.
BJJ Square · Updated on 09/21/2026
The industry's most quoted figure, and what it really said
The source is an analysis of Los Angeles Police Department data for 1988, compiled by Sergeant Greg Dossey. Across 316,525 arrests there were 5,617 use-of-force reports (1.7%), of which 2,031 involved an actual physical altercation.
Two different numbers come out of it, and the confusion comes from stacking them on top of each other:
- 95%: the share of altercations that followed five recurring patterns already known to trainers. It is not the share of fights that ended on the ground.
- 62%: the share of altercations that ended with officer and subject on the ground, the officer applying a joint lock and handcuffing.
Sixty-two isn't ninety, and more importantly it describes police officers who have to handcuff someone: their goal is to take the subject down and pin them. A civilian who wants to leave has the opposite goal. The figure is still interesting — it just doesn't transfer as is.
Note A second measurement, a 2003 Calibre Press survey of about 1,400 officers, reports that 52% had takedown attempts made on them, that 60% of those attempts succeeded, and that 77% of those taken down stayed in close-quarters grappling.
The Marietta case: real numbers, weak design
The police department of Marietta, Georgia made BJJ mandatory for new hires and published its 2020 results, comparing officers in the programme with those who weren't:
- −48% injuries to officers in use-of-force incidents
- −53% injuries to the person being arrested when force was used
- −23% taser deployments
⚠️ This is not a randomised study. Officers in the programme signed up for it: they may be younger, fitter, more motivated. These are figures published by a city administration, not a journal, and they read as a strong indication — not as proof.
The one serious experiment on self-defense training
On whether self-defense training works there is a randomised trial published in the New England Journal of Medicine in 2015 (Senn and colleagues), run on 899 women students at three Canadian universities. The programme, called EAAA, runs twelve hours across four parts: assessing risk, getting past the hesitation to react, acting — and sexual health in relationships.
One year later, the share who had been raped was 5.2% in the trained group against 9.8% in the control group: roughly half.
The detail almost nobody quotes: of those twelve hours, only two were physical technique. The rest was recognising risk early and overcoming the psychological resistance to fighting back against someone you know. That's why this page exists: the part that works best is not the part that sells best.
What we don't know
- No study exists comparing BJJ practitioners and non-practitioners on the outcomes of assaults in civilian life. Anyone quoting one made it up.
- Police data doesn't transfer automatically. An officer wants to pin; a civilian usually wants to leave.
- No number tells you how much training is enough. There is no measured threshold of months or classes.
Sources for this page: the analysis of Los Angeles police data published in the *Journal of Non-lethal Combatives* (2007), the 2003 Calibre Press survey, the City of Marietta release of 8 February 2021, and Senn et al. in the *New England Journal of Medicine* (2015). The overview is here.
Frequently asked questions
- Is it true that 90% of fights end up on the ground?
- No, and the original figure never said that. In the 1988 Los Angeles police data, 62% of altercations ended on the ground — and those were officers whose goal was to pin and handcuff.
- Is there evidence BJJ reduces injuries?
- There is police-department data, such as Marietta's, showing sizeable reductions among officers in the programme. These aren't randomised studies: participants self-select.
- Does a self-defense course actually work?
- A structured programme does, and it's documented by a 2015 randomised trial: it roughly halved the risk over a year. But the physical part was only two hours out of twelve.
Share this guide
We are on Instagram
The gyms on the map, the history of jiu-jitsu and the stories behind the names of the techniques.
Follow @bjjsquareRead next
- BJJ and self-defense: what actually works and what doesn'tBrazilian jiu-jitsu was born calling itself a self-defense method, and it's also the subject gym marketing shouts about the loudest. This page separates what the training actually gives you from what people selling courses promise.
- Injuries in BJJ: what the data actually saysPer hour spent on the mat, competition is ten times riskier than training. And yet nine injuries out of ten happen in training, and the reason is simple.