It would be violent crime trend correlation with de-escalation training. Or even with complaints of police aggression.

Even more useful would be to separate out assault + battery where the victim is random vs. non-random, IE domestic or gang.