On the use of police officers in randomized experiments: some lessons from the Milwaukee domestic violence experiment