Chicago data supports effectiveness of predictive policing. But maybe not how you expect:
[T]he number of complaints an officer receives in a certain year predicts whether and how many complaints he or she will have in the following year.1 Over multiple years, the signal becomes even stronger. Officers with a baseline history of one or two complaints in 2011-13 have a 30 percent to 37 percent chance of receiving a complaint in the following two years.2 But repeaters — those with 15 or 20 incidents in the first part of the data set — are almost certain to have a complaint against them in 2014-15.
… Even after controlling for neighborhood, however, individual officers with more complaints in 2011-13 remained more likely to have complaints filed against them in 2014-15.
… [C]omplaints were not only predictive of the number and type of future complaints — they also forecast whether the department would determine misconduct. Officers with 10 or more complaints in early years of the data set were about six times more likely to have a complaint from the last two years sustained against them.
… For all the complexity of policing, there is a clear signal in the data of who the bad actors are and, to a lesser extent, whether they are going to commit misconduct.