Early Warning Systems for Police Misconduct
We found that while the distribution of each indicator, such as personnel complaints, resembles that reported in previous research – that is, a small number of officers accounted for a disproportionately large fraction of the events – it was not, for the most part, the same officers year after year. Problem officers, the primary target of early warning systems, are very few in number. A larger number of officers display symptoms of problem behavior for limited periods of time – a year or two – and then not again, or only once again. Thus problem behavior that is captured by the indicators on which early warning system selection criteria rely is to a large degree evanescent and hence unpredictable. We tested various selection mechanisms; none performs very well. All of them yield fairly large numbers of false negatives; many of them also yield large numbers of false positives. Some perform better than others, however. A very simple time-and-numbers system, based only on complaints, with a threshold of 4 in 12 months, is arguably the best of those we examined. It is simple, it is straightforward to implement, it selects a small and thus manageable number of officers for intervention, a large fraction of which (albeit less than half) were by the definition adopted for this inquiry true positives. The need for systematic selection criteria of some kind is clear. Our evaluation of the EWS intervention – training in police-citizen interaction – suggests that it has not had the intended impacts in classes since the second or third, and the reason may be the composition of the trainees, whose selection is not based on explicit criteria. We find no reason to believe that the content of the training is deficient in any way, but rather that the needs of trainees have not been properly matched with the objectives of the training.
