全球智库信息集成服务系统

The Program Did Not Generate a Statistically Significant Reduction in Property Crime

• There were few participating districts over a limited duration, thus providing low statistical power to detect any true effect of the program.
• The null effect on crime may be due to treatment heterogeneity over time and across districts.
• The model's predictions may be insufficient to generate additional crime reductions over traditional crime mapping on their own.

Police Officers Across Districts Perceived Benefits from the Program

• Some prevention approaches reportedly improved community relations.
• Predictions provided additional information and further assisted commanders and officers in making day-to-day targeting decisions.

The Program May Have Been a More Efficient Use of Resources

• Treatment groups spent 6 percent to 10 percent less than the control groups to achieve similar levels of crime reduction.
• Using in-house crime analytics, program costs were likely offset by reductions in officer overtime hours.

Even though there is a growing interest in predictive policing, to date there have been few, if any, formal evaluations of these programs. This report documents an assessment of a predictive policing effort in Shreveport, Louisiana, in 2012, which was conducted to evaluate the crime reduction effects of policing guided by statistical predictions. RAND researchers led multiple interviews and focus groups with the Shreveport Police Department throughout the course of the trial to document the implementation of the statistical predictive and prevention models. In addition to a basic assessment of the process, the report shows the crime impacts and costs directly attributable to the strategy. It is hoped that this will provide a fuller picture for police departments considering if and how a predictive policing strategy should be adopted.

There was no statistically significant change in property crime in the experimental districts that applied the predictive models compared with the control districts; therefore, overall, the intervention was deemed to have no effect. There are both statistical and substantive possibilities to explain this null effect. In addition, it is likely that the predictive policing program did not cost any more than the status quo.

• Chapter One

  Introduction

• Chapter Two

  PILOT Process Evaluation

• Chapter Three

  PILOT Impact Evaluation

• Chapter Four

  PILOT Cost Evaluation

• Chapter Five

  Conclusions

• Appendix

  Technical Details