Development of Crime Forecasting and Mapping Systems for Use by Police

• NCJ Number: 211973
• Author(s): Jacqueline Cohen; Wilpen L. Gorr
• Date Published: February 2005
• Length: 155 pages
• Grant Number(s): 2001-IJ-CX-0018
• Agencies: NIJ-Sponsored
• Publication Type: Report (Study/Research)

Annotation

This report presents the results of research on the new field of crime forecasting and represents work on the second of two National Institute of Justice (NIJ) grants awarded for crime forecasting research.

Abstract

The research involved with the first NIJ grant established that crime forecasting was feasible using simple time series methods. The current research extends work from the first grant using new data and tests three advanced time series methods in order to improve the accuracy of crime forecasting. The three methods tested are the so-called naïve methods, univariate time series methods, and leading indicators models. The final goal of the research has been to develop crime forecasting capacity for police in support of the tactical deployment of resources. Data included 6 million offense reports and CAD calls for 24 crime types from Pittsburgh, PA, and Rochester, NY, which were calculated into monthly time series data for the period 1990 through 2001 across 5 geographies. Section 4 provides the details of the data collection and processing work, which were extensive and involved the use of geography to aggregate and forecast crime levels. The authors developed an experimental design involving a state-of-the-art rolling horizon forecast experiment. Two approaches for assessing forecast accuracy, the traditional average forecast error criterion and an innovative decision rule criterion, are described. The results demonstrate the potential of crime forecasting to support short-term decisionmaking by police. The best method for producing crime forecasts was determined to be exponential smoothing with seasonality estimated with pooled citywide data. Worst methods for crime forecasting were the commonly used naïve approach and the leading indicator models. The leading indicator models, however, were best for forecasting large crime changes. The accuracy attained was sufficient to support car beat-level crime forecasting as well as high-volume individual crime types at the precinct level. Recommendations are offered and include the advice of conceptualizing police decisionmaking and crime analysis into macro, meso, and micro levels. Figures, tables, references, appendixes