A Theory-Driven Algorithm for Real-Time Crime Hot Spot Forecasting

• Author: O SooHyun,YongJei Lee,John E. Eck
• Date: 01 June 2020
• DOI: 10.1177/1098611119887809
• Published date: 01 June 2020
• Subject Matter: Articles

Article

A Theory-Driven

Algorithm for

Real-Time Crime

Hot Spot Forecasting

YongJei Lee

1

, SooHyun O

2

, and

John E. Eck

3

Abstract

Real-time crime hot spot forecasting presents challenges to policing. There is a high

volume of hot spot misclassifications and a lack of theoretical support for forecasting

algorithms, especially in disciplines outside the fields of criminology and criminal

justice. Transparency is particularly important as most hot spot forecasting models

do not provide their underlying mechanisms. To address these challenges, we oper-

ationalize two different theories in our algorithm to forecast crime hot spots over

Portland and Cincinnati. First, we use a population heterogeneity framework to find

places that are consistent hot spots. Second, we use a state dependence model of

the number of crimes in the time periods prior to the predicted month. This algo-

rithm is implemented in Excel, making it extremely simple to apply and completely

transparent. Our forecasting models show high accuracy and high efficiency in hot

spot forecasting in both Portland and Cincinnati context. We suggest previously

developed hot spot forecasting models need to be reconsidered.

Keywords

crime hot spot, forecasting, population heterogeneity, state dependence, Excel

1

School of Public Affairs, University of Colorado, Colorado Springs, CO, USA

2

School of Criminology, Criminal Justice and Strategic Studies, Tarleton State University, RELLIS Campus,

TX, USA

3

School of Criminal Justice, University of Cincinnati, Cincinnati, OH, USA

Corresponding Author:

YongJei Lee, School of Public Affairs, University of Colorado, Colorado Springs, CO 80918, USA.

Email: ylee@uccs.edu

Police Quarterly

2020, Vol. 23(2) 174–201

!The Author(s) 2019

Article reuse guidelines:

sagepub.com/journals-permissions

DOI: 10.1177/1098611119887809

journals.sagepub.com/home/pqx

Introduction

With the advent of computer mapping and geographic information systems,

real-time crime hot spot forecasting has become important to policing

(Bowers, Johnson, & Pease, 2004; Gorr & Olligschlaeger, 2002; Groff &

La Vigne, 2002). Nevertheless, crime hot spot forecasting presents several

challenges. There is a high volume of crime hot spot misclassifications and a

lack of theoretical support for existing forecasting algorithms, especially in dis-

ciplines outside the fields of criminology and criminal justice (Gorr & Lee, 2012,

2015). In addition, many algorithms are complex and proprietary. Thus, they

require expensive contracts with private vendors and lack transparency

(Ferguson, 2012).

Transparency is particularly important when police use of stop-question-frisk

tactics in hot spots: A tactic linked to both racially disparate treatment of non-

Whites and excessive interference in the daily lives of people who are not

involved in crime (Sullivan & O’Keeffe, 2017). Recently, New York, Chicago,

and Los Angeles police departments were sued for not releasing information

about the algorithms used by their predictive policing programs (Collins, 2018;

Rieland, 2018). Therefore, transparency is just as important a criterion for eval-

uating hot spot forecasting models as predictive efficiency and accuracy.

To respond to these challenges, we create a forecasting algorithm based on

two models suggested by scholars in the field of predictive policing (Bowers

et al., 2004; Johnson, 2008). We use a population heterogeneity model to find

places that are consistently experiencing crimes in the forecasted month. This

narrows our focus to places with consistently high levels of crime. Second, we

apply a state dependence model to account for short-term elevations in risk at

these places. We implement this combination of models in Microsoft Excel,

making it extremely simple to apply and completely transparent. It does not

need highly specialized expertise to implement but can be modified by agencies

as needed. Importantly, people outside of policing can examine how it works,

should they be concerned that it produces undesired outcomes. In short, it is

completely transparent.

Our forecasting models show high efficiency and accuracy in hot spot fore-

casting relative to existing hot spot forecasting models. These results demon-

strate how basic theories could lead researchers and practitioners to build a

sound algorithm for hot spot forecasting. After we review a brief history and

utility of the past crime hot spot forecasting models in the following section, we

describe the two theoretical frameworks, both population heterogeneity and

state dependence in more detail. We also explain how we operationalize both

models in the methods section. Followed by the results section, we finally discuss

the implications of this simple, theory-driven forecasting method for policing

practices and its limitations as well.

Lee et al. 175