Applying Artificial Intelligence to Person-Based Policing Practices, 2019-2023 (ICPSR 39074)

Time Period(s) View help for Time Period(s)

Date of Collection View help for Date of Collection

Data Collection Notes View help for Data Collection Notes

Hide

Study Purpose View help for Study Purpose

The purpose of this study was to improve person-based focused deterrence policing practices using artificial intelligence, specifically:

1. To use artificial intelligence methods to better understand the behavioral patterns of high-risk individuals;
2. To better understand person-based focused deterrence policing strategy and its effectiveness; and
3. To assess the use of artificial intelligence in person-based policing strategies and its impacts on crime and criminal investigation.

Study Design View help for Study Design

Between 2019 and 2020, the research team developed an artificial intelligence software tool for crime analysis and risk evaluation (CARE), CAREsim, based on intelligence and data supplied by the Hampton (Virginia) Police Department (HPD). The CAREsim simulation environment uses an agent-based model (individual agents can make decisions and adapt actions independently), incorporating data from various sources such as the Census Bureau for creating the population based on demographic distributions, the National Oceanic and Atmospheric Administration (NOAA) for weather data, local commute and transportation data for routing, and crime data for hotspot calculations. Training and a short-term pilot experiment took place in January 2021, and the proposed study was conducted between February and December 2021.

An experimental design was used to evaluate CAREsim. At the start of each month, HPD crime intelligence analysts generated a list of high-risk individuals (n=20), who would be analyzed for 1 year. An individual was characterized as high-risk for criminal activity if they were unemployed adults with prior crime records and a strong tendency to roam in the city with constant, high motivation for seeking crime opportunities. Individuals were randomly assigned to the control condition (using HPD's current method for identifying and providing information to officers about high-risk individuals) or the treatment group (using CAREsim to simulate individual behavior to predict possible behavioral patterns in real-time). Outcomes were tested for both groups:

• If a high-risk individual was identified as a suspect
• If a high-risk individual was later determined to be not high-risk
• Number of citations per individual
• Number of arrests
• Number of victims per individual
• Amount of time spent per case
• Number of cases cleared per individual
• Number of connections or associates identified per individual
• Type and amount of evidence

The original experimental design used official crime and arrest data from the HPD to evaluate CAREsim. A total of 197 cases (control=100, experiment=97) were studied during this period. However, HPD terminated their participation in the study in January 2022. As a result, the team could not assess the impacts of CAREsim on criminal investigation or rely on the HPD crime data. Instead, they developed a simulation environment for predictive analysis, modeling the person-based strategy within the Hampton, VA jurisdiction. Initial information for simulated high-risk individuals in both control (n=120) and experimental (n=120) groups were based on the distributions of features from available high-risk individuals reported by HPD. The experiment was simulated for 23 months. Five different modes of focused deterrence patrol strategies were tested in the experiment:

1. Baseline group: 16 officers, random patrol
2. Experimental group with 8 added officers, focused patrol
3. Control group with 8 added officers, random patrol
4. Experimental group with 8 added undercover officers, focused patrol
5. Control group with 8 added undercover officers, random patrol

Sample View help for Sample

The criteria used to select potential high-risk individuals evolved to meet the dynamic needs of the Hampton Police Department (HPD). For February and March 2021, selected individuals met at least two of five criteria: history of gang affiliation, violent criminal history, history of illegal narcotics offenses, juvenile criminal history, and criminal history involving firearm offenses, gang activity, or illegal narcotics. Starting in April 2021, selected individuals met two of three criteria: charged with 2 or more crimes, a violent crime, or a property crime, all within in the past 10 years. These criteria remained in effect until January 2022.

Time Method View help for Time Method

Universe View help for Universe

Simulated individuals determined to be high-risk for criminal activity.

Unit(s) of Observation View help for Unit(s) of Observation

Data Source View help for Data Source

Data Type(s) View help for Data Type(s)

Description of Variables View help for Description of Variables

Simulated high-risk individual data (DS1) is at the individual level. Items include demographics (age, sex, race, Hispanic ethnicity, education level, employment level, poverty status, gang affiliation, if the individual used drugs), geographic coordinates for favorite areas, "home" address, and police field stops. An estimated risk value was calculated for each individual.

Simulated crime data (DS2) is at the incident level, including prior crime type (violent, property, or drug crime) and geographic coordinates for crime location.

Hide

Original Release Date View help for Original Release Date

2024-09-26

Version History View help for Version History

Hide