Predictive policing through risk assessment

Melissa Hamilton

Introduction

The emergence of big data, advancements in technologies and interest in experimentation have coalesced to usher in a new wave of crime control in the form of predictive policing. Intended as a progressive initiative, predictive policing refers to “the application of analytical techniques—particularly quantitative techniques—to identify likely targets for police intervention and prevent crime or solve past crimes by making statistical predictions” (Perry et al., 2013: xiii). One author finds that definition too broad and has suggested an alternative.

|Predictive policing means] the use of historical data to create a forecast of areas of criminality or crime hot spots, or high-risk offender characteristic profiles that will be one component of police resource allocation decisions. The resources will be allocated with the expectation that, with targeted deployment, criminal activity can be prevented, reduced, or disrupted.

(Ratcliffe, 2019: 349)

This reform requires “datification” in which social activity and human behaviours are turned into data points to be tracked, analysed and managed (Dencik et al., 2018: 7). Policing through data fits under the umbrella of intelligence-led policing, whereby objective information guides decisions in police operations (Vestby and Vestby, 2019).

Three main types of predictive policing exist: forecasting hot spots, in terms of where crime is likely to flourish; predicting victimization; and predicting who is likely to be offenders. This chapter is mainly concerned with the latter, predicting ‘hot people’. Predictive policing in the form of identifying likely individuals who will offend is a more recent endeavour than the identification of hot spots (Babuta, Oswald and Rinik, 2018). Much less is known about predicting who will offend. Further, far more policing resources are concerned with criminals than predicting who will be victims.

Certainly, individual risk prediction occurs in other areas of criminal justice decisions (e.g. pretrial bail, sentencing or parole). A major difference is that those other decision points are primarily about managing known offenders according to their risk level. At a high gradient of analysis, the ideal is to efficiently harness resources by diverting low risk offenders while saving precious cell space and intensive rehabilitative programming for high-risk offenders. In policing, though, individual risk relates to predictions of both known and unknown (potential) offenders. Plus, predictive policing concerns properly managing offenders, but it is also strongly about managing policing personnel resources (Shapiro, 2017). Resource management may include insight into where and when to deploy officers, whether to send specialised units (e.g. tactical teams, bomb squads, hostage negotiators), what level of response is reasonable in terms of force, and the type and amount of intervention that is most suitable to the predicted risk.

This chapter will review the expected advantages of offender-based predictive policing. Describing a selection of predictive tools in use will provide context for how recent efforts are taking shape and the variety of ways that policing agencies are experimenting with predictive technologies. Despite the potential benefits of predictive policing models, awareness of potential critical issues with them is important in order to manage unwanted consequences. One such issue is that, despite using scientific methods as a foundation for predictive policing, biases can still plague these tools. Various sources for how different biases may become imbedded into predictive policing models for assessing individual risk are discussed. Finally, the chapter offers some insights into the future, both near and far, of offender-based predictive policing. Overall, predictive policing will likely evolve quite quickly by drawing on emerging offerings from big data in social media and social network analysis to revolutionise crime prevention and pre-emption.

Projected benefits of predictive policing with individual risk

Predictive policing is a more recent evolution of a broader proactive policing movement that began in the 1980s and 1990s as a novel strategy to stem the then-rising crime rates (Weisburd et al., 2019).The theory was that preemptive, knowledge-based policing may yield greater gains than what was being achieved by a traditional, reactionary policing style (Bennett Moses and Chan, 2018). Manning (2018) contends that while the structure of policing in democratic societies has remained largely unchanged in the last century, policing practices have undergone significant shifts in recent years due to employing algorithmic technologies. In more recent years with extreme budget cuts, algorithm-led practices are also a response to such austerity measures (Ratcliffe and Kikuchi, 2019).

An algorithm simply provides “automated instructions to process data and produce outputs” (Dencik et al., 2018: 7). An algorithm, then, provides an equation that drives the predictions. These algorithms are generally developed by researchers who study large datasets to determine which factors are statistically associated with offending. For example, common risk factors for future offending are young age, being male, having a criminal history and associating with criminal friends. Overall, algorithmic predictions operate as evidence- based practices in relying upon scientific findings. More simply, predictive policing is criminal profiling using data and computer technology (Selbst, 2017). Importantly, models may need updating as follow-on studies may show that predictive factors change as environmental, cultural and individual links to criminal offending shift over time or place.

Predictive policing offers multiple advantages. Predictive policing methods promise to provide value to the public:

With a more precise focus, there is less inadvertent collateral damage to civilians unconnected to criminality, we benefit from improved efficiency within our criminal justice services, there is greater objectivity, and there may even be increased public trust and law enforcement legitimacy when people see the police are focused on the right people at the right places.

(Ratcliffe, 2019: 348)

Advance knowledge allows law enforcement officials to optimise their limited resources toward forecasting crimes (Brantingham, 2018) and thus permit the agency to operate more effectively (Perry et al., 2013).Where police actions are perceived as more objective in relying upon scientific study, these decisions may thereby be more defensible to both citizens and to courts.

The general methodology for creating an algorithmic predictive tool can be summarily explained. The developers of a tool statistically analyse historical data to isolate those factors that predict (e.g. correlate with) their outcome of interest (e.g. being involved in gun violence, homicide, or committing any new offense). Developers select significant factors and weight them, as some factors are more highly predictive than others. The selection and weighting create the resulting algorithmic equation on which the predictive model is based.

The algorithms underlying predictive tools often identify personal and social connections within individuals and across people that relate to criminal offending (Babuta, Oswald and Rinik, 2018). Algorithmic processing can locate patterns in social behaviour that humans are not capable of cognitively replicating. This is the basis of the pre-emptive ideal. “[T]he promise of the pattern is thus to serve as a basis for the extrapolation of possible criminal futures and to render those futures actionable for prevention programmes” (Kaufmann, Egbert and Leese, 2019: 674). In addition, predictive models single out which sociodemographic characteristics are related to committing crimes (Meijer and Wessels, 2019).

The goal of drawing on algorithmic tools to identify riskier individuals is justified as a large number of crimes are committed by a small number of people who are repeat offenders. Estimates are that, overall, six per cent of the population commit sixty per cent of crimes (Ratcliffe and Kikuchi, 2019). People often develop routines in their day-to-day activities, and criminal engagement may simply become a part of certain individuals’ lifestyles (Perry et al., 2013). Thus, the ability' to isolate the riskiest and thus focus resources on them serves to effectively protect the public from future harms. This focus also reduces unnecessary contacts with civilians by' police that less informed and thus poorly targeted missions cause, which otherwise may lead to questioning police legitimacy' and expertise (Papachristos and Sierra-Arevalo, 2018).

Another advantage of the predictive algorithm is its ability' to reduce human biases (Babuta, Oswald and Rinik, 2018). Human decision-making is replete with implicit and explicit biases. Humans can act out of prejudice. Their decisions can be affected by being tired, angry or distracted. Unlike a person, though, an algorithm cannot harbour animus or be diverted by' such distractions. Humans, as well, may not cognitively be capable of processing as many data points that a computer algorithm can efficiently handle. To illustrate an advantage for an algorithm, an initiative in Philadelphia aimed to triage repeat gun offenders using an algorithm that calculates a harm score based on past offenses with a time decay' adjustment. Results indicated that this algorithmic method identified dangerous gun offenders better than human anafysts (Ratcliffe and Kikuchi, 2019).

A side benefit of the automation offered by an algorithmic tool is the recordkeeping required (i.e. to score the inputs) to retrieve an output. These records mean that more information will be available to officers and serve to bolster big data for future studies to improve the algorithmic models. As some of the real-world examples of predictive policing programs that follow will attest, an intended consequence of the predictive policing turn is to improve communications about relevant facts and circumstances among officers within a police department and, in some cases, across agencies. Further, the recordkeeping that algorithms necessitate may bring greater accountability of decisions informed by predictive models.

The potential value here is not just in protecting the public by pre-empting crime. Several of the risk assessment tools may be useful for officer safety. For example, tools designed specifically to predict the threat of gang members or of domestic violence perpetrators may inform officers on the importance of greater precaution than they may otherwise take in responding to calls for assistance involving specific suspects. Domestic violence calls are, in particular, among the most confounding and dangerous encounters for the police (Campbell, Gill and Ballucci, 2018). Prior knowledge about the risk profile of the suspect assists officers in quickly adapting their response style as a result.

 
Source
< Prev   CONTENTS   Source   Next >