By Andrew Guthrie Ferguson[1]

The future of policing just arrived on a computer screen in downtown Los Angeles. As reported by National Public Radio, this month the Los Angeles Police Department’s (LAPD) Real Time Analysis and Critical Response Division has embraced a new computer program offers the holy grail of smart policing: the ability to predict where crime will happen and direct police resources to that location.

Predictive policing is based on the theory that by analyzing past crime patterns and crime data, police analysts can identify future hot spots of crime. Using this predictive modeling, police officers are directed to patrol areas of expected crime for additional police attention. These small hot spots of heightened police presence—in Los Angeles, a 500 foot by 500 foot area (one square block)—are poised to become the centerpiece of a new smart policing strategy in which law enforcement resources are directed to targeted locations before the crime occurs. The initial results are strikingly successful. As reported by the LAPD to NPR, for example, the first week of “predictive policing” reduced crime in certain areas by 50%.

Predicting the impact on constitutional rights in those targeted hotspots may be a bit more complicated. One of the unintended consequences of predictive policing technology may be the erosion of Fourth Amendment protections for individuals in those high crime areas.

Under current law, the character of a neighborhood as a “high crime area” can be a factor for “reasonable suspicion” to stop a suspect. In Illinois v. Wardlow, the Supreme Court held that the “high crime” designation “is among the relevant contextual considerations” to determine reasonable suspicion under a totality of circumstances analysis. In Wardlow, the Supreme Court upheld a stop based on a two-factor totality, “high crime area” and “unprovoked flight from police.” SinceWardlow, there have been thousands of federal and state cases that have used the characterization of the area as “high crime” as one of, if not the controlling, basis to justify a police stop of a citizen.[2]

So, imagine the case based on the predictive crime mapping scenario. Police officers receive official orders based on the predictive crime mapping program that a particular area is likely to have a higher incidence of crime. Officers respond to the area. They observe suspicious activity. They stop a suspect based on those suspicions and recover evidence of a crime. At the suppression hearing, the police officer will testify that the officer received official orders to respond to a 500 foot by 500 foot area to be alert for crime. This information reasonably influenced his observations and objectively supported his reason for stopping the suspect. The court will likely accept this rationale as objectively reasonable and will factor it into the totality of circumstances, thus satisfying one of a two-factor totality. While it is not automatic that a court will allow the character of the neighborhood alone to be sufficient for a finding of reasonable suspicion, it will certainly weigh heavily in its decision, and likely be enough to tip the scales to upholding the stop.

The Result: Predictive policing will shape Fourth Amendment analysis by officially designating one of the two factors of the totality of circumstances underWardlow. While police will still need some other factor to justify a stop, the thumbs will be on the scale weighted toward a finding of reasonable suspicion.

The Concern: Predictive policing will create an implicit high crime area exception to the Fourth Amendment in those targeted areas. Individuals in those areas will have a lesser expectation of privacythan those in other non-high crime area neighborhoods.[3]

Interviewed on NPR, former LAPD Police Commissioner William Bratton acknowledged that under the predictive policing paradigm, certain areas would be “redlined” and stigmatized as high crime areas, an unavoidable consequence of addressing real crime problems. While Commissioner Bratton was candid in his acknowledgment of the economic costs to targeted areas, unacknowledged was the constitutional effect of this official designation.

Thus, whether in Los Angeles or New York or anywhere in between, predictive policing offers a seemingly effective crime fighting strategy. While the practice raises concerns about perceived racial or socio-economic discrimination, stigmatizing effects on neighborhood development, and increased police-citizen interactions, it is the constitutional consequence that deserves an extended examination.

Simply stated, will predictive policing create targeted areas of lesser Fourth Amendment protection? If I am walking in an officially designated “hot spot,” does a police officer have more objective justification to stop me for behavior that would not constitute reasonable suspicion in a “non-high crime area”? If yes, would such an explicit “high crime area” exception to the Fourth Amendment survive Supreme Court review? Or more likely, will the combination of Wardlow and predictive policing create an implicit “high crime area” exception that needs to be addressed by the courts?

Such questions have not been adequately considered by scholars or courts reviewing the Fourth Amendment consequences of “high crime area” arguments. Further, as highlighted in the move toward predictive policing, these questions will only get more difficult as the technology advances with “real time” data updates, GPS coordinates, and improved mapping.

As a framework for analyzing these questions, I would suggest three questions that courts and litigants should consider in addressing the effect of crime mapping technologies on the Fourth Amendment.

First, what information does predictive policing provide that is relevant to the Fourth Amendment analysis? Are past acts really relevant to future observations? Central to the utility of this new technology must be a persuasive argument that past criminal activity has some objective relevance to current police observations in the area. Is the prediction focused on individuals, actions, or areas, as each prediction may be evaluated differently in a reasonable suspicion analysis? Are only certain crimes predictable (property crimes, as opposed to violent crimes) such that the technology has relevance? This emphasis on relevance is an often neglected concept in analyzing high crime areas.[4] The only reason why the “high crime area” is relevant for Fourth Amendment purposes is that an officer’s knowledge of the area makes the officer’s suspicions more objectively reasonable. For example, if the officer is informed that an area is known for car thefts, then a suspect testing the handles of parked cars becomes more objectively suspicious. However, if the officer is informed that the area is known for car thefts, and the officer observes a series of hand to hand transactions (signifying a possible drug deal) then the predictive information is not relevant to reasonable suspicion. Courts, thus, must ensure that there is a nexus between past crime data, future predictions, and police observations. More fundamentally, real questions must be asked about the accuracy and validity of the predictive programs.

Second, does predictive policing help define a functional geographical and temporal area of analysis?[5] In many reported cases, the geographic and temporal boundaries of the designated “high crime area” are not defined. Should courts look to a small area like the LAPD chose (one city block) or a larger area, such as an entire neighborhood, a police district, or city center? What is the appropriate timeframe for comparison? Should we be analyzing crimes per year, per month, per week? How high is high enough to be a “high crime area”? These questions get all the harder when you move away from a particular city and start comparing urban areas to rural areas, state crime statistics to county statistics etc. If limited to the 500 feet by 500 feet area utilized by LAPD, predictive policing might provide a usable area of analysis. If however the area is defined as a broader area, the usefulness of the predictive model may be illusory.   

Third, if predictive policing is based on crime data collection and analysis, how can we ensure that the data collection, analysis, and application are accurate, transparent, and accountable? Crime mapping technologies, including predictive policing, can only be as good as the data collected. Officers and analysts make errors in recording crime statistics, and much of the predictive value stems from precise recording. Further, not all crime is recorded. In fact, most recorded crime involves street crime as opposed to cybercrime, financial crime, or even possession offenses. Equally questionable, law enforcement maintains exclusive control over the data.[6] While jurisdictions like Los Angeles make the crime data available to their citizens through publically available websites, there is no independent check on the data or accountability for the analysis. Courts, thus, are without the independent ability to address the reliability or accuracy of the data on which they are being asked to judge the Fourth Amendment question. Figuring out how to improve this transparency should be a major concern for courts.

Finally, it must be noted that predictive policing itself may affect the statistics that shape the predictions. At one end of the spectrum, a successful predictive policing policy can create a self-perpetuating high crime area. For example, if police administrators direct police officers to a particular location, the increased police presence may mean more arrests. If there are more arrests, that fact will justify a continued designation as a high crime area, which will necessitate an even greater police presence and again more arrests. At the other end of the spectrum, a heavy police presence may deter crime (without greater arrests), which will mean that crime rates will be reduced causing the area to lose its “high crime area” designation even though it was the designation (and the extra police presence) that caused the decrease in crime. Courts addressing a Fourth Amendment question at either end of the spectrum will be asked to address the character of the neighborhood without the historical context. Courts will be asked to address a single case, and will not be privy to the history of crime patterns and perhaps the effect on policing strategies to create the very high crime statistics the courts are now being asked to rely upon.

In the face of these tensions of law and technology, I offer a simple suggestion for courts addressing the impact of new crime mapping technologies on the Fourth Amendment. At a minimum, courts should demand objective and particularized data about the predicted “high crime areas.” In fact, adopting a particularized approach to predictive policing may obviate the need for the “high crime area” terminology that courts have relied on for decades.

As I have explained in other contexts, a “particularized approach” would require that the police officer be informed of the particular crime patterns in a cabined area, and have that information inform his suspicions before a court would rely on any consideration of the neighborhood.[7] Generic labels like “high crime area” or “hot spots” would be replaced with information about the actual crime patterns, narrowed by temporal and geographic limits. A judge applying the particularized approach “would look at the objective data available to the officer and its relevance to the observation of alleged criminal activity, and factor that objective information into the reasonable suspicion analysis. If the officer did not base his decision on specific data about a specific crime problem in a specific area, or if the data relied upon did not demonstrate a specific and relevant crime problem, then reliance on this information should not be considered.”[8]

Thus, in the predictive policing example, if a police officer was dispatched to a localized area because there had been several car break-ins in the last few days, and the officer observed someone suspiciously casing cars, this particularized information about a specific area might be relevant, targeted, and reasonable for a court to consider in determining objective reasonable suspicion. However, if the police officer had simply been told there had been an uptick of crime (undifferentiated by type) then this information about a generic “high crime area” should not be used by the court in analyzing reasonable suspicion. Of course, if the officer was informed that there had been a statistically significant increase of car break-ins and drug arrests and prostitution arrests, then observations of any of those activities could form reasonable suspicion.

If done with the appropriate precision, predictive policing may solve many of the problems that have plagued courts trying to wrestle with how to define the high crime area question. At least in the ideal, if the predictions are narrowed to specific crime types, in localized areas, and based on accurate and timely data, this type of smart policing may help officers and answer some of the difficult definitional questions raised by the high crime area cases.  

Of course, real questions remain with the underlying theory and practical application of predictive policing. The Fourth Amendment’s requirement of individualized suspicion for a person exists in tension with even a particularized suspicion of an area. In addition, analysts’ predictions have not been evaluated in the constitutional context of reasonable suspicion determinations in the courts. For example, is a prediction more like a “hunch” that has been deemed inadequate for reasonable suspicion, or a reasonable fact to be accepted as part of the totality of circumstances. In addition, predictive policing modeling programs may not prove to be accurate, or criminals may change their patterns evolving to thwart this smart policing. Whether predictive policing will revolutionize policing or merely provide a new tool to the ongoing crime problems in certain jurisdictions, one prediction that will certainly come true is that it will have a real effect on the future of the Fourth Amendment.