A significant amount of attention has been focused on the topic of predictive analytics in policing over the past few years. Researchers, practitioners, government officials and vendors have all clamored to take part in the application and evaluation of various analytic approaches. Police chiefs, sheriffs and others in command have even run for office or based their promotional interviews on the premise of understanding and being able to implement predictive strategies to prevent crime and apprehend criminals.
However, crime analysts—although pleased to have a significant amount of attention drawn to their trade—must regularly address myths about what predictive policing efforts do and don't offer, as well as what the police and public should anticipate when engaging in this philosophy. Following are a few of the most common predictive policing myths, as well as some ammunition to dispel each myth.
Predictive policing is new.
Crime analysts, police officers and even the public have both formally and informally engaged in predictive policing for decades— really since the origins of policing. With differing levels of success, we have always been able to use information about past crimes, the environment, the criminal's behavior and time/space to create a logical "prediction" as to when the next crime might occur.
For example, if we know that someone is breaking into cars on Sunday mornings at churches in a particular area of town where there are only a handful of churches, it's pretty easy to predict where they'll strike next. More complex cases will, of course, be more challenging for the untrained to analyze.
The computer knows the future!
Despite the articles in the paper and the social media storm proclaiming repeated success across the nation, the computer does not actually "know" the future. Unfortunately, movies such as Minority Report with Tom Cruise continue to do a disservice to this policing approach. What the computer actually does is produce an algorithm that predicts risk—risk of possible future events, not guaranteed events.
All predictive software uses the same algorithm and data.
The algorithms are different from software to software (if calculated manually, of course, that will be different, too) and almost always involve different data sets. This means not only is the calculation itself different, but the data being calculated is different. One program may use five years of computer-aided dispatch (CAD) data, while another uses both CAD and records management system (RMS) data. Others will incorporate building permits, land use, weather forecasts and even school attendance data. Using different data sets or even different time periods for the same data sets is to be expected, since each vendor has their own approach to the most effective solution.
The computer does it all! Buy the software and you're good to go!
Crime analysts cringe when this is suggested. Even with the most contemporary research and technology, there is no substitute for the trained analyst's involvement and interpretation. Remember, it's the person who must find the necessary data; ensure its accuracy, completeness and reliability; process the data for analysis; and, using the output from the software, interpret the results and make recommendations for response to what has been identified. And remember, the environment is often changing. What had been an empty field for 20 years might now have a high-rise office building or shopping mall or have been paved into a commercial parking lot. Those types of environmental changes, which the computer often cannot understand, must be considered and included in the analysis and, in turn, the recommended responses.
"I was standing at the corner of 1st and Main on the prediction and nothing happened! You were wrong!"
Almost every crime analyst who has created a prediction, either manually or with the use of a software program, has heard some version of this comment—usually from a patrol officer. What the officer fails to understand is that their response may have changed the environment. An officer standing at the intersection in full uniform, patrol car nearby, may have actually prevented the crime—that is, the potential offender either saw the officer and chose to strike at another location, or chose not to strike at all. Admittedly, the prediction might have been wrong (the criminal did not intend to strike there), but it also might have been right, even absent a crime report, criminal apprehension or the admission of the crime in a post-arrest interview with a serial offender
Predictions are most valuable to "catch" a criminal and have nothing to do with crime prevention.
Ask any citizen which they would prefer: to be the victim of a crime, but the police arrest and prosecute the offender, or never to have been the victim of a crime at all. It's unlikely any citizen would choose the former.
Traditional police response has always been reactionary; the police intervene after the crime has been committed. More contemporary police responses have focused significantly on crime prevention. With predictive policing, knowing there is a strong likelihood of a crime occurring at a particular location during a certain timeframe gives police a critical "heads up" and allows the police time to develop strategies based on what is known to work to prevent the crime.
As the body of scientific research has deepened over the past decade, and police understand more about what is effective in preventing certain types of crime, departments have been able to leverage the science into practice and contribute to a reduction in crime rates. "If you can predict it, you can prevent it."
If you engage in predictive policing, you will quickly see crime reductions.
A recent news article quoted a police official as saying that after implementing a piece of software, their jurisdiction saw a 10% decrease in crime in just one day. Almost anyone reading that article, professional or layperson, who took a minute to think critically about the quote would have questioned the validity and/or applicability of the statistic.
Actual decreases in crime require scientific evaluations over a period of time, and a statistically significant level of increase or decrease in crime using the predictive approach. Being able to conduct these studies requires the careful gathering of data on the responses used regarding the prediction made. To reach a point where these conclusions can be safely drawn, longer term observation is required.
Prediction is just about "the next event"; it's tactical in nature. Although prediction is best known in law enforcement when the analyst is looking to predict "the next event" in a tactical (active) crime series, prediction can also be used in other ways. For instance, prediction can be used to determine how many additional officers need to be hired based on anticipated growth in a particular area of the jurisdiction. It can also be used to predict "when" a particular trending crime might reach various areas of the country, such as when oxycontin abuse would arrive in the Midwest. Prediction could also be used to determine the anticipated increase in reported crime following a mass release of offenders from an overcrowded prison into the community. The uses of prediction are numerous.
Predictive policing is done the same way in all sizes of jurisdictions.
Large jurisdictions with significant levels of crime will likely need more sophisticated software and large data sets. Smaller jurisdictions may infrequently encounter serial offenders or have significant changes in the environment (such as large-scale development) and may be able to use comparatively simple capabilities. Good prediction software need not be expensive or ridiculously complicated.
Moving in the Right Direction
Predictive policing continues to be an important topic in law enforcement, in research, in government and in the private sector. Despite the attention and progress thus far, there is still much to be gained through more research and evaluation in the area. Further, beyond making a prediction, understanding how to most effectively respond is key; knowing something is likely to happen but not knowing what to do about it is useless. All of this progress in the field, however, indicates that law enforcement continues to be more scientific and less random—an indication the field is moving in the right direction.