Criminal Justice Data Analysis

Can data analysis be used to improve the United States’ criminal justice system? 

        Right now, in the United States the most talked-about policy subject is crime and police reform. By far the biggest reason for this is that following the unnecessary deaths of several unarmed black Americans at the hands of law enforcement officers, the public has been demanding serious reform to the police system in the United States. But at the same time, law enforcement in this country was already potentially facing major changes for a different reason: the rise of predictive policing technology. This generally involves using predictive models to determine who is most likely to commit future crimes and where the crimes are probably going to be committed. Although systemic racial biases tend to manifest themselves to some degree in computer algorithms that were programmed by humans, it is possible that by combining data analysis strategies with certain political reforms suggested by activists today, local governments can decrease funding for the police by targeting their spending more efficiently and reducing bail recidivism. However, the benefits of these goals must be weighed against the downsides of the data analysis techniques, specifically, its risk to contribute to the racial inequality in the criminal justice system.

                                                                                        

                                                                                                                                     Image Credits: FiveThirtyEight

        Predictive policing methods tend to sound appealing on their surface, and for good reasons. Examples of what these techniques look like include those used by the University of Chicago Crime Lab in Chicago and by Palantir in New Orleans. These tactics aim to predict the most likely people to be the perpetrator or victim of a violent crime, then inform law enforcement about the neighborhoods where there are high concentrations of these high-risk individuals so the police can patrol those areas more frequently. A few years ago, using factors such as demographic data about prior violent crime offenders and others’ relationships with these people, the UChicago Crime Lab’s model created a list of the 5,000 people in Chicago most likely to be arrested for a violent crime with a gun or to be a victim of one. One year later, 17% of the homicide victims in the city came from that list, even though 5,000 people is just 0.2% of Chicago’s population. 

 

    Some predictive policing systems involve tracking population movements and digital communications from these people that are labeled as potentially dangerous, but even without these potentially invasive measures, if police officers used this system it would allow more effective management of their time and resources and force them to be proactive instead of reactive when dealing with crime. Since the simple presence of a police officer in an area can significantly decrease the criminal activity in that space, the predictive model is designed to give them a guide to the neighborhoods that they should patrol most often. And other organizations could then intervene with these people that are deemed likely to commit a violent crime, helping them with therapy, support services, or whatever else is needed. For example, the program Becoming A Man showed success in being able to reduce violent crime arrests among young men in Chicago by “blending the confessional aspect of support groups with the tough love of male mentoring and elements of cognitive behavioral therapy, a technique that helps people learn to change their patterns of thought, and, as a result, their behavior.” If this program or ones like it could scale up to be in cities nationwide and aim to enroll those who are deemed likely to be arrested in the future, it could go a long way towards significant declines in crime by giving these young people real guidance and an outlet to talk about their pain or anger. Together, the strategies discussed above could mean that local governments could potentially drastically reduce their spending on hiring and arming law enforcement and instead efficiently focus their efforts on the most dangerous neighborhoods and techniques to help at-risk people before a crime is ever committed.

        However, upon further inspection it becomes apparent that there are underlying issues with predictive policing that should be addressed before it is used. First, as a researcher from the Human Rights Data Analysis Group’s Policing Project states, data used in these predictive police systems “are collected by a criminal justice system in which race makes a big difference in the probability of arrest — even for people who behave identically”. This means that minorities will generally be deemed higher risks of getting arrested for a crime even if they are in the exact same circumstances as their white counterparts. And just because they are more likely to be arrested for a crime does not mean they are more likely to commit it. Furthermore, when police officers are sent to a specific neighborhood and warned that it is filled with people likely to commit a violent crime, they could perceive a non-threatening situation as being much more dangerous than it is and respond with unnecessary force. This is already something that occurs in predominantly black areas because many police officers instinctively (but often inaccurately) consider them unsafe, but when law enforcement has actual proof of certain areas being dangerous, the problem would most likely get worse. This would only exacerbate the problem we are facing in the United States of law enforcement officers in non-threatening situations killing or seriously injuring people, very frequently black men, who are not truly a danger to the officers or anyone around them. So although predictive policing could allow for significant reductions in funding for the police while still maintaining an effective law enforcement presence, local governments who are currently using these tactics or thinking about using them need to fully consider how to address their drawbacks.
    

        Another related idea that offers to improve the criminal justice system, but has issues attached as well, is predicting bail recidivism through data science. Currently the COMPAS score developed by the company Equivant aims to do just this, and is used in locations all over the country to help improve judges’ decisions regarding bail sentencing. This algorithm is developed essentially the same way as most other risk assessment tools, which, according to an article for the statistical analysis website FiveThirtyEight, means having the company “look at a large population of former prisoners, examine hundreds of facts about their lives, and then follow the individuals over several years to see what traits are associated with further criminal activity.” Because bail amounts are supposed to be adjusted based on how likely defendants are to commit another crime upon their release (recidivating), and statistical prediction techniques are very frequently more accurate than unaided human judgement, data-informed tools to determine the likelihood of recidivism can be very valuable. Judges can also use these algorithms in their decisions to put someone in jail or on probation, but as the Wisconsin Supreme Court ruled in 2016, the COMPAS scores should not be the determinative factor in this decision.

        This method’s growth in popularity in the United States has come with quite a bit of controversy, and it is not difficult to see why. The public currently does not have the right to see what data goes into any individual’s score, giving the algorithm a feel of secrecy that is unnerving to defendants when their entire prison sentence could depend on what risk rating they get. Furthermore, in a study done by ProPublica from 2013 to 2014, black defendants in Broward County were twice as likely to be wrongly labeled as a future criminal by the COMPAS model (meaning the algorithm said they would commit a crime in the future when they did not) as white defendants, while white defendants were more likely to be wrongly labeled as low risk. When the researchers separated the effect of race from other typically important factors such as age, gender, and previous criminal history, they found that blacks were still 77% more likely than whites to be labeled as high risk of committing a future violent crime and 45% more likely for all crimes. This leads judges to assign higher bail amounts to black defendants in the exact same situation, resulting in higher rates of incarceration for blacks. While this is only one study, and there has not been enough similar research to reach any concrete conclusions, the results suggest that bail recidivism prediction, like the predictive policing strategies discussed earlier, threatens to exacerbate the staggering racial inequality in the American criminal justice system rather than fix it. There is potential for it to improve the system by making more informed decisions regarding bail or probation sentences, but the facts that it is seemingly shrouded in mystery to the public and tends to make especially inaccurate decisions for minorities need to be addressed before moving forward. The former can be easily done by informing defendants and lawyers what types of data (and information about their own data) are factored into the model so they can correct any mistakes and are fully informed about why they received the decision that they did. Making the data fully public would also allow third-party sources to evaluate the algorithm for racial bias. But the latter issue is quite obviously more complicated, as the United States has struggled for decades, even centuries, over institutionalized racism, which is likely an enormous factor in the higher risk scores for black defendants. There will be no easy fix, and there are many complexities involving how to determine how and why racial biases are influencing the models as well as how to correct for this, but cooperation between local governments, law enforcement, and the public should be key to making sure everyone is as informed as possible in order to come to a collective solution.

        In summary, despite the great potential for data analysis to improve the criminal justice system in the United States, there are too many underlying problems and there has not been enough conclusive research on the subject to declare it successful. But if the companies developing these technologies can figure out ways to effectively address the issues mentioned earlier and then implement their models nationwide, it is certainly possible for our justice system to undergo major changes for the better.

Works Cited

Angwin, Julia, et al. "Machine Bias." ProPublica, 23 May 2016, www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing. Accessed 18 June 2020.

Barry-Jester, Anna Maria, et al. "Should Prison Sentences Be Based On Crimes That Haven't Been Committed Yet?" FiveThirtyEight, 4 Aug. 2015, fivethirtyeight.com/features/prison-reform-risk-assessment/. Accessed 23 June 2020.

Eckhouse, Laurel. "Big data may be reinforcing racial bias in the criminal justice system." The Washington Post, 10 Feb. 2017, www.washingtonpost.com/opinions/big-data-may-be-reinforcing-racial-bias-in-the-criminal-justice-system/2017/02/10/d63de518-ee3a-11e6-9973-c5efb7ccfb0d_story.html. Accessed 22 June 2020.

Ferguson, Andrew. "How data-driven policing threatens human freedom." The Economist: Open Future, 4 June 2018, www.economist.com/open-future/2018/06/04/how-data-driven-policing-threatens-human-freedom. Accessed 17 June 2020.

Heath, Dan. Upstream: The Quest to Solve Problems before They Happen. New York, Avid Reader Press, 2020.