Joanne Pohl

For many decades, the American criminal justice system has been exploring methods to accurately predict which criminals will commit more crimes. It is easy to understand why judges, prosecutors, defense attorneys and probation officers are keenly focused on the issue of recidivism. If criminal justice professionals were able to accurately predict the likelihood of a defendant committing more crimes, they could develop more effective sentencing practices and more effective supervision strategies. But predicting which criminal defendants will continue to commit crimes and which will not is proving to be a troublesome task.

As we have all experienced in our daily lives, predicting the future is fraught with difficulty. While some of our everyday predictions, based on past experience, common sense and intuition, prove to be correct, we understand that complex matters that involve unknown variables are very difficult to predict with accuracy. Human knowledge gained from experience can be extremely instructive when predicting what will happen in the future, but professionals in all fields of study are looking to science to provide the best tools for predicting future events.

We see this trust in science and reliance upon rigorous scientific methods in the field of Corrections. Because of the serious social consequences involved in crime and criminal justice, our communities hold criminal justice professionals to very high performance standards. Citizens and community leaders expect judges, prosecutors, defense attorneys and probation officers to utilize only the most effective practices in their dealings with criminal defendants. They want each individual decision about each individual defendant to be the correct decision. Not only is the community’s safety at stake, so too is the future of the defendant and his/her family.

One of the most significant factors bearing upon a criminal defendant’s ultimate sentencing outcome involves the defendant’s likelihood of committing another crime. While best practices provides that sentencing decisions should reflect a balanced approach addressing public safety, community restoration and offender competency, knowing a defendant’s likelihood for future criminal activity would result in a better informed sentencing response from the criminal justice system. Indeed, such predictions regarding a defendant’s future criminal behavior are routinely part of the arguments presented to the sentencing judge by the prosecutor and defense attorney at a sentencing hearing.

A defendant’s likelihood to reoffend comes to the forefront of judicial decision-making in areas in addition to sentencing. Identifying pre-trial release conditions also involves predicting a defendant’s behavior. Determining the level of a criminal defendant’s probation supervision is another area that is heavily dependent upon behavioral predictions.

With so many significant ramifications attendant to the matter of predicting a criminal defendant’s future behavior, we can understand why accurately predicting criminal recidivism has garnered such laser-type focus within the field of Corrections.

Corrections departments have utilized risk assessment tools widely across the country for over twenty-five years. There have been many risk assessment tools developed over this time, however, a report issued by Congressional Research Service in 2015 observed that no single risk assessment was found to be significantly superior to any other in predictive validity. Furthermore, the report stated, “in general, research suggests that the most commonly used assessment instruments can, with a moderate level of accuracy, predict who is at risk for violent recidivism.” Nevertheless, even with this “moderate level of accuracy,” risk assessments are not without detractors. Criticism of risk assessments centers on the discriminatory effects brought about by their use because some of the risk factors are correlated to race. Even the most recently developed tools, built upon predictive algorithms (and excluding an individual’s race information), have been found to produce racially biased predictions.

An article in the January 2018 Science Advances reports on research conducted by the Department of Computer Science of Dartmouth College that concluded that despite analyzing over 137 features of an individual criminal defendant and utilizing predictive algorithms, the widely used assessment tool was neither more fair nor more accurate than human assessment. In fact, the article highlighted a study conducted in Broward County Florida between 2013 and 2014 where 7,000 arrested defendants were tracked for recidivism and the assessment tool predictions were found to be unreliable and racially biased. The researchers proposed this final thought regarding the assessment tool which has been widely used (at least one million offenders) since its initial creation in 1998: “When considering using software such as COMPAS in making decisions that will significantly affect the lives and well-being of criminal defendants, it is valuable to ask whether we would put these decisions in the hands of random people who respond to an online survey because, in the end, the result from the two approaches appear to be indistinguishable.”

The researchers recommended caution in relying on algorithmic approaches to predicting recidivism because their meta-analysis of nine different tools found only moderate levels of predictive accuracy. So while more research is being conducted to establish accuracy and fairness of risk assessment tools, it is important for criminal justice professionals to understand and recognize their limits. Even in 2018, predicting recidivism, despite available scientific methodologies and specialized software, remains a speculative endeavor.

Load comments