Analytics and Operations Research

Many of the decisions associated with criminal justice systems can be addressed through the use of analytics, and a closely related field,

operations research. Analytics can be defined as the scientific process of transforming data into insights for the purpose of making better decisions (Best Definition of Analytics, n.d.). The emphasis in analytics is on making decisions, which happens to be the emphasis in this book.

Analytics is often categorized according to three types: descriptive analytics, predictive analytics, and prescriptive analytics (Operations Research and Analytics, n.d.). Prescriptive analytics, the focus of this book, yields guidance for making decisions. This guidance is accomplished using models or representations of systems.

Many decision makers employ a mental model of the situation when deciding. Typically, in these mental models, the decision maker will mentally assess the various respective outcomes associated with the alternative decisions and then select the decision which gives the best predicted outcome. These outcomes can correspond to the values associated with the various performance measures for the situation, so the decision maker will need to trade-off among these conflicting performance measures.

For example, a state could have a decision to make as to which private company will be selected to operate a new prison. (This issue of privatization of prisons is discussed in Chapter 3.) In making this decision, the state would naturally consider several conflicting performance measures, including those related to cost, safety, security, living conditions, rehabilitation programs, etc. Naturally, trade-offs would need to be made between cost and several of the quality-oriented measures.

In analytics and operations research, we use models which are developed and experimented with on a computer. These models can be any of several different types, including a decision tree, an optimization model, a simulation model, etc.

One of the advantages of such computerized models is that they allow for the experimentation with different alternatives without disturbing the original system. For example, consider the First Step Act, signed into law in December of 2018, and discussed in some detail in Chapter 4 of this book. This law had several parameters, including the funding of $244 million for recidivism-reducing programs for pre-release custody inmates and the earning of 10 days of earlyrelease credit for every 30 days of successful engagement in effective activities by the inmate. With an appropriate simulation model of the federal prison system, one could easily experiment with these and other values for the two parameters. Such a model would allow for the representation of many years of simulated time with a few seconds of actual time, thereby allowing for the simulation of many policies.

In this book we focus on the use of some methodologies from the area of multiple objective analytics; these methodologies include multiattribute value functions and multiattribute utility functions. The reason for the focus on multiple objectives is that the decision situations in criminal justice systems naturally lend themselves to these methods, as will be discussed in the next subsection.

Complexities Which Make Decisions Difficult for Criminal Justice Systems

There are several complexities associated with criminal justice systems which result in major difficulties for their design and operation. These complexities include the following:

  • • Criminal justice systems have many interacting subsystems, including those related to policing, courts, and corrections.
  • • Criminal justice systems have complex interactions with other systems, such as those related to social welfare, mental and physical health, education, etc.
  • • Criminal justice systems have many decision makers and stakeholders for their various decision situations, resulting in the existence of several conflicting performance measures.
  • • Criminal justice systems behave in a dynamic fashion, resulting from changing demographics, economics, laws, and morals.
  • • Criminal justice systems behave in a probabilistic manner due to many reasons, not the least of which has to do with the uncertainty of human behavior.
  • • Making changes to criminal justice systems, through for example new laws, is typically accomplished through an often highly partisan political environment.
  • • Good data for making decisions about criminal justice systems is often difficult to obtain.

As mentioned earlier, the three major subsystems of a criminal justice system are law enforcement, the court subsystem, and the prison/jail subsystem. Each of these subsystems can be further subdivided. For example, the prison/jail system can be divided into county jail systems, state prisons, federal prisons, probation services, etc. A change which initially affects one system/subsystem can subsequently affect others, which makes the accurate prediction of the effect on the overall system difficult to accomplish; hence, the need for multiple objective analytics.

As an example of a law that was enacted to change the operation of one part of a criminal justice system, but then resulted in (perhaps unanticipated) changes to other parts, consider the habitual offender law (often called the “three-strikes law”). This law has been enacted in several states over the last 30 years. There are many versions of the law; one version states that a severe violent felony, coupled with two previous convictions will result in a mandatory life sentence for the offender. Auerhahn (2008) illustrated how the three-strikes law in California resulted in a chain reaction in which fewer of the accused with two previous convictions entered guilty pleas, resulting in a much more congested jail and court system in California. This in turn resulted in court orders to cap the jail population, resulting in early release of many prisoners from county jails.

Criminal justice systems interface with and are greatly affected by other systems: social welfare systems, educational systems, economic systems, mental health systems, and health care systems in general. Addressing problems in these other systems can result in benefits to criminal justice systems. Not addressing many of the problems in these other systems can result in an overcrowded criminal justice system. For example, a poor educational system can lead to an increased dropout rate, which in turn can lead to increased criminal behavior.

Because there are typically many decision makers and stakeholders for most important decision situations of a criminal justice system, there can be different objectives and corresponding performance measures. The example discussed earlier in this chapter, involving which private company should be chosen to operate a state prison, had objectives related to cost, safety, security, living conditions, and rehabilitation programs. Making a decision in this situation obviously involves making trade-offs between cost and any of the objectives related to quality.

Changes to a criminal justice system affect the system over time; in addition, outside influences such as demographic characteristics of the population change over time, hence the dynamic nature of these systems.

The fact that criminal justice systems are dynamic in nature arises from the fact that demographics, laws, social customs, economic conditions, technology, and other important aspects of the human condition are changing over time. Certainly, decisions that have a strategic or long-term effect, such as the enactment of federal laws, must, at least implicitly, forecast these changing conditions in order to gage the outcome associated with the enactment.

Since system changes result in effects that occur in the future and which also depend on human behavior, the effects will typically have a large amount of inherent uncertainty. For example, with respect to early release programs, there will be uncertainty as to whether a released person will commit more crimes. Hence, modeling approaches w'hich accurately represent uncertainties in the behavior of persons released from prison are important to employ.

Human behavior is difficult to exactly predict, but one may be able to predict it in a probabilistic fashion. For example, a defendant released on bail or on his/her own recognizance may or may not appear for his/her court date and may or may not commit crime(s) prior to trial. However, by considering various characteristics of the situation and the defendant, a probability distribution over the various actions of the defendant may be assessed. Such a probability distribution may be input to a model which would aid the judge in deciding about the pre-trial disposition of the defendant.

The political nature of decisions associated with criminal justice systems is exacerbated by the interpretation of the data analysis associated with these systems. For example, Barnett (1988) provides three examples of misinterpretation of data involving (1) racism and the death penalty in Georgia; (2) the hypothetical increase in crime that would occur if all prisoners were released from jail or prison; and (3) whether defendants who were arrested for violence, would be more likely than other defendants (i.e., those arrested, but not for violent behavior) to engage in pretrial misconduct if released on their own recognizance.

The first example discussed by Barnett involved a Supreme Court case in the state of Georgia, discussed in an article in the New York Times. The contention was, as stated in the Times article and in Barnett’s paper, that:

other things being equal as statisticians can make them, someone who killed a white person in Georgia was four times as likely to receive a death sentence as someone who had killed a black.

Barnett showed how this statement could be easily misinterpreted, given the facts of the study. Even though the examples cited by Barnett were from more than 30 years ago, the practice of misinterpreting data for political gain continues to date. Examples can be seen in Chapter 4 of this book involving the recent discussion of the Violent Crime Control and Law Enforcement Act of 1994.

Finally, with respect to the lack of good data, as discussed in Chapter 1, many of the data collection systems involve self-reporting which is often inaccurate in nature.

The seven complexities discussed above make the prediction of the effect of a change to a criminal justice system, such as a new law, difficult to accomplish. In turn, this difficulty makes the choice of a best alternative (i.e., optimization) a difficult process. This difficulty occurs no matter what types of analyses are employed for the decision process.

Many of the complexities discussed above are typically found in what are termed ill-structured problems (Simon, 1960). Hence, it is important to consider problem structuring methods when addressing decision situations for criminal justice systems.

In the next section we will discuss four categories of methods that can aid in addressing several of the difficulties presented above. These are (1) problem structuring methods, (2) methods for generating the multiple performance measures for a decision situation, (3) methods for representing the preferences of a decision maker over multiple performance measures, and (4) simulation modeling and analysis as a way to represent the dynamics and probabilistic nature of a system.

< Prev   CONTENTS   Source   Next >