Furthermore, the discussion has not explored applications of machine learning methods for causal inference. The limitations of parametric models can, in many instances, be offset by computational statistical learning. Targeted learning, a technique that allows to estimate a single parameter of interest with machine learning methods has been successfully applied to estimate the effect of different policies at the individual level.
In this paper, I focus on the question of the extent to which machine learning is a tool to better understand data related to criminal justice policy . To this end, I organize this paper as follows. The first section provides working definitions of machine and statistical learning, and presents common statistical algorithms used in the field of criminology. Then, in the second section, I discuss a set of cases of criminology problems analyzed through the lenses of machine learning methods. These cases were selected to illustrate the practical application of machine learning to criminal justice and provide insight into the debate about the role of computational statistical techniques in dealing with real world policy issues.
Then, in the third section of this paper I review the case of assessment of risk of recidivism and violent behavior in order to illustrate the two poles of an undergoing debate about machine learning and criminal justice. At the center of the such debate is the guiding question of this article: are such statistical methods suitable to understand criminal behavior and design strategies to promote public safety?
The choice of risk assessment tools used in sentencing to analyze the usefulness of using machine learning in the field of criminology serves two purposes. One is to present the discussion in a practical fashion and not only in mathematical terms. Decision making processes such as sentencing are a practical and relevant for society as a whole, as it affects the lives of individuals and impacts public safety. The other objective is to exhibit the common misunderstandings and pitfalls that occur when academics and public officials communicate. Risk assessment in sentencing meets both aims. In the fourth section of this paper, I conclude by answering the guiding question and identifying ways to make the use of machine learning productive for criminal justice policy.
What is Machine Learning?
Machine learning is a sub-field of computer science based on the study of pattern recognition using computational tools (software and hardware) in order to identify mathematical rules. Such mathematical rules allow to predict future outcomes from existing data. Machine learning can be broken down into three actions, namely, defining an algorithm, training the algorithm on data, and collecting an expected outcome using using a computer and specialized software. The mediation of computers to learn from data explains the use of the term computational statistical learning as a synonym of machine learning.
Thus, the characteristic feature of machine learning is the use of computational tools understood as powerful computer hardware and statistical software, both needed to handle large datasets. Aside from this characteristic, the elements of machine learning are essentially the same than those of the statistical methods used in the last century in criminal justice. Some examples of such statistical methods of estimation include expert’s forecast of criminal behavior in a neighborhood, a linear projection of crime relative to a set of sociodemographic variables, an ordinary least squares regression procedure to establish crime incidence or the computation of a score of risk of violent behavior on the basis of a test designed by professionals with clinical expertise.
In social sciences, including criminology, research involving statistical analysis has shown a growing reliance on specialized software first developed in the 1960s. Below a table shows the most widely known specialized computer programs used in social sciences, including criminology, particularly in subfields such as quantitative criminology.
TABLE HERE
SPSS, Stata and R are arguably the most widely used programs to conduct statistical analysis in criminology studies using quantitative data. SPSS and Stata are paid programs, while R is an open source software accessible free of charge and crowd-sourced, as any individual can contribute with a library or package containing algorithms to perform computations and implement statistical model. A relevant caveat to examining the trend in use of these three programs is the fact that only R supports the implementation of machine learning algorithms and unlike Stata and SPSS is increasingly improving packages to efficiently conduct complex and recursive computations using big data. It is only in the past few years that Stata researchers have placed growing attention to the development of implementations of popular machine learning techniques such as random forest and support vector machine.
Thus data availability paired with computational power allow for the use of elaborated mathematical formulas, as opposed to simple linear ones, and a more frequent use of wide and big data to answers policy question (billions of observations of multiple variables in different formats), as opposed to small data (imaginable number of observations of a few defined variables).
DISTINGUISHING STATISTICAL METHODS
Methods for
Machine Learning and other methods
Machine Learning of Actuarial, Clinical