Abstract
Criminal justice forecasts of risk are increasingly being associated with machine learning and artificial intelligence. One might get the impression that constructing statistical forecasts of criminal behavior is primarily a technical enterprise. But the forecasted risks inform real decisions by criminal justice officials and other stakeholders. As a result, a wide range of matters can arise, only some of which are technical. Although for expositional purposes the issues must be examined one at a time, in practice each should be considered as part of a whole. Decisions about one necessarily affect decisions about another. Implied is the need to consider tradeoffs between competing priorities. We begin with some rather broad concerns and gradually narrow the discussion.
This is a preview of subscription content, log in via an institution.
Buying options
Tax calculation will be finalised at checkout
Purchases are for personal use only
Learn about institutional subscriptionsNotes
- 1.
There are many different psychological instruments that have been used on prison populations. Along the more venerable is the Minnesota Multiphasic Personality Inventory (MMPI) that contains scales measuring antisocial practices, self-control, and substance abuse vulnerabilities (Megargee et al. 1999). Even if such tools contain powerful predictors, they would be of no help unless routinely used with the target population, including offenders for whom forecasts are needed.
- 2.
Many of the leading figures associated with the lethality assessment instrument are well aware of these difficulties. Campbell and her colleagues (2003) show that there may be some associations between content included in the lethality assessment instrument and intimate partner violence leading to murder. That is certainly a step in the right direction. However, in a recent study of police use of the instrument, the outcomes are responses to survey questions based on a revised version of the well-known conflict tactics scale, which is essentially a check list of offenders actions reported by the victim (Messing et al. 2015). The conflict tactics scale is taken to be a proxy for what really needs to be measured. Another recent study focuses on the instrument itself, but again, the outcome is measured by a version of the conflict tactics scale (Messing et al. 2017).
- 3.
There is several decades of work on the costs of crime (Macmilian 2002; GAO 2017). Some costs are relatively easy to measure. Others are effectively impossible to measure. For example, the value of property taken is one obvious cost of a burglary. For residential burglaries, the value might be measured by market prices or using the proxy of the victim’s renters or home owners insurance. The time it takes for police officers to make an arrest, transport the offender to some central location, and complete the required paper work can be valued in part by each patrol officer’s pro-rated wage. But in the burglary illustration, there are no easy measure of the costs from a victim’s damaged psychological well-being, perhaps manifested in increased anxiety about crime. In the arrest illustration, it is very difficult to measure the opportunity costs when the arresting police officers are not on patrol.
- 4.
Insurance companies have a somewhat different problem. Because the premiums they charge different customers are monetized, they need to forecast the expected costs in monetary units. If for a given customer the expected loss is $650 per year, the yearly premium for that customer needs to exceed $650 to cover administrative costs and then on the average to make a profit. Insurance companies do not cover costs that are not monetized. Insurance companies work to be monetize costs accurately, but the challenges are substantial.
- 5.
Some might argue that these kinds of issues provide a role for social science theory. However, that depends on what one means by theory. With respect to the role of age, how many geriatric offenders are found in crime movies since the 1930s? In popular culture at least, serious crime is largely for the young. Is that theory? Likewise, anyone who had analyzed data on the biographical factors related to violent crime has likely found the same thing. Is that theory? And even when criminology theory is reviewed specifically for the role of age, the mechanisms by which individuals “age out” of crime are unclear. Everything from changing concentrations of certain steroidal hormones to marriage has been proposed. Is that theory? There is no doubt that as an empirical matter, age is related to crime. A forecasting procedure finding otherwise surely needs to be very carefully scrutinized. If there really is theory to help, all the better. Far too often, the help purported to be based on subject-matter theory is really not.
- 6.
As noted earlier, potential predictors are of no help unless they are routinely and consistently collected. They must be available in the training data, the test data, and the data used for forecasting. Given the very large number of individuals who may be subject to criminal justice risk assessments, the data will be in at least the medium term limited to information that is already ordinarily collected. The setting will matters too. Prisons often have the time and resources to collect a wide variety of data. Pre-sentence reports assembled for individuals who have never been incarcerated will unlikely to have access to the same quality of information.
References
Berk, R. A. (2017) An impact assessment of machine learning risk forecasts on parole board decisions and recidivism. Journal of Experimental Criminology 13(2): 193–242.
Berk, R. A., Kriegler, B., & Baek, J-H. (2006) Forecasting dangerous inmate misconduct: An application of ensemble statistical procedures. Journal of Quantitative Criminology 22(2) 135–145.
Berk, R. A., Sorenson, S. B., & He, Y. (2005) Developing a practical forecasting screener for domestic violence incidents. Evaluation Review 29(4): 358–382.
Berk, R. A., Sherman, L., Barnes, G., Kurtz, E., & Ahlman, L. (2009) Forecasting murder within a population of probationers and parolees: A high stakes application of statistical learning. Journal of the Royal Statistics Society — Series A 172 (part I): 191–211.
Berk, R. A., & Bleich, J. (2013) Statistical procedures for forecasting criminal behavior: a comparative assessment. Journal of Criminology and Public Policy 12(3): 515–544, 2013.
Berk, R. A., Sorenson, S. B., & Barnes, G. (2016) Forecasting domestic violence: a machine learning approach to help inform arraignment decisions. Journal of Empirical Legal Studies 13(1): 94–115.
Berk, R. A., Heirdari, H., Jabbari, S., Kearns, M., & Roth, A. (2018a) Fairness in criminal justice risk assessments: The State of the Art. Sociological Methods and Research, in press.
Berk, R. A., Bjua, A., Brown, L., George, E., Kuchibhotla, A.K., Sue, W., & Zhau, L. (2018b) Assumption lean regression. arXiv:1806.09014v1 [stat,ME].
Blumstein, A., & Nakamura, K. (2009) Redemption in the presence of widespread criminal background checks. Criminology 47(2): 327–359.
Buja, A., Berk, R., Brown, L., George, E., Pitkin, E., Traskin, M., Zhan, K., & Zhao, L. (2018a). Models as approximations — part I: a conspiracy of nonlinearity and random regressors in linear regression.” arXiv:1404.1578
Buja, A., Berk, R., Brown, L., George, E., Arun Kumar Kuchibhotla, & Zhao, L. (2018b). “Models as approximations — part II: a general theory of model-robust regression.” arXiv:1612.03257
Calderon, A. (2018) A dangerous mind: can neuroscience predict how likely someone is to commit another crime. The Marshall Project, filed 8/14/2018.
Campbell, J.C., Webster,D., Koziol-McLain, J., Block, C.,RhD, Doris Campbell, D., Curry. M.A.,. Gary, F. Glass, N., McFarlane, J., Sachs, C., Sharps, P., Ulrich, Y., Wilt, S.A., Manganello, J., Xu, X., Schollenberger, J., Frye, V., & Laughon, K. (2003) Risk factors for femicide in abusive relationships: results from a multisite case control study. American Journal of Public Health 93(7): 1089–1097.
Freedman, D.A. (2009) Statistical ModelsCambridge, UK: Cambridge University Press.
GAO. (2017) Costs of Crime: Experts Report Challenges Estimating Costs and Suggest Improvements to Better Inform Policy Decision. United States Government Accountability Office, September, 2017.
Heissel, J.A., Sharkey, P.T., Torrats-Espinosa, G., Grant, K., & Adam, E.K. (2017) Violence and vigilance: the acute effects of community violent crime on sleep and cortisol. Child Development, published online, https://doi.org/10.1111/cdev. 12889.
Kiehl, K.A., Anderson, N.E., Aharonie, E., Maurer, J.M., Harenski, K.A., Rao, V., Claus, E.D., Harenski, C., Koenigs, M., Decety, J., Kosson, D., Wager, T.D., Calhoun, V.D., & Steel, V.R. (2018) Age pf gray matters: Neuroprediction and recidivism. Neuroimage: Clinical 19: 813–823.
Kleinman, M., Ostrom, B. J., & Cheeman, F. L. (2007) Using risk assessment to inform sentencing decisions for nonviolent offenders in Virginia. Crime & Delinquency 53(1): 1–27.
Kleinberg, J., Mullainathan, S., & Raghavan, M. (2017b) Inherent tradeoffs in the fair determination of risk scores. Proc. 8th Conference on Innovations in Theoretical Computer Science (ITCS).
Macmilian, R. (2002) Statistics: costs of crime. Encyclopedia of Crime and Justice. Encyclopedia.com. 16 Aug. 2018 http://www.encyclopedia.com
Megargee, E.I., Carbonell, J.L. & Mercer, S.J. (1999) MMPI-2 with male and female state and federal prison inmate. Psychological Assessment 11(2): 177–185.
Messing, J.J., Campbell, J., Webster, D.W., Brown, S., Patchell, B., & Wilson, J.S. (2015) The Oklahoma lethality Assessment Study: A Euasi-Experimental Evaluation of the Lethality Assessment Program.Social Service Review89(3):499–530.
Messing, J.J., Campbell, J., Wilson, J.S., Brown, S., & Patchell, B. (2017) The lethality screen: the predictive validity of an intimate partner violence risk assessment for use by first responders. Journal of Interpersonal Violence 32(2): 205–226.
Roehl, J., O’Sullivan, C., Webster, D., & Campbell, J. (2005) Intimate partner violence risk assessment validation study, final report. National Institute of Justice, U.S. Department of Justice.
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 2019 Springer Nature Switzerland AG
About this chapter
Cite this chapter
Berk, R. (2019). Some Important Background Material. In: Machine Learning Risk Assessments in Criminal Justice Settings. Springer, Cham. https://doi.org/10.1007/978-3-030-02272-3_2
Download citation
DOI: https://doi.org/10.1007/978-3-030-02272-3_2
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-02271-6
Online ISBN: 978-3-030-02272-3
eBook Packages: Computer ScienceComputer Science (R0)