Machine Learning Forecasts of Risk to Inform Sentencing Decisions

Berk, Richard A.; Hyatt, Jordan M.

doi:10.1525/fsr.2015.27.4.222

Cited by 49 publications

(32 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…The regression‐based actuarial risk assessment instruments that currently dominate the field and which are relied on in this article may soon become obsolete. As we progress through the current technology revolution, we will increasingly see machine learning and artificial intelligence approaches in which computers harness more and more forms of data to make predictions based on intricate combinations of factors that are difficult for humans to parse (see Berk & Hyatt, 2015; Hess & Turner, 2017). These approaches will present new opportunities and raise new concerns.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Risk assessment savvy: The imperative of appreciating accuracy and outcome

Hester

2020

Behavioral Sci & The Law

View full text Add to dashboard Cite

This article focuses on two aspects of actuarial risk at sentencing, the accuracy of the instrument and the outcome it predicts. For theoretical reasons rooted in the cognitive decision-making and sentencing literature, there is a danger that judges and other practitioners might come to overly rely on a "high risk" label or designation without appreciating the accuracy of the prediction or the actual outcome being predicted. Using sentencing and recidivism data from Pennsylvania (n = 10,000), two simple risk instruments are constructed to illustrate the critical importance of understanding accuracy and outcome before relying on the risk tool information.

show abstract

Section: Discussionmentioning

confidence: 99%

“…3 3 In the conclusion I mention an alternative (and emerging) method of machine learning. For more on machine learning‐based risk assessments in the criminal justice context, see Berk and Hyatt (2015). …”

mentioning

confidence: 99%

Risk assessment savvy: The imperative of appreciating accuracy and outcome

Hester

2020

Behavioral Sci & The Law

View full text Add to dashboard Cite

show abstract

“…Second, our setting is particularly novel in that inspection data allow us control for food risk at the time complaints or Yelp reviews are made, hence allowing us to credibly assess bias of ordinary consumers. While much research documents racial disparities in algorithmic predictions, it is often much less clear how the alternative (typically, human judgment) fares (see, e.g., Berk and Hyatt, 2015). Third, our study is the first extension beyond P&S to study conditions under which their proposed marginalization can address questions of racial bias outside of the linear (or generalized linear) setting.…”

Section: Introductionmentioning

confidence: 95%

When Algorithms Import Private Bias into Public Enforcement: The Promise and Limitations ofStatistical Debiasing Solutions

Altenburger¹,

Ho²

2018

JITE

View full text Add to dashboard Cite

We make two contributions to understanding the role of algorithms in regulatory enforcement. First, we illustrate how big-data analytics can inadvertently import private biases into public policy. We show that a much-hyped use of predictive analytics-using consumer data to target food-safety enforcement-can disproportionately harm Asian establishments. Second, we study a solution by Pope and Sydnor (2011), which aims to debias predictors via marginalization, while still using information of contested predictors. We find the solution may be limited when protected groups have distinct predictor distributions, due to model extrapolation. Common machine-learning techniques heighten these problems.

show abstract

“…Indeed, error rates are the reciprocal, or flip side, of the discriminative and calibration accuracy statistics in the 2 × 2 framework in Table 2. Thus, Table 6 Because judges at sentencing are predominantly interested in making decisions based on predictions of the individual defendant's likelihood of future recidivism, 80 the discussion now focuses on the forecasting errors of FDR and FOR. The FDR is the forecasting false positive error rate.…”

Section: Error Ratesmentioning

confidence: 99%

Judicial gatekeeping on scientific validity with risk assessment tools

Hamilton

2020

Behavioral Sci & The Law

View full text Add to dashboard Cite

Risk assessment tools driven by algorithms offer promising advantages in predicting the recidivism risk of defendants.Jurisdictions are increasingly relying upon risk tool outcomes to help judges at sentencing with their decisions on whether to incarcerate or whether to use community-basedsanctions. Yet as sentencing has significant consequences for public safety and individual rights, care must be taken that the tools relied upon are appropriate for the task.Judges are encouraged to act as gatekeepers to evaluate whether the forensic risk assessment tool offered has a sufficient level of validity in that it is fit for the purposes of sentencing, provides an acceptable level of accuracy in its predictions, and achieves an adequate standard of reliability with regard to its outcomes.

show abstract

Machine Learning Forecasts of Risk to Inform Sentencing Decisions

Cited by 49 publications

References 12 publications

Risk assessment savvy: The imperative of appreciating accuracy and outcome

Risk assessment savvy: The imperative of appreciating accuracy and outcome

When Algorithms Import Private Bias into Public Enforcement: The Promise and Limitations ofStatistical Debiasing Solutions

Judicial gatekeeping on scientific validity with risk assessment tools

Contact Info

Product

Resources

About