The accuracy, fairness, and limits of predicting recidivism

See allHide authors and affiliations

Science Advances  17 Jan 2018:
Vol. 4, no. 1, eaao5580
DOI: 10.1126/sciadv.aao5580

eLetters is an online forum for ongoing peer review. Submission of eLetters are open to all . Please read our guidelines before submitting your own eLetter.

Compose eLetter

Plain text

  • Plain text
    No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.
Author Information
First or given name, e.g. 'Peter'.
Your last, or family, name, e.g. 'MacMoody'.
Your email address, e.g.
Your role and/or occupation, e.g. 'Orthopedic Surgeon'.
Your organization or institution (if applicable), e.g. 'Royal Free Hospital'.
Statement of Competing Interests

This question is for testing whether or not you are a human visitor and to prevent automated spam submissions.

Vertical Tabs

  • The Importance of Mass Trust in Validating Algorithms in a Public Space
    • Philip D. Waggoner, Political Scientist, College of William & Mary
    • Other Contributors:
      • Ryan Kennedy, Political Scientist, University of Houston

    Dressel and Farid (2018) have offered an important and timely assessment of recidivism algorithms used in hybrid decision making environments, such as in a court room where judges are making decisions of criminal sentencing with the “help” of algorithms.

    A thorough starting place, there exists the need to build on these findings by bringing in the role of the public, or mass trust, in this conversation. In such a public space as a court room coupled with the broader proliferation of algorithmic advice influencing so many tasks of daily life for virtually every person, the degrees to which people trust these algorithms to do what they are purported to do (e.g., accurately predict the likelihood of recidivism) is intimately tied to the level of trust placed in the algorithms by the mass public. If people, especially in a democracy such as America, do not trust the advice offered by algorithms, then the likelihood of continued algorithmic expansion and integration into public life, especially in public spaces like court rooms, will likely be limited.

    Therefore, it is worth pointing out that Dressel and Farid (2018) offer a valuable starting place to push the dialogue forward by exploring the role of mass trust in algorithms as a form of legitimization of algorithms in public spaces. Similar contexts such as military decision making/autonomous weapons systems (Horowitz 2016) and Congressional redistricting (Bernstein and Duchin 2017), which are prominent locati...

    Show More
    Competing Interests: None declared.
  • RE: The accuracy, fairness, and limits of predicting recidivism
    • Ashley Martin, Quality Assurance Manager, SC Department of Probation, Parole and Pardon Services

    I am the Quality Assurance Manager for the South Carolina Department of Probation, Parole and Pardon Services. Previously, I was the department's subject matter expert for risk assessments, during which time my department did and still does, use the COMPAS risk assessment.

    The main issue I see with this particular criticism, and many others, is the myopic view of the assessment. This assessment tool does more than simply provide a chance of re-arrest. It also provides risk levels for an array of criminogenic needs that can be informative when developing a supervision plan for a client under our supervision. It also provides important data for resource allocation through our department (we are a state department). For example, we are exploring ways to help clients afford their court ordered therapy by helping to pay for some of it. To ensure these resources are used by the clients that need it most, we rely on the risk level provided by the COMPAS assessment and the criminogenic need risk in particular to quickly locate those clients.

    Equivant actually provides sort of triage tool to alert staff of when a full assessment is required (the 137 question one, which only a minority of our clients require) which is 4 questions long and provides simply general risk of recidivism, should a department elect to use it. Many departments used a much smaller and quicker assessments before the 4th gen assessments like COMPAS became more mainstream to determine risk....

    Show More
    Competing Interests: None declared.

Stay Connected to Science Advances

Navigate This Article