Member-only story
Hidden Algorithms, Bad Science, and Racial Discrimination in the D.C. Juvenile Justice System
How D.C.’s juvenile system incarcerates its youth based on a shoddy SAVRY algorithm, unreliable analyses, and the “criminality” of their parents. Why and how this needs to change.
Since 2004, Washington, D.C.’s Juvenile Justice System has used a risk assessment algorithm that influences juvenile sentencing outcomes under the banner of scientific rigor, fairness, and public safety.
However, there are just two problems: the algorithm and methods that D.C. uses to send kids to jail are based on junk science, and the courts seem obsessed over incarcerating young people of color.
Sound like maybe we have a problem?
A Questionable Algorithm
As reported in AI Now’s 2018 Litigating Algorithms report, juvenile courts in D.C., like many other jurisdictions, increasingly rely on risk assessment tools to facilitate interventions for juvenile offenders. But as I learned from AI Now’s reporting and redacted court documents, there are serious issues with the algorithms behind the risk assessments and the juvenile system’s application of those risk assessments.
The algorithm in question is known as SAVRY or the Structured Assessment of Violence Risk in Youth. At surface-level, the argument for an algorithm like SAVRY appears to make sense — leverage a validated model to assess violence risk for youths and develop evidence-based interventions. The SAVRY algorithm takes into account 24 risk factors and produces a series of coded scores in the form of high, medium, or low. When used as intended, SAVRY can be used as part of an intervention planning process; according to SAVRY’s authors, “statistical formulas generally do as well or better than clinical judgments.”
However, a cursory study of court documents reveals SAVRY to be anything but a validated model. According to an expert witness, there are five central issues that raise…
