Abstract
Risk assessment algorithms lie at the heart of criminal justice reform
to tackle mass incarceration. The newest application of risk tools
centers on the pretrial stage as a means to reduce both reliance upon
wealth-based bail systems and rates of pretrial detention. Yet the ability
of risk assessment to achieve the reform movement’s goals will be challenged
if the risk tools do not perform equitably for minorities. To date,
little is known about the racial fairness of these algorithms as they are
used in the field. This Article offers an original empirical study of a popular
risk assessment tool to evaluate its race-based performance. The case
study is novel in employing a two-sample design with large datasets from
diverse jurisdictions, one with a supermajority white population and the
other a supermajority Black population.
Statistical analyses examine whether, in these jurisdictions, the
algorithmic risk tool results in disparate impact, exhibits test bias, or displays
differential validity in terms of unequal performance metrics for
white versus Black defendants. Implications of the study results are informative
to the broader knowledge base about risk assessment practices
in the field. Results contribute to the debate about the topic of algorithmic
fairness in an important setting where one’s liberty interests may be
infringed despite not being adjudicated guilty of any crime.