Abstract
In 2018, Congress passed the First Step Act, amending 18 U.S.C. § 3582(c)(1)(A) to permit federally incarcerated individuals to directly petition district courts for early release. In the wake of the COVID-19 pandemic, thousands of individuals did so, citing staggering prison death rates and asserting that further incarceration unjustly imperiled their health. Largely unprepared for the deluge, district courts granted 25.7 percent of those petitions. Without precedent to guide them, their decisions were and remain highly discretionary, based on dozens of interrelated variables and legal factors. Those decisions are also largely insulated from appellate review.
The FSA’s amendments and the COVID-19 pandemic created a “natural experiment” to evaluate compassionate-release decision making, igniting a maelstrom of sentencing- and prison-reform debate. For some, judicial discretion over compassionate release offers a new tool for decarceration and a second look at sentencing. Others, however, fear that the exercise of judicial discretion in compassionate release will cause disparate outcomes. Despite this recent attention, none have sought to comprehensively identify those variables that might drive such decision making, much less model compassionate-release decision making.
Until now. Using cutting-edge machine-learning tools in a mixed- methods research design, we built two algorithmic models to analyze the impact of 149 variables on 553 compassionate-release decisions. Our first, “all-inclusive” model determined with ninety-nine percent accuracy that variables like health, public safety, sentence length, and time served motivated compassionate-release outcomes. But our second, “forecasting” model—which included a subset of exclusively factual variables uninfluenced by judicial assessment—dug deeper, exposing with seventy-eight percent accuracy what factors actually impact compassionate-release outcomes. We determined that longer-tenured, Democrat-appointed judges were more likely to grant compassionate release, and that women, those who have served significant sentences, those who have low offense-level scores, and those who have been given short sentences were more likely to be released early.
Our findings are startling, confirming and challenging prior scholars’ hypotheses about compassionate release. Given the U.S. Sentencing Commission’s 2023 amendments to the Sentencing Guidelines—which codified the broad role that judicial discretion plays in compassionate release—our study contributes a novel theoretical model and methodological approach to better understand judicial decision making. It also aids judges and scholars alike when evaluating the normative and legal implications of compassionate-release and sentencing outcomes.
Recommended Citation
Max Alderman, Arash Barfar & Dana Edberg,
Explaining Compassionate Release: A Machine Learning Study of Sentencing Modifications & Judicial Decision Making,
59 Loy. L.A. L. Rev. 1
().
Available at: https://digitalcommons.lmu.edu/llr/vol59/iss1/1
