Order Michigan Supreme Court
Lansing, Michigan
July 23, 2021 Bridget M. McCormack,
Chief Justice
161915 & (14) Brian K. Zahra
David F. Viviano
Richard H. Bernstein
Elizabeth T. Clement
PEOPLE OF THE STATE OF MICHIGAN, Megan K. Cavanagh
Plaintiff-Appellee, Elizabeth M. Welch,
Justices
v SC: 161915
COA: 353965
Van Buren CC: 2019-021950-FC
JOSE HUMBERTO CANEDO,
Defendant-Appellant.
_________________________________________/
On order of the Court, the motion to amend the application for leave to appeal is
GRANTED. The application for leave to appeal the July 31, 2020 order of the Court of
Appeals is considered, and it is DENIED, because we are not persuaded that the
questions presented should be reviewed by this Court.
CAVANAGH, J. (concurring).
I concur in the order denying leave in this case, but write separately to discuss
some questions regarding the use of the Correctional Offender Management Profiling for
Alternative Sanctions (COMPAS) software tool during a defendant’s sentencing.
Generally, the Michigan Department of Corrections (MDOC) has used risk assessment to
make internal decisions about programming and placement. However, the extension of
that practice from informing the MDOC’s work after sentencing to informing the
sentencing decision seems consequential.
The COMPAS software tool creates a risk assessment “intended to measure the
‘likelihood of future Violent or Non-Violent Felony Offenses.’ ” People v Younglove,
unpublished per curiam opinion of the Court of Appeals, issued February 21, 2019
(Docket Nos. 341901, 342497, 342598, and 344475), quoting MDOC, Field Operations
Administration, Administration and Use of COMPAS in the Presentence Investigation
Report (March 2017), p 10, available at
https://www.michbar.org/file/news/releases/archives17/COMPAS-at-PSI-Manual-2-27-
17-Combined.pdf (accessed July 15, 2021) [https://perma.cc/YB5S-D3WL]. This
assessment is created through a proprietary algorithm that takes data inputs including
criminal history, age, employment status, education level, community ties, substance
abuse, and more. The algorithm’s output is an assessment that purports to represent the
probability a defendant will engage in future criminal conduct. See generally
Administration and Use of COMPAS; see also State v Loomis, 371 Wis 2d 235, 245
(2015).
Due process requires that a defendant be sentenced on the basis of accurate
2
information, People v Francisco, 474 Mich 82, 88 (2006), and a defendant must have “a
reasonable opportunity at sentencing to challenge the information” contained in the
presentence investigation report (PSIR), People v Zinn, 217 Mich App 340, 347-348
(1996). However, in the context of COMPAS risk assessments, it is unclear to me what it
might mean to measure the accuracy of a prediction about an individual’s future conduct
and how that prediction might be challenged without knowing how it was formulated.
See Loomis, 371 Wis 2d 235 (limiting the use of COMPAS at sentencing and mandating
that written warnings accompany any COMPAS attached to a PSIR). One evaluation of
the COMPAS tool, which was prepared for the California Department of Corrections and
Rehabilitation (CDCR), concluded there was “ ‘no sound evidence that the COMPAS can
be rated consistently by different evaluators, that it assesses the criminogenic needs it
purports to assess, and (most importantly) that it predicts inmates’ recidivism for CDCR
offenders.’ ” Id. at 262, quoting Skeem and Loudon, Assessment of Evidence on the
Quality of the Correctional Offender Management Profiling for Alternative Sanctions
(COMPAS) (2007), p 5, available at (accessed July 15, 2021) [https://perma.cc/PR5D-6N6A].
The algorithm COMPAS employs is proprietary, and undisclosed. Loomis, 371
Wis 2d at 258. The secretive nature of the algorithm raises questions. Without knowing
what the algorithm is, it is difficult to know whether and how race, class, and other
personal factors influence a potentially biased score. One investigation, for example,
concluded that Black defendants “ ‘were far more likely than white defendants to be
incorrectly judged to be at a higher risk of recidivism.’ ” Id. at 263, quoting Larson et al,
ProPublica, How We Analyzed the COMPAS Recidivism Algorithm (May 23, 2016),
available at (accessed July 15, 2021) [https://perma.cc/AP85-5EDE].
Additionally, in order to be accurate and mitigate these possible errors, risk assessment
tools “ ‘must be constantly re-normed for changing populations and subpopulations.’ ”
Loomis, 371 Wis 2d at 263-264, quoting Klingele, The Promises and Perils of Evidence-
Based Corrections, 91 Notre Dame L Rev 537, 576 (2015). It is unclear whether
COMPAS regularly updates its software accordingly.
The many criticisms that such risk assessments have drawn from public officials
and scholars create concerns about the use of COMPAS in sentencing. However,
defendant has not fully raised these issues in this Court. Accordingly, I concur in the
order denying leave to appeal.
I, Larry S. Royster, Clerk of the Michigan Supreme Court, certify that the
foregoing is a true and complete copy of the order entered at the direction of the Court.
July 23, 2021
b0720
Clerk