There is a story the system tells itself every time politicians, prosecutors, and judges want a shortcut.
When someone cycles through jail or prison, the narrative often lands on the individual: unstable, untreated, unpredictable. Mental illness becomes their convenient explanation that sounds compassionate, but functions like blame; one that never really gets properly addressed at that. It implies entirely that the revolving door is a clinical problem, not a structural one.
A 2024 study in PLOS ONE challenges that story with something the carceral state respects more than testimony: prediction. The researchers asked a narrow, measurable question. In a general U.S. prison sample, do mental illness variables add meaningful predictive power for who has a prior incarceration history, once you account for basic crime and demographic variables?
Their answer was straightforward: no.
That finding does not mean mental illness is rare in prison because it absolutely is not. It also does not mean treatment is optional; it is not. It simply means something sharper and more politically inconvenient.
It means the cycle of incarceration is being driven by the system’s usual levers, and mental illness is often just the label we attach after the fact, failing to acknowledge that societal failures are a bigger piece to the puzzle than we’d like to admit.
What the Study Actually Did
Study: Cohen TR, Fronk GE, Kiehl KA, Curtin JJ, Koenigs M (2024), Clarifying the relationship between mental illness and recidivism using machine learning: A retrospective study. PLOS ONE 19(2): e0297448.
Sample: 394 incarcerated adults (322 men, 72 women) recruited from three Midwest prisons.
Mental health measurement: Structured clinical interviews (SCID-5) with 28 diagnoses plus symptom-count variables, alongside substance use diagnoses and symptom counts.
Algorithms: Elastic net logistic regression (GLMnet), k-nearest neighbors, and random forests, with repeated cross-validation.
Outcome definition (important): “Recidivism” was defined retrospectively as having a prior adult incarceration before the current incarceration at the time of assessment.
This is not a small detail. It shapes what the model can “see,” and it matters for how we interpret the results.
The Headline Result
The study tested whether adding mental illness and substance use features improved prediction beyond crime and demographics.
It did not.
In their primary comparisons, the probability that adding mental illness and substance use meaningfully improved accuracy over crime and demographics was not compelling. In exploratory comparisons against a null model, only the crime-and-demographics feature set reliably improved prediction.
Also notable: another medical issue that is wrongly treated as a moral one, substance use disorders, were extremely common in the sample (86.29%), which reduces variance and can blunt predictive signal.
The Finding People Will Misuse, and the Finding We Actually Need
The misuse
Some will hear “mental illness did not predict recidivism” and translate it into: stop funding mental health care in jails and prisons. That is not just wrong. It is dangerous.
The authors explicitly state that treatment remains essential because mental illness is prevalent in prisons and because treatment affects safety, functioning, dignity, and outcomes beyond recidivism.
The real takeaway
If mental illness is not adding predictive utility once you include crime and demographics, then the system’s favorite assumption starts to collapse:
The revolving door is not primarily a psychiatric mystery. It is a policy outcome.
And that changes where the conversation on accountability belongs.
Why This Result Makes Sense
1) “Recidivism” here is a prior-incarceration flag, not a future outcome
Because the outcome is retrospective, older people have had more time to accumulate prior incarcerations. Age becomes structurally important, even if age is not the “cause” of anything. The authors discuss this directly.
So if someone tries to use this study to claim they predicted future reoffending, that is not what happened.
2) The sample excludes some of the most severe diagnoses
Participants with a history of psychosis or bipolar disorder were excluded in the parent study, which limits conclusions about those conditions.
That does not weaken the paper. It defines what it can honestly claim.
3) “Mental illness” was measured as diagnosis history, not timing, treatment, or crisis state
A lifetime diagnosis is not the same thing as symptoms during reentry, instability at the time of arrest, medication disruption, untreated trauma, or access to care. The authors note the limitation of limited temporal specificity.
Carceral systems love static checkboxes because they are easy. Human beings are not static; we are complex and no one is exactly the same from day to day. The human experience needs to play a bigger role in all of this.
4) Crime-and-demographics often proxy structural exposure
The criminal legal system is not a neutral measuring device. “Crime variables” and “demographics” can reflect enforcement patterns, supervision intensity, charging practices, and resource scarcity.
When those features predict prior incarceration, the model may be picking up system behavior as much as individual behavior. That is exactly why this paper is politically important.
What This Means for Risk Tools and “Objective” Decision-Making
A lot of jurisdictions treat risk assessments like neutral math, and more or less that is intellectually dishonest to operate that way.
This study reinforces a hard truth: the variables that predict incarceration history are often the variables most entangled with how the system has already treated someone.
If a tool is trained to predict “recidivism,” and “recidivism” is anchored to prior incarceration, the tool can become a machine for laundering the past into the future. It can turn history into destiny, while calling it science.
This paper does not build a new risk tool. It does something more valuable. It tests whether “mental illness” is really carrying predictive weight, or whether it is being used as narrative filler. In this sample, it is the filler.
What This Reveals About Societal Failure
If mental illness does not meaningfully predict who cycles back into prison once crime and basic demographics are accounted for, then the explanation politicians often rely on starts to fall apart. The problem is not primarily individual pathology. It is exposure to systems that fail repeatedly and predictably.
This study suggests that what drives reincarceration is less about who someone is and more about what they have already been subjected to. Education gaps, early system contact, age at first arrest, and offense categorization matter more than diagnostic labels. Those are not clinical failures. They are social ones.
For policymakers, this creates an uncomfortable reality. You cannot meaningfully reduce recidivism by expanding punitive controls, tightening supervision, or demanding personal responsibility while leaving housing instability, employment exclusion, underfunded education, and uneven enforcement untouched. The data does not support that theory of change.
Mental illness becomes a convenient explanation precisely because it keeps responsibility contained within the individual. And structural failure does the opposite; it does not care about politicians and their egos, and instead it forces acknowledgment that incarceration patterns reflect policy choices made upstream and sustained over time. This study does not excuse harm or wrongdoing. It simply makes clear that the conditions producing repeated incarceration are collective, not incidental, and far more difficult to campaign against than a diagnosis.
The Clutch Justice Policy Read
If we as a society want fewer people cycling back into cages, here is what this study suggests, indirectly but clearly:
Stop treating mental illness as the system’s alibi
Mental illness is real. Suffering is real. But the story that mental illness explains the revolving door is often how the system blatantly avoids looking at its own mechanics.
Treat mental illness because it is a legal and human imperative, not because it is a recidivism lever
Even when mental illness does not predict reincarceration, treatment matters for safety, function, dignity, and institutional harm reduction. We need to understand that mental illness is often a symptom. Do some mentally ill people commit crimes simply because they are mentally ill? Sure. But it is not the sole cause.
Move resources to the reentry choke points that actually drive returns
Employment barriers, housing instability, supervision tripwires, untreated substance use with real-time relapse risk, and enforcement patterns often do more to determine who returns than a diagnostic label ever will.
If your intervention plan starts and ends at the diagnosis, you are treating the paperwork, not the person.
A Cleaner Way to Say It
The system likes simple narratives and this paper disrupts one of the simplest.
Mental illness is not the sole engine of recidivism in the way the public has been taught to believe. In this study, the better predictors were basic crime and demographic factors.
So the next time someone says “we need tougher controls because mental illness makes people reoffend,” ask the quieter question:
Are we responding to risk, or are we rationalizing a cycle we refuse to reform?
