Study Risks Funneling Autistic Youth into Justice System
How Bad Measures Become Policy
A new study from the Portman Clinic (Tavistock and Portman NHS Foundation Trust) and the University of East Anglia Autism — Spectrum Disorder Traits and Other Offending Risk Factors in a London-Based Forensic Youth Population — claims that autistic traits, especially alongside substance use or adverse childhood experiences (ACEs), are linked to more violent offences and greater gang involvement in youth caught up in London's justice system. The title is clunky but the takeaway is dangerously simple — and the measures don’t support it. When a study counts thoughts and exposure as offences and folds “suspected” and diagnosed autism into one box, it stops answering the question it asked and starts manufacturing risk.
What Was Actually Counted
The outcome is not how often or how serious an offence was. It is a count of categories. Inside those categories sit items that are not acts at all. “Violence” can include having violent thoughts or viewing violent material. “Gang involvement” can include being at risk of exploitation. Once thoughts and exposure are treated as offences, the measure is no longer tracking criminality; it tallies a mixed pile of actions, exposures and private thoughts. That makes young people look more variously “offending” than they are.
Why This Breaks the Construct
In research, construct validity means what is being measured matches the hypothesis, the initial claim. Here it most definitely does not. You cannot claim “more violent offences” while counting non‑acts inside violence. You cannot claim “more gang involvement” while counting ‘risk of child criminal exploitation’ as if it were actual involvement. The construct is mis-specified at the foundation, so the headline is built to scare.
The Bucket Problem
The study merges confirmed autism with “suspected” autism into one “ASD traits” bucket. That may raise statistical power, but it also raises the risk of misclassification: people get put in the wrong bucket. In justice‑adjacent settings, where access to assessment is uneven and clinician bias is documented, a “suspected” note can move a young person into the autism column without a diagnosis. In practice, the model may be learning clinician labeling habits more than autistic status.
The Headline Outruns the Tables
Even if we bracket the construct errors, the internal signals are mixed. Some coefficients do not match the headline direction. Nuance inside the tables becomes certainty in the abstract. That is how stigma spreads: the simple line leaves the building while the caveats stay on page five.
Built to Travel into Punishment Systems
I use “carceral” to mean the network of punishment systems: school threat assessments, police triage, probation screens, court reports. A measure that counts thoughts as offences and treats suspicion as diagnosis is designed to be funneled into those systems. It creates a “risk” label that is cheap to apply and hard to shed. Racialized autistic youth will feel it first. The paper itself notes disparities and then proceeds with a model that can easily entrap them.
What a Fair Test Would Look Like
If the question is about offending, start with verified acts: charges, convictions or independently corroborated incidents. Keep exposure and internal experience as separate constructs and interpret them as such. Analyze confirmed diagnoses apart from suspected cases. Publish sensitivity analyses that remove the non‑act items and the suspected cases and report whether the claims survive. Pre‑register the primary outcomes and expected directions so the headline cannot drift after the fact. Above all, model exclusion, profiling and service failure as primary drivers, and test whether autism moderates those drivers rather than serving as a stand‑alone cause.
What Is Worth Keeping
The integrative ambition matters. Routine clinical data, adverse experiences and co‑occurring conditions belong in the same frame. The problem is not the scope; it is the mis-specified outcome and the fuzzy exposure. Fix those and the same data could support better questions and better care.
Editors, Please Fix This
This does not require impugning motive. It requires editorial repair. Correct the abstract and methods so that acts are distinguished from exposures and private thoughts, and confirmed diagnoses are separated from suspicion. If the main claims weaken under those corrections, issue a formal correction or an expression of concern. Readers and practitioners need a clear signal about what the study can and cannot show.
Bottom Line
I get that the researchers' point was to explore complex risk interactions in order to improve care and prevention. That aim is understandable. But the measurement choices (broad ‘offence’ definitions; merged diagnosis categories) risk reinforcing harmful stereotypes, regardless of the researchers’ intentions. Measurement choices like these are unacceptable when autistic youth are at risk of being caught up in the justice system.
Autism is not criminogenic. Bad measures are. When research collapses thoughts and exposure into “offences,” and merges suspicion with diagnosis, it builds a story that will be used to surveil autistic youth and call it care. That is intellectual malpractice. We can do better science — and we should, because autistic lives depend on it.