Suspension’s Hidden Toll: How School Exclusion Fuels Behavioral Problems in Autistic Teens

Autistic teenagers in the United Kingdom are more than four times as likely as their peers to be suspended or expelled from school, and a new study suggests that this disciplinary removal is not merely a marker of existing trouble but an active pathway toward worsening behavioral health. Drawing on more than a decade of data from a nationally representative birth cohort, researchers at University College London have shown that school exclusion partially explains why autistic adolescents go on to develop elevated externalising difficulties—problems such as hyperactivity, aggression, impulsivity and conduct issues—by the age of 17. The findings, published in the Journal of Autism and Developmental Disorders, position exclusion as a rare thing in autism research: a modifiable environmental mechanism that schools and policymakers could actually change.

The study, led by Molly Yeoman with colleagues including Vaso Totsika, analysed data from the UK Millennium Cohort Study, a longitudinal project following children born between 2000 and 2002 across seven waves of data collection spanning from 9 months to 17 years of age. Of the 10,228 participants with usable autism data, 160 had a parent-reported autism spectrum condition diagnosis recorded between the ages of 5 and 11. Parents also reported whether their child had been temporarily suspended or permanently excluded between ages 11 and 14, and the adolescents themselves completed the well-validated Strengths and Difficulties Questionnaire at 17, providing self-reported scores for both internalising difficulties—anxiety, depression and social withdrawal—and externalising difficulties.

The raw numbers were stark. By age 14, 21.9 percent of autistic children in the cohort had been excluded or suspended from school, compared with just 4.6 percent of their non-autistic peers. By 17, autistic adolescents reported significantly higher scores on both dimensions of mental health: internalising scores averaged 19.46 in the autism group versus 18.21 among peers, and externalising scores averaged 18.50 versus 17.34. After adjusting for a comprehensive set of confounders—sex, ethnicity, maternal age, socioeconomic deprivation, family structure, child health, prior behavioral difficulties, maternal psychological distress and parental school engagement—autism remained associated with 1.43 points higher internalising scores and 0.78 points higher externalising scores, and with more than double the odds of exclusion (an odds ratio of 2.32).

What distinguishes this study from earlier work is its statistical machinery. Simple associations between exclusion and poor mental health are notoriously difficult to interpret, because children who act out are more likely to be excluded in the first place, making the direction of cause and effect ambiguous. The research team confronted this head-on using causal mediation analysis under the counterfactual framework, employing the g-formula to estimate randomised interventional analogues of natural direct and indirect effects. This approach matters because the data contained intermediate confounders—prior externalising difficulties and maternal mental health—that are themselves influenced by autism and that also shape both exclusion and later outcomes. Simpler mediation methods would have produced biased estimates in this situation, so the g-computation procedure was essential for cleanly decomposing the total effect of autism into its direct and exclusion-mediated components.

The decomposition produced a clear and, in some ways, surprising result. For externalising symptoms, the total effect of autism was 0.69 points, of which 0.66 was a direct effect and 0.03 travelled through school exclusion—a small but statistically significant indirect pathway (95 percent confidence interval 0.004 to 0.06, p = 0.024). In other words, exclusion accounted for a modest slice of the association, estimated at roughly 4.9 percent, though the proportion itself was not statistically significant. For internalising symptoms, by contrast, the indirect effect was essentially zero (−0.0008, p = 0.936), with the entire association between autism and later anxiety and depression operating through the direct pathway. Exclusion, it seems, is a mechanism for behavioral deterioration specifically, not emotional deterioration generally.

The theoretical interpretation offered by the authors draws on the idea that schools function as protective environments, particularly for autistic young people who often depend heavily on routine, structure, predictable schedules and supportive adult supervision to regulate their behavior and emotions. When exclusion removes these scaffolds during adolescence—a developmental window already marked by heightened emotional reactivity and social pressure—the result may be a destabilising loss of exactly the supports that many autistic students rely on most. This aligns with Problem Behaviour Theory, which holds that adverse shifts in environmental context can interact with individual vulnerabilities, such as cognitive inflexibility, to elevate the risk of behavioral difficulties. Consistent with this account, research on residential mobility has shown that disrupting embedded routines and social networks predicts externalising problems more strongly than internalising ones, a pattern that mirrors the asymmetry observed here.

The findings also carry a subtler implication: exclusion does not simply reflect pre-existing behavioral problems but actively contributes to their development and escalation. Earlier studies in neurotypical populations, including work by Ford and colleagues and by Obsuth and colleagues, had already suggested that exclusion predicts later psychiatric difficulties even after accounting for prior vulnerabilities. The new study extends that logic to autistic adolescents, showing that exclusion operates as a pathway through which autism leads to externalising outcomes, rather than merely a consequence of them. A sensitivity analysis restricted to temporary suspensions only produced a similar-sized indirect effect that fell short of significance, possibly because fewer participants experienced only temporary exclusion, or because shorter removals from school are genuinely less damaging than permanent expulsion.

Why did exclusion fail to mediate internalising difficulties, when qualitative research has documented shame, low self-worth and social isolation among excluded autistic students? The authors offer several explanations. The direct pathway may reflect core features of autism itself, such as intolerance of uncertainty and sensory sensitivities, which generate emotional distress independently of school events. Measurement differences may also play a role: this study relied on adolescent self-report of internalising symptoms, whereas earlier work used parent reports, teacher reports, broader distress measures or retrospective interviews, each of which may capture different facets of emotional functioning. There is also the possibility of heterogeneity in individual experience—for some autistic adolescents facing chronic peer victimisation, which can affect a large majority of autistic students, exclusion may even offer temporary relief from bullying or the exhausting demands of camouflaging, potentially suppressing the group-level indirect effect.

The study is not without limitations, and the authors are candid about them. The autistic subsample of 160 participants is small, producing sparse data in some covariate strata and a positivity violation that the g-formula could only address through extrapolation, so the estimates warrant caution. Survey weights were not applied, limiting population representativeness, and residual confounding cannot be fully excluded despite the careful use of a directed acyclic graph to guide covariate selection. Missing data, ranging up to 34 percent across variables, were handled with multiple imputation by chained equations under a missing-at-random assumption, and the mediation models were run with 500 bootstrap replications to generate bias-corrected confidence intervals.

Even with those caveats, the practical message is pointed. Because exclusion is a modifiable environmental factor, the findings argue for preventative, autism-informed strategies that reduce reliance on exclusionary discipline altogether—better staff training, flexible policies, tailored support plans and structured reintegration programs such as low-stimulation learning hubs with one-to-one support. They also suggest that mental health outcomes in autism are not fixed by neurology alone but are shaped by social and institutional contexts that unfold over time. If a small but real share of behavioral difficulties in autistic adolescents traces back to the moment a school removes them from its community, then keeping that community intact, responsive and informed may be one of the most concrete levers available for improving autistic young people’s lives.

Subject of Research: The mediating role of school exclusion in the association between autism and adolescent internalising and externalising mental health outcomes

Article Title: School Exclusion as a Partial Mediator Between Autism and Adolescent Mental Health

Article References: School Exclusion as a Partial Mediator Between Autism and Adolescent Mental Health. (n.d.). https://doi.org/10.1007/s10803-026-07550-7

Image Credits: AI Generated

DOI: 10.1007/s10803-026-07550-7

Keywords: autism spectrum conditions, school exclusion, adolescent mental health, externalising difficulties, internalising difficulties, Millennium Cohort Study, causal mediation analysis, g-formula, Strengths and Difficulties Questionnaire, inclusive education, school discipline, UK cohort study