Exercise May Gently Boost Social Skills in Autistic Children, but Evidence Is Fragile

Can something as simple as running, swimming, or playing basketball help autistic children connect more easily with the people around them? A new systematic review and meta-analysis published in the Journal of Autism and Developmental Disorders suggests that physical exercise may offer a small benefit for social skills in children with autism spectrum disorder, but the authors urge caution: the overall finding hinges heavily on a single influential study, and when that study is removed, the statistical signal fades to insignificance. The work, led by Qianping Tan and Zhuoling Lei of Southwest University in Chongqing, together with Jindong Chang and Lei Yao, offers one of the most methodologically sober assessments to date of a question that parents, therapists, and coaches have been asking for decades.

The research team set out to answer a deceptively straightforward question with the most rigorous tools available. Social skill deficits are a core feature of autism spectrum disorder, affecting how children initiate interactions, interpret social cues, and maintain relationships. While behavioral therapies remain the mainstay of intervention, physical exercise has attracted growing interest as a complementary approach, partly because it is inexpensive, accessible, and carries well-documented physical health benefits. Previous meta-analyses, including work by Sowa and Meulenbroek in 2012 and Healy and colleagues in 2018, had reported encouraging effects, but questions persisted about study quality, publication bias, and whether the apparent benefits would survive stricter scrutiny.

To address those questions, the researchers searched four major databases—PubMed, Web of Science, Embase, and Scopus—from their inception through January 10, 2025, using an exhaustive search strategy that combined terms for exercise, autism, childhood, and social skills. They restricted inclusion to randomized controlled trials, the gold standard for testing causal effects, because non-randomized designs are vulnerable to selection bias and inflated effect estimates. After screening against predefined eligibility criteria, eleven randomized controlled trials involving 349 children made it into the final analysis. Methodological quality was evaluated with the Cochrane Risk of Bias Tool, and the statistical work was carried out in RevMan 5.3.5 and Stata 16.0, with subgroup, sensitivity, and publication-bias analyses built into the plan from the start.

The headline result was modest but statistically significant. Pooling the eleven trials under a random-effects model, the team calculated a standardized mean difference of 0.34, with a 95 percent confidence interval of 0.03 to 0.65 and a p-value of 0.03. In plain terms, children who participated in structured exercise programs scored somewhat higher on social skill measures than children in control conditions, and the heterogeneity statistic, I-squared, came in at 49 percent, indicating moderate variability among studies. A standardized mean difference of 0.34 falls into the small-effect range by conventional conventions, meaning the average treated child would fare better than roughly 63 percent of untreated children—a real but far from transformative shift.

Then came the twist that gives this meta-analysis its unusual honesty. In a post-hoc sensitivity analysis, the researchers removed one statistically influential study and re-ran the model. With ten trials remaining, the pooled effect shrank to a standardized mean difference of 0.19, the confidence interval widened to span zero, from minus 0.03 to 0.41, and the p-value rose to 0.09, which fails the conventional threshold for statistical significance. Notably, heterogeneity collapsed to zero, suggesting the remaining studies were internally consistent but collectively inconclusive. This kind of fragility analysis is increasingly recognized as essential in evidence synthesis, because a pooled estimate that depends on a single trial can mislead clinicians and policymakers if that trial happens to be an outlier for reasons unrelated to the intervention itself.

The team also probed whether the length of the intervention mattered, dividing studies into duration subgroups. The answer was no: differences among duration subgroups were not statistically significant, with a p-value of 0.61. Longer programs were not clearly superior to shorter ones, a finding that complicates the intuitive assumption that more practice should yield more social gains. On the question of publication bias—the tendency for small or null studies to go unpublished—Egger’s tests detected no significant small-study effects in either the primary analysis, with a p-value of 0.432, or the sensitivity analysis, with a p-value of 0.263. Even so, the authors note that publication bias could not be definitively excluded, a caveat that matters in a field where small trials of behavioral interventions are notoriously prone to selective reporting.

The eleven included trials spanned a striking diversity of exercise modalities, reflecting the creativity of researchers in this space. The broader literature from which these trials emerge includes mini-basketball training programs, which one cited study linked to improvements in social communication and even white matter integrity in the brain; structured physical activity curricula such as SPARK; martial arts and Kata techniques; therapeutic horseback riding; aquatic programs that use water as a medium of communication; sensory-motor integration exercises; and recreational ball games for preschoolers. The mechanistic rationale is plausible on several fronts: exercise offers structured, predictable social contexts; it engages motor circuits that overlap with social-cognitive networks; and animal work, including a well-known study on rapid synapse formation during motor learning published in Nature, hints at how physical activity might prime neural plasticity during development.

Yet the new analysis tempers the enthusiasm that has surrounded some of these individual findings. The GRADE quality assessment summarized in the paper reflects the limited evidence base, and the authors explicitly state that variability in protocols and outcome measures makes the findings difficult to generalize. Different trials measured social skills with different instruments, delivered exercise at different intensities and frequencies, and enrolled children across a wide range of ages and symptom severities. That heterogeneity of methods, even when statistical heterogeneity is low, means the pooled estimate averages over interventions that may work through different mechanisms or not at all in some subgroups of children.

What should families and practitioners take away? The authors’ conclusion is measured: physical exercise may have a small beneficial effect on social skills in children with autism, and given its low cost and broad health benefits, it remains a reasonable addition to a comprehensive support plan. But exercise should not be presented as an established treatment for core social deficits on the strength of the current evidence. The finding that the pooled effect dissolves without one influential study is a reminder that single dramatic trials, however compelling, need replication before they reshape clinical practice. The context is significant: according to surveillance data cited in the paper from the United States Centers for Disease Control and Prevention’s Autism and Developmental Disabilities Monitoring Network, autism is identified in a substantial and rising share of children, so even small, reliable interventions could have large public health implications if their effects were confirmed.

The path forward, the researchers argue, lies in larger and more rigorous randomized controlled trials designed to clarify both the magnitude of any effect and the moderators that shape it—factors such as child age, baseline social functioning, intervention type, dose, and the social richness of the exercise setting. Until those trials arrive, the honest summary of the science is this: exercise is good for autistic children for many reasons, and it might modestly help their social skills, but the strongest statistical case for that social benefit currently rests on shaky ground. In a research landscape often criticized for overselling positive findings, a meta-analysis willing to publish its own fragility is a quiet kind of progress—and perhaps the most trustworthy result of all.

Subject of Research: Effects of physical exercise interventions on social skills in children with autism spectrum disorder

Article Title: Effects of Physical Exercise Interventions on Social Skills in Children With Autism Spectrum Disorder: A Systematic Review and Meta-Analysis

Article References: Tan, Q., Lei, Z., Chang, J., & Yao, L. (2026). Effects of Physical Exercise Interventions on Social Skills in Children With Autism Spectrum Disorder: A Systematic Review and Meta-Analysis. Journal of Autism and Developmental Disorders. https://doi.org/10.1007/s10803-026-07462-6

Image Credits: AI Generated

DOI: 10.1007/s10803-026-07462-6

Keywords: autism spectrum disorder, physical exercise, social skills, meta-analysis, randomized controlled trials, children, intervention, publication bias, sensitivity analysis, motor skills, evidence synthesis, Effects