Autism, Intervention Evidence, and the Limits of Linear Frameworks
Center for Adaptive Stress | April 2026 Lori Hogenkamp, Founder and Director
This document synthesizes current evidence on autism intervention research from an Evolutionary Stress Framework (ESF) / complexity-science perspective. It is a conceptual analysis, not medical advice. Research and drafting assisted by Claude Opus 4.6 (Anthropic); evidence synthesis supported by ChatGPT deep research (OpenAI).
The Problem
There’s a pattern in autism research and intervention that cuts across almost every major approach, and it’s so foundational that most people working inside it don’t recognize it as a choice. It looks like this: we measure what we can see.
That sounds reasonable. Scientific. Empirical. But it has a consequence that shapes everything downstream — because in autism, what you can see is almost always behavior, and behavior is almost always the last thing that happens in a long chain of events that no one is measuring.
A child melts down. That’s visible. A teenager stops making eye contact. That’s visible. An adult can’t hold a job. That’s visible. And because these things are visible, they become the targets — of measurement, of intervention, of entire research programs designed to quantify them, reduce them, or replace them with something that looks more typical.
But here’s the question almost no one is asking at the level of the measurement paradigm itself: what if the visible behavior is the output of a system, and the system is what actually matters?
Three Surfaces, One Structural Limitation
The Behavioral Surface
Applied Behavior Analysis — ABA — is the most explicit version of this logic. ABA, in its classical form, treats behavior as the unit of analysis. It observes what a person does, identifies which behaviors to increase or decrease, and applies reinforcement contingencies to shape the output. The model is linear by design: stimulus in, behavior out, reinforcement modifies the ratio.
This isn’t a caricature. The foundational literature says this plainly. And the reason it has persisted as the dominant intervention model for decades isn’t that clinicians are indifferent to the inner lives of autistic people — many care deeply. It’s that the measurement framework only has access to the surface. When your instruments can see behavior and your theory says behavior is what matters, then behavior is what you treat.
The Social-Communication Surface
A significant proportion of autism research is organized around social-communication measurement. How much eye contact does the person make? How do they perform on standardized measures of conversational reciprocity? How well do they read facial expressions or navigate the pragmatics of social exchange?
These measures are clinically useful. They capture real patterns. But they share a structural feature with ABA: they are measuring downstream observables. Social-communication performance is the output of a regulatory system that is managing sensory load, interoceptive prediction, energy allocation, communicative mismatch, and contextual uncertainty — all at once, all in real time. The measures capture what the system produces under those conditions. They do not capture the conditions themselves.
A person who withdraws from social interaction under conditions of sensory overload and communicative mismatch isn’t demonstrating reduced social motivation. They’re demonstrating a system that has run out of regulatory bandwidth for an interaction that costs more than it gives back. The behavior is the same. The explanation is completely different. And the intervention that follows from each explanation is completely different too.
The Attachment Surface
More recent work in the psychodynamic tradition has tried to go deeper — moving past behavior and social performance toward the relational and emotional substrate underneath. This represents a genuine advance. When clinicians talk about attachment disruption, failures of object constancy, or the importance of connection as a therapeutic mechanism, they’re reaching toward something that the behavioral and social-communication frameworks miss: the relational context in which the person exists.
But this move introduces its own version of the same structural limitation. Attachment frameworks locate the explanatory weight in the relational domain — as if the relationship itself is the primary mechanism, and disruptions in attachment are primary deficits to be repaired.
From a systems perspective, this gets the direction of causation partially reversed. Attachment processes — the ability to register another person as safe, to sustain relational engagement, to internalize a stable model of connection — are not independent capacities. They are emergent properties of a regulatory system that is successfully managing the energetic and predictive demands of social interaction. When the system is operating within its viable range, attachment can emerge naturally. When those conditions aren’t met, attachment processes degrade — not because the capacity for connection is broken, but because the system that supports connection is operating outside its regulatory range.
The “Refrigerator Children” Problem
This brings us to a critical risk that is rarely named. The older psychoanalytic story said the mother was too cold (the “refrigerator mother”). Contemporary relational approaches have rightly rejected that framing. But without explicit complexity-science and neurodiversity-affirming grounding, they can inadvertently flip the direction while preserving the same linear logic: now it’s the child who is constitutionally hard to reach, hard to warm, hard to connect with. The directionality changes, but the deficit is still located in a single node.
Milton’s double empathy problem is the cleanest corrective. He argues that social difficulties in autism are not best understood as a unilateral failure inside the autistic person, but as a disjuncture in reciprocity between differently disposed people. Breakdown in mutual understanding is not owned by one side. It is relational and emergent.
Bolis and Schilbach push this further with the dialectical misattunement hypothesis: misattunement is a property of the unfolding relation, not simply a deficit in either participant. They stress that interaction can be improved not only by changing persons, but by changing communication media and environments. The question becomes less “how do we penetrate the autistic child?” and more “how do we reorganize the interactional ecology so that mutual attunement becomes more likely?”
What the Evidence Actually Shows
The ABA evidence base is far weaker than commonly claimed
The effects largely disappear when you control for bias. The 2023 Project AIM update (Sandbank et al., BMJ) — the largest meta-analysis of early autism interventions ever conducted, covering 252 studies and 13,304 children — found that when high risk-of-bias studies and caregiver/teacher-report outcomes were excluded, almost all significant effects for behavioral interventions vanished. After full bias-sensitive analysis, the only remaining significant effect was a modest result for naturalistic developmental behavioral interventions on autism diagnostic characteristics (g ≈ 0.30). Adverse events were poorly monitored but possibly common.
More hours don’t help. Sandbank et al. (2024, JAMA Pediatrics) tested the dose-response assumption across 144 studies and 9,038 children. None of the three core amount metrics — daily intensity, duration, or cumulative intensity — was significantly associated with outcomes. The foundational recommendation for 30–40 hours per week rests on a surprisingly thin base. A 2026 real-world study from the Catalight Research Institute (725 children) confirmed: higher ABA hours were not associated with improvements in communication, socialization, or daily living skills. Dangerous behaviors decreased regardless of hours.
The people doing the research are the people selling the service. Bottema-Beutel & Crowley (2021, Frontiers in Psychology) found that 84% of ABA studies had at least one author with a clinical/consultancy conflict of interest, yet these were disclosed only 2% of the time. 87% of “no conflict” statements were inaccurate. In group-design studies more broadly, Bottema-Beutel et al. (2021, JCPP) found COIs in 70% of 150 reports, with only 5.7% accurately disclosed.
Selective reporting compounds the problem. Sandbank & Bottema-Beutel (2024, Autism) documented widespread non-reporting of registered outcomes, reporting of unregistered outcomes, and data-sharing failures — creating a more favorable evidence picture than the underlying data warrant.
Cochrane remains cautious. The Cochrane review of EIBI found only weak evidence that intensive early behavioral intervention may help some children, based on very few studies, with no clear evidence for reducing autism severity or problem behavior.
Real-world monitoring confirms modest results. The TRICARE Autism Care Demonstration reports have repeatedly shown limited measurable change for many children, with higher service hours not clearly producing greater improvement.
Why ABA appears to have stronger evidence
This is a critical structural point: ABA appears stronger in part because behavioral outcomes are inherently easier to operationalize within traditional RCT frameworks. Linear interventions targeting discrete observable behaviors produce discrete measurable outputs. Project AIM found that distal outcomes were significantly smaller than proximal ones, and generalized outcomes were significantly smaller than context-bound ones. Interventions look better when they target narrow skills close to the treatment setting and worse when you ask whether change travels into everyday life. This creates systematic measurement bias favoring behavioral approaches — not because they’re more effective, but because they’re more measurable within the dominant research paradigm.
Contemporary developmental-relational work is not Bettelheim
The 2012 French HAS recommendation is often cited as proof that “psychoanalysis was rejected and ABA vindicated.” This misrepresents the document. HAS said psychoanalytic approaches could not be judged pertinent because efficacy data were lacking and expert opinion diverged — a statement about evidence gaps, not a philosophical rejection. Critically, DIR/Floortime (Greenspan) was explicitly allowed within a coordinated care plan; it was only excluded as a stand-alone exclusive method. The guidance recommended coordinated educational, behavioral, AND developmental approaches.
Contemporary developmental-relational approaches — formalized in 2024 by Cullinane et al. as Developmental Relationship-Based Interventions (DRBIs) — include DIR/Floortime, iBASIS-VIPP, Responsive Teaching, and related models. They share common features: a developmental framework grounded in the child’s motivation for social engagement, a focus on supporting relationships, and strategies including social play, following the child’s lead, and calibrated challenge. This is not Bettelheim.
The evidence is real, if mixed: multiple RCTs since 2011 show significant improvements in social-emotional functioning, communication, and parent-child interaction (Solomon et al. 2014; Casenhiser et al. 2013; Pajareya & Nopmaneejumruslers 2011, 2012). Whitehouse et al. (2021) found reduced autism diagnostic behaviors when iBASIS-VIPP was delivered preemptively in infancy.
The Third Position
The real alternative is not “ABA versus psychoanalysis.” It is a different level of analysis.
What connects ABA, social-communication research, and attachment-based models isn’t a shared ideology. What connects them is a shared level of analysis — the observable surface. ABA measures behavior. Social-communication research measures social performance. Attachment models measure relational patterns. All three are measuring real things. But all three are measuring outputs of a system rather than the dynamics of the system itself.
From a complexity-science standpoint, autism is better approached as architectural heterogeneity in a complex biological and social system — not as a deficit located on a single linear scale. Parametric variation means more or less of the same thing. Architectural variation means differently organized systems — different constraint structures, different signal priorities, different routes to stability, and different costs of adaptation. The emerging subtype literature (Litman et al. 2025, Nature Genetics) supports this: four robust classes linked to distinct genetic programs and developmental timing patterns.
The Evolutionary Stress Framework proposes a shift beneath behavior, social communication, and relational experience to ask: what is the regulatory system doing that produces these observable patterns? Many “behavior problems” can be redescribed as the visible edge of a high-cost regulatory economy. The question becomes less “why can’t this child regulate?” and more “what is the metabolic, sensory, interoceptive, and social cost of this environment for this nervous-system architecture?”
This doesn’t invalidate the observations that behavioral, social-communication, and attachment researchers have made. It recontextualizes them: you were measuring something real, but you were measuring it at the wrong level of the system. The data is good. The interpretation needs a larger frame.
And this matters for intervention — profoundly. If social disengagement reflects a system operating outside its viable regulatory range, then efforts to increase social motivation, remediate attachment deficits, or shape behavior through reinforcement may be insufficient or counterproductive if they don’t simultaneously address the conditions that constrain regulatory bandwidth. You can’t reinforce your way to a regulated system. You can’t love your way past an energy deficit. You can address the conditions that make regulation possible, and then watch the behaviors that everyone was trying to fix emerge on their own — not because they were trained, but because the system that produces them finally has the resources to do so.
The clinical implication is not therapeutic nihilism. It is a shift in design goals: from normalization to fit; from one-way correction to bidirectional adaptation; from trait suppression to regulatory support; from isolated child outcomes to ecological outcomes; from “more hours” to better-matched supports.
The behaviors are real. The social-communication patterns are real. The relational experiences are real. But they are the weather, not the climate. And until the field starts measuring the climate — the regulatory dynamics, the energetic constraints, the architectural features that make each person’s system respond to the world in the way it does — the interventions will keep targeting symptoms while the system that produces them remains unaddressed.
What we measure is what we see. What we miss is what matters.
References
ABA Evidence Base: Methodological Critiques
Sandbank, M., et al. (2020). Project AIM: Autism intervention meta-analysis for studies of young children. Psychological Bulletin, 146(1), 1–29. doi:10.1037/bul0000215
Sandbank, M., et al. (2023). Autism intervention meta-analysis of early childhood studies (Project AIM): Updated systematic review and secondary analysis. BMJ, 383, e076733. doi:10.1136/bmj-2023-076733
Sandbank, M., et al. (2024). Determining associations between intervention amount and outcomes for young autistic children: A meta-analysis. JAMA Pediatrics, 178(8), 763–773. doi:10.1001/jamapediatrics.2024.1832
Sandbank, M., et al. (2024). Evidence-b(i)ased practice: Selective and inadequate reporting in early childhood autism intervention research. Autism, 28(8), 2042–2058. doi:10.1177/13623613241231624
Samelson, D., Sneed, L., & Pfingston, B. (2026). Dosage in applied behavior analysis: Effect on adaptive behavior, goal attainment, and dangerous behavior. Journal of Autism and Developmental Disorders. doi:10.1007/s10803-025-07203-1
Reichow, B., et al. (2018). Early intensive behavioral intervention (EIBI) for young children with autism spectrum disorders (ASD). Cochrane Database of Systematic Reviews, Issue 5, CD009260. doi:10.1002/14651858.CD009260.pub3
Defense Health Agency. Annual Report on the Autism Care Demonstration Program (multiple years). U.S. Congress reporting requirements.
Conflicts of Interest
Bottema-Beutel, K., & Crowley, S. (2021). Pervasive undisclosed conflicts of interest in applied behavior analysis autism literature. Frontiers in Psychology, 12, 676303. doi:10.3389/fpsyg.2021.676303
Bottema-Beutel, K., et al. (2021). Research review: Conflicts of interest (COIs) in autism early intervention research. JCPP, 62(1), 5–15. doi:10.1111/jcpp.13249
Dawson, M., & Fletcher-Watson, S. (2021). Commentary: What conflicts of interest tell us about autism intervention research. JCPP, 62(1), 16–18. doi:10.1111/jcpp.13315
Bottema-Beutel, K. (2023). We must improve the low standards underlying “evidence-based practice.” Autism, 27(2), 269–274. doi:10.1177/13623613221146440
Double Empathy, Misattunement, and Autistic Cognition
Milton, D. E. M. (2012). On the ontological status of autism: The ‘double empathy problem.’ Disability & Society, 27(6), 883–887. doi:10.1080/09687599.2012.710008
Bolis, D., et al. (2017). Beyond autism: Introducing the dialectical misattunement hypothesis and a Bayesian account of intersubjectivity. Psychopathology, 50(6), 355–372. doi:10.1159/000484353
Bolis, D., & Schilbach, L. (2018). ‘I interact therefore I am’: The self as a historical product of dialectical attunement. Topoi, 39, 521–534. doi:10.1007/s11245-018-9574-0
Crompton, C. J., et al. (2020). Autistic peer-to-peer information transfer is highly effective. Autism, 24(7), 1704–1712. doi:10.1177/1362361320919286
Mottron, L. (2017). Should we change targets and methods of early intervention in autism, in favor of a strengths-based education? European Child & Adolescent Psychiatry, 26(7), 815–825. doi:10.1007/s00787-017-0955-5
Anderson, L. K. (2023). Autistic experiences of applied behavior analysis. Autism, 27(3), 737–750. doi:10.1177/13623613221118216
The French Context
Haute Autorité de Santé (HAS) / ANESM. (2012). Autisme et autres troubles envahissants du développement: interventions éducatives et thérapeutiques coordonnées chez l’enfant et l’adolescent.
Bishop, D. V. M., & Swendsen, J. (2021). Psychoanalysis in the treatment of autism: Why is France a cultural outlier? BJPsych Bulletin, 45(2), 89–93. doi:10.1192/bjb.2020.138
Developmental Relationship-Based Interventions
Cullinane, D. A., et al. (2024). Developmental relationship-based interventions for autistic children. Topics in Early Childhood Special Education. doi:10.1177/02711214241303695
Solomon, R., et al. (2014). PLAY Project Home Consultation intervention program for young children with ASD: A randomized controlled trial. JDBP, 35(8), 475–485. doi:10.1097/DBP.0000000000000096
Casenhiser, D. M., et al. (2013). Learning through interaction in children with autism. Autism, 17(2), 220–241. doi:10.1177/1362361311422052
Pajareya, K., & Nopmaneejumruslers, K. (2011). A pilot RCT of DIR/Floortime parent training intervention for pre-school children with ASD. Autism, 15(5), 563–577. doi:10.1177/1362361310386502
Whitehouse, A. J. O., et al. (2021). Effect of preemptive intervention on developmental outcomes among infants showing early signs of autism. JAMA Pediatrics, 175(11), e213298. doi:10.1001/jamapediatrics.2021.3298
Pickles, A., et al. (2016). Parent-mediated social communication therapy for young children with autism (PACT): long-term follow-up. Lancet, 388(10059), 2501–2509. doi:10.1016/S0140-6736(16)31229-6
Complexity Science, Allostasis, and Developmental Dynamics
Thelen, E., & Smith, L. B. (1994). A dynamic systems approach to the development of cognition and action. MIT Press.
Cicchetti, D., & Rogosch, F. A. (1996). Equifinality and multifinality in developmental psychopathology. Development and Psychopathology, 8(4), 597–600. doi:10.1017/S0954579400007318
Sameroff, A. J. (Ed.). (2009). The transactional model of development. American Psychological Association.
Barrett, L. F. (2017). The theory of constructed emotion: An active inference account. SCAN, 12(1), 1–23. doi:10.1093/scan/nsw154
Friston, K. (2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience, 11(2), 127–138. doi:10.1038/nrn2787
Kitano, H. (2004). Biological robustness. Nature Reviews Genetics, 5(11), 826–837. doi:10.1038/nrg1471
Del Giudice, M., Ellis, B. J., & Shirtcliff, E. A. (2011). The Adaptive Calibration Model of stress responsivity. Neuroscience & Biobehavioral Reviews, 35(7), 1562–1592. doi:10.1016/j.neubiorev.2010.11.007
Litman, A., et al. (2025). Decomposition of phenotypic heterogeneity in autism reveals underlying genetic programs. Nature Genetics, 57(7), 1611–1619. doi:10.1038/s41588-025-02224-z
ESF Published Work
Hogenkamp, L., Sanghavi, D., & Natri, H. (2026). Toward an emergent paradigm for neurodiversity and health. Autism in Adulthood. doi:10.1177/25739581261433443
The Center for Adaptive Stress (CAS) is a nonprofit conceptual systems research organization in Cincinnati, Ohio, developing theoretical scaffolding for integrating complexity science into medicine and neurodevelopmental research. This document represents a theoretical and evidence-synthesis perspective from the Evolutionary Stress Framework. It is not medical advice.
ndstress.org • evostress.blog
Leave a Reply