Why Autism Cannot Be Understood by Averaging People Away

Hatched by MGH

Apr 25, 2026

11 min read

3 views

91%

The mistake we keep making: turning a population into a prototype

What if the reason autism has remained so hard to understand is not that it is too mysterious, but that we have been asking the wrong kind of question?

For decades, medicine has often searched for the average autism case: the typical child, the standard symptom pattern, the single cause, the clean diagnostic boundary. That approach feels natural. It is also misleading. Autism does not behave like a single machine with one broken part. It behaves more like a landscape, with hills, valleys, overlapping paths, and different weather patterns depending on where you stand.

That is why one of the most important shifts in autism thinking is not merely biological. It is methodological. Autism is increasingly understood as a plural condition: a family of developmental trajectories, risk combinations, and functional outcomes that can look similar on the surface while arising from very different mechanisms underneath. The deeper lesson reaches beyond autism itself. When we study a highly heterogeneous human condition, the first challenge is not only to identify causes, but to choose a research design that can respect variation instead of flattening it.

The real puzzle is not whether autism has a cause. It is how a single label can contain so many different developmental stories.

From one disorder to many developmental paths

The history of autism is a history of category collapse and category expansion. What was once treated as a rare, narrow clinical entity is now recognized as relatively common across childhood and adulthood. But prevalence growth should not be read too simply. More diagnoses do not necessarily mean more cases in a pure biological sense. They may mean better recognition, broader criteria, changed clinical attention, improved services, and a willingness to see traits that used to be missed or reclassified.

This matters because autism does not begin at a single point and does not unfold in a single way. Some children show differences very early in life. Others appear to develop typically and then lose skills in a regressive phase. Some have intellectual disability, epilepsy, and language impairment. Others have average or high intelligence, strong vocabulary, and subtle but costly social difficulties. Some function well in one context and poorly in another. The label remains the same, but the lived reality changes dramatically.

This is why the move from autism as a singular disease to autisms as a spectrum of neurodevelopmental patterns is not semantic hair splitting. It is an attempt to match our concepts to reality. A diagnosis is useful only if it tracks something real in the world. If the world contains multiple overlapping pathways into similar social and behavioral outcomes, then a single name can help communication, but it can also obscure mechanism.

Think of the difference between calling every fever by the same name and learning whether the cause is flu, infection, inflammation, or heatstroke. The symptom may look familiar. The underlying action required is not.

That is where the tension begins. Autism is real as a clinical category, but it is not necessarily a single biological entity. The category is broad because the developmental routes into it are broad. The question is not whether that breadth is inconvenient. The question is whether we can design thinking and research that are broad enough to capture it without becoming vague.

Why the search for a single cause keeps failing

The temptation in medicine is always to hunt for the master key. One gene. One neurotransmitter imbalance. One prenatal exposure. One brain circuit. Autism has attracted every version of this impulse, and each has revealed something true, while none has been sufficient on its own.

The reason is simple: heterogeneity is not noise, it is the signal.

Genetic evidence strongly suggests substantial heritability. Twin studies, family recurrence patterns, and the accumulation of many associated genes all point to a major inherited contribution. Yet the genetic architecture is not the architecture of a single switch. Hundreds of genes may be associated, but each contributes only a small piece. Some rare syndromic conditions raise risk dramatically. Many more common variants appear to add tiny increments. No single autism gene explains autism on its own.

Environmental factors complicate the picture further. Advanced parental age, prenatal infection, metabolic conditions in pregnancy, some medication exposures, birth complications, and postnatal medical stressors have all been discussed as risk factors. But these do not operate like fate written in advance. They likely matter through interaction, not isolation. The same exposure may have different effects depending on genetic susceptibility, timing, and developmental context.

This creates a crucial insight: autism is not best understood as one cause producing one effect. It is better understood as a convergence zone. Different biological and developmental pressures can converge on overlapping patterns of social communication difficulty, repetitive behavior, and sensory or regulatory differences. In that sense, the label identifies a common endpoint more than a common origin.

That is also why the frequent presence of additional features, such as ADHD, anxiety, epilepsy, sleep disturbance, intellectual disability, and language impairment, is not a side note. It is a clue. These co-occurrences suggest that the person we call autistic may be carrying a broader developmental profile, not a single isolated disorder. The old model, in which one diagnosis should dominate all others, is too tidy for reality.

In a heterogeneous condition, comorbidity is not a distraction from the diagnosis. It is part of the diagnosis.

The hidden lesson of single subject design: why one person can tell the truth better than one average

Here is where methodology changes the story.

When a condition is highly variable, group averages can become deceptive. Averages are useful when people are similar enough that combining them reveals the underlying signal. But if the population contains multiple pathways, the average can hide more than it reveals. A classic group study may tell us what is true across many participants, yet still fail to explain what is true for any particular person.

This is where single subject design becomes intellectually important. Its central logic is underrated: study one case deeply, repeatedly, and systematically, and you may gain a clearer view of causal change than from a large but blurred group comparison. Such designs often offer strong internal validity, because the person serves as their own reference point across time, conditions, or interventions. The tradeoff is external validity, because conclusions about one individual do not automatically generalize to everyone else.

That tradeoff maps eerily well onto autism research and care. Autism is a condition where the group can be too broad to guide the individual, while the individual can be too specific to fit the group. A child with early regression, epilepsy, and severe language impairment may need a different developmental hypothesis than a verbally fluent teenager with social fatigue, anxiety, and masked sensory overload. Averages alone cannot distinguish these stories.

Imagine trying to understand climate by averaging the weather across an entire continent. The mean temperature may tell you something, but it tells you little about the storm system that hit one coast, the drought in another region, or the microclimate of a mountain valley. Autism looks more like climate than like a single weather event. The right scientific response is not to stop measuring patterns, but to measure them at multiple levels: population, subgroup, and person.

This is why the methodological debate is not just academic. If autism is plural, then research must become more plural too. Group studies can identify broad associations. Single subject designs can clarify individual response, causal timing, and intervention effects. Together, they create a more honest science than either alone.

The deeper synthesis is this: heterogeneous conditions require layered evidence. Population studies tell us where to look. Deep individual designs tell us what is actually happening in a life.

A better framework: autism as a developmental ecology

To make sense of autism without flattening it, we need a more useful mental model. The best one is not a single disease tree. It is a developmental ecology.

In an ecology, outcomes emerge from interactions among many elements: terrain, weather, species, timing, disturbance, and adaptation. None of these alone determines the system. Similarly, autism emerges from an interplay of inherited susceptibility, prenatal and early life influences, neurodevelopmental timing, and the demands of the surrounding environment.

This framework clarifies several things at once:

Same label, different pathways Two people can both meet diagnostic criteria while arriving there by different routes. One may have a syndromic genetic condition. Another may have many small genetic risks plus early environmental stress. The label captures overlap, not identity.
Symptoms are context dependent Function is not fixed in stone. It changes with age, language expectations, school demands, sensory load, family support, and mental health. A person who copes well in one context may struggle in another.
Comorbidity is structurally expected If development is globally altered, then overlapping difficulties in attention, sleep, anxiety, movement, or cognition are not surprising. They are often part of the same developmental ecosystem.
Intervention should be person specific The right question is not only “Does this treatment help autistic people?” but “For whom, under what conditions, and at what developmental stage?”

This ecological lens also changes how we think about prevalence. Rising diagnosis rates may reflect better detection of a broader phenotype, especially in girls, adults, and people with subtler presentations. The rise may also reflect a shift in what we are willing to call autism. That is not merely statistical inflation. It is a change in the boundaries of visibility.

But visibility has a cost. The broader the category becomes, the more important it is to distinguish diagnostic recognition from clinical need. Not every autistic trait requires intervention. Some traits are simply variations in cognition and social style. The decisive question is whether the traits create impairment, distress, or a mismatch with the environment.

That distinction is crucial. A label should help us allocate support, not force sameness onto people who are different.

Key Takeaways

Stop asking only what autism is. Start asking which autism is present. The label covers multiple developmental pathways, not one uniform condition.
Treat comorbidities as information, not clutter. ADHD, anxiety, epilepsy, language delay, and sleep problems often reveal the broader developmental context.
Use both population and person level evidence. Group studies help identify patterns; single subject approaches help explain causal change in individuals.
Focus on function, not just traits. The presence of autistic traits matters most when it affects daily life, support needs, or developmental trajectory.
Think in interactions, not isolated causes. Genetics, timing, environment, and context likely combine rather than act alone.

What this means for science, clinicians, and families

A more plural view of autism is not a retreat from explanation. It is an upgrade in explanatory ambition.

For scientists, it means designing studies that do not erase variation. That includes stratifying by cognitive profile, language level, sex, comorbidity, developmental course, and genetic subgroup. It also means valuing designs that track change within individuals over time, not only differences between averages.

For clinicians, it means resisting the urge to stop at the diagnosis. The label is a starting point, not an endpoint. The real work begins when one asks how this person learns, sleeps, communicates, regulates emotion, and responds to stress. A good assessment is less like stamping a passport and more like drawing a map.

For families, it means recognizing that an autism diagnosis does not predict a single future. It identifies a broad developmental vulnerability, but not a fixed destiny. What matters is the intersection of supports, demands, and strengths. Two children with the same diagnosis may need completely different forms of help, and both may be right.

This also explains why people can feel alienated by simplistic narratives around autism. Some are looking for a cure to a disabling condition. Others are defending a neurodevelopmental identity that is inseparable from their personality and way of knowing the world. Both perspectives can be sincere because autism is not one thing. It can be disabling, identity shaping, context dependent, and biologically diverse all at once.

That complexity is not a flaw in the concept. It is the point.

Conclusion: the diagnosis is real, but the average person is not

The deepest mistake in autism science was never the search for causes. It was the belief that a broad developmental phenomenon must hide a single, clean essence waiting to be extracted. Real human conditions are often messier than our categories. Autism is no exception.

The better reframing is this: autism is not one disease that appears in many people. It is a population of developmental paths that sometimes converge on a shared clinical language. That is why progress will come not from averaging people harder, but from learning how to see patterns at the right scale.

Once you see that, the methodological and clinical questions align. Group studies are useful, but incomplete. Single subject designs are precise, but local. Diagnostics are necessary, but not sufficient. And the search for meaning in autism shifts from hunting for one hidden cause to understanding how different forces produce similar human outcomes.

In the end, autism teaches a broader lesson about knowledge itself: when reality is heterogeneous, the most intelligent science is not the one that simplifies fastest. It is the one that can hold complexity without losing clarity.