Not Worth It: Procedural Fairness and Autistic Withdrawal from Competition

The question

In the laboratory, competition suits autistic people rather well. Put an autistic child in a competitive game and their performance improves. Embed a false-belief task inside a contest with a prize and the proportion passing rises from thirteen per cent to seventy-four. Sit autistic adults down to a purely strategic numbers game against other players and, on the best-known test of it, they hold their own.

Outside the laboratory, autistic adults are thinly represented in the arenas where adult life is openly competitive - the ranked, the rivalrous, the winner-takes-most. Where anyone has bothered to count, the gap is large.

Those two observations do not sit together, and the obvious reconciliation - that we lack competitive drive - is the one the laboratory rules out.

So the question is what is actually being declined. There is reason to think the answer differs for adults, on whom almost none of this research was done. A seven-year-old refusing a rigged game is responding to the game. A forty-seven-year-old refusing one is responding to the game plus three decades of evidence about how contests involving them tend to go.

Where the question came from

I started with a claim I was confident about and which turns out to be wrong.

The claim was that autistic people are less competitive. It is a common belief, held inside the autistic community as often as outside it, and I held it myself. The supporting observations are easy to assemble. We are less present in sport. We are less present in the parts of working life that run on ranking and rivalry. A great deal of male social life is organised around competition, either playing it or discussing it, and a great many autistic men are visibly outside that organisation.

I had a mechanism, too, and I was pleased with it. It went: we care unusually about fairness, so we tolerate competition when the contest is clean and withdraw when it is not. Cheating drives us out. So does a rigged field. And - this was the part I was most confident about - we would rather not play at all than play something we know we are going to lose.

Writing that last sentence down next to the others is what broke it.

Those are not one hypothesis. They are two, and they contradict each other. A contest you lose because your opponent is better is not unfair. It is the fairest thing a contest can be: the procedure worked and returned the correct answer. “I withdraw because the game is rigged” and “I withdraw because I will lose” are opposite claims wearing the same coat, and I had been carrying them around together for months without noticing.

That is worth saying plainly, because the second claim is much the more frugal. Autistic adults accumulate long histories of public failure, of effort that does not convert into result, and of being visibly bad at things in front of people who found them easy. Anticipatory withdrawal from likely failure requires no theory of justice whatsoever. If the fairness account is to be worth anyone’s time it has to beat that, on evidence rather than on how much I like it.

So the question narrows to something I can prosecute: when an autistic adult declines a contest, are they responding to the odds or to the procedure?

What the evidence actually says

The participation data is nearly all about sport, which is the problem

Sport is where participation gets counted, because sport is countable. It is also the worst available proxy for the question, because it bundles competition with physical exertion, sensory environment and motor coordination - and autistic people have independent reasons to withdraw from all three. Read what follows as the visible edge of the phenomenon rather than its definition. The domains I actually care about, where competition is cognitive and the body is irrelevant, are the ones nobody has measured.

A systematic review of physical activity and sedentary behaviour in autistic adults found the literature small - twelve studies met inclusion from an initial 2,834 - and consistent in direction.¹ Autistic adults are unlikely to meet national activity guidelines.

The most useful recent adult work is qualitative. McLeod and colleagues interviewed autistic adults in middle adulthood about physical activity, and did so using an explicitly trauma-informed frame - the first study in this area to do so.² That framing is not decoration. It reflects what the participants described: inaccessible environments, sensory overload, and unhelpful support from people outside their immediate circle, sitting on top of accumulated earlier experience.

A separate qualitative study of autistic adults produced three themes, and the third is the one I cannot stop looking at: support from others matters, it needs to be measurable, and it’s just not worth it.³

“Not worth it” is a cost-benefit verdict. It is the language of someone who has run the calculation before.

Everything about competition itself comes from children

This is the honest position and I would rather state it than manage it.

Team versus individual sport. O’Flaherty and colleagues followed a representative Australian cohort - 313 autistic and 6,346 non-autistic children - from ages eight to fifteen, using marginal structural models to handle time-varying mediators.⁴ Non-autistic children benefited from all forms of sport. For autistic children, team sport was associated with worse mental health and no physical benefit; individual sport improved physical health-related quality of life but not BMI or mental health. The authors attribute the split to the social and sensory environment of team sport.

Competitive drive. Yu and colleagues used gaze-contingent eye-tracking with 26 autistic preschoolers and 20 typically developing children.⁵ The autistic children oriented less toward competition - they looked at it less - and the presence of a parallel competitor improved their performance by the same margin as everyone else’s. They did not watch the contest and raced anyway.

Competition and social cognition. Peterson and colleagues built a false-belief task in which tracking another player’s beliefs is the route to a prize, and ran it against the standard Sally-Anne procedure in the same children.⁶ Seventy-four per cent passed the competitive version; thirteen per cent passed the standard one. Their reading is that conventional tests are motivationally barren and competition supplies the missing reason to model another mind.

Those children become autistic adults, and I see no reason to treat the findings as inapplicable on age grounds alone. But note the ages - preschool, primary, early adolescence - because they make the child data useful in a specific way. It is a baseline taken before the accumulation. If competitive drive is intact at five and participation has collapsed by forty-five, the gap between those two facts is not a nuisance in the evidence. It is the thing worth explaining.

The one competition-adjacent literature that is adult is contested. Pantelis and Kennedy tested thirty autistic adults on the “beauty contest” game and found no deficit in strategic depth.⁷ Król and Król ran a process-tracing version with a payoff calculator and concluded that autism limits strategic thinking after all.⁸ Two studies, opposite titles, unresolved.

Fairness: the weakest leg, and the reason for the study

Here I have to be careful, because this is the part I most want to be true.

The belief that autistic people have a heightened sense of justice is close to universal within the autistic community. It appears in community writing, in clinical description, in almost every popular account of autistic experience, and in a great deal of what autistic adults say about themselves. I believe it about myself.

It is also, as far as I can establish, largely uncited.

Justice sensitivity is a real and well-built construct. Schmitt and colleagues developed and validated an inventory measuring four facets - victim, observer, beneficiary and perpetrator sensitivity - on a representative sample of 2,510 people, with normative data.⁹ It is a serviceable adult instrument that has been in use for fifteen years. I cannot find a study that has administered it to an autistic sample. That is a strange gap, because the hypothesis is in plain view and the instrument is sitting next to it.

What circulates instead is a claim, repeated across a great many websites, that autistic children reject unequal rewards even at cost to themselves. It is consistently attributed to a 2021 paper by Shulman, Guberman and Ziv. I have not been able to find that paper.¹⁰ No journal, no volume, no DOI, no index entry - only a chain of sites citing one another. I raise it not to score a point but because a widely held community belief resting on a citation nobody can produce is a reason to do the study rather than abandon the idea.

One real result cuts against me, and it is - again - about children. In a study of sharing and reciprocity, children with ASD offered similarly to typically developing children, preferred equality over self-interest at similar rates, and referred verbally to “fairness” at similar rates, but were 37% less likely to reciprocate fair offers and 22% less likely to reciprocate unfair offers.¹¹ Fairness talk intact, fairness sharing intact, fairness-contingent responding reduced. That is close to the opposite of heightened reactivity to unfairness.

The one strand pointing my way is neural. Bolling and colleagues put children through comparable experiences of social exclusion and rule violation in the scanner.¹² During exclusion, right insula and ventral anterior cingulate were hypoactive in the autistic children, despite both groups reporting equal distress. During rule violation, right insula and dorsal prefrontal cortex were hyperactive.

Under-responsive to being left out. Over-responsive to the rules being broken. That is a dissociation in the right shape for my hypothesis, and it is the closest thing to direct support that exists.

Two cautions, neither fatal. A rule violation is a breach of a stated procedure, not an unfair outcome, and unfair outcomes frequently involve no rule being broken at all. And if this pattern is present in childhood, the interesting question is whether it strengthens with experience or is simply there from the start - which is a question the same paradigm could answer in adults, and nobody has run it.

The framing

The right frame is not aversion to anything. Aversion language relocates the problem inside the autistic person, which is both what I am arguing against and what makes the claim hard to test.

The frame is a precondition:

Autistic entry into competition may be conditional on the legitimacy of the contest, where non-autistic entry is conditional on the attractiveness of the prize.

This is not a construct I need to invent. Organisational justice research has distinguished distributive justice - was the outcome fair - from procedural justice - was the process fair - for decades, and Leventhal’s procedural rules supply a dimensional structure: consistency, bias suppression, accuracy, correctability, representativeness, ethicality. I am not proposing a new psychological quantity. I am asking whether a well-measured existing one has an unexamined population effect in adults.

It also makes a prediction that separates it from the obvious rival. “Autistic adults avoid sales roles because sales is socially demanding” and “autistic adults avoid sales roles because territory allocation, quota setting and credit attribution are procedurally opaque” both predict avoidance. Only the second predicts that a transparently commissioned, fixed-rules sales role would be tolerated.

The part where I argue against myself

I would rather state the weaknesses in my own idea than have a supervisor state them for me. There are four, and the second is new since I started writing.

Over-determination. Autistic withdrawal from sport is explained several times over before fairness is mentioned once. Sensory environment. Motor coordination, which co-occurs at high rates and independently explains withdrawal from anything requiring a body to be caught, thrown or dodged. Social complexity, which is what O’Flaherty’s team themselves reach for. Bullying in PE, close to universal in autistic recall. The executive load of getting somewhere fixed at a fixed time each week. Fairness is the last explanation a reasonable person would arrive at.

Worse, the obvious study - survey autistic adults, measure justice sensitivity, correlate against participation - is guaranteed to return a significant result and guaranteed to mean nothing, because justice-sensitive people are also people who were treated badly in PE, and in a self-report sample everything correlates with everything.

Learned withdrawal, which may be the real story. This is the objection that adult focus creates, and it is the strongest one against me.

Suppose procedural sensitivity is not constitutional at all. Suppose it is learned - the entirely rational output of decades of contests that were rigged. Autistic adults are un- or underemployed at rates far above the general population, are passed over for promotion and training, and describe workplace discrimination and bullying consistently. Someone who has repeatedly entered contests where the criteria were unstated, the assessment was social, and the outcome went against them regardless of performance has learned something true about contests. Declining the next one is not heightened justice sensitivity. It is induction.

McLeod’s trauma-informed framing points directly at this, and “it’s just not worth it” is what learned withdrawal sounds like when you ask someone to describe it.

This matters because the two accounts make different predictions and imply different interventions. Constitutional sensitivity says design fairer contests and autistic people will enter them. Learned withdrawal says design fairer contests and autistic people will still not enter them, because the prior takes years to update, and the intervention target is trust rather than procedure. Any study that cannot distinguish these will produce a finding that reads well and recommends the wrong thing.

The construct may be borrowed. If justice sensitivity mediates the whole effect, this is a study about justice-sensitive people who happen disproportionately to be autistic - a different and less interesting finding. The interoception literature went through exactly this with alexithymia, and “it isn’t autism, it’s the co-occurring trait” is now standard equipment in this field.

Self-report cannot reach it. Nobody declines a five-a-side invitation while thinking “insufficient procedural transparency.” If the effect exists it operates below the level at which people narrate their own choices.

And it may not be there at all. Autistic withdrawal may be fully accounted for by sensory load, motor difficulty, social complexity and anticipated failure, with fairness contributing nothing once those are held. That would be useful to know, and a study designed only to confirm my hypothesis could not produce it.

The design that solves the hard problem

The over-determination objection is fatal for as long as the study is about sport. So the study should not be about sport. It should not involve a body at all. If the question is about competition rather than exertion, the task should be one that a person could win lying still.

Move it to an abstract economic game with no sensory load, no motor demand and no real-time social interaction. A screen. A described opponent. A stated procedure. A decision to enter or decline.

If the withdrawal effect appears in an environment containing no noise, no crowd, no ball, no changing room and no face, then sensory aversion, motor coordination and social complexity are excluded by construction rather than by statistical adjustment. I am not controlling for the confounds; I am building a room they cannot enter.

The participation gap in adults is already established. Sport is the reason for the question, not the site of the answer.

What I would actually do

The two clauses I bundled at the start come apart if you cross procedural fairness with expected outcome, and let people choose whether to play at all.

Table 1
Predicted Entry Decisions Under Crossed Procedural Fairness and Win Probability

	Likely to win	Likely to lose
Fair procedure	Everyone enters. Floor condition.	Fairness account: enter. Failure-avoidance account: decline.
Rigged procedure	Fairness account: decline. Failure-avoidance account: enter.	Nobody enters. Ceiling condition.

Note: In two conditions the competing accounts predict the same behaviour and so cannot be told apart: a fair contest you expect to win attracts everyone, and a rigged contest you expect to lose attracts nobody. These establish the floor and ceiling of the entry rate. The test sits in the two conditions where the accounts disagree - a fair contest you expect to lose, and a rigged contest you expect to win.

Fair-but-losing asks whether a clean contest you will probably lose is still worth entering. Rigged-but-winning asks whether a corrupt contest you will probably win is worth refusing - and that is the cell I care about most, because walking away from a game you are winning on the grounds that the process is dishonest is a strange, specific prediction that very little else in the literature generates. If the effect is real it will be hard to explain away. If it is absent I have learned something early and cheaply.

In outline:

• Autistic and non-autistic adults, recruited through services and community organisations rather than social media, for the representativeness reasons I have written about before.
• 2 × 2 × 2 mixed factorial. Procedural fairness (fair / rigged) and win probability (favourable / unfavourable) within participants; group between participants. Trials blocked, order counterbalanced.
• Primary outcome: the entry decision - a binary opt-in per trial - not performance. The question is about participation, so measure participation.
• Manipulate procedure explicitly and visibly. In the rigged condition the participant is told the rule and shown that it favours the opponent: an extra card, ties awarded against them, a different scoring threshold. The rigging must be stated rather than discovered, so the manipulation is procedural knowledge rather than suspicion or detection skill.
• Manipulate win probability numerically and honestly. State the probability. Do not let it be inferred from opponent skill, which would reintroduce social judgement.
• Measure accumulated experience of unfair treatment - employment history, discrimination experience, whether the person has been through a grievance or tribunal, age at diagnosis. This is the variable that separates constitutional sensitivity from learned withdrawal, and without it the study cannot tell those two apart. If the fairness effect scales with accumulated adverse experience, it is learned. If it is flat across experience levels, it is more likely dispositional.
• Administer the Justice Sensitivity Inventory to both groups and test it as a mediator. If the group effect attenuates substantially once justice sensitivity is in the model, that is the finding, and a more interesting one than a bare group difference.
• Administer an alexithymia measure, because pre-empting the co-occurring-trait objection costs one questionnaire.
• Analysis: a 2 × 2 × 2 mixed ANOVA on entry rate, with the three-way interaction as the test of record. The hypothesis lives entirely in the interaction: a fairness × group effect that survives win probability. Main effects of fairness and of win probability are expected in both groups and tell me nothing. Given a binary trial-level outcome I would report the ANOVA for interpretability alongside a mixed-effects logistic regression with random intercepts for participant, which handles the dependent variable properly and preserves trial-level variance instead of averaging it away. Experience and justice sensitivity enter as covariates, then as moderators. Power on the interaction, not the main effects - which needs a larger sample than this literature usually recruits, and should be budgeted honestly rather than discovered at analysis.
• A small qualitative arm. Ask a subsample to explain each decline immediately after making it. If people decline rigged games while explaining it as boredom or fatigue, the gap between behaviour and account is itself the finding.
• Design it with autistic adults rather than about them.

My honest prediction: both groups track win probability, because everyone does. The autistic group shows a steeper decline in entry under rigged procedures. The effect is smaller than I want. A meaningful portion is absorbed by justice sensitivity as a mediator, and another meaningful portion scales with accumulated adverse experience - meaning the answer is both, dispositional and learned, in a proportion nobody currently knows. I also expect the fair-but-losing cell to show autistic entry below non-autistic, which would mean the clause I discarded was partly right after all.

I could be wrong on all of it. That is rather the point of doing the study.

Why this one is on the list

Because of what it would change about accommodation.

Workplace accommodation for autistic employees is almost entirely sensory and communicative. Lighting. Noise. Advance notice of meetings. Written instructions rather than verbal. A quiet room. All useful, all treating the environment as a physical space.

Procedural justice is on nobody’s accommodation list anywhere. If competitive participation is procedurally gated, then publishing the promotion criteria is an autism accommodation. So is a documented basis for allocating accounts and resources; a stated rule for who gets the stretch assignment, and a commission structure that does not change retroactively.

I nearly followed that with the observation that these cost nothing. It is wrong, and it is worth being clear about why.

They cost discretion. Vague criteria are not an administrative failure that better evidence would correct. They are frequently the point: ambiguity is what allows a manager to reward who they want to reward, retain who they need to retain, and account for the outcome afterwards. Written criteria remove that latitude, and they do something else besides - they create a record, and a record is discoverable. An organisation with documented promotion criteria can be shown to have departed from them. An organisation without them cannot.

If anyone doubts that transparency carries a price someone is unwilling to pay, consider the EU Pay Transparency Directive. Member states had three years’ notice of a transposition deadline of 7 June 2026. That date has passed with full implementing law in force in only a small minority of states, several major economies targeting 2027, and at least one seeking renegotiation of the instrument itself.¹³ That is the resistance encountered by a measure with an unambiguous equity rationale and the force of EU law behind it. A finding about autistic adults will not do better on persuasion alone.

So the honest claim is narrower, and it has three parts.

The first is legal rather than voluntary. Duties to make reasonable adjustments already exist across comparable jurisdictions. What does not exist is any evidence base establishing procedural opacity as a barrier with a disability-specific effect. That is what a finding supplies: not new legislation, but the evidentiary basis on which existing law might be argued to reach something it currently does not. Slower and more adversarial than I would like, and realistic.

The second is that this need not travel alone. Transparency mandates are arriving anyway, driven by gender pay equity rather than disability. A finding that procedural opacity also carries a neurodivergent cost attaches itself to a vehicle already moving - and to an argument that improves working life for a great many people who are not autistic at all. That is a far better position than asking employers to act unprompted.

The third requires nobody’s permission, and may matter most in practice. An autistic adult who knows this about themselves can ask at interview how territory is allocated and how promotion is decided, and can read a vague answer as information rather than as reassurance. They can select structures rather than simply endure them. That is not a policy outcome. It is a person being handed an accurate account of their own behaviour, which is the thing I keep insisting psychology is actually for.

And if the learned-withdrawal account wins instead, that is more actionable rather than less, because it says the problem is not autistic disposition but an accurate read of how these workplaces have actually behaved. That finding points at the employer, which is where a good deal of this should have been pointing all along.

There is a version of this that goes badly, and I would rather name it than let someone else build it. “Autistic people are unsuited to competitive roles” is one short step from a screening instrument, and a screening instrument is an exclusion mechanism with a citation attached. The same data supports “roles with opaque allocation lose their autistic staff,” which is an instruction to the employer, and “autistic people don’t belong in sales,” which is an instruction to HR about who not to interview. Identical evidence, opposite consequence, and the difference lies entirely in who is handed the result. A study in this area should state which one it is for.

Then the smaller thing, which is the one I actually care about.

An autistic adult who has spent thirty years being told they are a poor sport, a sore loser, not a team player, oddly intense about rules, or simply no fun, has been handed a character description. If what they were doing was declining contests they had judged dishonest - or had learned, correctly, to expect to be dishonest - then it is not a character description. It is a preference with a mechanism, and it can be explained to them in a sentence.

I keep arriving at the same place from different directions. Most of what a psychologist usefully does is tell somebody what is happening to them. You cannot do that for a pattern nobody has measured, and nobody has measured this one in anybody old enough to have been shaped by it.

---