That is why simple slogans fail.

Figure 14. Primary care funding architecture post 14 visual preview.

“More capitation” does not solve everything.

“More fee-for-service” does not solve everything.

“More telehealth” does not solve everything.

“More urgent care” does not solve everything.

“More hospital funding” definitely does not solve everything.

The system is a hybrid game.

Here is the plain-English map.

Game 1: the hospital salience game. Hospital pressure is visible and politically urgent. Upstream unmet need is quieter until it becomes hospital demand.

Game 2: the Health New Zealand allocation game. Separate appropriations exist, but operational pressure, baselines, targets and political attention still shape what becomes fundable.

Game 3: the capitation marginal-supply game. Capitation supports responsibility but weakly funds the next contact.

Game 4: the patient access pathway game. Patients choose between waiting, paying, telehealth, urgent care, ambulance, emergency department or giving up.

Game 5: the Primary Health Organisation intermediation game. Primary Health Organisations may add population-health value, but payment intermediation may add opacity and friction.

Game 6: the Accident Compensation Corporation / Health New Zealand cross-funder game. Injury and non-injury funding streams interact. Constraining one can shift pressure to another.

Game 7: the ambulance conveyance game. Ambulance can be access infrastructure, but if alternatives are not funded, emergency department conveyance becomes the default.

Game 8: the scope-of-practice supply game. If funding recognises too narrow a provider group, safe supply from nurses, nurse practitioners, pharmacists, therapists and paramedics is suppressed.

Game 9: the telehealth/local supply game. Telehealth can extend care or hollow out local in-person supply.

Game 10: the co-payment calibration game. Co-payments can signal demand or block necessary care.

Game 11: the key performance indicator salience game. What is measured at the top tier gets managed. What is invisible gets ignored.

Game 12: the equity and trust game. National benefits do not replace trust, kaupapa Māori provision, Pacific models, outreach and local relationships.

Game 13: the political economy game. The same policy can be framed as pro-market, anti-GP, anti-PHO, pro-patient, neoliberal, pragmatic, or equity-enhancing depending on who tells the story.

Game 14: the data observability game. Unmet need is hard to fund if nobody can see it.

Game 15: the place-based accountability game. Demand-led benefits can create cherry-picking unless someone remains responsible for whole populations.

Game 16: the current reform sufficiency game. Current reform is real. The question is whether it changes enough incentives.

Game 17: the formula-fixation game. Stakeholders can fight about weights forever while the deeper supply constraint remains.

Game 18: the urgent-care policy game. Urgent care can reduce emergency department pressure, but only if integrated with primary care, ambulance, data and workforce.

Game 19: the uncapped entitlement / fiscal-governance game. Eligible activity can be uncapped at the envelope level, but only if item prices, audit, scope, co-payment and place rules are strong.

This is why the recommendation is a hybrid.

Each game has a failure mode. The policy design needs to avoid all of them at once.

That is difficult.

But it is more honest than pretending one funding model can solve the whole system.

How to read the map

The map is not meant to be clever for the sake of it. It is meant to stop us arguing about only one piece of the system at a time.

If we argue only about capitation, we miss urgent care. If we argue only about urgent care, we miss ambulance. If we argue only about ambulance, we miss Accident Compensation Corporation. If we argue only about Accident Compensation Corporation, we miss provider scope. If we argue only about provider scope, we miss place-based accountability.

Each game is a small strategic trap. The policy problem is that the traps interact. A fix in one place can create a problem somewhere else.

That is why a hybrid model is needed. It is not a slogan. It is a way of admitting that no single lever is strong enough to solve the whole access problem.

Why nineteen games is not overcomplication

Nineteen sounds like a lot. But the health system is already playing these games. Naming them does not create complexity. It reveals it.

What would change my mind?

I would be less convinced if stakeholders scored most of these games as unreal, unimportant or already solved. That is exactly why the mapping should be tested with people inside the system.

---

Deep dive (optional, not required reading): I’ve kept the fuller explanation, game table, modelling notes and full source list in the [appendix for this post](../appendices-v1.6.0/appendix-14-the-19-games-a-map-of-the-new-zealand-primary-care-funding-problem-v1.6.0.md).

Useful links

• Ministry of Health: capitation reweighting

• Treasury: Vote Health 2025/26 Estimates

• Ministry of Health: Health Crown entities and Health New Zealand roles

• Ministry of Health: PHO finances briefing

• Ministry of Health: meeting with General Practice New Zealand, July 2025

---

Public companion links

• Interactive Streamlit dashboard and model lab

• GitHub Pages report, reading map and release model card

v1.8.1 model update

The current Streamlit model release is v1.8.1. Its public aggregate validation lane is public_aggregate_validated, and its claim level is empirically_supported_if_gated for registered gates only.

Tracks 050-074 turned the game map into a governed validation surface: public-only registries, parameter traceability, calibration gates, structural uncertainty, VOI, release audit, and claim-boundary review.

Claim boundary: The game map remains a decision-support frame. It does not prove causal effects or implementation performance.

One tribe says: targets save lives; stop whining and deliver.

The other says: targets are crude, political and gamed; abolish the lot.

Both are too tidy. Health systems are not tidy. They are messy, adaptive, queue-ridden organisms with clipboards.

The evidence from New Zealand (NZ) is mixed in exactly the way game theory would expect. In plain English, game theory just means thinking about how people respond to rules, rewards and pressure. The [plain-English glossary](../glossary/00plainenglish_glossary.md) has the short definitions used across the series.

Targets can improve things

The old emergency department (ED) target appears to have produced real improvements. The Health Research Council summarised evidence that the six-hour ED target was associated with reduced ED stays, reduced crowding, fewer patients leaving before care was completed, and around 700 fewer ED deaths than predicted if pre-target trends had continued. That summary is here.

A BMC Health Services Research study similarly found the target stimulated improvements in patient flow, while also noting that improvements plateaued and that use of short-stay units complicated the story. Read it here.

So yes, targets can matter.

But then the system learned the game.

Targets can also be gamed

Research on New Zealand’s ED target found that gaming increased over time, including clock-stopping and shifting patients into short-stay or observation units. The International Journal of Health Policy and Management paper is here.

This is not shocking. It is a repeated game: the same system faces the same target again and again, then learns which behaviours make the dashboard look better.

At first, a target can reveal slack and force coordination. People fix obvious bottlenecks. Flow improves. The low-hanging fruit is harvested.

Then the fruit gets higher. Demand keeps coming. Beds are full. Staff are tired. The target remains. The ministerial phone call still looms.

At that point, the system faces a choice:

• improve the underlying pathway;

• admit defeat; or

• redefine success.

Humans being humans, option three tends to make a cameo.

Goodhart is not an argument against measurement

Goodhart’s Law is usually paraphrased as: when a measure becomes a target, it stops being a good measure. More simply, a number can stop describing reality once people are rewarded or punished for hitting the number.

Fair. But this is not an argument for no targets. It is an argument for better target design.

A speedometer can be gamed too. You can disconnect it. You can only drive carefully past the camera. You can build a car culture around avoiding fines rather than driving safely.

The answer is not to abolish speedometers. It is to use speedometers alongside road design, enforcement, crash data, driver licensing and common sense.

Health targets are similar. They need companion measures.

The Ministry of Health (MoH) briefing on gaming and unintended consequences is refreshingly clear. It says targets can improve performance by influencing provider behaviour, but can also generate gaming. It recommends a “Goldilocks zone” for targets, clinical engagement, a range of measures, disaggregation, audit and review. In this series, audit means an independent check that the reported numbers match what is really happening. The briefing also notes that wait-time targets are more vulnerable to gaming through definitions, business rules and clock-stopping than immunisation targets.

That is exactly the right instinct.

Do not design targets for angels. Design them for people.

The behavioural economics bit

Targets create reputational incentives. Nobody wants to be the district at the bottom of the dashboard. That can be useful. Social comparison is powerful.

But shame is a blunt instrument. Too little pressure and nothing changes. Too much pressure and people protect themselves.

That is the Goldilocks problem:

• too soft: no one changes behaviour;

• too hard: people game, burn out or hide problems;

• just right: people focus, coordinate and improve.

The trick is to make honest improvement more rewarding than clever avoidance.

Red team: the anti-target case

A critic could say:

> Nice theory, but targets are political theatre. They narrow attention, encourage gaming, punish stretched clinicians, and allow governments to claim success without funding the system.

That critique is serious. Emergency nurses and perioperative nurses have warned that tough targets could backfire without capacity and staffing; Kaitiaki’s report is here.

Targets without capacity are just vibes with consequences. A target does not create an inpatient bed. It does not recruit a perioperative nurse. It does not build aged residential care capacity. It does not create a general-practitioner (GP) appointment.

So yes: targets can become a way to blame the delivery arm for a resource problem created elsewhere.

Reverse red team: the no-targets problem

But the anti-target position has its own weakness.

Without targets, performance failure does not vanish. It just becomes less visible.

The queue remains. The delay remains. The inequity remains. The difference is that the public loses a simple, repeatable way to see whether anything is improving.

That is not sophistication. That is darkness with better vocabulary.

Abolishing targets because they can be gamed is like abolishing financial accounts because fraud exists. The better answer is audit, verification and smarter reporting.

Respectful contrast

Richard Hamblin and Carl Shuker argue that tight performance measurement can worsen distrust and that a better approach may involve collaboration between consumers, local providers and central agencies around high-level national goals. Their commentary is here.

I agree with the trust point. But I would not throw away national targets. I would pair them with local co-design, independent audit and published balancing measures, meaning safety checks that show whether one improvement created a problem somewhere else.

The national target says: this is the service promise.

The local plan says: this is how we meet it without lying to ourselves.

Practical advice: every target needs a gaming-risk card

| Question | Why it matters | |---|---| | Can the clock be stopped? | If yes, audit the clock. | | Can patients be reclassified? | If yes, publish category shifts. | | Can easier cases be prioritised? | If yes, publish equity and clinical-priority breakdowns. | | Can quality worsen while timeliness improves? | If yes, publish readmissions, mortality, patient experience and staff indicators. | | Can the target hide a pre-target queue? | If yes, measure the full pathway. |

Closing line: The correct response to Goodhart is not panic. It is design.

Conceptual diagram

```mermaid flowchart TB A[Target pressure] --> B{System response} B --> C[Improve pathway] B --> D[Shift classification] B --> E[Stop or reset clock] B --> F[Prioritise easier cases] C --> G[True performance gain] D --> H[Gaming risk] E --> H F --> I[Equity risk] H --> J[Independent audit] I --> K[Equity-weighted reporting] J --> L[Target refinement] K --> L L --> A ```

Related in this series: [Plain-English glossary](../glossary/00plainenglish_glossary.md) | Previous: [The Queue Has Entered the Chat](../post-01/01-the-queue-has-entered-the-chat.md) | Next: [Little’s Law in the Waiting Room](../post-03/03-littles-law-in-the-waiting-room.md) | Practical companion: [The RareInsights Health Target Design Test](../bonus-design-test/05-health-target-design-test.md)

Positionality note: The health-policy lesson here is blunt: a system can look rational on paper while missing the person in front of it.

The central sentence of this series is simple enough to remember:

A thing can perform intelligence without inner life. A person can be inwardly present without performing intelligence in ways others recognise. Once you see the distinction, it appears everywhere.

In AI, performance is the source of fascination. The system writes, reasons, imitates tone, summarises books, offers comfort, and sometimes seems to understand the emotional shape of a conversation. It performs many outward signs of mindedness. But those signs do not settle whether anything is felt from within. The AI-consciousness literature is careful because the evidence is indirect.

In childhood, the problem reverses. The infant cannot explain herself. The toddler cannot offer a theory of her own mind. The child’s agency is partial, growing, uneven, and dependent. Yet no decent moral world waits for full performance before recognising the child as a subject of care. The UN Convention on the Rights of the Child does not make children’s rights depend on adult-like fluency.

In disability, the distinction becomes urgent. A person who communicates slowly, atypically, or not through speech may be misread as less present. A person who cannot work may be treated as less valuable. A person who needs extensive support may be seen as a burden rather than as a participant in a shared human world. The Convention on the Rights of Persons with Disabilities exists partly because legal dignity has to survive difference.

In genetics, performance may enter before birth. Future children are sometimes imagined through probabilities: risk scores, trait predictions, possible achievements, possible costs. Parents do not need to be cruel to be drawn into optimisation. Love itself can become anxious, managerial, and future-facing.

In economics, performance becomes price. Labour markets reward some capacities and ignore others. Care work is indispensable but undervalued. Dependency appears as expense. The person who produces measurable output is easier to count than the person whose value is relational, fragile, slow, or unmarketable.

This does not mean performance is irrelevant. It matters whether a patient can consent, whether a worker can do a job, whether an AI system can perform dangerous tasks, whether a genetic test can reliably reduce suffering. Competence and evidence matter.

The mistake is turning performance into the gatekeeper of worth.

That mistake has a long history. Many moral theories have been tempted by elegant thresholds: rational agency, self-consciousness, autonomy, preference formation, reciprocity, linguistic ability. These concepts are not useless. They help us think about responsibility, rights, consent, and justice. But they become dangerous when treated as the whole story.

Human life is full of people who matter before, after, beneath, or beyond those thresholds.

The baby matters before rational agency. The dying parent matters after independence fades. The disabled adult matters whether or not they can work. The person with dementia matters even when memory loosens its grip. Care does not begin when performance becomes impressive.

This is why a philosophy of consciousness after AI cannot be only a theory of consciousness. It needs a wider moral vocabulary: sentience, vulnerability, relation, embodiment, dependence, rights, history, and justice.

A society organised around performance will misrecognise both machines and humans. It may over-read the machine that performs humanity and under-read the human who does not perform according to social expectation.

The machine has forced an old question back into public life.

But the answer should not be a new intelligence test.

It should be a better way of seeing.

Three cautions: performance matters but is not moral worth; consciousness matters but is not the only moral category; dignity should not depend on being easy to measure.

Positionality note: Clinical systems make embodiment impossible to ignore. A patient is never just a stream of data; they arrive with sleep, pain, medication, money, transport, fear, and time.

A mind has a morning.

It wakes badly or well. It has a headache. It forgets to eat. It feels the pressure of an appointment, the ache in a joint, the embarrassment of needing help, the strange brightness of a room after poor sleep.

AI has no morning in this sense.

It may process time-stamped inputs. It may simulate fatigue in text. It may describe hunger with literary precision. But it does not wake into a body that must be maintained. It does not negotiate pain, hormones, medication, temperature, infection, digestion, grief, ageing, disability, or the social vulnerability of needing care.

That difference matters.

A lot of public talk about AI treats minds as if they were mainly information processors. The system receives inputs, transforms them, produces outputs. On that picture, consciousness becomes a puzzle about architecture: enough complexity, enough integration, enough self-modelling, and maybe experience appears.

Maybe. But embodiment reminds us that human consciousness is not merely computation happening somewhere. It is lived from a body, through a body, and often against the limits of a body.

The embodied cognition tradition has long argued that thinking is not sealed inside the skull. Perception, action, posture, environment, tools, and social interaction shape cognition itself. In health policy, the same point appears less romantically: housing, nutrition, disability support, transport, and clinical access change what a person can think, feel, and do.

This is why the body matters for a series about AI consciousness. It does not prove that machines can never be conscious. Some philosophers think a non-biological system could, in principle, have experience. But it does mean that human minds are not disembodied performances of intelligence.

A child’s consciousness is bodily from the start: hunger, touch, sound, warmth, distress, comfort. Disability often makes the politics of embodiment visible: pain, fatigue, sensory overload, access, assistive technology, medication, care routines. Ageing changes consciousness because the body changes the horizon of action.

A chatbot can sound calm at 3 am. A human being at 3 am may be frightened because the fever is worse, the child is crying, the shift starts in four hours, or the wheelchair battery is low.

The strongest critic will say: embodiment is not the same as consciousness. A machine might have a different route to experience. Biology may be familiar, not necessary.

That is fair. We should not use embodiment as a lazy veto. But we should also resist the opposite mistake: treating the absence of a body as a mere implementation detail. Bodies are not decorative shells around minds. They are part of how human minds become morally and socially real.

This matters for healthcare AI too. A model can predict deterioration, summarise a record, or triage a message. But the patient’s life is not the record. WHO’s AI ethics guidance is useful partly because it insists that AI systems sit inside human institutions, rights, responsibilities, and risks.

A mind has a morning. It also has a world.

And when we ask who counts as a mind, we should be careful not to mistake the fluency of a bodiless system for the presence of a life that has to get through the day.

Three cautions: embodiment is not a simple veto on machine consciousness; human consciousness is never just output; clinical data is not the same as a life.

Figure 13. Primary care funding architecture post 13 visual preview.

A demand signal means people face some cost when they use a service. In theory, this discourages low-value use and helps keep the system financially sustainable.

But health care is not like buying a coffee.

Patients often do not know whether their symptom is minor or serious. A parent with a feverish child may not be “shopping”. An older person with chest discomfort may be frightened. A person with mental distress may not have the capacity to navigate options. A rural patient may have only one practical choice. The tendency for policymakers, administrators and even some clinicians, to apply value-judgements to perceived “low value” presentations, forgets this information assymetry.

If co-payments are too high, people delay care. Delayed care can become more expensive care.

The New Zealand Health Survey shows cost is already a barrier for many people. In 2023/24, one in six adults reported not visiting a general practitioner because of cost. Waiting time was an even more common barrier.

So the co-payment question is not whether prices matter. They do. The question is how to use co-payments without letting them ration necessary care by income.

In an uncapped eligible fee-for-service model, co-payments could help manage demand. But they need guardrails.

The diagram below shows the basic equity problem. More price-sensitive patients reduce use more sharply as co-payments rise. Those patients are often the very people the system should most protect.

The Accident Compensation Corporation (ACC) system is useful here too. ACC contribution rates are capped by regulation. If provider costs rise faster than the regulated contribution, patients may face higher co-payments. The Ministry of Business, Innovation and Employment (MBIE) explicitly notes that rising co-payments may stop some claimants accessing treatment.

That is exactly the risk in primary care. An uncapped activity stream is only equitable if the patient price is controlled enough for necessary care to be used.

This is also why I do not think “free care for everything” is the only equity model. Free care can increase demand, but if supply is still capped, the rationing may move to waiting time. Waiting time is also an equity problem.

The goal is not simply low price. The goal is effective access.

A person should be able to get clinically appropriate care without being blocked by cost, distance, waiting time, digital exclusion or professional bottlenecks.

Co-payments can remain in the model. But they should be designed as a calibrated signal, not a blunt rationing weapon.

The uncomfortable trade-off

Co-payments are politically uncomfortable because they sit between two real concerns.

On one side, a zero-price service can create demand that is hard to manage, especially when supply is limited. On the other side, fees can stop people getting care they genuinely need. Both things can be true.

That is why the real question is not whether co-payments are good or bad in the abstract. The question is where they sit, who pays them, how large they are, what services are protected, and what happens to people with high need. In mathmatical terms, the question is where the equilibrium lies. In game theory terms, the question is what behaviours, incentives, rules and assumptions, influence the strategic games.

A co-payment for a low-risk convenience contact is not the same as a co-payment for a child, a person with multiple long-term conditions, a rural patient with transport costs, or someone delaying care because rent is due. If co-payments are used, they need to be calibrated. They should not become the main way the system rations care.

Why equity protections have to be designed up front

Equity protections cannot be an afterthought. Once a funding model is running, providers and patients adapt to it. Fees become normal. Workflows become fixed. Business models emerge. Changing them later is hard. That means the protections have to be designed into the schedule from the beginning.

Some groups may need zero or very low co-payments. Some contact types may need maximum fees. Some rural services may need extra public subsidy because travel and thin markets make care more expensive. Some high-need patients may need annual caps on out-of-pocket costs.

One commenter asked why should there be “equality” or equity protections. Putting aside normative arguments (which are important), the economic rationale is positive and negative externalities. That is, that where access and quality of care isn’t equitable, the negative effects of this on the individual can have negative effects on families, communities and societies as a whole. Examples include the spread of communicable diseases, but also social impacts such as homelessness, and lawlessness.

What would change my mind?

I would be less convinced if co-payments could be used as a demand signal without worsening unmet need, or if equity protections could not be designed tightly enough to protect high-need patients.

Useful links

• Ministry of Health: New Zealand Health Survey annual update

• Ministry of Health: capitation reweighting

• Accident Compensation Corporation: paying patient treatment

• Ministry of Business, Innovation and Employment: ACC regulated payments for treatment

• Ministry of Health: primary care health target

Public companion links

• Interactive Streamlit dashboard and model lab

• GitHub Pages report, reading map and release model card

v1.8.1 model update

The current Streamlit model release is v1.8.1. Its public aggregate validation lane is public_aggregate_validated, and its claim level is empirically_supported_if_gated for registered gates only.

Public aggregate cost-barrier evidence is now part of the calibration and validation surface. This supports treating co-payments as both a demand signal and an equity-risk mechanism.

Claim boundary: Do not claim a specific co-payment reform will produce measured access, equity, fiscal, or hospital outcomes. The evidence status supports risk framing and validation priorities.

This is the key queueing theory point. Queueing theory is just the maths of waiting lines: how many people arrive, how fast they can move through, and where the hold-up forms. In a simple queue, waiting time depends on arrivals, service capacity, variability and bottlenecks. More people arriving increases pressure. Slower processing increases pressure. But the killer is often the downstream bottleneck.

In hospital terms: ED gets crowded when patients cannot move to the next place. That next place might be an inpatient ward, theatre, imaging, aged residential care, community support, or home with services. A bottleneck is the slow point that holds up the whole pathway.

Blaming ED for ED waits is often like blaming the plughole for the plumbing.

’s University of Auckland commentary makes this point bluntly: the deeper driver of ED crowding is often access block, meaning patients are stuck because inpatient beds or downstream care are unavailable. Read it here.

The queue equation

Queuing theory gives us a simple way to think about this:

Little’s Law:

Work in progress = throughput × time in system

Human translation: the number of people stuck in the system is related to how many are moving through it and how long each person remains there.

So if more patients arrive, or each patient spends longer in the system, queues grow. If capacity is variable and demand is unpredictable, queues grow faster than intuition expects.

Health systems do not need to be “inefficient” in a moral sense to produce long waits. They can be operating close to capacity, which is exactly when small shocks create large queues. Specifically, there is evidence that the same patient activity distributed through the same number of beds, is tolerated less well when those hospitals are smaller.

This matters for targets.

An ED six-hour target can create useful urgency. It can force hospitals to improve flow. It can make senior leaders care about bottlenecks they previously left to the frontline.

But if the system has no inpatient beds, no discharge support, no aged care capacity, and no staff buffer, then the target starts pushing on a wall.

At that point the game becomes ugly. Move the patient to a short-stay unit. Reclassify. Stop the clock. Make the dashboard better.

The patient is still in the system. The queue has just changed costume.

Why percentages can mislead

A target like “95% within six hours” is easy to understand. That is useful. But it hides the shape of the queue.

Imagine two hospitals:

Same target result. Very different reality.

That is why we need the distribution, not just the threshold.

Publish:

This is not dashboard maximalism. It is preventing one number from cosplaying as the whole system. Unfortunately this is something many clinicians, managers/administrators and policymakers do not consider.

Electives and first specialist assessment

The same logic applies to first specialist assessment and elective treatment.

The headline target is less than four months. That is clear. Good.

But the pathway has more than one queue:

A target at one point in the pathway can improve that point while hiding delay elsewhere.

For cancer, the current target measures treatment within 31 days from decision to treat. That is important, but it does not fully answer whether the patient waited too long before diagnosis, staging or decision-to-treat. The Ministry of Health target definitions are here.

Again: the answer is not to abolish the target. It is to measure the pathway.

Red team: queue theory can get too tidy

A red-team critic could say:

Queue theory risks making everything sound technical. The real issue is political: not enough workforce, not enough beds, underfunded primary care, and too much pressure on hospitals.

That critique is valid but incomplete. Queue theory does not replace politics. It disciplines it.

It tells us that promising lower waits without changing capacity, variability, demand or bottlenecks is not a plan. It is a slogan with a numerator.

Reverse red team: complexity is not an excuse

The opposite mistake is to say:

Because the system is complex, targets are simplistic.

That sounds clever, but it can become a policy escape hatch. Complexity should change how we measure, not whether we measure.

A complex system needs more careful measurement, not less public accountability.

Practical advice

For each target, publish a companion flow measure:

Closing line: A health target should not ask only, “Did the patient cross the line in time?” It should ask, “Where did the queue go?”

Conceptual diagram

Figure 12. Primary care funding architecture post 12 visual preview.

But telehealth is not the whole answer. Some problems need physical contact.

A clinician may need to listen to a chest, examine an abdomen, look in an ear, check a rash, assess hydration, dress a wound, remove sutures, feel a pulse, test mobility, take bloods, give an injection or simply see how someone walks into the room. In clinical genetics, some findings may be genuinely novel and very difficult to characterise without getting up close in person.

Some care also needs local knowledge.

Local clinicians know which services are actually available, which hospital is under pressure, which pharmacy is open, which families need extra support, which roads are cut off, which community providers can help and which patient is more unwell than they sound on the phone.

Telehealth can extend local care. It can also undermine it if used badly.

If scalable virtual providers absorb the easier contacts, local practices may be left with complex, time-consuming, lower-margin work. That could weaken local viability. In rural areas, that matters. Once local in-person supply disappears, it is hard to rebuild.

So the telehealth game is a complement-versus-substitute game.

In the good equilibrium, telehealth supports local care. It provides overflow capacity, after-hours triage, advice, follow-up, prescription support and specialist input. It works with local records and local teams. It helps determine when in-person care is needed.

In the bad equilibrium, telehealth cherry-picks simple contacts, fragments records, weakens continuity and lets policymakers pretend rural access has been solved when it has not.

This is why the current reform pathway needs careful evaluation. A 24/7 digital general practitioner service may be valuable. But it should be measured against more than consultation numbers.

Questions to ask:

This also matters for funding. If fee-for-service benefits are created, they should distinguish between contact types. Some contacts can be safely virtual. Some should have an in-person loading. Some should require local follow-up. Some should not be claimable virtually except in defined circumstances.

A rural in-person loading may be needed. That means an extra payment signal for care that is physically present in rural or underserved communities. It recognises that local care has costs that digital care does not.

The goal is not to slow telehealth. The goal is to stop telehealth being mistaken for total supply.

Telehealth is a bridge. It should not become an excuse to remove the clinic, the nurse, the pharmacist, the paramedic, the visiting general practitioner or the rural hospital from the community.

The rural risk

Telehealth can be wonderful for rural communities. It can reduce travel, improve follow-up, support medication reviews and give people faster advice when local appointments are scarce.

But it can also mask a deeper problem. If the local service keeps shrinking while telehealth grows, the community may appear to have access until someone needs examination, procedures, urgent assessment or continuity with a team that knows them.

That is the rural hollowing-out risk.

A funding model should therefore ask two questions at the same time.

First: can digital care improve access? Yes.

Second: does the model support local in-person capacity when that capacity is clinically necessary? It must.

The goal is not nostalgia for old models of care. The goal is a mixed access system where digital care, local clinics, outreach, ambulance alternatives and urgent care are all funded for the jobs they do best.

Telehealth can also change provider behaviour

Telehealth does not only change patient access. It changes the economics of supply. A digital provider can centralise clinicians, standardise workflows, reduce room costs and operate across geography. That can be good. It can also outcompete local in-person services for easier contacts.

If local clinics lose the simple work but remain responsible for the complex, procedural and urgent in-person work, their economics worsen. That is a classic cream-skimming problem. The easy activity becomes scalable and remote. The hard activity stays local, expensive and underfunded.

So the policy design has to ask: does telehealth add capacity to the system, or does it pull the profitable activity away from local services?

What would change my mind?

I would be less convinced if telehealth demonstrably replaced local in-person capacity safely across rural, complex, procedural, frail and high-need populations. I suspect it mostly extends care, which is still valuable.

Deep dive (optional, not required reading): I’ve kept the fuller explanation, game table, modelling notes and full source list in the [appendix for this post](../appendices-v1.6.0/appendix-12-telehealth-is-an-extender-not-a-replacement-for-local-supply-v1.6.0.md).

Useful links

• Ministry of Health: primary care health target

• Health New Zealand: National Primary Care Dataset and new primary care health target

• Cabinet material: Primary Health Care Funding Improvements

• Beehive: new and improved urgent and after-hours healthcare

• Health New Zealand: the Ambulance Team

Public companion links

• Interactive Streamlit dashboard and model lab

• GitHub Pages report, reading map and release model card

v1.8.1 model update

The current Streamlit model release is v1.8.1. Its public aggregate validation lane is public_aggregate_validated, and its claim level is empirically_supported_if_gated for registered gates only.

Telehealth remains modelled as an extender rather than a replacement for local supply. The public aggregate validation lane does not validate a delivery-mode effect.

Claim boundary: Do not claim telehealth reduces demand, replaces local workforce, or resolves rural access constraints. Those are implementation claims outside the v1.8.1 evidence status.

For a long time, a tumour’s identity was strongly tied to two questions: where is it, and what does it look like under the microscope?

Those questions still matter. Any story in which genomics replaces the pathologist is a story told by someone who has not spent enough time near pathology. Tissue, architecture, staining, clinical context- these remain indispensable.

But cancer is also where genomic splitting has become clinically unavoidable. The microscope has gained a molecular layer.

Integrated diagnosis

The WHO classification of central nervous system tumours is a useful example because it formalises what is often called integrated diagnosis. Histology and molecular findings are not treated as enemies. They are combined into a layered judgment.

This is philosophically elegant, even if the clinical reality is more complicated. It says: the disease category is not located at one level only. It is assembled from tissue, morphology, molecular alteration, and clinical meaning.

In haematological malignancy, similar pressures are visible. The WHO classification of haematolymphoid tumours makes molecular, cytogenetic, and clinical features part of the classificatory landscape.

What is the tumour, really?

Cancer forces a blunt question. Is the “real” tumour the anatomical site? The morphology? The cell lineage? The driver mutation? The pathway? The therapeutic target? The prognosis?

The sensible answer is: do not make one layer do all the work.

A tumour’s tissue of origin may matter for surgery, symptoms, and spread. Histology may matter for recognition and grading. Molecular profile may matter for prognosis and therapy. Treatment response may matter for the category’s practical usefulness. None of these layers is fake. None is always sovereign.

This is why cancer is such a good case for constrained pluralism. The classification can be plural without being arbitrary. It is constrained by evidence and consequence.

The therapy problem

Molecular classification becomes especially powerful when it changes treatment. A mutation, fusion, or biomarker may define eligibility for a targeted therapy or clinical trial. In that context, molecular evidence can override older groupings.

But here too we should be careful. A treatment-actionable category is not necessarily a disease kind in the deepest metaphysical sense. It may be a practical category organised around intervention.

That does not make it less important. It may make it more important.

In health systems, categories that enable action often acquire authority quickly. Once a biomarker determines treatment access, it becomes more than a laboratory detail. It becomes a gate.

The danger of the clean story

The clean story says: old cancer classification was visual and crude; new cancer classification is molecular and true.

The better story says: cancer classification is becoming layered. Molecular evidence has gained authority, but not absolute authority. Diagnosis now involves negotiation among levels of evidence.

This matters because the old and new maps do different jobs. A molecular subtype may be excellent for trial design and poor for explaining symptoms. A histological category may be clinically useful even when molecularly heterogeneous. A treatment category may be urgent even if its boundaries shift as evidence evolves.

Why cancer belongs in a philosophy series

Cancer classification shows that disease names are not just nouns. They are decisions about which evidence is allowed to define the object.

That is why a tumour walking into a sequencer is not only a technical event. It is a philosophical one. The sequencer does not simply add data to a pre-existing category. Sometimes it helps decide what the category is.

The health-system angle

Cancer makes the health-system stakes unusually visible. If a molecular feature changes prognosis or therapy, then classification can become a route to treatment. But that route depends on access to testing, tissue handling, reporting standards, reimbursement, and clinician interpretation. A molecularly defined tumour entity is not only a biological discovery. It is also a practical achievement.

This is why I am wary of saying that the genome simply reveals the “real” tumour. In practice, a tumour diagnosis is assembled from tissue, imaging, histology, immunohistochemistry, molecular testing, clinical history, and available treatment pathways. The integrated diagnosis is not a slogan. It is an epistemic compromise that works because different kinds of evidence are allowed to carry different weights.

The claim I want to keep disciplined

Cancer makes the philosophical issue clinically obvious: molecular features can alter diagnosis, prognosis, treatment, and trial eligibility.

I am not saying histology has been replaced. The stronger point is that histology now often shares diagnostic authority with molecular evidence.

The useful version of the argument is not that genomics reveals the one true disease map. It is that genomics changes the evidence available for drawing maps, and therefore changes the arguments medicine has about what should be grouped together.

A visual way to hold the idea

Further reading

• WHO CNS5 overview

• WHO haematolymphoid tumours overview

A question to leave with

Is the real tumour the tissue, the morphology, the mutation, the pathway, or the treatment target?

Figure 11. Primary care funding architecture post 11 visual preview.

If the funding system only recognises general-practitioner-led activity, then the system will behave as if primary care supply depends mainly on general practitioner numbers.

General practitioners are essential. This is not an argument against them. But they are not the only clinicians who can safely provide primary care activity.

Nurse Practitioners can assess, diagnose, prescribe and manage many conditions. Pharmacists can support medicines optimisation, minor ailments, vaccination, prescribing in defined contexts and medication review. Physiotherapists can manage a wide range of musculoskeletal problems and I’ve worked with safe and effective physiotherapist medicine prescribing programs. Paramedics can support urgent assessment, triage, and alternative disposition, including a wide range of urgent care and community treatments. Nurses can provide a wide range of planned and acute services. Māori and Pacific providers can deliver care in ways that mainstream models often cannot or do not.

If all of that activity is clinically appropriate, why should funding be artificially constrained by one professional pathway? The answer should be clinical governance, not professional protection.

The funding system should ask:

Those are safety questions. They are different from saying only one profession can generate supply.

The current primary care access target is already framed around a general practice provider, not only a general practitioner. That is helpful. But the broader architecture should go further. It should recognise defined contact types across the primary, urgent, community and ambulance workforce.

For example:

Each of these could have different rules, item prices, co-payment protections, audit triggers and reporting requirements. That is not deregulation. It is smarter/right-sized/fit-for-purpose regulation.

The problem with a doctor-to-patient ratio mindset is that it can freeze the system around a workforce that is already short. The problem with a pure telehealth model is that it can scale simple contacts but hollow out local in-person supply. The problem with a pure capitation model is that it may fund responsibility without paying enough for activity. Each of these have already happened in New Zealand.

A better model asks what work needs doing, then asks who can safely do it. This is where the uncapped benefits schedule becomes useful.

A National Primary Medical Benefits Schedule (PMBS) could be provider-neutral where appropriate. That means the item is linked to the service, not automatically to a professional guild:

The point is to fund safe activity, not preserve artificial bottlenecks. That is particularly important in rural areas.

A rural community may not have enough general practitioners. But it may have a nurse practitioner, pharmacist, paramedics, physiotherapist, occupational therapist, visiting general practitioner sessions, telehealth support, rural hospital staff and community providers. A good funding model would assemble those capabilities rather than pretending supply only exists when a traditional general practitioner clinic is fully staffed.

The recommendation is simple:

Let scope of practice and clinical governance determine what providers can do. Do not let the payment model artificially restrict who can generate safe primary care supply.

Funding should not freeze old professional boundaries

Clinical safety absolutely matters. This matters much more than some people politicking about this either know about, or care to admit. Prescribing rules, scope of practice, competence, supervision, escalation pathways and audit all matter.

But funding rules should not make the workforce problem worse by pretending that only one professional group can generate useful primary care activity.

Some care needs a general practitioner. Some care needs a nurse practitioner. Some care can be provided by a pharmacist, physiotherapist, psychologist, paramedic, nurse, health coach, kaiāwhina or other member of a broader team. The right answer depends on the problem, the patient, the context and the risks. Figuring out who can do what is the role of health leadership and governance, and shouldn’t be artificially constrained by arbitrary upstream rules.

If the funding model only pays properly when a doctor is involved, it creates an artificial bottleneck. It may also waste scarce medical time on contacts that other professionals could handle safely. Not only does it constrain who can provide the service but it places increased burden on an increasingly limited number of providers, likely exacerbating market failure.

A better funding model would define eligible contact types, then allow the appropriate provider to generate the claim if they are accredited and working within scope. That is not anti-doctor. It is pro-access, pro-team and pro-safety.

What would change my mind?

I would be less convinced if provider scope expansion did not create safe additional supply, or if clinical governance could not distinguish appropriate pharmacist, nurse practitioner, nurse, allied health, paramedic and general practitioner roles.

Deep dive (optional, not required reading): I’ve kept the fuller explanation, game table, modelling notes and full source list in the appendix for this post.

Useful links

• Accident Compensation Corporation: paying patient treatment

• Ministry of Health: capitation reweighting

• Ministry of Health: primary care health target

• Health New Zealand: National Primary Care Dataset and new primary care health target

• RACGP/AJGP: understanding general practice funding models

Public companion links

• Interactive Streamlit dashboard and model lab

• GitHub Pages report, reading map and release model card

v1.8.1 model update

The current Streamlit model release is v1.8.1. Its public aggregate validation lane is public_aggregate_validated, and its claim level is empirically_supported_if_gated for registered gates only.

The v1.8.1 model sharpens the supply-generation question but does not validate workforce effects. Workforce remains a key uncertainty and evidence-priority domain.

Claim boundary: Do not claim workforce expansion, participation shifts, or scope-of-practice impacts. The model is not valid for workforce-effect estimates.

Figure: DSM-5 changed the clinical grammar of autism, even if formal criteria comparisons are complex.

Positionality note: This post is especially shaped by my clinical experience. In day-to-day paediatrics, DSM-5 did not feel like a minor paperwork change. It changed how autism was recognised, communicated and documented in practice. The economist in me, though, remains cautious about jumping from that clinical experience to claims about population prevalence.

DSM-5 did not merely change a checklist. In clinic, it felt more like a change in grammar- the difference between having a few scattered phrases and finally having a language.

That is not the kind of sentence that appears neatly in a prevalence paper, but it matters in practice. A diagnostic manual can change what clinicians ask, what parents remember, what teachers notice, what reports emphasise, and what service systems accept as legible. Manuals do not simply describe the world. Sometimes they teach a system how to see.

That is my clinical experience. DSM-5 helped consolidate autism as a spectrum. It gave sensory features a formal home. It allowed developmental history to matter. It allowed autism and ADHD to be diagnosed together. It introduced specifiers and severity levels. It changed how clinicians wrote reports, how schools read them, how families explained their child, and how service systems attached supports to documentation.

But the evidence is more complicated than the sentence “DSM-5 broadened autism.”

DSM-IV and DSM-IV-TR divided the field into Autistic Disorder, Asperger’s Disorder, PDD-NOS and related pervasive developmental disorder categories. DSM-5 consolidated these into Autism Spectrum Disorder, with two main domains: social communication/social interaction, and restricted/repetitive behaviours, including sensory features. Many clinicians experienced this as more coherent. Families did too. The child who might have been scattered across PDD-NOS, Asperger’s, sensory processing issues, anxiety and ADHD now had one more integrated developmental formulation.

At the same time, formal comparisons have sometimes suggested DSM-5 can be narrower under strict research criteria. A JAMA Psychiatry analysis of ADDM data estimated that applying DSM-5 criteria could produce lower prevalence estimates than DSM-IV-TR in some contexts (JAMA Psychiatry DSM-5 prevalence analysis). Earlier studies of proposed DSM-5 criteria raised concerns that some people previously diagnosed under Asperger’s disorder or PDD-NOS might not meet the new criteria if applied rigidly. CDC analysis of 2014 ADDM data found that most children met both DSM-IV-TR and DSM-5 case definitions, with DSM-IV-TR prevalence only modestly higher in the analysed population (CDC 2014 DSM comparison).

So how can DSM-5 feel broader in clinical rooms while appearing narrower in some formal comparisons?

Because formal criteria and clinical recognition are not the same thing.

A diagnostic manual is not just a threshold. It is a language. It tells clinicians what to look for, what combinations belong together, what comorbidities are allowed, and what histories count. It changes not only who qualifies but who is considered.

The inclusion of sensory criteria is a good example. Sensory sensitivities were not invented by DSM-5; Autistic people and families had described them for decades. But formal inclusion changed the way sensory symptoms were heard. A child covering their ears, refusing clothing textures, gagging at food smells, melting down under fluorescent lights, or seeking deep pressure could now be understood as having central autism features, not merely behaviour problems or parenting battles. That did not automatically increase formal prevalence in every dataset, but it changed recognition.

ADHD comorbidity is another example. DSM-IV discouraged diagnosing ADHD when autism was present. DSM-5 allowed both. Clinically, this mattered enormously. It meant that a child could be understood as autistic and impulsive, autistic and inattentive, autistic and hyperactive, rather than being pushed into one category at the expense of another. That may change documentation, treatment pathways and family understanding. It may also alter who is referred for autism assessment in the first place, because ADHD no longer excludes autism from consideration.

Developmental history also matters. Many children, especially girls and highly compensating children, do not display every feature in the consultation room. Some perform eye contact. Some script social language. Some suppress stimming. Some have learned rules through exhausting effort. DSM-5’s allowance for symptoms by history supports a more developmental assessment. This does not mean “anything goes.” It means the consultation room is not the whole child.

Severity levels and specifiers were intended to describe support needs and associated features, but they also became administrative language. Reports began to say ASD Level 1, Level 2 or Level 3; with or without intellectual impairment; with or without language impairment; associated with known medical or genetic conditions. In service systems, such wording can affect access, even when the clinical reality is more nuanced. Support needs vary by context. A child can look “Level 1” in a quiet clinic and require far more support in a chaotic school.

The Danish reporting-practice paper is useful here because it reminds us that changes in diagnostic criteria and reporting practices can materially affect measured prevalence. Hansen, Schendel and Parner found that changes in diagnostic criteria in 1994 and inclusion of outpatient contacts in 1995 explained a large proportion of increased reported ASD prevalence in Denmark (Hansen et al., JAMA Pediatrics 2015). The point is not that Denmark proves DSM-5 caused Australia’s rise. The point is that autism prevalence is sensitive to how diagnostic systems classify and record people.

DSM-5 also interacted with a broader cultural shift. Autism became more publicly recognisable as a spectrum. Autistic adults wrote and spoke more. Parents learned about masking. Clinicians became more alert to girls, high-IQ children, twice-exceptional children, and children previously labelled anxious or oppositional. The National Autism Strategy now explicitly recognises that girls, women, gender diverse people and people with less overt presentations may be diagnosed later (National Autism Strategy publication). That recognition is not solely DSM-5, but DSM-5 helped provide a shared clinical grammar.

The danger is overclaiming. DSM-5 did not single-handedly cause the increase in autism diagnosis. DSM-5 did not simply broaden the criteria in every formal comparison. It did not eliminate underdiagnosis. It did not solve diagnostic access. And it did not make autism a purely subjective label. Good diagnosis still requires careful developmental history, observation, collateral information, functional assessment, differential diagnosis and clinical judgement.

The stronger claim is this: DSM-5 plausibly contributed to broader practical recognition and documentation of autism, even though its formal prevalence effects vary depending on how the question is asked.

In clinical work, this distinction matters. I have seen children whose sensory distress was finally taken seriously. Children whose ADHD no longer blocked autism formulation. Girls whose apparent social competence collapsed under developmental history. Parents who stopped thinking of their child as wilfully difficult and began to understand the child’s nervous system. Reports that gave schools language to support rather than punish.

DSM-5 did more than rearrange diagnostic furniture. It changed what clinicians could say, what families could hear, and what institutions could record.

That is not the whole prevalence story. But it is one important chapter in how autism became more visible.

Clinical broadening is not the same as epidemiological broadening

The cleanest way to write about DSM-5 is to separate the checklist from the clinical culture around the checklist. In strict research comparisons, DSM-5 can look narrower than DSM-IV-TR, particularly for some people who previously met PDD-NOS or Asperger’s Disorder criteria. That finding should be stated plainly because it protects the essay from overclaiming. But the clinical reality is still different.

DSM-5 changed the grammar of autism by making several things easier to say. It made autism a single spectrum rather than a set of neighbouring diagnoses. It made sensory differences diagnostically explicit. It made developmental history more central, which matters for a child who masks in the assessment room but unravels at home. It allowed clinicians to diagnose autism and ADHD together, which matters enormously in paediatrics. It encouraged specifiers and severity levels, which in turn influenced reports, service pathways and school documentation.

In practice, those changes can broaden recognition even when the formal criteria do not broaden prevalence in a simple statistical sense. A teacher may now recognise sensory overload as part of autism rather than defiance. A paediatrician may now see the ADHD-autism combination rather than choosing one formulation and discarding the other. A parent may now describe developmental history in a way that the clinician knows how to use. A psychologist may write a report that connects traits, support needs and functional impact in language that a school or funding system understands.

This is why the author’s clinical experience belongs in the series, but it should be framed carefully. DSM-5 did not single-handedly cause the autism rise. It was one part of a larger change in diagnostic culture: a new vocabulary that made some previously unnamed children easier to recognise, describe and support.

The ICD story adds another layer. Australia does not live inside DSM alone. Clinicians, services and datasets move between DSM concepts, ICD coding, Medicare language, school categories and NDIS evidence requirements. A manual can influence practice even when it is not the only coding system in use, because it shapes training, research, reports and the examples clinicians carry in their heads.

This makes DSM-5 a cause of recognition rather than a sole cause of prevalence. Its effects travel through people: through paediatricians who ask about sensory features, psychologists who document developmental history, psychiatrists who stop treating ADHD as an exclusion, teachers who read reports differently, and parents who finally see a pattern rather than a collection of moral failures.

So the honest formulation is not “DSM-5 made autism explode.” It is: DSM-5 was one important part of a broader shift in diagnostic language, and that shift made some real children more diagnosable.

Clinical grammar is not the same as criteria maths

This is why I would be careful about a sentence like “DSM-5 caused the increase.” It is too blunt. DSM-5 did not operate alone, and the formal comparison papers do not all point in one direction. Some people who fitted DSM-IV categories may be harder to classify under a strict DSM-5 reading, especially if the older presentation sat near PDD-NOS or Asperger’s Disorder.

But clinical practice is not only strict criteria arithmetic. Manuals also teach clinicians what to notice. DSM-5 made sensory experience part of the core conversation. It allowed developmental history to matter. It allowed autism and ADHD to be recognised together. It consolidated a fragmented family of categories into a single spectrum. Those shifts changed reports, referrals, school conversations and family explanations. The honest claim is therefore narrower and stronger: DSM-5 plausibly contributed to broader practical recognition and documentation, even if its formal epidemiological effect depends on the study design and population being examined.

So yes, I think DSM-5 mattered. I think it mattered in the practical, clinical way that often escapes tidy arguments about criteria tables. It changed what could be seen, and how it could be named. That is not the whole story of rising diagnosis, but it is a bigger part of the story than many people admit.

This essay is part of Counting Autism: diagnosis, data, incentives and dignity, a Rare Insights series on autism diagnosis, causation, incentives and dignity, written from my dual position as a paediatrician and economist. The linked series guide keeps the full reading order in one place, and you can follow the broader work at Rare Insights.

Further sources

• CDC DSM-IV-TR vs DSM-5 comparison — Quantifies overlap and differences between criteria.

• JAMA Psychiatry DSM-5 criteria analysis — Evidence that strict DSM-5 criteria may be narrower in some analyses.

• AAP clinical report on ASD — Clinical discussion of diagnosis and comorbidity.

• Hansen et al. 2015, JAMA Pediatrics — Reporting and diagnostic-change contribution.

• National Autism Strategy 2025–2031 — Contemporary spectrum and language framing.

Figure 8. Primary care funding architecture post 08 visual preview.

But the patient’s need can change. They might become frail. They might develop diabetes. They might need mental health care. They might have a child with repeated infections. They might need forms, follow-up, prescriptions, referrals and coordination.

The workload rises. The fixed payment does not necessarily rise with each contact.

At some point, the practice faces a choice. It can absorb the extra work, charge more, shorten appointments, limit enrolments, increase waits, rely on triage, use telehealth where possible, or direct people elsewhere.

That is not a moral failure. It is a predictable supply response.

New Zealand has public evidence of this pressure. A New Zealand Medical Journal study found that in 2022 only 28 percent of surveyed general practices were freely enrolling new people, while 79 percent had closed or limited enrolments at some point since 2019. Another area-based analysis found that 33 percent of general practices had closed books in June 2022.

Closed books are important because they show a boundary in the capitation model.

In theory, capitation gives a practice a financial incentive to enrol more patients. In practice, a practice may decide that more enrolments would overload staff, reduce quality or worsen burnout.

A wealthy, digitally connected person in an urban area has more options. A rural person, disabled person, older person, Māori or Pacific patient, or a person with low income may have fewer practical options.

This is where equity enters the model. When a system says it is controlling costs, it must ask: controlling costs for whom?

If the public budget is controlled by pushing cost, delay or risk onto patients, practices, ambulance or hospitals, the saving may be an illusion.

The need still exists.

The New Zealand Health Survey shows this access problem in plain terms. In 2023/24, one in four adults reported that the time taken to get an appointment was too long as a barrier to visiting a general practitioner. One in six adults reported cost as a barrier. Over five years, general practitioner visits decreased while emergency department visits increased for adults and children.

That does not prove one caused the other. But it supports the hypothesis that upstream access constraints and downstream hospital demand need to be analysed together.

The microeconomic point is straightforward:

A fixed funding envelope can ration by waiting time.

If patient need rises faster than funded upstream capacity, something has to give.

In a well-designed system, some of that increased need should be met in primary care, urgent care, community pharmacy, nurse practitioner services, allied health, paramedic pathways and virtual care.

In a poorly designed system, the need leaks into the emergency department.

That is why capitation should be kept, but not asked to do everything.

Capitation is for responsibility.

Fee-for-service is for eligible activity.

Place accountability is for whole populations.

Co-payment protections are for equity.

Data is for visibility.

The patient does not care what the funding model is called.

They care whether care exists.

The patient game is not theoretical

Patients play the game too, although usually not by choice. They make decisions inside the rules the system gives them.

If the general practice appointment is two weeks away, the patient may wait. If they are worried, they may go to urgent care. If they cannot afford the fee, they may delay. If they cannot travel, they may choose telehealth. If the problem gets worse, they may call an ambulance or present to an emergency department.

None of those choices is irrational. The patient is responding to access, cost, worry, transport, time off work and trust.

That is why access targets have to be interpreted carefully. A target that rewards fast appointments can help. But it can also create pressure to prioritise simple, bookable activity over complex planned care. It can change booking behaviour without actually solving capacity. So in this specific example, composite access targets need to be complimented by measures that speak to unwanted effects- such as total/average waiting time, referral rejection rate (a massive issue in New Zealand), etc.

For anyone wanting to see a better (not not neccesarily best) example of multi-tier system performance monitoring and management, look at the New South Walves Bureau of Health Information hospital performance dashboard. For comparison, NZ has ~5M and NSW has ~8M people. NSW is a highly hybrid public-private system, with most acute care provided publicly, thereby making it more difficult to be efficient compared with the NZ system, which is predominantly public.

What would change my mind?

I would be less convinced if fixed capitation did not affect the marginal decision to add appointments, or if consumers did not shift between delay, payment, telehealth, ambulance and emergency departments when access changes.

Deep dive (optional, not required reading): I’ve kept the fuller explanation, game table, modelling notes and full source list in the [appendix for this post](../appendices-v1.6.0/appendix-08-the-capitation-marginal-supply-game-and-the-consumer-access-game-v1.6.0.md).

Useful links

• Ministry of Health: capitation reweighting

• Ministry of Health: primary care health target

• Health New Zealand: National Primary Care Dataset and new primary care health target

• Ministry of Health: New Zealand Health Survey annual update

• Cochrane: payment methods for outpatient healthcare providers

---

Public companion links

• Interactive Streamlit dashboard and model lab

• GitHub Pages report, reading map and release model card

v1.8.1 model update

The current Streamlit model release is v1.8.1. Its public aggregate validation lane is public_aggregate_validated, and its claim level is empirically_supported_if_gated for registered gates only.

Public aggregate access evidence now supports the bounded validation lane for the capitation and consumer-access argument. The strongest claim is still architectural: payment design should be tested against access constraints and supply response, not assumed to work.

Claim boundary: Do not claim patient-level access improvements, appointment gains, or distributional effects. The update remains publicaggregatevalidated only for registered aggregate evidence gates.

The current reform pathway is more substantial than simply tweaking capitation.

Figure 5. Primary care funding architecture post 05 visual preview.

The Government is reweighting capitation. It is introducing a primary care access target. Health New Zealand is building a National Primary Care Dataset. There is performance-based funding. There is 24/7 digital general practitioner access. There is a large urgent and after-hours programme. There are workforce initiatives. There is policy work on Primary Health Organisations. There is a separate appropriation for primary, community, public and population health services.

That matters.

If the critique is written as if the only thing happening is capitation reweighting, it becomes too easy to dismiss.

So the better question is not:

Why is New Zealand only changing the capitation formula?

The better question is:

Does the current reform pathway change the game enough?

That is a much stronger question.

The current reform has some obvious strengths.

First, the new capitation formula should be fairer. It includes factors such as age, sex, multimorbidity, rurality and deprivation. That is better than an old formula built from late-1990s utilisation patterns.

Second, the primary care access target makes access visible. For the first time, primary care has a national access target. The proposed target is that more than 80 percent of people can access an appointment with a general practice provider within one week.

Third, the National Primary Care Dataset should improve observability. If we can see when appointments are booked, when people are seen, and what the outcome was, we can start to understand access rather than guessing.

Fourth, urgent and after-hours care is being expanded. The Government has announced investment to support a goal of 98 percent of New Zealanders being able to access urgent care within one hour’s drive.

Fifth, official policy is starting to talk about the broader general practice team, not only doctors.

All of that is important.

But there are still gaps.

A target does not create appointments by itself.

A dataset does not create supply by itself.

A better capitation formula does not necessarily fund the next urgent contact.

A digital service does not replace all local, in-person care.

Urgent care does not solve routine continuity.

Performance payments do not necessarily fix the base economics of practices.

Separate appropriations do not automatically prevent hospital pressure from dominating political attention.

This is why I think the current reform should be treated as the comparator, not the endpoint.

The current reform may improve allocation, visibility and some access. The question is whether it removes the hard cap on eligible primary medical activity.

I do not think it fully does.

The most important missing mechanism is still the marginal supply signal.

If a practice has extra patient demand but no viable way to fund the next clinically useful contact, the system still rations. If a nurse practitioner, pharmacist, paramedic, physiotherapist, general practitioner or other clinician can safely deal with a defined problem, but the funding architecture does not let that activity be generated and paid for, supply is still artificially constrained.

That is why the proposal is to add an uncapped, rules-based fee-for-service stream for eligible primary medical care.

Not all care. Not all providers doing anything they want. Not all volume without controls.

Eligible medical activity.

Scheduled contribution rates.

Scope-based provider eligibility.

Clinical necessity rules.

Documentation.

Audit.

Co-payment protections.

Place-based accountability.

This would sit beside capitation, not replace it.

In other words, the current reform pathway may be the foundation. The hybrid model is the next layer.

The diagram below compares the current reform pathway with the hybrid architecture I am suggesting. It is not a statistical forecast. It is a structured way of thinking about where each architecture is strong and weak.

Figure 6. Current reform versus the proposed hybrid. Use this as a comparator map, not as a forecast.

A useful caution: The current reform pathway is the serious comparator. It may work better than sceptics expect. The test is whether it changes access, closed books, co-payment barriers and upstream-to-hospital flow.

The current reform is strongest on allocation fairness, data, targets and some urgent-care access. The hybrid model is stronger where the current pathway may remain weaker: marginal supply, provider-scope flexibility, whole-population accountability and uncapped eligible activity.

So the critique is not “the Government has done nothing”.

The critique is:

The Government has started to change the system. But it may still be managing upstream activity too tightly, while hospital demand remains the pressure valve.

That is the game we need to test.

Why this matters for criticism

If we ignore the current reform agenda, the critique becomes weak. It sounds like we are arguing against a system that no longer exists. That is why the current reform pathway should be the real comparator.

What would change my mind?

I would be less convinced if the current reform pathway proves it can expand safe upstream supply, not only measure access or redistribute existing capitation funding.

---

Deep dive (optional, not required reading): I’ve kept the fuller explanation, game table, modelling notes and full source list in the [appendix for this post](../appendices-v1.7.2/appendix-05-the-current-reform-pathway-stronger-than-a-straw-man-but-maybe-still-incomplete-v1.7.2.md).

Note: This series is exploratory policy analysis. It is not a party-political argument, not a position sponsored by an external body, not a claim that any single funding model is perfect, and not a calibrated prediction of savings. The central question is whether New Zealand's current funding architecture lets lower-cost upstream care expand safely before need becomes hospital demand.

Useful links

• Ministry of Health: capitation reweighting

• Cabinet material: Primary Health Care Funding Improvements

• Ministry of Health: primary care health target

• Health New Zealand: National Primary Care Dataset and new primary care health target

• Ministry of Health: PHO finances briefing

---

Public companion links

• Interactive Streamlit dashboard and model lab

• GitHub Pages report, reading map and release model card

v1.8.1 model update

The current Streamlit model release is v1.8.1. Its public aggregate validation lane is public_aggregate_validated, and its claim level is empirically_supported_if_gated for registered gates only.

The current reform pathway remains a serious comparator, not a straw man. The v1.8.1 validation work improves the comparison surface but does not turn any alternative package into an implementation recommendation.

Claim boundary: Do not claim the model proves current reform is wrong or that an alternative will outperform it in practice. The update supports aggregate policy learning only.

When a person cannot get a primary care appointment, the system often sees it much later.

That difference matters.

This post is about the visibility game: hospital pressure becomes urgent because it is counted, public and politically exposed. Primary care failure can be slower, quieter and scattered across thousands of missed appointments, delayed follow-ups, closed books and unaffordable co-payments.

Game theory sounds more dramatic than it needs to be.

For this series, a “game” just means a situation where different players respond to incentives, rules, constraints and each other’s behaviour.

The first game is the hospital rescue game.

Hospitals have a special kind of political visibility. When hospital pressure rises, it is hard to ignore. Emergency departments fill. Waiting lists grow. Media stories appear. Ambulances queue. Elective surgery is delayed. Ministers get briefed. The public notices.

Primary care failure is different.

A person who could not get an appointment is not always counted. A person who paid more than they could afford is not always visible. A person who gave up and waited is not always in the dataset. A practice that quietly closed its books may be a local crisis but not a national headline.

So hospital demand is a visible crisis, while upstream unmet need is often invisible until it becomes hospital demand.

That creates a repeated game.

• Round one: primary care access is constrained.

• Round two: some unmet need flows to urgent care, ambulance and emergency departments.

• Round three: hospital pressure becomes visible.

• Round four: funding and management attention go to the hospital pressure.

• Round five: primary care remains constrained.

Then the cycle repeats.

This is not about hospital managers being villains. It is about salience. The hospital problem is immediate, measurable and politically urgent.

The second game is the Health New Zealand allocation game.

New Zealand now has formal separate appropriations for hospital/specialist services and primary/community/public/population health services. That matters. It means the old simple claim- that primary care is just residual inside a hospital budget- is too crude.

But separate appropriations do not eliminate the game.

Health New Zealand still operates in a world of workforce shortages, hospital deficits, urgent public expectations, ministerial targets, capital pressures, and competing service demands. It must manage hospital delivery while also commissioning primary and community services.

So the allocation game is not simply “money can be moved anywhere”. It is more subtle.

The question is:

Which pressures become visible, urgent and fundable?

Hospital pressure usually wins that contest.

That is why upstream access needs top-tier visibility. Primary care and ambulance outcomes should not be treated merely as inputs to hospital performance. They should be visible system performance domains in their own right.

A primary care access target is a start. But the target should be interpreted carefully. If it only measures appointment timing within general practice, it may miss urgent care, ambulance alternatives, digital care, rural in-person access, co-payment burden and closed books.

The hospital rescue game can only be changed if upstream access is measured and funded in a way that matters.

That means:

• primary care access measures;

• urgent-care access measures;

• ambulance non-conveyance and safe alternative-disposition measures;

• co-payment burden measures;

• closed-books measures;

• rural in-person access measures;

• continuity measures;

• avoidable hospitalisation measures;

• lower-acuity emergency department flow measures;

• patient experience by deprivation, ethnicity, rurality and disability.

Once upstream failure is visible, it becomes harder to ignore.

But visibility is still not enough.

The system also needs a funding pathway that allows upstream supply to grow.

That is where the uncapped eligible primary medical fee-for-service stream fits. It gives the system a way to fund clinically useful activity before it becomes hospital demand.

The goal is not to starve hospitals.

Hospitals need funding. Hospitals are essential.

The goal is to stop using hospitals as the default growth pathway for needs that could have been met earlier, cheaper and closer to home.

The subtle version of the hospital game

It is tempting to describe this as hospitals stealing money from primary care. That is too crude. Hospitals are not the villain. Hospital managers are not sitting around trying to undermine general practice.

The problem is structural. Hospital pressure is immediate. If elective surgery slows, people notice. If emergency department waiting times blow out, people notice. If beds are full, the system has no choice but to respond.

Primary care pressure is different. It is distributed across thousands of small moments: the appointment not available, the follow-up delayed, the fee that is too high, the practice that closes its books, the rural clinic that reduces sessions.

Those moments are real, but they are less visible. By the time they become visible, they often appear as hospital pressure.

That is the hospital game. It is not about bad intent. It is about what the system can see, what it is forced to rescue, and what it can postpone.

Why this belongs in the main post, not just the appendix

The appendix keeps the longer game table, source list and modelling notes. The public argument is shorter: if upstream failure is less visible than hospital failure, the system will keep rescuing hospitals after the pressure has already arrived.

That is why the next policy question is not only "how much money is in primary care?" It is also "what does the system see early enough to fund, manage and fix?"

What would change my mind?

I would be less convinced if Health New Zealand's internal incentives gave primary and community care the same operational salience as emergency departments, hospitals, waiting lists and deficits.

I would also be less convinced if the primary care access target, National Primary Care Dataset, urgent-care expansion and ambulance performance measures made upstream access failure visible early enough to shift funding and operational decisions before hospital pressure rose.

---

Deep dive (optional, not required reading): I’ve kept the fuller explanation, game table, modelling notes and full source list in the [appendix for this post](../appendices-v1.6.0/appendix-07-the-hospital-salience-game-and-the-health-new-zealand-allocation-game-v1.6.0.md).

Useful links

• Ministry of Health: Health Crown entities and Health New Zealand roles

• Treasury: Vote Health 2025/26 Estimates

• Cabinet material: Primary Health Care Funding Improvements

• Health New Zealand: National Primary Care Dataset and new primary care health target

• Ministry of Health: primary care health target

---

v1.8.1 model update

The current Streamlit model release is v1.8.1. Its public aggregate validation lane is public_aggregate_validated, and its claim level is empirically_supported_if_gated for registered gates only.

Hospital salience remains central, but the validation status is deliberately narrow. The v1.8.1 model can support aggregate plausibility about upstream architecture; it cannot claim measured ED or hospital-demand reduction.

Claim boundary: No ED reduction, hospital-demand reduction, or fiscal-savings claim should be added from v1.8.1. The release remains not valid for those impact claims.

But formulas do not remove politics. They often concentrate it. Their complexity can obfuscate understanding.

New Zealand has seen this before with the Population-Based Funding Formula, which was used to distribute District Health Board funding. A New Zealand Medical Journal article analysed 487 newspaper articles about that formula between 2003 and 2016. The formula became a public flashpoint, especially in the South Island. A central theme was dissatisfaction with allocations and concern about transparency.

That should not surprise anyone.

Health officials, perhaps unwisely, responded to emphasise the formula’s technical legitimacy and chide Canterbury District Health Board leaders, for being “special", rather than listen and respond to the actual concerns.

Figure 4. A better allocation formula can distribute a fixed pool more fairly. It does not, by itself, remove the supply constraint.

To clarify: Formula reweighting is necessary. The argument is not that formulas do not matter. It is that formulas alone cannot settle questions of supply, accountability, gaming, and unmet need. Nor are they effective at quelling community sentiment.

A funding formula is a way of deciding shares. Once shares are at stake, everyone has a reason to argue that the formula misses something important.

Rurality. Deprivation. Age. Ethnicity. Unmet need. Complexity. Diseconomies of scale. Transport. Workforce costs. Growth. Decline. Fixed infrastructure. Historical underfunding. Future demand.

All of those things matter.

But the more variables you add, the more the debate becomes a contest about weights. One group says deprivation is underweighted. Another says rurality is underweighted. Another says age is underweighted. Another says historical utilisation bakes in past access failure. Another says the model punishes efficient providers. Another says it rewards providers who generate activity. Then an opaque actuarial process is undertaken that in reality few decision-makers, let alone consumers and clinicians, understand.

That does not mean formulas are useless. They are necessary. If public money is being allocated across populations, there must be some logic to the allocation.

But a formula can only answer one kind of question:

How should a funding pool be distributed?

It cannot fully answer a different question:

Should the pool itself be capped in a way that suppresses clinically useful activity?

That is the distinction I think matters in primary care. I've only come across one New Zealand politician who has asked that question: then Minister for Health, Simon Power.

The current capitation reweighting work is sensible. The Ministry of Health says the old formula was based on how people used general practice in the late 1990s. Since then New Zealand has changed: more long-term conditions, more multimorbidity, more treatment options, more complexity managed in the community, and different rural and deprivation patterns. A ~30 year delay in revision does itself say something about the machinery of government in NZ.

So yes, the formula should change.

But reweighting capitation does not solve the marginal-supply problem by itself. Capitation’s dominance in our funding system, perpetuates the marginal supply problem.

It can make funding distribution fairer across practices. It can move more funding toward practices with higher-need enrolled populations. It can reduce some inequity. Those are good things.

But if the overall architecture remains heavily capped, the next appointment may still be weakly funded.

This is why I worry about “missing the wood from the trees”:

The architecture should include:

• capitation for continuity and population accountability;

• uncapped scheduled fee-for-service for eligible primary medical contacts;

• targeted funding for priority programmes;

• place-based accountability to prevent cherry-picking;

• co-payment protections;

• transparent data;

• urgent-care and ambulance integration;

• audit and clinical governance.

That is more complicated than a formula. But the system is complicated.

The danger is that we spend years arguing over capitation weights while the real supply constraint remains intact.

A better formula may distribute scarcity more fairly. It will not remove the scarcity.

The trap in formula politics

In plain English: a formula can divide money, but it cannot remove incentives. Once the money is divided, organisations still respond to pressure, visibility and risk.

Formula fights are seductive because they look technical. Everyone can point to a variable. Age. Deprivation. Rurality. Ethnicity. Multimorbidity. Workforce cost. Practice size. Travel time.

All of those variables matter. But the deeper problem is that no formula can carry all the political expectations placed on it. If the total envelope is fixed, every added weight creates a redistribution. Someone gains. Someone loses. The debate then becomes a fight over the denominator, the coefficients and the evidence base.

What would change my mind?

I would be less convinced if capitation reweighting alone materially improved access, reduced closed books, protected rural in-person care and reduced avoidable hospital demand. That is testable: the data exists, accessing it and using it is another story altogether.

Deep dive (optional, not required reading): I’ve included the fuller explanation, modelling notes and source list in the appendix below.

Note: This series is exploratory policy analysis. It is not a party-political argument, not a position sponsored by an external body, not a claim that any single funding model is perfect, and not a calibrated prediction of savings. The central question is whether New Zealand's current funding architecture lets lower-cost upstream care expand safely before need becomes hospital demand.

Useful links

• MoH capitation reweighting

• Primary-care Cabinet material

• New Zealand Medical Journal: media content

• New Zealand Medical Journal: Population-Based Funding

• Ministry of Health: PHO finances briefing

Deep dive appendix for Post 04: Why formulas do not solve games

This appendix is supporting material for the public post. It carries the longer explanation, sources and assumptions for readers who want the detail.

The game underneath the policy

Every post in this series is built around a game. A game is simply a situation where each player responds to the rules and to what the other players do.

Table summary: Player | What they are trying to avoid | What they may do under pressure

• Patients: What they are trying to avoid: Delay, cost, uncertainty, worsening illness; What they may do under pressure: Wait, pay, delay, use telehealth, call ambulance, go to hospital

• Providers: What they are trying to avoid: Unfunded work, burnout, financial risk; What they may do under pressure: Close books, shorten appointments, raise fees, limit extra activity

• Health New Zealand: What they are trying to avoid: Visible failure, deficits, hospital pressure; What they may do under pressure: Prioritise urgent hospital pressures

• Primary Health Organisations or locality bodies: What they are trying to avoid: Loss of role, loss of funding, accountability risk; What they may do under pressure: Defend functions, manage pass-through, shape provider incentives

• Accident Compensation Corporation: What they are trying to avoid: Uncontrolled claims cost, poor outcomes; What they may do under pressure: Tighten payment rules or shift toward commissioning

• Ministers: What they are trying to avoid: Publicly visible service failure; What they may do under pressure: Fund the pressure people can see

This is why an apparently technical funding issue becomes a political economy issue very quickly.

How this fits the hybrid model

The hybrid model has five parts:

• capitation for continuity and population responsibility;

• uncapped scheduled fee-for-service for eligible primary medical activity;

• place-based accountability so providers cannot simply cherry-pick easy activity;

• scope-enabled supply so safe care can be generated by the right provider, not only the traditional provider;

• data, audit and top-tier key performance indicators so the system can see access failure before it becomes hospital pressure.

The model is deliberately not a blank cheque. The point is to remove the global cap on eligible primary medical activity, while keeping item prices, clinical eligibility, provider scope, documentation, audit, co-payment protections and place accountability.

What this adds to the modelling

In the demonstrative model, this post corresponds to one or more component games. The model asks what happens if the system stays in the current equilibrium, and what happens if the policy architecture shifts the equilibrium.

The model does not claim, yet, that the preferred architecture will reduce emergency department presentations by a precise number. That would require linked data, calibration and validation. What the model does show is the logic of the mechanism and the assumptions that need to be tested.

The most important empirical tests are:

1. whether scheduled activity payments increase safe primary care supply;

2. whether unmet primary care need flows into urgent care, ambulance and hospitals;

3. whether Accident Compensation Corporation activity payments help sustain local primary care capacity;

4. whether Primary Health Organisation payment arrangements create material pass-through, transparency or entry barriers;

5. whether scope-enabled providers can expand supply safely and equitably.

Sources and further reading

• MoH capitation reweighting

• Primary-care Cabinet material

• New Zealand Medical Journal: media content

• New Zealand Medical Journal: Population-Based Funding

• Ministry of Health: PHO finances briefing

• Health system review

• meeting with General Practice New Zealand,

• Vote Health 2025/26

• National Primary Care Dataset

• Primary-care health target

• NZ Health Survey update

• Cochrane payment review

Appendix note: Supporting material for readers who want the longer explanation, sources and assumptions.

Positionality note: I am not a historian by primary training, but as a paediatrician I work in the long shadow of earlier disability systems, and as an economist I am interested in how institutions make some people visible and others invisible. This post is written with that humility in mind: carefully, and without pretending historical categories map neatly onto today’s autism language.

One reason the autism increase feels mysterious is that the past looks emptier than it was. We search old classrooms for today’s diagnostic categories and cannot find them. So we assume the people were not there.

But diagnostic categories are historical objects. They appear, change, merge, split, migrate and disappear. People do not wait for the category before they exist. They live first. Societies name them later.

This is one of the reasons I am careful with historical claims. The temptation is to rummage through old institutions and retrospectively diagnose everyone. That would be bad history and bad medicine. But the opposite mistake is just as serious: to treat the absence of a modern label as proof that the underlying people were not there.

This is where the language needs to slow down. It would be wrong to say that everyone historically labelled “mentally deficient,” “feeble-minded,” “childhood psychotic,” “emotionally disturbed,” “slow,” “eccentric” or “difficult” was autistic. Many were not. Some had intellectual disability, trauma, epilepsy, cerebral palsy, hearing impairment, poverty, institutional deprivation, mental illness, language disorders, or no disability at all. Historical diagnostic labels were often crude, moralised and coercive.

But it would also be wrong to imagine that autistic people only appeared when autism appeared in diagnostic manuals.

The history of disability in Australia and Aotearoa New Zealand shows how many people with developmental and intellectual disability were hidden, segregated, mislabelled or institutionalised. The Disability Royal Commission’s historical report describes the use of asylums, hospitals and institutions for people with disability in Australia in the nineteenth and twentieth centuries, often because disability was misunderstood and people were deemed unfit to live in the community (Disability Royal Commission history report). In Aotearoa New Zealand, current resources such as the Aotearoa New Zealand Autism Guideline: He Waka Huia Takiwātanga Rau and KidsHealth’s autism-takiwātanga information sit in a very different era of language and rights. That is not autism history in a narrow sense. It is disability history. But autism history cannot be separated from disability history, because many autistic people would have been understood through broader disability categories.

Find & Connect’s entry on “mental deficiency” is particularly sobering. It describes a term used in the first half of the twentieth century for intellectual or developmental disability, treated as a disease and shaped by eugenic ideas about segregation and sterilisation (Find & Connect, mental deficiency). It also warns that people labelled “mentally deficient” were not necessarily intellectually disabled. Some were classified because of poverty, institutional upbringing, behaviour, moral judgement or social nonconformity.

That is the first lesson: the old categories were not simply less precise medical versions of our categories. They were administrative and moral tools.

The second lesson is that invisibility has many forms. Some people were literally hidden in institutions. Some were hidden in families through shame and secrecy. Some were hidden in special schools. Some were hidden by other diagnoses. Some were hidden because their autism was interpreted as naughtiness, bad parenting, laziness, anxiety, psychosis, obstinacy, eccentricity, or lack of discipline. Some were hidden because they were girls who copied their peers, collapsed at home, and held themselves together in public.

This matters for the vaccine debate because the claim “there were no autistic people before widespread vaccination” depends on a fantasy of perfect historical visibility. There was no such visibility. There was no universal developmental surveillance. There were no autism teams in every region. There was no National Disability Insurance Scheme. There were no parent Facebook groups, no autism language in schools, no sensory criteria in DSM-5, no adult diagnosis clinics, no wide recognition of masking.

Even Kanner and Asperger, whose 1940s descriptions now sit near the beginning of formal autism history, did not discover an entirely new kind of person. They described children who were visible to them through particular clinics, at a particular time, with particular theories. After them, autism did not become a stable category. It moved through childhood schizophrenia, infantile autism, pervasive developmental disorders, Asperger’s disorder, PDD-NOS, autism spectrum disorder, and modern neurodevelopmental frameworks.

The moral history matters too. Eugenics was not specifically aimed at “autism” as a modern category. But eugenic policies and attitudes affected people with intellectual and developmental disability, neurodevelopmental difference, mental illness, poverty and socially unacceptable behaviour. Find & Connect describes eugenics as ableist, classist and racist, influencing policy toward disabled people, especially people with intellectual disability (Find & Connect, eugenics). Those histories shaped what families disclosed, what institutions recorded, and who became legible as a citizen with rights rather than a problem to be managed.

The story is not only dark. Deinstitutionalisation, disability rights, inclusive education, self-advocacy, and neurodiversity have changed what can be said. Australia’s National Autism Strategy 2025–2031 reflects a very different moral world: one that uses identity-first language, rejects simplistic functioning labels, and frames autism within rights, inclusion and self-determination. New Zealand’s Health NZ adult autism information and national guideline likewise treat autism as a lifelong neurodevelopmental difference rather than a childhood moral failing. That does not mean either system is good enough. It means more people are now visible under language that can support dignity rather than concealment.

When older Australians or New Zealanders say they did not know autistic children at school, they may be remembering accurately. They may not have known them as autistic. The child may have been in a special school. Or at home. Or in an institution. Or called naughty. Or called odd. Or called intellectually disabled. Or called anxious. Or never called anything because the family knew that naming difference made life worse.

Aotearoa, takiwātanga and Pacific language

One reason Aotearoa New Zealand belongs in this essay is that it makes the language question visible. In many New Zealand health, education and disability resources, autism is now paired with the Māori word takiwātanga, often explained through the phrase “in his, her or my own time and space” (KidsHealth). The word is used in the national autism guideline and in Māori-led work with whānau and tamariki takiwātanga.

That is a generous and mana-enhancing word. It resists the idea that autism is only deficit, disorder or failure. But it should be handled carefully. Takiwātanga is not proof that pre-colonial Māori society had a clinical category identical to modern autism spectrum diagnosis. Nor should it be flattened into a decorative translation for an English medical term. It is better understood as a contemporary kupu that can hold autistic experience in a Māori frame: time, space, personhood, whānau, dignity and difference.

That distinction matters. Autism, as used in diagnosis, is a clinical category with particular histories, thresholds and service consequences. Takiwātanga may overlap with that category, and many people use it that way. But it may also carry cultural, relational and spiritual meanings that are not exhausted by the diagnostic label. The humane stance is to let the words sit near each other without pretending they are always the same thing. A culturally resonant word can improve recognition and dignity, but it does not remove the need for careful clinical assessment, support planning and service access.

There is also a data point here. A New Zealand study, Autism spectrum disorder/Takiwātanga: An Integrated Data Infrastructure-based approach to autism spectrum disorder research in New Zealand, used linked administrative data to explore how autism could be identified in national datasets. Administrative data can show who touched a system; it cannot show everyone who existed outside it. That caution reinforces the broader argument: what a system can count depends on its language, services, coding habits and reasons to notice.

Something similar is happening in Pacific communities, but the language is not settled. I am relying here on published Pasifika voices, not claiming authority over Sāmoan terminology. A RNZ Pacific report on Pasifika autism advocacy describes concern that some existing Sāmoan wording for autism can feel negative or shadowed by stigma, and notes the hope that Pacific linguists, elders, churches and families will develop better words. A Samoa Observer/Pacific Media Network story similarly describes the phrase sometimes used for autism, “vaivai o la’ua mafaufau”, as carrying a meaning about weakness or struggle of mind, while many families prefer the more affirming feel of takiwātanga.

There are also broader strengths-based Pacific disability terms. For example, Tagata sa’ilimalo is used for Sāmoan disabled people and is often glossed around seeking success or victory. That is close in spirit to the point of this essay, but I would not call it the Sāmoan word for autism. It is broader than autism. The careful point is not that Māori, Sāmoan and English words all map neatly onto one another. They do not. The point is that language can either hide people, shame them, or make room for them to be recognised with dignity.

The same principle applies to older records: the word available at the time shaped who could be seen.

The question “where were autistic people?” therefore has several answers.

They were in institutions, sometimes under other labels. They were in mainstream classrooms, sometimes punished for traits we now recognise. They were in families, sometimes protected, sometimes blamed, sometimes hidden. They were in disability datasets under different categories. They were in psychiatric systems. They were in jobs that tolerated routine, precision and solitude. They were in churches, farms, workshops, kitchens, libraries, laboratories, train stations and bedrooms. They were also absent from the data because the data had no reason to see them.

As a paediatrician, I want the diagnosis to help a child be understood. As an economist, I know that systems count people when counting has a purpose. As a citizen, I think we owe the past more humility.

The past did not necessarily have fewer autistic people simply because it had fewer autism diagnoses. It had fewer autism diagnoses because it had different categories, different institutions, different stigma, different incentives, and different ways of making people disappear.

The risk of re-diagnosing the dead

There is a tempting but hazardous move in this history: to look backwards and declare that particular people in older institutions, schools or families were “really autistic.” Sometimes that may be true. Often we cannot know. The diagnostic category did not exist in the same way, the records are incomplete, and the people themselves cannot tell us how they understood their lives.

A more honest historical claim is narrower and stronger. We can say that many people with developmental, communication, sensory, learning and social differences were classified under other systems. We can say that intellectual and developmental disability were often treated through segregation, institutionalisation and stigma. We can say that older categories such as “mental deficiency” were entangled with eugenic thinking and social control. We can say that some people now recognisable as Autistic would likely have been hidden inside those older categories. We do not have to pretend we can sort the past with a modern DSM checklist.

This matters because the series should not use history merely as a rhetorical weapon against vaccine claims. It should use history to restore humility. The absence of a diagnosis in an archive is not proof of the absence of a person. But neither is every unusual child in the archive ours to claim. The better line is that modern autism diagnosis emerged from a long history of changing disability categories, and those categories shaped who was visible, who was blamed, who was helped, and who was removed from ordinary life.

That history also explains why diagnosis can be emotionally complicated. For some families, a diagnosis is a door to help. For some Autistic adults, diagnosis is a language of identity. For many disabled people historically, diagnosis was a tool of exclusion. A humane account has to hold all of that at once.

The institutional story is not just an old cruelty we can congratulate ourselves for escaping. It is a reminder that classification has consequences. A label can open a door, close a door, or decide whether someone is imagined as educable, employable, lovable or dangerous. Modern autism diagnosis can be a route to understanding, but only because we have slowly, incompletely and unevenly changed the social meaning of developmental difference.

This is also why the history post should sit near the beginning of the series. Before readers are asked to think about DSM-5, NDIS or vaccines, they need to see that the baseline was never clean. Older societies did not have a hidden, well-measured autism rate waiting in a spreadsheet. They had families, schools, institutions, asylums, special classes, diagnostic habits and social shame. Some people were counted under categories we would now avoid. Some were deliberately kept out of sight. Some were present in ordinary life but never made legible to medicine.

That history changes the vaccine debate. If we do not know how many Autistic people were hidden by older systems, we should be careful before claiming that the modern rise in diagnosis requires a modern biological insult.

The historical caution

The most important sentence in this essay is a cautious one: we cannot simply re-diagnose the past. It would be wrong to say that everyone institutionalised under older disability categories was Autistic, or that every person described as eccentric, withdrawn, intellectually disabled, psychotic or behaviourally disturbed would meet modern autism criteria. Historical humility matters.

But caution cuts both ways. We also cannot treat the absence of an autism label as proof that autism was absent. Older societies had different diagnostic languages, different institutions, different schooling arrangements, different family expectations and different levels of stigma. Some people were hidden because disability was shameful. Some were hidden because services were segregated. Some were hidden because no one had a name that would help rather than harm. The ethical task is not to claim them all, but to notice how easily people disappear when a society lacks the language and the will to see them.

One of the crueller tricks of history is that invisibility can later be mistaken for absence. It is easier to imagine there was less autism in the past than to reckon with how many people were hidden, misnamed, institutionalised, or simply never given words that fit. I think we owe those people, and their families, more honesty than that.

This essay is part of [Counting Autism: diagnosis, data, incentives and dignity](https://rareinsights.substack.com/p/counting-autism-series-landing-page), a Rare Insights series on autism diagnosis, causation, incentives and dignity, written from my dual position as a paediatrician and economist. The linked [series guide](https://rareinsights.substack.com/p/series-guide-counting-autism) keeps the full reading order in one place, and you can follow the broader work at [Rare Insights](https://rareinsights.substack.com/).

Further sources

• Find & Connect: Mental Deficiency — Historical category and cautions.

• Find & Connect: Eugenics — Australian eugenics context.

• Disability Royal Commission historical report — Disability history and institutionalisation.

• Hansen et al. 2015, JAMA Pediatrics — Diagnostic/reporting shifts as a prevalence mechanism.

• National Autism Strategy 2025–2031 — Current Australian language and inclusion frame.

• Aotearoa New Zealand Autism Guideline: He Waka Huia Takiwātanga Rau — Current New Zealand autism guideline.

• KidsHealth: Autism-takiwātanga in children — New Zealand child and whānau-facing explanation.

• Health NZ: Autism in adults — New Zealand adult autism information.

• Bowden et al. 2020: Autism spectrum disorder/Takiwātanga and the Integrated Data Infrastructure — New Zealand administrative-data approach to autism research.

• Altogether Autism: Takiwātanga — Explanation of takiwātanga and its mana-enhancing intent.

• RNZ Pacific: Pasifika autism group a “vehicle for change” — Pasifika advocacy and language concerns.

• Samoa Observer/Pacific Media Network: Pacific children with autism and their families — Pasifika autism stigma and language.

• VisAble glossary: Tagata sa’ilimalo — Strengths-based Sāmoan disability terminology.

I do not mean every provider should be able to bill anything, for anything, at any price.

I do not mean abolishing capitation.

I do not mean removing clinical governance.

Figure 7. Uncapped does not mean uncontrolled. The proposal is uncapped volume only inside rules, audit and clinical governance.

A useful caution: The weak-control version of uncapped fee-for-service is explicitly not the proposal. The proposal is scheduled, rules-based, audited, clinically governed, equity-protected and place-accountable.

I do not mean ignoring equity. In fact, I don’t think capitation can or does produce equitable outcomes. At best, it could be considered an equal-ish input.

I mean something narrower and more useful:

Eligible primary medical activity should not be limited by a hard global funding envelope. It should be demand-led within rules.

That is an important difference.

A capped envelope says: here is the total pool; once the pool is exhausted, the system must ration.

A rules-based benefit says: if the service is eligible, the provider is qualified, the patient meets criteria, the documentation is adequate and the claim is not suspicious, the public contribution flows.

That is closer to how Accident Compensation Corporation treatment funding works.

Accident Compensation Corporation does not pay for everything. It pays or contributes to injury treatment when the treatment is clinically appropriate, delivered by an appropriately qualified provider, documented, necessary, and within the relevant regulation or contract.

There are rules. There are item codes. There are contribution rates. There are contracts. Some services require pre-approval. There are expectations about quality and the number of treatments needed.

This governance, does not require an intermediary. So it reduces the cost of market entry, reduces the cost of administration to practices, and transactional cost to treasury.

That is why the Accident Compensation Corporation analogy is useful, albeit their approach could also be optimized. Fortunately they have had one of New Zealand’s most senior health funders recently join their executive.

It shows that activity-sensitive funding can be controlled without using a fixed global cap as the main rationing tool. This isn’t novel, it’s done in lots of places (including other parts of NZ). It also solves the measurement and data access problem, because the funding scheme would be government-administered, which would enable measurement and remove the current barriers to access inherent to having PHOs control access to system performance data. This is billing data, not clinical data.

For primary medical care, an uncapped benefit stream could start with specific contact types:

• same-day or next-day urgent primary medical assessment;

• complex consultations requiring longer time;

• rural in-person assessment;

• minor procedures that prevent escalation;

• follow-up after ambulance non-conveyance;

• care after emergency department discharge;

• clinically necessary reviews for frailty or multimorbidity;

• defined nurse practitioner, pharmacist, paramedic, physiotherapist or general practitioner services within scope.

The public contribution would be scheduled. The patient might pay a co-payment, depending on policy settings. For high-need groups, children, Community Services Card holders, rural patients or priority services, the co-payment could be reduced or removed. It also means that some of the odd funding rules that currently exist, such as a minimum number of certain practitioner-types, can be removed.

The key point is that the total number of eligible services would not be fixed in advance.

If patients need care, and providers can safely deliver it, the system should not suppress that activity just because the envelope has been capped.

The controls should be smarter:

• item definitions;

• clinical eligibility;

• scope of practice;

• documentation;

• audit;

• unusual-billing detection;

• outcome monitoring;

• co-payment caps;

• locality obligations;

• minimum service coverage;

• reporting against access and equity.

This is what I mean by “uncapped does not mean uncontrolled”.

Figure 8. Hybrid architecture. The design uses multiple controls because no single lever solves access, equity, fiscal and gaming risk at once.

Why does this matter?

Because if the main control is a hard cap, the system often controls the budget by suppressing care upstream.

That does not mean people stop needing care.

It means they wait. Or pay. Or go elsewhere. Or deteriorate. Or call an ambulance. Or end up in hospital.

The hospital system is then funded to deal with the pressure because hospital pressure is more visible and less avoidable.

That is not good economics. It is delayed spending at a higher cost. Overplaying one’s hand in one part of the system, simply because control is easier, not because it makes any sense to.

A better model would let eligible primary medical activity grow safely.

It would still keep capitation. Capitation is important for continuity and population accountability. But capitation cannot prevent the unpreventable. It cannot manage issues such as social determinants. It would still keep place-based responsibility. Otherwise providers could cherry-pick easy work and leave hard-to-reach populations behind.

The proposal is a hybrid:

• capitation for having responsibility;

• fee-for-service for doing eligible medical work;

• place-based accountability for reaching everyone;

• data and audit for controlling gaming;

• co-payment protections for equity.

That is not neoliberal. It is not a blank cheque. It is not an anti-public model.

It is a way of paying for the care we want to happen before the hospital becomes the only place left to go.

What would still be controlled?

In plain English: uncapping primary care funding does not mean uncontrolled spending. It means clear rules for when extra upstream care is funded, measured and audited before pressure becomes hospital demand.

This is the part people often miss. If eligible activity is uncapped, almost everything else still needs rules.

The item price is scheduled. The provider must be eligible. The service must match a defined contact type. The provider must act within legal scope. The record must show clinical need. Repeated patterns can be audited. Co-payment protections can be applied. Some services can require pre-approval or additional documentation.

That is why Accident Compensation Corporation is such a useful analogy. It does not mean every provider can bill anything they want forever. It means there is a rules-based stream where eligible injury-related treatment can generate payment.

What would change my mind?

I would be less convinced if an uncapped scheduled stream could not be governed through item rules, scope, audit, co-payment protections and place accountability. The weak-control version is not the proposal.

Deep dive (optional, not required reading): I’ve included the fuller explanation, modelling notes and source list in the appendix below.

Note: This series is exploratory policy analysis. It is not a party-political argument, not a position sponsored by an external body, not a claim that any single funding model is perfect, and not a calibrated prediction of savings. The central question is whether New Zealand's current funding architecture lets lower-cost upstream care expand safely before need becomes hospital demand.

Useful links

• MoH capitation reweighting

• ACC treatment payments

• Ministry of Business, Innovation and Employment:

• Cochrane payment review

• RACGP funding models

Deep dive appendix for Post 06: What I mean by uncapping primary care funding

This appendix is supporting material for the public post. It carries the longer explanation, sources and assumptions for readers who want the detail.

The political misunderstanding

This proposal can easily be misunderstood as a market free-for-all. That is not what I mean.

The public system would still decide what services are eligible, how much it contributes, who can claim, what records are required, which groups are protected from excessive fees, and what patterns trigger review.

The difference is that the system would stop rationing useful medical primary care mainly through a hard total cap. It would ration through rules, clinical need, scheduled prices, audit, scope and patient protections. That is a more transparent form of control.

The game underneath the policy

Every post in this series is built around a game. A game is simply a situation where each player responds to the rules and to what the other players do.

Table summary: Player | What they are trying to avoid | What they may do under pressure

• Patients: What they are trying to avoid: Delay, cost, uncertainty, worsening illness; What they may do under pressure: Wait, pay, delay, use telehealth, call ambulance, go to hospital

• Providers: What they are trying to avoid: Unfunded work, burnout, financial risk; What they may do under pressure: Close books, shorten appointments, raise fees, limit extra activity

• Health New Zealand: What they are trying to avoid: Visible failure, deficits, hospital pressure; What they may do under pressure: Prioritise urgent hospital pressures

• Primary Health Organisations or locality bodies: What they are trying to avoid: Loss of role, loss of funding, accountability risk; What they may do under pressure: Defend functions, manage pass-through, shape provider incentives

• Accident Compensation Corporation: What they are trying to avoid: Uncontrolled claims cost, poor outcomes; What they may do under pressure: Tighten payment rules or shift toward commissioning

• Ministers: What they are trying to avoid: Publicly visible service failure; What they may do under pressure: Fund the pressure people can see

This is why an apparently technical funding issue becomes a political economy issue very quickly.

How this fits the hybrid model

The hybrid model has five parts:

• capitation for continuity and population responsibility;

• uncapped scheduled fee-for-service for eligible primary medical activity;

• place-based accountability so providers cannot simply cherry-pick easy activity;

• scope-enabled supply so safe care can be generated by the right provider, not only the traditional provider;

• data, audit and top-tier key performance indicators so the system can see access failure before it becomes hospital pressure.

The model is deliberately not a blank cheque. The point is to remove the global cap on eligible primary medical activity, while keeping item prices, clinical eligibility, provider scope, documentation, audit, co-payment protections and place accountability.

What this adds to the modelling

In the demonstrative model, this post corresponds to one or more component games. The model asks what happens if the system stays in the current equilibrium, and what happens if the policy architecture shifts the equilibrium.

The model does not claim, yet, that the preferred architecture will reduce emergency department presentations by a precise number. That would require linked data, calibration and validation. What the model does show is the logic of the mechanism and the assumptions that need to be tested.

The most important empirical tests are:

1. whether scheduled activity payments increase safe primary care supply;

2. whether unmet primary care need flows into urgent care, ambulance and hospitals;

3. whether Accident Compensation Corporation activity payments help sustain local primary care capacity;

4. whether Primary Health Organisation payment arrangements create material pass-through, transparency or entry barriers;

5. whether scope-enabled providers can expand supply safely and equitably.

Sources and further reading

• MoH capitation reweighting

• ACC treatment payments

• Ministry of Business, Innovation and Employment:

• Cochrane payment review

• RACGP funding models

• Primary-care health target

• National Primary Care Dataset

• Vote Health 2025/26

• Primary-care Cabinet material

• Beehive: new and improved urgent and

• Health New Zealand: the Ambulance Team

• Australian incentives review

Appendix note: Supporting material for readers who want the longer explanation, sources and assumptions.

Figure: A child becomes countable only after a chain of recognition, assessment and recording.

Positionality note: As a paediatrician, I know children do not arrive in datasets by magic. They arrive through parents, teachers, referrals, waiting lists and reports. As an economist, I know data are produced by systems with rules and incentives. This post is really about that intersection: diagnostic reality and the architecture of counting.

The question is usually phrased as a memory: When I was at school, there weren’t this many autistic children.

That sentence is doing a lot of work. It is a recollection, but also a claim about measurement. It assumes the children who were present were recognisable to the adults of the time, that their differences were named accurately, and that somebody had a reason to write the name down. Those are large assumptions, and in health policy large assumptions deserve to be made visible.

Perhaps. But before we can answer it, we need to ask a more uncomfortable question: how would we know?

Autism prevalence is not observed from a mountaintop. It is observed through clinics, schools, parents, teachers, diagnostic manuals, surveys, funding schemes and administrative databases. Each of those systems has a field of vision. Each has blind spots. When the blind spots change, the numbers change too.

Mortality is relatively easy to count. Birth weight is measured at a defined event. Vaccination status is recorded against a schedule. Autism is different. It is a neurodevelopmental difference inferred from behaviour, history, context and impairment. It is often identified only after someone worries, refers, assesses, codes and documents. A child does not enter an autism dataset when they become autistic. They enter it when someone recognises, assesses, diagnoses, codes and records them.

That means autism prevalence is not a single thing. We need to separate true prevalence, diagnosed prevalence, administrative prevalence, service-linked prevalence and knowable prevalence.

True prevalence is the underlying number of people who meet a meaningful clinical or neurodevelopmental definition of autism, whether or not they have been recognised. Diagnosed prevalence is the number of people who have received a diagnosis. Administrative prevalence is the number captured in records. Service-linked prevalence is the number visible because they receive supports. Knowable prevalence is what our systems are capable of detecting.

These are not the same.

The CDC’s ADDM Network is valuable because it uses systematic surveillance across selected sites, but it is still record-based. Its 2022 report identified autism among eight-year-old children using health and special education records in 16 sites. The resulting estimate was one in 31 eight-year-olds across those sites, but with wide variation by location and record source (CDC ADDM 2022; CDC autism data summary). The NIMH summary of the same data emphasises a measurement caveat: the figures should not be read as an overall US national prevalence rate, and they cannot be extrapolated to other age cohorts (NIMH ASD statistics).

Australia’s ABS data have different strengths and limits. The 2022 Survey of Disability, Ageing and Carers estimated 290,900 Autistic Australians, but it is a survey-based estimate that depends on reporting, household participation, disability concepts and survey methods (ABS Autism in Australia 2022). NDIS data answer yet another question: who has entered a service and funding system. They are not a population census of autism. The NDIA’s autism dashboard is useful for understanding participants with autism in the Scheme, but the dashboard itself is Scheme data (NDIS Autism dashboard).

New Zealand makes the same point from a different angle. Aotearoa does not have one simple national autism counter. The Aotearoa New Zealand Autism Guideline: Third Edition is a living guideline for support, assessment, diagnosis, education, mental health, Māori perspectives and Pacific perspectives; it is not a prevalence register. Stats NZ’s 2023 Household Disability Survey includes autism among the topics it can describe, but Stats NZ explicitly warns that the 2023 survey is not a valid time series with earlier disability surveys because the way disabled people are identified changed. That is exactly the measurement problem in miniature: better questions may produce better data, but they also break the illusion of a clean historical trend line.

Economists are used to this problem. Data are not just facts; they are facts that a system found worth recording. When incentives change, records change.

A diagnosis may become worth recording because it unlocks school adjustments, therapy, disability support, insurance, family explanation, identity, workplace accommodations, research participation, or simply a better way of understanding a life. In New Zealand, the Ministry of Education’s current learning-support work programme is framed around growing demand, access and nationally consistent support (Ministry of Education learning support, updated 2025). Again, this does not mean diagnoses are fake. It means the pathway to being counted is not random.

The missed people are also not random. Autistic adults who were never assessed as children are less visible in child surveillance systems. Girls and women who mask may be missed in childhood and diagnosed later. Autistic people without intellectual disability may be labelled anxious, obsessive, gifted, oppositional or eccentric. Autistic people with intellectual disability may have their autism hidden behind broader disability categories. Rural families may have less diagnostic access. Culturally and linguistically diverse families may face language, stigma or service barriers. Families who cannot afford assessment, tolerate long waiting lists, or find a clinician who understands masked presentations may never make it into the count.

The Australian National Autism Strategy notes that diagnosis is likely to occur later for girls, women and gender diverse people, people with less overt presentations, and people in regional, rural and remote areas (National Autism Strategy publication). Recent health-system data point in the same direction: a 2024 JAMA Network Open study found the largest relative increases in autism diagnosis rates among young adults, female people and some minority groups in children (Grosvenor et al., 2024). A 2026 Swedish population study in The BMJ adds a powerful warning: male-female diagnosis ratios narrowed with age, suggesting that childhood diagnosis data can miss later recognition in females (Fyfe et al., BMJ 2026). A review of the female autism phenotype and camouflaging explains why some Autistic girls and women may appear more typical externally while paying a high internal cost (Hull, Petrides and Mandy, 2020).

The knowability problem also matters historically. If a person in 1935 was institutionalised under a label such as “mental deficiency,” we cannot simply reclassify them as autistic. That would be bad history. But we also cannot treat the absence of an autism diagnosis as evidence that autism was absent. The Find & Connect history of “mental deficiency” in Australia describes a broad, morally loaded, eugenics-influenced category that swept together people with intellectual disability, developmental difference, poverty, trauma, unconventional behaviour and institutional disadvantage (Find & Connect, mental deficiency). Historical categories were not neutral containers of clinical truth.

The absence of diagnosis may mean absence of autism. It may also mean absence of language, absence of assessment, absence of services, absence of safety, absence of clinician awareness, absence of schooling pathways, or absence of a reason to write it down.

This is why I am cautious about neat claims. “Autism has risen because of DSM-5” is too simple. “Autism has risen because of NDIS” is too simple. “Autism has risen because of vaccines” is not supported by the evidence. The better claim is that recorded autism has risen because the data-generating process has changed: diagnostic categories, clinical recognition, social awareness, school systems, funding systems, stigma, and administrative records have all changed.

The right response to imperfect data is not to give up on evidence. It is to become more honest about what kind of evidence we have.

We can still know a great deal. We can compare surveillance systems. We can study diagnostic substitution. We can examine changes in criteria. We can use cohort studies. We can triangulate ABS, NDIS, school, health and survey data. We can ask who is missing. We can ask whether rising counts are concentrated in certain ages, sexes, severity profiles or service systems. We can distinguish local service-linked patterns from population-level estimates.

But we should stop treating the autism prevalence curve as if it were a thermometer. It is more like a ledger. Ledgers reflect what institutions choose, need and are able to record.

The child is real. The developmental difference is real. The family’s need for support is real. But the number in the prevalence table is not a direct photograph of reality. It is reality filtered through recognition, access, criteria, documentation and incentives.

Missing, but not randomly missing

The people outside the count are not a random sample. This is the part that matters most. If a dataset misses Autistic adults, girls who mask, children in rural areas, families who cannot afford assessment, people whose intellectual disability swallows the autism formulation, or children whose schools do not ask the question, then the dataset does not merely undercount. It undercounts in patterned ways.

That pattern can make historical comparisons misleading. A child who was quiet, academically capable and distressed in 1980 might have disappeared into “anxiety” or “shyness.” A non-speaking child with epilepsy and intellectual disability might have been counted under intellectual disability alone. A disruptive child might have been labelled behaviourally disturbed. A child in a family worried about stigma might never have been assessed at all. Each missing person changes not just the numerator, but our story about what autism looked like.

Economists would call this selection. Epidemiologists would call it ascertainment. Clinicians might call it the waiting room problem: the children we see are shaped by who gets referred, who can attend, who can pay, who waits long enough, and who is considered diagnosable by the local clinician. None of those filters makes autism unreal. They make autism data contingent.

The implication is not that prevalence studies are useless. Quite the opposite. Good surveillance becomes more valuable when we understand the filters. Active case finding, record review, registry linkage, household surveys, school data, health records and NDIS data each see different parts of the picture. The task is triangulation, not surrender.

So when a politician, journalist or campaigner points to a prevalence curve and says, “Look how much autism has increased,” the first response should be gentle but firm: increased in which dataset, by which method, among which children, under which rules, and with what incentive to record the diagnosis?

That is why a good prevalence essay should be modest about the past and demanding about the present. We may never reconstruct a perfect denominator for 1975, but we can build better systems now: active surveillance, linked datasets, culturally safe assessment pathways, public diagnostic services, adult pathways, and data that record support needs without requiring people to perform catastrophe.

The practical test

A useful test for any prevalence claim is to ask: what had to happen before this person entered the dataset? In one system, a child may need a paediatrician, a psychologist, a school report, a Medicare item, a diagnostic code and an NDIS application. In another, the same child may be counted through a special education eligibility category. In another, an adult may appear only if they answer a survey question or seek a late assessment.

Those pathways are not neutral. They depend on money, geography, waiting lists, stigma, clinician confidence and the practical value of the diagnosis. This is why prevalence estimates from different countries, eras and datasets should not be read as if they are measuring autism from the same position. They are measuring autism as seen by a particular system. The system may be seeing more clearly than before, but it is still a system with edges.

I realise this is not the kind of answer people usually want. People like numbers that look like clean facts. But in health policy the most honest sentence is often the least glamorous one: the number is real, and it is also produced by a system. Once you see that, the whole autism-prevalence debate becomes less mystical and more intelligible.

This essay is part of Counting Autism: diagnosis, data, incentives and dignity, a Rare Insights series on autism diagnosis, causation, incentives and dignity, written from my dual position as a paediatrician and economist. The linked series guide keeps the full reading order in one place, and you can follow the broader work at Rare Insights.

Further sources

• CDC ADDM 2022 methods and results — Record-based surveillance example.

• ABS: Autism in Australia 2022 — Survey-based Australian example.

• NDIS Autism dashboard — Service-linked administrative data.

• Aotearoa New Zealand Autism Guideline: Third Edition — New Zealand living guideline for assessment, diagnosis, education and support.

• Stats NZ 2023 Household Disability Survey data collection — New Zealand disability survey context and comparability cautions.

• Ministry of Education learning support work programme — New Zealand service-system context for demand and access.

• Grosvenor et al. 2024, JAMA Network Open — Health-system diagnosis trends across children and adults.

• Hansen et al. 2015, JAMA Pediatrics — Reporting-practice changes and prevalence.

• NIMH ASD statistics — Helpful caution about site-specific surveillance.

“Marginal” means the next one.

The next appointment. The next prescription review. The next same-day urgent slot. The next rural clinic. The next wound dressing. The next complex consultation. The next follow-up after an ambulance crew decides not to take someone to hospital.

The question is simple:

A useful caution: marginal supply is not the whole story. Rural infrastructure, workforce, housing, clinical governance and co-payment protections also matter. But even with those protections, the next appointment still needs a viable funding signal.

Is the next clinically useful contact financially viable for the provider to deliver?

That does not mean clinicians think only about money. They do not. But clinics are not abstract moral machines. They are organisations with finite staff, rooms, phones, software, admin support, rent, indemnity, clinical risk and exhaustion.

If the next contact has a cost, but little or no payment signal, the system will ration it.

That is especially true when workforce is tight.

Under a mostly capitated system, a practice receives a fixed payment per enrolled person. That payment helps support the practice. But after the patient is enrolled, seeing them more often may not bring enough additional revenue to cover the extra time and cost.

Under fee-for-service, the next eligible contact brings a payment. That payment may not fully cover cost either, but it creates a marginal signal. It tells the provider that expanding activity is possible.

The diagram below is deliberately simple. It shows marginal cost rising as a practice takes on more contacts. Early contacts may be easy to absorb. Later ones are harder because the practice needs extra staff, longer hours, more rooms or more administration.

Under weak marginal revenue, supply stops early. Under scheduled fee-for-service, more contacts become viable.

Figure 4. Capitation and scheduled fee-for-service at the margin. The plot shows why extra contacts stop earlier when marginal revenue is weak, and why an activity-sensitive payment can make more contacts viable. This is a stylised plot, not an empirical estimate.

*Figure 4. Capitation and scheduled fee-for-service at the margin. This uses the same simplified decision map to keep the marginal-supply point visually consistent.

This is not an argument that fee-for-service is always good. It is an argument that every system needs some way to pay for the next clinically necessary contact.

If it does not, the rationing still happens. It just happens less transparently.

The system rations by:

• waiting time;

• closed books;

• appointment length;

• phone triage;

• co-payment increases;

• referral thresholds;

• telling patients to use emergency departments;

• shifting work to ambulance or urgent care;

• using telehealth for problems that may still need hands-on examination.

Some of those tools are useful. Triage is useful. Telehealth is useful. Co-payments can play a role. But when they become the main rationing mechanism, the system is in trouble. NZ has had a number of these for quite some time.

A hard funding envelope can look tidy from the centre. The budget is controlled. The line item is stable. The formula is updated. The target is announced. This solves a political economy problem for Government: the appearance of control and positive performance.

But from the patient’s point of view, the question is simpler: can I get care when I need it?

From the provider’s point of view: can we afford to open another slot without burning out staff or bankrupting the practice?

From the hospital’s point of view: why are more people turning up at the emergency department?

Those are all marginal-supply questions. They are also distant enough from central health administrators and politicians, to deny accountability for it.

This is why I think New Zealand should explore an uncapped, scheduled, rules-based primary medical fee-for-service stream. Not uncapped prices. Not uncapped provider behaviour. Uncapped eligible activity.

The public contribution would be scheduled. The service would need to be eligible. The provider would need to be working within scope. Documentation would be required. Patterns of overuse would be audited. Co-payment protections would be needed for children, Community Services Card holders, rural patients, high-need groups and people with complex long-term conditions.

The point is to stop using a capped envelope as the main control. Particularly where it's being used as a control as overreach.

A capped envelope is simple, but it pushes pressure elsewhere.

A capped envelope gives the appearance of control, only if you look at it in isolation.

A rules-based activity stream is more complicated, but it can let supply grow where care is lower cost and earlier.

Microeconomics does not tell us the exact policy answer. But it does tell us what to look for.

If the next clinically useful contact is unfunded, that contact will eventually disappear.

And if enough contacts disappear upstream, they reappear downstream as ambulance calls, urgent care demand and hospital pressure.

Why this matters for rural areas

In plain English: marginal supply asks whether the next appointment is worth creating. A system can look funded on average and still fail at the edge where the next patient is trying to get in.

Marginal supply is especially important in rural areas. A city practice may be able to absorb a bit more work by using a larger team, extending hours or shifting some care to telehealth. A rural service may have fewer staff, fewer rooms, longer travel times, fewer locums and less backup.

And increasingly, that makes working for a telehealth provider, look more and more appealing. Same pay, less fuss. Except that the service/product isn't the same. It extracts extracts local supply, restricts its range, and make it national. Telehealth exchanges a wide, local service for a narrower national service. That sustains a national system at the cost of a local one.

That means the marginal cost of the next in-person clinic can rise very quickly. If the payment signal does not rise with it, the system will drift toward remote-only care or no care at all. For local rural care, it is coverging towards the latter.

What would change my mind?

I would be less convinced if practices could reliably increase timely appointments under fixed capitation alone, without higher co-payments, shorter consultations, hidden rationing, or extra activity-linked funding.

Deep dive (optional, not required reading): I’ve included the fuller explanation, modelling notes and source list in the appendix below.

Note: This series is exploratory policy analysis. It is not a party-political argument, not a position sponsored by an external body, not a claim that any single funding model is perfect, and not a calibrated prediction of savings. The central question is whether New Zealand's current funding architecture lets lower-cost upstream care expand safely before need becomes hospital demand.

Useful links

• MoH capitation reweighting

• ACC treatment payments

• Cochrane payment review

• RACGP funding models

• Primary-care Cabinet material

Deep dive appendix for Post 03: Marginal supply: the tiny economic idea that decides whether appointments exist

This appendix is supporting material for the public post. It carries the longer explanation, sources and assumptions for readers who want the detail.

The game underneath the policy

Every post in this series is built around a game. A game is simply a situation where each player responds to the rules and to what the other players do.

Summary: Player | What they are trying to avoid | What they may do under pressure

• Patients: What they are trying to avoid: Delay, cost, uncertainty, worsening illness; What they may do under pressure: Wait, pay, delay, use telehealth, call ambulance, go to hospital

• Providers: What they are trying to avoid: Unfunded work, burnout, financial risk; What they may do under pressure: Close books, shorten appointments, raise fees, limit extra activity

• Health New Zealand: What they are trying to avoid: Visible failure, deficits, hospital pressure; What they may do under pressure: Prioritise urgent hospital pressures

• Primary Health Organisations or locality bodies: What they are trying to avoid: Loss of role, loss of funding, accountability risk; What they may do under pressure: Defend functions, manage pass-through, shape provider incentives

• Accident Compensation Corporation: What they are trying to avoid: Uncontrolled claims cost, poor outcomes; What they may do under pressure: Tighten payment rules or shift toward commissioning

• Ministers: What they are trying to avoid: Publicly visible service failure; What they may do under pressure: Fund the pressure people can see

This is why an apparently technical funding issue becomes a political economy issue very quickly.

How this fits the hybrid model

The hybrid model has five parts:

• capitation for continuity and population responsibility;

• uncapped scheduled fee-for-service for eligible primary medical activity;

• place-based accountability so providers cannot simply cherry-pick easy activity;

• scope-enabled supply so safe care can be generated by the right provider, not only the traditional provider;

• data, audit and top-tier key performance indicators so the system can see access failure before it becomes hospital pressure.

The model is deliberately not a blank cheque. The point is to remove the global cap on eligible primary medical activity, while keeping item prices, clinical eligibility, provider scope, documentation, audit, co-payment protections and place accountability.

What this adds to the modelling

In the demonstrative model, this post corresponds to one or more component games. The model asks what happens if the system stays in the current equilibrium, and what happens if the policy architecture shifts the equilibrium.

The model does not claim, yet, that the preferred architecture will reduce emergency department presentations by a precise number. That would require linked data, calibration and validation. What the model does show is the logic of the mechanism and the assumptions that need to be tested.

The most important empirical tests are:

1. whether scheduled activity payments increase safe primary care supply;

2. whether unmet primary care need flows into urgent care, ambulance and hospitals;

3. whether Accident Compensation Corporation activity payments help sustain local primary care capacity;

4. whether Primary Health Organisation payment arrangements create material pass-through, transparency or entry barriers;

5. whether scope-enabled providers can expand supply safely and equitably.

Sources and further reading

• MoH capitation reweighting

• ACC treatment payments

• Cochrane payment review

• RACGP funding models

• Primary-care Cabinet material

• National Primary Care Dataset

• Primary-care health target

• Vote Health 2025/26

• Health New Zealand: the Ambulance Team

• Beehive: new and improved urgent and

• Australian incentives review

• NZ Health Survey update

Appendix note: Supporting material for readers who want the longer explanation, sources and assumptions.

The first is fee-for-service. A provider does a service and receives a payment. The service might be a consultation, a procedure, a review, a wound dressing, a vaccination, a minor operation or an urgent assessment.

The plain strength of fee-for-service is that it pays for the next piece of work.

If a clinic sees more patients, it receives more revenue. If a nurse practitioner provides more eligible consultations, those consultations can be funded. If a rural clinician does more urgent-care sessions, there is a payment signal attached to that work.

That can increase supply. It tells the system that doing more work is financially possible.

But fee-for-service has a danger. If it is poorly designed, it can reward volume rather than value. It can make short, simple, repeat contacts more attractive than slow, complex, relationship-based care. It can also encourage providers to focus on services that are easy to bill, rather than services that matter most.

So fee-for-service is useful, but can be problematic.

The second model is capitation. A provider receives a fixed payment for each enrolled person, usually adjusted by patient characteristics such as age, sex, deprivation, rurality and illness burden.

Capitation earns its place for different reasons. It supports continuity, gives a clinic responsibility for a defined population, and can fund preventive or team-based work that does not fit neatly into a single consultation. Funders also like it because the total cost is more predictable.

But capitation has a different danger. Once the patient is enrolled, the next contact is often a cost rather than a revenue event, to the provider. If a person needs more visits, more phone calls, more follow-up, more complexity and more risk management, the practice may receive little additional payment.

That is where rationing can creep in, often without anyone intending it. Time and workforce are finite.

The third model is programme or targeted funding. This is money for specific activities, such as immunisation, long-term condition care, screening, care coordination or access programmes. This is the category that Professor Des Gorman advocated for increased investment in, particularly through the Leveraged Primary Care model pioneered by Kaiser Permanente.

Targeted funding can be useful when government wants to promote particular outcomes. But it can become complicated. It can fragment care into many little funding streams, each with its own forms, rules, audits and reporting. Often targeted funds are introduced without a clear implementation plan, and usually with no exit plan. Typically health services executives are left scrambling to fund these, without Ministry or Minister interest or support. Targeted funding is most useful for high-value interventions like vaccinations or high-cost diseases like diabetes.

That is why blended systems exist.

A blended model uses capitation for baseline responsibility, fee-for-service for eligible activity, and targeted funding for priority programmes. The practical job is not to choose one model as if it were perfect. None of them is.

The payment signal has to match the type of care.

For example:

• continuity and preventive care suit capitation;

• urgent appointments and procedures suit fee-for-service;

• immunisation catch-up or outreach may suit targeted funding;

• rural access may need loading or place-based support;

• complex long-term care may need both capitation and activity-sensitive payments.

In New Zealand, the Government itself describes primary care funding as blended: capitation, co-payments and targeted streams. That is true. But the important question is whether the blend is strong enough at the margin.

The margin is the next appointment.

If the next appointment is clinically needed but weakly funded, the system may still ration access even if the overall model is technically “blended”.

That is why I keep coming back to an Accident Compensation Corporation-style analogy. Accident Compensation Corporation payments are not unlimited chaos. They are rules-based. Providers must be qualified. Treatment must be clinically appropriate. Documentation is required. Some items require approval. There are scheduled contributions. Rules can be regularly adjusted to accommodate a range of factors.

In other words, activity can be demand-led without being uncontrolled.

That is the distinction I think New Zealand needs to explore for primary medical care.

That does not mean a blank cheque.

It does not mean returning to 1980s medicine.

And it does not mean replacing capitation.

It means a properly designed hybrid.

Capitation for population responsibility.

Fee-for-service for eligible medical activity.

Place-based commissioning for people and communities who might otherwise be left behind.

And transparent data so we can see whether the system is actually improving access, rather than just shifting patients around.

Figure 2. Funding models in plain English. The point is not that one model is good and the others are bad; each model rewards a different behaviour.

A useful caution: This is not a claim that fee-for-service is good and capitation is bad. Each model solves one problem and creates another. The practical question is the blend.

The useful way to think about all three

In plain English: this post is about what each funding rule makes easier or harder. The question is not which label sounds best, but what the rule makes clinics, funders and patients do next.

A simple way to remember the three models is this:

• capitation pays for responsibility;

• fee-for-service pays for activity;

• programme funding pays for specific organised work.

A mature system usually needs all three. The argument starts when one of them is asked to do a job it cannot do well.

What would change my mind?

I would be less convinced if one payment model consistently delivered access, equity, continuity, fiscal control and rural supply without needing the others. I do not think the evidence or lived system experience shows that.

Deep dive (optional, not required reading): I’ve included the fuller explanation, game table, modelling notes and full source list in the appendix below.

Note: This series is exploratory policy analysis. It is not a party-political argument, not a position sponsored by an external body, not a claim that any single funding model is perfect, and not a calibrated prediction of savings. The central question is whether New Zealand's current funding architecture lets lower-cost upstream care expand safely before need becomes hospital demand.

Useful links

• MoH capitation reweighting

• RACGP funding models

• Cochrane payment review

• Australian incentives review

• ACC treatment payments

Deep dive appendix for Post 02: Fee-for-service, capitation and blended funding: the plain-English version

This appendix is supporting material for the public post. It carries the longer explanation, sources and assumptions for readers who want the detail.

Visual shorthand for the three funding types

For visual readers, the three models differ mainly in what triggers payment: the service, the enrolled person, or the programme rule.

The game underneath the policy

Every post in this series is built around a game. A game is simply a situation where each player responds to the rules and to what the other players do.

A compact player map:

• Patients: avoids Delay, cost, uncertainty, worsening illness; under pressure may wait, pay, delay, use telehealth, call ambulance, go to hospital.

• Providers: avoids Unfunded work, burnout, financial risk; under pressure may close books, shorten appointments, raise fees, limit extra activity.

• Health New Zealand: avoids Visible failure, deficits, hospital pressure; under pressure may prioritise urgent hospital pressures.

• Primary Health Organisations or locality bodies: avoids Loss of role, loss of funding, accountability risk; under pressure may defend functions, manage pass-through, shape provider incentives.

• Accident Compensation Corporation: avoids Uncontrolled claims cost, poor outcomes; under pressure may tighten payment rules or shift toward commissioning.

• Ministers: avoids Publicly visible service failure; under pressure may fund the pressure people can see.

This is why an apparently technical funding issue becomes a political economy issue very quickly.

How this fits the hybrid model

The hybrid model has five parts:

• capitation for continuity and population responsibility;

• uncapped scheduled fee-for-service for eligible primary medical activity;

• place-based accountability so providers cannot simply cherry-pick easy activity;

• scope-enabled supply so safe care can be generated by the right provider, not only the traditional provider;

• data, audit and top-tier key performance indicators so the system can see access failure before it becomes hospital pressure.

The model is deliberately not a blank cheque. The point is to remove the global cap on eligible primary medical activity, while keeping item prices, clinical eligibility, provider scope, documentation, audit, co-payment protections and place accountability.

What this adds to the modelling

In the demonstrative model, this post corresponds to one or more component games. The model asks what happens if the system stays in the current equilibrium, and what happens if the policy architecture shifts the equilibrium.

The model does not claim, yet, that the preferred architecture will reduce emergency department presentations by a precise number. That would require linked data, calibration and validation. What the model does show is the logic of the mechanism and the assumptions that need to be tested.

The most important empirical tests are:

1. whether scheduled activity payments increase safe primary care supply;

2. whether unmet primary care need flows into urgent care, ambulance and hospitals;

3. whether Accident Compensation Corporation activity payments help sustain local primary care capacity;

4. whether Primary Health Organisation payment arrangements create material pass-through, transparency or entry barriers;

5. whether scope-enabled providers can expand supply safely and equitably.

Sources and further reading

• MoH capitation reweighting

• RACGP funding models

• Cochrane payment review

• Australian incentives review

• ACC treatment payments

• Health system review

• Primary-care Cabinet material

• Primary-care health target

• National Primary Care Dataset

• Vote Health 2025/26

• NZ Health Survey update

• Health NZ roles

Appendix note: Supporting material for readers who want the longer explanation, sources and assumptions.