The arithmetic of the NDIS cuts
Two numbers go in. Every step is shown, so you can check it yourself. Budget figures are stated over four years, so the totals are too.
Professor Paul A. Watters OAM · Honorary Professor, Macquarie University
The 2026 reforms cut NDIS spending by about $35–38bn over four years — roughly $9bn a year.
Annual participant growth on the pre-reform trajectory. National Cabinet's agreed cap was 8% growth a year.
Every $1 of NDIS spending generates about $2.25 in economic activity — Per Capita, False Economy (2021), commissioned by National Disability Services.
There is no four-year total. Jobs are a stock, not a flow.
10,200 jobs per $1bn underfunded — same source. The workforce falls by this much and stays down, so multiplying by four would count the same lost jobs four times. Headcount, not full-time equivalent; the sector is heavily part-time, so FTE would be materially lower. Bloomberg Economics separately puts up to 140,000 disability sector jobs at risk by 2030.
Shallow cuts remove modest amounts from many people; deep cuts remove large amounts from few. The split above reflects that. The scheme was also growing, so holding it flat shuts out two separate groups: those who lose supports they already have, and those who would have qualified and now never enter. The second group is larger, and is usually left out of the debate entirely.
Unlike the figures above, this is an assumption rather than a published coefficient.
No Australian study measures how many people die when disability supports are withdrawn. Two mechanisms are modelled, because cuts do not fall evenly. The first $13bn comes from discretionary supports — community participation and coordination — at 5 additional deaths per 10,000 people, a 10% increase on an annual mortality rate near 0.55%. Beyond that the discretionary pool is exhausted and the money must come from core daily-living and SIL supports. That does not remove whole packages from a few people; it removes hours from all 80,000 of them. For someone who cannot eat, turn or manage continence unaided, withdrawn hours are not reduced amenity — they are aspiration, pressure injury, urinary sepsis and dehydration, on a timescale of days. Complete withdrawal is costed at raising annual mortality from about 2% to about 15%, scaled linearly by the share of support removed. The UK LeDeR review found about one in four deaths in this group were treatable. Counts only people losing existing supports, not those denied entry.
| 2.25 | Economic activity generated per $1 of NDIS spending | 1 |
| 10,200 | Jobs per $1bn of underfunding — headcount, not FTE | 1 |
| $13bn | Discretionary pool — community participation and coordination spending available to cut before core supports are reached | 3, 7 |
| $50,000 | Average value withdrawn per person, discretionary cuts | 2, 3 |
| $25bn | Core daily-living and SIL spending — the pool deep cuts must come from | 3 |
| 80,000 | Participants dependent on core daily support, including about 33,000 in supported independent living | 3 |
| 2% | Annual mortality, people with severe or profound disability | 4, 5 |
| 13 pts | Rise in annual mortality at complete withdrawal of core support, from 2% to 15% — the single most consequential assumption on this page | 5, 6 |
| 760,000 | Participants before the reforms | 3 |
| 0.55% | Annual mortality rate, adults with intellectual disability | 4 |
| 5 | Deaths per 10,000 when discretionary support is withdrawn — assumption; 0.55% baseline mortality × 10% increase | 4, 5, 6 |
| 4 years | Budget period over which the reductions are stated | 2, 3 |
This is a research draft. Its purpose is to stimulate thinking about how the effects of funding changes on people with disability might be modelled, and to make the arithmetic of those effects visible and contestable. It is not a forecast, an official estimate, or a finding. It is a set of calculations that follow from stated inputs, published where anyone can check them.
This model has not been peer reviewed. It has not been independently verified or validated against observed outcomes, and no such validation is currently possible, because the policy changes it models have not yet taken full effect and Australia does not systematically record the deaths of people with disability in a way that would permit it.
The code has not been independently audited. It may contain errors. The arithmetic shown on each card is provided precisely so that readers can check the calculations themselves rather than trusting the implementation, and any reader who finds a discrepancy should assume the implementation is wrong and say so.
Every constant listed above sits on a chain of assumptions, and those assumptions may be sound, unsound, or partly sound. Some are drawn from published research conducted in other jurisdictions, most importantly the United Kingdom, where service systems, funding structures and clinical pathways differ from Australia's in ways that may or may not matter. Some are derived by combining published figures in ways their original authors did not test and might not endorse. At least one — the increase in mortality following withdrawal of core daily support — has no published estimate behind it at all and is a judgement. Where the model reports a range, that range reflects the spread of the inputs, not a statistical confidence interval, and should never be described as one.
The relationships modelled here are linear. Real systems are not. Multipliers estimated at the margin are applied across a much wider span than the source material establishes; mortality is scaled in direct proportion to support withdrawn, when the true relationship is more likely to involve thresholds and step changes; and the boundary between discretionary and core support is treated as a clean line when in practice it is contested, variable between participants, and administratively blurred.
The model converts dollars into people using average package values. Averages of this kind describe a distribution; they do not describe anyone. The distribution of NDIS package values is extremely skewed: a large number of participants receive modest plans while a small number receive packages an order of magnitude larger, and the people at that end are precisely those whose survival depends on the support being funded. Any figure produced by dividing an aggregate dollar amount by an average package is therefore an accounting convenience rather than a count of individuals, and results expressed as numbers of people should be read accordingly.
The same caution applies with greater force to mortality. The model produces population-level statistical expectations under stated assumptions. It says nothing whatever about any individual, and it cannot be used to predict, explain or attribute the death of any particular person. Individual outcomes depend on clinical circumstances, informal support, housing, provider quality and chance, none of which appear here.
It does not model cost-shifting onto state health, mental health, housing, justice or emergency services, which means the fiscal picture is incomplete in a direction that understates total cost. It does not model non-fatal harm, which is by far the largest category of consequence and includes pain, infection, hospitalisation, institutionalisation, loss of housing, carer breakdown and loss of community participation. It does not model behavioural responses, provider market dynamics, substitution into other funded programs, or offsetting government action. It does not distinguish between jurisdictions, age groups, disability types or remoteness. It assumes the announced policy parameters are implemented as announced, which they may not be.
Figures are conditional on the inputs shown on screen and must be cited together with those inputs and with the assumption set that produced them. Quoting a number from this page without its parameters misrepresents it. The mortality figures in particular must be presented as scenario calculations under stated assumptions and never as predicted, expected or forecast deaths.
Nothing here constitutes medical, legal, financial, actuarial or policy advice. No warranty is given as to accuracy, completeness or fitness for any purpose, and no reliance should be placed on these figures for any decision affecting an individual. The model does not represent the position of the National Disability Insurance Agency, any government, or any organisation cited in the sources above. Sources are cited as they stood at the date of this draft and may since have been superseded, corrected or withdrawn.
Correction and criticism are actively invited. If a parameter is wrong, an inference unsound, or a calculation mistaken, that should be said publicly, and the model changed or withdrawn accordingly.