A $51.5 Billion NDIS Should Show What Outcomes It Delivers

Participants report some gains. But public reporting still cannot show which supports produced them—or what outcomes the Scheme’s $51.5 billion annual spend is delivering.

NoteExecutive summary

• The NDIS made $51.5 billion in nominal cash-basis payments in the 12 months to June 2026. Spending is traceable by category; outcomes are not linked to the supports and dollars behind them.
• This is not a claim that there are no benefits. Participants report improvement in some areas, and independent studies find credible gains in selected domains.
• The accountability gap is that public evidence still cannot reliably show which funded supports work, for whom, and whether the outcomes are commensurate with the spend.

The NDIS can tell the public—in considerable detail—how many participants it supports and how much it pays. In the 12 months to June 2026, nominal cash-basis payments reached $51.5 billion, as set out in the companion analysis of NDIS payments and utilisation. For a program of that scale, the reasonable question is: where is the equally clear public evidence of the outcomes that spending delivers?

The answer is not “there are no measured outcomes”. The NDIA’s 2024 Participant Outcomes Summary Report examined 296 indicators and found 30% with statistically significant and material change, separating favourable from adverse movements. Some participants report meaningful improvement and several independent analyses find credible scheme-level benefits.

The problem is traceability. Public outcome reporting rarely lets us connect a particular therapy, support worker, assistive technology purchase or other funded support—and the money spent on it—to a better outcome. The freshest downloadable extract also publishes rounded aggregates without the denominators and uncertainty needed to judge the precision of individual movements.

The cohort consists of matched respondents whose latest Short Form assessment was their first reassessment, generally around one year after first-plan approval. The data can show where reported outcomes moved; it cannot establish that the NDIS caused the movement.

We can see some change, but not what the spending produced

The argument becomes clearer when it is split into three questions:

1. Did reported outcomes change after people entered the Scheme? Yes, for some measures and cohorts. This is the strongest and most current public evidence.
2. Did the NDIS cause those changes? There is some credible scheme-level evidence, but it is mixed across outcomes and methods.
3. Which supports caused better outcomes for which participants—and at what cost? The public evidence is not yet reliable enough to answer this.

evidence_ladder <- tibble(
  x = 1:3,
  question = c(
    "Reported outcomes\nchanged",
    "The NDIS caused\nthe change",
    "A particular support\ncaused the change"
  ),
  assessment = c(
    "Current descriptive evidence",
    "Some credible but mixed evidence",
    "Not established in public data"
  ),
  fill = c("#246b8e", "#b77b2b", "#8b4a45")
)

ggplot(evidence_ladder, aes(x, 1)) +
  annotate("segment", x = 1.36, xend = 1.64, y = 1, yend = 1, colour = "grey60", linewidth = 1) +
  annotate("segment", x = 2.36, xend = 2.64, y = 1, yend = 1, colour = "grey60", linewidth = 1) +
  geom_tile(aes(fill = fill), width = 0.72, height = 0.8, show.legend = FALSE) +
  geom_text(aes(label = question), colour = "white", fontface = "bold", lineheight = 0.95, size = 4) +
  geom_text(aes(y = 0.48, label = assessment), colour = "grey25", size = 3.3) +
  scale_fill_identity() +
  coord_cartesian(xlim = c(0.55, 3.45), ylim = c(0.25, 1.45), clip = "off") +
  labs(title = "The evidence thins as the question gets more specific") +
  theme_void(base_size = 12) +
  theme(plot.title = element_text(face = "bold"), plot.margin = margin(10, 20, 15, 20))

Public evidence is strongest for reported change, mixed for scheme-level attribution and weakest for intervention-level effectiveness.

The broader evidence base supports that middle position. The independent trial evaluation found that the Scheme delivered intended outcomes overall, including improved choice and control and generally modest participant-wellbeing gains, but that benefits were uneven. It found no statistically detectable effect on overall child wellbeing—a broader measure that does not negate the reported communication and friendship movements below, but does stop us treating them as causal evidence of a child-level benefit. Taylor Fry’s modelling for the NDIS Review found positive results for some participant outcomes—including life satisfaction, choice and control, housing, transport and hospital use—but not self-rated health, safety or school outcomes. Its separate HILDA analysis found improved participant financial security and stronger carer results, but no significant participant improvement in life satisfaction, employment, income, school outcomes or psychological distress.

A 2025 quasi-experimental staggered-rollout study provides a useful counterweight: its difference-in-differences design found no statistically significant short-run area-level effects on formal-service use, caring hours, carer employment or social participation. The causal interpretation depends on the design’s identifying assumptions; the study also had limited statistical power and measured local NDIS availability rather than confirmed individual receipt. It is a null result—not proof of no benefit.

Taken together, the evidence says that benefits are plausible and sometimes visible, but not universal. It is much stronger at identifying broad change than at showing what the Scheme’s spending produced.

The newest public data shows most first-reassessment changes are small

For each outcome, I compare the baseline percentage with the first reassessment percentage for the group recorded as having one reassessment. Three indicators are coded adverse on their face—broad development concerns, inability to access wanted training, and having no friends beyond family or paid staff—and reversed so that positive means favourable movement. For development concerns, improved recognition may contribute to the movement, so its interpretation remains uncertain. The two “want more choice and control” indicators are excluded because their direction is ambiguous: wanting more choice can reveal unmet need, greater agency, or both.

Across the 40 direction-classified participant indicators, most movements cluster close to zero. Early childhood has 4 of 5 indicators moving favourably, but it also contains the largest movement coded adverse on its face. Among adult indicators, 8 improve, 6 are unchanged and none worsen at the published one-percentage-point precision. The red medians summarise heterogeneous indicators; they are not participant medians and should not be used to rank cohorts.

plot_reassessment <- reassessment_one |>
  mutate(cohort = factor(cohort, levels = expected_cohorts))

ggplot(plot_reassessment, aes(cohort, adjusted_change)) +
  geom_hline(yintercept = 0, colour = "grey55", linewidth = 0.6) +
  geom_jitter(width = 0.10, height = 0, alpha = 0.55, size = 2.2, colour = "#246b8e", seed = 2026) +
  stat_summary(fun = median, geom = "crossbar", width = 0.48, linewidth = 0.7, colour = "#9b3f2f") +
  scale_y_continuous(labels = label_percent(accuracy = 1), breaks = seq(-0.05, 0.15, 0.05)) +
  labs(title = "Most first-reassessment indicators moved only a few points", x = NULL, y = "Favourable change from baseline")

Direction-adjusted change from baseline to first reassessment. Each point is one published outcome; positive values indicate favourable movement. Red bars are medians across indicators, not participant-level effects.

The largest reported movements are concentrated

The largest favourable change is Can communicate what they want for early childhood, up 15 percentage points. The largest movement coded adverse on its face is Fewer broad development concerns for early childhood, down 7 percentage points after accounting for direction. The NDIA’s 2024 report notes that greater use of specialist services can also reflect improved recognition of developmental concerns, so that indicator’s interpretation is uncertain.

The chart makes the concentration visible: the two double-digit favourable movements both relate to children before school age. Adult employment, education and volunteering do not show comparable movement in the first-reassessment cohort.

largest_movements <- reassessment_one |>
  slice_max(abs(adjusted_change), n = 10, with_ties = FALSE) |>
  mutate(
    label = str_c(cohort, ": ", outcome_short),
    label = fct_reorder(label, adjusted_change),
    direction_label = if_else(adjusted_change >= 0, "Favourable", "Adverse")
  )

ggplot(largest_movements, aes(adjusted_change, label, colour = direction_label)) +
  geom_vline(xintercept = 0, colour = "grey65") +
  geom_segment(aes(x = 0, xend = adjusted_change, yend = label), linewidth = 0.9) +
  geom_point(size = 3) +
  geom_text(
    aes(label = percent(adjusted_change, accuracy = 1)),
    hjust = if_else(largest_movements$adjusted_change >= 0, -0.25, 1.25),
    show.legend = FALSE,
    size = 3.4
  ) +
  scale_x_continuous(labels = percent, limits = c(-0.09, 0.17)) +
  scale_colour_manual(values = c(Favourable = "#16735c", Adverse = "#b0433c")) +
  labs(title = "The largest reported movements were in early childhood", x = "Direction-adjusted change", y = NULL, colour = NULL)

A descriptive selection of the ten largest rounded absolute movements from baseline to first reassessment. Positive values indicate favourable movement. The source provides no sample counts or uncertainty estimates, so this is not a statistical ranking.

Entry outcomes provide context, not a current scorecard

The baseline file is often tempting to describe as “current outcomes”. It is not. It records outcomes when participants entered the Scheme, for people represented in the June 2026 extract. It is a starting-point profile.

Among adults, the entry picture separates relatively common measures of home and day-to-day choice from much lower economic and community participation. Only 23% had a paid job, 10% volunteered and 9% participated in education, training or skill development at baseline.

adult_baseline_patterns <- c(
  "choose what they do each day",
  "choose who supports them",
  "feel safe or very safe",
  "did not have any difficulties accessing health services",
  "rate their health as good",
  "actively involved in a community",
  "have a paid job",
  "participate in education, training",
  "volunteer"
)

adult_baseline <- baseline |>
  filter(cohort == "Adults") |>
  filter(str_detect(str_to_lower(outcome), str_c(adult_baseline_patterns, collapse = "|"))) |>
  mutate(
    outcome_short = short_outcome(outcome),
    outcome_short = fct_reorder(outcome_short, baseline_rate)
  )

assert_true(nrow(adult_baseline) == length(adult_baseline_patterns), "The selected adult baseline outcomes changed.")

ggplot(adult_baseline, aes(baseline_rate, outcome_short)) +
  geom_segment(aes(x = 0, xend = baseline_rate, yend = outcome_short), colour = "grey80", linewidth = 0.8) +
  geom_point(size = 3, colour = "#246b8e") +
  geom_text(aes(label = percent(baseline_rate, accuracy = 1)), hjust = -0.35, size = 3.5) +
  scale_x_continuous(labels = percent, limits = c(0, 0.85), expand = expansion(mult = c(0, 0.03))) +
  labs(title = "Adult entry outcomes were lowest for work, study and volunteering", x = "Share at Scheme entry", y = NULL)

Selected outcomes for adults at NDIS entry, in the data current to 30 June 2026. These are baseline—not current—levels.

The missing public account is spend to support to outcome

These published movements are useful monitoring signals, not participant-level effect estimates. They are not uniformly large and they are not spread evenly across ages or domains. The largest reported favourable movements are in early-childhood communication and friendship or social connection. Broad development concerns move adversely on their face, although better recognition may be part of that movement. For adults, the indicators tied to work, learning and community participation remain low at entry and generally change little in the first-reassessment cohort.

The Independent NDIS Review reached the same deeper problem: the Outcomes Framework was not fit for purpose and it was difficult to link supports, investment and outputs to participant outcomes. Taylor Fry also warned that observational spending models can mistake participant need, goals or declining function for the effect of a support. More spending can correlate with a worse outcome because people with greater needs receive more support; an employment-support payment can correlate with employment because the person already had an employment goal.

So the core conclusion is not that the NDIS fails. It is that Australia spends and reports tens of billions of dollars without yet having a reliable public learning system that tells participants, providers and governments what outcomes that spending produced.

What a real learning system would need

• participant-level links between the support received, its intensity and timing, and later outcomes;
• published sample sizes and uncertainty, with comparable outcomes defined before analysis;
• credible comparison groups or phased evaluations where practical and ethical;
• results split by disability, age, geography and starting need so average effects do not hide who benefits; and
• participant-defined outcomes and cost-effectiveness, not spending or plan utilisation alone.

Three-line takeaway

• NDIS payments are traceable by category; public outcomes are not linked to the supports and dollars behind them.
• This does not mean there are no benefits: participants report some gains and independent studies find credible improvements in selected domains.
• The unresolved public-accountability question is which supports produced which outcomes, for whom, and whether those outcomes justify the spend.

Appendix

NoteCohort definition, caveats and method

The first-reassessment comparison describes matched respondents whose latest Short Form assessment was their first reassessment, generally around one year after first-plan approval. It is not a representative panel of all NDIS participants and has no untreated counterfactual.

Four limits shape the interpretation:

• The extract has no participant-level records or sample counts, so uncertainty cannot be calculated.
• Common questions can be linked across age-based Short Form questionnaires, but a question introduced only in an older-age form receives a new baseline. The observations underpinning individual indicators can therefore differ.
• Percentages are rounded to whole percentage points. A reported zero change may conceal a smaller movement.
• The direction-adjusted charts combine heterogeneous indicators. Their red medians are medians across published indicators, not people, and the top-ten chart is a descriptive selection of rounded extremes rather than a statistical ranking.

For direction coding, three indicators are coded adverse on their face—broad development concerns, inability to access wanted training, and having no friends beyond family or paid staff—and reversed so positive means favourable movement. Improved recognition may contribute to the development-concerns movement, so that indicator’s interpretation remains uncertain. The two “want more choice and control” indicators are excluded because their direction is ambiguous.

NoteBroader evidence and spending-source audit

The broader sources answer different questions and therefore should not be treated as interchangeable estimates. The Flinders comparison-group evaluation and Taylor Fry models are the most informative here for scheme-level attribution. The NDIS Review is an expert synthesis diagnosing the measurement gap rather than an estimate of the Scheme’s effect. The latest NDIA extract is strongest for current descriptive monitoring, while the staggered-rollout study is a quasi-experimental causal-inference design estimating narrow area-level carer and service-use effects.

evidence_sources |>
  transmute(
    Source = str_c("[", title, "](", source_url, ")"),
    Year = publication_year,
    Design = study_design,
    `What it supports` = what_it_supports,
    `Key limit` = key_limit
  ) |>
  kable(align = c("l", "r", "l", "l", "l"), escape = FALSE)

Source	Year	Design	What it supports	Key limit
Payments data to 30 June 2026	2026	Administrative cash-basis payment data	Shows nominal NDIS payments reached 51.5 billion dollars in the trailing 12 months	Spending does not itself measure participant benefit
Participant datasets: baseline and longitudinal outcomes	2026	Matched longitudinal survey aggregates	Tracks reported participant outcomes from Scheme entry to reassessment	No denominators uncertainty estimates or non-NDIS counterfactual
Participant Outcomes Summary Report to 30 June 2024	2024	Longitudinal survey analysis	Shows statistically significant and material change in some domains alongside adverse or unchanged results	Change over time is not proof that the Scheme caused the change
Evaluation of the NDIS: Final Report	2018	Mixed-method independent trial evaluation with a comparison group	Found scheme-level gains in choice and control and generally modest participant wellbeing gains	Early trial sites with uneven effects and differential survey response
NDIS Review: Costs benefits and frameworks	2023	NDIA outcome modelling plus HILDA longitudinal modelling	Finds benefits in some participant and carer domains but null or mixed results in others	Imperfect counterfactual small HILDA NDIS sample and assumption-dependent valuations
Working together to deliver the NDIS: Final Report	2023	Independent policy and evidence review	Concludes current systems cannot adequately show what works for whom and why	Expert synthesis rather than a causal impact study
Exploring the impact of a personalised disability reform on people with disability and their primary carers	2025	Quasi-experimental staggered-rollout difference-in-differences study	Tests short-run area-level effects on formal-service use and carer outcomes and finds no statistically significant effects	Area availability is not confirmed receipt and the 1054-dyad sample had limited power

The evidence is consistent with benefits for some participants. It is not sufficient to rank individual supports or providers by effectiveness.

NoteDirection-coding audit for all indicators

reassessment_all |>
  arrange(factor(cohort, levels = expected_cohorts), indicator_number) |>
  transmute(
    Cohort = cohort,
    Indicator = indicator_number,
    Outcome = outcome,
    Baseline = percent(baseline_rate, accuracy = 1),
    Reassessment = percent(reassessment_rate, accuracy = 1),
    `Raw change` = paste0(number(100 * raw_change, accuracy = 1), " pp"),
    `Assigned direction` = if_else(direction_class == "Ambiguous", "Ambiguous—excluded", direction_class),
    `Direction-adjusted change` = if_else(
      is.na(adjusted_change),
      "Excluded",
      paste0(number(100 * adjusted_change, accuracy = 1), " pp")
    )
  ) |>
  kable(align = c("l", "r", "l", "r", "r", "r", "l", "r"))

Cohort	Indicator	Outcome	Baseline	Reassessment	Raw change	Assigned direction	Direction-adjusted change
Early childhood	1	% with concerns in 6 or more of the areas: gross motor skills, fine motor skills, self-care, eating, social interaction, communication, cognitive development, sensory processing	69%	76%	7 pp	Adverse	-7 pp
Early childhood	2	% who say their child is able to tell them what he/she wants	62%	77%	15 pp	Favourable	15 pp
Early childhood	3	% of children who can make friends with people outside the family	48%	59%	11 pp	Favourable	11 pp
Early childhood	4	% of children who participate in age appropriate community, cultural or religious activities	45%	47%	2 pp	Favourable	2 pp
Early childhood	5	Of these, % who are welcomed or actively included	62%	65%	3 pp	Favourable	3 pp
School age	1	% developing functional, learning and coping skills appropriate to their ability and circumstances	28%	30%	2 pp	Favourable	2 pp
School age	2	% who say their child is becoming more independent	43%	49%	6 pp	Favourable	6 pp
School age	3	% of children who have a genuine say in decisions about themselves	79%	80%	1 pp	Favourable	1 pp
School age	4	% of children who can make friends with people outside the family	64%	67%	3 pp	Favourable	3 pp
School age	5	% of children who spend time after school and on weekends with friends and/or in mainstream programs	38%	39%	1 pp	Favourable	1 pp
School age	6	Of those, % who are welcomed or actively include	75%	75%	0 pp	Favourable	0 pp
School age	7	% of children who spend time with friends without an adult present	11%	12%	1 pp	Favourable	1 pp
School age	8	% of children attending school in a mainstream class	81%	77%	-4 pp	Favourable	-4 pp
Young people	1	% who are happy with the level of independence/control they have now	29%	32%	3 pp	Favourable	3 pp
Young people	2	% who choose who supports them	36%	39%	3 pp	Favourable	3 pp
Young people	3	% who choose what they do each day	46%	49%	3 pp	Favourable	3 pp
Young people	4	% who had been given the opportunity to participate in a self-advocacy group meeting	18%	18%	0 pp	Favourable	0 pp
Young people	5	% who want more choice and control in their life	82%	86%	4 pp	Ambiguous<U+2014>excluded	Excluded
Young people	6	% with no friends other than family or paid staff	34%	33%	-1 pp	Adverse	1 pp
Young people	7	% who have been actively involved in a community, cultural or religious group in the last 12 months	28%	31%	3 pp	Favourable	3 pp
Young people	8	% who are happy with their home	82%	81%	-1 pp	Favourable	-1 pp
Young people	9	% who feel safe or very safe in their home	84%	84%	0 pp	Favourable	0 pp
Young people	10	% who rate their health as good, very good or excellent	67%	66%	-1 pp	Favourable	-1 pp
Young people	11	% who did not have any difficulties accessing health services	70%	72%	2 pp	Favourable	2 pp
Young people	12	% who currently attend or previously attended school in a mainstream class	46%	45%	-1 pp	Favourable	-1 pp
Young people	13	% who have a paid job	15%	19%	4 pp	Favourable	4 pp
Young people	14	% who volunteer	7%	8%	1 pp	Favourable	1 pp
Adults	1	% who choose who supports them	68%	68%	0 pp	Favourable	0 pp
Adults	2	% who choose what they do each day	74%	75%	1 pp	Favourable	1 pp
Adults	3	% who had been given the opportunity to participate in a self-advocacy group meeting	23%	24%	1 pp	Favourable	1 pp
Adults	4	% who want more choice and control in their life	78%	81%	3 pp	Ambiguous<U+2014>excluded	Excluded
Adults	5	% with no friends other than family or paid staff	35%	35%	0 pp	Adverse	0 pp
Adults	6	% who have been actively involved in a community, cultural or religious group in the last 12 months	29%	32%	3 pp	Favourable	3 pp
Adults	7	% who are happy with their home	65%	68%	3 pp	Favourable	3 pp
Adults	8	% who feel safe or very safe in their home	64%	65%	1 pp	Favourable	1 pp
Adults	9	% who rate their health as good, very good or excellent	38%	38%	0 pp	Favourable	0 pp
Adults	10	% who did not have any difficulties accessing health services	58%	61%	3 pp	Favourable	3 pp
Adults	11	% who participate in education, training or skill development	8%	9%	1 pp	Favourable	1 pp
Adults	12	Of those who participate, % who do so in mainstream settings	75%	75%	0 pp	Favourable	0 pp
Adults	13	% unable to do a course or training they wanted to do in the last 12 months	37%	35%	-2 pp	Adverse	2 pp
Adults	14	% who have a paid job	22%	22%	0 pp	Favourable	0 pp
Adults	15	% who volunteer	9%	9%	0 pp	Favourable	0 pp

The two “want more choice and control” indicators are the ambiguous rows excluded from the direction-adjusted charts.

NoteSources and reproducibility

The analysis uses the NDIA’s baseline and longitudinal participant outcomes datasets and their data-rules documents, all current to 30 June 2026. The source files are pinned by size and SHA-256 checksum in sources.csv; set BLOG_REFRESH_DATA=true to force a fresh download and checksum validation. In the registry, release_date records the downloadable asset’s Last-Modified date.

source_manifest |>
  transmute(
    source = dataset,
    source_revision_date = release_date,
    coverage_end,
    file = local_file,
    bytes = comma(bytes),
    sha256_prefix = substr(sha256, 1, 12)
  ) |>
  kable(col.names = c("Source", "Source/revision date", "Coverage end", "Cached file", "Bytes", "SHA-256 prefix"))

Source	Source/revision date	Coverage end	Cached file	Bytes	SHA-256 prefix
Baseline outcomes data	2026-08-14	2026-06-30	baseline_outcomes.csv	67,976	02ac858b8800
Longitudinal outcomes data	2026-08-14	2026-06-30	longitudinal_outcomes.csv	1,395,120	1218998cae0b
Baseline outcomes data rules	2026-08-14	2026-06-30	baseline-outcomes-rules.docx	91,387	8c477e8fbc1a
Longitudinal outcomes data rules	2026-08-14	2026-06-30	longitudinal-outcomes-rules.docx	97,560	df6915b822ac