Answerability
Statistics Canada asks every Indigenous identity group the same 2,430 health questions. For non-Indigenous adults, 91 per cent come back with a number. For Inuit adults, 11 per cent do. This is not a story about anyone withholding anything. It is a measurement of what a survey built at a national sample size can and cannot see, and of what that costs.
The same questions, asked of everyone
Table 13100924 is unusually well suited to this because it is a complete cross product. Five identity groups, each asked an identical set of 2,430 question shapes, 77,760 cells apiece. Not approximately matched. Identical. That is what makes the comparison below a comparison rather than an impression.
Restricting to a single statistic, so that an estimate and its confidence bounds are not counted three times:
Health questions refused as too small, all jurisdictions pooled, percent statistic only
| Group | Questions asked | Refused as too small |
|---|---|---|
| Inuk (Inuit) | 12,960 | 89.4% |
| Métis | 12,960 | 54.2% |
| First Nations (North American Indian) | 12,960 | 51.8% |
| Total, Indigenous identity | 12,960 | 27.6% |
| Non-Indigenous people | 12,960 | 8.7% |
The pattern holds in a second table. Food security among First Nations people living off reserve, table 41100063, refuses 54.9 per cent of its cells against 54.8 per cent in the health table. Two instruments, built separately, arriving at the same place.
What this is not
Suppression is correct practice. When a cell contains too few people, publishing it can identify them, and a statistical agency that published anyway would be doing something worse than declining. The refusals counted here are an agency following its own confidentiality and reliability rules, and this page is not an accusation that anyone broke them.
It is also not evidence that Indigenous data is suppressed preferentially. The most obvious counter-argument is that refusal tracks population size, and when tested against the survey's own population counts, it does:
Population against refusal rate, same questions, same table
| Group | Population | Refused |
|---|---|---|
| Inuk (Inuit) | 29,600 | 89.4% |
| First Nations (North American Indian) | 370,900 | 51.8% |
| Métis | 407,900 | 54.2% |
| Total, Indigenous identity | 828,300 | 27.6% |
| Non-Indigenous people | 25,406,700 | 8.7% |
The relationship is monotonic. Size explains most of it, and stating that plainly is more useful than the version of this finding that leaves it out. The interesting claim is not that something improper happened. It is that a survey's resolution is a function of the population it is asking about, that this is entirely predictable, and that nobody publishes what it costs.
The cost, at each level of resolution
The same battery of questions, asked at every geographic level the table covers. Answerability falls as the question narrows.
Share of the same 24,300 questions returning a published value, by jurisdiction
| Jurisdiction | Answered |
|---|---|
| Canada | 85.2% |
| Ontario | 63.9% |
| Alberta | 52.6% |
| British Columbia | 51.3% |
| Nova Scotia | 32.4% |
| Northwest Territories | 30.8% |
| New Brunswick | 26.3% |
| Nunavut | 22.7% |
| Yukon | 19.8% |
| Prince Edward Island | 17.6% |
Nunavut is worth pausing on. It is a territory whose population is largely Inuit, and the national health survey can answer 22.7 per cent of questions about it. Prince Edward Island, a province of about 175,000 people, sits at 17.6.
Below the province line, this stops being a rate and becomes an absence. Nine provinces and regions return no published value at all for Inuit adults: 4,860 questions asked in each of Alberta, British Columbia, Manitoba, New Brunswick, Nova Scotia, Ontario, Prince Edward Island, Quebec and the Prairies, and not one answer. First Nations adults in Nunavut, and Métis adults in Nunavut and Prince Edward Island, are in the same position.
Anyone arguing for municipal or community level health data should know this before they start. The instrument does not thin out at that resolution. It has already stopped.
Why this warehouse can measure it
Because it kept the refusals. A cell that comes back withheld is loaded here as a row with a status, not dropped on the way in. That is an unusual choice and it is the only reason any of the above can be counted. Across every source held, 7,146,622 cells were asked for, 1,427,413 came back refused, and 705,499 of those were refused specifically because the group was too small to report on safely.
Nothing about this is hidden. Every number is in published tables that anyone can download. What is missing everywhere else is the counting.
What follows
A refusal rate is a measurement, which means it can be tracked, targeted and moved. It falls when a survey oversamples a population, when a community holds and reports its own data, and when an instrument is designed for the question rather than borrowed from a national frame. It is therefore a reasonable way to ask whether a data investment worked: not whether a report was published, but whether the share of questions a community can answer about itself went up.
Some of that answer will not come from Statistics Canada, and should not. Where the questions are about First Nations, Inuit or Métis populations, the instruments that close this gap are the ones those nations and communities govern themselves. The number on this page is not an argument for extracting more. It is a measurement of how little the current arrangement can see.