Who counts as a child

In 2022, Manitoba had 29.60 children in out-of-home care for every thousand children. Ontario had 2.72. That is an elevenfold difference between two provinces in the same country, under the same constitution, in the same year.

It is a real number from a good source: the Canadian Child Welfare Research Portal's statistics matrix, drawn from Pollock and colleagues, 2024. Thirteen jurisdictions, each with its rate, its count of children in care, and the child population that rate is calculated against. I checked all thirteen; every published rate reconciles exactly with its own count and population. Nothing is fudged.

The trouble starts one column to the right.

Five definitions of a child

The matrix publishes something most reproductions of it drop: the age span each jurisdiction counts.

Five jurisdictions count children up to 17. Four count to 18. Newfoundland and Labrador and Saskatchewan count to 21. Nova Scotia counts to 20. Yukon counts to 25.

Both numbers are labelled "per 1,000 children". They are not measuring the same thing.

Yukon's denominator runs eight birth years longer than Manitoba's. The difference is not an error by anyone; it is thirteen legislatures having made thirteen decisions about when childhood ends for the purposes of a child welfare statute. The portal publishes the span. It sits right beside the rate. What happens is that the rate travels and the span does not.

Established findings

You cannot simply restate every jurisdiction on a common age span, and it is worth being exact about why, because the reason is the whole point.

To standardise properly you would need each jurisdiction's in-care count broken down by age, so that narrowing the denominator to 0 to 17 could be matched by removing the 18-and-overs from the numerator. No jurisdiction publishes that. One missing column is the difference between thirteen comparable numbers and thirteen numbers that merely look comparable.

What can be established is a bound. Narrow each jurisdiction's denominator to a common 0 to 17 while keeping its whole numerator, and you get the highest rate that jurisdiction could possibly have on that span, because every child aged 18 or over genuinely in its count is left in the numerator. The true standardised rate lies somewhere between the published figure and the bound.

A bound is not an answer. It is the size of the room the answer is in.

Results

The spread survives. Published, the highest rate is 10.9 times the lowest. At the bound it is 11.1 times. The gap between Manitoba and Ontario does not close at all, for a simple reason: both already count to 17. The elevenfold spread is not an artefact of who counts as a child. I had assumed a good part of it would be, and that assumption was wrong.

The ranking does not survive. This is where the age span does its damage, and it is the thing people actually use, because nobody quotes a rate without placing it against the others.

Yukon publishes 16.49, fifth highest in the country. On a common denominator its upper bound is 24.45, a rise of 48%, and it moves to second. Saskatchewan rises 23%, and Newfoundland and Labrador rises 27%. Meanwhile Nunavut and the Northwest Territories each fall two places in the ordering, not because their rates went down, but because they went up more slowly than the wide-span jurisdictions above them.

Anybody ranking provinces on these figures is ranking them partly by legislative drafting.

The table, both ways

Switch the ordering and watch what moves. The rows marked at the floor are the ones whose movement is smaller than the measurement noise, explained below.

Movement, or noise

Five jurisdictions already count to 17. For them the bound should equal the published rate exactly, because nothing changes. What they actually show is a movement of up to 1.48%, which is Statistics Canada revising its population estimates after Pollock and colleagues published.

That gives a floor, and it is the reason I trust the rest. Yukon at 48%, Newfoundland at 27%, Saskatchewan at 23% and Nova Scotia at 18% are far clear of it. New Brunswick and the Northwest Territories at around 6%, British Columbia at 4% and Nunavut at 3% are close enough to the floor that I would report them as movements consistent with the age span rather than as findings.

Without that control I would have written all eight up the same way, and four of them would not have deserved it.

An interesting observation

Nova Scotia publishes 5.98 per 1,000 on a span of 0 to 20, second lowest in the country. On a common 0 to 17 denominator its upper bound is 7.04, a rise of 18%.

That is real, and worth saying plainly: part of what makes Nova Scotia's rate look low is that it is divided by three extra birth years. But its position does not move. It is second lowest either way.

So the tempting sentence, that Nova Scotia only appears restrained because its denominator is wide, is true about the number and false about the conclusion people draw from it. I am stating both halves because the first is the interesting one and the second is the honest one.

None of this says whether any province removes too many children or too few. The public record cannot support that judgment, and a rate was never going to answer it.

Two fixes

Publish the age span beside every rate. It costs one column. It is already collected, since it is what the denominator was built from. It would stop the most common misreading of these figures at no analytical cost whatsoever.

Publish in-care counts by age. This is the harder ask, and the one that matters. It is the single missing column standing between these thirteen numbers and genuine comparability.

It is worth being clear about what kind of problem this is, because it is easy to assume the data does not exist. It does. I have collected age-disaggregated counts of children in care myself, as part of a quality improvement plan; the ages are in the file, the file is in the system, and somebody runs the report. The gap is not in the case files, and it is not a limitation of the front line. It sits between what is collected and what is released, which makes it a publication decision rather than a measurement problem, and publication decisions can simply be made differently.

And the general form, which is why this series is called what it is: a rate is a quotient, and an argument about a rate is almost always an argument about its divisor. When a number arrives without its denominator, that is not a detail that got left out. That is the part someone decided you did not need.