The denominator problem

Now the part that turned out to be wrong.

Look at the window everyone was looking at: 2015 to 2020. The line goes from 228 to 211 and then flattens. Progress appears to stop.

That reading launched a thousand op-eds about donor fatigue and a stalled agenda. We nearly published it ourselves.

The problem was the vintage of the data, not the data.

The series most sources were using ended in 2020 — right on top of the pandemic. Extend it to 2023 and the flatline resolves: 197, and falling again. From 2015 to 2023 the world improved 13.6%.

The stall was a virus, not a verdict. A series that stops one year early can turn a disruption into a diagnosis.

So if the world is still improving, who is doing the improving?

Split it by income group, 2015 to 2023. Low income: 491 down to 346, a fall of 29.5% — the fastest-improving group on the chart. Lower-middle: 291 to 235. Upper-middle: 61 to 57. High income: 11 to 10.

Every single group moved the right way.

Which leaves something that should not be possible.

Low income improved 29.5%. Lower-middle improved 19.2%. Those two groups are where roughly two-thirds of the world's babies are born. And yet the world as a whole improved only 13.6% — slower than either of them.

An average cannot fall more slowly than nearly all of its parts. Unless the parts are changing size.

Here is the second dataset, and the answer.

These are births per year, by income group. Low-income births rose from 21.8 million in 2015 to 25.3 million in 2023. Over the same years upper-middle births fell from 39.8 million to 28.5 million, and high-income births from 13.2 million to 11.5 million.

The low-income share of the world's births went from 15.2% to 19.2%. More of the world's childbirth is happening in the places where childbirth is most dangerous.

Put numbers on it.

Weight each group's mortality rate by its births and the world improves 14.5% from 2015 to 2023 — which matches the published world figure closely. Now hold the 2015 birth-mix fixed, so the only thing changing is how safe each group got: the same improvements produce 21.6%.

The gap between those two is the composition drag. About a third of the achievable improvement is being absorbed by where babies are now born. Every country can get safer while the world barely moves.

None of which makes the gap less stark.

Sub-Saharan Africa sits at 448 deaths per 100,000 live births in 2023, against a world figure of 197. It carries roughly 70% of all maternal deaths. The low-income group's rate is more than thirty times the high-income group's.

The composition effect is not a consolation. It is the reason the headline understates how much work is actually being done — and why the work has to keep accelerating just to hold the average still.

One last figure, because it is the one people ask about.

The United States reads 17 in 2015, 22 in 2020, 31 in 2021, and 17 again in 2023. The spike reversed. The claim that America is a uniquely and durably worsening rich country does not survive the return to its 2015 value.

A caveat we will not skip: US figures across 2003–2017 carry a measurement-change annotation — states phased in a pregnancy checkbox on death certificates at different times, which mechanically raised recorded numbers. We have not verified how the modelled series handles it, so we make no precise claim about the size of American change across those years.

The story we started with was: progress stopped, and someone is to blame.

The story the data tells is: progress did not stop, nobody is obviously to blame, and the world's headline number is being held back by a demographic shift rather than by any country getting worse.

That is a harder thing to put in a headline. It is also the more useful thing to know, because it changes what you would do about it — and it only shows up if you look at the denominator.