Outline — Most Countries Now Have Birth Rates Below Replacement

The skeleton. The prose is a rendering of this; if the prose drifts from the argument here, the prose is wrong.

Question from the owner's capture (bead datapressr-nh7): "Falling birth rates". Open mode: the data was found for this story, not taken from a DataPressr dataset. Source snapshot and choices: birth-rates-src/PROVENANCE.md and birth-rates-src/DATA.md. Revision 2, after the outline review (round 1: 13 corrections, recorded at the end).

The argument, in one sentence

Fertility below the replacement level of 2.1 children per woman has gone from a minority condition to the majority one (130 of 237 countries and territories in 2023, up from 60 in 1990), and most places already below it have kept falling.

What this story is about

What has happened to birth rates since 1950, how fast, and where: the count of places below replacement, the time each took to fall from five children per woman to below 2.1, the continued fall after 2010 in places that were already below 2.1, and the one large region where the fall is far from finished (sub-Saharan Africa). All charted numbers come from one source, UN World Population Prospects 2024 (WPP 2024).

It does not show why. The data are rates and counts; they cannot test any cause. Explanations are given in one short section, attributed to named sources and marked as their account. Not about projections (WPP's 2024–2100 numbers are not used), population decline, ageing, policy advice or the wrangling.

Argument, in order

The chart, first. Number of countries and territories (of WPP's 237) with a total fertility rate (TFR) below 2.1, 1950–2023, with a second line for below 1.4 (what the UN calls "ultra-low"). 4 in 1950 (the exception), 60 in 1990 (a large minority), 130 in 2023: more than twice as many, and more than half. Below 1.4: 5 in 1990, 43 in 2023.
- Population share, given only at the endpoints: places below 2.1 held 24% of the world's people in 1990 and 67% in 2023 (the UN's own Summary of Results, Box 2.2, says "two thirds"). Places below 1.4 held 1.9% in 1990 and 25% in 2023.
- Two giants drive the share: China (first below 2.1 in 1991) and India (2020) account for about 35 of the 43.6-point rise. Outside China and India, the share of people living below 2.1 went from 38% to 49%. This is why the count, not the share, leads, and why the title is about countries.
- Robustness, places of 1 million or more people only: 40 of 152 in 1990, 81 of 161 in 2023, so just over half. The count of 237 weights the Holy See and Saint Barthélemy the same as India, so this check is on the chart.
What these numbers are. TFR is the average number of children a woman would have if she lived through her childbearing years at the age-specific birth rates of that one year. It is a period measure, not the number any actual cohort of women had: when births are postponed, the period TFR falls even if women eventually have as many children. Replacement is about 2.1 where few children die young: the level at which each generation, without migration, is replaced by one of the same size; it is higher where child mortality is high. Robustness: counting places with WPP's net reproduction rate below 1 (which builds in each country's own mortality) also gives 130 in 2023, though not the same 130: two places differ at the margin each way, Indonesia and Myanmar in and Guadeloupe and the US Virgin Islands out (checked once on the full UN file; not reproducible from the committed extracts). WPP 2024 values for 1950–2023 are UN estimates, revised with each edition; 2024 onwards are projections and are not used. "Countries" means WPP's 237 countries and areas, including territories.
What it says.
- The world rate has more than halved: 4.85 children per woman in 1950, a peak of 5.31 in 1963, 3.31 in 1990, 2.25 in 2023 (the UN summary gives the same 3.31 and 2.25).
- How fast: about a generation, often less. Of the 37 countries with 1 million or more people in 2023 that went from a TFR of 5 or more to below 2.1 within the record, the time from the last year at 5 or above to the first year below 2.1 had a median of 31 years. Fastest: Iran, 10 (1989–1999); Singapore 12; China 19 (1972–1991); South Korea 20 (1964–1984). Slower: Brazil 33, Mexico 38, India 43 (1977–2020), Sri Lanka 51. Limits, stated: (a) the rule counts places that have ever been below 2.1; 8 of the 37 later went back above 2.1 for at least a year (Mongolia, Tunisia, Mauritius, Viet Nam, Azerbaijan, Kuwait, Bahrain, Türkiye; Mongolia is at 2.69 in 2023). They are marked on the chart. (b) Places that have never been below 2.1 are excluded: Bangladesh (2.16 in 2023) has been below 5 since 1988. So 31 years describes falls that reached 2.1, not a forecast. (c) Places already below 5 in 1950 (all of Europe, the US, Japan) cannot be timed: the record starts too late, so there is no comparison with Europe's own historical fall. (d) Later starters show only a weak tendency to fall faster (correlation of start year with duration, r = −0.22): the 1960s starters include both Singapore (12) and Sri Lanka (51).
- It has not stopped at 2.1. Of the 103 places below 2.1 in 2010, 87 were lower still in 2023, and 72 of those 87 were already lower by 2019, before COVID. Named countries, 2010 → 2023: South Korea 1.23 → 0.72 (0.720: among the lowest anywhere; only Macao, 0.662, and Hong Kong, 0.717, are lower); China 1.69 → 1.00 (−0.69); Chile 1.84 → 1.17; Finland 1.86 → 1.28 (−0.58); Norway 1.94 → 1.40; Sweden 1.97 → 1.43; France 2.02 → 1.64; United Kingdom 1.92 → 1.56; United States 1.92 → 1.62 (−0.29); Italy 1.44 → 1.20; Japan 1.36 → 1.21; Brazil 1.79 → 1.62. The US, Norway and Finland had each stayed within a narrow band through 1990–2010 (US 1.92–2.10; Norway 1.76–1.97; Finland 1.70–1.86), which is what makes their fall after 2010 notable. (Sweden, 1.50–2.12, and France, 1.66–2.02, moved more, so they are not called "stable".) Two countries that were above 2.1 in 2010 crossed it since and are shown separately on the chart: Mexico 2.34 → 1.91, India 2.60 → 1.98.
- Counter-numbers: 16 of the 103 did not fall. Most are in central and eastern Europe (Hungary 1.26 → 1.49, Bulgaria, Romania, Serbia, Slovakia, Bosnia and Herzegovina, Montenegro, Moldova, Armenia), plus Germany (1.40 → 1.44), Portugal, Singapore, Viet Nam, Liechtenstein, the Falkland Islands and the Holy See (a constant modelled 1.00). Hungary and Germany are on the chart. This data cannot say why Hungary rose. The World Bank's figures, from national statistics, have Hungary at 1.55 in 2023 and 1.41 in 2024, so the rise may not be holding; the story says that and attributes no cause.
The wrinkle — where the fall is far from finished. Sub-Saharan Africa's TFR was 4.32 in 2023 (world 2.25), down from 6.30 in 1990: falling, but from higher and later. Its share of the world's births: 9.1% in 1950, 15.5% in 1990, 30.4% in 2023. Part of that rise is simply its growing share of the world's people: 9.4% in 1990, 15.0% in 2023. Its share of births grew faster (×1.96) than its share of population (×1.59), and in 2023 is about twice its population share. Eastern and South-Eastern Asia went the other way, 29.9% → 15.7% of births, 1990–2023, roughly halving. Central and Southern Asia (which includes India) held at about 30% (29.5% → 29.6%).
Outside context: why, according to others. Not tested by this data; one or two sentences each, attributed:
- The long fall from five children to two: Max Roser, "Until the late 1960s, the total fertility rate was five — since then, it has halved" (Our World in Data, 2019): the rate "has steeply declined as a result of women's empowerment, declining child mortality, and the rising cost of bringing up children".
- Low-fertility countries now: UN Population Division, WPP 2024 Summary of Results, Box 2.2: in several low-fertility countries women have fewer children than they had expected; obstacles named include "demands of higher education, high costs of childcare, challenges to work-family balance, unequal division of household tasks between partners, care responsibilities for ageing parents and biological limits to the reproductive lifespan".
- United States: Kearney, Levine and Pardue, "The Puzzle of Falling US Birth Rates since the Great Recession" (Journal of Economic Perspectives, 2022): "The Great Recession contributed to the decline in the early part of this period, but we are unable to identify any other economic, policy, or social factor that has changed since 2007 that is responsible for much of the decline beyond that." They conjecture "shifting priorities" of more recent cohorts.
- Nordic countries: Hellstrand and colleagues, "Not Just Later, but Fewer" (Demography, 2021): postponement (the period-measure effect in beat 2) "explains only part of the decline"; they forecast completed family size falling from 2 children for women born in 1970 to around 1.8 for those born in the late 1980s.
How this was made. WPP 2024 medium-variant demographic indicators (one CSV), snapshotted by birth-rates-src/fetch.mjs into two small extracts; charts by birth-rates-make-charts.mjs; a World Bank WDI cross-check (18 places: 16 countries, the world and sub-Saharan Africa) agrees with WPP to within 0.07 on every value checked (largest gap 0.064, Hungary 2023). Licence CC BY 3.0 IGO.

Numbers that weaken or limit the argument (collected)

The population share is lumpy: China (1991) and India (2020) account for about 35 of its 43.6-point rise; outside them the share went from 38% to 49%.
The count of 237 weights tiny territories equally; for places of 1 million or more it is 81 of 161, only just over half.
India is only just below: 1.98 in 2023. Mexico 1.91, Viet Nam 1.91. A small revision could move large countries back above 2.1, and 8 of the 37 "completed" falls have already gone back above it at least once.
2023 values are UN estimates and differ from national figures by up to 0.064 in the World Bank check (Hungary 1.49 in WPP, 1.55 in WDI; Germany 1.44 vs 1.39).
Some of the post-2010 fall in period TFR is postponement, not fewer children in the end (Hellstrand et al.: "only part").
Not every country fell after 2010: 16 of 103 did not, mostly in central and eastern Europe; Hungary rose. The World Bank's 2024 figures, outside this story's window, show South Korea up slightly (0.72 → 0.75) and Hungary down (1.55 → 1.41).
Speed figure (31 years) covers falls that reached 2.1 only.
Sub-Saharan Africa's rising share of births is partly its rising share of population (9.4% → 15.0%).

Reader questions

Before writing the beats: what a curious reader brings, and where it is answered.

Are birth rates really falling worldwide, and by how much? — beat 3 (4.85 → 2.25).
How many countries are now below replacement, and what is "replacement"? — beats 1–2.
How fast did this happen, and is that faster than Europe was? — beat 3, chart 2; speed answered, the Europe comparison explicitly not made (record starts too late).
Which countries are lowest? — chart 3 (South Korea 0.72; Macao and Hong Kong lower).
Did it level off once countries reached 2.1? — beat 3, chart 3.
Has anywhere reversed it? — beat 3 counter-numbers: 16 of 103 did not fall; Hungary rose, with no cause claimed and the 2024 figure noted.
Where are birth rates still high? — beat 4, chart 4.
Why? — beat 5, attributed only, covering both the long fall and the recent one.
Is this a COVID or recession blip? — partly: 72 of the 87 were already lower by 2019, before COVID; Kearney et al. put part of the US fall on the Great Recession and most of it on something they cannot identify.
Will populations shrink? — not answered; out of scope (projections not used). Said in one line.

Chart plan

All from birth-rates-src/wpp2024-countries.csv (iso3, location, year, tfr, population_thousands; 237 places × 1950–2023, no gaps) and birth-rates-src/wpp2024-regions.csv (location, year, tfr, births_thousands, population_thousands; World + 8 SDG regions × 1950–2023). Annotations read their values from the rows at build time; the build fails if any SVG contains NaN, Infinity, undefined or a stringified function.

#	Chart	Data	Transform, gaps, dates	Purpose
1	Two lines, 1950–2023: count of places with TFR < 2.1, and with TFR < 1.4 (y: number of the 237 countries and territories). Points at 1990 and 2023 labelled with the count and population share ("60 places, 24% of world population"; "130, 67%"; below 1.4: "5", "43, 25%"). Dashed reference at 118.5 (half of 237). A note in the empty upper left: places of 1 million+ people, 40 of 152 (1990) and 81 of 161 (2023); outside China and India, 38% → 49% of people.	countries CSV: `year`, `tfr`, `population_thousands`	Per year: count `tfr < 2.1` and `tfr < 1.4` over all 237 rows (every place has every year, so the denominator is constant). Population share = sum of `population_thousands` where below / sum over all 237, same year. 1m+ check: places with `population_thousands ≥ 1000` in that year. Outside China and India: the same share with `CHN` and `IND` removed from numerator and denominator. Strict `<`. y-domain 0–237. Direct labels, no legend.	Beat 1: below replacement went from a minority to the majority, with the robustness checks visible.
2	Range chart, one row per country (37 rows): a bar from the last year at TFR ≥ 5 to the first year below 2.1, labelled with the years taken; Iran, Singapore, China, South Korea, Brazil, Mexico, India and Sri Lanka highlighted with their start and end years, the rest grey; the 8 that later went back above 2.1 marked with an asterisk and a note; "37 countries … median 31 years" in the header.	countries CSV	Countries with `population_thousands` ≥ 1000 in 2023. `y21` = first year with `tfr < 2.1`; `y5` = last year before `y21` with `tfr ≥ 5`; keep only countries with both. Duration = `y21 − y5`. "Went back above" = any later year with `tfr ≥ 2.1`. Sort by duration. x = calendar year 1950–2023.	Beat 3, speed: about a generation, often less, with the range and the caveat shown.
3	Dumbbell, one row per country: TFR in 2010 (grey dot) and 2023 (red if lower, blue if higher), joined, with both values labelled; a vertical rule at 2.1; change labelled at the right (e.g. "−0.51"). Two groups: 14 countries below 2.1 in 2010 (sorted by 2023 value), then, under a separator labelled "Above 2.1 in 2010", Mexico and India. Header note: "87 of the 103 places below 2.1 in 2010 were lower in 2023; 72 of them already by 2019".	countries CSV	Group 1: South Korea, China, Chile, Italy, Japan, Finland, Norway, Sweden, Germany, Hungary, United Kingdom, Brazil, United States, France (the countries the question names, the largest Latin American economies, and two of the 16 that did not fall). Group 2: Mexico, India. 87/103/72 counted over all 237 with `tfr < 2.1` in 2010. x-domain 0.5–2.85 (includes 0.72 and 2.60 and room for labels).	Beat 3: the fall did not stop at 2.1; Hungary and Germany shown as counter-cases.
4	Lines, 1950–2023: each region's share of world births (%). Sub-Saharan Africa highlighted (red), with 1950, 1990 and 2023 shares labelled; Eastern and South-Eastern Asia (dark) labelled at 1990 and 2023; other regions in grey with end labels. Australia/New Zealand and the rest of Oceania summed as "Oceania". A note: "Sub-Saharan Africa in 2023: 4.32 children per woman (world 2.25); share of world population 9.4% in 1990, 15.0% in 2023".	regions CSV: `births_thousands`, `population_thousands`, `tfr`	Share = region `births_thousands` / World `births_thousands`, same year; SSA population share = region `population_thousands` / World, same year. The eight SDG regions sum exactly to World for births and population (checked at build, tolerance 0.1%). y-domain 0–40% (East and South-East Asia peaks at 39.4% in 1963). End-label collisions resolved by nudging labels, not data.	Beat 4: where the fall is far from finished, and what that means for where babies are born, with the population denominator shown.

Rendered by birth-rates-make-charts.mjs with Observable Plot per skills/story/references/charting.md. Build twice, compare SHA-256; check no NaN or Infinity in any SVG.

Voice

Plain and factual, per the voice guide. No "crisis", "collapse" or "demographic time bomb": the numbers carry it. State differences and ratios (more than twice as many; about twice its population share; −0.58) rather than leaving two numbers side by side. A "sounds like me" pass is a separate step the author runs.

Outline review

Round 1 — independent AI reviewer (Claude subagent, general-purpose, no hand in the outline), revision 1, SHA-256 b0ec3fc3201be8392115790e330a3284928acce8ed764f49829cc73fa24aa098, 10 October 2026. Not approved, 13 corrections. Every count, share, speed and 2010 → 2023 value reproduced. Corrections, all applied in this revision: Korea is not the lowest of 237 (Macao, Hong Kong); chart 4 y-domain clipped East Asia (39.4% in 1963); NRR check gives 130 but not the same places; "exception to the norm" overstated 1990 and "so that" implied a causal link (argument sentence rewritten, SSA population share added); title led with the lumpy population share (retitled on the count; China and India's part stated); speed analysis needed the 8 that went back above 2.1 and Bangladesh's 1988; Mexico and India were not among the 103; "the two that did not fall" (16 did); "stable" applied only to the US, Norway and Finland; Kearney et al.'s Great Recession sentence restored; the UN "girls' education" line misrepresented policy advice as explanation (replaced by an attributed OWID account); Hungary's rise needs a no-cause line; World Bank gap is 0.064; 1m+ robustness count added; Singapore and Sri Lanka highlighted; "no pattern" softened to a weak tendency (r = −0.22); 72 of 87 already lower by 2019 added.
Round 2 — same reviewer, revision 2, SHA-256 15370d1c4c5dbaa3c3c8d8fa49e8bf21dda6ffe43affd904f5804bd775241000, 10 October 2026. APPROVED. All 13 corrections verified against the data; new numbers reproduced (below-1.4 share 1.91% → 25.11%; 1m+ 40/152 → 81/161; outside China and India 38.12% → 49.36%, China and India contributing 35.35 of the 43.6-point rise; 72 of 87 lower by 2019; 8 of 37 back above 2.1; SSA population share 9.40% → 14.98%; the 16 non-fallers match, 9 in central and eastern Europe); OWID quote verbatim. Two non-blocking notes: keep the OWID line framed as Roser's summary; do not set the World Bank's 2024 Hungary figure next to WPP's 1.49. This line was added after approval; nothing else changed.

Friction notes

Open mode: no DataPressr dataset existed, so the story snapshots its own source extract. The UN file is 16.6 MB gzipped; only two extracts (about 0.7 MB) are committed and the fetch script checks the original's SHA-256.
The headline measure was chosen after seeing that the population-weighted share moves in two steps (China, India): the count is the steadier measure and leads, the share is reported only at the endpoints. Worth a line in story-craft: a population-weighted share can be driven by one or two giants; say so, and give the share without them.
A count of "countries" over WPP's 237 weights the Holy See like India. A 1-million-people floor is a cheap robustness check worth making routine for country counts.
The "historical Europe was slower" comparison the brief suggested is not supported by this record (it starts in 1950, after most of Europe's fall). Dropped rather than imported from a second source with a different start point.
Observable Plot's dx, textAnchor and fontWeight are constants, not channels: passing a function writes the function's source into the SVG (one chart became invalid XML). The build now refuses any SVG containing =>.

Outline: Most Countries Now Have Birth Rates Below Replacement