Dylan Slagh
Conceived the question and 2150 horizon, chose the scientific architecture, adjudicated modelling decisions and reviewed the analysis and manuscript.
A 237-country demographic model · working paper v1.2.2
Population forecasts forget who has the children.
Families differ in size, and part of that difference persists between generations. A national fertility average cannot represent the resulting change in population composition. This project puts that missing mechanism inside a full cohort-component model.
The finding
Who has the children adds 1.8 billion people.
Hold fertility low and flat and the model reaches 8.54 billion people. Now allow what is actually true — families differ in size, and larger families tend to run in families — and the same model reaches 10.36 billion. That is 21% more people, and nothing in it assumes fertility bounces back.
Fertility low and flat
8.54 bn
Who has the children
1.82 bn
This model
10.36 bn
World population, 1950–2150
The UN record, its medium projection, this project's post-2100 extension, and the paper's selection benchmark.
- UN estimates
- UN projection
- Post-2100 extension
- Selection model
- Extension range
Show the numbers
Watch
150 years of population change in 106 seconds.
Twelve countries race from 1950 to 2100. When the video reaches 2024, history becomes projection and the uncertainty appears.
Watch on YouTube ↗
The bars follow the UN. The uncertainty ranges come from the University of Washington's Bayesian projection.
Why the average misses it
An average hides who is having the children.
Two countries can share the same average family size and be nothing alike underneath. In one, nearly everyone has two children. In the other, many people have none and some have five — and the next generation is drawn from the ones who had five. Across 19 low-fertility countries — 29.1 million women who have finished having children — family size varies by 0.57 of the average, and how many children people have tracks their parents at 0.15: weak, but not zero.
01
Variation
Some families have more children than others.
02
Persistence
Children resemble their parents imperfectly.
03
Composition
Higher-fertility lineages become more common.
The useful answer
How much further would fertility have to fall to erase it?
Nobody knows where fertility finally settles, so the model does not guess. It asks a sharper question instead: how much further would fertility have to keep falling for those extra people never to arrive? After 2050, the answer is 1.52% more decline per decade.
1.52%
per decade
1.51%–1.53% across 50 paired migration paths
How much decline it would take
The two things we measured could each be higher or lower than our best estimate. Every dot is one plausible pair, placed at the answer that pair gives.
Show the numbers
Nobody knows yet which side of that line the world is on. The point is that it is now a number someone can go and measure, instead of an argument about the far future.
How the paper was made
One question, three authors, a complete record.
The project was developed through 15 recorded conversations over 15 days, and every one of them is published. Dylan set the scientific direction and accepted responsibility for the result; Claude and ChatGPT made substantive contributions to the model, evidence and paper.
Claude Opus 5
Built the cohort-component engine, backtest, Bayesian baseline, selection mechanism, stochastic extension and first complete manuscript.
ChatGPT 5.6 Sol
Developed the hierarchical selection specification, parameter calibration, migration and boundary analysis, then independently reviewed and edited the final paper.
The work itself
The model is open. The argument is testable.
The paper gives the derivation and substantive case. The supplement carries the engine validation, parameter audit, historical backtest, and reproducibility record. Every number on this page is checked against the same result files used to build both documents.