How population is measured, estimated, and projected.
The methods behind the numbers — from archaeological reconstruction to UN modelling — and where confidence breaks down.
Terms in this category
The steady yearly growth rate that would carry a population from its start value to its end value over a given period.
CAGR = (end / start)^(1/years) − 1. Unlike a simple average rate, CAGR accounts for compounding. It is useful for comparing growth across countries and periods of different lengths.
Population figures pieced together from historical census records and administrative data, roughly 1800 to 1949.
This regime fills the gap between pre-modern reconstruction and modern UN estimates. Sources vary by country — some have near-complete census records from the 19th century; others rely on interpolated estimates. Gapminder v7 is our primary source for this era.
The UN's central scenario for future population, derived from modelled fertility, mortality, and migration assumptions.
The UN publishes several variants (low, medium, high, and constant-fertility). The medium variant is the most-cited. It is not a prediction — it is the outcome of a specific set of demographic assumptions, and uncertainty grows with distance from today.
A population estimate built from indirect evidence for periods before systematic records existed.
Archaeological data, land-use models, pollen counts, and written chronicles are combined to reconstruct plausible population levels. HYDE 3.3 uses land-use change as its primary proxy. Uncertainty ranges are wide — often ±30% or more for ancient periods.
A new release of a dataset that replaces earlier figures, often with updated methods, new censuses, or corrected errors.
Dataset revisions can shift historical numbers significantly. When a new UN WPP edition is released, past figures are not frozen — censuses that arrive after a previous edition can revise earlier estimates. We track the revision year for each data point.
The year where one source dataset ends and the next begins; methodology changes at this boundary.
We have two seams: circa 1800 (HYDE to Gapminder) and 1950 (Gapminder to UN WPP). Values just before and just after a seam may differ slightly because of different methodologies, not a real change in population. Seam lines are marked on every chart.
A shaded region around a line that shows the range in which the true value likely falls.
For projections, the UN publishes 80% and 95% prediction intervals. For reconstructions, HYDE provides implicit uncertainty through alternative scenario files. We encode confidence visually through line weight and hatch patterns rather than explicit bands in V1.
Common questions
Why are historical population figures sometimes revised significantly?
New archaeological discoveries, digitised census archives, and updated modelling methods can shift estimates substantially. A census that was previously unavailable might reveal that earlier figures were 10–20% off. Revisions are more common and larger for developing regions where historical records are sparse.
How far can projections be trusted?
Short-term projections (5–10 years) are generally reliable because most people who will be alive then are already born. Long-term projections (50+ years) are highly sensitive to fertility assumptions. The UN publishes wide prediction intervals — by 2100, world population could plausibly range from around 7 billion to over 16 billion. The full probabilistic range is available on the UN WPP probabilistic projections page.
What is the difference between a projection and a forecast?
A projection shows what happens under specific stated assumptions (e.g. "if fertility follows this trajectory"). A forecast is a best-guess prediction of what will actually occur. UN population projections are technically projections, though the medium variant is often used as a de facto forecast.
Why does the chart show two different data regimes for before and after 1950?
Before 1950 we use Gapminder v7 (compiled long-run history) and HYDE 3.3 (reconstruction). From 1950 onward, the UN World Population Prospects provides harmonised, consistently-methodologised estimates based on national censuses and surveys. The methodological shift at 1950 means the seam line is especially meaningful.
What does CAGR tell me that a raw population change number doesn't?
Raw change is affected by time period length. A country that grew by 20 million over 10 years and one that grew by 20 million over 30 years look the same in absolute terms, but CAGR reveals the first grew roughly three times faster. CAGR enables apples-to-apples comparison across different time windows. Our World in Data's population growth explainer shows CAGR comparisons across countries and regions.