Newer UN projections: more of the error lies in the starting population
For the 2023 population, the starting count made 12% of the combined error sizes in the 1998 projection and 57% in 2022, on average across countries. This is a share, not the size of the error.
More
Even a short projection starts from an estimate of how many people are already alive. That estimate can later change. A short horizon leaves less time for errors in births, deaths and migration to accumulate.
The horizontal axis is the year of the projection, not the year of the population count. Each point is the mean share across the countries available for that edition; intermediate years are not additional forecasts.
- 1998 starting-error share
11.6%; 155 countries
- 2000 starting-error share
14.7%; 156 countries
- 2002 starting-error share
13.3%; 156 countries
- 2004 starting-error share
19.1%; 156 countries
- 2006 starting-error share
19%; 157 countries
- 2008 starting-error share
29%; 157 countries
- 2010 starting-error share
27.4%; 158 countries
- 2012 starting-error share
28%; 160 countries
- 2015 starting-error share
38.3%; 160 countries
- 2017 starting-error share
35.6%; 160 countries
- 2019 starting-error share
58%; 160 countries
- 2022 starting-error share
57.2%; 161 countries
|Starting error| / (|starting error| + |birth error| + |death error| + |net migration error|), then an unweighted country mean. Population at least one million; editions with at least 100 countries. Compared with revised 2023 estimates.
Parts can cancel out: this is not a share of the net population error. Country coverage and projection horizons differ by edition. The line is not strictly increasing and does not establish a causal effect of publication date.
A question for this picture
Does the pattern hold for the same countries and the same forecast horizon?