June 25, 2013

Error in population observations and population forecast errors

When compared with ex-post observed real trends, the twelve sets of populations projections that the United Nations prepared between the 1950s and the end of the 1980s show a clear tendency over time towards better accuracy. Part of this improvement is due to better data for base populations. The number of elderly persons for the world as a whole has been systematically underestimated by up to 8 percentage points in the base populations for the forecasts made in the 1960s and 1970s. This feature disappeared in the 1980s, when errors in the world's base population became as small as between -1 and +1 per cent in almost every five-year age group. Not only the base population, but also implied crude death rates have become more accurate: forecast errors for the CDR were particularly large in the 1950s and 1960s. On the other hand, errors in long term (10-15 years) crude birth rate forecasts were still at a relatively high level up to the period 1980-85.

1950 population observation and counts required many revisions over several years. They could not accurately count and estimate the number people in the world for the year the estimate was made.

Age structures in the former USSR and in Asia have been more difficult to predict than those in other regions. The fact that errors were smallest in Europe is mainly due to relative good data quality for the base year populations and not so much due to better assumptions concerning mortality and fertility. Indeed, average errors in European age structures caused by wrong assumptions alone are relatively high (1.5 per cent at a forecast duration of 10 years) - larger than those for the world as a whole (0.7 per cent). They are lowest in Africa (1.2 per cent) and Asia (1.3 per cent), and highest in India (2.8 per cent), Northern America, and the former USSR (both 2.4 per cent).

A first recommendation is that the UN consider to include more than one variant for mortality. A comparison between observed and projected life expectancies at birth shows that mortality assumptions have almost invariably been too pessimistic.

A second suggestion is to include base population variants for countries for which such data are not reliable. The forecasts since 1965 for which we have data showed considerable errors in the base population of Africa, Asia (including China, and, in particular, India), and the former USSR. Although the situation clearly has improved over time, the available data for a number of developing countries are still of poor quality.


Demographic Research is a peer reviewed and open access journal.

There is a roughly 270 page document that describes the methodology for population forecasting [A Guide to Population Projection - 2001].

The UN projects a single mortality path for all scenarios of its population projections to 2050. In the 1998 Revision, it assumes that all countries will eventually reach a maximum life expectancy of 87.5 years for males and 92.5 years for females. The pace of change is determined by assigning one of three models of change in life expectancy to each country (fast, medium, and slow change) based on recent experience and on the idea that improvements in life expectancy will slow as life expectancy grows. Countries may be assumed to switch from one model to another in 2025. No country actually reaches the maximum life expectancy by 2050, and only North America reaches this maximum by 2150 in the long-term projections, but it is the implicit end point of all paths.

A recent revision to the UN methodology was the addition of five-year age groups between 80 and 100 years of age. Previously, the oldest age group had been the open ended 80+ group. Currently, it is the 100+ age group. The addition of these groups was made difficult by the scarcity of mortality data for the elderly.

UN projections of life expectancy have changed little in recent years, with a slightly less optimistic outlook for LDCs. The projection of a rise in LDC life expectancy from its current level of about 62 years to 75.5 years in 2050 is about 1 year less than the rise to 76.4 years envisioned in the 1994 revision. This change is
primarily due to new data on the extent and implications of HIV/AIDS mortality.

Evaluating US census projections based on the 1990 census (by ESRI and US census department) against the 2000 census. Errors at the census track level were reduced from 14.5% level in 1990 to 6.8% in 2000. Population changes and population movement at the smallest level are tough to predict. Overall US accuracy was about 2.5% but was higher for smaller areas. Compared ESRI population estimates for the year 2000 in 1999 against the US census for 2000.

Economic uncertainty and family dynamics in Europe PUBLISHED 20 December 2012 The collection comprises studies on how various dimensions of employment uncertainty, such as temporary working contracts and individual and aggregate unemployment, are related to the fertility and family formation of women and men in contexts across Europe. It covers studies on Germany, the U.K., France, Russia, Estonia, Sweden, Italy, Spain, and Israel.

If you liked this article, please give it a quick review on ycombinator or StumbleUpon. Thanks
blog comments powered by Disqus