Beyond Six Billion Forecasting the World's Population (2000) / Chapter Skim
Currently Skimming:
The Accuracy of Past Projections
The Accuracy of Past Projections
Pages 37-52
The Chapter Skim interface presents what we've algorithmically identified as the most significant single chunk of text within every page in the chapter.
Select key terms on the right to highlight them within pages of the chapter.
From page 37... ...
and World Bank projections, some from as far back as the 1950s, which are assessed against past population trends, also estimated by the U.N.
|
From page 38... ...
The positive and negative errors in country projections partly cancel one another, in all these cases, when world totals are calculated. Second, country projections themselves benefit from the fact that populations do not turn over quickly; at current life expectancy levels, most people alive at a given date are still alive three or four decades later.3 Finally, demographers have become reasonably proficient at predicting broad trends in fertility and mortality, the key determinants of future population size.
|
From page 39... ...
An alternative summary measure can be calculated by averaging percentage errors, allowing positive errors and negative errors to cancel each other. The resulting average percentage error can be called bias: a positive value indicates that the projections were too high on average, and a negative one indicates they were too low.4 Estimates of the biases in past 4The literature often applies the term "mean percentage error" (MPE)
|
From page 40... ...
These bias estimates are similar in magnitude to the errors in the world totals plotted in Figure 21. The agreement between country biases and world errors would have been exact if every country had the same population size, but that is not the case.5 Error in world population projections is influenced less by error in small countries and more by error in large countries, such as China, India, and Nigeria.
|
From page 41... ...
Whichever view is correct, projection error can be shown to vary in systematic ways, being sensitive to specific factors.6 Projection Length The mean absolute percentage errors for countries in Figure 2-2 indicate that more recent forecasts, which involve a shorter time horizon, have been more precise than those further in the past. A similar pattern of increasing precision can be observed if one compares projections to other years 1995, 1990, 1985, and so on.
|
From page 42... ...
Figure 2-3 plots mean absolute errors by projection length for all countries, as well as for countries in six world regions. As expected, the magnitude of these absolute percentage errors increases steadily as the projection interval lengthens.
|
From page 43... ...
In general, the contribution of baseline error to the total error declines as projection length increases, and in the long run the length effect dominates.9 9Base population error accounts for most of the error in short projections. In a 5-year projection, at least 60 percent of the bias in the projected population can be explained by base population bias.
|
From page 44... ...
Since these huge countries account for about 60 percent of world population, this relative accuracy contributes substantially to the accuracy of world projections. This is shown in Appendix B
|
From page 45... ...
population projections for Europe and Northern America combined illustrate the concentration of error at extreme ages (Figure 2-4~. Forecasts dating back to 1965 have overprotected the number of children in these regions, and the upward bias has been greater the longer the projection.
|
From page 46... ...
forecasts from 1965 to 1990, following Keilman (1998~. PROJECTED COMPONENT RATES Besides being due to inaccurate baseline data, errors in projections of population size and age structure are also due to misspecified trends in fertility, mortality, and migration.
|
From page 47... ...
In this region, weak economic growth, slow development of health systems, and the HIV/AIDS epidemic may all have contributed to slower than expected mortality declined 13If the former Soviet Union were treated as a region, it would also be an exception. Life expectancy has stayed in a narrow range between 67.5 and 69.5 since 1970, instead of increasing, as all projections since then have assumed.
|
From page 48... ...
Misspecification of fertility accounts for three to four times as much projection error as does misspecification of mortality. Migration error explains as much or more projection error as fertility error, although its effect is relatively constant across Projection lengths, unlike the increasing effect of fertility error.l5 - r -- ~Misspecified trends in fertility, mortality, and migration are visible as errors in different parts of the projected age structure.
|
From page 49... ...
Component trends appear to have been more accurately specified when data were more available when recent surveys were available to provide total fertility rates, or recent censuses from which one could estimate life expectancies. A second factor is the level of the component rates in question.
|
From page 50... ...
To be equally accurate, however, such direct world projections would have to have substantially less proportional error than typical country projections, just as projections for the largest countries have less proportional error than projections for the smallest countries. Presumably, therefore, the base data for the world as a whole would have to be better, or the prediction of future trends in component rates more precise, than for the largest countries.
|
From page 51... ...
Practice in this regard varies among international agencies making world projections,~7 but continual updating does appear to be possible. The fastest possible updating schedule, consistent with thorough examination of reported data, is clearly to be preferred and should be coupled with rapid dissemination of basic results, perhaps over the Internet.
|
From page 52... ...
13~3~:245-277. 1998 How accurate are the United Nations world population projections?
|
Key Terms
This material may be derived from roughly machine-read images, and so is provided only to facilitate research.
More
information on Chapter Skim is available.