Future trends in population size, age structure, births, and other demographic variables are of interest to a wide range of analysts, including policymakers, scientists, and planners in industry and government. For example, global and national trends in population size are needed to project the future demand for food, water, and energy and the environmental impact of rising consumption of natural resources. Subnational projections help planners decide where to build new schools and where investments in roads and other infrastructure are required. Reliable estimates of the number of retired people in need of pensions and health care are essential to the optimal design of social security systems.
To address the needs of such a variety of potential users, global as well as national population projections for all countries have been produced in recent decades by various agencies, such as the United Nations (U.N.) Population Division, the World Bank, and the U.S. Census Bureau. (The International Institute for Applied Systems Analysis has also made world and regional but not country projections.) The Panel on Population Projections was asked to examine these projections: to assess their assumptions, estimate their accuracy and uncertainty, evaluate the implications of current demographic research for projection procedures, and recommend changes where appropriate as well as research that might improve projections. Generally, the panel finds current world projections up to 2050 to be plausible, although they could indeed be improved in some ways, and their uncertainty deserves more precise quantification.
We expand on this broad conclusion below, summarizing, in order, what current projections say about future population trends and how
their conclusions were arrived at; how accurate such projections have been in the past; how the projected components of population growth—fertility, mortality, and migration—compare with historical trends; and what degree of uncertainty should be attached to these forecasts. Then we detail our conclusions and suggest how research might improve population projections.
CURRENT WORLD PROJECTIONS
From the 6 billion that had been reached by the end of 1999, world population is now projected to approach 9 billion by 2050 (see Chapter 1). This increase of 3 billion in the next 50 years will be only slightly smaller than the increase of 3.5 billion in world population in the past 50 years. Beyond 2050, forecasts involve so much uncertainty that we do not examine them.
Nearly all world growth to 2050 is projected to occur in developing regions, that is, in Africa, Asia, and Latin America. The population of the industrial (or more developed) world is expected to remain close to its current size, and, in some countries, population is likely to decline. Thus the distribution of world population will shift significantly. Over the next 50 years, the share of world population in Sub-Saharan Africa, in particular, will rise from 10 to 17 percent of the world total, while the share in Europe will decline from 13 to 7 percent. The population in industrial regions as a whole, now outnumbered almost 4 to 1 by the population in developing countries, will be outnumbered by about 7 to 1 by the year 2050.
The projections also indicate a rise in the proportion of people aged 65 and older in every major region of the world. In the next half-century, the proportion of the population aged 65 and older is expected to rise from 5 to 15 percent in the developing regions, and from 14 to 26 percent in the industrial regions.
Future population growth and changes in the age composition of the population are determined by levels and trends in the following factors:
Fertility. In most developing regions, the number of births per woman is still well above the level required for each generation to exactly replace itself (about 2.1 births per woman). This is a key reason why population growth in the developing world continues. Birth rates in these regions are projected to decline to around the replacement level over the next few decades. In the industrial regions, women are on average having fewer than two children each. If this low level of childbearing is maintained in future decades, declines in population size will occur unless the deficit in natural growth is offset with a flow of immigrants.
Mortality. Life expectancy levels have risen worldwide for a long period and are projected to continue to do so, adding somewhat to future population growth and substantially to population aging (which is also accentuated by falling birth rates).
Migration. Movement of people between countries has no direct effect on world population growth, but it affects growth in particular countries and regions. Net migration levels are projected to continue to vary across countries, slightly retarding shifts in the regional balance of populations.
Age distribution. Even if the number of births per woman stayed at the replacement level of around two and future mortality levels were unchanging, world population would still grow, because of the high ratio of young to old people. Half the world's population is under age 27, because high fertility and low mortality in recent decades have increased the young population in particular. Growth caused by a youthful age structure is known as population momentum and is expected to account for more than half of world population growth to 2050.
Fertility, mortality, and migration constitute the components of population growth. U.N. and other forecasters determine levels and the likely future path of each component, combine these with information on the existing distribution of the population by sex and age (or birth cohort), and then, through extensive though straightforward calculations called the cohort-component method, project future populations.
Projections are inevitably uncertain. The present demographic situation is not known perfectly, and future trends in births, deaths, and net migrants are subject to unpredictable influences. At the start of the 20th century, forecasters would have had difficulty foreseeing such technological achievements as the development of antibiotics or such social trends as women's increased participation in the labor force. Many other social, economic, political, technological, and scientific developments have influenced population growth by affecting birth, death, or migration rates. Growth has also been influenced by deliberate social policy, such as decisions about public health services, policies affecting the availability of family planning methods, and regulations on immigration. Policies themselves may result from consideration of population projections, which complicates the attempt to make accurate forecasts.
ACCURACY OF PAST PROJECTIONS
The accuracy of current projections cannot be directly evaluated, but older global and country-level projections can be assessed against current estimates (see Chapter 2). For instance, the U.N. has been making projec
tions since the 1950s of world population size for the year 2000. These projections have almost all been off by less than 4 percent.
Errors in past projections of the population for specific countries have typically been larger. Across several sets of U.N. and World Bank forecasts, absolute error in projected country populations averaged 4.8 percent in 5-year projections but 17 percent in 30-year projections. As these figures suggest, projection error increases systematically as the projection interval lengthens. Other factors affecting projection accuracy include level of development and size. Errors have been larger for developing countries than for industrial countries, and for smaller countries (especially those under 1 million) than for larger countries.
Country projections became progressively more accurate during the 1950s and 1960s, as demographic data for developing countries improved. Since then, no significant further improvements can be demonstrated. Better data quality played a large role in the earlier improvements. Erroneous estimates of initial population, fertility, mortality, and migration are the dominant cause of error in projections up to 10 years long, although longer projections are more sensitive to misspecified trends in population growth components.
Error in projecting a country's total population is generally accompanied by errors in regard to the sizes of particular age groups. Projections of the youngest and the oldest age groups tend to be the least reliable. In the past, these errors have been the result of too high projections of fertility (resulting in too many infants and young children) and too high projections of mortality (resulting in too few elderly).
Current projections are not necessarily subject to the same errors as past projections. Past forecasts, for instance, produced slightly too high world projections for 2000, due mainly to larger than expected fertility declines in a few major countries. Such unexpected fertility declines will never be exactly replicated, and future demographic conditions generally are likely to diverge from conditions that prevailed during the periods covered by past forecasts.
Recognizing this limitation on the conclusions that can be drawn from reviewing past forecasts, the panel also reviewed historical trends and current levels for each component of population growth —fertility, mortality, and migration—in order to determine what inferences might be drawn about likely future trends.
Fertility in developing regions is in transition from high to low levels (see Chapter 3). In the 1950s, the average woman in the developing regions of Africa, Asia, and Latin America gave birth to about six children. By the early 1990s, this average had fallen to 3.3 births, with considerable variation across regions. In Sub-Saharan Africa, the average number of births per woman was still 5.9, but in East Asia, it had fallen to 1.9.
Fertility has declined because married couples have decided to limit births. These decisions are implemented primarily by adopting contraception, but substantial numbers of women also rely on induced abortion to prevent unwanted births. Couples want smaller families for several reasons: children have become less household assets than household liabilities and require increasingly costly parental investments; the idea of and the means for birth control have been increasingly acceptable after diffusion through media, health services, and interpersonal channels; and the odds of child survival have substantially improved, so that fewer births are needed to achieve a desired family size. The relative importance of each factor in couple decisions has been much debated, but that each at least plays some role is broadly accepted.
Assuming that such factors will continue to operate in developing countries, forecasters have projected continuing declines in fertility. This trend is supported by the record of recent decades, which indicates increasingly widespread fertility change. Fertility has dropped in poor, largely illiterate, agrarian populations to a degree that, 30 years ago, most experts would have thought highly unlikely. Once initiated, this decline spreads rapidly within countries and even across national borders when countries are linked by geography, culture, and trade.
The assumption that forecasters make of continued fertility decline therefore appears sound. In fact, forecasters' projections of fertility decline in the 1970s, 1980s, and 1990s have generally fallen a little short of actual declines. In particular, forecasters missed two historical turning points at which fertility in developing regions fell faster than expected. The forecasts of the 1970s missed the sharp downturn in fertility in China between the late 1960s and the early 1970s. These forecasts and later ones as well also missed an acceleration in fertility decline in developing regions from the late 1980s to the early 1990s, attributable mainly to trends in China, Bangladesh, and India.
What pace of fertility decline can be expected for the future? The record of recent decades indicates that the pace of decline generally slows as countries attain lower levels of fertility. A slower pace can also be
expected because countries that are still in transition from high to low levels appear self-selected for slower decline. This is especially the case for countries—mostly in Sub-Saharan Africa—that have not started their transition. But if fertility decline may slow slightly in the aggregate, it will certainly continue to spread.
Once fertility reaches low levels—2.5 births per woman or below—where will it head? This question is immediately relevant for demographically advanced developing countries as well as for all industrial countries, which are already at these levels. This group of countries includes fully 50 percent of current world population. In fact, 15 percent of world population lives in countries where fertility is already below 1.8 births per woman (see Chapter 4).
Countries have achieved low fertility primarily by eliminating higher-order births. Third births, fourth births, and still higher-order births have diminished rapidly, while first and second births have been reduced much less. The reduction in higher-order births is likely to be permanent, since most couples in modern settings have few good reasons for planning large families.
Among the main factors responsible for low fertility are socioeconomic changes that have expanded the options for women that compete with motherhood, thus raising the opportunity costs (that is, the opportunities forgone) in having children. A related consequence is the continuing “retreat from marriage,” which has led to falling marriage rates and rising rates of separation and divorce.
Another factor contributing to low fertility is the ongoing postponement of births to later ages. Mean ages at childbearing have risen substantially in most industrial countries since the 1970s. This postponement of childbearing reduces current numbers of births; the fertility effect is temporary and will disappear when the average age at childbearing stops rising. It is not possible to predict when this will happen in specific countries, but when it does, observed fertility could rise modestly.
Public policy could affect fertility trends, particularly as the demographic consequences of low fertility become increasingly evident. For instance, current projections show that the population of Italy will fall 30 percent in 50 years, and other effects of low fertility, such as the aging of the population, are already highly visible. Such potentially important demographic changes could elicit policy initiatives or popular movements to raise fertility. For example, new governmental programs or incentives could make it easier for women to combine childbearing with the pursuit of advanced education and careers. In contemporary low-fertility set
tings, young women on average want two children. If policy or social changes removed the obstacles they face in implementing their preferences, future fertility could move closer to replacement.
At what level will future fertility settle? It is unlikely to be the same in all countries, or to settle at a single level in any country for an extended period. Variation and fluctuations should be expected. The average level around which fertility will vary is largely indeterminate but will probably be around or somewhat below two births per woman in most countries.
Mortality is declining in most countries, propelling life expectancy to higher and higher levels (see Chapter 5). For example, at 81 years, life expectancy in Japan is higher than it has ever been in any country. But some countries remain like Malawi, with a life expectancy, at 39 years, only half as high as in Japan.
The industrial countries attained their unprecedented levels of life expectancy through a centuries-long transition from high to low mortality. Before the transition started, life expectancy rarely exceeded 40 years, and year-to-year mortality would fluctuate sharply. With progress in standards of living, nutrition, medicine, and public health, annual fluctuations in mortality were gradually reduced, and life expectancy began a long, steady rise.
Most developing countries are working through the same mortality transition process experienced by industrial countries. These contemporary transitions are often moving at a more rapid pace than in the past. As in industrial countries, gains in survival have been due to socioeconomic progress, control of infectious diseases, the diffusion of public health and medical technology, and changes in health-related behaviors. Progress in mortality reduction has depended to some extent on effective government intervention in all these processes.
Life expectancy should continue to rise in the future, perhaps indefinitely. Some analysts have broached the possibility that it will reach some natural ceiling because improvements in industrial countries have come more slowly in recent years. This slowdown, however, has an explanation. In these countries, survival through infancy, childhood, and young adulthood is so likely that further gains in years lived at these ages have greatly diminished. In contrast, mortality rates at older ages have continued to decline steadily since the 1960s. Although gains may slow for developing regions as they approach industrial-country levels, further breakthroughs in medicine and biotechnology as well as behavioral changes are likely to sustain a continuing upward trend in life expectancy for most countries and for the world as a whole.
Past forecasts have somewhat underestimated actual gains in life expectancy. The consequences have been slight for population totals but more substantial for estimates of the elderly. As a result, the potential challenges to retirement and social security programs have been understated. Removing assumed ceilings to life expectancy in projections may prevent this problem from recurring.
Mortality transitions have been interrupted in some developing countries. Most of these are in Sub-Saharan Africa, which has been substantially affected by the HIV/AIDS epidemic. In a few countries with very severe epidemics, life expectancy is already falling and could drop eventually by 10 years or more. Nevertheless, population size has not declined in these countries and is not projected to do so.
The HIV/AIDS epidemic constitutes an unexpected mortality crisis. Mostly as a result of this crisis, Sub-Saharan Africa is the only world region for which past life expectancy projections have been too high. In recent decades, another crisis with somewhat similar broad impact was the collapse of the Communist bloc, which accelerated the fall in life expectancies in Eastern Europe. Crises of this type and scale do affect long-term trends in life expectancy, unlike most natural disasters, which merely produce fluctuations in trend.
The number of crises and their global impact have diminished over time as life expectancy has risen and as populations have grown. But, as the example of Eastern Europe illustrates, such crises cannot be ruled out altogether for the future. Nor can they really be foreseen. The best that can be done is to revise projections often, and certainly soon after such events are recognized and their potential impact can be assessed.
Worldwide, the number of people born in one country but resident in another has increased from 75 million in 1965 to 120 million in 1990. Today, the foreign-born constitute slightly over 2 percent of world population. This proportion has remained stable for the last 30 years as world population size has risen correspondingly (see Chapter 6).
International migrants have consistently moved toward areas of economic opportunity. From the beginning of the 19th century until World War I, the main flow was from Europe to six “traditional countries of immigration”: the United States and Canada, Australia and New Zealand, and Brazil and Argentina. After World War II, immigrants increasingly came from the developing regions, and their destinations multiplied. Besides the traditional countries of immigration, important additional destinations have emerged in Western Europe, the Persian Gulf, Japan, and other rapidly growing East Asian economies.
The effect of such migration on population growth is small for the majority of countries. However, in about 15 countries, the recent net inflow of migrants exceeded 1 percent of the population annually, and in about as many countries the net outflow was equally large. Net migration into the industrial world as a whole (excluding migration within it) was about 0.4 per thousand population annually in the 1950s and 1960s, and then rose to 1.6-1.9 per thousand by the late 1980s and early 1990s—around 2 million people annually.
Once established, the flow of migrants between two countries tends to be sustained and even expanded by networks of transportation, communication, politics, and culture. The flow usually does not decline until incentives for migration, especially international wage gaps, diminish. The flow may be constrained by public policy in the receiving countries, which has become increasingly restrictive. However, policy may itself be undercut not simply by weak enforcement but also by the forces of globalization. These forces promote international movement of capital, goods, services, and ideas and thus also facilitate international movements of people. One might therefore expect no substantial decrease in net migration into the main receiving countries, probably for decades.
International migration is difficult to predict partly because it is affected by the complex process of policy development and enforcement in the main receiving countries. One type of migration that has been especially problematic to predict has been sudden surges of migrants that often result from political, economic, or environmental crises. In the 1990s, unexpected flows of migrants have occurred after humanitarian catastrophes such as the mass killings in Rwanda, civil war in Liberia, and the invasion of Kuwait. Mass migrations of this sort have been the primary cause of “demographic quakes”—sudden and extreme changes in population growth rates—and have been a major source of error in past population projections in the affected countries. No adequate methodology exists to predict such events or the demographic response to them. As is the case for crisis mortality, the best that forecasters can do is to update their projections to take account of such events as quickly as possible.
THE UNCERTAINTY OF PROJECTIONS
While broad trends in fertility, mortality, and migration can be discerned and projected into the future with reasonable confidence, substantial uncertainty is attached to the specific trend for any particular country or region. Quantifying this uncertainty is helpful to users of projection results, such as social security actuaries or environmental modelers, because it focuses their attention on alternative population futures that may
have different implications and requires them to decide what forecast horizon to take seriously (see Chapter 7).
In current forecasts, uncertainty is typically expressed by providing alternative scenarios, varying the trajectory for fertility (and, rarely, for mortality and migration). “High” and “low” scenarios are used to indicate a range of possible futures. However, no specific probability is attached to the range, and what it means is therefore unclear.
Probability distributions for projection error can be estimated using an ex post approach. We analyze the distribution of past errors in U.N. forecasts over two decades and use this information, by way of stochastic simulations, to define predictive intervals for the current medium U.N. projection. The approach assumes that the accuracy of current forecasts will be closely related to that of past forecasts.
We estimate that a 95-percent prediction interval for world population in 2030 would extend from 7.5 to 8.9 billion, and a similar interval for world population in 2050 would extend from 7.9 to 10.9 billion. The intervals are asymmetric around the U.N. medium projection of 8.9 billion in 2050. This indicates that, based on the record of previous projections, a greater risk exists of a large understatement of future world growth than of a large overstatement. The intervals suggest that world population decline between 2000 and 2050 is quite unlikely.
Because many country errors cancel each other after aggregation, these prediction intervals for world population are proportionally much narrower than those for individual countries. Across 13 large countries, the median prediction interval for population in 50 years runs from 30 percent below the point forecast to 43 percent above it, for a total width of 73 percent. This width is more than three times the width of the corresponding projection interval for world population. The width of intervals for regional projections is intermediate, reflecting an intermediate degree of aggregation. The width of prediction intervals does vary greatly across countries—in line with the errors in past projections—and tends to be larger for smaller countries, especially in developing regions.
The historical record on which these prediction intervals are based includes some major unanticipated influences on demographic behavior, such as the HIV/AIDS epidemic, civil wars, and other disturbances that produced crisis migration and mortality in several countries. To some extent, therefore, these predictive intervals allow for unexpected events. But it is always possible that the future will see developments different in kind from those in the past few decades. These probability distributions do not and cannot allow for such unprecedented catastrophes as nuclear war. If such events occurred, the planning that projections are intended to inform would be of little relevance.
Current world population projections from the U.N. and the World Bank incorporate the major expected trends in population growth components: continuing decline in fertility in developing countries to low levels; persistence of fertility at these levels in demographically advanced countries; continued rise in life expectancy, although at a slower pace globally than in previous decades; and the persistence of migration into the major receiving countries. The panel's review finds that the projection assumptions regarding future trends in fertility, mortality, and migration are generally supported by available scientific evidence.
The panel therefore concludes that these current world population projections to 2050 are based on reasonable assumptions and provide plausible forecasts of world demographic trends for the next few decades. The relatively small global errors made in past projections are consistent with this conclusion. This conclusion does not imply any endorsement of the projections made for specific countries; the panel has examined the general methodology of world projections, not the particular input data and assumed trends applied to each country. Projections made by other agencies, such as the U.S. Census Bureau and the International Institute for Applied Systems Analysis, could not be analyzed in the same detail.
The implication of this finding is that the population of the world will probably grow from 6 billion today to between 8 and 11 billion in 2050. Nearly all this growth will take place in developing regions. In contrast, population size is expected to change little from its current level in the industrial world as a whole and will probably decline in a number of countries. Expected trends in fertility and mortality will lead to substantial changes in the age structure of populations and especially to increasing proportions of elderly in all regions of the world.
The panel has identified several improvements that could be made in current projection methodology. The simplifying assumptions that projections typically make about future trends in fertility, mortality, and migration could be refined. Projections are likely to improve if forecasters:
Reduce the assumed pace of fertility decline as fertility approaches the replacement level in countries now in transition.
Impose no assumed ceiling on life expectancy.
Maintain net migration around current levels for several decades for large receiving countries.
Use more reliable baseline data. This requires further investments in censuses, surveys, and vital registration.
Update projections quickly as new information on current demographic trends becomes available.
Current projections for some countries would have been somewhat different if these recommended improvements had been adopted earlier. However, the overall effect at the world and regional level would have been minor.
Users of projections would also benefit from clearer presentation of the underlying methodology and assumptions and from rapid dissemination of projections in electronic formats. How they actually make use of projections is a question worth investigating, which could lead to improvements in presentation.
Official projections have neglected the important issue of the uncertainty surrounding forecasts. The potential for error in projections rises with the length of the projection interval. Projections of population size are relatively accurate up to one or two decades into the future, but beyond that period uncertainty accumulates rapidly and nonlinearly. For example, the upper bound of the estimated 95-percent prediction interval for world population size is 10 percent higher than the medium estimate for 2030 but 22 percent higher for 2050. Prediction intervals are substantially wider for country-level projections. Future projections should explicitly acknowledge this uncertainty and develop the methodology necessary to quantify it. The report illustrates how this can be done.
Various types of population research could make population forecasts more accurate. Better estimates of demographic parameters have been critical in improving forecasts in the past, especially for developing countries. More accurate data not only improve the base estimates from which projections start but also enhance understanding of the demographic dynamics of specific countries. Better data would be especially useful for smaller countries and for international migration. In neither case are baseline errors of much global consequence, but projections of small countries and of international migration are of interest in their own right.
More research on trends in the components of population change, their determinants, and their statistical modeling would be valuable to forecasters. While much attention has been paid to determinants of and
differentials in fertility levels, for instance, much less attention has been given to how the pace of fertility decline varies during fertility transition and to the reasons for any variation. Some approaches to modeling trends are mentioned in the report, including projection of age-specific mortality rates, the use of dynamic models for migration flows, and the application of time-series methods. Each of these approaches requires more research before being applied to world projections.
Long-term projections are strongly influenced by assumptions about levels of fertility and mortality decades from now. Direct research on these long-term levels is impossible, but analyses of past and current trends have already clarified the prospects and could make them even clearer. Research, interdisciplinary when necessary, on the reasons for low fertility levels, on patterns in and causes of birth delay, and on the nature and causes of gains in life expectancy, especially at older ages, would be highly desirable.
Considerable uncertainty in projections stems from unexpected events. Wars, natural disasters, economic crises, and similar events can generate streams of migrants, suppress fertility temporarily, and produce many premature deaths. On the positive side, unexpected biomedical breakthroughs may lead to large increases in life expectancy or provide new fertility options. Environmental crises that may loom in the future are an additional concern. Events of such types are not within the competence of demographers to predict. Their involvement in interdisciplinary work is essential to obtain a better appreciation of the likelihood and demographic implications of such events.
The components of population change can be influenced by public policy, and this issue invites further research on a number of questions. How do policies and programs affect the pace of fertility decline in developing regions? Could any new policy initiatives substantially raise very low fertility? Will constraints in public expenditures slow mortality reductions in old age? Are further policy restrictions on immigrants in receiving countries probable, and what effects are policy changes likely to have?
Uncertainty is inevitable in forecasts, but if it cannot be avoided, to some extent it can be quantified. The panel has attempted to estimate prediction intervals for projected population based on an analysis of errors in past projections. The analysis involved assessments of the accuracy of more than 1,000 country projections made in recent decades but was still limited because the latest projections cannot yet be assessed. The modeling of uncertainty illustrated here should be considered a first step, subject to subsequent deepening and refinement. Extending the analysis and the modeling should have high priority.