Fertility, mortality, migration, and population scenarios for 195 countries and territories from 2017 to 2100: a forecasting analysis for the Global Burden of Disease Study
Research in context
Evidence before this study
Global population projections have been produced by the Population Division of the Department of Economic and Social Affairs of the UN Secretariat (UNPD) since the 1950s. For many years, UNPD used a deterministic model for fertility, mortality, and migration. Structural scenarios were also computed by assuming a fixed difference of 0·5 children in the total fertility rate (TFR) in each time period and country. Beginning in 2010, UNPD adopted a statistical model for the TFR and life expectancy as functions of calendar year and a deterministic model for migration. This blend of statistical models for two of the components of population growth has been used to generate uncertainty intervals (UIs). In fitting their global model for low fertility recovery, UNPD has excluded countries with sustained low fertility such as Thailand, South Korea, Canada, and Greece. Estimated in this way, the UNPD predicts TFRs will rebound to approximately 1·75 in all countries with TFR lower than the replacement level (<2·1).
Since the 1990s, the International Institute for Applied Systems Analysis–Wittgenstein Centre has generated alternative population projections. The Wittgenstein Centre fertility forecasts are a blend of expert opinions about future fertility patterns and statistical modelling. For low-fertility countries, they assume that the TFR will converge to 1·75 in the year 2200. Expert judgment is also used by the Wittgenstein Centre to set assumptions of future mortality, migration, and education that are combined with statistical modelling to produce future population scenarios. Their hybrid approach does not generate UIs for population projections. By accounting for educational attainment in the qualitative assessment, Wittgenstein predicts much faster declines in the TFR in sub-Saharan Africa than those by UNPD.
Added value of this study
In our study, we improved on UNPD and Wittgenstein forecasts in seven important ways. First, we modelled completed cohort fertility at age 50 years (CCF50) rather than the TFR. CCF50 is much less affected by the delay of childbearing that occurs as females become more educated, which leads the period measure of the TFR to initially decline to low levels and then increase. By contrast, completed cohort fertility rarely increases, making the modelling of CCF50 much more stable. Second, we modelled CCF50 as a function of educational attainment and contraceptive met need. These two variables alone account for 80·5% of the variance in CCF50 over time and location. Third, we used the causal model to explore the effect of faster or slower than expected changes in educational attainment and contraceptive met need. These scenarios, unlike structural scenarios, can provide direct guidance to policy debates on the impact of faster or slower scale-up of educational attainment or access to reproductive health services. Fourth, we leveraged the previously published future health scenarios model for cause-specific and all-cause mortality; this model also allows the effect of faster scale-up of educational attainment on mortality to be captured. Fifth, rather than assume deterministic patterns of migration, we fitted a time-series model with covariates (Socio-demographic Index, crude population growth rate, and deaths from war and natural disasters) to national net migration rates. By making explicit the pathways through which fertility, mortality, and migration patterns can change, our model is able to identify where future time trends might be different from past trends. Sixth, uncertainty in all three components (fertility, mortality, and migration) were propagated into the uncertainty distributions for each country and territory in each year. Seventh, we traced the changes in age structure expected in the reference and alternative scenarios on total gross domestic product (GDP) using previously published forecasts of GDP per adult of working age.
Implications of all the available evidence
Our reference forecast of the global population in 2100 was lower than the Wittgenstein Centre forecast and much lower than the UNPD forecast. Our findings suggest that, because of progress in female educational attainment and access to contraception contributing to declining fertility rates, continued global population growth through the century is no longer the most likely trajectory for the world's population. By contrast, world population might peak just after mid-century and substantially decline by 2100. The difference in population forecasts between our reference scenario and the UNPD forecasts is a third due to faster declines in sub-Saharan African fertility and two thirds due to the lower level of TFR expected in populations with fertility lower than the replacement level, especially China and India. Our findings show that some countries with fertility lower than replacement level, such as the USA, Australia, and Canada, will probably maintain their working-age populations through net immigration. Our forecasts for a shrinking global population have positive implications for the environment, climate change, and food production, but possible negative implications for labour forces, economic growth, and social support systems in parts of the world with the greatest fertility declines.