2001 Probabilistic Population Projections

Introduction to Probabilistic Forecasting

"Over the past two decades, roughly two billion people were added to the world's population. No other decade in human history has seen such an increase in numbers. And the coming decades will continue to see significant world population growth. Population size will most likely increase by at least two more billion to over 8 billion people. But during the last part of this century the world will likely experience no further population growth and possibly even see the beginning of a decline. Sometime before 2100, the great centuries-long expansion of the world's population will have come to an end."  (Lutz, Sanderson, and Scherbov, 2004)

Forecast results:

• Short descriptive summary of the projection results
• Download the IIASA-2001-projections-data:
      (a) by region, median and, decile distribution
      (b) by region and single simulation runs

Why do we use probabilistic forecasting methodology?

The common way to forecast future population is based on the assumptions of "variants" (low, medium and high variants). These variants are said to cover a "plausible range of future population trends." The medium variant is usually the best "guess" of what will happen. This approach is imprecise in the sense that it does not tell the user what "plausible" means. And this approach is incomplete because it ignores uncertainties in mortality, fertility, and migration assumptions. And it is statistically deficient, because when high or low variants for world regions are computed by adding up the low or high variants of the country projections, the likelihood that all countries follow the same variant paths simultaneously is implausible. The population projections presented here avoid these problems of the "variant approach" because they are explicit in stating probabilities; they fully incorporate uncertainties in fertility, mortality and migration; and they scale up from regions to the world in a statistically consistent manner.  

What is a probabilistic projection? A basic description

The word "probabilistic" means that we try to quantify the uncertainties involved in the population projections presented here. Basically, the cohort component population projection approach, which is used in all IIASA population projections, begins with a starting population and moves forward in time using age-specific birth and death rates, and where relevant, age-specific migration and educational transition rates. When those rates, at each time period, are given as distributions rather than as points, the projections become probabilistic. The outputs of probabilistic projections are distributions of variables of interest, such as population size, at each date. For example, a probabilistic projection of the world's population in 2050 might say that it has an 80 percent probability of being between 8 and 10 billion people.

What is, and why do we use argument-based assumptions?

There are different views among population experts about what is the most likely future path of life expectancy, fertility rates and migration rates in any particular country. The process of the evaluation of alternative views is almost never documented for the user. The approach used here to produce this set of new population forecasts is called expert argument-based probabilistic forecasting. We refer here to what is called subjective or judgmental probabilities. That means that there is no perfect objective model for the definition of future trends of the three determinants of population change. Instead, the task is to collect the best available information to make an informed judgment about the likely uncertainty distribution of certain future demographic trends, and goes a step further in the direction of making the expert-based assumption more objective. That means it does not derive the median and the associated distribution from entirely different resources. It uses additional checks against underestimation of uncertainty by experts by including empirical errors of past population projections (ex post error analysis) in the uncertainty distribution for the future. The ex post error analysis enters the study in two ways: the substantive assumptions made on fertility and mortality changes are informed by the analyses of past errors in those components, and our results at the regional level have been compared to the results of an ex post error analysis of global UN projections documented in the NRC report (2000).

A small non-specified sample of forecast results    

When you move the mouse over this image, you will see a sample of probabilistic forecast results for different IIASA world regions. The sample includes the figures of the proportion of population 80 years and older, or 20 years and younger, of the total population, fertility, mortality, and others. The plots illustrate the 95% probability range: the central line is the median, the blue area is the 40-60 percentile range, the green is the 20-80, and the yellow is the 2.5-97.5 percentile range.

Click here to download selected probabilistic population forecast results by 13 world regions as spreadsheets.

Generating 2,000 simulations and inter-regional correlations

The projections are carried out using the cohort component method for single years of age and single years in time. The population is calculated by age and sex as it changes from one year to the next, based on the assumptions of fertility and mortality rates, as well as migration. The forecasts are carried out for 13 world regions. The forecasts presented here are not alternative scenarios or variants, but the distribution of the results of 2,000 different cohort component projections. For these stochastic simulations the fertility, mortality and migration paths underlying the individual projection runs were derived randomly from the described uncertainty distribution for fertility, mortality and migration in the different world regions.

Due to increasing globalization of medical technology as well as of new threats to life, it is assumed that the inter-regional correlations of the deviations from the expected trends in life expectancy improvements are very high. For fertility, increasing globalization of the media as well as reproductive technology are likely to result in high global correlations. For the results presented here, the inter-regional correlations in the deviations from expected trends were assumed to be 0.9 in the case of life expectancy and 0.7 in the case of fertility.

Mathematical details are published in

Last edited: 06 November 2012